Initial checkin

BeefRT/gperftools/src/addressmap-inl.h

@@ -0,0 +1,421 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//
+// A fast map from addresses to values.  Assumes that addresses are
+// clustered.  The main use is intended to be for heap-profiling.
+// May be too memory-hungry for other uses.
+//
+// We use a user-defined allocator/de-allocator so that we can use
+// this data structure during heap-profiling.
+//
+// IMPLEMENTATION DETAIL:
+//
+// Some default definitions/parameters:
+//  * Block      -- aligned 128-byte region of the address space
+//  * Cluster    -- aligned 1-MB region of the address space
+//  * Block-ID   -- block-number within a cluster
+//  * Cluster-ID -- Starting address of cluster divided by cluster size
+//
+// We use a three-level map to represent the state:
+//  1. A hash-table maps from a cluster-ID to the data for that cluster.
+//  2. For each non-empty cluster we keep an array indexed by
+//     block-ID tht points to the first entry in the linked-list
+//     for the block.
+//  3. At the bottom, we keep a singly-linked list of all
+//     entries in a block (for non-empty blocks).
+//
+//    hash table
+//  +-------------+
+//  | id->cluster |---> ...
+//  |     ...     |
+//  | id->cluster |--->  Cluster
+//  +-------------+     +-------+    Data for one block
+//                      |  nil  |   +------------------------------------+
+//                      |   ----+---|->[addr/value]-->[addr/value]-->... |
+//                      |  nil  |   +------------------------------------+
+//                      |   ----+--> ...
+//                      |  nil  |
+//                      |  ...  |
+//                      +-------+
+//
+// Note that we require zero-bytes of overhead for completely empty
+// clusters.  The minimum space requirement for a cluster is the size
+// of the hash-table entry plus a pointer value for each block in
+// the cluster.  Empty blocks impose no extra space requirement.
+//
+// The cost of a lookup is:
+//      a. A hash-table lookup to find the cluster
+//      b. An array access in the cluster structure
+//      c. A traversal over the linked-list for a block
+
+#ifndef BASE_ADDRESSMAP_INL_H_
+#define BASE_ADDRESSMAP_INL_H_
+
+#include "tcconfig.h"
+#include <stddef.h>
+#include <string.h>
+#if defined HAVE_STDINT_H
+#include <stdint.h>             // to get uint16_t (ISO naming madness)
+#elif defined HAVE_INTTYPES_H
+#include <inttypes.h>           // another place uint16_t might be defined
+#else
+#include <sys/types.h>          // our last best hope
+#endif
+
+// This class is thread-unsafe -- that is, instances of this class can
+// not be accessed concurrently by multiple threads -- because the
+// callback function for Iterate() may mutate contained values. If the
+// callback functions you pass do not mutate their Value* argument,
+// AddressMap can be treated as thread-compatible -- that is, it's
+// safe for multiple threads to call "const" methods on this class,
+// but not safe for one thread to call const methods on this class
+// while another thread is calling non-const methods on the class.
+template <class Value>
+class AddressMap {
+ public:
+  typedef void* (*Allocator)(size_t size);
+  typedef void  (*DeAllocator)(void* ptr);
+  typedef const void* Key;
+
+  // Create an AddressMap that uses the specified allocator/deallocator.
+  // The allocator/deallocator should behave like malloc/free.
+  // For instance, the allocator does not need to return initialized memory.
+  AddressMap(Allocator alloc, DeAllocator dealloc);
+  ~AddressMap();
+
+  // If the map contains an entry for "key", return it. Else return NULL.
+  inline const Value* Find(Key key) const;
+  inline Value* FindMutable(Key key);
+
+  // Insert <key,value> into the map.  Any old value associated
+  // with key is forgotten.
+  void Insert(Key key, Value value);
+
+  // Remove any entry for key in the map.  If an entry was found
+  // and removed, stores the associated value in "*removed_value"
+  // and returns true.  Else returns false.
+  bool FindAndRemove(Key key, Value* removed_value);
+
+  // Similar to Find but we assume that keys are addresses of non-overlapping
+  // memory ranges whose sizes are given by size_func.
+  // If the map contains a range into which "key" points
+  // (at its start or inside of it, but not at the end),
+  // return the address of the associated value
+  // and store its key in "*res_key".
+  // Else return NULL.
+  // max_size specifies largest range size possibly in existence now.
+  typedef size_t (*ValueSizeFunc)(const Value& v);
+  const Value* FindInside(ValueSizeFunc size_func, size_t max_size,
+                          Key key, Key* res_key);
+
+  // Iterate over the address map calling 'callback'
+  // for all stored key-value pairs and passing 'arg' to it.
+  // We don't use full Closure/Callback machinery not to add
+  // unnecessary dependencies to this class with low-level uses.
+  template<class Type>
+  inline void Iterate(void (*callback)(Key, Value*, Type), Type arg) const;
+
+ private:
+  typedef uintptr_t Number;
+
+  // The implementation assumes that addresses inserted into the map
+  // will be clustered.  We take advantage of this fact by splitting
+  // up the address-space into blocks and using a linked-list entry
+  // for each block.
+
+  // Size of each block.  There is one linked-list for each block, so
+  // do not make the block-size too big.  Oterwise, a lot of time
+  // will be spent traversing linked lists.
+  static const int kBlockBits = 7;
+  static const int kBlockSize = 1 << kBlockBits;
+
+  // Entry kept in per-block linked-list
+  struct Entry {
+    Entry* next;
+    Key    key;
+    Value  value;
+  };
+
+  // We further group a sequence of consecutive blocks into a cluster.
+  // The data for a cluster is represented as a dense array of
+  // linked-lists, one list per contained block.
+  static const int kClusterBits = 13;
+  static const Number kClusterSize = 1 << (kBlockBits + kClusterBits);
+  static const int kClusterBlocks = 1 << kClusterBits;
+
+  // We use a simple chaining hash-table to represent the clusters.
+  struct Cluster {
+    Cluster* next;                      // Next cluster in hash table chain
+    Number   id;                        // Cluster ID
+    Entry*   blocks[kClusterBlocks];    // Per-block linked-lists
+  };
+
+  // Number of hash-table entries.  With the block-size/cluster-size
+  // defined above, each cluster covers 1 MB, so an 4K entry
+  // hash-table will give an average hash-chain length of 1 for 4GB of
+  // in-use memory.
+  static const int kHashBits = 12;
+  static const int kHashSize = 1 << 12;
+
+  // Number of entry objects allocated at a time
+  static const int ALLOC_COUNT = 64;
+
+  Cluster**     hashtable_;              // The hash-table
+  Entry*        free_;                   // Free list of unused Entry objects
+
+  // Multiplicative hash function:
+  // The value "kHashMultiplier" is the bottom 32 bits of
+  //    int((sqrt(5)-1)/2 * 2^32)
+  // This is a good multiplier as suggested in CLR, Knuth.  The hash
+  // value is taken to be the top "k" bits of the bottom 32 bits
+  // of the muliplied value.
+  static const uint32_t kHashMultiplier = 2654435769u;
+  static int HashInt(Number x) {
+    // Multiply by a constant and take the top bits of the result.
+    const uint32_t m = static_cast<uint32_t>(x) * kHashMultiplier;
+    return static_cast<int>(m >> (32 - kHashBits));
+  }
+
+  // Find cluster object for specified address.  If not found
+  // and "create" is true, create the object.  If not found
+  // and "create" is false, return NULL.
+  //
+  // This method is bitwise-const if create is false.
+  Cluster* FindCluster(Number address, bool create) {
+    // Look in hashtable
+    const Number cluster_id = address >> (kBlockBits + kClusterBits);
+    const int h = HashInt(cluster_id);
+    for (Cluster* c = hashtable_[h]; c != NULL; c = c->next) {
+      if (c->id == cluster_id) {
+        return c;
+      }
+    }
+
+    // Create cluster if necessary
+    if (create) {
+      Cluster* c = New<Cluster>(1);
+      c->id = cluster_id;
+      c->next = hashtable_[h];
+      hashtable_[h] = c;
+      return c;
+    }
+    return NULL;
+  }
+
+  // Return the block ID for an address within its cluster
+  static int BlockID(Number address) {
+    return (address >> kBlockBits) & (kClusterBlocks - 1);
+  }
+
+  //--------------------------------------------------------------
+  // Memory management -- we keep all objects we allocate linked
+  // together in a singly linked list so we can get rid of them
+  // when we are all done.  Furthermore, we allow the client to
+  // pass in custom memory allocator/deallocator routines.
+  //--------------------------------------------------------------
+  struct Object {
+    Object* next;
+    // The real data starts here
+  };
+
+  Allocator     alloc_;                 // The allocator
+  DeAllocator   dealloc_;               // The deallocator
+  Object*       allocated_;             // List of allocated objects
+
+  // Allocates a zeroed array of T with length "num".  Also inserts
+  // the allocated block into a linked list so it can be deallocated
+  // when we are all done.
+  template <class T> T* New(int num) {
+    void* ptr = (*alloc_)(sizeof(Object) + num*sizeof(T));
+    memset(ptr, 0, sizeof(Object) + num*sizeof(T));
+    Object* obj = reinterpret_cast<Object*>(ptr);
+    obj->next = allocated_;
+    allocated_ = obj;
+    return reinterpret_cast<T*>(reinterpret_cast<Object*>(ptr) + 1);
+  }
+};
+
+// More implementation details follow:
+
+template <class Value>
+AddressMap<Value>::AddressMap(Allocator alloc, DeAllocator dealloc)
+  : free_(NULL),
+    alloc_(alloc),
+    dealloc_(dealloc),
+    allocated_(NULL) {
+  hashtable_ = New<Cluster*>(kHashSize);
+}
+
+template <class Value>
+AddressMap<Value>::~AddressMap() {
+  // De-allocate all of the objects we allocated
+  for (Object* obj = allocated_; obj != NULL; /**/) {
+    Object* next = obj->next;
+    (*dealloc_)(obj);
+    obj = next;
+  }
+}
+
+template <class Value>
+inline const Value* AddressMap<Value>::Find(Key key) const {
+  return const_cast<AddressMap*>(this)->FindMutable(key);
+}
+
+template <class Value>
+inline Value* AddressMap<Value>::FindMutable(Key key) {
+  const Number num = reinterpret_cast<Number>(key);
+  const Cluster* const c = FindCluster(num, false/*do not create*/);
+  if (c != NULL) {
+    for (Entry* e = c->blocks[BlockID(num)]; e != NULL; e = e->next) {
+      if (e->key == key) {
+        return &e->value;
+      }
+    }
+  }
+  return NULL;
+}
+
+template <class Value>
+void AddressMap<Value>::Insert(Key key, Value value) {
+  const Number num = reinterpret_cast<Number>(key);
+  Cluster* const c = FindCluster(num, true/*create*/);
+
+  // Look in linked-list for this block
+  const int block = BlockID(num);
+  for (Entry* e = c->blocks[block]; e != NULL; e = e->next) {
+    if (e->key == key) {
+      e->value = value;
+      return;
+    }
+  }
+
+  // Create entry
+  if (free_ == NULL) {
+    // Allocate a new batch of entries and add to free-list
+    Entry* array = New<Entry>(ALLOC_COUNT);
+    for (int i = 0; i < ALLOC_COUNT-1; i++) {
+      array[i].next = &array[i+1];
+    }
+    array[ALLOC_COUNT-1].next = free_;
+    free_ = &array[0];
+  }
+  Entry* e = free_;
+  free_ = e->next;
+  e->key = key;
+  e->value = value;
+  e->next = c->blocks[block];
+  c->blocks[block] = e;
+}
+
+template <class Value>
+bool AddressMap<Value>::FindAndRemove(Key key, Value* removed_value) {
+  const Number num = reinterpret_cast<Number>(key);
+  Cluster* const c = FindCluster(num, false/*do not create*/);
+  if (c != NULL) {
+    for (Entry** p = &c->blocks[BlockID(num)]; *p != NULL; p = &(*p)->next) {
+      Entry* e = *p;
+      if (e->key == key) {
+        *removed_value = e->value;
+        *p = e->next;         // Remove e from linked-list
+        e->next = free_;      // Add e to free-list
+        free_ = e;
+        return true;
+      }
+    }
+  }
+  return false;
+}
+
+template <class Value>
+const Value* AddressMap<Value>::FindInside(ValueSizeFunc size_func,
+                                           size_t max_size,
+                                           Key key,
+                                           Key* res_key) {
+  const Number key_num = reinterpret_cast<Number>(key);
+  Number num = key_num;  // we'll move this to move back through the clusters
+  while (1) {
+    const Cluster* c = FindCluster(num, false/*do not create*/);
+    if (c != NULL) {
+      while (1) {
+        const int block = BlockID(num);
+        bool had_smaller_key = false;
+        for (const Entry* e = c->blocks[block]; e != NULL; e = e->next) {
+          const Number e_num = reinterpret_cast<Number>(e->key);
+          if (e_num <= key_num) {
+            if (e_num == key_num  ||  // to handle 0-sized ranges
+                key_num < e_num + (*size_func)(e->value)) {
+              *res_key = e->key;
+              return &e->value;
+            }
+            had_smaller_key = true;
+          }
+        }
+        if (had_smaller_key) return NULL;  // got a range before 'key'
+                                           // and it did not contain 'key'
+        if (block == 0) break;
+        // try address-wise previous block
+        num |= kBlockSize - 1;  // start at the last addr of prev block
+        num -= kBlockSize;
+        if (key_num - num > max_size) return NULL;
+      }
+    }
+    if (num < kClusterSize) return NULL;  // first cluster
+    // go to address-wise previous cluster to try
+    num |= kClusterSize - 1;  // start at the last block of previous cluster
+    num -= kClusterSize;
+    if (key_num - num > max_size) return NULL;
+      // Having max_size to limit the search is crucial: else
+      // we have to traverse a lot of empty clusters (or blocks).
+      // We can avoid needing max_size if we put clusters into
+      // a search tree, but performance suffers considerably
+      // if we use this approach by using stl::set.
+  }
+}
+
+template <class Value>
+template <class Type>
+inline void AddressMap<Value>::Iterate(void (*callback)(Key, Value*, Type),
+                                       Type arg) const {
+  // We could optimize this by traversing only non-empty clusters and/or blocks
+  // but it does not speed up heap-checker noticeably.
+  for (int h = 0; h < kHashSize; ++h) {
+    for (const Cluster* c = hashtable_[h]; c != NULL; c = c->next) {
+      for (int b = 0; b < kClusterBlocks; ++b) {
+        for (Entry* e = c->blocks[b]; e != NULL; e = e->next) {
+          callback(e->key, &e->value, arg);
+        }
+      }
+    }
+  }
+}
+
+#endif  // BASE_ADDRESSMAP_INL_H_

BeefRT/gperftools/src/base/arm_instruction_set_select.h

@@ -0,0 +1,79 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+// Author: Alexander Levitskiy
+//
+// Generalizes the plethora of ARM flavors available to an easier to manage set
+// Defs reference is at https://wiki.edubuntu.org/ARM/Thumb2PortingHowto
+
+#ifndef ARM_INSTRUCTION_SET_SELECT_H_
+#define ARM_INSTRUCTION_SET_SELECT_H_
+
+#if defined(__ARM_ARCH_7__) || \
+    defined(__ARM_ARCH_7R__) || \
+    defined(__ARM_ARCH_7A__)
+# define ARMV7 1
+#endif
+
+#if defined(ARMV7) || \
+    defined(__ARM_ARCH_6__) || \
+    defined(__ARM_ARCH_6J__) || \
+    defined(__ARM_ARCH_6K__) || \
+    defined(__ARM_ARCH_6Z__) || \
+    defined(__ARM_ARCH_6T2__) || \
+    defined(__ARM_ARCH_6ZK__)
+# define ARMV6 1
+#endif
+
+#if defined(ARMV6) || \
+    defined(__ARM_ARCH_5T__) || \
+    defined(__ARM_ARCH_5E__) || \
+    defined(__ARM_ARCH_5TE__) || \
+    defined(__ARM_ARCH_5TEJ__)
+# define ARMV5 1
+#endif
+
+#if defined(ARMV5) || \
+    defined(__ARM_ARCH_4__) || \
+    defined(__ARM_ARCH_4T__)
+# define ARMV4 1
+#endif
+
+#if defined(ARMV4) || \
+    defined(__ARM_ARCH_3__) || \
+    defined(__ARM_ARCH_3M__)
+# define ARMV3 1
+#endif
+
+#if defined(ARMV3) || \
+    defined(__ARM_ARCH_2__)
+# define ARMV2 1
+#endif
+
+#endif  // ARM_INSTRUCTION_SET_SELECT_H_

BeefRT/gperftools/src/base/atomicops-internals-arm-generic.h

@@ -0,0 +1,259 @@
+// Copyright (c) 2003, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+// ---
+//
+// Author: Lei Zhang, Sasha Levitskiy
+//
+// This file is an internal atomic implementation, use base/atomicops.h instead.
+//
+// LinuxKernelCmpxchg and Barrier_AtomicIncrement are from Google Gears.
+
+#ifndef BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_
+#define BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_
+
+#include <stdio.h>
+#include <stdlib.h>
+#include "base/basictypes.h"
+
+typedef int32_t Atomic32;
+
+namespace base {
+namespace subtle {
+
+typedef int64_t Atomic64;
+
+// 0xffff0fc0 is the hard coded address of a function provided by
+// the kernel which implements an atomic compare-exchange. On older
+// ARM architecture revisions (pre-v6) this may be implemented using
+// a syscall. This address is stable, and in active use (hard coded)
+// by at least glibc-2.7 and the Android C library.
+// pLinuxKernelCmpxchg has both acquire and release barrier sematincs.
+typedef Atomic32 (*LinuxKernelCmpxchgFunc)(Atomic32 old_value,
+                                           Atomic32 new_value,
+                                           volatile Atomic32* ptr);
+LinuxKernelCmpxchgFunc pLinuxKernelCmpxchg ATTRIBUTE_WEAK =
+    (LinuxKernelCmpxchgFunc) 0xffff0fc0;
+
+typedef void (*LinuxKernelMemoryBarrierFunc)(void);
+LinuxKernelMemoryBarrierFunc pLinuxKernelMemoryBarrier ATTRIBUTE_WEAK =
+    (LinuxKernelMemoryBarrierFunc) 0xffff0fa0;
+
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  Atomic32 prev_value = *ptr;
+  do {
+    if (!pLinuxKernelCmpxchg(old_value, new_value,
+                             const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr,
+                                         Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (pLinuxKernelCmpxchg(old_value, new_value,
+                               const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  // pLinuxKernelCmpxchg already has acquire and release barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  // pLinuxKernelCmpxchg already has acquire and release barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32* ptr,
+                                        Atomic32 increment) {
+  for (;;) {
+    // Atomic exchange the old value with an incremented one.
+    Atomic32 old_value = *ptr;
+    Atomic32 new_value = old_value + increment;
+    if (pLinuxKernelCmpxchg(old_value, new_value,
+                            const_cast<Atomic32*>(ptr)) == 0) {
+      // The exchange took place as expected.
+      return new_value;
+    }
+    // Otherwise, *ptr changed mid-loop and we need to retry.
+  }
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32* ptr,
+                                          Atomic32 increment) {
+  return Barrier_AtomicIncrement(ptr, increment);
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+inline void MemoryBarrier() {
+  pLinuxKernelMemoryBarrier();
+}
+
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) {
+  MemoryBarrier();
+  *ptr = value;
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) {
+  Atomic32 value = *ptr;
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32* ptr) {
+  MemoryBarrier();
+  return *ptr;
+}
+
+
+// 64-bit versions are not implemented yet.
+
+inline void NotImplementedFatalError(const char *function_name) {
+  fprintf(stderr, "64-bit %s() not implemented on this platform\n",
+          function_name);
+  abort();
+}
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  NotImplementedFatalError("NoBarrier_CompareAndSwap");
+  return 0;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+  NotImplementedFatalError("NoBarrier_AtomicExchange");
+  return 0;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  // pLinuxKernelCmpxchg already has acquire and release barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  // pLinuxKernelCmpxchg already has acquire and release barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  NotImplementedFatalError("NoBarrier_AtomicIncrement");
+  return 0;
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  NotImplementedFatalError("Barrier_AtomicIncrement");
+  return 0;
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NotImplementedFatalError("NoBarrier_Store");
+}
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NotImplementedFatalError("Acquire_Store64");
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NotImplementedFatalError("Release_Store");
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  NotImplementedFatalError("NoBarrier_Load");
+  return 0;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  NotImplementedFatalError("Atomic64 Acquire_Load");
+  return 0;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+  NotImplementedFatalError("Atomic64 Release_Load");
+  return 0;
+}
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  NotImplementedFatalError("Atomic64 Acquire_CompareAndSwap");
+  return 0;
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  NotImplementedFatalError("Atomic64 Release_CompareAndSwap");
+  return 0;
+}
+
+}  // namespace base::subtle
+}  // namespace base
+
+#endif  // BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_

BeefRT/gperftools/src/base/atomicops-internals-arm-v6plus.h

@@ -0,0 +1,411 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+// ---
+//
+// Author: Sasha Levitskiy
+// based on atomicops-internals by Sanjay Ghemawat
+//
+// This file is an internal atomic implementation, use base/atomicops.h instead.
+//
+// This code implements ARM atomics for architectures V6 and  newer.
+
+#ifndef BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_
+#define BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_
+
+#include <stdio.h>
+#include <stdlib.h>
+#include "base/basictypes.h"  // For COMPILE_ASSERT
+
+// The LDREXD and STREXD instructions in ARM all v7 variants or above.  In v6,
+// only some variants support it.  For simplicity, we only use exclusive
+// 64-bit load/store in V7 or above.
+#if defined(ARMV7)
+# define BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD
+#endif
+
+typedef int32_t Atomic32;
+
+namespace base {
+namespace subtle {
+
+typedef int64_t Atomic64;
+
+// 32-bit low-level ops
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  Atomic32 oldval, res;
+  do {
+    __asm__ __volatile__(
+    "ldrex   %1, [%3]\n"
+    "mov     %0, #0\n"
+    "teq     %1, %4\n"
+    // The following IT (if-then) instruction is needed for the subsequent
+    // conditional instruction STREXEQ when compiling in THUMB mode.
+    // In ARM mode, the compiler/assembler will not generate any code for it.
+    "it      eq\n"
+    "strexeq %0, %5, [%3]\n"
+        : "=&r" (res), "=&r" (oldval), "+Qo" (*ptr)
+        : "r" (ptr), "Ir" (old_value), "r" (new_value)
+        : "cc");
+  } while (res);
+  return oldval;
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr,
+                                         Atomic32 new_value) {
+  Atomic32 tmp, old;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrex  %1, [%2]\n"
+      "strex  %0, %3, [%2]\n"
+      "teq    %0, #0\n"
+      "bne    1b"
+      : "=&r" (tmp), "=&r" (old)
+      : "r" (ptr), "r" (new_value)
+      : "cc", "memory");
+  return old;
+}
+
+inline void MemoryBarrier() {
+#if defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6KZ__) || defined(__ARM_ARCH_6T2__)
+  uint32_t dest = 0;
+  __asm__ __volatile__("mcr p15,0,%0,c7,c10,5" :"=&r"(dest) : : "memory");
+#else
+  __asm__ __volatile__("dmb" : : : "memory");
+#endif
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  Atomic32 old_value = NoBarrier_AtomicExchange(ptr, new_value);
+  MemoryBarrier();
+  return old_value;
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  MemoryBarrier();
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32* ptr,
+                                          Atomic32 increment) {
+  Atomic32 tmp, res;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrex  %1, [%2]\n"
+      "add    %1, %1, %3\n"
+      "strex  %0, %1, [%2]\n"
+      "teq    %0, #0\n"
+      "bne    1b"
+      : "=&r" (tmp), "=&r"(res)
+      : "r" (ptr), "r"(increment)
+      : "cc", "memory");
+  return res;
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32* ptr,
+                                        Atomic32 increment) {
+  Atomic32 tmp, res;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrex  %1, [%2]\n"
+      "add    %1, %1, %3\n"
+      "dmb\n"
+      "strex  %0, %1, [%2]\n"
+      "teq    %0, #0\n"
+      "bne    1b"
+      : "=&r" (tmp), "=&r"(res)
+      : "r" (ptr), "r"(increment)
+      : "cc", "memory");
+  return res;
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  Atomic32 value = NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  MemoryBarrier();
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) {
+  MemoryBarrier();
+  *ptr = value;
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) {
+  Atomic32 value = *ptr;
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32* ptr) {
+  MemoryBarrier();
+  return *ptr;
+}
+
+// 64-bit versions are only available if LDREXD and STREXD instructions
+// are available.
+#ifdef BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD
+
+#define BASE_HAS_ATOMIC64 1
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  Atomic64 oldval, res;
+  do {
+    __asm__ __volatile__(
+    "ldrexd   %1, [%3]\n"
+    "mov      %0, #0\n"
+    "teq      %Q1, %Q4\n"
+    // The following IT (if-then) instructions are needed for the subsequent
+    // conditional instructions when compiling in THUMB mode.
+    // In ARM mode, the compiler/assembler will not generate any code for it.
+    "it       eq\n"
+    "teqeq    %R1, %R4\n"
+    "it       eq\n"
+    "strexdeq %0, %5, [%3]\n"
+        : "=&r" (res), "=&r" (oldval), "+Q" (*ptr)
+        : "r" (ptr), "Ir" (old_value), "r" (new_value)
+        : "cc");
+  } while (res);
+  return oldval;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+  int store_failed;
+  Atomic64 old;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrexd  %1, [%2]\n"
+      "strexd  %0, %3, [%2]\n"
+      "teq     %0, #0\n"
+      "bne     1b"
+      : "=&r" (store_failed), "=&r" (old)
+      : "r" (ptr), "r" (new_value)
+      : "cc", "memory");
+  return old;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  Atomic64 old_value = NoBarrier_AtomicExchange(ptr, new_value);
+  MemoryBarrier();
+  return old_value;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  MemoryBarrier();
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  int store_failed;
+  Atomic64 res;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrexd  %1, [%2]\n"
+      "adds    %Q1, %Q1, %Q3\n"
+      "adc     %R1, %R1, %R3\n"
+      "strexd  %0, %1, [%2]\n"
+      "teq     %0, #0\n"
+      "bne     1b"
+      : "=&r" (store_failed), "=&r"(res)
+      : "r" (ptr), "r"(increment)
+      : "cc", "memory");
+  return res;
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  int store_failed;
+  Atomic64 res;
+  __asm__ __volatile__(
+      "1:\n"
+      "ldrexd  %1, [%2]\n"
+      "adds    %Q1, %Q1, %Q3\n"
+      "adc     %R1, %R1, %R3\n"
+      "dmb\n"
+      "strexd  %0, %1, [%2]\n"
+      "teq     %0, #0\n"
+      "bne     1b"
+      : "=&r" (store_failed), "=&r"(res)
+      : "r" (ptr), "r"(increment)
+      : "cc", "memory");
+  return res;
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  int store_failed;
+  Atomic64 dummy;
+  __asm__ __volatile__(
+      "1:\n"
+      // Dummy load to lock cache line.
+      "ldrexd  %1, [%3]\n"
+      "strexd  %0, %2, [%3]\n"
+      "teq     %0, #0\n"
+      "bne     1b"
+      : "=&r" (store_failed), "=&r"(dummy)
+      : "r"(value), "r" (ptr)
+      : "cc", "memory");
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  Atomic64 res;
+  __asm__ __volatile__(
+  "ldrexd   %0, [%1]\n"
+  "clrex\n"
+      : "=r" (res)
+      : "r"(ptr), "Q"(*ptr));
+  return res;
+}
+
+#else // BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD
+
+inline void NotImplementedFatalError(const char *function_name) {
+  fprintf(stderr, "64-bit %s() not implemented on this platform\n",
+          function_name);
+  abort();
+}
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  NotImplementedFatalError("NoBarrier_CompareAndSwap");
+  return 0;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+  NotImplementedFatalError("NoBarrier_AtomicExchange");
+  return 0;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  NotImplementedFatalError("Acquire_AtomicExchange");
+  return 0;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  NotImplementedFatalError("Release_AtomicExchange");
+  return 0;
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  NotImplementedFatalError("NoBarrier_AtomicIncrement");
+  return 0;
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  NotImplementedFatalError("Barrier_AtomicIncrement");
+  return 0;
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NotImplementedFatalError("NoBarrier_Store");
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  NotImplementedFatalError("NoBarrier_Load");
+  return 0;
+}
+
+#endif // BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NoBarrier_Store(ptr, value);
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  MemoryBarrier();
+  NoBarrier_Store(ptr, value);
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  Atomic64 value = NoBarrier_Load(ptr);
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+  MemoryBarrier();
+  return NoBarrier_Load(ptr);
+}
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  Atomic64 value = NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  MemoryBarrier();
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+}  // namespace subtle ends
+}  // namespace base ends
+
+#endif  // BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_

BeefRT/gperftools/src/base/atomicops-internals-linuxppc.h

@@ -0,0 +1,456 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ */
+
+// Implementation of atomic operations for ppc-linux.  This file should not
+// be included directly.  Clients should instead include
+// "base/atomicops.h".
+
+#ifndef BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_
+#define BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_
+
+typedef int32_t Atomic32;
+
+#ifdef __PPC64__
+#define BASE_HAS_ATOMIC64 1
+#endif
+
+namespace base {
+namespace subtle {
+
+static inline void _sync(void) {
+  __asm__ __volatile__("sync": : : "memory");
+}
+
+static inline void _lwsync(void) {
+  // gcc defines __NO_LWSYNC__ when appropriate; see
+  //    http://gcc.gnu.org/ml/gcc-patches/2006-11/msg01238.html
+#ifdef __NO_LWSYNC__
+  __asm__ __volatile__("msync": : : "memory");
+#else
+  __asm__ __volatile__("lwsync": : : "memory");
+#endif
+}
+
+static inline void _isync(void) {
+  __asm__ __volatile__("isync": : : "memory");
+}
+
+static inline Atomic32 OSAtomicAdd32(Atomic32 amount, Atomic32 *value) {
+  Atomic32 t;
+  __asm__ __volatile__(
+"1:		lwarx   %0,0,%3\n\
+		add     %0,%2,%0\n\
+		stwcx.  %0,0,%3 \n\
+		bne-    1b"
+		: "=&r" (t), "+m" (*value)
+		: "r" (amount), "r" (value)
+                : "cc");
+  return t;
+}
+
+static inline Atomic32 OSAtomicAdd32Barrier(Atomic32 amount, Atomic32 *value) {
+  Atomic32 t;
+  _lwsync();
+  t = OSAtomicAdd32(amount, value);
+  // This is based on the code snippet in the architecture manual (Vol
+  // 2, Appendix B).  It's a little tricky: correctness depends on the
+  // fact that the code right before this (in OSAtomicAdd32) has a
+  // conditional branch with a data dependency on the update.
+  // Otherwise, we'd have to use sync.
+  _isync();
+  return t;
+}
+
+static inline bool OSAtomicCompareAndSwap32(Atomic32 old_value,
+                                            Atomic32 new_value,
+                                            Atomic32 *value) {
+  Atomic32 prev;
+  __asm__ __volatile__(
+"1:		lwarx   %0,0,%2\n\
+		cmpw    0,%0,%3\n\
+		bne-    2f\n\
+		stwcx.  %4,0,%2\n\
+		bne-    1b\n\
+2:"
+                : "=&r" (prev), "+m" (*value)
+                : "r" (value), "r" (old_value), "r" (new_value)
+                : "cc");
+  return prev == old_value;
+}
+
+static inline Atomic32 OSAtomicCompareAndSwap32Acquire(Atomic32 old_value,
+                                                       Atomic32 new_value,
+                                                       Atomic32 *value) {
+  Atomic32 t;
+  t = OSAtomicCompareAndSwap32(old_value, new_value, value);
+  // This is based on the code snippet in the architecture manual (Vol
+  // 2, Appendix B).  It's a little tricky: correctness depends on the
+  // fact that the code right before this (in
+  // OSAtomicCompareAndSwap32) has a conditional branch with a data
+  // dependency on the update.  Otherwise, we'd have to use sync.
+  _isync();
+  return t;
+}
+
+static inline Atomic32 OSAtomicCompareAndSwap32Release(Atomic32 old_value,
+                                                       Atomic32 new_value,
+                                                       Atomic32 *value) {
+  _lwsync();
+  return OSAtomicCompareAndSwap32(old_value, new_value, value);
+}
+
+typedef int64_t Atomic64;
+
+inline void MemoryBarrier() {
+  // This can't be _lwsync(); we need to order the immediately
+  // preceding stores against any load that may follow, but lwsync
+  // doesn't guarantee that.
+  _sync();
+}
+
+// 32-bit Versions.
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32 *ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  Atomic32 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap32(old_value, new_value,
+                                 const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32 *ptr,
+                                         Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap32(old_value, new_value,
+                                     const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32 *ptr,
+                                       Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap32Acquire(old_value, new_value,
+                                            const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32 *ptr,
+                                       Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap32Release(old_value, new_value,
+                                            const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32 *ptr,
+                                          Atomic32 increment) {
+  return OSAtomicAdd32(increment, const_cast<Atomic32*>(ptr));
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32 *ptr,
+                                        Atomic32 increment) {
+  return OSAtomicAdd32Barrier(increment, const_cast<Atomic32*>(ptr));
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32 *ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  Atomic32 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap32Acquire(old_value, new_value,
+                                        const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32 *ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  Atomic32 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap32Release(old_value, new_value,
+                                        const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+#ifdef __PPC64__
+
+// 64-bit Versions.
+
+static inline Atomic64 OSAtomicAdd64(Atomic64 amount, Atomic64 *value) {
+  Atomic64 t;
+  __asm__ __volatile__(
+"1:		ldarx   %0,0,%3\n\
+		add     %0,%2,%0\n\
+		stdcx.  %0,0,%3 \n\
+		bne-    1b"
+		: "=&r" (t), "+m" (*value)
+		: "r" (amount), "r" (value)
+                : "cc");
+  return t;
+}
+
+static inline Atomic64 OSAtomicAdd64Barrier(Atomic64 amount, Atomic64 *value) {
+  Atomic64 t;
+  _lwsync();
+  t = OSAtomicAdd64(amount, value);
+  // This is based on the code snippet in the architecture manual (Vol
+  // 2, Appendix B).  It's a little tricky: correctness depends on the
+  // fact that the code right before this (in OSAtomicAdd64) has a
+  // conditional branch with a data dependency on the update.
+  // Otherwise, we'd have to use sync.
+  _isync();
+  return t;
+}
+
+static inline bool OSAtomicCompareAndSwap64(Atomic64 old_value,
+                                            Atomic64 new_value,
+                                            Atomic64 *value) {
+  Atomic64 prev;
+  __asm__ __volatile__(
+"1:		ldarx   %0,0,%2\n\
+		cmpd    0,%0,%3\n\
+		bne-    2f\n\
+		stdcx.  %4,0,%2\n\
+		bne-    1b\n\
+2:"
+                : "=&r" (prev), "+m" (*value)
+                : "r" (value), "r" (old_value), "r" (new_value)
+                : "cc");
+  return prev == old_value;
+}
+
+static inline Atomic64 OSAtomicCompareAndSwap64Acquire(Atomic64 old_value,
+                                                       Atomic64 new_value,
+                                                       Atomic64 *value) {
+  Atomic64 t;
+  t = OSAtomicCompareAndSwap64(old_value, new_value, value);
+  // This is based on the code snippet in the architecture manual (Vol
+  // 2, Appendix B).  It's a little tricky: correctness depends on the
+  // fact that the code right before this (in
+  // OSAtomicCompareAndSwap64) has a conditional branch with a data
+  // dependency on the update.  Otherwise, we'd have to use sync.
+  _isync();
+  return t;
+}
+
+static inline Atomic64 OSAtomicCompareAndSwap64Release(Atomic64 old_value,
+                                                       Atomic64 new_value,
+                                                       Atomic64 *value) {
+  _lwsync();
+  return OSAtomicCompareAndSwap64(old_value, new_value, value);
+}
+
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64 *ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  Atomic64 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap64(old_value, new_value,
+                                 const_cast<Atomic64*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64 *ptr,
+                                         Atomic64 new_value) {
+  Atomic64 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap64(old_value, new_value,
+                                     const_cast<Atomic64*>(ptr)));
+  return old_value;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64 *ptr,
+                                       Atomic64 new_value) {
+  Atomic64 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap64Acquire(old_value, new_value,
+                                            const_cast<Atomic64*>(ptr)));
+  return old_value;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64 *ptr,
+                                       Atomic64 new_value) {
+  Atomic64 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap64Release(old_value, new_value,
+                                            const_cast<Atomic64*>(ptr)));
+  return old_value;
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64 *ptr,
+                                          Atomic64 increment) {
+  return OSAtomicAdd64(increment, const_cast<Atomic64*>(ptr));
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64 *ptr,
+                                        Atomic64 increment) {
+  return OSAtomicAdd64Barrier(increment, const_cast<Atomic64*>(ptr));
+}
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64 *ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  Atomic64 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap64Acquire(old_value, new_value,
+                                        const_cast<Atomic64*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64 *ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  Atomic64 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap64Release(old_value, new_value,
+                                        const_cast<Atomic64*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+#endif
+
+inline void NoBarrier_Store(volatile Atomic32 *ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic32 *ptr, Atomic32 value) {
+  *ptr = value;
+  // This can't be _lwsync(); we need to order the immediately
+  // preceding stores against any load that may follow, but lwsync
+  // doesn't guarantee that.
+  _sync();
+}
+
+inline void Release_Store(volatile Atomic32 *ptr, Atomic32 value) {
+  _lwsync();
+  *ptr = value;
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32 *ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32 *ptr) {
+  Atomic32 value = *ptr;
+  _lwsync();
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32 *ptr) {
+  // This can't be _lwsync(); we need to order the immediately
+  // preceding stores against any load that may follow, but lwsync
+  // doesn't guarantee that.
+  _sync();
+  return *ptr;
+}
+
+#ifdef __PPC64__
+
+// 64-bit Versions.
+
+inline void NoBarrier_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  *ptr = value;
+  // This can't be _lwsync(); we need to order the immediately
+  // preceding stores against any load that may follow, but lwsync
+  // doesn't guarantee that.
+  _sync();
+}
+
+inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  _lwsync();
+  *ptr = value;
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64 *ptr) {
+  return *ptr;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) {
+  Atomic64 value = *ptr;
+  _lwsync();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64 *ptr) {
+  // This can't be _lwsync(); we need to order the immediately
+  // preceding stores against any load that may follow, but lwsync
+  // doesn't guarantee that.
+  _sync();
+  return *ptr;
+}
+
+#endif
+
+}   // namespace base::subtle
+}   // namespace base
+
+#endif  // BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_

BeefRT/gperftools/src/base/atomicops-internals-macosx.h

@@ -0,0 +1,390 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+// Implementation of atomic operations for Mac OS X.  This file should not
+// be included directly.  Clients should instead include
+// "base/atomicops.h".
+
+#ifndef BASE_ATOMICOPS_INTERNALS_MACOSX_H_
+#define BASE_ATOMICOPS_INTERNALS_MACOSX_H_
+
+typedef int32_t Atomic32;
+
+// MacOS uses long for intptr_t, AtomicWord and Atomic32 are always different
+// on the Mac, even when they are the same size.  Similarly, on __ppc64__,
+// AtomicWord and Atomic64 are always different.  Thus, we need explicit
+// casting.
+#ifdef __LP64__
+#define AtomicWordCastType base::subtle::Atomic64
+#else
+#define AtomicWordCastType Atomic32
+#endif
+
+#if defined(__LP64__) || defined(__i386__)
+#define BASE_HAS_ATOMIC64 1  // Use only in tests and base/atomic*
+#endif
+
+#include <libkern/OSAtomic.h>
+
+namespace base {
+namespace subtle {
+
+#if !defined(__LP64__) && defined(__ppc__)
+
+// The Mac 64-bit OSAtomic implementations are not available for 32-bit PowerPC,
+// while the underlying assembly instructions are available only some
+// implementations of PowerPC.
+
+// The following inline functions will fail with the error message at compile
+// time ONLY IF they are called.  So it is safe to use this header if user
+// code only calls AtomicWord and Atomic32 operations.
+//
+// NOTE(vchen): Implementation notes to implement the atomic ops below may
+// be found in "PowerPC Virtual Environment Architecture, Book II,
+// Version 2.02", January 28, 2005, Appendix B, page 46.  Unfortunately,
+// extra care must be taken to ensure data are properly 8-byte aligned, and
+// that data are returned correctly according to Mac OS X ABI specs.
+
+inline int64_t OSAtomicCompareAndSwap64(
+    int64_t oldValue, int64_t newValue, int64_t *theValue) {
+  __asm__ __volatile__(
+      "_OSAtomicCompareAndSwap64_not_supported_for_32_bit_ppc\n\t");
+  return 0;
+}
+
+inline int64_t OSAtomicAdd64(int64_t theAmount, int64_t *theValue) {
+  __asm__ __volatile__(
+      "_OSAtomicAdd64_not_supported_for_32_bit_ppc\n\t");
+  return 0;
+}
+
+inline int64_t OSAtomicCompareAndSwap64Barrier(
+    int64_t oldValue, int64_t newValue, int64_t *theValue) {
+  int64_t prev = OSAtomicCompareAndSwap64(oldValue, newValue, theValue);
+  OSMemoryBarrier();
+  return prev;
+}
+
+inline int64_t OSAtomicAdd64Barrier(
+    int64_t theAmount, int64_t *theValue) {
+  int64_t new_val = OSAtomicAdd64(theAmount, theValue);
+  OSMemoryBarrier();
+  return new_val;
+}
+#endif
+
+typedef int64_t Atomic64;
+
+inline void MemoryBarrier() {
+  OSMemoryBarrier();
+}
+
+// 32-bit Versions.
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32 *ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  Atomic32 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap32(old_value, new_value,
+                                 const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32 *ptr,
+                                         Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap32(old_value, new_value,
+                                     const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32 *ptr,
+                                       Atomic32 new_value) {
+  Atomic32 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap32Barrier(old_value, new_value,
+                                            const_cast<Atomic32*>(ptr)));
+  return old_value;
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32 *ptr,
+                                       Atomic32 new_value) {
+  return Acquire_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32 *ptr,
+                                          Atomic32 increment) {
+  return OSAtomicAdd32(increment, const_cast<Atomic32*>(ptr));
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32 *ptr,
+                                          Atomic32 increment) {
+  return OSAtomicAdd32Barrier(increment, const_cast<Atomic32*>(ptr));
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32 *ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  Atomic32 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap32Barrier(old_value, new_value,
+                                        const_cast<Atomic32*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32 *ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return Acquire_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic32 *ptr, Atomic32 value) {
+  *ptr = value;
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic32 *ptr, Atomic32 value) {
+  MemoryBarrier();
+  *ptr = value;
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32 *ptr) {
+  Atomic32 value = *ptr;
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32 *ptr) {
+  MemoryBarrier();
+  return *ptr;
+}
+
+// 64-bit version
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64 *ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  Atomic64 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap64(old_value, new_value,
+                                 const_cast<Atomic64*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64 *ptr,
+                                         Atomic64 new_value) {
+  Atomic64 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap64(old_value, new_value,
+                                     const_cast<Atomic64*>(ptr)));
+  return old_value;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64 *ptr,
+                                       Atomic64 new_value) {
+  Atomic64 old_value;
+  do {
+    old_value = *ptr;
+  } while (!OSAtomicCompareAndSwap64Barrier(old_value, new_value,
+                                            const_cast<Atomic64*>(ptr)));
+  return old_value;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64 *ptr,
+                                       Atomic64 new_value) {
+  return Acquire_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64 *ptr,
+                                          Atomic64 increment) {
+  return OSAtomicAdd64(increment, const_cast<Atomic64*>(ptr));
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64 *ptr,
+                                        Atomic64 increment) {
+  return OSAtomicAdd64Barrier(increment, const_cast<Atomic64*>(ptr));
+}
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64 *ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  Atomic64 prev_value;
+  do {
+    if (OSAtomicCompareAndSwap64Barrier(old_value, new_value,
+                                        const_cast<Atomic64*>(ptr))) {
+      return old_value;
+    }
+    prev_value = *ptr;
+  } while (prev_value == old_value);
+  return prev_value;
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64 *ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  // The lib kern interface does not distinguish between
+  // Acquire and Release memory barriers; they are equivalent.
+  return Acquire_CompareAndSwap(ptr, old_value, new_value);
+}
+
+#ifdef __LP64__
+
+// 64-bit implementation on 64-bit platform
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  *ptr = value;
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  MemoryBarrier();
+  *ptr = value;
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  return *ptr;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) {
+  Atomic64 value = *ptr;
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64 *ptr) {
+  MemoryBarrier();
+  return *ptr;
+}
+
+#else
+
+// 64-bit implementation on 32-bit platform
+
+#if defined(__ppc__)
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+   __asm__ __volatile__(
+       "_NoBarrier_Store_not_supported_for_32_bit_ppc\n\t");
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+   __asm__ __volatile__(
+       "_NoBarrier_Load_not_supported_for_32_bit_ppc\n\t");
+   return 0;
+}
+
+#elif defined(__i386__)
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  __asm__ __volatile__("movq %1, %%mm0\n\t"    // Use mmx reg for 64-bit atomic
+                       "movq %%mm0, %0\n\t"  // moves (ptr could be read-only)
+                       "emms\n\t"              // Reset FP registers
+                       : "=m" (*ptr)
+                       : "m" (value)
+                       : // mark the FP stack and mmx registers as clobbered
+                         "st", "st(1)", "st(2)", "st(3)", "st(4)",
+                         "st(5)", "st(6)", "st(7)", "mm0", "mm1",
+                         "mm2", "mm3", "mm4", "mm5", "mm6", "mm7");
+
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  Atomic64 value;
+  __asm__ __volatile__("movq %1, %%mm0\n\t"  // Use mmx reg for 64-bit atomic
+                       "movq %%mm0, %0\n\t"  // moves (ptr could be read-only)
+                       "emms\n\t"            // Reset FP registers
+                       : "=m" (value)
+                       : "m" (*ptr)
+                       : // mark the FP stack and mmx registers as clobbered
+                         "st", "st(1)", "st(2)", "st(3)", "st(4)",
+                         "st(5)", "st(6)", "st(7)", "mm0", "mm1",
+                         "mm2", "mm3", "mm4", "mm5", "mm6", "mm7");
+
+  return value;
+}
+#endif
+
+/*
+inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  NoBarrier_Store(ptr, value);
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) {
+  MemoryBarrier();
+  NoBarrier_Store(ptr, value);
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) {
+  Atomic64 value = NoBarrier_Load(ptr);
+  MemoryBarrier();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64 *ptr) {
+  MemoryBarrier();
+  return NoBarrier_Load(ptr);
+}*/
+    
+#endif  // __LP64__
+
+}   // namespace base::subtle
+}   // namespace base
+
+#endif  // BASE_ATOMICOPS_INTERNALS_MACOSX_H_

BeefRT/gperftools/src/base/atomicops-internals-windows.h

@@ -0,0 +1,520 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ */
+
+// Implementation of atomic operations using Windows API
+// functions.  This file should not be included directly.  Clients
+// should instead include "base/atomicops.h".
+
+#ifndef BASE_ATOMICOPS_INTERNALS_WINDOWS_H_
+#define BASE_ATOMICOPS_INTERNALS_WINDOWS_H_
+
+#include <windows.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include "base/basictypes.h"  // For COMPILE_ASSERT
+
+typedef int32 Atomic32;
+
+#if defined(_WIN64)
+#define BASE_HAS_ATOMIC64 1  // Use only in tests and base/atomic*
+#endif
+
+namespace base {
+namespace subtle {
+
+typedef int64 Atomic64;
+
+// 32-bit low-level operations on any platform
+
+extern "C" {
+// We use windows intrinsics when we can (they seem to be supported
+// well on MSVC 8.0 and above).  Unfortunately, in some
+// environments, <windows.h> and <intrin.h> have conflicting
+// declarations of some other intrinsics, breaking compilation:
+//   http://connect.microsoft.com/VisualStudio/feedback/details/262047
+// Therefore, we simply declare the relevant intrinsics ourself.
+
+// MinGW has a bug in the header files where it doesn't indicate the
+// first argument is volatile -- they're not up to date.  See
+//   http://readlist.com/lists/lists.sourceforge.net/mingw-users/0/3861.html
+// We have to const_cast away the volatile to avoid compiler warnings.
+// TODO(csilvers): remove this once MinGW has updated MinGW/include/winbase.h
+#if defined(__MINGW32__)
+inline LONG FastInterlockedCompareExchange(volatile LONG* ptr,
+                                           LONG newval, LONG oldval) {
+  return ::InterlockedCompareExchange(const_cast<LONG*>(ptr), newval, oldval);
+}
+inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) {
+  return ::InterlockedExchange(const_cast<LONG*>(ptr), newval);
+}
+inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) {
+  return ::InterlockedExchangeAdd(const_cast<LONG*>(ptr), increment);
+}
+
+#elif _MSC_VER >= 1400   // intrinsics didn't work so well before MSVC 8.0
+// Unfortunately, in some environments, <windows.h> and <intrin.h>
+// have conflicting declarations of some intrinsics, breaking
+// compilation.  So we declare the intrinsics we need ourselves.  See
+//   http://connect.microsoft.com/VisualStudio/feedback/details/262047
+LONG _InterlockedCompareExchange(volatile LONG* ptr, LONG newval, LONG oldval);
+#pragma intrinsic(_InterlockedCompareExchange)
+inline LONG FastInterlockedCompareExchange(volatile LONG* ptr,
+                                           LONG newval, LONG oldval) {
+  return _InterlockedCompareExchange(ptr, newval, oldval);
+}
+
+LONG _InterlockedExchange(volatile LONG* ptr, LONG newval);
+#pragma intrinsic(_InterlockedExchange)
+inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) {
+  return _InterlockedExchange(ptr, newval);
+}
+
+LONG _InterlockedExchangeAdd(volatile LONG* ptr, LONG increment);
+#pragma intrinsic(_InterlockedExchangeAdd)
+inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) {
+  return _InterlockedExchangeAdd(ptr, increment);
+}
+
+#else
+inline LONG FastInterlockedCompareExchange(volatile LONG* ptr,
+                                           LONG newval, LONG oldval) {
+  return ::InterlockedCompareExchange(ptr, newval, oldval);
+}
+inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) {
+  return ::InterlockedExchange(ptr, newval);
+}
+inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) {
+  return ::InterlockedExchangeAdd(ptr, increment);
+}
+
+#endif  // ifdef __MINGW32__
+}  // extern "C"
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  LONG result = FastInterlockedCompareExchange(
+      reinterpret_cast<volatile LONG*>(ptr),
+      static_cast<LONG>(new_value),
+      static_cast<LONG>(old_value));
+  return static_cast<Atomic32>(result);
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr,
+                                         Atomic32 new_value) {
+  LONG result = FastInterlockedExchange(
+      reinterpret_cast<volatile LONG*>(ptr),
+      static_cast<LONG>(new_value));
+  return static_cast<Atomic32>(result);
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  // FastInterlockedExchange has both acquire and release memory barriers.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  // FastInterlockedExchange has both acquire and release memory barriers.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32* ptr,
+                                        Atomic32 increment) {
+  return FastInterlockedExchangeAdd(
+      reinterpret_cast<volatile LONG*>(ptr),
+      static_cast<LONG>(increment)) + increment;
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32* ptr,
+                                          Atomic32 increment) {
+  return Barrier_AtomicIncrement(ptr, increment);
+}
+
+}  // namespace base::subtle
+}  // namespace base
+
+
+// In msvc8/vs2005, winnt.h already contains a definition for
+// MemoryBarrier in the global namespace.  Add it there for earlier
+// versions and forward to it from within the namespace.
+#if !(defined(_MSC_VER) && _MSC_VER >= 1400)
+inline void GMemoryBarrier() {
+  Atomic32 value = 0;
+  base::subtle::NoBarrier_AtomicExchange(&value, 0);
+                        // actually acts as a barrier in thisd implementation
+}
+#endif
+
+namespace base {
+namespace subtle {
+
+inline void GMemoryBarrier() {
+  MemoryBarrier();
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  Acquire_AtomicExchange(ptr, value);
+}
+
+inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value; // works w/o barrier for current Intel chips as of June 2005
+  // See comments in Atomic64 version of Release_Store() below.
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) {
+  Atomic32 value = *ptr;
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32* ptr) {
+  GMemoryBarrier();
+  return *ptr;
+}
+
+// 64-bit operations
+
+#if defined(_WIN64) || defined(__MINGW64__)
+
+// 64-bit low-level operations on 64-bit platform.
+
+COMPILE_ASSERT(sizeof(Atomic64) == sizeof(PVOID), atomic_word_is_atomic);
+
+// These are the intrinsics needed for 64-bit operations.  Similar to the
+// 32-bit case above.
+
+extern "C" {
+#if defined(__MINGW64__)
+inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr,
+                                                   PVOID newval, PVOID oldval) {
+  return ::InterlockedCompareExchangePointer(const_cast<PVOID*>(ptr),
+                                             newval, oldval);
+}
+inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) {
+  return ::InterlockedExchangePointer(const_cast<PVOID*>(ptr), newval);
+}
+inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr,
+                                             LONGLONG increment) {
+  return ::InterlockedExchangeAdd64(const_cast<LONGLONG*>(ptr), increment);
+}
+
+#elif _MSC_VER >= 1400   // intrinsics didn't work so well before MSVC 8.0
+// Like above, we need to declare the intrinsics ourselves.
+PVOID _InterlockedCompareExchangePointer(volatile PVOID* ptr,
+                                         PVOID newval, PVOID oldval);
+#pragma intrinsic(_InterlockedCompareExchangePointer)
+inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr,
+                                                   PVOID newval, PVOID oldval) {
+  return _InterlockedCompareExchangePointer(const_cast<PVOID*>(ptr),
+                                            newval, oldval);
+}
+
+PVOID _InterlockedExchangePointer(volatile PVOID* ptr, PVOID newval);
+#pragma intrinsic(_InterlockedExchangePointer)
+inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) {
+  return _InterlockedExchangePointer(const_cast<PVOID*>(ptr), newval);
+}
+
+LONGLONG _InterlockedExchangeAdd64(volatile LONGLONG* ptr, LONGLONG increment);
+#pragma intrinsic(_InterlockedExchangeAdd64)
+inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr,
+                                             LONGLONG increment) {
+  return _InterlockedExchangeAdd64(const_cast<LONGLONG*>(ptr), increment);
+}
+
+#else
+inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr,
+                                                   PVOID newval, PVOID oldval) {
+  return ::InterlockedCompareExchangePointer(ptr, newval, oldval);
+}
+inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) {
+  return ::InterlockedExchangePointer(ptr, newval);
+}
+inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr,
+                                         LONGLONG increment) {
+  return ::InterlockedExchangeAdd64(ptr, increment);
+}
+
+#endif  // ifdef __MINGW64__
+}  // extern "C"
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  PVOID result = FastInterlockedCompareExchangePointer(
+    reinterpret_cast<volatile PVOID*>(ptr),
+    reinterpret_cast<PVOID>(new_value), reinterpret_cast<PVOID>(old_value));
+  return reinterpret_cast<Atomic64>(result);
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+  PVOID result = FastInterlockedExchangePointer(
+    reinterpret_cast<volatile PVOID*>(ptr),
+    reinterpret_cast<PVOID>(new_value));
+  return reinterpret_cast<Atomic64>(result);
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  return FastInterlockedExchangeAdd64(
+      reinterpret_cast<volatile LONGLONG*>(ptr),
+      static_cast<LONGLONG>(increment)) + increment;
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  return Barrier_AtomicIncrement(ptr, increment);
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NoBarrier_AtomicExchange(ptr, value);
+              // acts as a barrier in this implementation
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  *ptr = value; // works w/o barrier for current Intel chips as of June 2005
+
+  // When new chips come out, check:
+  //  IA-32 Intel Architecture Software Developer's Manual, Volume 3:
+  //  System Programming Guide, Chatper 7: Multiple-processor management,
+  //  Section 7.2, Memory Ordering.
+  // Last seen at:
+  //   http://developer.intel.com/design/pentium4/manuals/index_new.htm
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  return *ptr;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  Atomic64 value = *ptr;
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+	GMemoryBarrier();
+  return *ptr;
+}
+
+#else  // defined(_WIN64) || defined(__MINGW64__)
+
+// 64-bit low-level operations on 32-bit platform
+
+// TODO(vchen): The GNU assembly below must be converted to MSVC inline
+// assembly.  Then the file should be renamed to ...-x86-msvc.h, probably.
+
+inline void NotImplementedFatalError(const char *function_name) {
+  fprintf(stderr, "64-bit %s() not implemented on this platform\n",
+          function_name);
+  abort();
+}
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+#if 0 // Not implemented
+  Atomic64 prev;
+  __asm__ __volatile__("movl (%3), %%ebx\n\t"    // Move 64-bit new_value into
+                       "movl 4(%3), %%ecx\n\t"   // ecx:ebx
+                       "lock; cmpxchg8b %1\n\t"  // If edx:eax (old_value) same
+                       : "=A" (prev)             // as contents of ptr:
+                       : "m" (*ptr),             //   ecx:ebx => ptr
+                         "0" (old_value),        // else:
+                         "r" (&new_value)        //   old *ptr => edx:eax
+                       : "memory", "%ebx", "%ecx");
+  return prev;
+#else
+  NotImplementedFatalError("NoBarrier_CompareAndSwap");
+  return 0;
+#endif
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+#if 0 // Not implemented
+  __asm__ __volatile__(
+                       "movl (%2), %%ebx\n\t"    // Move 64-bit new_value into
+                       "movl 4(%2), %%ecx\n\t"   // ecx:ebx
+                       "0:\n\t"
+                       "movl %1, %%eax\n\t"      // Read contents of ptr into
+                       "movl 4%1, %%edx\n\t"     // edx:eax
+                       "lock; cmpxchg8b %1\n\t"  // Attempt cmpxchg; if *ptr
+                       "jnz 0b\n\t"              // is no longer edx:eax, loop
+                       : "=A" (new_value)
+                       : "m" (*ptr),
+                         "r" (&new_value)
+                       : "memory", "%ebx", "%ecx");
+  return new_value;  // Now it's the previous value.
+#else
+  NotImplementedFatalError("NoBarrier_AtomicExchange");
+  return 0;
+#endif
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+#if 0 // Not implemented
+  Atomic64 temp = increment;
+  __asm__ __volatile__(
+                       "0:\n\t"
+                       "movl (%3), %%ebx\n\t"    // Move 64-bit increment into
+                       "movl 4(%3), %%ecx\n\t"   // ecx:ebx
+                       "movl (%2), %%eax\n\t"    // Read contents of ptr into
+                       "movl 4(%2), %%edx\n\t"   // edx:eax
+                       "add %%eax, %%ebx\n\t"    // sum => ecx:ebx
+                       "adc %%edx, %%ecx\n\t"    // edx:eax still has old *ptr
+                       "lock; cmpxchg8b (%2)\n\t"// Attempt cmpxchg; if *ptr
+                       "jnz 0b\n\t"              // is no longer edx:eax, loop
+                       : "=A"(temp), "+m"(*ptr)
+                       : "D" (ptr), "S" (&increment)
+                       : "memory", "%ebx", "%ecx");
+  // temp now contains the previous value of *ptr
+  return temp + increment;
+#else
+  NotImplementedFatalError("NoBarrier_AtomicIncrement");
+  return 0;
+#endif
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+#if 0 // Not implemented
+  Atomic64 new_val = NoBarrier_AtomicIncrement(ptr, increment);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return new_val;
+#else
+  NotImplementedFatalError("Barrier_AtomicIncrement");
+  return 0;
+#endif
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptrValue, Atomic64 value)
+{
+ 	__asm {
+    	movq mm0, value;  // Use mmx reg for 64-bit atomic moves
+    	mov eax, ptrValue;
+    	movq [eax], mm0;
+    	emms;            // Empty mmx state to enable FP registers
+  	}
+}
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NoBarrier_AtomicExchange(ptr, value);
+              // acts as a barrier in this implementation
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NoBarrier_Store(ptr, value);
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptrValue)
+{
+  	Atomic64 value;
+  	__asm {
+    	mov eax, ptrValue;
+    	movq mm0, [eax]; // Use mmx reg for 64-bit atomic moves
+    	movq value, mm0;
+    	emms; // Empty mmx state to enable FP registers
+  }
+  return value;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  Atomic64 value = NoBarrier_Load(ptr);
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+	GMemoryBarrier();
+  return NoBarrier_Load(ptr);
+}
+
+#endif  // defined(_WIN64) || defined(__MINGW64__)
+
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  // FastInterlockedExchange has both acquire and release memory barriers.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  // FastInterlockedExchange has both acquire and release memory barriers.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+}  // namespace base::subtle
+}  // namespace base
+
+#endif  // BASE_ATOMICOPS_INTERNALS_WINDOWS_H_

BeefRT/gperftools/src/base/atomicops-internals-x86.cc

@@ -0,0 +1,125 @@
+/* Copyright (c) 2007, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * This module gets enough CPU information to optimize the
+ * atomicops module on x86.
+ */
+
+#include "base/atomicops.h"
+#include "base/basictypes.h"
+#include "base/googleinit.h"
+#include "base/logging.h"
+#include <string.h>
+
+// This file only makes sense with atomicops-internals-x86.h -- it
+// depends on structs that are defined in that file.  If atomicops.h
+// doesn't sub-include that file, then we aren't needed, and shouldn't
+// try to do anything.
+#ifdef BASE_ATOMICOPS_INTERNALS_X86_H_
+
+// Inline cpuid instruction.  In PIC compilations, %ebx contains the address
+// of the global offset table.  To avoid breaking such executables, this code
+// must preserve that register's value across cpuid instructions.
+#if defined(__i386__)
+#define cpuid(a, b, c, d, inp) \
+  asm ("mov %%ebx, %%edi\n"    \
+       "cpuid\n"               \
+       "xchg %%edi, %%ebx\n"   \
+       : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp))
+#elif defined (__x86_64__)
+#define cpuid(a, b, c, d, inp) \
+  asm ("mov %%rbx, %%rdi\n"    \
+       "cpuid\n"               \
+       "xchg %%rdi, %%rbx\n"   \
+       : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp))
+#endif
+
+#if defined(cpuid)        // initialize the struct only on x86
+
+// Set the flags so that code will run correctly and conservatively
+// until InitGoogle() is called.
+struct AtomicOps_x86CPUFeatureStruct AtomicOps_Internalx86CPUFeatures = {
+  false,          // bug can't exist before process spawns multiple threads
+  false,          // no SSE2
+  false,          // no cmpxchg16b
+};
+
+// Initialize the AtomicOps_Internalx86CPUFeatures struct.
+static void AtomicOps_Internalx86CPUFeaturesInit() {
+  uint32 eax;
+  uint32 ebx;
+  uint32 ecx;
+  uint32 edx;
+
+  // Get vendor string (issue CPUID with eax = 0)
+  cpuid(eax, ebx, ecx, edx, 0);
+  char vendor[13];
+  memcpy(vendor, &ebx, 4);
+  memcpy(vendor + 4, &edx, 4);
+  memcpy(vendor + 8, &ecx, 4);
+  vendor[12] = 0;
+
+  // get feature flags in ecx/edx, and family/model in eax
+  cpuid(eax, ebx, ecx, edx, 1);
+
+  int family = (eax >> 8) & 0xf;        // family and model fields
+  int model = (eax >> 4) & 0xf;
+  if (family == 0xf) {                  // use extended family and model fields
+    family += (eax >> 20) & 0xff;
+    model += ((eax >> 16) & 0xf) << 4;
+  }
+
+  // Opteron Rev E has a bug in which on very rare occasions a locked
+  // instruction doesn't act as a read-acquire barrier if followed by a
+  // non-locked read-modify-write instruction.  Rev F has this bug in 
+  // pre-release versions, but not in versions released to customers,
+  // so we test only for Rev E, which is family 15, model 32..63 inclusive.
+  if (strcmp(vendor, "AuthenticAMD") == 0 &&       // AMD
+      family == 15 &&
+      32 <= model && model <= 63) {
+    AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug = true;
+  } else {
+    AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug = false;
+  }
+
+  // edx bit 26 is SSE2 which we use to tell use whether we can use mfence
+  AtomicOps_Internalx86CPUFeatures.has_sse2 = ((edx >> 26) & 1);
+
+  // ecx bit 13 indicates whether the cmpxchg16b instruction is supported
+  AtomicOps_Internalx86CPUFeatures.has_cmpxchg16b = ((ecx >> 13) & 1);
+}
+
+REGISTER_MODULE_INITIALIZER(atomicops_x86, {
+  AtomicOps_Internalx86CPUFeaturesInit();
+});
+
+#endif
+
+#endif  /* ifdef BASE_ATOMICOPS_INTERNALS_X86_H_ */

BeefRT/gperftools/src/base/atomicops-internals-x86.h

@@ -0,0 +1,471 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ */
+
+// Implementation of atomic operations for x86.  This file should not
+// be included directly.  Clients should instead include
+// "base/atomicops.h".
+
+#ifndef BASE_ATOMICOPS_INTERNALS_X86_H_
+#define BASE_ATOMICOPS_INTERNALS_X86_H_
+
+typedef int32_t Atomic32;
+#define BASE_HAS_ATOMIC64 1  // Use only in tests and base/atomic*
+
+
+// NOTE(vchen): x86 does not need to define AtomicWordCastType, because it
+// already matches Atomic32 or Atomic64, depending on the platform.
+
+
+// This struct is not part of the public API of this module; clients may not
+// use it.
+// Features of this x86.  Values may not be correct before main() is run,
+// but are set conservatively.
+struct AtomicOps_x86CPUFeatureStruct {
+  bool has_amd_lock_mb_bug; // Processor has AMD memory-barrier bug; do lfence
+                            // after acquire compare-and-swap.
+  bool has_sse2;            // Processor has SSE2.
+  bool has_cmpxchg16b;      // Processor supports cmpxchg16b instruction.
+};
+extern struct AtomicOps_x86CPUFeatureStruct AtomicOps_Internalx86CPUFeatures;
+
+
+#define ATOMICOPS_COMPILER_BARRIER() __asm__ __volatile__("" : : : "memory")
+
+
+namespace base {
+namespace subtle {
+
+typedef int64_t Atomic64;
+
+// 32-bit low-level operations on any platform.
+
+inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr,
+                                         Atomic32 old_value,
+                                         Atomic32 new_value) {
+  Atomic32 prev;
+  __asm__ __volatile__("lock; cmpxchgl %1,%2"
+                       : "=a" (prev)
+                       : "q" (new_value), "m" (*ptr), "0" (old_value)
+                       : "memory");
+  return prev;
+}
+
+inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr,
+                                         Atomic32 new_value) {
+  __asm__ __volatile__("xchgl %1,%0"  // The lock prefix is implicit for xchg.
+                       : "=r" (new_value)
+                       : "m" (*ptr), "0" (new_value)
+                       : "memory");
+  return new_value;  // Now it's the previous value.
+}
+
+inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  Atomic32 old_val = NoBarrier_AtomicExchange(ptr, new_value);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return old_val;
+}
+
+inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr,
+                                       Atomic32 new_value) {
+  // xchgl already has release memory barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32* ptr,
+                                          Atomic32 increment) {
+  Atomic32 temp = increment;
+  __asm__ __volatile__("lock; xaddl %0,%1"
+                       : "+r" (temp), "+m" (*ptr)
+                       : : "memory");
+  // temp now holds the old value of *ptr
+  return temp + increment;
+}
+
+inline Atomic32 Barrier_AtomicIncrement(volatile Atomic32* ptr,
+                                        Atomic32 increment) {
+  Atomic32 temp = increment;
+  __asm__ __volatile__("lock; xaddl %0,%1"
+                       : "+r" (temp), "+m" (*ptr)
+                       : : "memory");
+  // temp now holds the old value of *ptr
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return temp + increment;
+}
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  Atomic32 x = NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return x;
+}
+
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+}
+
+#if defined(__x86_64__)
+
+// 64-bit implementations of memory barrier can be simpler, because it
+// "mfence" is guaranteed to exist.
+inline void GMemoryBarrier() {
+  __asm__ __volatile__("mfence" : : : "memory");
+}
+
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  *ptr = value;
+  GMemoryBarrier();
+}
+
+#else
+
+inline void MemoryBarrier() {
+  if (AtomicOps_Internalx86CPUFeatures.has_sse2) {
+    __asm__ __volatile__("mfence" : : : "memory");
+  } else { // mfence is faster but not present on PIII
+    Atomic32 x = 0;
+    Acquire_AtomicExchange(&x, 0);
+  }
+}
+
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  if (AtomicOps_Internalx86CPUFeatures.has_sse2) {
+    *ptr = value;
+    __asm__ __volatile__("mfence" : : : "memory");
+  } else {
+    Acquire_AtomicExchange(ptr, value);
+  }
+}
+#endif
+
+inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) {
+  ATOMICOPS_COMPILER_BARRIER();
+  *ptr = value; // An x86 store acts as a release barrier.
+  // See comments in Atomic64 version of Release_Store(), below.
+}
+
+inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) {
+  return *ptr;
+}
+
+inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) {
+  Atomic32 value = *ptr; // An x86 load acts as a acquire barrier.
+  // See comments in Atomic64 version of Release_Store(), below.
+  ATOMICOPS_COMPILER_BARRIER();
+  return value;
+}
+
+inline Atomic32 Release_Load(volatile const Atomic32* ptr) {
+  GMemoryBarrier();
+  return *ptr;
+}
+
+#if defined(__x86_64__)
+
+// 64-bit low-level operations on 64-bit platform.
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_value,
+                                         Atomic64 new_value) {
+  Atomic64 prev;
+  __asm__ __volatile__("lock; cmpxchgq %1,%2"
+                       : "=a" (prev)
+                       : "q" (new_value), "m" (*ptr), "0" (old_value)
+                       : "memory");
+  return prev;
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_value) {
+  __asm__ __volatile__("xchgq %1,%0"  // The lock prefix is implicit for xchg.
+                       : "=r" (new_value)
+                       : "m" (*ptr), "0" (new_value)
+                       : "memory");
+  return new_value;  // Now it's the previous value.
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  Atomic64 old_val = NoBarrier_AtomicExchange(ptr, new_value);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return old_val;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_value) {
+  // xchgq already has release memory barrier semantics.
+  return NoBarrier_AtomicExchange(ptr, new_value);
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  Atomic64 temp = increment;
+  __asm__ __volatile__("lock; xaddq %0,%1"
+                       : "+r" (temp), "+m" (*ptr)
+                       : : "memory");
+  // temp now contains the previous value of *ptr
+  return temp + increment;
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  Atomic64 temp = increment;
+  __asm__ __volatile__("lock; xaddq %0,%1"
+                       : "+r" (temp), "+m" (*ptr)
+                       : : "memory");
+  // temp now contains the previous value of *ptr
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return temp + increment;
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  *ptr = value;
+}
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  *ptr = value;
+  GMemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  ATOMICOPS_COMPILER_BARRIER();
+
+  *ptr = value; // An x86 store acts as a release barrier
+                // for current AMD/Intel chips as of Jan 2008.
+                // See also Acquire_Load(), below.
+
+  // When new chips come out, check:
+  //  IA-32 Intel Architecture Software Developer's Manual, Volume 3:
+  //  System Programming Guide, Chatper 7: Multiple-processor management,
+  //  Section 7.2, Memory Ordering.
+  // Last seen at:
+  //   http://developer.intel.com/design/pentium4/manuals/index_new.htm
+  //
+  // x86 stores/loads fail to act as barriers for a few instructions (clflush
+  // maskmovdqu maskmovq movntdq movnti movntpd movntps movntq) but these are
+  // not generated by the compiler, and are rare.  Users of these instructions
+  // need to know about cache behaviour in any case since all of these involve
+  // either flushing cache lines or non-temporal cache hints.
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  return *ptr;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  Atomic64 value = *ptr; // An x86 load acts as a acquire barrier,
+                         // for current AMD/Intel chips as of Jan 2008.
+                         // See also Release_Store(), above.
+  ATOMICOPS_COMPILER_BARRIER();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+  GMemoryBarrier();
+  return *ptr;
+}
+
+#else // defined(__x86_64__)
+
+// 64-bit low-level operations on 32-bit platform.
+
+#if !((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1))
+// For compilers older than gcc 4.1, we use inline asm.
+//
+// Potential pitfalls:
+//
+// 1. %ebx points to Global offset table (GOT) with -fPIC.
+//    We need to preserve this register.
+// 2. When explicit registers are used in inline asm, the
+//    compiler may not be aware of it and might try to reuse
+//    the same register for another argument which has constraints
+//    that allow it ("r" for example).
+
+inline Atomic64 __sync_val_compare_and_swap(volatile Atomic64* ptr,
+                                            Atomic64 old_value,
+                                            Atomic64 new_value) {
+  Atomic64 prev;
+  __asm__ __volatile__("push %%ebx\n\t"
+                       "movl (%3), %%ebx\n\t"    // Move 64-bit new_value into
+                       "movl 4(%3), %%ecx\n\t"   // ecx:ebx
+                       "lock; cmpxchg8b (%1)\n\t"// If edx:eax (old_value) same
+                       "pop %%ebx\n\t"
+                       : "=A" (prev)             // as contents of ptr:
+                       : "D" (ptr),              //   ecx:ebx => ptr
+                         "0" (old_value),        // else:
+                         "S" (&new_value)        //   old *ptr => edx:eax
+                       : "memory", "%ecx");
+  return prev;
+}
+#endif  // Compiler < gcc-4.1
+
+inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                         Atomic64 old_val,
+                                         Atomic64 new_val) {
+  return __sync_val_compare_and_swap(ptr, old_val, new_val);
+}
+
+inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr,
+                                         Atomic64 new_val) {
+  Atomic64 old_val;
+
+  do {
+    old_val = *ptr;
+  } while (__sync_val_compare_and_swap(ptr, old_val, new_val) != old_val);
+
+  return old_val;
+}
+
+inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_val) {
+  Atomic64 old_val = NoBarrier_AtomicExchange(ptr, new_val);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return old_val;
+}
+
+inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr,
+                                       Atomic64 new_val) {
+ return NoBarrier_AtomicExchange(ptr, new_val);
+}
+
+inline Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr,
+                                          Atomic64 increment) {
+  Atomic64 old_val, new_val;
+
+  do {
+    old_val = *ptr;
+    new_val = old_val + increment;
+  } while (__sync_val_compare_and_swap(ptr, old_val, new_val) != old_val);
+
+  return old_val + increment;
+}
+
+inline Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr,
+                                        Atomic64 increment) {
+  Atomic64 new_val = NoBarrier_AtomicIncrement(ptr, increment);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return new_val;
+}
+
+inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) {
+  __asm__ __volatile__("movq %1, %%mm0\n\t"  // Use mmx reg for 64-bit atomic
+                       "movq %%mm0, %0\n\t"  // moves (ptr could be read-only)
+                       "emms\n\t"            // Empty mmx state/Reset FP regs
+                       : "=m" (*ptr)
+                       : "m" (value)
+                       : // mark the FP stack and mmx registers as clobbered
+			 "st", "st(1)", "st(2)", "st(3)", "st(4)",
+                         "st(5)", "st(6)", "st(7)", "mm0", "mm1",
+                         "mm2", "mm3", "mm4", "mm5", "mm6", "mm7");
+}
+
+inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) {
+  NoBarrier_Store(ptr, value);
+  MemoryBarrier();
+}
+
+inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) {
+  ATOMICOPS_COMPILER_BARRIER();
+  NoBarrier_Store(ptr, value);
+}
+
+inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) {
+  Atomic64 value;
+  __asm__ __volatile__("movq %1, %%mm0\n\t"  // Use mmx reg for 64-bit atomic
+                       "movq %%mm0, %0\n\t"  // moves (ptr could be read-only)
+                       "emms\n\t"            // Empty mmx state/Reset FP regs
+                       : "=m" (value)
+                       : "m" (*ptr)
+                       : // mark the FP stack and mmx registers as clobbered
+                         "st", "st(1)", "st(2)", "st(3)", "st(4)",
+                         "st(5)", "st(6)", "st(7)", "mm0", "mm1",
+                         "mm2", "mm3", "mm4", "mm5", "mm6", "mm7");
+  return value;
+}
+
+inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) {
+  Atomic64 value = NoBarrier_Load(ptr);
+  ATOMICOPS_COMPILER_BARRIER();
+  return value;
+}
+
+inline Atomic64 Release_Load(volatile const Atomic64* ptr) {
+  MemoryBarrier();
+  return NoBarrier_Load(ptr);
+}
+
+#endif // defined(__x86_64__)
+
+inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  Atomic64 x = NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+  if (AtomicOps_Internalx86CPUFeatures.has_amd_lock_mb_bug) {
+    __asm__ __volatile__("lfence" : : : "memory");
+  }
+  return x;
+}
+
+inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr,
+                                       Atomic64 old_value,
+                                       Atomic64 new_value) {
+  return NoBarrier_CompareAndSwap(ptr, old_value, new_value);
+}
+
+} // namespace base::subtle
+} // namespace base
+
+#undef ATOMICOPS_COMPILER_BARRIER
+
+#endif  // BASE_ATOMICOPS_INTERNALS_X86_H_

BeefRT/gperftools/src/base/atomicops.h

@@ -0,0 +1,406 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ */
+
+// For atomic operations on statistics counters, see atomic_stats_counter.h.
+// For atomic operations on sequence numbers, see atomic_sequence_num.h.
+// For atomic operations on reference counts, see atomic_refcount.h.
+
+// Some fast atomic operations -- typically with machine-dependent
+// implementations.  This file may need editing as Google code is
+// ported to different architectures.
+
+// The routines exported by this module are subtle.  If you use them, even if
+// you get the code right, it will depend on careful reasoning about atomicity
+// and memory ordering; it will be less readable, and harder to maintain.  If
+// you plan to use these routines, you should have a good reason, such as solid
+// evidence that performance would otherwise suffer, or there being no
+// alternative.  You should assume only properties explicitly guaranteed by the
+// specifications in this file.  You are almost certainly _not_ writing code
+// just for the x86; if you assume x86 semantics, x86 hardware bugs and
+// implementations on other archtectures will cause your code to break.  If you
+// do not know what you are doing, avoid these routines, and use a Mutex.
+//
+// These following lower-level operations are typically useful only to people
+// implementing higher-level synchronization operations like spinlocks,
+// mutexes, and condition-variables.  They combine CompareAndSwap(), a load, or
+// a store with appropriate memory-ordering instructions.  "Acquire" operations
+// ensure that no later memory access can be reordered ahead of the operation.
+// "Release" operations ensure that no previous memory access can be reordered
+// after the operation.  "Barrier" operations have both "Acquire" and "Release"
+// semantics.   A MemoryBarrier() has "Barrier" semantics, but does no memory
+// access.
+//
+// It is incorrect to make direct assignments to/from an atomic variable.
+// You should use one of the Load or Store routines.  The NoBarrier
+// versions are provided when no barriers are needed:
+//   NoBarrier_Store()
+//   NoBarrier_Load()
+// Although there are currently no compiler enforcement, you are encouraged
+// to use these.  Moreover, if you choose to use base::subtle::Atomic64 type,
+// you MUST use one of the Load or Store routines to get correct behavior
+// on 32-bit platforms.
+//
+// The intent is eventually to put all of these routines in namespace
+// base::subtle
+
+#ifndef THREAD_ATOMICOPS_H_
+#define THREAD_ATOMICOPS_H_
+
+#include "tcconfig.h"
+#ifdef HAVE_STDINT_H
+#include <stdint.h>
+#endif
+
+// ------------------------------------------------------------------------
+// Include the platform specific implementations of the types
+// and operations listed below.  Implementations are to provide Atomic32
+// and Atomic64 operations. If there is a mismatch between intptr_t and
+// the Atomic32 or Atomic64 types for a platform, the platform-specific header
+// should define the macro, AtomicWordCastType in a clause similar to the
+// following:
+// #if ...pointers are 64 bits...
+// # define AtomicWordCastType base::subtle::Atomic64
+// #else
+// # define AtomicWordCastType Atomic32
+// #endif
+// TODO(csilvers): figure out ARCH_PIII/ARCH_K8 (perhaps via ./configure?)
+// ------------------------------------------------------------------------
+
+#include "base/arm_instruction_set_select.h"
+
+// TODO(csilvers): match piii, not just __i386.  Also, match k8
+#if defined(__MACH__) && defined(__APPLE__)
+#include "base/atomicops-internals-macosx.h"
+#elif defined(__GNUC__) && defined(ARMV6)
+#include "base/atomicops-internals-arm-v6plus.h"
+#elif defined(ARMV3)
+#include "base/atomicops-internals-arm-generic.h"
+#elif defined(__GNUC__) && (defined(__i386) || defined(__x86_64__))
+#include "base/atomicops-internals-x86.h"
+#elif defined(_WIN32)
+#include "base/atomicops-internals-windows.h"
+#elif defined(__linux__) && defined(__PPC__)
+#include "base/atomicops-internals-linuxppc.h"
+#else
+// Assume x86 for now.  If you need to support a new architecture and
+// don't know how to implement atomic ops, you can probably get away
+// with using pthreads, since atomicops is only used by spinlock.h/cc
+//#error You need to implement atomic operations for this architecture
+#include "base/atomicops-internals-x86.h"
+#endif
+
+// Signed type that can hold a pointer and supports the atomic ops below, as
+// well as atomic loads and stores.  Instances must be naturally-aligned.
+typedef intptr_t AtomicWord;
+
+#ifdef AtomicWordCastType
+// ------------------------------------------------------------------------
+// This section is needed only when explicit type casting is required to
+// cast AtomicWord to one of the basic atomic types (Atomic64 or Atomic32).
+// It also serves to document the AtomicWord interface.
+// ------------------------------------------------------------------------
+
+namespace base {
+namespace subtle {
+
+// Atomically execute:
+//      result = *ptr;
+//      if (*ptr == old_value)
+//        *ptr = new_value;
+//      return result;
+//
+// I.e., replace "*ptr" with "new_value" if "*ptr" used to be "old_value".
+// Always return the old value of "*ptr"
+//
+// This routine implies no memory barriers.
+inline AtomicWord NoBarrier_CompareAndSwap(volatile AtomicWord* ptr,
+                                           AtomicWord old_value,
+                                           AtomicWord new_value) {
+  return NoBarrier_CompareAndSwap(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr),
+      old_value, new_value);
+}
+
+// Atomically store new_value into *ptr, returning the previous value held in
+// *ptr.  This routine implies no memory barriers.
+inline AtomicWord NoBarrier_AtomicExchange(volatile AtomicWord* ptr,
+                                           AtomicWord new_value) {
+  return NoBarrier_AtomicExchange(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), new_value);
+}
+
+inline AtomicWord Acquire_AtomicExchange(volatile AtomicWord* ptr,
+                                         AtomicWord new_value) {
+  return Acquire_AtomicExchange(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), new_value);
+}
+
+inline AtomicWord Release_AtomicExchange(volatile AtomicWord* ptr,
+                                         AtomicWord new_value) {
+  return Release_AtomicExchange(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), new_value);
+}
+
+// Atomically increment *ptr by "increment".  Returns the new value of
+// *ptr with the increment applied.  This routine implies no memory
+// barriers.
+inline AtomicWord NoBarrier_AtomicIncrement(volatile AtomicWord* ptr,
+                                            AtomicWord increment) {
+  return NoBarrier_AtomicIncrement(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), increment);
+}
+
+inline AtomicWord Barrier_AtomicIncrement(volatile AtomicWord* ptr,
+                                          AtomicWord increment) {
+  return Barrier_AtomicIncrement(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), increment);
+}
+
+inline AtomicWord Acquire_CompareAndSwap(volatile AtomicWord* ptr,
+                                         AtomicWord old_value,
+                                         AtomicWord new_value) {
+  return base::subtle::Acquire_CompareAndSwap(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr),
+      old_value, new_value);
+}
+
+inline AtomicWord Release_CompareAndSwap(volatile AtomicWord* ptr,
+                                         AtomicWord old_value,
+                                         AtomicWord new_value) {
+  return base::subtle::Release_CompareAndSwap(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr),
+      old_value, new_value);
+}
+
+inline void NoBarrier_Store(volatile AtomicWord *ptr, AtomicWord value) {
+  NoBarrier_Store(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), value);
+}
+
+inline void Acquire_Store(volatile AtomicWord* ptr, AtomicWord value) {
+  return base::subtle::Acquire_Store(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), value);
+}
+
+inline void Release_Store(volatile AtomicWord* ptr, AtomicWord value) {
+  return base::subtle::Release_Store(
+      reinterpret_cast<volatile AtomicWordCastType*>(ptr), value);
+}
+
+inline AtomicWord NoBarrier_Load(volatile const AtomicWord *ptr) {
+  return NoBarrier_Load(
+      reinterpret_cast<volatile const AtomicWordCastType*>(ptr));
+}
+
+inline AtomicWord Acquire_Load(volatile const AtomicWord* ptr) {
+  return base::subtle::Acquire_Load(
+      reinterpret_cast<volatile const AtomicWordCastType*>(ptr));
+}
+
+inline AtomicWord Release_Load(volatile const AtomicWord* ptr) {
+  return base::subtle::Release_Load(
+      reinterpret_cast<volatile const AtomicWordCastType*>(ptr));
+}
+
+}  // namespace base::subtle
+}  // namespace base
+#endif  // AtomicWordCastType
+
+// ------------------------------------------------------------------------
+// Commented out type definitions and method declarations for documentation
+// of the interface provided by this module.
+// ------------------------------------------------------------------------
+
+#if 0
+
+// Signed 32-bit type that supports the atomic ops below, as well as atomic
+// loads and stores.  Instances must be naturally aligned.  This type differs
+// from AtomicWord in 64-bit binaries where AtomicWord is 64-bits.
+typedef int32_t Atomic32;
+
+// Corresponding operations on Atomic32
+namespace base {
+namespace subtle {
+
+// Signed 64-bit type that supports the atomic ops below, as well as atomic
+// loads and stores.  Instances must be naturally aligned.  This type differs
+// from AtomicWord in 32-bit binaries where AtomicWord is 32-bits.
+typedef int64_t Atomic64;
+
+Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr,
+                                  Atomic32 old_value,
+                                  Atomic32 new_value);
+Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value);
+Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value);
+Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value);
+Atomic32 NoBarrier_AtomicIncrement(volatile Atomic32* ptr, Atomic32 increment);
+Atomic32 Barrier_AtomicIncrement(volatile Atomic32* ptr,
+                                 Atomic32 increment);
+Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                Atomic32 old_value,
+                                Atomic32 new_value);
+Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                Atomic32 old_value,
+                                Atomic32 new_value);
+void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value);
+void Acquire_Store(volatile Atomic32* ptr, Atomic32 value);
+void Release_Store(volatile Atomic32* ptr, Atomic32 value);
+Atomic32 NoBarrier_Load(volatile const Atomic32* ptr);
+Atomic32 Acquire_Load(volatile const Atomic32* ptr);
+Atomic32 Release_Load(volatile const Atomic32* ptr);
+
+// Corresponding operations on Atomic64
+Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr,
+                                  Atomic64 old_value,
+                                  Atomic64 new_value);
+Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value);
+Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value);
+Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value);
+Atomic64 NoBarrier_AtomicIncrement(volatile Atomic64* ptr, Atomic64 increment);
+Atomic64 Barrier_AtomicIncrement(volatile Atomic64* ptr, Atomic64 increment);
+
+Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr,
+                                Atomic64 old_value,
+                                Atomic64 new_value);
+Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr,
+                                Atomic64 old_value,
+                                Atomic64 new_value);
+void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value);
+void Acquire_Store(volatile Atomic64* ptr, Atomic64 value);
+void Release_Store(volatile Atomic64* ptr, Atomic64 value);
+Atomic64 NoBarrier_Load(volatile const Atomic64* ptr);
+Atomic64 Acquire_Load(volatile const Atomic64* ptr);
+Atomic64 Release_Load(volatile const Atomic64* ptr);
+}  // namespace base::subtle
+}  // namespace base
+
+void GMemoryBarrier();
+
+#endif  // 0
+
+
+// ------------------------------------------------------------------------
+// The following are to be deprecated when all uses have been changed to
+// use the base::subtle namespace.
+// ------------------------------------------------------------------------
+
+#ifdef AtomicWordCastType
+// AtomicWord versions to be deprecated
+inline AtomicWord Acquire_CompareAndSwap(volatile AtomicWord* ptr,
+                                         AtomicWord old_value,
+                                         AtomicWord new_value) {
+  return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline AtomicWord Release_CompareAndSwap(volatile AtomicWord* ptr,
+                                         AtomicWord old_value,
+                                         AtomicWord new_value) {
+  return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value);
+}
+
+inline void Acquire_Store(volatile AtomicWord* ptr, AtomicWord value) {
+  return base::subtle::Acquire_Store(ptr, value);
+}
+
+inline void Release_Store(volatile AtomicWord* ptr, AtomicWord value) {
+  return base::subtle::Release_Store(ptr, value);
+}
+
+inline AtomicWord Acquire_Load(volatile const AtomicWord* ptr) {
+  return base::subtle::Acquire_Load(ptr);
+}
+
+inline AtomicWord Release_Load(volatile const AtomicWord* ptr) {
+  return base::subtle::Release_Load(ptr);
+}
+#endif  // AtomicWordCastType
+
+// 32-bit Acquire/Release operations to be deprecated.
+
+inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value);
+}
+inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr,
+                                       Atomic32 old_value,
+                                       Atomic32 new_value) {
+  return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value);
+}
+inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) {
+  base::subtle::Acquire_Store(ptr, value);
+}
+inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) {
+  return base::subtle::Release_Store(ptr, value);
+}
+inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) {
+  return base::subtle::Acquire_Load(ptr);
+}
+inline Atomic32 Release_Load(volatile const Atomic32* ptr) {
+  return base::subtle::Release_Load(ptr);
+}
+
+#ifdef BASE_HAS_ATOMIC64
+
+// 64-bit Acquire/Release operations to be deprecated.
+
+inline base::subtle::Atomic64 Acquire_CompareAndSwap(
+    volatile base::subtle::Atomic64* ptr,
+    base::subtle::Atomic64 old_value, base::subtle::Atomic64 new_value) {
+  return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value);
+}
+inline base::subtle::Atomic64 Release_CompareAndSwap(
+    volatile base::subtle::Atomic64* ptr,
+    base::subtle::Atomic64 old_value, base::subtle::Atomic64 new_value) {
+  return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value);
+}
+/*inline void Acquire_Store(
+    volatile base::subtle::Atomic64* ptr, base::subtle::Atomic64 value) {
+  base::subtle::Acquire_Store(ptr, value);
+}
+inline void Release_Store(
+    volatile base::subtle::Atomic64* ptr, base::subtle::Atomic64 value) {
+  return base::subtle::Release_Store(ptr, value);
+}
+inline base::subtle::Atomic64 Acquire_Load(
+    volatile const base::subtle::Atomic64* ptr) {
+  return base::subtle::Acquire_Load(ptr);
+}
+inline base::subtle::Atomic64 Release_Load(
+    volatile const base::subtle::Atomic64* ptr) {
+  return base::subtle::Release_Load(ptr);
+}*/
+
+#endif  // BASE_HAS_ATOMIC64
+
+#endif  // THREAD_ATOMICOPS_H_

BeefRT/gperftools/src/base/basictypes.h

@@ -0,0 +1,377 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef _BASICTYPES_H_
+#define _BASICTYPES_H_
+
+#include "tcconfig.h"
+#include <string.h>       // for memcpy()
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h>     // gets us PRId64, etc
+#endif
+
+// To use this in an autoconf setting, make sure you run the following
+// autoconf macros:
+//    AC_HEADER_STDC              /* for stdint_h and inttypes_h */
+//    AC_CHECK_TYPES([__int64])   /* defined in some windows platforms */
+
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h>           // uint16_t might be here; PRId64 too.
+#endif
+#ifdef HAVE_STDINT_H
+#include <stdint.h>             // to get uint16_t (ISO naming madness)
+#endif
+#include <sys/types.h>          // our last best hope for uint16_t
+
+// Standard typedefs
+// All Google code is compiled with -funsigned-char to make "char"
+// unsigned.  Google code therefore doesn't need a "uchar" type.
+// TODO(csilvers): how do we make sure unsigned-char works on non-gcc systems?
+typedef signed char         schar;
+typedef int8_t              int8;
+typedef int16_t             int16;
+typedef int32_t             int32;
+typedef int64_t             int64;
+
+// NOTE: unsigned types are DANGEROUS in loops and other arithmetical
+// places.  Use the signed types unless your variable represents a bit
+// pattern (eg a hash value) or you really need the extra bit.  Do NOT
+// use 'unsigned' to express "this value should always be positive";
+// use assertions for this.
+
+typedef uint8_t            uint8;
+typedef uint16_t           uint16;
+typedef uint32_t           uint32;
+typedef uint64_t           uint64;
+
+const uint16 kuint16max = (   (uint16) 0xFFFF);
+const uint32 kuint32max = (   (uint32) 0xFFFFFFFF);
+const uint64 kuint64max = ( (((uint64) kuint32max) << 32) | kuint32max );
+
+const  int8  kint8max   = (   (  int8) 0x7F);
+const  int16 kint16max  = (   ( int16) 0x7FFF);
+const  int32 kint32max  = (   ( int32) 0x7FFFFFFF);
+const  int64 kint64max =  ( ((( int64) kint32max) << 32) | kuint32max );
+
+const  int8  kint8min   = (   (  int8) 0x80);
+const  int16 kint16min  = (   ( int16) 0x8000);
+const  int32 kint32min  = (   ( int32) 0x80000000);
+const  int64 kint64min =  ( (((uint64) kint32min) << 32) | 0 );
+
+// Define the "portable" printf and scanf macros, if they're not
+// already there (via the inttypes.h we #included above, hopefully).
+// Mostly it's old systems that don't support inttypes.h, so we assume
+// they're 32 bit.
+#ifndef PRIx64
+#define PRIx64 "llx"
+#endif
+#ifndef SCNx64
+#define SCNx64 "llx"
+#endif
+#ifndef PRId64
+#define PRId64 "lld"
+#endif
+#ifndef SCNd64
+#define SCNd64 "lld"
+#endif
+#ifndef PRIu64
+#define PRIu64 "llu"
+#endif
+#ifndef PRIxPTR
+#define PRIxPTR "lx"
+#endif
+
+// Also allow for printing of a pthread_t.
+#define GPRIuPTHREAD "lu"
+#define GPRIxPTHREAD "lx"
+#if defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__APPLE__) || defined(__FreeBSD__)
+#define PRINTABLE_PTHREAD(pthreadt) reinterpret_cast<uintptr_t>(pthreadt)
+#else
+#define PRINTABLE_PTHREAD(pthreadt) pthreadt
+#endif
+
+// A macro to disallow the evil copy constructor and operator= functions
+// This should be used in the private: declarations for a class
+#define DISALLOW_EVIL_CONSTRUCTORS(TypeName)    \
+  TypeName(const TypeName&);                    \
+  void operator=(const TypeName&)
+
+// An alternate name that leaves out the moral judgment... :-)
+#define DISALLOW_COPY_AND_ASSIGN(TypeName) DISALLOW_EVIL_CONSTRUCTORS(TypeName)
+
+// The COMPILE_ASSERT macro can be used to verify that a compile time
+// expression is true. For example, you could use it to verify the
+// size of a static array:
+//
+//   COMPILE_ASSERT(sizeof(num_content_type_names) == sizeof(int),
+//                  content_type_names_incorrect_size);
+//
+// or to make sure a struct is smaller than a certain size:
+//
+//   COMPILE_ASSERT(sizeof(foo) < 128, foo_too_large);
+//
+// The second argument to the macro is the name of the variable. If
+// the expression is false, most compilers will issue a warning/error
+// containing the name of the variable.
+//
+// Implementation details of COMPILE_ASSERT:
+//
+// - COMPILE_ASSERT works by defining an array type that has -1
+//   elements (and thus is invalid) when the expression is false.
+//
+// - The simpler definition
+//
+//     #define COMPILE_ASSERT(expr, msg) typedef char msg[(expr) ? 1 : -1]
+//
+//   does not work, as gcc supports variable-length arrays whose sizes
+//   are determined at run-time (this is gcc's extension and not part
+//   of the C++ standard).  As a result, gcc fails to reject the
+//   following code with the simple definition:
+//
+//     int foo;
+//     COMPILE_ASSERT(foo, msg); // not supposed to compile as foo is
+//                               // not a compile-time constant.
+//
+// - By using the type CompileAssert<(bool(expr))>, we ensures that
+//   expr is a compile-time constant.  (Template arguments must be
+//   determined at compile-time.)
+//
+// - The outter parentheses in CompileAssert<(bool(expr))> are necessary
+//   to work around a bug in gcc 3.4.4 and 4.0.1.  If we had written
+//
+//     CompileAssert<bool(expr)>
+//
+//   instead, these compilers will refuse to compile
+//
+//     COMPILE_ASSERT(5 > 0, some_message);
+//
+//   (They seem to think the ">" in "5 > 0" marks the end of the
+//   template argument list.)
+//
+// - The array size is (bool(expr) ? 1 : -1), instead of simply
+//
+//     ((expr) ? 1 : -1).
+//
+//   This is to avoid running into a bug in MS VC 7.1, which
+//   causes ((0.0) ? 1 : -1) to incorrectly evaluate to 1.
+
+template <bool>
+struct CompileAssert {
+};
+
+#ifdef HAVE___ATTRIBUTE__
+# define ATTRIBUTE_UNUSED __attribute__((unused))
+#else
+# define ATTRIBUTE_UNUSED
+#endif
+
+#define COMPILE_ASSERT(expr, msg)                               \
+  typedef CompileAssert<(bool(expr))> msg[bool(expr) ? 1 : -1] ATTRIBUTE_UNUSED
+
+#define arraysize(a)  (sizeof(a) / sizeof(*(a)))
+
+#define OFFSETOF_MEMBER(strct, field)                                   \
+   (reinterpret_cast<char*>(&reinterpret_cast<strct*>(16)->field) -     \
+    reinterpret_cast<char*>(16))
+
+// bit_cast<Dest,Source> implements the equivalent of
+// "*reinterpret_cast<Dest*>(&source)".
+//
+// The reinterpret_cast method would produce undefined behavior
+// according to ISO C++ specification section 3.10 -15 -.
+// bit_cast<> calls memcpy() which is blessed by the standard,
+// especially by the example in section 3.9.
+//
+// Fortunately memcpy() is very fast.  In optimized mode, with a
+// constant size, gcc 2.95.3, gcc 4.0.1, and msvc 7.1 produce inline
+// code with the minimal amount of data movement.  On a 32-bit system,
+// memcpy(d,s,4) compiles to one load and one store, and memcpy(d,s,8)
+// compiles to two loads and two stores.
+
+template <class Dest, class Source>
+inline Dest bit_cast(const Source& source) {
+  COMPILE_ASSERT(sizeof(Dest) == sizeof(Source), bitcasting_unequal_sizes);
+  Dest dest;
+  memcpy(&dest, &source, sizeof(dest));
+  return dest;
+}
+
+#ifdef HAVE___ATTRIBUTE__
+# define ATTRIBUTE_WEAK      __attribute__((weak))
+# define ATTRIBUTE_NOINLINE  __attribute__((noinline))
+#else
+# define ATTRIBUTE_WEAK
+# define ATTRIBUTE_NOINLINE
+#endif
+
+// Section attributes are supported for both ELF and Mach-O, but in
+// very different ways.  Here's the API we provide:
+// 1) ATTRIBUTE_SECTION: put this with the declaration of all functions
+//    you want to be in the same linker section
+// 2) DEFINE_ATTRIBUTE_SECTION_VARS: must be called once per unique
+//    name.  You want to make sure this is executed before any
+//    DECLARE_ATTRIBUTE_SECTION_VARS; the easiest way is to put them
+//    in the same .cc file.  Put this call at the global level.
+// 3) INIT_ATTRIBUTE_SECTION_VARS: you can scatter calls to this in
+//    multiple places to help ensure execution before any
+//    DECLARE_ATTRIBUTE_SECTION_VARS.  You must have at least one
+//    DEFINE, but you can have many INITs.  Put each in its own scope.
+// 4) DECLARE_ATTRIBUTE_SECTION_VARS: must be called before using
+//    ATTRIBUTE_SECTION_START or ATTRIBUTE_SECTION_STOP on a name.
+//    Put this call at the global level.
+// 5) ATTRIBUTE_SECTION_START/ATTRIBUTE_SECTION_STOP: call this to say
+//    where in memory a given section is.  All functions declared with
+//    ATTRIBUTE_SECTION are guaranteed to be between START and STOP.
+
+#if defined(HAVE___ATTRIBUTE__) && defined(__ELF__)
+# define ATTRIBUTE_SECTION(name) __attribute__ ((section (#name)))
+
+  // Weak section declaration to be used as a global declaration
+  // for ATTRIBUTE_SECTION_START|STOP(name) to compile and link
+  // even without functions with ATTRIBUTE_SECTION(name).
+# define DECLARE_ATTRIBUTE_SECTION_VARS(name) \
+    extern char __start_##name[] ATTRIBUTE_WEAK; \
+    extern char __stop_##name[] ATTRIBUTE_WEAK
+# define INIT_ATTRIBUTE_SECTION_VARS(name)     // no-op for ELF
+# define DEFINE_ATTRIBUTE_SECTION_VARS(name)   // no-op for ELF
+
+  // Return void* pointers to start/end of a section of code with functions
+  // having ATTRIBUTE_SECTION(name), or 0 if no such function exists.
+  // One must DECLARE_ATTRIBUTE_SECTION(name) for this to compile and link.
+# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast<void*>(__start_##name))
+# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast<void*>(__stop_##name))
+# define HAVE_ATTRIBUTE_SECTION_START 1
+
+#elif defined(HAVE___ATTRIBUTE__) && defined(__MACH__)
+# define ATTRIBUTE_SECTION(name) __attribute__ ((section ("__TEXT, " #name)))
+
+#include <mach-o/getsect.h>
+#include <mach-o/dyld.h>
+class AssignAttributeStartEnd {
+ public:
+  AssignAttributeStartEnd(const char* name, char** pstart, char** pend) {
+    // Find out what dynamic library name is defined in
+      
+#if 0
+    if (_dyld_present())
+    {
+      for (int i = _dyld_image_count() - 1; i >= 0; --i) {
+        const mach_header* hdr = _dyld_get_image_header(i);
+#ifdef MH_MAGIC_64
+        if (hdr->magic == MH_MAGIC_64) {
+          uint64_t len;
+          *pstart = getsectdatafromheader_64((mach_header_64*)hdr,
+                                             "__TEXT", name, &len);
+          if (*pstart) {   // NULL if not defined in this dynamic library
+            *pstart += _dyld_get_image_vmaddr_slide(i);   // correct for reloc
+            *pend = *pstart + len;
+            return;
+          }
+        }
+#endif
+        if (hdr->magic == MH_MAGIC) {
+          uint32_t len;
+          *pstart = getsectdatafromheader(hdr, "__TEXT", name, &len);
+          if (*pstart) {   // NULL if not defined in this dynamic library
+            *pstart += _dyld_get_image_vmaddr_slide(i);   // correct for reloc
+            *pend = *pstart + len;
+            return;
+          }
+        }
+      }
+    }
+#endif
+      
+    // If we get here, not defined in a dll at all.  See if defined statically.
+    unsigned long len;    // don't ask me why this type isn't uint32_t too...
+    *pstart = getsectdata("__TEXT", name, &len);
+    *pend = *pstart + len;
+  }
+};
+
+#define DECLARE_ATTRIBUTE_SECTION_VARS(name)    \
+  extern char* __start_##name;                  \
+  extern char* __stop_##name
+
+#define INIT_ATTRIBUTE_SECTION_VARS(name)               \
+  DECLARE_ATTRIBUTE_SECTION_VARS(name);                 \
+  static const AssignAttributeStartEnd __assign_##name( \
+    #name, &__start_##name, &__stop_##name)
+
+#define DEFINE_ATTRIBUTE_SECTION_VARS(name)     \
+  char* __start_##name, *__stop_##name;         \
+  INIT_ATTRIBUTE_SECTION_VARS(name)
+
+# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast<void*>(__start_##name))
+# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast<void*>(__stop_##name))
+# define HAVE_ATTRIBUTE_SECTION_START 1
+
+#else  // not HAVE___ATTRIBUTE__ && __ELF__, nor HAVE___ATTRIBUTE__ && __MACH__
+# define ATTRIBUTE_SECTION(name)
+# define DECLARE_ATTRIBUTE_SECTION_VARS(name)
+# define INIT_ATTRIBUTE_SECTION_VARS(name)
+# define DEFINE_ATTRIBUTE_SECTION_VARS(name)
+# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast<void*>(0))
+# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast<void*>(0))
+
+#endif  // HAVE___ATTRIBUTE__ and __ELF__ or __MACH__
+
+#if defined(HAVE___ATTRIBUTE__)
+# if (defined(__i386__) || defined(__x86_64__))
+#   define CACHELINE_ALIGNED __attribute__((aligned(64)))
+# elif (defined(__PPC__) || defined(__PPC64__))
+#   define CACHELINE_ALIGNED __attribute__((aligned(16)))
+# elif (defined(__arm__))
+#   define CACHELINE_ALIGNED __attribute__((aligned(64)))
+    // some ARMs have shorter cache lines (ARM1176JZF-S is 32 bytes for example) but obviously 64-byte aligned implies 32-byte aligned
+# else
+#   error Could not determine cache line length - unknown architecture
+# endif
+#else
+# define CACHELINE_ALIGNED
+#endif  // defined(HAVE___ATTRIBUTE__) && (__i386__ || __x86_64__)
+
+
+// The following enum should be used only as a constructor argument to indicate
+// that the variable has static storage class, and that the constructor should
+// do nothing to its state.  It indicates to the reader that it is legal to
+// declare a static nistance of the class, provided the constructor is given
+// the base::LINKER_INITIALIZED argument.  Normally, it is unsafe to declare a
+// static variable that has a constructor or a destructor because invocation
+// order is undefined.  However, IF the type can be initialized by filling with
+// zeroes (which the loader does for static variables), AND the destructor also
+// does nothing to the storage, then a constructor declared as
+//       explicit MyClass(base::LinkerInitialized x) {}
+// and invoked as
+//       static MyClass my_variable_name(base::LINKER_INITIALIZED);
+namespace base {
+enum LinkerInitialized { LINKER_INITIALIZED };
+}
+
+#endif  // _BASICTYPES_H_

BeefRT/gperftools/src/base/commandlineflags.h

@@ -0,0 +1,131 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// This file is a compatibility layer that defines Google's version of
+// command line flags that are used for configuration.
+//
+// We put flags into their own namespace.  It is purposefully
+// named in an opaque way that people should have trouble typing
+// directly.  The idea is that DEFINE puts the flag in the weird
+// namespace, and DECLARE imports the flag from there into the
+// current namespace.  The net result is to force people to use
+// DECLARE to get access to a flag, rather than saying
+//   extern bool FLAGS_logtostderr;
+// or some such instead.  We want this so we can put extra
+// functionality (like sanity-checking) in DECLARE if we want,
+// and make sure it is picked up everywhere.
+//
+// We also put the type of the variable in the namespace, so that
+// people can't DECLARE_int32 something that they DEFINE_bool'd
+// elsewhere.
+#ifndef BASE_COMMANDLINEFLAGS_H_
+#define BASE_COMMANDLINEFLAGS_H_
+
+#include "tcconfig.h"
+#include <string>
+#include <string.h>               // for memchr
+#include <stdlib.h>               // for getenv
+#include "base/basictypes.h"
+
+#define DECLARE_VARIABLE(type, name)                                          \
+  namespace FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead {  \
+  extern PERFTOOLS_DLL_DECL type FLAGS_##name;                                \
+  }                                                                           \
+  using FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead::FLAGS_##name
+
+#define DEFINE_VARIABLE(type, name, value, meaning) \
+  namespace FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead {  \
+  PERFTOOLS_DLL_DECL type FLAGS_##name(value);                                \
+  char FLAGS_no##name;                                                        \
+  }                                                                           \
+  using FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead::FLAGS_##name
+
+// bool specialization
+#define DECLARE_bool(name) \
+  DECLARE_VARIABLE(bool, name)
+#define DEFINE_bool(name, value, meaning) \
+  DEFINE_VARIABLE(bool, name, value, meaning)
+
+// int32 specialization
+#define DECLARE_int32(name) \
+  DECLARE_VARIABLE(int32, name)
+#define DEFINE_int32(name, value, meaning) \
+  DEFINE_VARIABLE(int32, name, value, meaning)
+
+// int64 specialization
+#define DECLARE_int64(name) \
+  DECLARE_VARIABLE(int64, name)
+#define DEFINE_int64(name, value, meaning) \
+  DEFINE_VARIABLE(int64, name, value, meaning)
+
+#define DECLARE_uint64(name) \
+  DECLARE_VARIABLE(uint64, name)
+#define DEFINE_uint64(name, value, meaning) \
+  DEFINE_VARIABLE(uint64, name, value, meaning)
+
+// double specialization
+#define DECLARE_double(name) \
+  DECLARE_VARIABLE(double, name)
+#define DEFINE_double(name, value, meaning) \
+  DEFINE_VARIABLE(double, name, value, meaning)
+
+// Special case for string, because we have to specify the namespace
+// std::string, which doesn't play nicely with our FLAG__namespace hackery.
+#define DECLARE_string(name)                                          \
+  namespace FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead {  \
+  extern std::string FLAGS_##name;                                                   \
+  }                                                                           \
+  using FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead::FLAGS_##name
+#define DEFINE_string(name, value, meaning) \
+  namespace FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead {  \
+  std::string FLAGS_##name(value);                                                   \
+  char FLAGS_no##name;                                                        \
+  }                                                                           \
+  using FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead::FLAGS_##name
+
+// These macros (could be functions, but I don't want to bother with a .cc
+// file), make it easier to initialize flags from the environment.
+
+#define EnvToString(envname, dflt)   \
+  (!getenv(envname) ? (dflt) : getenv(envname))
+
+#define EnvToBool(envname, dflt)   \
+  (!getenv(envname) ? (dflt) : memchr("tTyY1\0", getenv(envname)[0], 6) != NULL)
+
+#define EnvToInt(envname, dflt)  \
+  (!getenv(envname) ? (dflt) : strtol(getenv(envname), NULL, 10))
+
+#define EnvToInt64(envname, dflt)  \
+  (!getenv(envname) ? (dflt) : strtoll(getenv(envname), NULL, 10))
+
+#define EnvToDouble(envname, dflt)  \
+  (!getenv(envname) ? (dflt) : strtod(getenv(envname), NULL))
+
+#endif  // BASE_COMMANDLINEFLAGS_H_

BeefRT/gperftools/src/base/cycleclock.h

@@ -0,0 +1,172 @@
+// Copyright (c) 2004, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ----------------------------------------------------------------------
+// CycleClock
+//    A CycleClock tells you the current time in Cycles.  The "time"
+//    is actually time since power-on.  This is like time() but doesn't
+//    involve a system call and is much more precise.
+//
+// NOTE: Not all cpu/platform/kernel combinations guarantee that this
+// clock increments at a constant rate or is synchronized across all logical
+// cpus in a system.
+//
+// Also, in some out of order CPU implementations, the CycleClock is not 
+// serializing. So if you're trying to count at cycles granularity, your
+// data might be inaccurate due to out of order instruction execution.
+// ----------------------------------------------------------------------
+
+#ifndef GOOGLE_BASE_CYCLECLOCK_H_
+#define GOOGLE_BASE_CYCLECLOCK_H_
+
+#include "base/basictypes.h"   // make sure we get the def for int64
+#include "base/arm_instruction_set_select.h"
+// base/sysinfo.h is really big and we don't want to include it unless
+// it is necessary.
+#if defined(__arm__) || defined(__mips__)
+# include "base/sysinfo.h"
+#endif
+#if defined(__MACH__) && defined(__APPLE__)
+# include <mach/mach_time.h>
+#endif
+// For MSVC, we want to use '_asm rdtsc' when possible (since it works
+// with even ancient MSVC compilers), and when not possible the
+// __rdtsc intrinsic, declared in <intrin.h>.  Unfortunately, in some
+// environments, <windows.h> and <intrin.h> have conflicting
+// declarations of some other intrinsics, breaking compilation.
+// Therefore, we simply declare __rdtsc ourselves. See also
+// http://connect.microsoft.com/VisualStudio/feedback/details/262047
+#if defined(_MSC_VER) && !defined(_M_IX86)
+extern "C" uint64 __rdtsc();
+#pragma intrinsic(__rdtsc)
+#endif
+#if defined(ARMV3) || defined(__mips__)
+#include <sys/time.h>
+#endif
+
+// NOTE: only i386 and x86_64 have been well tested.
+// PPC, sparc, alpha, and ia64 are based on
+//    http://peter.kuscsik.com/wordpress/?p=14
+// with modifications by m3b.  See also
+//    https://setisvn.ssl.berkeley.edu/svn/lib/fftw-3.0.1/kernel/cycle.h
+struct CycleClock {
+  // This should return the number of cycles since power-on.  Thread-safe.
+  static inline int64 Now() {
+#if defined(__MACH__) && defined(__APPLE__)
+    // this goes at the top because we need ALL Macs, regardless of
+    // architecture, to return the number of "mach time units" that
+    // have passed since startup.  See sysinfo.cc where
+    // InitializeSystemInfo() sets the supposed cpu clock frequency of
+    // macs to the number of mach time units per second, not actual
+    // CPU clock frequency (which can change in the face of CPU
+    // frequency scaling).  Also note that when the Mac sleeps, this
+    // counter pauses; it does not continue counting, nor does it
+    // reset to zero.
+    return mach_absolute_time();
+#elif defined(_MSC_VER) && defined(_M_IX86)
+	  // Older MSVC compilers (like 7.x) don't seem to support the
+	  // __rdtsc intrinsic properly, so I prefer to use _asm instead
+	  // when I know it will work.  Otherwise, I'll use __rdtsc and hope
+	  // the code is being compiled with a non-ancient compiler.
+	  _asm rdtsc
+#elif defined(_MSC_VER)
+	  return __rdtsc();
+#elif defined(__i386__)
+    int64 ret;
+    __asm__ volatile ("rdtsc" : "=A" (ret) );
+    return ret;
+#elif defined(__x86_64__) || defined(__amd64__)
+    uint64 low, high;
+    __asm__ volatile ("rdtsc" : "=a" (low), "=d" (high));
+    return (high << 32) | low;
+#elif defined(__powerpc64__) || defined(__ppc64__)
+    uint64 tb;
+    __asm__ volatile (\
+      "mfspr %0, 268"
+      : "=r" (tb));
+    return tb;
+#elif defined(__powerpc__) || defined(__ppc__)
+    // This returns a time-base, which is not always precisely a cycle-count.
+    uint32 tbu, tbl, tmp;
+    __asm__ volatile (\
+      "0:\n"
+      "mftbu %0\n"
+      "mftbl %1\n"
+      "mftbu %2\n"
+      "cmpw %0, %2\n"
+      "bne- 0b"
+      : "=r" (tbu), "=r" (tbl), "=r" (tmp));
+    return (((uint64) tbu << 32) | tbl);
+#elif defined(__sparc__)
+    int64 tick;
+    asm(".byte 0x83, 0x41, 0x00, 0x00");
+    asm("mov   %%g1, %0" : "=r" (tick));
+    return tick;
+#elif defined(__ia64__)
+    int64 itc;
+    asm("mov %0 = ar.itc" : "=r" (itc));
+    return itc;
+#elif defined(ARMV3)
+#if defined(ARMV6)  // V6 is the earliest arch that has a standard cyclecount
+    uint32 pmccntr;
+    uint32 pmuseren;
+    uint32 pmcntenset;
+    // Read the user mode perf monitor counter access permissions.
+    asm volatile ("mrc p15, 0, %0, c9, c14, 0" : "=r" (pmuseren));
+    if (pmuseren & 1) {  // Allows reading perfmon counters for user mode code.
+      asm volatile ("mrc p15, 0, %0, c9, c12, 1" : "=r" (pmcntenset));
+      if (pmcntenset & 0x80000000ul) {  // Is it counting?
+        asm volatile ("mrc p15, 0, %0, c9, c13, 0" : "=r" (pmccntr));
+        // The counter is set up to count every 64th cycle
+        return static_cast<int64>(pmccntr) * 64;  // Should optimize to << 6
+      }
+    }
+#endif
+    struct timeval tv;
+    gettimeofday(&tv, NULL);
+    return static_cast<int64>((tv.tv_sec + tv.tv_usec * 0.000001)
+                              * CyclesPerSecond());
+#elif defined(__mips__)
+    // mips apparently only allows rdtsc for superusers, so we fall
+    // back to gettimeofday.  It's possible clock_gettime would be better.
+    struct timeval tv;
+    gettimeofday(&tv, NULL);
+    return static_cast<int64>((tv.tv_sec + tv.tv_usec * 0.000001)
+                              * CyclesPerSecond());
+#else
+// The soft failover to a generic implementation is automatic only for ARM.
+// For other platforms the developer is expected to make an attempt to create
+// a fast implementation and use generic version if nothing better is available.
+#error You need to define CycleTimer for your O/S and CPU
+#endif
+  }
+};
+
+
+#endif  // GOOGLE_BASE_CYCLECLOCK_H_

BeefRT/gperftools/src/base/dynamic_annotations.cc

@@ -0,0 +1,186 @@
+/* Copyright (c) 2008-2009, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Kostya Serebryany
+ */
+
+#include "tcconfig.h"
+#include <stdlib.h>
+#include <string.h>
+
+#include "base/dynamic_annotations.h"
+
+#ifdef __GNUC__
+/* valgrind.h uses gcc extensions so it won't build with other compilers */
+# ifdef HAVE_VALGRIND_H    /* prefer the user's copy if they have it */
+#  include <valgrind.h>
+# else                     /* otherwise just use the copy that we have */
+#  include "third_party/valgrind.h"
+# endif
+#endif
+
+/* Compiler-based ThreadSanitizer defines
+   DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL = 1
+   and provides its own definitions of the functions. */
+
+#ifndef DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL
+# define DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL 0
+#endif
+
+/* Each function is empty and called (via a macro) only in debug mode.
+   The arguments are captured by dynamic tools at runtime. */
+
+#if DYNAMIC_ANNOTATIONS_ENABLED == 1 \
+    && DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0
+
+void AnnotateRWLockCreate(const char *file, int line,
+                          const volatile void *lock){}
+void AnnotateRWLockDestroy(const char *file, int line,
+                           const volatile void *lock){}
+void AnnotateRWLockAcquired(const char *file, int line,
+                            const volatile void *lock, long is_w){}
+void AnnotateRWLockReleased(const char *file, int line,
+                            const volatile void *lock, long is_w){}
+void AnnotateBarrierInit(const char *file, int line,
+                         const volatile void *barrier, long count,
+                         long reinitialization_allowed) {}
+void AnnotateBarrierWaitBefore(const char *file, int line,
+                               const volatile void *barrier) {}
+void AnnotateBarrierWaitAfter(const char *file, int line,
+                              const volatile void *barrier) {}
+void AnnotateBarrierDestroy(const char *file, int line,
+                            const volatile void *barrier) {}
+
+void AnnotateCondVarWait(const char *file, int line,
+                         const volatile void *cv,
+                         const volatile void *lock){}
+void AnnotateCondVarSignal(const char *file, int line,
+                           const volatile void *cv){}
+void AnnotateCondVarSignalAll(const char *file, int line,
+                              const volatile void *cv){}
+void AnnotatePublishMemoryRange(const char *file, int line,
+                                const volatile void *address,
+                                long size){}
+void AnnotateUnpublishMemoryRange(const char *file, int line,
+                                  const volatile void *address,
+                                  long size){}
+void AnnotatePCQCreate(const char *file, int line,
+                       const volatile void *pcq){}
+void AnnotatePCQDestroy(const char *file, int line,
+                        const volatile void *pcq){}
+void AnnotatePCQPut(const char *file, int line,
+                    const volatile void *pcq){}
+void AnnotatePCQGet(const char *file, int line,
+                    const volatile void *pcq){}
+void AnnotateNewMemory(const char *file, int line,
+                       const volatile void *mem,
+                       long size){}
+void AnnotateExpectRace(const char *file, int line,
+                        const volatile void *mem,
+                        const char *description){}
+void AnnotateBenignRace(const char *file, int line,
+                        const volatile void *mem,
+                        const char *description){}
+void AnnotateBenignRaceSized(const char *file, int line,
+                             const volatile void *mem,
+                             long size,
+                             const char *description) {}
+void AnnotateMutexIsUsedAsCondVar(const char *file, int line,
+                                  const volatile void *mu){}
+void AnnotateTraceMemory(const char *file, int line,
+                         const volatile void *arg){}
+void AnnotateThreadName(const char *file, int line,
+                        const char *name){}
+void AnnotateIgnoreReadsBegin(const char *file, int line){}
+void AnnotateIgnoreReadsEnd(const char *file, int line){}
+void AnnotateIgnoreWritesBegin(const char *file, int line){}
+void AnnotateIgnoreWritesEnd(const char *file, int line){}
+void AnnotateEnableRaceDetection(const char *file, int line, int enable){}
+void AnnotateNoOp(const char *file, int line,
+                  const volatile void *arg){}
+void AnnotateFlushState(const char *file, int line){}
+
+#endif  /* DYNAMIC_ANNOTATIONS_ENABLED == 1
+    && DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 */
+
+#if DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0
+
+static int GetRunningOnValgrind(void) {
+#ifdef RUNNING_ON_VALGRIND
+  if (RUNNING_ON_VALGRIND) return 1;
+#endif
+#ifdef _MSC_VER
+  /* Visual Studio can complain about getenv, so use a windows equivalent. */
+  char value[100] = "1";    /* something that is not "0" */
+  int res = GetEnvironmentVariableA("RUNNING_ON_VALGRIND",
+                                    value, sizeof(value));
+  /* value will remain "1" if the called failed for some reason. */
+  return (res > 0 && strcmp(value, "0") != 0);
+#else
+  /* TODO(csilvers): use GetenvBeforeMain() instead?  Will need to
+   *                 change it to be extern "C".
+   */
+  char *running_on_valgrind_str = getenv("RUNNING_ON_VALGRIND");
+  if (running_on_valgrind_str) {
+    return strcmp(running_on_valgrind_str, "0") != 0;
+  }
+  return 0;
+#endif
+}
+
+/* See the comments in dynamic_annotations.h */
+int RunningOnValgrind(void) {
+  static volatile int running_on_valgrind = -1;
+  int local_running_on_valgrind = running_on_valgrind;
+  /* C doesn't have thread-safe initialization of statics, and we
+     don't want to depend on pthread_once here, so hack it. */
+  ANNOTATE_BENIGN_RACE(&running_on_valgrind, "safe hack");
+  if (local_running_on_valgrind == -1)
+    running_on_valgrind = local_running_on_valgrind = GetRunningOnValgrind();
+  return local_running_on_valgrind;
+}
+
+#endif  /* DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 */
+
+/* See the comments in dynamic_annotations.h */
+double ValgrindSlowdown(void) {
+  /* Same initialization hack as in RunningOnValgrind(). */
+  static volatile double slowdown = 0.0;
+  double local_slowdown = slowdown;
+  ANNOTATE_BENIGN_RACE(&slowdown, "safe hack");
+  if (RunningOnValgrind() == 0) {
+    return 1.0;
+  }
+  if (local_slowdown == 0.0) {
+    char *env = getenv("VALGRIND_SLOWDOWN");
+    slowdown = local_slowdown = env ? atof(env) : 50.0;
+  }
+  return local_slowdown;
+}

BeefRT/gperftools/src/base/dynamic_annotations.h

@@ -0,0 +1,627 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Kostya Serebryany
+ */
+
+/* This file defines dynamic annotations for use with dynamic analysis
+   tool such as valgrind, PIN, etc.
+
+   Dynamic annotation is a source code annotation that affects
+   the generated code (that is, the annotation is not a comment).
+   Each such annotation is attached to a particular
+   instruction and/or to a particular object (address) in the program.
+
+   The annotations that should be used by users are macros in all upper-case
+   (e.g., ANNOTATE_NEW_MEMORY).
+
+   Actual implementation of these macros may differ depending on the
+   dynamic analysis tool being used.
+
+   See http://code.google.com/p/data-race-test/  for more information.
+
+   This file supports the following dynamic analysis tools:
+   - None (DYNAMIC_ANNOTATIONS_ENABLED is not defined or zero).
+      Macros are defined empty.
+   - ThreadSanitizer, Helgrind, DRD (DYNAMIC_ANNOTATIONS_ENABLED is 1).
+      Macros are defined as calls to non-inlinable empty functions
+      that are intercepted by Valgrind. */
+
+#ifndef BASE_DYNAMIC_ANNOTATIONS_H_
+#define BASE_DYNAMIC_ANNOTATIONS_H_
+
+#ifndef DYNAMIC_ANNOTATIONS_ENABLED
+# define DYNAMIC_ANNOTATIONS_ENABLED 0
+#endif
+
+#if DYNAMIC_ANNOTATIONS_ENABLED != 0
+
+  /* -------------------------------------------------------------
+     Annotations useful when implementing condition variables such as CondVar,
+     using conditional critical sections (Await/LockWhen) and when constructing
+     user-defined synchronization mechanisms.
+
+     The annotations ANNOTATE_HAPPENS_BEFORE() and ANNOTATE_HAPPENS_AFTER() can
+     be used to define happens-before arcs in user-defined synchronization
+     mechanisms:  the race detector will infer an arc from the former to the
+     latter when they share the same argument pointer.
+
+     Example 1 (reference counting):
+
+     void Unref() {
+       ANNOTATE_HAPPENS_BEFORE(&refcount_);
+       if (AtomicDecrementByOne(&refcount_) == 0) {
+         ANNOTATE_HAPPENS_AFTER(&refcount_);
+         delete this;
+       }
+     }
+
+     Example 2 (message queue):
+
+     void MyQueue::Put(Type *e) {
+       MutexLock lock(&mu_);
+       ANNOTATE_HAPPENS_BEFORE(e);
+       PutElementIntoMyQueue(e);
+     }
+
+     Type *MyQueue::Get() {
+       MutexLock lock(&mu_);
+       Type *e = GetElementFromMyQueue();
+       ANNOTATE_HAPPENS_AFTER(e);
+       return e;
+     }
+
+     Note: when possible, please use the existing reference counting and message
+     queue implementations instead of inventing new ones. */
+
+  /* Report that wait on the condition variable at address "cv" has succeeded
+     and the lock at address "lock" is held. */
+  #define ANNOTATE_CONDVAR_LOCK_WAIT(cv, lock) \
+    AnnotateCondVarWait(__FILE__, __LINE__, cv, lock)
+
+  /* Report that wait on the condition variable at "cv" has succeeded.  Variant
+     w/o lock. */
+  #define ANNOTATE_CONDVAR_WAIT(cv) \
+    AnnotateCondVarWait(__FILE__, __LINE__, cv, NULL)
+
+  /* Report that we are about to signal on the condition variable at address
+     "cv". */
+  #define ANNOTATE_CONDVAR_SIGNAL(cv) \
+    AnnotateCondVarSignal(__FILE__, __LINE__, cv)
+
+  /* Report that we are about to signal_all on the condition variable at "cv". */
+  #define ANNOTATE_CONDVAR_SIGNAL_ALL(cv) \
+    AnnotateCondVarSignalAll(__FILE__, __LINE__, cv)
+
+  /* Annotations for user-defined synchronization mechanisms. */
+  #define ANNOTATE_HAPPENS_BEFORE(obj) ANNOTATE_CONDVAR_SIGNAL(obj)
+  #define ANNOTATE_HAPPENS_AFTER(obj)  ANNOTATE_CONDVAR_WAIT(obj)
+
+  /* Report that the bytes in the range [pointer, pointer+size) are about
+     to be published safely. The race checker will create a happens-before
+     arc from the call ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size) to
+     subsequent accesses to this memory.
+     Note: this annotation may not work properly if the race detector uses
+     sampling, i.e. does not observe all memory accesses.
+     */
+  #define ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size) \
+    AnnotatePublishMemoryRange(__FILE__, __LINE__, pointer, size)
+
+  /* DEPRECATED. Don't use it. */
+  #define ANNOTATE_UNPUBLISH_MEMORY_RANGE(pointer, size) \
+    AnnotateUnpublishMemoryRange(__FILE__, __LINE__, pointer, size)
+
+  /* DEPRECATED. Don't use it. */
+  #define ANNOTATE_SWAP_MEMORY_RANGE(pointer, size)   \
+    do {                                              \
+      ANNOTATE_UNPUBLISH_MEMORY_RANGE(pointer, size); \
+      ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size);   \
+    } while (0)
+
+  /* Instruct the tool to create a happens-before arc between mu->Unlock() and
+     mu->Lock(). This annotation may slow down the race detector and hide real
+     races. Normally it is used only when it would be difficult to annotate each
+     of the mutex's critical sections individually using the annotations above.
+     This annotation makes sense only for hybrid race detectors. For pure
+     happens-before detectors this is a no-op. For more details see
+     http://code.google.com/p/data-race-test/wiki/PureHappensBeforeVsHybrid . */
+  #define ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX(mu) \
+    AnnotateMutexIsUsedAsCondVar(__FILE__, __LINE__, mu)
+
+  /* Deprecated. Use ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX. */
+  #define ANNOTATE_MUTEX_IS_USED_AS_CONDVAR(mu) \
+    AnnotateMutexIsUsedAsCondVar(__FILE__, __LINE__, mu)
+
+  /* -------------------------------------------------------------
+     Annotations useful when defining memory allocators, or when memory that
+     was protected in one way starts to be protected in another. */
+
+  /* Report that a new memory at "address" of size "size" has been allocated.
+     This might be used when the memory has been retrieved from a free list and
+     is about to be reused, or when a the locking discipline for a variable
+     changes. */
+  #define ANNOTATE_NEW_MEMORY(address, size) \
+    AnnotateNewMemory(__FILE__, __LINE__, address, size)
+
+  /* -------------------------------------------------------------
+     Annotations useful when defining FIFO queues that transfer data between
+     threads. */
+
+  /* Report that the producer-consumer queue (such as ProducerConsumerQueue) at
+     address "pcq" has been created.  The ANNOTATE_PCQ_* annotations
+     should be used only for FIFO queues.  For non-FIFO queues use
+     ANNOTATE_HAPPENS_BEFORE (for put) and ANNOTATE_HAPPENS_AFTER (for get). */
+  #define ANNOTATE_PCQ_CREATE(pcq) \
+    AnnotatePCQCreate(__FILE__, __LINE__, pcq)
+
+  /* Report that the queue at address "pcq" is about to be destroyed. */
+  #define ANNOTATE_PCQ_DESTROY(pcq) \
+    AnnotatePCQDestroy(__FILE__, __LINE__, pcq)
+
+  /* Report that we are about to put an element into a FIFO queue at address
+     "pcq". */
+  #define ANNOTATE_PCQ_PUT(pcq) \
+    AnnotatePCQPut(__FILE__, __LINE__, pcq)
+
+  /* Report that we've just got an element from a FIFO queue at address "pcq". */
+  #define ANNOTATE_PCQ_GET(pcq) \
+    AnnotatePCQGet(__FILE__, __LINE__, pcq)
+
+  /* -------------------------------------------------------------
+     Annotations that suppress errors.  It is usually better to express the
+     program's synchronization using the other annotations, but these can
+     be used when all else fails. */
+
+  /* Report that we may have a benign race at "pointer", with size
+     "sizeof(*(pointer))". "pointer" must be a non-void* pointer.  Insert at the
+     point where "pointer" has been allocated, preferably close to the point
+     where the race happens.  See also ANNOTATE_BENIGN_RACE_STATIC. */
+  #define ANNOTATE_BENIGN_RACE(pointer, description) \
+    AnnotateBenignRaceSized(__FILE__, __LINE__, pointer, \
+                            sizeof(*(pointer)), description)
+
+  /* Same as ANNOTATE_BENIGN_RACE(address, description), but applies to
+     the memory range [address, address+size). */
+  #define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) \
+    AnnotateBenignRaceSized(__FILE__, __LINE__, address, size, description)
+
+  /* Request the analysis tool to ignore all reads in the current thread
+     until ANNOTATE_IGNORE_READS_END is called.
+     Useful to ignore intentional racey reads, while still checking
+     other reads and all writes.
+     See also ANNOTATE_UNPROTECTED_READ. */
+  #define ANNOTATE_IGNORE_READS_BEGIN() \
+    AnnotateIgnoreReadsBegin(__FILE__, __LINE__)
+
+  /* Stop ignoring reads. */
+  #define ANNOTATE_IGNORE_READS_END() \
+    AnnotateIgnoreReadsEnd(__FILE__, __LINE__)
+
+  /* Similar to ANNOTATE_IGNORE_READS_BEGIN, but ignore writes. */
+  #define ANNOTATE_IGNORE_WRITES_BEGIN() \
+    AnnotateIgnoreWritesBegin(__FILE__, __LINE__)
+
+  /* Stop ignoring writes. */
+  #define ANNOTATE_IGNORE_WRITES_END() \
+    AnnotateIgnoreWritesEnd(__FILE__, __LINE__)
+
+  /* Start ignoring all memory accesses (reads and writes). */
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() \
+    do {\
+      ANNOTATE_IGNORE_READS_BEGIN();\
+      ANNOTATE_IGNORE_WRITES_BEGIN();\
+    }while(0)\
+
+  /* Stop ignoring all memory accesses. */
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_END() \
+    do {\
+      ANNOTATE_IGNORE_WRITES_END();\
+      ANNOTATE_IGNORE_READS_END();\
+    }while(0)\
+
+  /* Enable (enable!=0) or disable (enable==0) race detection for all threads.
+     This annotation could be useful if you want to skip expensive race analysis
+     during some period of program execution, e.g. during initialization. */
+  #define ANNOTATE_ENABLE_RACE_DETECTION(enable) \
+    AnnotateEnableRaceDetection(__FILE__, __LINE__, enable)
+
+  /* -------------------------------------------------------------
+     Annotations useful for debugging. */
+
+  /* Request to trace every access to "address". */
+  #define ANNOTATE_TRACE_MEMORY(address) \
+    AnnotateTraceMemory(__FILE__, __LINE__, address)
+
+  /* Report the current thread name to a race detector. */
+  #define ANNOTATE_THREAD_NAME(name) \
+    AnnotateThreadName(__FILE__, __LINE__, name)
+
+  /* -------------------------------------------------------------
+     Annotations useful when implementing locks.  They are not
+     normally needed by modules that merely use locks.
+     The "lock" argument is a pointer to the lock object. */
+
+  /* Report that a lock has been created at address "lock". */
+  #define ANNOTATE_RWLOCK_CREATE(lock) \
+    AnnotateRWLockCreate(__FILE__, __LINE__, lock)
+
+  /* Report that the lock at address "lock" is about to be destroyed. */
+  #define ANNOTATE_RWLOCK_DESTROY(lock) \
+    AnnotateRWLockDestroy(__FILE__, __LINE__, lock)
+
+  /* Report that the lock at address "lock" has been acquired.
+     is_w=1 for writer lock, is_w=0 for reader lock. */
+  #define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) \
+    AnnotateRWLockAcquired(__FILE__, __LINE__, lock, is_w)
+
+  /* Report that the lock at address "lock" is about to be released. */
+  #define ANNOTATE_RWLOCK_RELEASED(lock, is_w) \
+    AnnotateRWLockReleased(__FILE__, __LINE__, lock, is_w)
+
+  /* -------------------------------------------------------------
+     Annotations useful when implementing barriers.  They are not
+     normally needed by modules that merely use barriers.
+     The "barrier" argument is a pointer to the barrier object. */
+
+  /* Report that the "barrier" has been initialized with initial "count".
+   If 'reinitialization_allowed' is true, initialization is allowed to happen
+   multiple times w/o calling barrier_destroy() */
+  #define ANNOTATE_BARRIER_INIT(barrier, count, reinitialization_allowed) \
+    AnnotateBarrierInit(__FILE__, __LINE__, barrier, count, \
+                        reinitialization_allowed)
+
+  /* Report that we are about to enter barrier_wait("barrier"). */
+  #define ANNOTATE_BARRIER_WAIT_BEFORE(barrier) \
+    AnnotateBarrierWaitBefore(__FILE__, __LINE__, barrier)
+
+  /* Report that we just exited barrier_wait("barrier"). */
+  #define ANNOTATE_BARRIER_WAIT_AFTER(barrier) \
+    AnnotateBarrierWaitAfter(__FILE__, __LINE__, barrier)
+
+  /* Report that the "barrier" has been destroyed. */
+  #define ANNOTATE_BARRIER_DESTROY(barrier) \
+    AnnotateBarrierDestroy(__FILE__, __LINE__, barrier)
+
+  /* -------------------------------------------------------------
+     Annotations useful for testing race detectors. */
+
+  /* Report that we expect a race on the variable at "address".
+     Use only in unit tests for a race detector. */
+  #define ANNOTATE_EXPECT_RACE(address, description) \
+    AnnotateExpectRace(__FILE__, __LINE__, address, description)
+
+  /* A no-op. Insert where you like to test the interceptors. */
+  #define ANNOTATE_NO_OP(arg) \
+    AnnotateNoOp(__FILE__, __LINE__, arg)
+
+  /* Force the race detector to flush its state. The actual effect depends on
+   * the implementation of the detector. */
+  #define ANNOTATE_FLUSH_STATE() \
+    AnnotateFlushState(__FILE__, __LINE__)
+
+
+#else  /* DYNAMIC_ANNOTATIONS_ENABLED == 0 */
+
+  #define ANNOTATE_RWLOCK_CREATE(lock) /* empty */
+  #define ANNOTATE_RWLOCK_DESTROY(lock) /* empty */
+  #define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) /* empty */
+  #define ANNOTATE_RWLOCK_RELEASED(lock, is_w) /* empty */
+  #define ANNOTATE_BARRIER_INIT(barrier, count, reinitialization_allowed) /* */
+  #define ANNOTATE_BARRIER_WAIT_BEFORE(barrier) /* empty */
+  #define ANNOTATE_BARRIER_WAIT_AFTER(barrier) /* empty */
+  #define ANNOTATE_BARRIER_DESTROY(barrier) /* empty */
+  #define ANNOTATE_CONDVAR_LOCK_WAIT(cv, lock) /* empty */
+  #define ANNOTATE_CONDVAR_WAIT(cv) /* empty */
+  #define ANNOTATE_CONDVAR_SIGNAL(cv) /* empty */
+  #define ANNOTATE_CONDVAR_SIGNAL_ALL(cv) /* empty */
+  #define ANNOTATE_HAPPENS_BEFORE(obj) /* empty */
+  #define ANNOTATE_HAPPENS_AFTER(obj) /* empty */
+  #define ANNOTATE_PUBLISH_MEMORY_RANGE(address, size) /* empty */
+  #define ANNOTATE_UNPUBLISH_MEMORY_RANGE(address, size)  /* empty */
+  #define ANNOTATE_SWAP_MEMORY_RANGE(address, size)  /* empty */
+  #define ANNOTATE_PCQ_CREATE(pcq) /* empty */
+  #define ANNOTATE_PCQ_DESTROY(pcq) /* empty */
+  #define ANNOTATE_PCQ_PUT(pcq) /* empty */
+  #define ANNOTATE_PCQ_GET(pcq) /* empty */
+  #define ANNOTATE_NEW_MEMORY(address, size) /* empty */
+  #define ANNOTATE_EXPECT_RACE(address, description) /* empty */
+  #define ANNOTATE_BENIGN_RACE(address, description) /* empty */
+  #define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) /* empty */
+  #define ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX(mu) /* empty */
+  #define ANNOTATE_MUTEX_IS_USED_AS_CONDVAR(mu) /* empty */
+  #define ANNOTATE_TRACE_MEMORY(arg) /* empty */
+  #define ANNOTATE_THREAD_NAME(name) /* empty */
+  #define ANNOTATE_IGNORE_READS_BEGIN() /* empty */
+  #define ANNOTATE_IGNORE_READS_END() /* empty */
+  #define ANNOTATE_IGNORE_WRITES_BEGIN() /* empty */
+  #define ANNOTATE_IGNORE_WRITES_END() /* empty */
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() /* empty */
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_END() /* empty */
+  #define ANNOTATE_ENABLE_RACE_DETECTION(enable) /* empty */
+  #define ANNOTATE_NO_OP(arg) /* empty */
+  #define ANNOTATE_FLUSH_STATE() /* empty */
+
+#endif  /* DYNAMIC_ANNOTATIONS_ENABLED */
+
+/* Macro definitions for GCC attributes that allow static thread safety
+   analysis to recognize and use some of the dynamic annotations as
+   escape hatches.
+   TODO(lcwu): remove the check for __SUPPORT_DYN_ANNOTATION__ once the
+   default crosstool/GCC supports these GCC attributes.  */
+
+#define ANNOTALYSIS_STATIC_INLINE
+#define ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY ;
+#define ANNOTALYSIS_IGNORE_READS_BEGIN
+#define ANNOTALYSIS_IGNORE_READS_END
+#define ANNOTALYSIS_IGNORE_WRITES_BEGIN
+#define ANNOTALYSIS_IGNORE_WRITES_END
+#define ANNOTALYSIS_UNPROTECTED_READ
+
+#if defined(__GNUC__) && (!defined(SWIG)) && (!defined(__clang__)) && \
+    defined(__SUPPORT_TS_ANNOTATION__) && defined(__SUPPORT_DYN_ANNOTATION__)
+
+#if DYNAMIC_ANNOTATIONS_ENABLED == 0
+#define ANNOTALYSIS_ONLY 1
+#undef ANNOTALYSIS_STATIC_INLINE
+#define ANNOTALYSIS_STATIC_INLINE static inline
+#undef ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+#define ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY { (void)file; (void)line; }
+#endif
+
+/* Only emit attributes when annotalysis is enabled. */
+#if defined(__SUPPORT_TS_ANNOTATION__) && defined(__SUPPORT_DYN_ANNOTATION__)
+#undef  ANNOTALYSIS_IGNORE_READS_BEGIN
+#define ANNOTALYSIS_IGNORE_READS_BEGIN  __attribute__ ((ignore_reads_begin))
+#undef  ANNOTALYSIS_IGNORE_READS_END
+#define ANNOTALYSIS_IGNORE_READS_END    __attribute__ ((ignore_reads_end))
+#undef  ANNOTALYSIS_IGNORE_WRITES_BEGIN
+#define ANNOTALYSIS_IGNORE_WRITES_BEGIN __attribute__ ((ignore_writes_begin))
+#undef  ANNOTALYSIS_IGNORE_WRITES_END
+#define ANNOTALYSIS_IGNORE_WRITES_END   __attribute__ ((ignore_writes_end))
+#undef  ANNOTALYSIS_UNPROTECTED_READ
+#define ANNOTALYSIS_UNPROTECTED_READ    __attribute__ ((unprotected_read))
+#endif
+
+#endif // defined(__GNUC__) && (!defined(SWIG)) && (!defined(__clang__))
+
+/* Use the macros above rather than using these functions directly. */
+#ifdef __cplusplus
+extern "C" {
+#endif
+void AnnotateRWLockCreate(const char *file, int line,
+                          const volatile void *lock);
+void AnnotateRWLockDestroy(const char *file, int line,
+                           const volatile void *lock);
+void AnnotateRWLockAcquired(const char *file, int line,
+                            const volatile void *lock, long is_w);
+void AnnotateRWLockReleased(const char *file, int line,
+                            const volatile void *lock, long is_w);
+void AnnotateBarrierInit(const char *file, int line,
+                         const volatile void *barrier, long count,
+                         long reinitialization_allowed);
+void AnnotateBarrierWaitBefore(const char *file, int line,
+                               const volatile void *barrier);
+void AnnotateBarrierWaitAfter(const char *file, int line,
+                              const volatile void *barrier);
+void AnnotateBarrierDestroy(const char *file, int line,
+                            const volatile void *barrier);
+void AnnotateCondVarWait(const char *file, int line,
+                         const volatile void *cv,
+                         const volatile void *lock);
+void AnnotateCondVarSignal(const char *file, int line,
+                           const volatile void *cv);
+void AnnotateCondVarSignalAll(const char *file, int line,
+                              const volatile void *cv);
+void AnnotatePublishMemoryRange(const char *file, int line,
+                                const volatile void *address,
+                                long size);
+void AnnotateUnpublishMemoryRange(const char *file, int line,
+                                  const volatile void *address,
+                                  long size);
+void AnnotatePCQCreate(const char *file, int line,
+                       const volatile void *pcq);
+void AnnotatePCQDestroy(const char *file, int line,
+                        const volatile void *pcq);
+void AnnotatePCQPut(const char *file, int line,
+                    const volatile void *pcq);
+void AnnotatePCQGet(const char *file, int line,
+                    const volatile void *pcq);
+void AnnotateNewMemory(const char *file, int line,
+                       const volatile void *address,
+                       long size);
+void AnnotateExpectRace(const char *file, int line,
+                        const volatile void *address,
+                        const char *description);
+void AnnotateBenignRace(const char *file, int line,
+                        const volatile void *address,
+                        const char *description);
+void AnnotateBenignRaceSized(const char *file, int line,
+                        const volatile void *address,
+                        long size,
+                        const char *description);
+void AnnotateMutexIsUsedAsCondVar(const char *file, int line,
+                                  const volatile void *mu);
+void AnnotateTraceMemory(const char *file, int line,
+                         const volatile void *arg);
+void AnnotateThreadName(const char *file, int line,
+                        const char *name);
+ANNOTALYSIS_STATIC_INLINE
+void AnnotateIgnoreReadsBegin(const char *file, int line)
+    ANNOTALYSIS_IGNORE_READS_BEGIN ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+ANNOTALYSIS_STATIC_INLINE
+void AnnotateIgnoreReadsEnd(const char *file, int line)
+    ANNOTALYSIS_IGNORE_READS_END ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+ANNOTALYSIS_STATIC_INLINE
+void AnnotateIgnoreWritesBegin(const char *file, int line)
+    ANNOTALYSIS_IGNORE_WRITES_BEGIN ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+ANNOTALYSIS_STATIC_INLINE
+void AnnotateIgnoreWritesEnd(const char *file, int line)
+    ANNOTALYSIS_IGNORE_WRITES_END ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+void AnnotateEnableRaceDetection(const char *file, int line, int enable);
+void AnnotateNoOp(const char *file, int line,
+                  const volatile void *arg);
+void AnnotateFlushState(const char *file, int line);
+
+/* Return non-zero value if running under valgrind.
+
+  If "valgrind.h" is included into dynamic_annotations.c,
+  the regular valgrind mechanism will be used.
+  See http://valgrind.org/docs/manual/manual-core-adv.html about
+  RUNNING_ON_VALGRIND and other valgrind "client requests".
+  The file "valgrind.h" may be obtained by doing
+     svn co svn://svn.valgrind.org/valgrind/trunk/include
+
+  If for some reason you can't use "valgrind.h" or want to fake valgrind,
+  there are two ways to make this function return non-zero:
+    - Use environment variable: export RUNNING_ON_VALGRIND=1
+    - Make your tool intercept the function RunningOnValgrind() and
+      change its return value.
+ */
+int RunningOnValgrind(void);
+
+/* ValgrindSlowdown returns:
+    * 1.0, if (RunningOnValgrind() == 0)
+    * 50.0, if (RunningOnValgrind() != 0 && getenv("VALGRIND_SLOWDOWN") == NULL)
+    * atof(getenv("VALGRIND_SLOWDOWN")) otherwise
+   This function can be used to scale timeout values:
+   EXAMPLE:
+   for (;;) {
+     DoExpensiveBackgroundTask();
+     SleepForSeconds(5 * ValgrindSlowdown());
+   }
+ */
+double ValgrindSlowdown(void);
+
+#ifdef __cplusplus
+}
+#endif
+
+#if DYNAMIC_ANNOTATIONS_ENABLED != 0 && defined(__cplusplus)
+
+  /* ANNOTATE_UNPROTECTED_READ is the preferred way to annotate racey reads.
+
+     Instead of doing
+        ANNOTATE_IGNORE_READS_BEGIN();
+        ... = x;
+        ANNOTATE_IGNORE_READS_END();
+     one can use
+        ... = ANNOTATE_UNPROTECTED_READ(x); */
+  template <class T>
+  inline T ANNOTATE_UNPROTECTED_READ(const volatile T &x)
+      ANNOTALYSIS_UNPROTECTED_READ {
+    ANNOTATE_IGNORE_READS_BEGIN();
+    T res = x;
+    ANNOTATE_IGNORE_READS_END();
+    return res;
+  }
+  /* Apply ANNOTATE_BENIGN_RACE_SIZED to a static variable. */
+  #define ANNOTATE_BENIGN_RACE_STATIC(static_var, description)        \
+    namespace {                                                       \
+      class static_var ## _annotator {                                \
+       public:                                                        \
+        static_var ## _annotator() {                                  \
+          ANNOTATE_BENIGN_RACE_SIZED(&static_var,                     \
+                                      sizeof(static_var),             \
+            # static_var ": " description);                           \
+        }                                                             \
+      };                                                              \
+      static static_var ## _annotator the ## static_var ## _annotator;\
+    }
+#else /* DYNAMIC_ANNOTATIONS_ENABLED == 0 */
+
+  #define ANNOTATE_UNPROTECTED_READ(x) (x)
+  #define ANNOTATE_BENIGN_RACE_STATIC(static_var, description)  /* empty */
+
+#endif /* DYNAMIC_ANNOTATIONS_ENABLED */
+
+/* Annotalysis, a GCC based static analyzer, is able to understand and use
+   some of the dynamic annotations defined in this file. However, dynamic
+   annotations are usually disabled in the opt mode (to avoid additional
+   runtime overheads) while Annotalysis only works in the opt mode.
+   In order for Annotalysis to use these dynamic annotations when they
+   are disabled, we re-define these annotations here. Note that unlike the
+   original macro definitions above, these macros are expanded to calls to
+   static inline functions so that the compiler will be able to remove the
+   calls after the analysis. */
+
+#ifdef ANNOTALYSIS_ONLY
+
+  #undef ANNOTALYSIS_ONLY
+
+  /* Undefine and re-define the macros that the static analyzer understands. */
+  #undef ANNOTATE_IGNORE_READS_BEGIN
+  #define ANNOTATE_IGNORE_READS_BEGIN()           \
+    AnnotateIgnoreReadsBegin(__FILE__, __LINE__)
+
+  #undef ANNOTATE_IGNORE_READS_END
+  #define ANNOTATE_IGNORE_READS_END()             \
+    AnnotateIgnoreReadsEnd(__FILE__, __LINE__)
+
+  #undef ANNOTATE_IGNORE_WRITES_BEGIN
+  #define ANNOTATE_IGNORE_WRITES_BEGIN()          \
+    AnnotateIgnoreWritesBegin(__FILE__, __LINE__)
+
+  #undef ANNOTATE_IGNORE_WRITES_END
+  #define ANNOTATE_IGNORE_WRITES_END()            \
+    AnnotateIgnoreWritesEnd(__FILE__, __LINE__)
+
+  #undef ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN()       \
+    do {                                                 \
+      ANNOTATE_IGNORE_READS_BEGIN();                     \
+      ANNOTATE_IGNORE_WRITES_BEGIN();                    \
+    }while(0)                                            \
+
+  #undef ANNOTATE_IGNORE_READS_AND_WRITES_END
+  #define ANNOTATE_IGNORE_READS_AND_WRITES_END()  \
+    do {                                          \
+      ANNOTATE_IGNORE_WRITES_END();               \
+      ANNOTATE_IGNORE_READS_END();                \
+    }while(0)                                     \
+
+  #if defined(__cplusplus)
+    #undef ANNOTATE_UNPROTECTED_READ
+    template <class T>
+    inline T ANNOTATE_UNPROTECTED_READ(const volatile T &x)
+         ANNOTALYSIS_UNPROTECTED_READ {
+      ANNOTATE_IGNORE_READS_BEGIN();
+      T res = x;
+      ANNOTATE_IGNORE_READS_END();
+      return res;
+    }
+  #endif /* __cplusplus */
+
+#endif /* ANNOTALYSIS_ONLY */
+
+/* Undefine the macros intended only in this file. */
+#undef ANNOTALYSIS_STATIC_INLINE
+#undef ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY
+
+#endif  /* BASE_DYNAMIC_ANNOTATIONS_H_ */

BeefRT/gperftools/src/base/elf_mem_image.cc

@@ -0,0 +1,433 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Pluzhnikov
+//
+// Allow dynamic symbol lookup in an in-memory Elf image.
+//
+
+#include "base/elf_mem_image.h"
+
+#ifdef HAVE_ELF_MEM_IMAGE  // defined in elf_mem_image.h
+
+#include <stddef.h>   // for size_t, ptrdiff_t
+#include "base/logging.h"
+
+// From binutils/include/elf/common.h (this doesn't appear to be documented
+// anywhere else).
+//
+//   /* This flag appears in a Versym structure.  It means that the symbol
+//      is hidden, and is only visible with an explicit version number.
+//      This is a GNU extension.  */
+//   #define VERSYM_HIDDEN           0x8000
+//
+//   /* This is the mask for the rest of the Versym information.  */
+//   #define VERSYM_VERSION          0x7fff
+
+#define VERSYM_VERSION 0x7fff
+
+namespace base {
+
+namespace {
+template <int N> class ElfClass {
+ public:
+  static const int kElfClass = -1;
+  static int ElfBind(const ElfW(Sym) *) {
+    CHECK(false); // << "Unexpected word size";
+    return 0;
+  }
+  static int ElfType(const ElfW(Sym) *) {
+    CHECK(false); // << "Unexpected word size";
+    return 0;
+  }
+};
+
+template <> class ElfClass<32> {
+ public:
+  static const int kElfClass = ELFCLASS32;
+  static int ElfBind(const ElfW(Sym) *symbol) {
+    return ELF32_ST_BIND(symbol->st_info);
+  }
+  static int ElfType(const ElfW(Sym) *symbol) {
+    return ELF32_ST_TYPE(symbol->st_info);
+  }
+};
+
+template <> class ElfClass<64> {
+ public:
+  static const int kElfClass = ELFCLASS64;
+  static int ElfBind(const ElfW(Sym) *symbol) {
+    return ELF64_ST_BIND(symbol->st_info);
+  }
+  static int ElfType(const ElfW(Sym) *symbol) {
+    return ELF64_ST_TYPE(symbol->st_info);
+  }
+};
+
+typedef ElfClass<__WORDSIZE> CurrentElfClass;
+
+// Extract an element from one of the ELF tables, cast it to desired type.
+// This is just a simple arithmetic and a glorified cast.
+// Callers are responsible for bounds checking.
+template <class T>
+const T* GetTableElement(const ElfW(Ehdr) *ehdr,
+                         ElfW(Off) table_offset,
+                         ElfW(Word) element_size,
+                         size_t index) {
+  return reinterpret_cast<const T*>(reinterpret_cast<const char *>(ehdr)
+                                    + table_offset
+                                    + index * element_size);
+}
+}  // namespace
+
+const void *const ElfMemImage::kInvalidBase =
+    reinterpret_cast<const void *>(~0L);
+
+ElfMemImage::ElfMemImage(const void *base) {
+  CHECK(base != kInvalidBase);
+  Init(base);
+}
+
+int ElfMemImage::GetNumSymbols() const {
+  if (!hash_) {
+    return 0;
+  }
+  // See http://www.caldera.com/developers/gabi/latest/ch5.dynamic.html#hash
+  return hash_[1];
+}
+
+const ElfW(Sym) *ElfMemImage::GetDynsym(int index) const {
+  CHECK_LT(index, GetNumSymbols());
+  return dynsym_ + index;
+}
+
+const ElfW(Versym) *ElfMemImage::GetVersym(int index) const {
+  CHECK_LT(index, GetNumSymbols());
+  return versym_ + index;
+}
+
+const ElfW(Phdr) *ElfMemImage::GetPhdr(int index) const {
+  CHECK_LT(index, ehdr_->e_phnum);
+  return GetTableElement<ElfW(Phdr)>(ehdr_,
+                                     ehdr_->e_phoff,
+                                     ehdr_->e_phentsize,
+                                     index);
+}
+
+const char *ElfMemImage::GetDynstr(ElfW(Word) offset) const {
+  CHECK_LT(offset, strsize_);
+  return dynstr_ + offset;
+}
+
+const void *ElfMemImage::GetSymAddr(const ElfW(Sym) *sym) const {
+  if (sym->st_shndx == SHN_UNDEF || sym->st_shndx >= SHN_LORESERVE) {
+    // Symbol corresponds to "special" (e.g. SHN_ABS) section.
+    return reinterpret_cast<const void *>(sym->st_value);
+  }
+  CHECK_LT(link_base_, sym->st_value);
+  return GetTableElement<char>(ehdr_, 0, 1, sym->st_value) - link_base_;
+}
+
+const ElfW(Verdef) *ElfMemImage::GetVerdef(int index) const {
+  CHECK_LE(index, verdefnum_);
+  const ElfW(Verdef) *version_definition = verdef_;
+  while (version_definition->vd_ndx < index && version_definition->vd_next) {
+    const char *const version_definition_as_char =
+        reinterpret_cast<const char *>(version_definition);
+    version_definition =
+        reinterpret_cast<const ElfW(Verdef) *>(version_definition_as_char +
+                                               version_definition->vd_next);
+  }
+  return version_definition->vd_ndx == index ? version_definition : NULL;
+}
+
+const ElfW(Verdaux) *ElfMemImage::GetVerdefAux(
+    const ElfW(Verdef) *verdef) const {
+  return reinterpret_cast<const ElfW(Verdaux) *>(verdef+1);
+}
+
+const char *ElfMemImage::GetVerstr(ElfW(Word) offset) const {
+  CHECK_LT(offset, strsize_);
+  return dynstr_ + offset;
+}
+
+void ElfMemImage::Init(const void *base) {
+  ehdr_      = NULL;
+  dynsym_    = NULL;
+  dynstr_    = NULL;
+  versym_    = NULL;
+  verdef_    = NULL;
+  hash_      = NULL;
+  strsize_   = 0;
+  verdefnum_ = 0;
+  link_base_ = ~0L;  // Sentinel: PT_LOAD .p_vaddr can't possibly be this.
+  if (!base) {
+    return;
+  }
+  const intptr_t base_as_uintptr_t = reinterpret_cast<uintptr_t>(base);
+  // Fake VDSO has low bit set.
+  const bool fake_vdso = ((base_as_uintptr_t & 1) != 0);
+  base = reinterpret_cast<const void *>(base_as_uintptr_t & ~1);
+  const char *const base_as_char = reinterpret_cast<const char *>(base);
+  if (base_as_char[EI_MAG0] != ELFMAG0 || base_as_char[EI_MAG1] != ELFMAG1 ||
+      base_as_char[EI_MAG2] != ELFMAG2 || base_as_char[EI_MAG3] != ELFMAG3) {
+    RAW_DCHECK(false, "no ELF magic"); // at %p", base);
+    return;
+  }
+  int elf_class = base_as_char[EI_CLASS];
+  if (elf_class != CurrentElfClass::kElfClass) {
+    DCHECK_EQ(elf_class, CurrentElfClass::kElfClass);
+    return;
+  }
+  switch (base_as_char[EI_DATA]) {
+    case ELFDATA2LSB: {
+      if (__LITTLE_ENDIAN != __BYTE_ORDER) {
+        DCHECK_EQ(__LITTLE_ENDIAN, __BYTE_ORDER); // << ": wrong byte order";
+        return;
+      }
+      break;
+    }
+    case ELFDATA2MSB: {
+      if (__BIG_ENDIAN != __BYTE_ORDER) {
+        DCHECK_EQ(__BIG_ENDIAN, __BYTE_ORDER); // << ": wrong byte order";
+        return;
+      }
+      break;
+    }
+    default: {
+      RAW_DCHECK(false, "unexpected data encoding"); // << base_as_char[EI_DATA];
+      return;
+    }
+  }
+
+  ehdr_ = reinterpret_cast<const ElfW(Ehdr) *>(base);
+  const ElfW(Phdr) *dynamic_program_header = NULL;
+  for (int i = 0; i < ehdr_->e_phnum; ++i) {
+    const ElfW(Phdr) *const program_header = GetPhdr(i);
+    switch (program_header->p_type) {
+      case PT_LOAD:
+        if (link_base_ == ~0L) {
+          link_base_ = program_header->p_vaddr;
+        }
+        break;
+      case PT_DYNAMIC:
+        dynamic_program_header = program_header;
+        break;
+    }
+  }
+  if (link_base_ == ~0L || !dynamic_program_header) {
+    RAW_DCHECK(~0L != link_base_, "no PT_LOADs in VDSO");
+    RAW_DCHECK(dynamic_program_header, "no PT_DYNAMIC in VDSO");
+    // Mark this image as not present. Can not recur infinitely.
+    Init(0);
+    return;
+  }
+  ptrdiff_t relocation =
+      base_as_char - reinterpret_cast<const char *>(link_base_);
+  ElfW(Dyn) *dynamic_entry =
+      reinterpret_cast<ElfW(Dyn) *>(dynamic_program_header->p_vaddr +
+                                    relocation);
+  for (; dynamic_entry->d_tag != DT_NULL; ++dynamic_entry) {
+    ElfW(Xword) value = dynamic_entry->d_un.d_val;
+    if (fake_vdso) {
+      // A complication: in the real VDSO, dynamic entries are not relocated
+      // (it wasn't loaded by a dynamic loader). But when testing with a
+      // "fake" dlopen()ed vdso library, the loader relocates some (but
+      // not all!) of them before we get here.
+      if (dynamic_entry->d_tag == DT_VERDEF) {
+        // The only dynamic entry (of the ones we care about) libc-2.3.6
+        // loader doesn't relocate.
+        value += relocation;
+      }
+    } else {
+      // Real VDSO. Everything needs to be relocated.
+      value += relocation;
+    }
+    switch (dynamic_entry->d_tag) {
+      case DT_HASH:
+        hash_ = reinterpret_cast<ElfW(Word) *>(value);
+        break;
+      case DT_SYMTAB:
+        dynsym_ = reinterpret_cast<ElfW(Sym) *>(value);
+        break;
+      case DT_STRTAB:
+        dynstr_ = reinterpret_cast<const char *>(value);
+        break;
+      case DT_VERSYM:
+        versym_ = reinterpret_cast<ElfW(Versym) *>(value);
+        break;
+      case DT_VERDEF:
+        verdef_ = reinterpret_cast<ElfW(Verdef) *>(value);
+        break;
+      case DT_VERDEFNUM:
+        verdefnum_ = dynamic_entry->d_un.d_val;
+        break;
+      case DT_STRSZ:
+        strsize_ = dynamic_entry->d_un.d_val;
+        break;
+      default:
+        // Unrecognized entries explicitly ignored.
+        break;
+    }
+  }
+  if (!hash_ || !dynsym_ || !dynstr_ || !versym_ ||
+      !verdef_ || !verdefnum_ || !strsize_) {
+    RAW_DCHECK(hash_, "invalid VDSO (no DT_HASH)");
+    RAW_DCHECK(dynsym_, "invalid VDSO (no DT_SYMTAB)");
+    RAW_DCHECK(dynstr_, "invalid VDSO (no DT_STRTAB)");
+    RAW_DCHECK(versym_, "invalid VDSO (no DT_VERSYM)");
+    RAW_DCHECK(verdef_, "invalid VDSO (no DT_VERDEF)");
+    RAW_DCHECK(verdefnum_, "invalid VDSO (no DT_VERDEFNUM)");
+    RAW_DCHECK(strsize_, "invalid VDSO (no DT_STRSZ)");
+    // Mark this image as not present. Can not recur infinitely.
+    Init(0);
+    return;
+  }
+}
+
+bool ElfMemImage::LookupSymbol(const char *name,
+                               const char *version,
+                               int type,
+                               SymbolInfo *info) const {
+  for (SymbolIterator it = begin(); it != end(); ++it) {
+    if (strcmp(it->name, name) == 0 && strcmp(it->version, version) == 0 &&
+        CurrentElfClass::ElfType(it->symbol) == type) {
+      if (info) {
+        *info = *it;
+      }
+      return true;
+    }
+  }
+  return false;
+}
+
+bool ElfMemImage::LookupSymbolByAddress(const void *address,
+                                        SymbolInfo *info_out) const {
+  for (SymbolIterator it = begin(); it != end(); ++it) {
+    const char *const symbol_start =
+        reinterpret_cast<const char *>(it->address);
+    const char *const symbol_end = symbol_start + it->symbol->st_size;
+    if (symbol_start <= address && address < symbol_end) {
+      if (info_out) {
+        // Client wants to know details for that symbol (the usual case).
+        if (CurrentElfClass::ElfBind(it->symbol) == STB_GLOBAL) {
+          // Strong symbol; just return it.
+          *info_out = *it;
+          return true;
+        } else {
+          // Weak or local. Record it, but keep looking for a strong one.
+          *info_out = *it;
+        }
+      } else {
+        // Client only cares if there is an overlapping symbol.
+        return true;
+      }
+    }
+  }
+  return false;
+}
+
+ElfMemImage::SymbolIterator::SymbolIterator(const void *const image, int index)
+    : index_(index), image_(image) {
+}
+
+const ElfMemImage::SymbolInfo *ElfMemImage::SymbolIterator::operator->() const {
+  return &info_;
+}
+
+const ElfMemImage::SymbolInfo& ElfMemImage::SymbolIterator::operator*() const {
+  return info_;
+}
+
+bool ElfMemImage::SymbolIterator::operator==(const SymbolIterator &rhs) const {
+  return this->image_ == rhs.image_ && this->index_ == rhs.index_;
+}
+
+bool ElfMemImage::SymbolIterator::operator!=(const SymbolIterator &rhs) const {
+  return !(*this == rhs);
+}
+
+ElfMemImage::SymbolIterator &ElfMemImage::SymbolIterator::operator++() {
+  this->Update(1);
+  return *this;
+}
+
+ElfMemImage::SymbolIterator ElfMemImage::begin() const {
+  SymbolIterator it(this, 0);
+  it.Update(0);
+  return it;
+}
+
+ElfMemImage::SymbolIterator ElfMemImage::end() const {
+  return SymbolIterator(this, GetNumSymbols());
+}
+
+void ElfMemImage::SymbolIterator::Update(int increment) {
+  const ElfMemImage *image = reinterpret_cast<const ElfMemImage *>(image_);
+  CHECK(image->IsPresent() || increment == 0);
+  if (!image->IsPresent()) {
+    return;
+  }
+  index_ += increment;
+  if (index_ >= image->GetNumSymbols()) {
+    index_ = image->GetNumSymbols();
+    return;
+  }
+  const ElfW(Sym)    *symbol = image->GetDynsym(index_);
+  const ElfW(Versym) *version_symbol = image->GetVersym(index_);
+  CHECK(symbol && version_symbol);
+  const char *const symbol_name = image->GetDynstr(symbol->st_name);
+  const ElfW(Versym) version_index = version_symbol[0] & VERSYM_VERSION;
+  const ElfW(Verdef) *version_definition = NULL;
+  const char *version_name = "";
+  if (symbol->st_shndx == SHN_UNDEF) {
+    // Undefined symbols reference DT_VERNEED, not DT_VERDEF, and
+    // version_index could well be greater than verdefnum_, so calling
+    // GetVerdef(version_index) may trigger assertion.
+  } else {
+    version_definition = image->GetVerdef(version_index);
+  }
+  if (version_definition) {
+    // I am expecting 1 or 2 auxiliary entries: 1 for the version itself,
+    // optional 2nd if the version has a parent.
+    CHECK_LE(1, version_definition->vd_cnt);
+    CHECK_LE(version_definition->vd_cnt, 2);
+    const ElfW(Verdaux) *version_aux = image->GetVerdefAux(version_definition);
+    version_name = image->GetVerstr(version_aux->vda_name);
+  }
+  info_.name    = symbol_name;
+  info_.version = version_name;
+  info_.address = image->GetSymAddr(symbol);
+  info_.symbol  = symbol;
+}
+
+}  // namespace base
+
+#endif  // HAVE_ELF_MEM_IMAGE

BeefRT/gperftools/src/base/elf_mem_image.h

@@ -0,0 +1,134 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Pluzhnikov
+//
+// Allow dynamic symbol lookup for in-memory Elf images.
+
+#ifndef BASE_ELF_MEM_IMAGE_H_
+#define BASE_ELF_MEM_IMAGE_H_
+
+#include <config.h>
+#ifdef HAVE_FEATURES_H
+#include <features.h>   // for __GLIBC__
+#endif
+
+// Maybe one day we can rewrite this file not to require the elf
+// symbol extensions in glibc, but for right now we need them.
+#if defined(__ELF__) && defined(__GLIBC__) && !defined(__native_client__)
+
+#define HAVE_ELF_MEM_IMAGE 1
+
+#include <stdlib.h>
+#include <link.h>  // for ElfW
+
+namespace base {
+
+// An in-memory ELF image (may not exist on disk).
+class ElfMemImage {
+ public:
+  // Sentinel: there could never be an elf image at this address.
+  static const void *const kInvalidBase;
+
+  // Information about a single vdso symbol.
+  // All pointers are into .dynsym, .dynstr, or .text of the VDSO.
+  // Do not free() them or modify through them.
+  struct SymbolInfo {
+    const char      *name;      // E.g. "__vdso_getcpu"
+    const char      *version;   // E.g. "LINUX_2.6", could be ""
+                                // for unversioned symbol.
+    const void      *address;   // Relocated symbol address.
+    const ElfW(Sym) *symbol;    // Symbol in the dynamic symbol table.
+  };
+
+  // Supports iteration over all dynamic symbols.
+  class SymbolIterator {
+   public:
+    friend class ElfMemImage;
+    const SymbolInfo *operator->() const;
+    const SymbolInfo &operator*() const;
+    SymbolIterator& operator++();
+    bool operator!=(const SymbolIterator &rhs) const;
+    bool operator==(const SymbolIterator &rhs) const;
+   private:
+    SymbolIterator(const void *const image, int index);
+    void Update(int incr);
+    SymbolInfo info_;
+    int index_;
+    const void *const image_;
+  };
+
+
+  explicit ElfMemImage(const void *base);
+  void                 Init(const void *base);
+  bool                 IsPresent() const { return ehdr_ != NULL; }
+  const ElfW(Phdr)*    GetPhdr(int index) const;
+  const ElfW(Sym)*     GetDynsym(int index) const;
+  const ElfW(Versym)*  GetVersym(int index) const;
+  const ElfW(Verdef)*  GetVerdef(int index) const;
+  const ElfW(Verdaux)* GetVerdefAux(const ElfW(Verdef) *verdef) const;
+  const char*          GetDynstr(ElfW(Word) offset) const;
+  const void*          GetSymAddr(const ElfW(Sym) *sym) const;
+  const char*          GetVerstr(ElfW(Word) offset) const;
+  int                  GetNumSymbols() const;
+
+  SymbolIterator begin() const;
+  SymbolIterator end() const;
+
+  // Look up versioned dynamic symbol in the image.
+  // Returns false if image is not present, or doesn't contain given
+  // symbol/version/type combination.
+  // If info_out != NULL, additional details are filled in.
+  bool LookupSymbol(const char *name, const char *version,
+                    int symbol_type, SymbolInfo *info_out) const;
+
+  // Find info about symbol (if any) which overlaps given address.
+  // Returns true if symbol was found; false if image isn't present
+  // or doesn't have a symbol overlapping given address.
+  // If info_out != NULL, additional details are filled in.
+  bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const;
+
+ private:
+  const ElfW(Ehdr) *ehdr_;
+  const ElfW(Sym) *dynsym_;
+  const ElfW(Versym) *versym_;
+  const ElfW(Verdef) *verdef_;
+  const ElfW(Word) *hash_;
+  const char *dynstr_;
+  size_t strsize_;
+  size_t verdefnum_;
+  ElfW(Addr) link_base_;     // Link-time base (p_vaddr of first PT_LOAD).
+};
+
+}  // namespace base
+
+#endif  // __ELF__ and __GLIBC__ and !__native_client__
+
+#endif  // BASE_ELF_MEM_IMAGE_H_

BeefRT/gperftools/src/base/elfcore.h

@@ -0,0 +1,400 @@
+/* Copyright (c) 2005-2008, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Markus Gutschke, Carl Crous
+ */
+
+#ifndef _ELFCORE_H
+#define _ELFCORE_H
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* We currently only support x86-32, x86-64, ARM, MIPS, PPC on Linux.
+ * Porting to other related platforms should not be difficult.
+ */
+#if (defined(__i386__) || defined(__x86_64__) || defined(__ARM_ARCH_3__) || \
+     defined(__mips__) || defined(__PPC__)) && defined(__linux)
+
+#include <stdarg.h>
+#include <stdint.h>
+#include <sys/types.h>
+#include <config.h>
+
+
+/* Define the DUMPER symbol to make sure that there is exactly one
+ * core dumper built into the library.
+ */
+#define DUMPER "ELF"
+
+/* By the time that we get a chance to read CPU registers in the
+ * calling thread, they are already in a not particularly useful
+ * state. Besides, there will be multiple frames on the stack that are
+ * just making the core file confusing. To fix this problem, we take a
+ * snapshot of the frame pointer, stack pointer, and instruction
+ * pointer at an earlier time, and then insert these values into the
+ * core file.
+ */
+
+#if defined(__i386__) || defined(__x86_64__)
+  typedef struct i386_regs {    /* Normal (non-FPU) CPU registers            */
+  #ifdef __x86_64__
+    #define BP rbp
+    #define SP rsp
+    #define IP rip
+    uint64_t  r15,r14,r13,r12,rbp,rbx,r11,r10;
+    uint64_t  r9,r8,rax,rcx,rdx,rsi,rdi,orig_rax;
+    uint64_t  rip,cs,eflags;
+    uint64_t  rsp,ss;
+    uint64_t  fs_base, gs_base;
+    uint64_t  ds,es,fs,gs;
+  #else
+    #define BP ebp
+    #define SP esp
+    #define IP eip
+    uint32_t  ebx, ecx, edx, esi, edi, ebp, eax;
+    uint16_t  ds, __ds, es, __es;
+    uint16_t  fs, __fs, gs, __gs;
+    uint32_t  orig_eax, eip;
+    uint16_t  cs, __cs;
+    uint32_t  eflags, esp;
+    uint16_t  ss, __ss;
+  #endif
+  } i386_regs;
+#elif defined(__ARM_ARCH_3__)
+  typedef struct arm_regs {     /* General purpose registers                 */
+    #define BP uregs[11]        /* Frame pointer                             */
+    #define SP uregs[13]        /* Stack pointer                             */
+    #define IP uregs[15]        /* Program counter                           */
+    #define LR uregs[14]        /* Link register                             */
+    long uregs[18];
+  } arm_regs;
+#elif defined(__mips__)
+  typedef struct mips_regs {
+    unsigned long pad[6];       /* Unused padding to match kernel structures */
+    unsigned long uregs[32];    /* General purpose registers.                */
+    unsigned long hi;           /* Used for multiplication and division.     */
+    unsigned long lo;
+    unsigned long cp0_epc;      /* Program counter.                          */
+    unsigned long cp0_badvaddr;
+    unsigned long cp0_status;
+    unsigned long cp0_cause;
+    unsigned long unused;
+  } mips_regs;
+#elif defined (__PPC__)
+  typedef struct ppc_regs {
+    #define SP uregs[1]         /* Stack pointer                             */
+    #define IP rip              /* Program counter                           */
+    #define LR lr               /* Link register                             */
+    unsigned long uregs[32];	/* General Purpose Registers - r0-r31.       */
+    double        fpr[32];	/* Floating-Point Registers - f0-f31.        */
+    unsigned long rip;		/* Program counter.                          */
+    unsigned long msr;
+    unsigned long ccr;
+    unsigned long lr;
+    unsigned long ctr;
+    unsigned long xeq;
+    unsigned long mq;
+  } ppc_regs;
+#endif
+
+#if defined(__i386__) && defined(__GNUC__)
+  /* On x86 we provide an optimized version of the FRAME() macro, if the
+   * compiler supports a GCC-style asm() directive. This results in somewhat
+   * more accurate values for CPU registers.
+   */
+  typedef struct Frame {
+    struct i386_regs uregs;
+    int              errno_;
+    pid_t            tid;
+  } Frame;
+  #define FRAME(f) Frame f;                                           \
+                   do {                                               \
+                     f.errno_ = errno;                                \
+                     f.tid    = sys_gettid();                         \
+                     __asm__ volatile (                               \
+                       "push %%ebp\n"                                 \
+                       "push %%ebx\n"                                 \
+                       "mov  %%ebx,0(%%eax)\n"                        \
+                       "mov  %%ecx,4(%%eax)\n"                        \
+                       "mov  %%edx,8(%%eax)\n"                        \
+                       "mov  %%esi,12(%%eax)\n"                       \
+                       "mov  %%edi,16(%%eax)\n"                       \
+                       "mov  %%ebp,20(%%eax)\n"                       \
+                       "mov  %%eax,24(%%eax)\n"                       \
+                       "mov  %%ds,%%ebx\n"                            \
+                       "mov  %%ebx,28(%%eax)\n"                       \
+                       "mov  %%es,%%ebx\n"                            \
+                       "mov  %%ebx,32(%%eax)\n"                       \
+                       "mov  %%fs,%%ebx\n"                            \
+                       "mov  %%ebx,36(%%eax)\n"                       \
+                       "mov  %%gs,%%ebx\n"                            \
+                       "mov  %%ebx, 40(%%eax)\n"                      \
+                       "call 0f\n"                                    \
+                     "0:pop %%ebx\n"                                  \
+                       "add  $1f-0b,%%ebx\n"                          \
+                       "mov  %%ebx,48(%%eax)\n"                       \
+                       "mov  %%cs,%%ebx\n"                            \
+                       "mov  %%ebx,52(%%eax)\n"                       \
+                       "pushf\n"                                      \
+                       "pop  %%ebx\n"                                 \
+                       "mov  %%ebx,56(%%eax)\n"                       \
+                       "mov  %%esp,%%ebx\n"                           \
+                       "add  $8,%%ebx\n"                              \
+                       "mov  %%ebx,60(%%eax)\n"                       \
+                       "mov  %%ss,%%ebx\n"                            \
+                       "mov  %%ebx,64(%%eax)\n"                       \
+                       "pop  %%ebx\n"                                 \
+                       "pop  %%ebp\n"                                 \
+                     "1:"                                             \
+                       : : "a" (&f) : "memory");                      \
+                     } while (0)
+  #define SET_FRAME(f,r)                                              \
+                     do {                                             \
+                       errno = (f).errno_;                            \
+                       (r)   = (f).uregs;                             \
+                     } while (0)
+#elif defined(__x86_64__) && defined(__GNUC__)
+  /* The FRAME and SET_FRAME macros for x86_64.  */
+  typedef struct Frame {
+    struct i386_regs uregs;
+    int              errno_;
+    pid_t            tid;
+  } Frame;
+  #define FRAME(f) Frame f;                                           \
+                   do {                                               \
+                     f.errno_ = errno;                                \
+                     f.tid    = sys_gettid();                         \
+                     __asm__ volatile (                               \
+                       "push %%rbp\n"                                 \
+                       "push %%rbx\n"                                 \
+                       "mov  %%r15,0(%%rax)\n"                        \
+                       "mov  %%r14,8(%%rax)\n"                        \
+                       "mov  %%r13,16(%%rax)\n"                       \
+                       "mov  %%r12,24(%%rax)\n"                       \
+                       "mov  %%rbp,32(%%rax)\n"                       \
+                       "mov  %%rbx,40(%%rax)\n"                       \
+                       "mov  %%r11,48(%%rax)\n"                       \
+                       "mov  %%r10,56(%%rax)\n"                       \
+                       "mov  %%r9,64(%%rax)\n"                        \
+                       "mov  %%r8,72(%%rax)\n"                        \
+                       "mov  %%rax,80(%%rax)\n"                       \
+                       "mov  %%rcx,88(%%rax)\n"                       \
+                       "mov  %%rdx,96(%%rax)\n"                       \
+                       "mov  %%rsi,104(%%rax)\n"                      \
+                       "mov  %%rdi,112(%%rax)\n"                      \
+                       "mov  %%ds,%%rbx\n"                            \
+                       "mov  %%rbx,184(%%rax)\n"                      \
+                       "mov  %%es,%%rbx\n"                            \
+                       "mov  %%rbx,192(%%rax)\n"                      \
+                       "mov  %%fs,%%rbx\n"                            \
+                       "mov  %%rbx,200(%%rax)\n"                      \
+                       "mov  %%gs,%%rbx\n"                            \
+                       "mov  %%rbx,208(%%rax)\n"                      \
+                       "call 0f\n"                                    \
+                     "0:pop %%rbx\n"                                  \
+                       "add  $1f-0b,%%rbx\n"                          \
+                       "mov  %%rbx,128(%%rax)\n"                      \
+                       "mov  %%cs,%%rbx\n"                            \
+                       "mov  %%rbx,136(%%rax)\n"                      \
+                       "pushf\n"                                      \
+                       "pop  %%rbx\n"                                 \
+                       "mov  %%rbx,144(%%rax)\n"                      \
+                       "mov  %%rsp,%%rbx\n"                           \
+                       "add  $16,%%ebx\n"                             \
+                       "mov  %%rbx,152(%%rax)\n"                      \
+                       "mov  %%ss,%%rbx\n"                            \
+                       "mov  %%rbx,160(%%rax)\n"                      \
+                       "pop  %%rbx\n"                                 \
+                       "pop  %%rbp\n"                                 \
+                     "1:"                                             \
+                       : : "a" (&f) : "memory");                      \
+                     } while (0)
+  #define SET_FRAME(f,r)                                              \
+                     do {                                             \
+                       errno = (f).errno_;                            \
+                       (f).uregs.fs_base = (r).fs_base;               \
+                       (f).uregs.gs_base = (r).gs_base;               \
+                       (r)   = (f).uregs;                             \
+                     } while (0)
+#elif defined(__ARM_ARCH_3__) && defined(__GNUC__)
+  /* ARM calling conventions are a little more tricky. A little assembly
+   * helps in obtaining an accurate snapshot of all registers.
+   */
+  typedef struct Frame {
+    struct arm_regs arm;
+    int             errno_;
+    pid_t           tid;
+  } Frame;
+  #define FRAME(f) Frame f;                                           \
+                   do {                                               \
+                     long cpsr;                                       \
+                     f.errno_ = errno;                                \
+                     f.tid    = sys_gettid();                         \
+                     __asm__ volatile(                                \
+                       "stmia %0, {r0-r15}\n" /* All integer regs   */\
+                       : : "r"(&f.arm) : "memory");                   \
+                     f.arm.uregs[16] = 0;                             \
+                     __asm__ volatile(                                \
+                       "mrs %0, cpsr\n"       /* Condition code reg */\
+                       : "=r"(cpsr));                                 \
+                     f.arm.uregs[17] = cpsr;                          \
+                   } while (0)
+  #define SET_FRAME(f,r)                                              \
+                     do {                                             \
+                       /* Don't override the FPU status register.   */\
+                       /* Use the value obtained from ptrace(). This*/\
+                       /* works, because our code does not perform  */\
+                       /* any FPU operations, itself.               */\
+                       long fps      = (f).arm.uregs[16];             \
+                       errno         = (f).errno_;                    \
+                       (r)           = (f).arm;                       \
+                       (r).uregs[16] = fps;                           \
+                     } while (0)
+#elif defined(__mips__) && defined(__GNUC__)
+  typedef struct Frame {
+    struct mips_regs mips_regs;
+    int              errno_;
+    pid_t            tid;
+  } Frame;
+  #define MIPSREG(n) ({ register unsigned long r __asm__("$"#n); r; })
+  #define FRAME(f) Frame f = { 0 };                                   \
+                   do {                                               \
+                     unsigned long hi, lo;                            \
+                     register unsigned long pc __asm__("$31");        \
+                     f.mips_regs.uregs[ 0] = MIPSREG( 0);             \
+                     f.mips_regs.uregs[ 1] = MIPSREG( 1);             \
+                     f.mips_regs.uregs[ 2] = MIPSREG( 2);             \
+                     f.mips_regs.uregs[ 3] = MIPSREG( 3);             \
+                     f.mips_regs.uregs[ 4] = MIPSREG( 4);             \
+                     f.mips_regs.uregs[ 5] = MIPSREG( 5);             \
+                     f.mips_regs.uregs[ 6] = MIPSREG( 6);             \
+                     f.mips_regs.uregs[ 7] = MIPSREG( 7);             \
+                     f.mips_regs.uregs[ 8] = MIPSREG( 8);             \
+                     f.mips_regs.uregs[ 9] = MIPSREG( 9);             \
+                     f.mips_regs.uregs[10] = MIPSREG(10);             \
+                     f.mips_regs.uregs[11] = MIPSREG(11);             \
+                     f.mips_regs.uregs[12] = MIPSREG(12);             \
+                     f.mips_regs.uregs[13] = MIPSREG(13);             \
+                     f.mips_regs.uregs[14] = MIPSREG(14);             \
+                     f.mips_regs.uregs[15] = MIPSREG(15);             \
+                     f.mips_regs.uregs[16] = MIPSREG(16);             \
+                     f.mips_regs.uregs[17] = MIPSREG(17);             \
+                     f.mips_regs.uregs[18] = MIPSREG(18);             \
+                     f.mips_regs.uregs[19] = MIPSREG(19);             \
+                     f.mips_regs.uregs[20] = MIPSREG(20);             \
+                     f.mips_regs.uregs[21] = MIPSREG(21);             \
+                     f.mips_regs.uregs[22] = MIPSREG(22);             \
+                     f.mips_regs.uregs[23] = MIPSREG(23);             \
+                     f.mips_regs.uregs[24] = MIPSREG(24);             \
+                     f.mips_regs.uregs[25] = MIPSREG(25);             \
+                     f.mips_regs.uregs[26] = MIPSREG(26);             \
+                     f.mips_regs.uregs[27] = MIPSREG(27);             \
+                     f.mips_regs.uregs[28] = MIPSREG(28);             \
+                     f.mips_regs.uregs[29] = MIPSREG(29);             \
+                     f.mips_regs.uregs[30] = MIPSREG(30);             \
+                     f.mips_regs.uregs[31] = MIPSREG(31);             \
+                     __asm__ volatile ("mfhi %0" : "=r"(hi));         \
+                     __asm__ volatile ("mflo %0" : "=r"(lo));         \
+                     __asm__ volatile ("jal 1f; 1:nop" : "=r"(pc));   \
+                     f.mips_regs.hi       = hi;                       \
+                     f.mips_regs.lo       = lo;                       \
+                     f.mips_regs.cp0_epc  = pc;                       \
+                     f.errno_             = errno;                    \
+                     f.tid                = sys_gettid();             \
+                   } while (0)
+  #define SET_FRAME(f,r)                                              \
+                   do {                                               \
+                     errno       = (f).errno_;                        \
+                     memcpy((r).uregs, (f).mips_regs.uregs,           \
+                            32*sizeof(unsigned long));                \
+                     (r).hi      = (f).mips_regs.hi;                  \
+                     (r).lo      = (f).mips_regs.lo;                  \
+                     (r).cp0_epc = (f).mips_regs.cp0_epc;             \
+                   } while (0)
+#else
+  /* If we do not have a hand-optimized assembly version of the FRAME()
+   * macro, we cannot reliably unroll the stack. So, we show a few additional
+   * stack frames for the coredumper.
+   */
+  typedef struct Frame {
+    pid_t tid;
+  } Frame;
+  #define FRAME(f) Frame f; do { f.tid = sys_gettid(); } while (0)
+  #define SET_FRAME(f,r) do { } while (0)
+#endif
+
+
+/* Internal function for generating a core file. This API can change without
+ * notice and is only supposed to be used internally by the core dumper.
+ *
+ * This function works for both single- and multi-threaded core
+ * dumps. If called as
+ *
+ *   FRAME(frame);
+ *   InternalGetCoreDump(&frame, 0, NULL, ap);
+ *
+ * it creates a core file that only contains information about the
+ * calling thread.
+ *
+ * Optionally, the caller can provide information about other threads
+ * by passing their process ids in "thread_pids". The process id of
+ * the caller should not be included in this array. All of the threads
+ * must have been attached to with ptrace(), prior to calling this
+ * function. They will be detached when "InternalGetCoreDump()" returns.
+ *
+ * This function either returns a file handle that can be read for obtaining
+ * a core dump, or "-1" in case of an error. In the latter case, "errno"
+ * will be set appropriately.
+ *
+ * While "InternalGetCoreDump()" is not technically async signal safe, you
+ * might be tempted to invoke it from a signal handler. The code goes to
+ * great lengths to make a best effort that this will actually work. But in
+ * any case, you must make sure that you preserve the value of "errno"
+ * yourself. It is guaranteed to be clobbered otherwise.
+ *
+ * Also, "InternalGetCoreDump" is not strictly speaking re-entrant. Again,
+ * it makes a best effort to behave reasonably when called in a multi-
+ * threaded environment, but it is ultimately the caller's responsibility
+ * to provide locking.
+ */
+int InternalGetCoreDump(void *frame, int num_threads, pid_t *thread_pids,
+                        va_list ap
+                     /* const struct CoreDumpParameters *params,
+                        const char *file_name,
+                        const char *PATH
+                      */);
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* _ELFCORE_H */

BeefRT/gperftools/src/base/googleinit.h

@@ -0,0 +1,73 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Jacob Hoffman-Andrews
+
+#ifndef _GOOGLEINIT_H
+#define _GOOGLEINIT_H
+
+#include "base/logging.h"
+
+class GoogleInitializer {
+ public:
+  typedef void (*VoidFunction)(void);
+  GoogleInitializer(const char* name, VoidFunction ctor, VoidFunction dtor)
+      : name_(name), destructor_(dtor) {
+    RAW_VLOG(10, "<GoogleModuleObject> constructing: %s\n", name_);
+    if (ctor)
+      ctor();
+  }
+  ~GoogleInitializer() {
+    RAW_VLOG(10, "<GoogleModuleObject> destroying: %s\n", name_);
+    if (destructor_)
+      destructor_();
+  }
+
+ private:
+  const char* const name_;
+  const VoidFunction destructor_;
+};
+
+#define REGISTER_MODULE_INITIALIZER(name, body)                 \
+  namespace {                                                   \
+    static void google_init_module_##name () { body; }          \
+    GoogleInitializer google_initializer_module_##name(#name,   \
+            google_init_module_##name, NULL);                   \
+  }
+
+#define REGISTER_MODULE_DESTRUCTOR(name, body)                  \
+  namespace {                                                   \
+    static void google_destruct_module_##name () { body; }      \
+    GoogleInitializer google_destructor_module_##name(#name,    \
+            NULL, google_destruct_module_##name);               \
+  }
+
+
+#endif /* _GOOGLEINIT_H */

BeefRT/gperftools/src/base/linuxthreads.cc

@@ -0,0 +1,706 @@
+/* Copyright (c) 2005-2007, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Markus Gutschke
+ */
+
+#include "base/linuxthreads.h"
+
+#ifdef THREADS
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <sched.h>
+#include <signal.h>
+#include <stdlib.h>
+#include <string.h>
+#include <fcntl.h>
+#include <sys/socket.h>
+#include <sys/wait.h>
+#include <sys/prctl.h>
+#include <semaphore.h>
+
+#include "base/linux_syscall_support.h"
+#include "base/thread_lister.h"
+
+#ifndef CLONE_UNTRACED
+#define CLONE_UNTRACED 0x00800000
+#endif
+
+
+/* Synchronous signals that should not be blocked while in the lister thread.
+ */
+static const int sync_signals[]  = { SIGABRT, SIGILL, SIGFPE, SIGSEGV, SIGBUS,
+                                     SIGXCPU, SIGXFSZ };
+
+/* itoa() is not a standard function, and we cannot safely call printf()
+ * after suspending threads. So, we just implement our own copy. A
+ * recursive approach is the easiest here.
+ */
+static char *local_itoa(char *buf, int i) {
+  if (i < 0) {
+    *buf++ = '-';
+    return local_itoa(buf, -i);
+  } else {
+    if (i >= 10)
+      buf = local_itoa(buf, i/10);
+    *buf++ = (i%10) + '0';
+    *buf   = '\000';
+    return buf;
+  }
+}
+
+
+/* Wrapper around clone() that runs "fn" on the same stack as the
+ * caller! Unlike fork(), the cloned thread shares the same address space.
+ * The caller must be careful to use only minimal amounts of stack until
+ * the cloned thread has returned.
+ * There is a good chance that the cloned thread and the caller will share
+ * the same copy of errno!
+ */
+#ifdef __GNUC__
+#if __GNUC__ == 3 && __GNUC_MINOR__ >= 1 || __GNUC__ > 3
+/* Try to force this function into a separate stack frame, and make sure
+ * that arguments are passed on the stack.
+ */
+static int local_clone (int (*fn)(void *), void *arg, ...)
+  __attribute__ ((noinline));
+#endif
+#endif
+
+/* To avoid the gap cross page boundaries, increase by the large parge
+ * size mostly PowerPC system uses.  */
+#ifdef __PPC64__
+#define CLONE_STACK_SIZE 65536
+#else
+#define CLONE_STACK_SIZE 4096
+#endif
+
+static int local_clone (int (*fn)(void *), void *arg, ...) {
+  /* Leave 4kB of gap between the callers stack and the new clone. This
+   * should be more than sufficient for the caller to call waitpid() until
+   * the cloned thread terminates.
+   *
+   * It is important that we set the CLONE_UNTRACED flag, because newer
+   * versions of "gdb" otherwise attempt to attach to our thread, and will
+   * attempt to reap its status codes. This subsequently results in the
+   * caller hanging indefinitely in waitpid(), waiting for a change in
+   * status that will never happen. By setting the CLONE_UNTRACED flag, we
+   * prevent "gdb" from stealing events, but we still expect the thread
+   * lister to fail, because it cannot PTRACE_ATTACH to the process that
+   * is being debugged. This is OK and the error code will be reported
+   * correctly.
+   */
+  return sys_clone(fn, (char *)&arg - CLONE_STACK_SIZE,
+                   CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_UNTRACED, arg, 0, 0, 0);
+}
+
+
+/* Local substitute for the atoi() function, which is not necessarily safe
+ * to call once threads are suspended (depending on whether libc looks up
+ * locale information,  when executing atoi()).
+ */
+static int local_atoi(const char *s) {
+  int n   = 0;
+  int neg = *s == '-';
+  if (neg)
+    s++;
+  while (*s >= '0' && *s <= '9')
+    n = 10*n + (*s++ - '0');
+  return neg ? -n : n;
+}
+
+
+/* Re-runs fn until it doesn't cause EINTR
+ */
+#define NO_INTR(fn)   do {} while ((fn) < 0 && errno == EINTR)
+
+
+/* Wrap a class around system calls, in order to give us access to
+ * a private copy of errno. This only works in C++, but it has the
+ * advantage of not needing nested functions, which are a non-standard
+ * language extension.
+ */
+#ifdef __cplusplus
+namespace {
+  class SysCalls {
+   public:
+    #define SYS_CPLUSPLUS
+    #define SYS_ERRNO     my_errno
+    #define SYS_INLINE    inline
+    #define SYS_PREFIX    -1
+    #undef  SYS_LINUX_SYSCALL_SUPPORT_H
+    #include "linux_syscall_support.h"
+    SysCalls() : my_errno(0) { }
+    int my_errno;
+  };
+}
+#define ERRNO sys.my_errno
+#else
+#define ERRNO my_errno
+#endif
+
+
+/* Wrapper for open() which is guaranteed to never return EINTR.
+ */
+static int c_open(const char *fname, int flags, int mode) {
+  ssize_t rc;
+  NO_INTR(rc = sys_open(fname, flags, mode));
+  return rc;
+}
+
+
+/* abort() is not safely reentrant, and changes it's behavior each time
+ * it is called. This means, if the main application ever called abort()
+ * we cannot safely call it again. This would happen if we were called
+ * from a SIGABRT signal handler in the main application. So, document
+ * that calling SIGABRT from the thread lister makes it not signal safe
+ * (and vice-versa).
+ * Also, since we share address space with the main application, we
+ * cannot call abort() from the callback and expect the main application
+ * to behave correctly afterwards. In fact, the only thing we can do, is
+ * to terminate the main application with extreme prejudice (aka
+ * PTRACE_KILL).
+ * We set up our own SIGABRT handler to do this.
+ * In order to find the main application from the signal handler, we
+ * need to store information about it in global variables. This is
+ * safe, because the main application should be suspended at this
+ * time. If the callback ever called ResumeAllProcessThreads(), then
+ * we are running a higher risk, though. So, try to avoid calling
+ * abort() after calling ResumeAllProcessThreads.
+ */
+static volatile int *sig_pids, sig_num_threads, sig_proc, sig_marker;
+
+
+/* Signal handler to help us recover from dying while we are attached to
+ * other threads.
+ */
+static void SignalHandler(int signum, siginfo_t *si, void *data) {
+  if (sig_pids != NULL) {
+    if (signum == SIGABRT) {
+      while (sig_num_threads-- > 0) {
+        /* Not sure if sched_yield is really necessary here, but it does not */
+        /* hurt, and it might be necessary for the same reasons that we have */
+        /* to do so in sys_ptrace_detach().                                  */
+        sys_sched_yield();
+        sys_ptrace(PTRACE_KILL, sig_pids[sig_num_threads], 0, 0);
+      }
+    } else if (sig_num_threads > 0) {
+      ResumeAllProcessThreads(sig_num_threads, (int *)sig_pids);
+    }
+  }
+  sig_pids = NULL;
+  if (sig_marker >= 0)
+    NO_INTR(sys_close(sig_marker));
+  sig_marker = -1;
+  if (sig_proc >= 0)
+    NO_INTR(sys_close(sig_proc));
+  sig_proc = -1;
+
+  sys__exit(signum == SIGABRT ? 1 : 2);
+}
+
+
+/* Try to dirty the stack, and hope that the compiler is not smart enough
+ * to optimize this function away. Or worse, the compiler could inline the
+ * function and permanently allocate the data on the stack.
+ */
+static void DirtyStack(size_t amount) {
+  char buf[amount];
+  memset(buf, 0, amount);
+  sys_read(-1, buf, amount);
+}
+
+
+/* Data structure for passing arguments to the lister thread.
+ */
+#define ALT_STACKSIZE (MINSIGSTKSZ + 4096)
+
+struct ListerParams {
+  int         result, err;
+  char        *altstack_mem;
+  ListAllProcessThreadsCallBack callback;
+  void        *parameter;
+  va_list     ap;
+  sem_t       *lock;
+};
+
+
+static void ListerThread(struct ListerParams *args) {
+  int                found_parent = 0;
+  pid_t              clone_pid  = sys_gettid(), ppid = sys_getppid();
+  char               proc_self_task[80], marker_name[48], *marker_path;
+  const char         *proc_paths[3];
+  const char *const  *proc_path = proc_paths;
+  int                proc = -1, marker = -1, num_threads = 0;
+  int                max_threads = 0, sig;
+  struct kernel_stat marker_sb, proc_sb;
+  stack_t            altstack;
+
+  /* Wait for parent thread to set appropriate permissions
+   * to allow ptrace activity
+   */
+  if (sem_wait(args->lock) < 0) {
+    goto failure;
+  }
+
+  /* Create "marker" that we can use to detect threads sharing the same
+   * address space and the same file handles. By setting the FD_CLOEXEC flag
+   * we minimize the risk of misidentifying child processes as threads;
+   * and since there is still a race condition,  we will filter those out
+   * later, anyway.
+   */
+  if ((marker = sys_socket(PF_LOCAL, SOCK_DGRAM, 0)) < 0 ||
+      sys_fcntl(marker, F_SETFD, FD_CLOEXEC) < 0) {
+  failure:
+    args->result = -1;
+    args->err    = errno;
+    if (marker >= 0)
+      NO_INTR(sys_close(marker));
+    sig_marker = marker = -1;
+    if (proc >= 0)
+      NO_INTR(sys_close(proc));
+    sig_proc = proc = -1;
+    sys__exit(1);
+  }
+
+  /* Compute search paths for finding thread directories in /proc            */
+  local_itoa(strrchr(strcpy(proc_self_task, "/proc/"), '\000'), ppid);
+  strcpy(marker_name, proc_self_task);
+  marker_path = marker_name + strlen(marker_name);
+  strcat(proc_self_task, "/task/");
+  proc_paths[0] = proc_self_task; /* /proc/$$/task/                          */
+  proc_paths[1] = "/proc/";       /* /proc/                                  */
+  proc_paths[2] = NULL;
+
+  /* Compute path for marker socket in /proc                                 */
+  local_itoa(strcpy(marker_path, "/fd/") + 4, marker);
+  if (sys_stat(marker_name, &marker_sb) < 0) {
+    goto failure;
+  }
+
+  /* Catch signals on an alternate pre-allocated stack. This way, we can
+   * safely execute the signal handler even if we ran out of memory.
+   */
+  memset(&altstack, 0, sizeof(altstack));
+  altstack.ss_sp    = args->altstack_mem;
+  altstack.ss_flags = 0;
+  altstack.ss_size  = ALT_STACKSIZE;
+  sys_sigaltstack(&altstack, (const stack_t *)NULL);
+
+  /* Some kernels forget to wake up traced processes, when the
+   * tracer dies.  So, intercept synchronous signals and make sure
+   * that we wake up our tracees before dying. It is the caller's
+   * responsibility to ensure that asynchronous signals do not
+   * interfere with this function.
+   */
+  sig_marker = marker;
+  sig_proc   = -1;
+  for (sig = 0; sig < sizeof(sync_signals)/sizeof(*sync_signals); sig++) {
+    struct kernel_sigaction sa;
+    memset(&sa, 0, sizeof(sa));
+    sa.sa_sigaction_ = SignalHandler;
+    sys_sigfillset(&sa.sa_mask);
+    sa.sa_flags      = SA_ONSTACK|SA_SIGINFO|SA_RESETHAND;
+    sys_sigaction(sync_signals[sig], &sa, (struct kernel_sigaction *)NULL);
+  }
+  
+  /* Read process directories in /proc/...                                   */
+  for (;;) {
+    /* Some kernels know about threads, and hide them in "/proc"
+     * (although they are still there, if you know the process
+     * id). Threads are moved into a separate "task" directory. We
+     * check there first, and then fall back on the older naming
+     * convention if necessary.
+     */
+    if ((sig_proc = proc = c_open(*proc_path, O_RDONLY|O_DIRECTORY, 0)) < 0) {
+      if (*++proc_path != NULL)
+        continue;
+      goto failure;
+    }
+    if (sys_fstat(proc, &proc_sb) < 0)
+      goto failure;
+    
+    /* Since we are suspending threads, we cannot call any libc
+     * functions that might acquire locks. Most notably, we cannot
+     * call malloc(). So, we have to allocate memory on the stack,
+     * instead. Since we do not know how much memory we need, we
+     * make a best guess. And if we guessed incorrectly we retry on
+     * a second iteration (by jumping to "detach_threads").
+     *
+     * Unless the number of threads is increasing very rapidly, we
+     * should never need to do so, though, as our guestimate is very
+     * conservative.
+     */
+    if (max_threads < proc_sb.st_nlink + 100)
+      max_threads = proc_sb.st_nlink + 100;
+    
+    /* scope */ {
+      pid_t pids[max_threads];
+      int   added_entries = 0;
+      sig_num_threads     = num_threads;
+      sig_pids            = pids;
+      for (;;) {
+        struct kernel_dirent *entry;
+        char buf[4096];
+        ssize_t nbytes = sys_getdents(proc, (struct kernel_dirent *)buf,
+                                      sizeof(buf));
+        if (nbytes < 0)
+          goto failure;
+        else if (nbytes == 0) {
+          if (added_entries) {
+            /* Need to keep iterating over "/proc" in multiple
+             * passes until we no longer find any more threads. This
+             * algorithm eventually completes, when all threads have
+             * been suspended.
+             */
+            added_entries = 0;
+            sys_lseek(proc, 0, SEEK_SET);
+            continue;
+          }
+          break;
+        }
+        for (entry = (struct kernel_dirent *)buf;
+             entry < (struct kernel_dirent *)&buf[nbytes];
+             entry = (struct kernel_dirent *)((char *)entry+entry->d_reclen)) {
+          if (entry->d_ino != 0) {
+            const char *ptr = entry->d_name;
+            pid_t pid;
+            
+            /* Some kernels hide threads by preceding the pid with a '.'     */
+            if (*ptr == '.')
+              ptr++;
+            
+            /* If the directory is not numeric, it cannot be a
+             * process/thread
+             */
+            if (*ptr < '0' || *ptr > '9')
+              continue;
+            pid = local_atoi(ptr);
+
+            /* Attach (and suspend) all threads                              */
+            if (pid && pid != clone_pid) {
+              struct kernel_stat tmp_sb;
+              char fname[entry->d_reclen + 48];
+              strcat(strcat(strcpy(fname, "/proc/"),
+                            entry->d_name), marker_path);
+              
+              /* Check if the marker is identical to the one we created      */
+              if (sys_stat(fname, &tmp_sb) >= 0 &&
+                  marker_sb.st_ino == tmp_sb.st_ino) {
+                long i;
+
+                /* Found one of our threads, make sure it is no duplicate    */
+                for (i = 0; i < num_threads; i++) {
+                  /* Linear search is slow, but should not matter much for
+                   * the typically small number of threads.
+                   */
+                  if (pids[i] == pid) {
+                    /* Found a duplicate; most likely on second pass         */
+                    goto next_entry;
+                  }
+                }
+                
+                /* Check whether data structure needs growing                */
+                if (num_threads >= max_threads) {
+                  /* Back to square one, this time with more memory          */
+                  NO_INTR(sys_close(proc));
+                  goto detach_threads;
+                }
+
+                /* Attaching to thread suspends it                           */
+                pids[num_threads++] = pid;
+                sig_num_threads     = num_threads;
+                if (sys_ptrace(PTRACE_ATTACH, pid, (void *)0,
+                               (void *)0) < 0) {
+                  /* If operation failed, ignore thread. Maybe it
+                   * just died?  There might also be a race
+                   * condition with a concurrent core dumper or
+                   * with a debugger. In that case, we will just
+                   * make a best effort, rather than failing
+                   * entirely.
+                   */
+                  num_threads--;
+                  sig_num_threads = num_threads;
+                  goto next_entry;
+                }
+                /* Attaching to a process doesn't guarantee it'll stop before
+                 * ptrace returns; you have to wait on it.  Specifying __WCLONE
+                 * means it will only wait for clone children (i.e. threads,
+                 * not processes).
+                 */
+                while (sys_waitpid(pid, (int *)0, __WCLONE) < 0) {
+                  if (errno != EINTR) {
+                    /* Assumes ECHILD                                        */
+                    if (pid == ppid) {
+                        /* The parent is not a clone                         */
+                        found_parent = true;
+                        break;
+                    } else {
+                        sys_ptrace_detach(pid);
+                        num_threads--;
+                        sig_num_threads = num_threads;
+                        goto next_entry;
+                    }
+                  }
+                }
+
+                added_entries++;
+              }
+            }
+          }
+        next_entry:;
+        }
+      }
+      NO_INTR(sys_close(proc));
+      sig_proc = proc = -1;
+
+      /* If we failed to find any threads, try looking somewhere else in
+       * /proc. Maybe, threads are reported differently on this system.
+       */
+      if (num_threads > 1 || !*++proc_path) {
+        NO_INTR(sys_close(marker));
+        sig_marker = marker = -1;
+
+        /* If we never found the parent process, something is very wrong.
+         * Most likely, we are running in debugger. Any attempt to operate
+         * on the threads would be very incomplete. Let's just report an
+         * error to the caller.
+         */
+        if (!found_parent) {
+          ResumeAllProcessThreads(num_threads, pids);
+          sys__exit(3);
+        }
+
+        /* Now we are ready to call the callback,
+         * which takes care of resuming the threads for us.
+         */
+        args->result = args->callback(args->parameter, num_threads,
+                                      pids, args->ap);
+        args->err = errno;
+
+        /* Callback should have resumed threads, but better safe than sorry  */
+        if (ResumeAllProcessThreads(num_threads, pids)) {
+          /* Callback forgot to resume at least one thread, report error     */
+          args->err    = EINVAL;
+          args->result = -1;
+        }
+
+        sys__exit(0);
+      }
+    detach_threads:
+      /* Resume all threads prior to retrying the operation                  */
+      ResumeAllProcessThreads(num_threads, pids);
+      sig_pids = NULL;
+      num_threads = 0;
+      sig_num_threads = num_threads;
+      max_threads += 100;
+    }
+  }
+}
+
+
+/* This function gets the list of all linux threads of the current process
+ * passes them to the 'callback' along with the 'parameter' pointer; at the
+ * call back call time all the threads are paused via
+ * PTRACE_ATTACH.
+ * The callback is executed from a separate thread which shares only the
+ * address space, the filesystem, and the filehandles with the caller. Most
+ * notably, it does not share the same pid and ppid; and if it terminates,
+ * the rest of the application is still there. 'callback' is supposed to do
+ * or arrange for ResumeAllProcessThreads. This happens automatically, if
+ * the thread raises a synchronous signal (e.g. SIGSEGV); asynchronous
+ * signals are blocked. If the 'callback' decides to unblock them, it must
+ * ensure that they cannot terminate the application, or that
+ * ResumeAllProcessThreads will get called.
+ * It is an error for the 'callback' to make any library calls that could
+ * acquire locks. Most notably, this means that most system calls have to
+ * avoid going through libc. Also, this means that it is not legal to call
+ * exit() or abort().
+ * We return -1 on error and the return value of 'callback' on success.
+ */
+int ListAllProcessThreads(void *parameter,
+                          ListAllProcessThreadsCallBack callback, ...) {
+  char                   altstack_mem[ALT_STACKSIZE];
+  struct ListerParams    args;
+  pid_t                  clone_pid;
+  int                    dumpable = 1, sig;
+  struct kernel_sigset_t sig_blocked, sig_old;
+  sem_t                  lock;
+
+  va_start(args.ap, callback);
+
+  /* If we are short on virtual memory, initializing the alternate stack
+   * might trigger a SIGSEGV. Let's do this early, before it could get us
+   * into more trouble (i.e. before signal handlers try to use the alternate
+   * stack, and before we attach to other threads).
+   */
+  memset(altstack_mem, 0, sizeof(altstack_mem));
+
+  /* Some of our cleanup functions could conceivable use more stack space.
+   * Try to touch the stack right now. This could be defeated by the compiler
+   * being too smart for it's own good, so try really hard.
+   */
+  DirtyStack(32768);
+
+  /* Make this process "dumpable". This is necessary in order to ptrace()
+   * after having called setuid().
+   */
+  dumpable = sys_prctl(PR_GET_DUMPABLE, 0);
+  if (!dumpable)
+    sys_prctl(PR_SET_DUMPABLE, 1);
+
+  /* Fill in argument block for dumper thread                                */
+  args.result       = -1;
+  args.err          = 0;
+  args.altstack_mem = altstack_mem;
+  args.parameter    = parameter;
+  args.callback     = callback;
+  args.lock         = &lock;
+
+  /* Before cloning the thread lister, block all asynchronous signals, as we */
+  /* are not prepared to handle them.                                        */
+  sys_sigfillset(&sig_blocked);
+  for (sig = 0; sig < sizeof(sync_signals)/sizeof(*sync_signals); sig++) {
+    sys_sigdelset(&sig_blocked, sync_signals[sig]);
+  }
+  if (sys_sigprocmask(SIG_BLOCK, &sig_blocked, &sig_old)) {
+    args.err = errno;
+    args.result = -1;
+    goto failed;
+  }
+
+  /* scope */ {
+    /* After cloning, both the parent and the child share the same instance
+     * of errno. We must make sure that at least one of these processes
+     * (in our case, the parent) uses modified syscall macros that update
+     * a local copy of errno, instead.
+     */
+    #ifdef __cplusplus
+      #define sys0_sigprocmask sys.sigprocmask
+      #define sys0_waitpid     sys.waitpid
+      SysCalls sys;
+    #else
+      int my_errno;
+      #define SYS_ERRNO        my_errno
+      #define SYS_INLINE       inline
+      #define SYS_PREFIX       0
+      #undef  SYS_LINUX_SYSCALL_SUPPORT_H
+      #include "linux_syscall_support.h"
+    #endif
+
+    /* Lock before clone so that parent can set
+	 * ptrace permissions (if necessary) prior
+     * to ListerThread actually executing
+     */
+    if (sem_init(&lock, 0, 0) == 0) {
+
+      int clone_errno;
+      clone_pid = local_clone((int (*)(void *))ListerThread, &args);
+      clone_errno = errno;
+
+      sys_sigprocmask(SIG_SETMASK, &sig_old, &sig_old);
+
+      if (clone_pid >= 0) {
+#ifdef PR_SET_PTRACER
+        /* In newer versions of glibc permission must explicitly
+         * be given to allow for ptrace.
+         */
+        prctl(PR_SET_PTRACER, clone_pid, 0, 0, 0);
+#endif
+        /* Releasing the lock here allows the
+         * ListerThread to execute and ptrace us.
+		 */
+        sem_post(&lock);
+        int status, rc;
+        while ((rc = sys0_waitpid(clone_pid, &status, __WALL)) < 0 &&
+               ERRNO == EINTR) {
+                /* Keep waiting                                                 */
+        }
+        if (rc < 0) {
+          args.err = ERRNO;
+          args.result = -1;
+        } else if (WIFEXITED(status)) {
+          switch (WEXITSTATUS(status)) {
+            case 0: break;             /* Normal process termination           */
+            case 2: args.err = EFAULT; /* Some fault (e.g. SIGSEGV) detected   */
+                    args.result = -1;
+                    break;
+            case 3: args.err = EPERM;  /* Process is already being traced      */
+                    args.result = -1;
+                    break;
+            default:args.err = ECHILD; /* Child died unexpectedly              */
+                    args.result = -1;
+                    break;
+          }
+        } else if (!WIFEXITED(status)) {
+          args.err    = EFAULT;        /* Terminated due to an unhandled signal*/
+          args.result = -1;
+        }
+        sem_destroy(&lock);
+      } else {
+        args.result = -1;
+        args.err    = clone_errno;
+      }
+    } else {
+      args.result = -1;
+      args.err    = errno;
+    }
+  }
+
+  /* Restore the "dumpable" state of the process                             */
+failed:
+  if (!dumpable)
+    sys_prctl(PR_SET_DUMPABLE, dumpable);
+
+  va_end(args.ap);
+
+  errno = args.err;
+  return args.result;
+}
+
+/* This function resumes the list of all linux threads that
+ * ListAllProcessThreads pauses before giving to its callback.
+ * The function returns non-zero if at least one thread was
+ * suspended and has now been resumed.
+ */
+int ResumeAllProcessThreads(int num_threads, pid_t *thread_pids) {
+  int detached_at_least_one = 0;
+  while (num_threads-- > 0) {
+    detached_at_least_one |= sys_ptrace_detach(thread_pids[num_threads]) >= 0;
+  }
+  return detached_at_least_one;
+}
+
+#ifdef __cplusplus
+}
+#endif
+#endif

BeefRT/gperftools/src/base/linuxthreads.h

@@ -0,0 +1,53 @@
+/* Copyright (c) 2005-2007, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Markus Gutschke
+ */
+
+#ifndef _LINUXTHREADS_H
+#define _LINUXTHREADS_H
+
+/* Include thread_lister.h to get the interface that we implement for linux.
+ */
+
+/* We currently only support x86-32 and x86-64 on Linux. Porting to other
+ * related platforms should not be difficult.
+ */
+#if (defined(__i386__) || defined(__x86_64__) || defined(__ARM_ARCH_3__) || \
+     defined(__mips__) || defined(__PPC__)) && defined(__linux)
+
+/* Define the THREADS symbol to make sure that there is exactly one core dumper
+ * built into the library.
+ */
+#define THREADS "Linux /proc"
+
+#endif
+
+#endif  /* _LINUXTHREADS_H */

BeefRT/gperftools/src/base/logging.cc

@@ -0,0 +1,107 @@
+// Copyright (c) 2007, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// This file just provides storage for FLAGS_verbose.
+
+#include <config.h>
+#include "base/logging.h"
+#include "base/commandlineflags.h"
+
+DEFINE_int32(verbose, EnvToInt("PERFTOOLS_VERBOSE", 0),
+             "Set to numbers >0 for more verbose output, or <0 for less.  "
+             "--verbose == -4 means we log fatal errors only.");
+
+
+#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__)
+
+// While windows does have a POSIX-compatible API
+// (_open/_write/_close), it acquires memory.  Using this lower-level
+// windows API is the closest we can get to being "raw".
+RawFD RawOpenForWriting(const char* filename) {
+  // CreateFile allocates memory if file_name isn't absolute, so if
+  // that ever becomes a problem then we ought to compute the absolute
+  // path on its behalf (perhaps the ntdll/kernel function isn't aware
+  // of the working directory?)
+  RawFD fd = CreateFileA(filename, GENERIC_WRITE, 0, NULL,
+                         CREATE_ALWAYS, 0, NULL);
+  if (fd != kIllegalRawFD && GetLastError() == ERROR_ALREADY_EXISTS)
+    SetEndOfFile(fd);    // truncate the existing file
+  return fd;
+}
+
+void RawWrite(RawFD handle, const char* buf, size_t len) {
+  while (len > 0) {
+    DWORD wrote;
+    BOOL ok = WriteFile(handle, buf, len, &wrote, NULL);
+    // We do not use an asynchronous file handle, so ok==false means an error
+    if (!ok) break;
+    buf += wrote;
+    len -= wrote;
+  }
+}
+
+void RawClose(RawFD handle) {
+  CloseHandle(handle);
+}
+
+#else  // _WIN32 || __CYGWIN__ || __CYGWIN32__
+
+#ifdef HAVE_SYS_TYPES_H
+#include <sys/types.h>
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#ifdef HAVE_FCNTL_H
+#include <fcntl.h>
+#endif
+
+// Re-run fn until it doesn't cause EINTR.
+#define NO_INTR(fn)  do {} while ((fn) < 0 && errno == EINTR)
+
+RawFD RawOpenForWriting(const char* filename) {
+  return open(filename, O_WRONLY|O_CREAT|O_TRUNC, 0664);
+}
+
+void RawWrite(RawFD fd, const char* buf, size_t len) {
+  while (len > 0) {
+    ssize_t r;
+    NO_INTR(r = write(fd, buf, len));
+    if (r <= 0) break;
+    buf += r;
+    len -= r;
+  }
+}
+
+void RawClose(RawFD fd) {
+  NO_INTR(close(fd));
+}
+
+#endif  // _WIN32 || __CYGWIN__ || __CYGWIN32__

BeefRT/gperftools/src/base/logging.h

@@ -0,0 +1,258 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// This file contains #include information about logging-related stuff.
+// Pretty much everybody needs to #include this file so that they can
+// log various happenings.
+//
+#ifndef _LOGGING_H_
+#define _LOGGING_H_
+
+#include "config.h"
+#include <stdarg.h>
+#include <stdlib.h>
+#include <stdio.h>
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>    // for write()
+#endif
+#include <string.h>    // for strlen(), strcmp()
+#include <assert.h>
+#include <errno.h>     // for errno
+#include "base/commandlineflags.h"
+
+// On some systems (like freebsd), we can't call write() at all in a
+// global constructor, perhaps because errno hasn't been set up.
+// (In windows, we can't call it because it might call malloc.)
+// Calling the write syscall is safer (it doesn't set errno), so we
+// prefer that.  Note we don't care about errno for logging: we just
+// do logging on a best-effort basis.
+#if defined(_MSC_VER)
+#define WRITE_TO_STDERR(buf, len) WriteToStderr(buf, len);  // in port.cc
+#elif defined(HAVE_SYS_SYSCALL_H)
+#include <sys/syscall.h>
+#define WRITE_TO_STDERR(buf, len) syscall(SYS_write, STDERR_FILENO, buf, len)
+#else
+#define WRITE_TO_STDERR(buf, len) _write(STDERR_FILENO, buf, len)
+#endif
+
+// MSVC and mingw define their own, safe version of vnsprintf (the
+// windows one in broken) in port.cc.  Everyone else can use the
+// version here.  We had to give it a unique name for windows.
+#ifndef _WIN32
+# define perftools_vsnprintf vsnprintf
+#endif
+
+
+// We log all messages at this log-level and below.
+// INFO == -1, WARNING == -2, ERROR == -3, FATAL == -4
+DECLARE_int32(verbose);
+
+// CHECK dies with a fatal error if condition is not true.  It is *not*
+// controlled by NDEBUG, so the check will be executed regardless of
+// compilation mode.  Therefore, it is safe to do things like:
+//    CHECK(fp->Write(x) == 4)
+// Note we use write instead of printf/puts to avoid the risk we'll
+// call malloc().
+#define CHECK(condition)                                                \
+  do {                                                                  \
+    if (!(condition)) {                                                 \
+      WRITE_TO_STDERR("Check failed: " #condition "\n",                 \
+                      sizeof("Check failed: " #condition "\n")-1);      \
+      abort();                                                          \
+    }                                                                   \
+  } while (0)
+
+// This takes a message to print.  The name is historical.
+#define RAW_CHECK(condition, message)                                          \
+  do {                                                                         \
+    if (!(condition)) {                                                        \
+      WRITE_TO_STDERR("Check failed: " #condition ": " message "\n",           \
+                      sizeof("Check failed: " #condition ": " message "\n")-1);\
+      abort();                                                                 \
+    }                                                                          \
+  } while (0)
+
+// This is like RAW_CHECK, but only in debug-mode
+#ifdef NDEBUG
+enum { DEBUG_MODE = 0 };
+#define RAW_DCHECK(condition, message)
+#else
+enum { DEBUG_MODE = 1 };
+#define RAW_DCHECK(condition, message)  RAW_CHECK(condition, message)
+#endif
+
+// This prints errno as well.  Note we use write instead of printf/puts to
+// avoid the risk we'll call malloc().
+#define PCHECK(condition)                                               \
+  do {                                                                  \
+    if (!(condition)) {                                                 \
+      const int err_no = errno;                                         \
+      WRITE_TO_STDERR("Check failed: " #condition ": ",                 \
+                      sizeof("Check failed: " #condition ": ")-1);      \
+      WRITE_TO_STDERR(strerror(err_no), strlen(strerror(err_no)));      \
+      WRITE_TO_STDERR("\n", sizeof("\n")-1);                            \
+      abort();                                                          \
+    }                                                                   \
+  } while (0)
+
+// Helper macro for binary operators; prints the two values on error
+// Don't use this macro directly in your code, use CHECK_EQ et al below
+
+// WARNING: These don't compile correctly if one of the arguments is a pointer
+// and the other is NULL. To work around this, simply static_cast NULL to the
+// type of the desired pointer.
+
+// TODO(jandrews): Also print the values in case of failure.  Requires some
+// sort of type-sensitive ToString() function.
+#define CHECK_OP(op, val1, val2)                                        \
+  do {                                                                  \
+    if (!((val1) op (val2))) {                                          \
+      fprintf(stderr, "Check failed: %s %s %s\n", #val1, #op, #val2);   \
+      abort();                                                          \
+    }                                                                   \
+  } while (0)
+
+#define CHECK_EQ(val1, val2) CHECK_OP(==, val1, val2)
+#define CHECK_NE(val1, val2) CHECK_OP(!=, val1, val2)
+#define CHECK_LE(val1, val2) CHECK_OP(<=, val1, val2)
+#define CHECK_LT(val1, val2) CHECK_OP(< , val1, val2)
+#define CHECK_GE(val1, val2) CHECK_OP(>=, val1, val2)
+#define CHECK_GT(val1, val2) CHECK_OP(> , val1, val2)
+
+// Synonyms for CHECK_* that are used in some unittests.
+#define EXPECT_EQ(val1, val2) CHECK_EQ(val1, val2)
+#define EXPECT_NE(val1, val2) CHECK_NE(val1, val2)
+#define EXPECT_LE(val1, val2) CHECK_LE(val1, val2)
+#define EXPECT_LT(val1, val2) CHECK_LT(val1, val2)
+#define EXPECT_GE(val1, val2) CHECK_GE(val1, val2)
+#define EXPECT_GT(val1, val2) CHECK_GT(val1, val2)
+#define ASSERT_EQ(val1, val2) EXPECT_EQ(val1, val2)
+#define ASSERT_NE(val1, val2) EXPECT_NE(val1, val2)
+#define ASSERT_LE(val1, val2) EXPECT_LE(val1, val2)
+#define ASSERT_LT(val1, val2) EXPECT_LT(val1, val2)
+#define ASSERT_GE(val1, val2) EXPECT_GE(val1, val2)
+#define ASSERT_GT(val1, val2) EXPECT_GT(val1, val2)
+// As are these variants.
+#define EXPECT_TRUE(cond)     CHECK(cond)
+#define EXPECT_FALSE(cond)    CHECK(!(cond))
+#define EXPECT_STREQ(a, b)    CHECK(strcmp(a, b) == 0)
+#define ASSERT_TRUE(cond)     EXPECT_TRUE(cond)
+#define ASSERT_FALSE(cond)    EXPECT_FALSE(cond)
+#define ASSERT_STREQ(a, b)    EXPECT_STREQ(a, b)
+
+// Used for (libc) functions that return -1 and set errno
+#define CHECK_ERR(invocation)  PCHECK((invocation) != -1)
+
+// A few more checks that only happen in debug mode
+#ifdef NDEBUG
+#define DCHECK_EQ(val1, val2)
+#define DCHECK_NE(val1, val2)
+#define DCHECK_LE(val1, val2)
+#define DCHECK_LT(val1, val2)
+#define DCHECK_GE(val1, val2)
+#define DCHECK_GT(val1, val2)
+#else
+#define DCHECK_EQ(val1, val2)  CHECK_EQ(val1, val2)
+#define DCHECK_NE(val1, val2)  CHECK_NE(val1, val2)
+#define DCHECK_LE(val1, val2)  CHECK_LE(val1, val2)
+#define DCHECK_LT(val1, val2)  CHECK_LT(val1, val2)
+#define DCHECK_GE(val1, val2)  CHECK_GE(val1, val2)
+#define DCHECK_GT(val1, val2)  CHECK_GT(val1, val2)
+#endif
+
+
+#ifdef ERROR
+#undef ERROR      // may conflict with ERROR macro on windows
+#endif
+enum LogSeverity {INFO = -1, WARNING = -2, ERROR = -3, FATAL = -4};
+
+// NOTE: we add a newline to the end of the output if it's not there already
+inline void LogPrintf(int severity, const char* pat, va_list ap) {
+  // We write directly to the stderr file descriptor and avoid FILE
+  // buffering because that may invoke malloc()
+  char buf[600];
+  perftools_vsnprintf(buf, sizeof(buf)-1, pat, ap);
+  if (buf[0] != '\0' && buf[strlen(buf)-1] != '\n') {
+    assert(strlen(buf)+1 < sizeof(buf));
+    strcat(buf, "\n");
+  }
+  WRITE_TO_STDERR(buf, strlen(buf));
+  if ((severity) == FATAL)
+    abort(); // LOG(FATAL) indicates a big problem, so don't run atexit() calls
+}
+
+// Note that since the order of global constructors is unspecified,
+// global code that calls RAW_LOG may execute before FLAGS_verbose is set.
+// Such code will run with verbosity == 0 no matter what.
+#define VLOG_IS_ON(severity) (FLAGS_verbose >= severity)
+
+// In a better world, we'd use __VA_ARGS__, but VC++ 7 doesn't support it.
+#define LOG_PRINTF(severity, pat) do {          \
+  if (VLOG_IS_ON(severity)) {                   \
+    va_list ap;                                 \
+    va_start(ap, pat);                          \
+    LogPrintf(severity, pat, ap);               \
+    va_end(ap);                                 \
+  }                                             \
+} while (0)
+
+// RAW_LOG is the main function; some synonyms are used in unittests.
+inline void RAW_LOG(int lvl, const char* pat, ...)  { LOG_PRINTF(lvl, pat); }
+inline void RAW_VLOG(int lvl, const char* pat, ...) { LOG_PRINTF(lvl, pat); }
+inline void LOG(int lvl, const char* pat, ...)      { LOG_PRINTF(lvl, pat); }
+inline void VLOG(int lvl, const char* pat, ...)     { LOG_PRINTF(lvl, pat); }
+inline void LOG_IF(int lvl, bool cond, const char* pat, ...) {
+  if (cond)  LOG_PRINTF(lvl, pat);
+}
+
+// This isn't technically logging, but it's also IO and also is an
+// attempt to be "raw" -- that is, to not use any higher-level libc
+// routines that might allocate memory or (ideally) try to allocate
+// locks.  We use an opaque file handle (not necessarily an int)
+// to allow even more low-level stuff in the future.
+// Like other "raw" routines, these functions are best effort, and
+// thus don't return error codes (except RawOpenForWriting()).
+#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__)
+#ifndef NOMINMAX
+#define NOMINMAX     // @#!$& windows
+#endif
+#include <windows.h>
+typedef HANDLE RawFD;
+const RawFD kIllegalRawFD = INVALID_HANDLE_VALUE;
+#else
+typedef int RawFD;
+const RawFD kIllegalRawFD = -1;   // what open returns if it fails
+#endif  // defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__)
+
+RawFD RawOpenForWriting(const char* filename);   // uses default permissions
+void RawWrite(RawFD fd, const char* buf, size_t len);
+void RawClose(RawFD fd);
+
+#endif // _LOGGING_H_

BeefRT/gperftools/src/base/low_level_alloc.cc

@@ -0,0 +1,522 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+// A low-level allocator that can be used by other low-level
+// modules without introducing dependency cycles.
+// This allocator is slow and wasteful of memory;
+// it should not be used when performance is key.
+
+#include "base/low_level_alloc.h"
+#include "base/dynamic_annotations.h"
+#include "base/spinlock.h"
+#include "base/logging.h"
+#include "malloc_hook-inl.h"
+#include <gperftools/malloc_hook.h>
+#include <errno.h>
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#ifdef HAVE_MMAP
+#include <sys/mman.h>
+#endif
+#include <new>                   // for placement-new
+
+// On systems (like freebsd) that don't define MAP_ANONYMOUS, use the old
+// form of the name instead.
+#ifndef MAP_ANONYMOUS
+# define MAP_ANONYMOUS MAP_ANON
+#endif
+
+// A first-fit allocator with amortized logarithmic free() time.
+
+// ---------------------------------------------------------------------------
+static const int kMaxLevel = 30;
+
+// We put this class-only struct in a namespace to avoid polluting the
+// global namespace with this struct name (thus risking an ODR violation).
+namespace low_level_alloc_internal {
+  // This struct describes one allocated block, or one free block.
+  struct AllocList {
+    struct Header {
+      intptr_t size;  // size of entire region, including this field. Must be
+                      // first.  Valid in both allocated and unallocated blocks
+      intptr_t magic; // kMagicAllocated or kMagicUnallocated xor this
+      LowLevelAlloc::Arena *arena; // pointer to parent arena
+      void *dummy_for_alignment;   // aligns regions to 0 mod 2*sizeof(void*)
+    } header;
+
+    // Next two fields: in unallocated blocks: freelist skiplist data
+    //                  in allocated blocks: overlaps with client data
+    int levels;           // levels in skiplist used
+    AllocList *next[kMaxLevel];   // actually has levels elements.
+                                  // The AllocList node may not have room for
+                                  // all kMaxLevel entries.  See max_fit in
+                                  // LLA_SkiplistLevels()
+  };
+}
+using low_level_alloc_internal::AllocList;
+
+
+// ---------------------------------------------------------------------------
+// A trivial skiplist implementation.  This is used to keep the freelist
+// in address order while taking only logarithmic time per insert and delete.
+
+// An integer approximation of log2(size/base)
+// Requires size >= base.
+static int IntLog2(size_t size, size_t base) {
+  int result = 0;
+  for (size_t i = size; i > base; i >>= 1) { // i == floor(size/2**result)
+    result++;
+  }
+  //    floor(size / 2**result) <= base < floor(size / 2**(result-1))
+  // =>     log2(size/(base+1)) <= result < 1+log2(size/base)
+  // => result ~= log2(size/base)
+  return result;
+}
+
+// Return a random integer n:  p(n)=1/(2**n) if 1 <= n; p(n)=0 if n < 1.
+static int Random() {
+  static int32 r = 1;         // no locking---it's not critical
+  ANNOTATE_BENIGN_RACE(&r, "benign race, not critical.");
+  int result = 1;
+  while ((((r = r*1103515245 + 12345) >> 30) & 1) == 0) {
+    result++;
+  }
+  return result;
+}
+
+// Return a number of skiplist levels for a node of size bytes, where
+// base is the minimum node size.  Compute level=log2(size / base)+n
+// where n is 1 if random is false and otherwise a random number generated with
+// the standard distribution for a skiplist:  See Random() above.
+// Bigger nodes tend to have more skiplist levels due to the log2(size / base)
+// term, so first-fit searches touch fewer nodes.  "level" is clipped so
+// level<kMaxLevel and next[level-1] will fit in the node.
+// 0 < LLA_SkiplistLevels(x,y,false) <= LLA_SkiplistLevels(x,y,true) < kMaxLevel
+static int LLA_SkiplistLevels(size_t size, size_t base, bool random) {
+  // max_fit is the maximum number of levels that will fit in a node for the
+  // given size.   We can't return more than max_fit, no matter what the
+  // random number generator says.
+  int max_fit = (size-OFFSETOF_MEMBER(AllocList, next)) / sizeof (AllocList *);
+  int level = IntLog2(size, base) + (random? Random() : 1);
+  if (level > max_fit)     level = max_fit;
+  if (level > kMaxLevel-1) level = kMaxLevel - 1;
+  RAW_CHECK(level >= 1, "block not big enough for even one level");
+  return level;
+}
+
+// Return "atleast", the first element of AllocList *head s.t. *atleast >= *e.
+// For 0 <= i < head->levels, set prev[i] to "no_greater", where no_greater
+// points to the last element at level i in the AllocList less than *e, or is
+// head if no such element exists.
+static AllocList *LLA_SkiplistSearch(AllocList *head,
+                                     AllocList *e, AllocList **prev) {
+  AllocList *p = head;
+  for (int level = head->levels - 1; level >= 0; level--) {
+    for (AllocList *n; (n = p->next[level]) != 0 && n < e; p = n) {
+    }
+    prev[level] = p;
+  }
+  return (head->levels == 0) ?  0 : prev[0]->next[0];
+}
+
+// Insert element *e into AllocList *head.  Set prev[] as LLA_SkiplistSearch.
+// Requires that e->levels be previously set by the caller (using
+// LLA_SkiplistLevels())
+static void LLA_SkiplistInsert(AllocList *head, AllocList *e,
+                               AllocList **prev) {
+  LLA_SkiplistSearch(head, e, prev);
+  for (; head->levels < e->levels; head->levels++) { // extend prev pointers
+    prev[head->levels] = head;                       // to all *e's levels
+  }
+  for (int i = 0; i != e->levels; i++) { // add element to list
+    e->next[i] = prev[i]->next[i];
+    prev[i]->next[i] = e;
+  }
+}
+
+// Remove element *e from AllocList *head.  Set prev[] as LLA_SkiplistSearch().
+// Requires that e->levels be previous set by the caller (using
+// LLA_SkiplistLevels())
+static void LLA_SkiplistDelete(AllocList *head, AllocList *e,
+                               AllocList **prev) {
+  AllocList *found = LLA_SkiplistSearch(head, e, prev);
+  RAW_CHECK(e == found, "element not in freelist");
+  for (int i = 0; i != e->levels && prev[i]->next[i] == e; i++) {
+    prev[i]->next[i] = e->next[i];
+  }
+  while (head->levels > 0 && head->next[head->levels - 1] == 0) {
+    head->levels--;   // reduce head->levels if level unused
+  }
+}
+
+// ---------------------------------------------------------------------------
+// Arena implementation
+
+struct LowLevelAlloc::Arena {
+  Arena() : mu(SpinLock::LINKER_INITIALIZED) {} // does nothing; for static init
+  explicit Arena(int) : pagesize(0) {}  // set pagesize to zero explicitly
+                                        // for non-static init
+
+  SpinLock mu;            // protects freelist, allocation_count,
+                          // pagesize, roundup, min_size
+  AllocList freelist;     // head of free list; sorted by addr (under mu)
+  int32 allocation_count; // count of allocated blocks (under mu)
+  int32 flags;            // flags passed to NewArena (ro after init)
+  size_t pagesize;        // ==getpagesize()  (init under mu, then ro)
+  size_t roundup;         // lowest power of 2 >= max(16,sizeof (AllocList))
+                          // (init under mu, then ro)
+  size_t min_size;        // smallest allocation block size
+                          // (init under mu, then ro)
+};
+
+// The default arena, which is used when 0 is passed instead of an Arena
+// pointer.
+static struct LowLevelAlloc::Arena default_arena;
+
+// Non-malloc-hooked arenas: used only to allocate metadata for arenas that
+// do not want malloc hook reporting, so that for them there's no malloc hook
+// reporting even during arena creation.
+static struct LowLevelAlloc::Arena unhooked_arena;
+static struct LowLevelAlloc::Arena unhooked_async_sig_safe_arena;
+
+// magic numbers to identify allocated and unallocated blocks
+static const intptr_t kMagicAllocated = 0x4c833e95;
+static const intptr_t kMagicUnallocated = ~kMagicAllocated;
+
+namespace {
+  class SCOPED_LOCKABLE ArenaLock {
+   public:
+    explicit ArenaLock(LowLevelAlloc::Arena *arena)
+        EXCLUSIVE_LOCK_FUNCTION(arena->mu)
+        : left_(false), mask_valid_(false), arena_(arena) {
+      if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) != 0) {
+      // We've decided not to support async-signal-safe arena use until
+      // there a demonstrated need.  Here's how one could do it though
+      // (would need to be made more portable).
+#if 0
+        sigset_t all;
+        sigfillset(&all);
+        this->mask_valid_ =
+            (pthread_sigmask(SIG_BLOCK, &all, &this->mask_) == 0);
+#else
+        RAW_CHECK(false, "We do not yet support async-signal-safe arena.");
+#endif
+      }
+      this->arena_->mu.Lock();
+    }
+    ~ArenaLock() { RAW_CHECK(this->left_, "haven't left Arena region"); }
+    void Leave() /*UNLOCK_FUNCTION()*/ {
+      this->arena_->mu.Unlock();
+#if 0
+      if (this->mask_valid_) {
+        pthread_sigmask(SIG_SETMASK, &this->mask_, 0);
+      }
+#endif
+      this->left_ = true;
+    }
+   private:
+    bool left_;       // whether left region
+    bool mask_valid_;
+#if 0
+    sigset_t mask_;   // old mask of blocked signals
+#endif
+    LowLevelAlloc::Arena *arena_;
+    DISALLOW_COPY_AND_ASSIGN(ArenaLock);
+  };
+} // anonymous namespace
+
+// create an appropriate magic number for an object at "ptr"
+// "magic" should be kMagicAllocated or kMagicUnallocated
+inline static intptr_t Magic(intptr_t magic, AllocList::Header *ptr) {
+  return magic ^ reinterpret_cast<intptr_t>(ptr);
+}
+
+// Initialize the fields of an Arena
+static void ArenaInit(LowLevelAlloc::Arena *arena) {
+  if (arena->pagesize == 0) {
+    arena->pagesize = getpagesize();
+    // Round up block sizes to a power of two close to the header size.
+    arena->roundup = 16;
+    while (arena->roundup < sizeof (arena->freelist.header)) {
+      arena->roundup += arena->roundup;
+    }
+    // Don't allocate blocks less than twice the roundup size to avoid tiny
+    // free blocks.
+    arena->min_size = 2 * arena->roundup;
+    arena->freelist.header.size = 0;
+    arena->freelist.header.magic =
+        Magic(kMagicUnallocated, &arena->freelist.header);
+    arena->freelist.header.arena = arena;
+    arena->freelist.levels = 0;
+    memset(arena->freelist.next, 0, sizeof (arena->freelist.next));
+    arena->allocation_count = 0;
+    if (arena == &default_arena) {
+      // Default arena should be hooked, e.g. for heap-checker to trace
+      // pointer chains through objects in the default arena.
+      arena->flags = LowLevelAlloc::kCallMallocHook;
+    } else if (arena == &unhooked_async_sig_safe_arena) {
+      arena->flags = LowLevelAlloc::kAsyncSignalSafe;
+    } else {
+      arena->flags = 0;   // other arenas' flags may be overridden by client,
+                          // but unhooked_arena will have 0 in 'flags'.
+    }
+  }
+}
+
+// L < meta_data_arena->mu
+LowLevelAlloc::Arena *LowLevelAlloc::NewArena(int32 flags,
+                                              Arena *meta_data_arena) {
+  RAW_CHECK(meta_data_arena != 0, "must pass a valid arena");
+  if (meta_data_arena == &default_arena) {
+    if ((flags & LowLevelAlloc::kAsyncSignalSafe) != 0) {
+      meta_data_arena = &unhooked_async_sig_safe_arena;
+    } else if ((flags & LowLevelAlloc::kCallMallocHook) == 0) {
+      meta_data_arena = &unhooked_arena;
+    }
+  }
+  // Arena(0) uses the constructor for non-static contexts
+  Arena *result =
+    new (AllocWithArena(sizeof (*result), meta_data_arena)) Arena(0);
+  ArenaInit(result);
+  result->flags = flags;
+  return result;
+}
+
+// L < arena->mu, L < arena->arena->mu
+bool LowLevelAlloc::DeleteArena(Arena *arena) {
+  RAW_CHECK(arena != 0 && arena != &default_arena && arena != &unhooked_arena,
+            "may not delete default arena");
+  ArenaLock section(arena);
+  bool empty = (arena->allocation_count == 0);
+  section.Leave();
+  if (empty) {
+    while (arena->freelist.next[0] != 0) {
+      AllocList *region = arena->freelist.next[0];
+      size_t size = region->header.size;
+      arena->freelist.next[0] = region->next[0];
+      RAW_CHECK(region->header.magic ==
+                Magic(kMagicUnallocated, &region->header),
+                "bad magic number in DeleteArena()");
+      RAW_CHECK(region->header.arena == arena,
+                "bad arena pointer in DeleteArena()");
+      RAW_CHECK(size % arena->pagesize == 0,
+                "empty arena has non-page-aligned block size");
+      RAW_CHECK(reinterpret_cast<intptr_t>(region) % arena->pagesize == 0,
+                "empty arena has non-page-aligned block");
+      int munmap_result;
+      if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) == 0) {
+        munmap_result = munmap(region, size);
+      } else {
+        munmap_result = MallocHook::UnhookedMUnmap(region, size);
+      }
+      RAW_CHECK(munmap_result == 0,
+                "LowLevelAlloc::DeleteArena:  munmap failed address");
+    }
+    Free(arena);
+  }
+  return empty;
+}
+
+// ---------------------------------------------------------------------------
+
+// Return value rounded up to next multiple of align.
+// align must be a power of two.
+static intptr_t RoundUp(intptr_t addr, intptr_t align) {
+  return (addr + align - 1) & ~(align - 1);
+}
+
+// Equivalent to "return prev->next[i]" but with sanity checking
+// that the freelist is in the correct order, that it
+// consists of regions marked "unallocated", and that no two regions
+// are adjacent in memory (they should have been coalesced).
+// L < arena->mu
+static AllocList *Next(int i, AllocList *prev, LowLevelAlloc::Arena *arena) {
+  RAW_CHECK(i < prev->levels, "too few levels in Next()");
+  AllocList *next = prev->next[i];
+  if (next != 0) {
+    RAW_CHECK(next->header.magic == Magic(kMagicUnallocated, &next->header),
+              "bad magic number in Next()");
+    RAW_CHECK(next->header.arena == arena,
+              "bad arena pointer in Next()");
+    if (prev != &arena->freelist) {
+      RAW_CHECK(prev < next, "unordered freelist");
+      RAW_CHECK(reinterpret_cast<char *>(prev) + prev->header.size <
+                reinterpret_cast<char *>(next), "malformed freelist");
+    }
+  }
+  return next;
+}
+
+// Coalesce list item "a" with its successor if they are adjacent.
+static void Coalesce(AllocList *a) {
+  AllocList *n = a->next[0];
+  if (n != 0 && reinterpret_cast<char *>(a) + a->header.size ==
+                    reinterpret_cast<char *>(n)) {
+    LowLevelAlloc::Arena *arena = a->header.arena;
+    a->header.size += n->header.size;
+    n->header.magic = 0;
+    n->header.arena = 0;
+    AllocList *prev[kMaxLevel];
+    LLA_SkiplistDelete(&arena->freelist, n, prev);
+    LLA_SkiplistDelete(&arena->freelist, a, prev);
+    a->levels = LLA_SkiplistLevels(a->header.size, arena->min_size, true);
+    LLA_SkiplistInsert(&arena->freelist, a, prev);
+  }
+}
+
+// Adds block at location "v" to the free list
+// L >= arena->mu
+static void AddToFreelist(void *v, LowLevelAlloc::Arena *arena) {
+  AllocList *f = reinterpret_cast<AllocList *>(
+                        reinterpret_cast<char *>(v) - sizeof (f->header));
+  RAW_CHECK(f->header.magic == Magic(kMagicAllocated, &f->header),
+            "bad magic number in AddToFreelist()");
+  RAW_CHECK(f->header.arena == arena,
+            "bad arena pointer in AddToFreelist()");
+  f->levels = LLA_SkiplistLevels(f->header.size, arena->min_size, true);
+  AllocList *prev[kMaxLevel];
+  LLA_SkiplistInsert(&arena->freelist, f, prev);
+  f->header.magic = Magic(kMagicUnallocated, &f->header);
+  Coalesce(f);                  // maybe coalesce with successor
+  Coalesce(prev[0]);            // maybe coalesce with predecessor
+}
+
+// Frees storage allocated by LowLevelAlloc::Alloc().
+// L < arena->mu
+void LowLevelAlloc::Free(void *v) {
+  if (v != 0) {
+    AllocList *f = reinterpret_cast<AllocList *>(
+                        reinterpret_cast<char *>(v) - sizeof (f->header));
+    RAW_CHECK(f->header.magic == Magic(kMagicAllocated, &f->header),
+              "bad magic number in Free()");
+    LowLevelAlloc::Arena *arena = f->header.arena;
+    if ((arena->flags & kCallMallocHook) != 0) {
+      MallocHook::InvokeDeleteHook(v);
+    }
+    ArenaLock section(arena);
+    AddToFreelist(v, arena);
+    RAW_CHECK(arena->allocation_count > 0, "nothing in arena to free");
+    arena->allocation_count--;
+    section.Leave();
+  }
+}
+
+// allocates and returns a block of size bytes, to be freed with Free()
+// L < arena->mu
+static void *DoAllocWithArena(size_t request, LowLevelAlloc::Arena *arena) {
+  void *result = 0;
+  if (request != 0) {
+    AllocList *s;       // will point to region that satisfies request
+    ArenaLock section(arena);
+    ArenaInit(arena);
+    // round up with header
+    size_t req_rnd = RoundUp(request + sizeof (s->header), arena->roundup);
+    for (;;) {      // loop until we find a suitable region
+      // find the minimum levels that a block of this size must have
+      int i = LLA_SkiplistLevels(req_rnd, arena->min_size, false) - 1;
+      if (i < arena->freelist.levels) {   // potential blocks exist
+        AllocList *before = &arena->freelist;  // predecessor of s
+        while ((s = Next(i, before, arena)) != 0 && s->header.size < req_rnd) {
+          before = s;
+        }
+        if (s != 0) {       // we found a region
+          break;
+        }
+      }
+      // we unlock before mmap() both because mmap() may call a callback hook,
+      // and because it may be slow.
+      arena->mu.Unlock();
+      // mmap generous 64K chunks to decrease
+      // the chances/impact of fragmentation:
+      size_t new_pages_size = RoundUp(req_rnd, arena->pagesize * 16);
+      void *new_pages;
+      if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) != 0) {
+        new_pages = MallocHook::UnhookedMMap(0, new_pages_size,
+            PROT_WRITE|PROT_READ, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+      } else {
+        new_pages = mmap(0, new_pages_size,
+            PROT_WRITE|PROT_READ, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
+      }
+      RAW_CHECK(new_pages != MAP_FAILED, "mmap error");
+      arena->mu.Lock();
+      s = reinterpret_cast<AllocList *>(new_pages);
+      s->header.size = new_pages_size;
+      // Pretend the block is allocated; call AddToFreelist() to free it.
+      s->header.magic = Magic(kMagicAllocated, &s->header);
+      s->header.arena = arena;
+      AddToFreelist(&s->levels, arena);  // insert new region into free list
+    }
+    AllocList *prev[kMaxLevel];
+    LLA_SkiplistDelete(&arena->freelist, s, prev);    // remove from free list
+    // s points to the first free region that's big enough
+    if (req_rnd + arena->min_size <= s->header.size) {  // big enough to split
+      AllocList *n = reinterpret_cast<AllocList *>
+                        (req_rnd + reinterpret_cast<char *>(s));
+      n->header.size = s->header.size - req_rnd;
+      n->header.magic = Magic(kMagicAllocated, &n->header);
+      n->header.arena = arena;
+      s->header.size = req_rnd;
+      AddToFreelist(&n->levels, arena);
+    }
+    s->header.magic = Magic(kMagicAllocated, &s->header);
+    RAW_CHECK(s->header.arena == arena, "");
+    arena->allocation_count++;
+    section.Leave();
+    result = &s->levels;
+  }
+  ANNOTATE_NEW_MEMORY(result, request);
+  return result;
+}
+
+void *LowLevelAlloc::Alloc(size_t request) {
+  void *result = DoAllocWithArena(request, &default_arena);
+  if ((default_arena.flags & kCallMallocHook) != 0) {
+    // this call must be directly in the user-called allocator function
+    // for MallocHook::GetCallerStackTrace to work properly
+    MallocHook::InvokeNewHook(result, request);
+  }
+  return result;
+}
+
+void *LowLevelAlloc::AllocWithArena(size_t request, Arena *arena) {
+  RAW_CHECK(arena != 0, "must pass a valid arena");
+  void *result = DoAllocWithArena(request, arena);
+  if ((arena->flags & kCallMallocHook) != 0) {
+    // this call must be directly in the user-called allocator function
+    // for MallocHook::GetCallerStackTrace to work properly
+    MallocHook::InvokeNewHook(result, request);
+  }
+  return result;
+}
+
+LowLevelAlloc::Arena *LowLevelAlloc::DefaultArena() {
+  return &default_arena;
+}

BeefRT/gperftools/src/base/low_level_alloc.h

@@ -0,0 +1,106 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#if !defined(_BASE_LOW_LEVEL_ALLOC_H_)
+#define _BASE_LOW_LEVEL_ALLOC_H_
+
+// A simple thread-safe memory allocator that does not depend on
+// mutexes or thread-specific data.  It is intended to be used
+// sparingly, and only when malloc() would introduce an unwanted
+// dependency, such as inside the heap-checker.
+
+#include <config.h>
+#include <stddef.h>             // for size_t
+#include "base/basictypes.h"
+
+class LowLevelAlloc {
+ public:
+  struct Arena;       // an arena from which memory may be allocated
+
+  // Returns a pointer to a block of at least "request" bytes
+  // that have been newly allocated from the specific arena.
+  // for Alloc() call the DefaultArena() is used.
+  // Returns 0 if passed request==0.
+  // Does not return 0 under other circumstances; it crashes if memory
+  // is not available.
+  static void *Alloc(size_t request)
+    ATTRIBUTE_SECTION(malloc_hook);
+  static void *AllocWithArena(size_t request, Arena *arena)
+    ATTRIBUTE_SECTION(malloc_hook);
+
+  // Deallocates a region of memory that was previously allocated with
+  // Alloc().   Does nothing if passed 0.   "s" must be either 0,
+  // or must have been returned from a call to Alloc() and not yet passed to
+  // Free() since that call to Alloc().  The space is returned to the arena
+  // from which it was allocated.
+  static void Free(void *s) ATTRIBUTE_SECTION(malloc_hook);
+
+    // ATTRIBUTE_SECTION(malloc_hook) for Alloc* and Free
+    // are to put all callers of MallocHook::Invoke* in this module
+    // into special section,
+    // so that MallocHook::GetCallerStackTrace can function accurately.
+
+  // Create a new arena.
+  // The root metadata for the new arena is allocated in the
+  // meta_data_arena; the DefaultArena() can be passed for meta_data_arena.
+  // These values may be ored into flags:
+  enum {
+    // Report calls to Alloc() and Free() via the MallocHook interface.
+    // Set in the DefaultArena.
+    kCallMallocHook = 0x0001,
+
+    // Make calls to Alloc(), Free() be async-signal-safe.  Not set in
+    // DefaultArena().
+    kAsyncSignalSafe = 0x0002,
+
+    // When used with DefaultArena(), the NewArena() and DeleteArena() calls
+    // obey the flags given explicitly in the NewArena() call, even if those
+    // flags differ from the settings in DefaultArena().  So the call
+    // NewArena(kAsyncSignalSafe, DefaultArena()) is itself async-signal-safe,
+    // as well as generatating an arena that provides async-signal-safe
+    // Alloc/Free.
+  };
+  static Arena *NewArena(int32 flags, Arena *meta_data_arena);
+
+  // Destroys an arena allocated by NewArena and returns true,
+  // provided no allocated blocks remain in the arena.
+  // If allocated blocks remain in the arena, does nothing and
+  // returns false.
+  // It is illegal to attempt to destroy the DefaultArena().
+  static bool DeleteArena(Arena *arena);
+
+  // The default arena that always exists.
+  static Arena *DefaultArena();
+
+ private:
+  LowLevelAlloc();      // no instances
+};
+
+#endif

BeefRT/gperftools/src/base/simple_mutex.h

@@ -0,0 +1,331 @@
+// Copyright (c) 2007, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+// 
+// ---
+// Author: Craig Silverstein.
+//
+// A simple mutex wrapper, supporting locks and read-write locks.
+// You should assume the locks are *not* re-entrant.
+//
+// To use: you should define the following macros in your configure.ac:
+//   ACX_PTHREAD
+//   AC_RWLOCK
+// The latter is defined in ../autoconf.
+//
+// This class is meant to be internal-only and should be wrapped by an
+// internal namespace.  Before you use this module, please give the
+// name of your internal namespace for this module.  Or, if you want
+// to expose it, you'll want to move it to the Google namespace.  We
+// cannot put this class in global namespace because there can be some
+// problems when we have multiple versions of Mutex in each shared object.
+//
+// NOTE: TryLock() is broken for NO_THREADS mode, at least in NDEBUG
+//       mode.
+//
+// CYGWIN NOTE: Cygwin support for rwlock seems to be buggy:
+//    http://www.cygwin.com/ml/cygwin/2008-12/msg00017.html
+// Because of that, we might as well use windows locks for
+// cygwin.  They seem to be more reliable than the cygwin pthreads layer.
+//
+// TRICKY IMPLEMENTATION NOTE:
+// This class is designed to be safe to use during
+// dynamic-initialization -- that is, by global constructors that are
+// run before main() starts.  The issue in this case is that
+// dynamic-initialization happens in an unpredictable order, and it
+// could be that someone else's dynamic initializer could call a
+// function that tries to acquire this mutex -- but that all happens
+// before this mutex's constructor has run.  (This can happen even if
+// the mutex and the function that uses the mutex are in the same .cc
+// file.)  Basically, because Mutex does non-trivial work in its
+// constructor, it's not, in the naive implementation, safe to use
+// before dynamic initialization has run on it.
+//
+// The solution used here is to pair the actual mutex primitive with a
+// bool that is set to true when the mutex is dynamically initialized.
+// (Before that it's false.)  Then we modify all mutex routines to
+// look at the bool, and not try to lock/unlock until the bool makes
+// it to true (which happens after the Mutex constructor has run.)
+//
+// This works because before main() starts -- particularly, during
+// dynamic initialization -- there are no threads, so a) it's ok that
+// the mutex operations are a no-op, since we don't need locking then
+// anyway; and b) we can be quite confident our bool won't change
+// state between a call to Lock() and a call to Unlock() (that would
+// require a global constructor in one translation unit to call Lock()
+// and another global constructor in another translation unit to call
+// Unlock() later, which is pretty perverse).
+//
+// That said, it's tricky, and can conceivably fail; it's safest to
+// avoid trying to acquire a mutex in a global constructor, if you
+// can.  One way it can fail is that a really smart compiler might
+// initialize the bool to true at static-initialization time (too
+// early) rather than at dynamic-initialization time.  To discourage
+// that, we set is_safe_ to true in code (not the constructor
+// colon-initializer) and set it to true via a function that always
+// evaluates to true, but that the compiler can't know always
+// evaluates to true.  This should be good enough.
+//
+// A related issue is code that could try to access the mutex
+// after it's been destroyed in the global destructors (because
+// the Mutex global destructor runs before some other global
+// destructor, that tries to acquire the mutex).  The way we
+// deal with this is by taking a constructor arg that global
+// mutexes should pass in, that causes the destructor to do no
+// work.  We still depend on the compiler not doing anything
+// weird to a Mutex's memory after it is destroyed, but for a
+// static global variable, that's pretty safe.
+
+#ifndef GOOGLE_MUTEX_H_
+#define GOOGLE_MUTEX_H_
+
+#include <config.h>
+
+#if defined(NO_THREADS)
+  typedef int MutexType;      // to keep a lock-count
+#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__)
+# ifndef WIN32_LEAN_AND_MEAN
+#   define WIN32_LEAN_AND_MEAN  // We only need minimal includes
+# endif
+  // We need Windows NT or later for TryEnterCriticalSection().  If you
+  // don't need that functionality, you can remove these _WIN32_WINNT
+  // lines, and change TryLock() to assert(0) or something.
+# ifndef _WIN32_WINNT
+#   define _WIN32_WINNT 0x0400
+# endif
+# include <windows.h>
+  typedef CRITICAL_SECTION MutexType;
+#elif defined(HAVE_PTHREAD) && defined(HAVE_RWLOCK)
+  // Needed for pthread_rwlock_*.  If it causes problems, you could take it
+  // out, but then you'd have to unset HAVE_RWLOCK (at least on linux -- it
+  // *does* cause problems for FreeBSD, or MacOSX, but isn't needed
+  // for locking there.)
+# ifdef __linux__
+#   define _XOPEN_SOURCE 500  // may be needed to get the rwlock calls
+# endif
+# include <pthread.h>
+  typedef pthread_rwlock_t MutexType;
+#elif defined(HAVE_PTHREAD)
+# include <pthread.h>
+  typedef pthread_mutex_t MutexType;
+#else
+# error Need to implement mutex.h for your architecture, or #define NO_THREADS
+#endif
+
+#include <assert.h>
+#include <stdlib.h>      // for abort()
+
+#define MUTEX_NAMESPACE perftools_mutex_namespace
+
+namespace MUTEX_NAMESPACE {
+
+class Mutex {
+ public:
+  // This is used for the single-arg constructor
+  enum LinkerInitialized { LINKER_INITIALIZED };
+
+  // Create a Mutex that is not held by anybody.  This constructor is
+  // typically used for Mutexes allocated on the heap or the stack.
+  inline Mutex();
+  // This constructor should be used for global, static Mutex objects.
+  // It inhibits work being done by the destructor, which makes it
+  // safer for code that tries to acqiure this mutex in their global
+  // destructor.
+  inline Mutex(LinkerInitialized);
+
+  // Destructor
+  inline ~Mutex();
+
+  inline void Lock();    // Block if needed until free then acquire exclusively
+  inline void Unlock();  // Release a lock acquired via Lock()
+  inline bool TryLock(); // If free, Lock() and return true, else return false
+  // Note that on systems that don't support read-write locks, these may
+  // be implemented as synonyms to Lock() and Unlock().  So you can use
+  // these for efficiency, but don't use them anyplace where being able
+  // to do shared reads is necessary to avoid deadlock.
+  inline void ReaderLock();   // Block until free or shared then acquire a share
+  inline void ReaderUnlock(); // Release a read share of this Mutex
+  inline void WriterLock() { Lock(); }     // Acquire an exclusive lock
+  inline void WriterUnlock() { Unlock(); } // Release a lock from WriterLock()
+
+ private:
+  MutexType mutex_;
+  // We want to make sure that the compiler sets is_safe_ to true only
+  // when we tell it to, and never makes assumptions is_safe_ is
+  // always true.  volatile is the most reliable way to do that.
+  volatile bool is_safe_;
+  // This indicates which constructor was called.
+  bool destroy_;
+
+  inline void SetIsSafe() { is_safe_ = true; }
+
+  // Catch the error of writing Mutex when intending MutexLock.
+  Mutex(Mutex* /*ignored*/) {}
+  // Disallow "evil" constructors
+  Mutex(const Mutex&);
+  void operator=(const Mutex&);
+};
+
+// Now the implementation of Mutex for various systems
+#if defined(NO_THREADS)
+
+// When we don't have threads, we can be either reading or writing,
+// but not both.  We can have lots of readers at once (in no-threads
+// mode, that's most likely to happen in recursive function calls),
+// but only one writer.  We represent this by having mutex_ be -1 when
+// writing and a number > 0 when reading (and 0 when no lock is held).
+//
+// In debug mode, we assert these invariants, while in non-debug mode
+// we do nothing, for efficiency.  That's why everything is in an
+// assert.
+
+Mutex::Mutex() : mutex_(0) { }
+Mutex::Mutex(Mutex::LinkerInitialized) : mutex_(0) { }
+Mutex::~Mutex()            { assert(mutex_ == 0); }
+void Mutex::Lock()         { assert(--mutex_ == -1); }
+void Mutex::Unlock()       { assert(mutex_++ == -1); }
+bool Mutex::TryLock()      { if (mutex_) return false; Lock(); return true; }
+void Mutex::ReaderLock()   { assert(++mutex_ > 0); }
+void Mutex::ReaderUnlock() { assert(mutex_-- > 0); }
+
+#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__)
+
+Mutex::Mutex() : destroy_(true) {
+  InitializeCriticalSection(&mutex_);
+  SetIsSafe();
+}
+Mutex::Mutex(LinkerInitialized) : destroy_(false) {
+  InitializeCriticalSection(&mutex_);
+  SetIsSafe();
+}
+Mutex::~Mutex()            { if (destroy_) DeleteCriticalSection(&mutex_); }
+void Mutex::Lock()         { if (is_safe_) EnterCriticalSection(&mutex_); }
+void Mutex::Unlock()       { if (is_safe_) LeaveCriticalSection(&mutex_); }
+bool Mutex::TryLock()      { return is_safe_ ?
+                                 TryEnterCriticalSection(&mutex_) != 0 : true; }
+void Mutex::ReaderLock()   { Lock(); }      // we don't have read-write locks
+void Mutex::ReaderUnlock() { Unlock(); }
+
+#elif defined(HAVE_PTHREAD) && defined(HAVE_RWLOCK)
+
+#define SAFE_PTHREAD(fncall)  do {   /* run fncall if is_safe_ is true */  \
+  if (is_safe_ && fncall(&mutex_) != 0) abort();                           \
+} while (0)
+
+Mutex::Mutex() : destroy_(true) {
+  SetIsSafe();
+  if (is_safe_ && pthread_rwlock_init(&mutex_, NULL) != 0) abort();
+}
+Mutex::Mutex(Mutex::LinkerInitialized) : destroy_(false) {
+  SetIsSafe();
+  if (is_safe_ && pthread_rwlock_init(&mutex_, NULL) != 0) abort();
+}
+Mutex::~Mutex()       { if (destroy_) SAFE_PTHREAD(pthread_rwlock_destroy); }
+void Mutex::Lock()         { SAFE_PTHREAD(pthread_rwlock_wrlock); }
+void Mutex::Unlock()       { SAFE_PTHREAD(pthread_rwlock_unlock); }
+bool Mutex::TryLock()      { return is_safe_ ?
+                               pthread_rwlock_trywrlock(&mutex_) == 0 : true; }
+void Mutex::ReaderLock()   { SAFE_PTHREAD(pthread_rwlock_rdlock); }
+void Mutex::ReaderUnlock() { SAFE_PTHREAD(pthread_rwlock_unlock); }
+#undef SAFE_PTHREAD
+
+#elif defined(HAVE_PTHREAD)
+
+#define SAFE_PTHREAD(fncall)  do {   /* run fncall if is_safe_ is true */  \
+  if (is_safe_ && fncall(&mutex_) != 0) abort();                           \
+} while (0)
+
+Mutex::Mutex() : destroy_(true) {
+  SetIsSafe();
+  if (is_safe_ && pthread_mutex_init(&mutex_, NULL) != 0) abort();
+}
+Mutex::Mutex(Mutex::LinkerInitialized) : destroy_(false) {
+  SetIsSafe();
+  if (is_safe_ && pthread_mutex_init(&mutex_, NULL) != 0) abort();
+}
+Mutex::~Mutex()       { if (destroy_) SAFE_PTHREAD(pthread_mutex_destroy); }
+void Mutex::Lock()         { SAFE_PTHREAD(pthread_mutex_lock); }
+void Mutex::Unlock()       { SAFE_PTHREAD(pthread_mutex_unlock); }
+bool Mutex::TryLock()      { return is_safe_ ?
+                                 pthread_mutex_trylock(&mutex_) == 0 : true; }
+void Mutex::ReaderLock()   { Lock(); }
+void Mutex::ReaderUnlock() { Unlock(); }
+#undef SAFE_PTHREAD
+
+#endif
+
+// --------------------------------------------------------------------------
+// Some helper classes
+
+// MutexLock(mu) acquires mu when constructed and releases it when destroyed.
+class MutexLock {
+ public:
+  explicit MutexLock(Mutex *mu) : mu_(mu) { mu_->Lock(); }
+  ~MutexLock() { mu_->Unlock(); }
+ private:
+  Mutex * const mu_;
+  // Disallow "evil" constructors
+  MutexLock(const MutexLock&);
+  void operator=(const MutexLock&);
+};
+
+// ReaderMutexLock and WriterMutexLock do the same, for rwlocks
+class ReaderMutexLock {
+ public:
+  explicit ReaderMutexLock(Mutex *mu) : mu_(mu) { mu_->ReaderLock(); }
+  ~ReaderMutexLock() { mu_->ReaderUnlock(); }
+ private:
+  Mutex * const mu_;
+  // Disallow "evil" constructors
+  ReaderMutexLock(const ReaderMutexLock&);
+  void operator=(const ReaderMutexLock&);
+};
+
+class WriterMutexLock {
+ public:
+  explicit WriterMutexLock(Mutex *mu) : mu_(mu) { mu_->WriterLock(); }
+  ~WriterMutexLock() { mu_->WriterUnlock(); }
+ private:
+  Mutex * const mu_;
+  // Disallow "evil" constructors
+  WriterMutexLock(const WriterMutexLock&);
+  void operator=(const WriterMutexLock&);
+};
+
+// Catch bug where variable name is omitted, e.g. MutexLock (&mu);
+#define MutexLock(x) COMPILE_ASSERT(0, mutex_lock_decl_missing_var_name)
+#define ReaderMutexLock(x) COMPILE_ASSERT(0, rmutex_lock_decl_missing_var_name)
+#define WriterMutexLock(x) COMPILE_ASSERT(0, wmutex_lock_decl_missing_var_name)
+
+}  // namespace MUTEX_NAMESPACE
+
+using namespace MUTEX_NAMESPACE;
+
+#undef MUTEX_NAMESPACE
+
+#endif  /* #define GOOGLE_SIMPLE_MUTEX_H_ */

BeefRT/gperftools/src/base/spinlock.cc

@@ -0,0 +1,182 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ */
+
+#include <config.h>
+#include "base/spinlock.h"
+#include "base/synchronization_profiling.h"
+#include "base/spinlock_internal.h"
+#include "base/cycleclock.h"
+#include "base/sysinfo.h"   /* for NumCPUs() */
+
+// NOTE on the Lock-state values:
+//
+//   kSpinLockFree represents the unlocked state
+//   kSpinLockHeld represents the locked state with no waiters
+//
+// Values greater than kSpinLockHeld represent the locked state with waiters,
+// where the value is the time the current lock holder had to
+// wait before obtaining the lock.  The kSpinLockSleeper state is a special
+// "locked with waiters" state that indicates that a sleeper needs to
+// be woken, but the thread that just released the lock didn't wait.
+
+static int adaptive_spin_count = 0;
+
+const base::LinkerInitialized SpinLock::LINKER_INITIALIZED =
+    base::LINKER_INITIALIZED;
+
+namespace {
+struct SpinLock_InitHelper {
+  SpinLock_InitHelper() {
+    // On multi-cpu machines, spin for longer before yielding
+    // the processor or sleeping.  Reduces idle time significantly.
+    if (NumCPUs() > 1) {
+      adaptive_spin_count = 1000;
+    }
+  }
+};
+
+// Hook into global constructor execution:
+// We do not do adaptive spinning before that,
+// but nothing lock-intensive should be going on at that time.
+static SpinLock_InitHelper init_helper;
+
+}  // unnamed namespace
+
+// Monitor the lock to see if its value changes within some time period
+// (adaptive_spin_count loop iterations).  A timestamp indicating
+// when the thread initially started waiting for the lock is passed in via
+// the initial_wait_timestamp value.  The total wait time in cycles for the
+// lock is returned in the wait_cycles parameter.  The last value read
+// from the lock is returned from the method.
+Atomic32 SpinLock::SpinLoop(int64 initial_wait_timestamp,
+                            Atomic32* wait_cycles) {
+  int c = adaptive_spin_count;
+  while (base::subtle::NoBarrier_Load(&lockword_) != kSpinLockFree && --c > 0) {
+  }
+  Atomic32 spin_loop_wait_cycles = CalculateWaitCycles(initial_wait_timestamp);
+  Atomic32 lock_value =
+      base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree,
+                                           spin_loop_wait_cycles);
+  *wait_cycles = spin_loop_wait_cycles;
+  return lock_value;
+}
+
+void SpinLock::SlowLock() {
+  // The lock was not obtained initially, so this thread needs to wait for
+  // it.  Record the current timestamp in the local variable wait_start_time
+  // so the total wait time can be stored in the lockword once this thread
+  // obtains the lock.
+  int64 wait_start_time = CycleClock::Now();
+  Atomic32 wait_cycles;
+  Atomic32 lock_value = SpinLoop(wait_start_time, &wait_cycles);
+
+  int lock_wait_call_count = 0;
+  while (lock_value != kSpinLockFree) {
+    // If the lock is currently held, but not marked as having a sleeper, mark
+    // it as having a sleeper.
+    if (lock_value == kSpinLockHeld) {
+      // Here, just "mark" that the thread is going to sleep.  Don't store the
+      // lock wait time in the lock as that will cause the current lock
+      // owner to think it experienced contention.
+      lock_value = base::subtle::Acquire_CompareAndSwap(&lockword_,
+                                                        kSpinLockHeld,
+                                                        kSpinLockSleeper);
+      if (lock_value == kSpinLockHeld) {
+        // Successfully transitioned to kSpinLockSleeper.  Pass
+        // kSpinLockSleeper to the SpinLockWait routine to properly indicate
+        // the last lock_value observed.
+        lock_value = kSpinLockSleeper;
+      } else if (lock_value == kSpinLockFree) {
+        // Lock is free again, so try and acquire it before sleeping.  The
+        // new lock state will be the number of cycles this thread waited if
+        // this thread obtains the lock.
+        lock_value = base::subtle::Acquire_CompareAndSwap(&lockword_,
+                                                          kSpinLockFree,
+                                                          wait_cycles);
+        continue;  // skip the delay at the end of the loop
+      }
+    }
+
+    // Wait for an OS specific delay.
+    base::internal::SpinLockDelay(&lockword_, lock_value,
+                                  ++lock_wait_call_count);
+    // Spin again after returning from the wait routine to give this thread
+    // some chance of obtaining the lock.
+    lock_value = SpinLoop(wait_start_time, &wait_cycles);
+  }
+}
+
+// The wait time for contentionz lock profiling must fit into 32 bits.
+// However, the lower 32-bits of the cycle counter wrap around too quickly
+// with high frequency processors, so a right-shift by 7 is performed to
+// quickly divide the cycles by 128.  Using these 32 bits, reduces the
+// granularity of time measurement to 128 cycles, and loses track
+// of wait time for waits greater than 109 seconds on a 5 GHz machine
+// [(2^32 cycles/5 Ghz)*128 = 109.95 seconds]. Waits this long should be
+// very rare and the reduced granularity should not be an issue given
+// processors in the Google fleet operate at a minimum of one billion
+// cycles/sec.
+enum { PROFILE_TIMESTAMP_SHIFT = 7 };
+
+void SpinLock::SlowUnlock(uint64 wait_cycles) {
+  base::internal::SpinLockWake(&lockword_, false);  // wake waiter if necessary
+
+  // Collect contentionz profile info, expanding the wait_cycles back out to
+  // the full value.  If wait_cycles is <= kSpinLockSleeper, then no wait
+  // was actually performed, so don't record the wait time.  Note, that the
+  // CalculateWaitCycles method adds in kSpinLockSleeper cycles
+  // unconditionally to guarantee the wait time is not kSpinLockFree or
+  // kSpinLockHeld.  The adding in of these small number of cycles may
+  // overestimate the contention by a slight amount 50% of the time.  However,
+  // if this code tried to correct for that addition by subtracting out the
+  // kSpinLockSleeper amount that would underestimate the contention slightly
+  // 50% of the time.  Both ways get the wrong answer, so the code
+  // overestimates to be more conservative. Overestimating also makes the code
+  // a little simpler.
+  //
+  if (wait_cycles > kSpinLockSleeper) {
+    base::SubmitSpinLockProfileData(this,
+                                    wait_cycles << PROFILE_TIMESTAMP_SHIFT);
+  }
+}
+
+inline int32 SpinLock::CalculateWaitCycles(int64 wait_start_time) {
+  int32 wait_cycles = ((CycleClock::Now() - wait_start_time) >>
+                       PROFILE_TIMESTAMP_SHIFT);
+  // The number of cycles waiting for the lock is used as both the
+  // wait_cycles and lock value, so it can't be kSpinLockFree or
+  // kSpinLockHeld.  Make sure the value returned is at least
+  // kSpinLockSleeper.
+  wait_cycles |= kSpinLockSleeper;
+  return wait_cycles;
+}

BeefRT/gperftools/src/base/spinlock.h

@@ -0,0 +1,145 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ */
+
+// SpinLock is async signal safe.
+// If used within a signal handler, all lock holders
+// should block the signal even outside the signal handler.
+
+#ifndef BASE_SPINLOCK_H_
+#define BASE_SPINLOCK_H_
+
+#include "tcconfig.h"
+#include "base/atomicops.h"
+#include "base/basictypes.h"
+#include "base/dynamic_annotations.h"
+#include "base/thread_annotations.h"
+
+class LOCKABLE SpinLock {
+ public:
+  SpinLock() : lockword_(kSpinLockFree) { }
+
+  // Special constructor for use with static SpinLock objects.  E.g.,
+  //
+  //    static SpinLock lock(base::LINKER_INITIALIZED);
+  //
+  // When intialized using this constructor, we depend on the fact
+  // that the linker has already initialized the memory appropriately.
+  // A SpinLock constructed like this can be freely used from global
+  // initializers without worrying about the order in which global
+  // initializers run.
+  explicit SpinLock(base::LinkerInitialized /*x*/) {
+    // Does nothing; lockword_ is already initialized
+  }
+
+  // Acquire this SpinLock.
+  // TODO(csilvers): uncomment the annotation when we figure out how to
+  //                 support this macro with 0 args (see thread_annotations.h)
+  inline void Lock() /*EXCLUSIVE_LOCK_FUNCTION()*/ {
+    if (base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree,
+                                             kSpinLockHeld) != kSpinLockFree) {
+      SlowLock();
+    }
+    ANNOTATE_RWLOCK_ACQUIRED(this, 1);
+  }
+
+  // Try to acquire this SpinLock without blocking and return true if the
+  // acquisition was successful.  If the lock was not acquired, false is
+  // returned.  If this SpinLock is free at the time of the call, TryLock
+  // will return true with high probability.
+  inline bool TryLock() EXCLUSIVE_TRYLOCK_FUNCTION(true) {
+    bool res =
+        (base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree,
+                                              kSpinLockHeld) == kSpinLockFree);
+    if (res) {
+      ANNOTATE_RWLOCK_ACQUIRED(this, 1);
+    }
+    return res;
+  }
+
+  // Release this SpinLock, which must be held by the calling thread.
+  // TODO(csilvers): uncomment the annotation when we figure out how to
+  //                 support this macro with 0 args (see thread_annotations.h)
+  inline void Unlock() /*UNLOCK_FUNCTION()*/ {
+    ANNOTATE_RWLOCK_RELEASED(this, 1);
+    uint64 wait_cycles = static_cast<uint64>(
+        base::subtle::Release_AtomicExchange(&lockword_, kSpinLockFree));
+    if (wait_cycles != kSpinLockHeld) {
+      // Collect contentionz profile info, and speed the wakeup of any waiter.
+      // The wait_cycles value indicates how long this thread spent waiting
+      // for the lock.
+      SlowUnlock(wait_cycles);
+    }
+  }
+
+  // Determine if the lock is held.  When the lock is held by the invoking
+  // thread, true will always be returned. Intended to be used as
+  // CHECK(lock.IsHeld()).
+  inline bool IsHeld() const {
+    return base::subtle::NoBarrier_Load(&lockword_) != kSpinLockFree;
+  }
+
+  static const base::LinkerInitialized LINKER_INITIALIZED;  // backwards compat
+ private:
+  enum { kSpinLockFree = 0 };
+  enum { kSpinLockHeld = 1 };
+  enum { kSpinLockSleeper = 2 };
+
+  volatile Atomic32 lockword_;
+
+  void SlowLock();
+  void SlowUnlock(uint64 wait_cycles);
+  Atomic32 SpinLoop(int64 initial_wait_timestamp, Atomic32* wait_cycles);
+  inline int32 CalculateWaitCycles(int64 wait_start_time);
+
+  DISALLOW_COPY_AND_ASSIGN(SpinLock);
+};
+
+// Corresponding locker object that arranges to acquire a spinlock for
+// the duration of a C++ scope.
+class SCOPED_LOCKABLE SpinLockHolder {
+ private:
+  SpinLock* lock_;
+ public:
+  inline explicit SpinLockHolder(SpinLock* l) EXCLUSIVE_LOCK_FUNCTION(l)
+      : lock_(l) {
+    l->Lock();
+  }
+  // TODO(csilvers): uncomment the annotation when we figure out how to
+  //                 support this macro with 0 args (see thread_annotations.h)
+  inline ~SpinLockHolder() /*UNLOCK_FUNCTION()*/ { lock_->Unlock(); }
+};
+// Catch bug where variable name is omitted, e.g. SpinLockHolder (&lock);
+#define SpinLockHolder(x) COMPILE_ASSERT(0, spin_lock_decl_missing_var_name)
+
+
+#endif  // BASE_SPINLOCK_H_

BeefRT/gperftools/src/base/spinlock_internal.cc

@@ -0,0 +1,121 @@
+/* Copyright (c) 2010, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+// The OS-specific header included below must provide two calls:
+// base::internal::SpinLockDelay() and base::internal::SpinLockWake().
+// See spinlock_internal.h for the spec of SpinLockWake().
+
+// void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop)
+// SpinLockDelay() generates an apprproate spin delay on iteration "loop" of a
+// spin loop on location *w, whose previously observed value was "value".
+// SpinLockDelay() may do nothing, may yield the CPU, may sleep a clock tick,
+// or may wait for a delay that can be truncated by a call to SpinlockWake(w).
+// In all cases, it must return in bounded time even if SpinlockWake() is not
+// called.
+
+#include "base/spinlock_internal.h"
+
+// forward declaration for use by spinlock_*-inl.h
+namespace base { namespace internal { static int SuggestedDelayNS(int loop); }}
+
+#if defined(_WIN32)
+#include "base/spinlock_win32-inl.h"
+#elif defined(__linux__)
+#include "base/spinlock_linux-inl.h"
+#else
+#include "base/spinlock_posix-inl.h"
+#endif
+
+namespace base {
+namespace internal {
+
+// See spinlock_internal.h for spec.
+int32 SpinLockWait(volatile Atomic32 *w, int n,
+                   const SpinLockWaitTransition trans[]) {
+  int32 v;
+  bool done = false;
+  for (int loop = 0; !done; loop++) {
+    v = base::subtle::Acquire_Load(w);
+    int i;
+    for (i = 0; i != n && v != trans[i].from; i++) {
+    }
+    if (i == n) {
+      SpinLockDelay(w, v, loop);     // no matching transition
+    } else if (trans[i].to == v ||   // null transition
+               base::subtle::Acquire_CompareAndSwap(w, v, trans[i].to) == v) {
+      done = trans[i].done;
+    }
+  }
+  return v;
+}
+
+// Return a suggested delay in nanoseconds for iteration number "loop"
+static int SuggestedDelayNS(int loop) {
+  // Weak pseudo-random number generator to get some spread between threads
+  // when many are spinning.
+#ifdef BASE_HAS_ATOMIC64
+  static base::subtle::Atomic64 rand;
+  uint64 r = base::subtle::NoBarrier_Load(&rand);
+  r = 0x5deece66dLL * r + 0xb;   // numbers from nrand48()
+  base::subtle::NoBarrier_Store(&rand, r);
+
+  r <<= 16;   // 48-bit random number now in top 48-bits.
+  if (loop < 0 || loop > 32) {   // limit loop to 0..32
+    loop = 32;
+  }
+  // loop>>3 cannot exceed 4 because loop cannot exceed 32.
+  // Select top 20..24 bits of lower 48 bits,
+  // giving approximately 0ms to 16ms.
+  // Mean is exponential in loop for first 32 iterations, then 8ms.
+  // The futex path multiplies this by 16, since we expect explicit wakeups
+  // almost always on that path.
+  return r >> (44 - (loop >> 3));
+#else
+  static Atomic32 rand;
+  uint32 r = base::subtle::NoBarrier_Load(&rand);
+  r = 0x343fd * r + 0x269ec3;   // numbers from MSVC++
+  base::subtle::NoBarrier_Store(&rand, r);
+
+  r <<= 1;   // 31-bit random number now in top 31-bits.
+  if (loop < 0 || loop > 32) {   // limit loop to 0..32
+    loop = 32;
+  }
+  // loop>>3 cannot exceed 4 because loop cannot exceed 32.
+  // Select top 20..24 bits of lower 31 bits,
+  // giving approximately 0ms to 16ms.
+  // Mean is exponential in loop for first 32 iterations, then 8ms.
+  // The futex path multiplies this by 16, since we expect explicit wakeups
+  // almost always on that path.
+  return r >> (12 - (loop >> 3));
+#endif
+}
+
+} // namespace internal
+} // namespace base

BeefRT/gperftools/src/base/spinlock_internal.h

@@ -0,0 +1,64 @@
+/* Copyright (c) 2010, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * This file is an internal part spinlock.cc and once.cc
+ * It may not be used directly by code outside of //base.
+ */
+
+#ifndef BASE_SPINLOCK_INTERNAL_H_
+#define BASE_SPINLOCK_INTERNAL_H_
+
+#include <config.h>
+#include "base/basictypes.h"
+#include "base/atomicops.h"
+
+namespace base {
+namespace internal {
+
+// SpinLockWait() waits until it can perform one of several transitions from
+// "from" to "to".  It returns when it performs a transition where done==true.
+struct SpinLockWaitTransition {
+  int32 from;
+  int32 to;
+  bool done;
+};
+
+// Wait until *w can transition from trans[i].from to trans[i].to for some i
+// satisfying 0<=i<n && trans[i].done, atomically make the transition,
+// then return the old value of *w.   Make any other atomic tranistions
+// where !trans[i].done, but continue waiting.
+int32 SpinLockWait(volatile Atomic32 *w, int n,
+                   const SpinLockWaitTransition trans[]);
+void SpinLockWake(volatile Atomic32 *w, bool all);
+void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop);
+
+} // namespace internal
+} // namespace base
+#endif

BeefRT/gperftools/src/base/spinlock_linux-inl.h

@@ -0,0 +1,103 @@
+/* Copyright (c) 2009, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * This file is a Linux-specific part of spinlock_internal.cc
+ */
+
+#include <errno.h>
+#include <sched.h>
+#include <time.h>
+#include <limits.h>
+#include "base/linux_syscall_support.h"
+
+#define FUTEX_WAIT 0
+#define FUTEX_WAKE 1
+#define FUTEX_PRIVATE_FLAG 128
+
+static bool have_futex;
+static int futex_private_flag = FUTEX_PRIVATE_FLAG;
+
+namespace {
+static struct InitModule {
+  InitModule() {
+    int x = 0;
+    // futexes are ints, so we can use them only when
+    // that's the same size as the lockword_ in SpinLock.
+#ifdef __arm__
+    // ARM linux doesn't support sys_futex1(void*, int, int, struct timespec*);
+    have_futex = 0;
+#else
+    have_futex = (sizeof (Atomic32) == sizeof (int) &&
+                  sys_futex(&x, FUTEX_WAKE, 1, 0) >= 0);
+#endif
+    if (have_futex &&
+        sys_futex(&x, FUTEX_WAKE | futex_private_flag, 1, 0) < 0) {
+      futex_private_flag = 0;
+    }
+  }
+} init_module;
+
+}  // anonymous namespace
+
+
+namespace base {
+namespace internal {
+
+void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) {
+  if (loop != 0) {
+    int save_errno = errno;
+    struct timespec tm;
+    tm.tv_sec = 0;
+    if (have_futex) {
+      tm.tv_nsec = base::internal::SuggestedDelayNS(loop);
+    } else {
+      tm.tv_nsec = 2000001;   // above 2ms so linux 2.4 doesn't spin
+    }
+    if (have_futex) {
+      tm.tv_nsec *= 16;  // increase the delay; we expect explicit wakeups
+      sys_futex(reinterpret_cast<int *>(const_cast<Atomic32 *>(w)),
+                FUTEX_WAIT | futex_private_flag,
+                value, reinterpret_cast<struct kernel_timespec *>(&tm));
+    } else {
+      nanosleep(&tm, NULL);
+    }
+    errno = save_errno;
+  }
+}
+
+void SpinLockWake(volatile Atomic32 *w, bool all) {
+  if (have_futex) {
+    sys_futex(reinterpret_cast<int *>(const_cast<Atomic32 *>(w)),
+              FUTEX_WAKE | futex_private_flag, all? INT_MAX : 1, 0);
+  }
+}
+
+} // namespace internal
+} // namespace base

BeefRT/gperftools/src/base/spinlock_posix-inl.h

@@ -0,0 +1,62 @@
+/* Copyright (c) 2009, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * This file is a Posix-specific part of spinlock_internal.cc
+ */
+
+#include <config.h>
+#include <errno.h>
+#ifdef HAVE_SCHED_H
+#include <sched.h>      /* For sched_yield() */
+#endif
+#include <time.h>       /* For nanosleep() */
+
+namespace base {
+namespace internal {
+
+void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) {
+  int save_errno = errno;
+  if (loop == 0) {
+  } else if (loop == 1) {
+    sched_yield();
+  } else {
+    struct timespec tm;
+    tm.tv_sec = 0;
+    tm.tv_nsec = base::internal::SuggestedDelayNS(loop);
+    nanosleep(&tm, NULL);
+  }
+  errno = save_errno;
+}
+
+void SpinLockWake(volatile Atomic32 *w, bool all) {
+}
+
+} // namespace internal
+} // namespace base

BeefRT/gperftools/src/base/spinlock_win32-inl.h

@@ -0,0 +1,53 @@
+/* Copyright (c) 2009, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * This file is a Win32-specific part of spinlock_internal.cc
+ */
+
+
+#include <windows.h>
+
+namespace base {
+namespace internal {
+
+void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) {
+  if (loop == 0) {
+  } else if (loop == 1) {
+    Sleep(0);
+  } else {
+    Sleep(base::internal::SuggestedDelayNS(loop) / 1000000);
+  }
+}
+
+void SpinLockWake(volatile Atomic32 *w, bool all) {
+}
+
+} // namespace internal
+} // namespace base

BeefRT/gperftools/src/base/stl_allocator.h

@@ -0,0 +1,97 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Maxim Lifantsev
+ */
+
+
+#ifndef BASE_STL_ALLOCATOR_H_
+#define BASE_STL_ALLOCATOR_H_
+
+#include <config.h>
+
+#include <stddef.h>   // for ptrdiff_t
+#include <limits>
+
+#include "base/logging.h"
+
+// Generic allocator class for STL objects
+// that uses a given type-less allocator Alloc, which must provide:
+//   static void* Alloc::Allocate(size_t size);
+//   static void Alloc::Free(void* ptr, size_t size);
+//
+// STL_Allocator<T, MyAlloc> provides the same thread-safety
+// guarantees as MyAlloc.
+//
+// Usage example:
+//   set<T, less<T>, STL_Allocator<T, MyAlloc> > my_set;
+// CAVEAT: Parts of the code below are probably specific
+//         to the STL version(s) we are using.
+//         The code is simply lifted from what std::allocator<> provides.
+template <typename T, class Alloc>
+class STL_Allocator {
+ public:
+  typedef size_t     size_type;
+  typedef ptrdiff_t  difference_type;
+  typedef T*         pointer;
+  typedef const T*   const_pointer;
+  typedef T&         reference;
+  typedef const T&   const_reference;
+  typedef T          value_type;
+
+  template <class T1> struct rebind {
+    typedef STL_Allocator<T1, Alloc> other;
+  };
+
+  STL_Allocator() { }
+  STL_Allocator(const STL_Allocator&) { }
+  template <class T1> STL_Allocator(const STL_Allocator<T1, Alloc>&) { }
+  ~STL_Allocator() { }
+
+  pointer address(reference x) const { return &x; }
+  const_pointer address(const_reference x) const { return &x; }
+
+  pointer allocate(size_type n, const void* = 0) {
+    RAW_DCHECK((n * sizeof(T)) / sizeof(T) == n, "n is too big to allocate");
+    return static_cast<T*>(Alloc::Allocate(n * sizeof(T)));
+  }
+  void deallocate(pointer p, size_type n) { Alloc::Free(p, n * sizeof(T)); }
+
+  size_type max_size() const { return size_t(-1) / sizeof(T); }
+
+  void construct(pointer p, const T& val) { ::new(p) T(val); }
+  void construct(pointer p) { ::new(p) T(); }
+  void destroy(pointer p) { p->~T(); }
+
+  // There's no state, so these allocators are always equal
+  bool operator==(const STL_Allocator&) const { return true; }
+};
+
+#endif  // BASE_STL_ALLOCATOR_H_

BeefRT/gperftools/src/base/synchronization_profiling.h

@@ -0,0 +1,50 @@
+/* Copyright (c) 2010, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Chris Ruemmler
+ */
+
+#ifndef BASE_AUXILIARY_SYNCHRONIZATION_PROFILING_H_
+#define BASE_AUXILIARY_SYNCHRONIZATION_PROFILING_H_
+
+#include "base/basictypes.h"
+
+namespace base {
+
+// We can do contention-profiling of SpinLocks, but the code is in
+// mutex.cc, which is not always linked in with spinlock.  Hence we
+// provide a weak definition, which are used if mutex.cc isn't linked in.
+
+// Submit the number of cycles the spinlock spent contending.
+ATTRIBUTE_WEAK extern void SubmitSpinLockProfileData(const void *, int64);
+extern void SubmitSpinLockProfileData(const void *contendedlock,
+                                      int64 wait_cycles) {}
+}
+#endif  // BASE_AUXILIARY_SYNCHRONIZATION_PROFILING_H_

BeefRT/gperftools/src/base/sysinfo.h

@@ -0,0 +1,235 @@
+// Copyright (c) 2006, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// All functions here are thread-hostile due to file caching unless
+// commented otherwise.
+
+#ifndef _SYSINFO_H_
+#define _SYSINFO_H_
+
+#include <config.h>
+
+#include <time.h>
+#if (defined(_WIN32) || defined(__MINGW32__)) && (!defined(__CYGWIN__) && !defined(__CYGWIN32__))
+#include <windows.h>   // for DWORD
+#include <tlhelp32.h>  // for CreateToolhelp32Snapshot
+#endif
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>    // for pid_t
+#endif
+#include <stddef.h>    // for size_t
+#include <limits.h>    // for PATH_MAX
+#include "base/basictypes.h"
+#include "base/logging.h"   // for RawFD
+
+// This getenv function is safe to call before the C runtime is initialized.
+// On Windows, it utilizes GetEnvironmentVariable() and on unix it uses
+// /proc/self/environ instead calling getenv().  It's intended to be used in
+// routines that run before main(), when the state required for getenv() may
+// not be set up yet.  In particular, errno isn't set up until relatively late
+// (after the pthreads library has a chance to make it threadsafe), and
+// getenv() doesn't work until then. 
+// On some platforms, this call will utilize the same, static buffer for
+// repeated GetenvBeforeMain() calls. Callers should not expect pointers from
+// this routine to be long lived.
+// Note that on unix, /proc only has the environment at the time the
+// application was started, so this routine ignores setenv() calls/etc.  Also
+// note it only reads the first 16K of the environment.
+extern const char* GetenvBeforeMain(const char* name);
+
+// This takes as an argument an environment-variable name (like
+// CPUPROFILE) whose value is supposed to be a file-path, and sets
+// path to that path, and returns true.  Non-trivial for surprising
+// reasons, as documented in sysinfo.cc.  path must have space PATH_MAX.
+extern bool GetUniquePathFromEnv(const char* env_name, char* path);
+
+extern int NumCPUs();
+
+void SleepForMilliseconds(int milliseconds);
+
+// processor cycles per second of each processor.  Thread-safe.
+extern double CyclesPerSecond(void);
+
+
+//  Return true if we're running POSIX (e.g., NPTL on Linux) threads,
+//  as opposed to a non-POSIX thread library.  The thing that we care
+//  about is whether a thread's pid is the same as the thread that
+//  spawned it.  If so, this function returns true.
+//  Thread-safe.
+//  Note: We consider false negatives to be OK.
+bool HasPosixThreads();
+
+#ifndef SWIG  // SWIG doesn't like struct Buffer and variable arguments.
+
+// A ProcMapsIterator abstracts access to /proc/maps for a given
+// process. Needs to be stack-allocatable and avoid using stdio/malloc
+// so it can be used in the google stack dumper, heap-profiler, etc.
+//
+// On Windows and Mac OS X, this iterator iterates *only* over DLLs
+// mapped into this process space.  For Linux, FreeBSD, and Solaris,
+// it iterates over *all* mapped memory regions, including anonymous
+// mmaps.  For other O/Ss, it is unlikely to work at all, and Valid()
+// will always return false.  Also note: this routine only works on
+// FreeBSD if procfs is mounted: make sure this is in your /etc/fstab:
+//    proc            /proc   procfs  rw 0 0
+class ProcMapsIterator {
+ public:
+  struct Buffer {
+#ifdef __FreeBSD__
+    // FreeBSD requires us to read all of the maps file at once, so
+    // we have to make a buffer that's "always" big enough
+    static const size_t kBufSize = 102400;
+#else   // a one-line buffer is good enough
+    static const size_t kBufSize = PATH_MAX + 1024;
+#endif
+    char buf_[kBufSize];
+  };
+
+
+  // Create a new iterator for the specified pid.  pid can be 0 for "self".
+  explicit ProcMapsIterator(pid_t pid);
+
+  // Create an iterator with specified storage (for use in signal
+  // handler). "buffer" should point to a ProcMapsIterator::Buffer
+  // buffer can be NULL in which case a bufer will be allocated.
+  ProcMapsIterator(pid_t pid, Buffer *buffer);
+
+  // Iterate through maps_backing instead of maps if use_maps_backing
+  // is true.  Otherwise the same as above.  buffer can be NULL and
+  // it will allocate a buffer itself.
+  ProcMapsIterator(pid_t pid, Buffer *buffer,
+                   bool use_maps_backing);
+
+  // Returns true if the iterator successfully initialized;
+  bool Valid() const;
+
+  // Returns a pointer to the most recently parsed line. Only valid
+  // after Next() returns true, and until the iterator is destroyed or
+  // Next() is called again.  This may give strange results on non-Linux
+  // systems.  Prefer FormatLine() if that may be a concern.
+  const char *CurrentLine() const { return stext_; }
+
+  // Writes the "canonical" form of the /proc/xxx/maps info for a single
+  // line to the passed-in buffer. Returns the number of bytes written,
+  // or 0 if it was not able to write the complete line.  (To guarantee
+  // success, buffer should have size at least Buffer::kBufSize.)
+  // Takes as arguments values set via a call to Next().  The
+  // "canonical" form of the line (taken from linux's /proc/xxx/maps):
+  //    <start_addr(hex)>-<end_addr(hex)> <perms(rwxp)> <offset(hex)>   +
+  //    <major_dev(hex)>:<minor_dev(hex)> <inode> <filename> Note: the
+  // eg
+  //    08048000-0804c000 r-xp 00000000 03:01 3793678    /bin/cat
+  // If you don't have the dev_t (dev), feel free to pass in 0.
+  // (Next() doesn't return a dev_t, though NextExt does.)
+  //
+  // Note: if filename and flags were obtained via a call to Next(),
+  // then the output of this function is only valid if Next() returned
+  // true, and only until the iterator is destroyed or Next() is
+  // called again.  (Since filename, at least, points into CurrentLine.)
+  static int FormatLine(char* buffer, int bufsize,
+                        uint64 start, uint64 end, const char *flags,
+                        uint64 offset, int64 inode, const char *filename,
+                        dev_t dev);
+
+  // Find the next entry in /proc/maps; return true if found or false
+  // if at the end of the file.
+  //
+  // Any of the result pointers can be NULL if you're not interested
+  // in those values.
+  //
+  // If "flags" and "filename" are passed, they end up pointing to
+  // storage within the ProcMapsIterator that is valid only until the
+  // iterator is destroyed or Next() is called again. The caller may
+  // modify the contents of these strings (up as far as the first NUL,
+  // and only until the subsequent call to Next()) if desired.
+
+  // The offsets are all uint64 in order to handle the case of a
+  // 32-bit process running on a 64-bit kernel
+  //
+  // IMPORTANT NOTE: see top-of-class notes for details about what
+  // mapped regions Next() iterates over, depending on O/S.
+  // TODO(csilvers): make flags and filename const.
+  bool Next(uint64 *start, uint64 *end, char **flags,
+            uint64 *offset, int64 *inode, char **filename);
+
+  bool NextExt(uint64 *start, uint64 *end, char **flags,
+               uint64 *offset, int64 *inode, char **filename,
+               uint64 *file_mapping, uint64 *file_pages,
+               uint64 *anon_mapping, uint64 *anon_pages,
+               dev_t *dev);
+
+  ~ProcMapsIterator();
+
+ private:
+  void Init(pid_t pid, Buffer *buffer, bool use_maps_backing);
+
+  char *ibuf_;        // input buffer
+  char *stext_;       // start of text
+  char *etext_;       // end of text
+  char *nextline_;    // start of next line
+  char *ebuf_;        // end of buffer (1 char for a nul)
+#if (defined(_WIN32) || defined(__MINGW32__)) && (!defined(__CYGWIN__) && !defined(__CYGWIN32__))
+  HANDLE snapshot_;   // filehandle on dll info
+  // In a change from the usual W-A pattern, there is no A variant of
+  // MODULEENTRY32.  Tlhelp32.h #defines the W variant, but not the A.
+  // We want the original A variants, and this #undef is the only
+  // way I see to get them.  Redefining it when we're done prevents us
+  // from affecting other .cc files.
+# ifdef MODULEENTRY32  // Alias of W
+#   undef MODULEENTRY32
+  MODULEENTRY32 module_;   // info about current dll (and dll iterator)
+#   define MODULEENTRY32 MODULEENTRY32W
+# else  // It's the ascii, the one we want.
+  MODULEENTRY32 module_;   // info about current dll (and dll iterator)
+# endif
+#elif defined(__MACH__)
+  int current_image_; // dll's are called "images" in macos parlance
+  int current_load_cmd_;   // the segment of this dll we're examining
+#elif defined(__sun__)     // Solaris
+  int fd_;
+  char current_filename_[PATH_MAX];
+#else
+  int fd_;            // filehandle on /proc/*/maps
+#endif
+  pid_t pid_;
+  char flags_[10];
+  Buffer* dynamic_buffer_;  // dynamically-allocated Buffer
+  bool using_maps_backing_; // true if we are looking at maps_backing instead of maps.
+};
+
+#endif  /* #ifndef SWIG */
+
+// Helper routines
+
+namespace TCMALLOC_NAMESPACE {
+int FillProcSelfMaps(char buf[], int size, bool* wrote_all);
+void DumpProcSelfMaps(RawFD fd);
+}
+
+#endif   /* #ifndef _SYSINFO_H_ */

BeefRT/gperftools/src/base/thread_annotations.h

@@ -0,0 +1,134 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Le-Chun Wu
+//
+// This header file contains the macro definitions for thread safety
+// annotations that allow the developers to document the locking policies
+// of their multi-threaded code. The annotations can also help program
+// analysis tools to identify potential thread safety issues.
+//
+// The annotations are implemented using GCC's "attributes" extension.
+// Using the macros defined here instead of the raw GCC attributes allows
+// for portability and future compatibility.
+//
+// This functionality is not yet fully implemented in perftools,
+// but may be one day.
+
+#ifndef BASE_THREAD_ANNOTATIONS_H_
+#define BASE_THREAD_ANNOTATIONS_H_
+
+
+#if defined(__GNUC__) \
+  && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)) \
+  && defined(__SUPPORT_TS_ANNOTATION__) && (!defined(SWIG))
+#define THREAD_ANNOTATION_ATTRIBUTE__(x)   __attribute__((x))
+#else
+#define THREAD_ANNOTATION_ATTRIBUTE__(x)   // no-op
+#endif
+
+
+// Document if a shared variable/field needs to be protected by a lock.
+// GUARDED_BY allows the user to specify a particular lock that should be
+// held when accessing the annotated variable, while GUARDED_VAR only
+// indicates a shared variable should be guarded (by any lock). GUARDED_VAR
+// is primarily used when the client cannot express the name of the lock.
+#define GUARDED_BY(x)          THREAD_ANNOTATION_ATTRIBUTE__(guarded_by(x))
+#define GUARDED_VAR            THREAD_ANNOTATION_ATTRIBUTE__(guarded)
+
+// Document if the memory location pointed to by a pointer should be guarded
+// by a lock when dereferencing the pointer. Similar to GUARDED_VAR,
+// PT_GUARDED_VAR is primarily used when the client cannot express the name
+// of the lock. Note that a pointer variable to a shared memory location
+// could itself be a shared variable. For example, if a shared global pointer
+// q, which is guarded by mu1, points to a shared memory location that is
+// guarded by mu2, q should be annotated as follows:
+//     int *q GUARDED_BY(mu1) PT_GUARDED_BY(mu2);
+#define PT_GUARDED_BY(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded_by(x))
+#define PT_GUARDED_VAR \
+  THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded)
+
+// Document the acquisition order between locks that can be held
+// simultaneously by a thread. For any two locks that need to be annotated
+// to establish an acquisition order, only one of them needs the annotation.
+// (i.e. You don't have to annotate both locks with both ACQUIRED_AFTER
+// and ACQUIRED_BEFORE.)
+#define ACQUIRED_AFTER(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(acquired_after(x))
+#define ACQUIRED_BEFORE(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(acquired_before(x))
+
+// The following three annotations document the lock requirements for
+// functions/methods.
+
+// Document if a function expects certain locks to be held before it is called
+#define EXCLUSIVE_LOCKS_REQUIRED(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(exclusive_locks_required(x))
+
+#define SHARED_LOCKS_REQUIRED(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(shared_locks_required(x))
+
+// Document the locks acquired in the body of the function. These locks
+// cannot be held when calling this function (as google3's Mutex locks are
+// non-reentrant).
+#define LOCKS_EXCLUDED(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(locks_excluded(x))
+
+// Document the lock the annotated function returns without acquiring it.
+#define LOCK_RETURNED(x)       THREAD_ANNOTATION_ATTRIBUTE__(lock_returned(x))
+
+// Document if a class/type is a lockable type (such as the Mutex class).
+#define LOCKABLE               THREAD_ANNOTATION_ATTRIBUTE__(lockable)
+
+// Document if a class is a scoped lockable type (such as the MutexLock class).
+#define SCOPED_LOCKABLE        THREAD_ANNOTATION_ATTRIBUTE__(scoped_lockable)
+
+// The following annotations specify lock and unlock primitives.
+#define EXCLUSIVE_LOCK_FUNCTION(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(exclusive_lock(x))
+
+#define SHARED_LOCK_FUNCTION(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(shared_lock(x))
+
+#define EXCLUSIVE_TRYLOCK_FUNCTION(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(exclusive_trylock(x))
+
+#define SHARED_TRYLOCK_FUNCTION(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(shared_trylock(x))
+
+#define UNLOCK_FUNCTION(x) \
+  THREAD_ANNOTATION_ATTRIBUTE__(unlock(x))
+
+// An escape hatch for thread safety analysis to ignore the annotated function.
+#define NO_THREAD_SAFETY_ANALYSIS \
+  THREAD_ANNOTATION_ATTRIBUTE__(no_thread_safety_analysis)
+
+#endif  // BASE_THREAD_ANNOTATIONS_H_

BeefRT/gperftools/src/base/thread_lister.c

@@ -0,0 +1,77 @@
+/* Copyright (c) 2005-2007, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Markus Gutschke
+ */
+
+#include "config.h"
+#include <stdio.h>         /* needed for NULL on some powerpc platforms (?!) */
+#ifdef HAVE_SYS_PRCTL
+# include <sys/prctl.h>
+#endif
+#include "base/thread_lister.h"
+#include "base/linuxthreads.h"
+/* Include other thread listers here that define THREADS macro
+ * only when they can provide a good implementation.
+ */
+
+#ifndef THREADS
+
+/* Default trivial thread lister for single-threaded applications,
+ * or if the multi-threading code has not been ported, yet.
+ */
+
+int ListAllProcessThreads(void *parameter,
+                          ListAllProcessThreadsCallBack callback, ...) {
+  int rc;
+  va_list ap;
+  pid_t pid;
+
+#ifdef HAVE_SYS_PRCTL
+  int dumpable = prctl(PR_GET_DUMPABLE, 0);
+  if (!dumpable)
+    prctl(PR_SET_DUMPABLE, 1);
+#endif
+  va_start(ap, callback);
+  pid = getpid();
+  rc = callback(parameter, 1, &pid, ap);
+  va_end(ap);
+#ifdef HAVE_SYS_PRCTL
+  if (!dumpable)
+    prctl(PR_SET_DUMPABLE, 0);
+#endif
+  return rc;
+}
+
+int ResumeAllProcessThreads(int num_threads, pid_t *thread_pids) {
+  return 1;
+}
+
+#endif   /* ifndef THREADS */

BeefRT/gperftools/src/base/thread_lister.h

@@ -0,0 +1,82 @@
+/* Copyright (c) 2005-2007, Google Inc.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Markus Gutschke
+ */
+
+#ifndef _THREAD_LISTER_H
+#define _THREAD_LISTER_H
+
+#include <stdarg.h>
+#include <sys/types.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef int (*ListAllProcessThreadsCallBack)(void *parameter,
+                                             int num_threads,
+                                             pid_t *thread_pids,
+                                             va_list ap);
+
+/* This function gets the list of all linux threads of the current process
+ * passes them to the 'callback' along with the 'parameter' pointer; at the
+ * call back call time all the threads are paused via
+ * PTRACE_ATTACH.
+ * The callback is executed from a separate thread which shares only the
+ * address space, the filesystem, and the filehandles with the caller. Most
+ * notably, it does not share the same pid and ppid; and if it terminates,
+ * the rest of the application is still there. 'callback' is supposed to do
+ * or arrange for ResumeAllProcessThreads. This happens automatically, if
+ * the thread raises a synchronous signal (e.g. SIGSEGV); asynchronous
+ * signals are blocked. If the 'callback' decides to unblock them, it must
+ * ensure that they cannot terminate the application, or that
+ * ResumeAllProcessThreads will get called.
+ * It is an error for the 'callback' to make any library calls that could
+ * acquire locks. Most notably, this means that most system calls have to
+ * avoid going through libc. Also, this means that it is not legal to call
+ * exit() or abort().
+ * We return -1 on error and the return value of 'callback' on success.
+ */
+int ListAllProcessThreads(void *parameter,
+                          ListAllProcessThreadsCallBack callback, ...);
+
+/* This function resumes the list of all linux threads that
+ * ListAllProcessThreads pauses before giving to its callback.
+ * The function returns non-zero if at least one thread was
+ * suspended and has now been resumed.
+ */
+int ResumeAllProcessThreads(int num_threads, pid_t *thread_pids);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif  /* _THREAD_LISTER_H */

BeefRT/gperftools/src/base/vdso_support.cc

@@ -0,0 +1,189 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Pluzhnikov
+//
+// Allow dynamic symbol lookup in the kernel VDSO page.
+//
+// VDSOSupport -- a class representing kernel VDSO (if present).
+//
+
+#include "base/vdso_support.h"
+
+#ifdef HAVE_VDSO_SUPPORT     // defined in vdso_support.h
+
+#include <fcntl.h>
+#include <stddef.h>   // for ptrdiff_t
+
+#include "base/atomicops.h"  // for MemoryBarrier
+#include "base/linux_syscall_support.h"
+#include "base/logging.h"
+#include "base/dynamic_annotations.h"
+#include "base/basictypes.h"  // for COMPILE_ASSERT
+
+using base::subtle::MemoryBarrier;
+
+#ifndef AT_SYSINFO_EHDR
+#define AT_SYSINFO_EHDR 33
+#endif
+
+namespace base {
+
+const void *VDSOSupport::vdso_base_ = ElfMemImage::kInvalidBase;
+VDSOSupport::GetCpuFn VDSOSupport::getcpu_fn_ = &InitAndGetCPU;
+VDSOSupport::VDSOSupport()
+    // If vdso_base_ is still set to kInvalidBase, we got here
+    // before VDSOSupport::Init has been called. Call it now.
+    : image_(vdso_base_ == ElfMemImage::kInvalidBase ? Init() : vdso_base_) {
+}
+
+// NOTE: we can't use GoogleOnceInit() below, because we can be
+// called by tcmalloc, and none of the *once* stuff may be functional yet.
+//
+// In addition, we hope that the VDSOSupportHelper constructor
+// causes this code to run before there are any threads, and before
+// InitGoogle() has executed any chroot or setuid calls.
+//
+// Finally, even if there is a race here, it is harmless, because
+// the operation should be idempotent.
+const void *VDSOSupport::Init() {
+  if (vdso_base_ == ElfMemImage::kInvalidBase) {
+    // Valgrind zaps AT_SYSINFO_EHDR and friends from the auxv[]
+    // on stack, and so glibc works as if VDSO was not present.
+    // But going directly to kernel via /proc/self/auxv below bypasses
+    // Valgrind zapping. So we check for Valgrind separately.
+    if (RunningOnValgrind()) {
+      vdso_base_ = NULL;
+      getcpu_fn_ = &GetCPUViaSyscall;
+      return NULL;
+    }
+    int fd = open("/proc/self/auxv", O_RDONLY);
+    if (fd == -1) {
+      // Kernel too old to have a VDSO.
+      vdso_base_ = NULL;
+      getcpu_fn_ = &GetCPUViaSyscall;
+      return NULL;
+    }
+    ElfW(auxv_t) aux;
+    while (read(fd, &aux, sizeof(aux)) == sizeof(aux)) {
+      if (aux.a_type == AT_SYSINFO_EHDR) {
+        COMPILE_ASSERT(sizeof(vdso_base_) == sizeof(aux.a_un.a_val),
+                       unexpected_sizeof_pointer_NE_sizeof_a_val);
+        vdso_base_ = reinterpret_cast<void *>(aux.a_un.a_val);
+        break;
+      }
+    }
+    close(fd);
+    if (vdso_base_ == ElfMemImage::kInvalidBase) {
+      // Didn't find AT_SYSINFO_EHDR in auxv[].
+      vdso_base_ = NULL;
+    }
+  }
+  GetCpuFn fn = &GetCPUViaSyscall;  // default if VDSO not present.
+  if (vdso_base_) {
+    VDSOSupport vdso;
+    SymbolInfo info;
+    if (vdso.LookupSymbol("__vdso_getcpu", "LINUX_2.6", STT_FUNC, &info)) {
+      // Casting from an int to a pointer is not legal C++.  To emphasize
+      // this, we use a C-style cast rather than a C++-style cast.
+      fn = (GetCpuFn)(info.address);
+    }
+  }
+  // Subtle: this code runs outside of any locks; prevent compiler
+  // from assigning to getcpu_fn_ more than once.
+  base::subtle::MemoryBarrier();
+  getcpu_fn_ = fn;
+  return vdso_base_;
+}
+
+const void *VDSOSupport::SetBase(const void *base) {
+  CHECK(base != ElfMemImage::kInvalidBase);
+  const void *old_base = vdso_base_;
+  vdso_base_ = base;
+  image_.Init(base);
+  // Also reset getcpu_fn_, so GetCPU could be tested with simulated VDSO.
+  getcpu_fn_ = &InitAndGetCPU;
+  return old_base;
+}
+
+bool VDSOSupport::LookupSymbol(const char *name,
+                               const char *version,
+                               int type,
+                               SymbolInfo *info) const {
+  return image_.LookupSymbol(name, version, type, info);
+}
+
+bool VDSOSupport::LookupSymbolByAddress(const void *address,
+                                        SymbolInfo *info_out) const {
+  return image_.LookupSymbolByAddress(address, info_out);
+}
+
+// NOLINT on 'long' because this routine mimics kernel api.
+long VDSOSupport::GetCPUViaSyscall(unsigned *cpu, void *, void *) { // NOLINT
+#if defined(__NR_getcpu)
+  return sys_getcpu(cpu, NULL, NULL);
+#else
+  // x86_64 never implemented sys_getcpu(), except as a VDSO call.
+  errno = ENOSYS;
+  return -1;
+#endif
+}
+
+// Use fast __vdso_getcpu if available.
+long VDSOSupport::InitAndGetCPU(unsigned *cpu, void *x, void *y) { // NOLINT
+  Init();
+  CHECK_NE(getcpu_fn_, &InitAndGetCPU); // << "Init() did not set getcpu_fn_";
+  return (*getcpu_fn_)(cpu, x, y);
+}
+
+// This function must be very fast, and may be called from very
+// low level (e.g. tcmalloc). Hence I avoid things like
+// GoogleOnceInit() and ::operator new.
+int GetCPU(void) {
+  unsigned cpu;
+  int ret_code = (*VDSOSupport::getcpu_fn_)(&cpu, NULL, NULL);
+  return ret_code == 0 ? cpu : ret_code;
+}
+
+// We need to make sure VDSOSupport::Init() is called before
+// the main() runs, since it might do something like setuid or
+// chroot.  If VDSOSupport
+// is used in any global constructor, this will happen, since
+// VDSOSupport's constructor calls Init.  But if not, we need to
+// ensure it here, with a global constructor of our own.  This
+// is an allowed exception to the normal rule against non-trivial
+// global constructors.
+static class VDSOInitHelper {
+ public:
+  VDSOInitHelper() { VDSOSupport::Init(); }
+} vdso_init_helper;
+}
+
+#endif  // HAVE_VDSO_SUPPORT

BeefRT/gperftools/src/base/vdso_support.h

@@ -0,0 +1,155 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Pluzhnikov
+//
+// Allow dynamic symbol lookup in the kernel VDSO page.
+//
+// VDSO stands for "Virtual Dynamic Shared Object" -- a page of
+// executable code, which looks like a shared library, but doesn't
+// necessarily exist anywhere on disk, and which gets mmap()ed into
+// every process by kernels which support VDSO, such as 2.6.x for 32-bit
+// executables, and 2.6.24 and above for 64-bit executables.
+//
+// More details could be found here:
+// http://www.trilithium.com/johan/2005/08/linux-gate/
+//
+// VDSOSupport -- a class representing kernel VDSO (if present).
+//
+// Example usage:
+//  VDSOSupport vdso;
+//  VDSOSupport::SymbolInfo info;
+//  typedef (*FN)(unsigned *, void *, void *);
+//  FN fn = NULL;
+//  if (vdso.LookupSymbol("__vdso_getcpu", "LINUX_2.6", STT_FUNC, &info)) {
+//     fn = reinterpret_cast<FN>(info.address);
+//  }
+
+#ifndef BASE_VDSO_SUPPORT_H_
+#define BASE_VDSO_SUPPORT_H_
+
+#include <config.h>
+#include "base/basictypes.h"
+#include "base/elf_mem_image.h"
+
+#ifdef HAVE_ELF_MEM_IMAGE
+
+#define HAVE_VDSO_SUPPORT 1
+
+#include <stdlib.h>     // for NULL
+
+namespace base {
+
+// NOTE: this class may be used from within tcmalloc, and can not
+// use any memory allocation routines.
+class VDSOSupport {
+ public:
+  VDSOSupport();
+
+  typedef ElfMemImage::SymbolInfo SymbolInfo;
+  typedef ElfMemImage::SymbolIterator SymbolIterator;
+
+  // Answers whether we have a vdso at all.
+  bool IsPresent() const { return image_.IsPresent(); }
+
+  // Allow to iterate over all VDSO symbols.
+  SymbolIterator begin() const { return image_.begin(); }
+  SymbolIterator end() const { return image_.end(); }
+
+  // Look up versioned dynamic symbol in the kernel VDSO.
+  // Returns false if VDSO is not present, or doesn't contain given
+  // symbol/version/type combination.
+  // If info_out != NULL, additional details are filled in.
+  bool LookupSymbol(const char *name, const char *version,
+                    int symbol_type, SymbolInfo *info_out) const;
+
+  // Find info about symbol (if any) which overlaps given address.
+  // Returns true if symbol was found; false if VDSO isn't present
+  // or doesn't have a symbol overlapping given address.
+  // If info_out != NULL, additional details are filled in.
+  bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const;
+
+  // Used only for testing. Replace real VDSO base with a mock.
+  // Returns previous value of vdso_base_. After you are done testing,
+  // you are expected to call SetBase() with previous value, in order to
+  // reset state to the way it was.
+  const void *SetBase(const void *s);
+
+  // Computes vdso_base_ and returns it. Should be called as early as
+  // possible; before any thread creation, chroot or setuid.
+  static const void *Init();
+
+ private:
+  // image_ represents VDSO ELF image in memory.
+  // image_.ehdr_ == NULL implies there is no VDSO.
+  ElfMemImage image_;
+
+  // Cached value of auxv AT_SYSINFO_EHDR, computed once.
+  // This is a tri-state:
+  //   kInvalidBase   => value hasn't been determined yet.
+  //              0   => there is no VDSO.
+  //           else   => vma of VDSO Elf{32,64}_Ehdr.
+  //
+  // When testing with mock VDSO, low bit is set.
+  // The low bit is always available because vdso_base_ is
+  // page-aligned.
+  static const void *vdso_base_;
+
+  // NOLINT on 'long' because these routines mimic kernel api.
+  // The 'cache' parameter may be used by some versions of the kernel,
+  // and should be NULL or point to a static buffer containing at
+  // least two 'long's.
+  static long InitAndGetCPU(unsigned *cpu, void *cache,     // NOLINT 'long'.
+                            void *unused);
+  static long GetCPUViaSyscall(unsigned *cpu, void *cache,  // NOLINT 'long'.
+                               void *unused);
+  typedef long (*GetCpuFn)(unsigned *cpu, void *cache,      // NOLINT 'long'.
+                           void *unused);
+
+  // This function pointer may point to InitAndGetCPU,
+  // GetCPUViaSyscall, or __vdso_getcpu at different stages of initialization.
+  static GetCpuFn getcpu_fn_;
+
+  friend int GetCPU(void);  // Needs access to getcpu_fn_.
+
+  DISALLOW_COPY_AND_ASSIGN(VDSOSupport);
+};
+
+// Same as sched_getcpu() on later glibc versions.
+// Return current CPU, using (fast) __vdso_getcpu@LINUX_2.6 if present,
+// otherwise use syscall(SYS_getcpu,...).
+// May return -1 with errno == ENOSYS if the kernel doesn't
+// support SYS_getcpu.
+int GetCPU();
+}  // namespace base
+
+#endif  // HAVE_ELF_MEM_IMAGE
+
+#endif  // BASE_VDSO_SUPPORT_H_

BeefRT/gperftools/src/central_freelist.cc

@@ -0,0 +1,442 @@
+// Copyright (c) 2008, Google Inc. 
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com> 
+
+#include "tcconfig.h" 
+#include <algorithm>
+#include "central_freelist.h"
+#include "internal_logging.h"  // for ASSERT, MESSAGE
+#include "linked_list.h"       // for SLL_Next, SLL_Push, etc
+#include "page_heap.h"         // for PageHeap
+#include "static_vars.h"       // for Static
+
+#include "BeefySysLib/Common.h"
+
+using std::min;
+using std::max;
+
+namespace TCMALLOC_NAMESPACE {
+
+void CentralFreeList::Init(size_t cl) {
+  size_class_ = cl;
+  TCMALLOC_NAMESPACE::DLL_Init(&empty_);
+  TCMALLOC_NAMESPACE::DLL_Init(&nonempty_);
+  num_spans_ = 0;
+  counter_ = 0;
+
+  max_cache_size_ = kMaxNumTransferEntries;
+#ifdef TCMALLOC_SMALL_BUT_SLOW
+  // Disable the transfer cache for the small footprint case.
+  cache_size_ = 0;
+#else
+  cache_size_ = 16;
+#endif
+  if (cl > 0) {
+    // Limit the maximum size of the cache based on the size class.  If this
+    // is not done, large size class objects will consume a lot of memory if
+    // they just sit in the transfer cache.
+    int32_t bytes = Static::sizemap()->ByteSizeForClass(cl);
+    int32_t objs_to_move = Static::sizemap()->num_objects_to_move(cl);
+
+    ASSERT(objs_to_move > 0 && bytes > 0);
+    // Limit each size class cache to at most 1MB of objects or one entry,
+    // whichever is greater. Total transfer cache memory used across all
+    // size classes then can't be greater than approximately
+    // 1MB * kMaxNumTransferEntries.
+    // min and max are in parens to avoid macro-expansion on windows.
+    max_cache_size_ = (min)(max_cache_size_,
+                          (max)(1, (1024 * 1024) / (bytes * objs_to_move)));
+    cache_size_ = (min)(cache_size_, max_cache_size_);
+  }
+  used_slots_ = 0;
+  ASSERT(cache_size_ <= max_cache_size_);
+}
+
+void CentralFreeList::ReleaseListToSpans(void* start) {
+  while (start) {
+    void *next = SLL_Next(start);
+    ReleaseToSpans(start);
+    start = next;
+  }
+}
+
+// MapObjectToSpan should logically be part of ReleaseToSpans.  But
+// this triggers an optimization bug in gcc 4.5.0.  Moving to a
+// separate function, and making sure that function isn't inlined,
+// seems to fix the problem.  It also should be fixed for gcc 4.5.1.
+static
+#if __GNUC__ == 4 && __GNUC_MINOR__ == 5 && __GNUC_PATCHLEVEL__ == 0
+__attribute__ ((noinline))
+#endif
+Span* MapObjectToSpan(void* object) {
+  const PageID p = reinterpret_cast<uintptr_t>(object) >> kPageShift;
+  Span* span = Static::pageheap()->GetDescriptor(p);
+  return span;
+}
+
+void CentralFreeList::ReleaseToSpans(void* object) {
+  Span* span = MapObjectToSpan(object);
+  ASSERT(span != NULL);
+  ASSERT(span->refcount > 0);
+
+  // If span is empty, move it to non-empty list
+  if (span->objects == NULL) {
+    TCMALLOC_NAMESPACE::DLL_Remove(span);
+    TCMALLOC_NAMESPACE::DLL_Prepend(&nonempty_, span);
+    Event(span, 'N', 0);
+  }
+
+  // The following check is expensive, so it is disabled by default
+  if (false) {
+    // Check that object does not occur in list
+    int got = 0;
+    for (void* p = span->objects; p != NULL; p = *((void**) p)) {
+      ASSERT(p != object);
+      got++;
+    }
+    ASSERT(got + span->refcount ==
+           (span->length<<kPageShift) /
+           Static::sizemap()->ByteSizeForClass(span->sizeclass));
+  }
+
+  counter_++;
+  span->refcount--;
+  if (span->refcount == 0) {
+    Event(span, '#', 0);
+    counter_ -= ((span->length<<kPageShift) /
+                 Static::sizemap()->ByteSizeForClass(span->sizeclass));
+    TCMALLOC_NAMESPACE::DLL_Remove(span);
+    --num_spans_;
+
+    // Release central list lock while operating on pageheap
+    lock_.Unlock();
+    {
+      SpinLockHolder h(Static::pageheap_lock());
+      Static::pageheap()->Delete(span);
+    }
+    lock_.Lock();
+  } else {
+    *(reinterpret_cast<void**>(object)) = span->objects;
+    span->objects = object;
+  }
+}
+    
+void CentralFreeList::ReleasePendingSpanObjects(Span* span) {
+    ASSERT(span != NULL);
+    
+    int freeingCount = *((intptr_t*)span->freeingObjectsTail);
+    
+    ASSERT(span->refcount >= freeingCount);
+    
+    SpinLockHolder h(&lock_);
+    // If span is empty, move it to non-empty list
+    if (span->objects == NULL) {
+        TCMALLOC_NAMESPACE::DLL_Remove(span);
+        TCMALLOC_NAMESPACE::DLL_Prepend(&nonempty_, span);
+        Event(span, 'N', 0);
+    }
+    
+    counter_ += freeingCount;
+    span->refcount -= freeingCount;
+    if (span->refcount == 0) {
+        Event(span, '#', 0);
+        counter_ -= ((span->length<<kPageShift) /
+                     Static::sizemap()->ByteSizeForClass(span->sizeclass));
+        TCMALLOC_NAMESPACE::DLL_Remove(span);
+        --num_spans_;
+        
+        // Release central list lock while operating on pageheap
+        lock_.Unlock();
+        {
+            SpinLockHolder h(Static::pageheap_lock());
+            Static::pageheap()->Delete(span, true);
+        }
+        lock_.Lock();
+    } else {
+        *(reinterpret_cast<void**>(span->freeingObjectsTail)) = span->objects;
+        span->objects = span->freeingObjects;
+    }
+}
+
+
+bool CentralFreeList::EvictRandomSizeClass(
+    int locked_size_class, bool force) {
+  static int race_counter = 0;
+  int t = race_counter++;  // Updated without a lock, but who cares.
+  if (t >= kNumClasses) {
+    while (t >= kNumClasses) {
+      t -= kNumClasses;
+    }
+    race_counter = t;
+  }
+  ASSERT(t >= 0);
+  ASSERT(t < kNumClasses);
+  if (t == locked_size_class) return false;
+  return Static::central_cache()[t].ShrinkCache(locked_size_class, force);
+}
+
+bool CentralFreeList::MakeCacheSpace() {
+  // Is there room in the cache?
+  if (used_slots_ < cache_size_) return true;
+  // Check if we can expand this cache?
+  if (cache_size_ == max_cache_size_) return false;
+  // Ok, we'll try to grab an entry from some other size class.
+  if (EvictRandomSizeClass(size_class_, false) ||
+      EvictRandomSizeClass(size_class_, true)) {
+    // Succeeded in evicting, we're going to make our cache larger.
+    // However, we may have dropped and re-acquired the lock in
+    // EvictRandomSizeClass (via ShrinkCache and the LockInverter), so the
+    // cache_size may have changed.  Therefore, check and verify that it is
+    // still OK to increase the cache_size.
+    if (cache_size_ < max_cache_size_) {
+      cache_size_++;
+      return true;
+    }
+  }
+  return false;
+}
+
+
+namespace {
+class LockInverter {
+ private:
+  SpinLock *held_, *temp_;
+ public:
+  inline explicit LockInverter(SpinLock* held, SpinLock *temp)
+    : held_(held), temp_(temp) { held_->Unlock(); temp_->Lock(); }
+  inline ~LockInverter() { temp_->Unlock(); held_->Lock();  }
+};
+}
+
+// This function is marked as NO_THREAD_SAFETY_ANALYSIS because it uses
+// LockInverter to release one lock and acquire another in scoped-lock
+// style, which our current annotation/analysis does not support.
+bool CentralFreeList::ShrinkCache(int locked_size_class, bool force)
+    NO_THREAD_SAFETY_ANALYSIS {
+  // Start with a quick check without taking a lock.
+  if (cache_size_ == 0) return false;
+  // We don't evict from a full cache unless we are 'forcing'.
+  if (force == false && used_slots_ == cache_size_) return false;
+
+  // Grab lock, but first release the other lock held by this thread.  We use
+  // the lock inverter to ensure that we never hold two size class locks
+  // concurrently.  That can create a deadlock because there is no well
+  // defined nesting order.
+  LockInverter li(&Static::central_cache()[locked_size_class].lock_, &lock_);
+  ASSERT(used_slots_ <= cache_size_);
+  ASSERT(0 <= cache_size_);
+  if (cache_size_ == 0) return false;
+  if (used_slots_ == cache_size_) {
+    if (force == false) return false;
+    // ReleaseListToSpans releases the lock, so we have to make all the
+    // updates to the central list before calling it.
+    cache_size_--;
+    used_slots_--;
+    ReleaseListToSpans(tc_slots_[used_slots_].head);
+    return true;
+  }
+  cache_size_--;
+  return true;
+}
+
+void CentralFreeList::InsertRange(void *start, void *end, int N) {
+  SpinLockHolder h(&lock_);
+  if (N == Static::sizemap()->num_objects_to_move(size_class_) &&
+    MakeCacheSpace()) {
+    int slot = used_slots_++;
+    ASSERT(slot >=0);
+    ASSERT(slot < max_cache_size_);
+    TCEntry *entry = &tc_slots_[slot];
+    entry->head = start;
+    entry->tail = end;
+    return;
+  }
+  ReleaseListToSpans(start);
+}
+
+//BCF- The intel compiler crashes in the initial call to RemoveRange
+//  The compiler fence fixed this, but I'm not sure why
+#ifdef __INTEL_COMPILER
+#define FIXER BF_COMPILER_FENCE()
+#else
+#define FIXER
+#endif
+
+int CentralFreeList::RemoveRange(void **start, void **end, int N) {
+  ASSERT(N > 0);
+  
+  lock_.Lock();
+  if (N == Static::sizemap()->num_objects_to_move(size_class_) &&
+      used_slots_ > 0) {	  
+    int slot = --used_slots_;
+    ASSERT(slot >= 0);
+    TCEntry *entry = &tc_slots_[slot];
+    *start = entry->head;
+    *end = entry->tail;
+    lock_.Unlock();	
+    return N;
+  }
+
+  int result = 0;
+  void* head = NULL;
+  void* tail = NULL;
+  // TODO: Prefetch multiple TCEntries?
+  
+  tail = FetchFromSpansSafe();
+  if (tail != NULL) {	  
+    SLL_SetNext(tail, NULL);
+    head = tail;
+	FIXER;
+    result = 1;
+    while (result < N) {
+		FIXER;
+      void *t = FetchFromSpans();
+      if (!t) break;
+      SLL_Push(&head, t);
+      result++;
+    }
+  }
+  FIXER;
+  lock_.Unlock();
+  *start = head;
+  *end = tail;
+  return result;
+}
+
+
+void* CentralFreeList::FetchFromSpansSafe() {
+  void *t = FetchFromSpans();
+  if (!t) {
+      
+    Populate();
+    t = FetchFromSpans();
+  }
+  return t;
+}
+
+void* CentralFreeList::FetchFromSpans() {
+  if (TCMALLOC_NAMESPACE::DLL_IsEmpty(&nonempty_)) return NULL;
+  Span* span = nonempty_.next;
+
+  ASSERT(span->objects != NULL);
+  span->refcount++;
+  void* result = span->objects;
+  span->objects = *(reinterpret_cast<void**>(result));
+  if (span->objects == NULL) {
+    // Move to empty list
+    TCMALLOC_NAMESPACE::DLL_Remove(span);
+    TCMALLOC_NAMESPACE::DLL_Prepend(&empty_, span);
+    Event(span, 'E', 0);
+  }
+  counter_--;
+  return result;
+}
+
+// Fetch memory from the system and add to the central cache freelist.
+void CentralFreeList::Populate() {
+  // Release central list lock while operating on pageheap
+  lock_.Unlock();
+    
+  const size_t npages = Static::sizemap()->class_to_pages(size_class_);
+
+  Span* span;
+  {
+    SpinLockHolder h(Static::pageheap_lock());
+    span = Static::pageheap()->New(npages);
+	
+    if (span) Static::pageheap()->RegisterSizeClass(span, size_class_);
+  }  
+
+  //void* spanStart = (void*)((intptr_t)span->start << kPageShift);
+  //int spanSize = span->length << kPageShift;
+  //memset(spanStart, 0, spanSize);
+
+  //if (size_class_ == 2)
+  {
+	  //printf("sc:%d ptr:%08X\n", size_class_, spanStart);
+  }
+
+  if (span == NULL) {
+    Log(kLog, __FILE__, __LINE__,
+        "tcmalloc: allocation failed", npages << kPageShift);
+    lock_.Lock();
+    return;
+  }
+  ASSERT(span->length == npages);
+  // Cache sizeclass info eagerly.  Locking is not necessary.
+  // (Instead of being eager, we could just replace any stale info
+  // about this span, but that seems to be no better in practice.)
+  for (int i = 0; i < npages; i++) {
+    Static::pageheap()->CacheSizeClass(span->start + i, size_class_);
+  }
+
+  // Split the block into pieces and add to the free-list
+  // TODO: coloring of objects to avoid cache conflicts?
+  void** tail = &span->objects;
+  char* ptr = reinterpret_cast<char*>(span->start << kPageShift);
+  char* limit = ptr + (npages << kPageShift);
+  const size_t size = Static::sizemap()->ByteSizeForClass(size_class_);
+  int num = 0;
+  while (ptr + size <= limit) {
+    *tail = ptr;
+    tail = reinterpret_cast<void**>(ptr);
+    ptr += size;
+    num++;
+  }
+  ASSERT(ptr <= limit);
+  *tail = NULL;
+  span->refcount = 0; // No sub-object in use yet
+
+  // Add span to list of non-empty spans
+  lock_.Lock();
+  TCMALLOC_NAMESPACE::DLL_Prepend(&nonempty_, span);
+  ++num_spans_;
+  counter_ += num;
+}
+
+int CentralFreeList::tc_length() {
+  SpinLockHolder h(&lock_);
+  return used_slots_ * Static::sizemap()->num_objects_to_move(size_class_);
+}
+
+size_t CentralFreeList::OverheadBytes() {
+  SpinLockHolder h(&lock_);
+  if (size_class_ == 0) {  // 0 holds the 0-sized allocations
+    return 0;
+  }
+  const size_t pages_per_span = Static::sizemap()->class_to_pages(size_class_);
+  const size_t object_size = Static::sizemap()->class_to_size(size_class_);
+  ASSERT(object_size > 0);
+  const size_t overhead_per_span = (pages_per_span * kPageSize) % object_size;
+  return num_spans_ * overhead_per_span;
+}
+
+}  // namespace TCMALLOC_NAMESPACE

BeefRT/gperftools/src/central_freelist.h

@@ -0,0 +1,213 @@
+// Copyright (c) 2008, Google Inc.  
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without 
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com> 
+
+#ifndef TCMALLOC_CENTRAL_FREELIST_H_
+#define TCMALLOC_CENTRAL_FREELIST_H_
+
+#include "tcconfig.h"
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_STDINT_H
+#include <stdint.h>                     // for int32_t
+#endif
+#include "base/spinlock.h"
+#include "base/thread_annotations.h"
+#include "common.h"
+#include "span.h"
+
+namespace TCMALLOC_NAMESPACE {
+
+// Data kept per size-class in central cache.
+class CentralFreeList {
+ public:
+  // A CentralFreeList may be used before its constructor runs.
+  // So we prevent lock_'s constructor from doing anything to the
+  // lock_ state.
+  CentralFreeList() : lock_(base::LINKER_INITIALIZED) { }
+
+  void Init(size_t cl);
+
+  // These methods all do internal locking.
+
+  // Insert the specified range into the central freelist.  N is the number of
+  // elements in the range.  RemoveRange() is the opposite operation.
+  void InsertRange(void *start, void *end, int N);
+
+  // Returns the actual number of fetched elements and sets *start and *end.
+  int RemoveRange(void **start, void **end, int N);
+
+  //BCF- releases the whole pending list in one block to minimize time the lock needs to be held
+  void ReleasePendingSpanObjects(Span* span);
+    
+  // Returns the number of free objects in cache.
+  int length() {
+    SpinLockHolder h(&lock_);
+    return counter_;
+  }
+
+  // Returns the number of free objects in the transfer cache.
+  int tc_length();
+
+  // Returns the memory overhead (internal fragmentation) attributable
+  // to the freelist.  This is memory lost when the size of elements
+  // in a freelist doesn't exactly divide the page-size (an 8192-byte
+  // page full of 5-byte objects would have 2 bytes memory overhead).
+  size_t OverheadBytes();
+
+  // Lock/Unlock the internal SpinLock. Used on the pthread_atfork call
+  // to set the lock in a consistent state before the fork.
+  void Lock() {
+    lock_.Lock();
+  }
+
+  void Unlock() {
+    lock_.Unlock();
+  }
+
+ private:
+  // TransferCache is used to cache transfers of
+  // sizemap.num_objects_to_move(size_class) back and forth between
+  // thread caches and the central cache for a given size class.
+  struct TCEntry {
+    void *head;  // Head of chain of objects.
+    void *tail;  // Tail of chain of objects.
+  };
+
+  // A central cache freelist can have anywhere from 0 to kMaxNumTransferEntries
+  // slots to put link list chains into.
+#ifdef TCMALLOC_SMALL_BUT_SLOW
+  // For the small memory model, the transfer cache is not used.
+  static const int kMaxNumTransferEntries = 0;
+#else
+  // Starting point for the the maximum number of entries in the transfer cache.
+  // This actual maximum for a given size class may be lower than this
+  // maximum value.
+  static const int kMaxNumTransferEntries = 64;
+#endif
+
+  // REQUIRES: lock_ is held
+  // Remove object from cache and return.
+  // Return NULL if no free entries in cache.
+  void* FetchFromSpans() EXCLUSIVE_LOCKS_REQUIRED(lock_);
+
+  // REQUIRES: lock_ is held
+  // Remove object from cache and return.  Fetches
+  // from pageheap if cache is empty.  Only returns
+  // NULL on allocation failure.
+  void* FetchFromSpansSafe() EXCLUSIVE_LOCKS_REQUIRED(lock_);
+
+  // REQUIRES: lock_ is held
+  // Release a linked list of objects to spans.
+  // May temporarily release lock_.
+  void ReleaseListToSpans(void *start) EXCLUSIVE_LOCKS_REQUIRED(lock_);
+    
+  // REQUIRES: lock_ is held
+  // Release an object to spans.
+  // May temporarily release lock_.
+  void ReleaseToSpans(void* object) EXCLUSIVE_LOCKS_REQUIRED(lock_);
+
+  // REQUIRES: lock_ is held
+  // Populate cache by fetching from the page heap.
+  // May temporarily release lock_.
+  void Populate() EXCLUSIVE_LOCKS_REQUIRED(lock_);
+
+  // REQUIRES: lock is held.
+  // Tries to make room for a TCEntry.  If the cache is full it will try to
+  // expand it at the cost of some other cache size.  Return false if there is
+  // no space.
+  bool MakeCacheSpace() EXCLUSIVE_LOCKS_REQUIRED(lock_);
+
+  // REQUIRES: lock_ for locked_size_class is held.
+  // Picks a "random" size class to steal TCEntry slot from.  In reality it
+  // just iterates over the sizeclasses but does so without taking a lock.
+  // Returns true on success.
+  // May temporarily lock a "random" size class.
+  static bool EvictRandomSizeClass(int locked_size_class, bool force);
+
+  // REQUIRES: lock_ is *not* held.
+  // Tries to shrink the Cache.  If force is true it will relase objects to
+  // spans if it allows it to shrink the cache.  Return false if it failed to
+  // shrink the cache.  Decrements cache_size_ on succeess.
+  // May temporarily take lock_.  If it takes lock_, the locked_size_class
+  // lock is released to keep the thread from holding two size class locks
+  // concurrently which could lead to a deadlock.
+  bool ShrinkCache(int locked_size_class, bool force) LOCKS_EXCLUDED(lock_);
+
+  // This lock protects all the data members.  cached_entries and cache_size_
+  // may be looked at without holding the lock.
+  SpinLock lock_;
+
+  // We keep linked lists of empty and non-empty spans.
+  size_t   size_class_;     // My size class
+  Span     empty_;          // Dummy header for list of empty spans
+  Span     nonempty_;       // Dummy header for list of non-empty spans
+  size_t   num_spans_;      // Number of spans in empty_ plus nonempty_
+  size_t   counter_;        // Number of free objects in cache entry
+
+  // Here we reserve space for TCEntry cache slots.  Space is preallocated
+  // for the largest possible number of entries than any one size class may
+  // accumulate.  Not all size classes are allowed to accumulate
+  // kMaxNumTransferEntries, so there is some wasted space for those size
+  // classes.
+  TCEntry tc_slots_[kMaxNumTransferEntries];
+
+  // Number of currently used cached entries in tc_slots_.  This variable is
+  // updated under a lock but can be read without one.
+  int32_t used_slots_;
+  // The current number of slots for this size class.  This is an
+  // adaptive value that is increased if there is lots of traffic
+  // on a given size class.
+  int32_t cache_size_;
+  // Maximum size of the cache for a given size class.
+  int32_t max_cache_size_;
+};
+
+// Pads each CentralCache object to multiple of 64 bytes.  Since some
+// compilers (such as MSVC) don't like it when the padding is 0, I use
+// template specialization to remove the padding entirely when
+// sizeof(CentralFreeList) is a multiple of 64.
+template<int kFreeListSizeMod64>
+class CentralFreeListPaddedTo : public CentralFreeList {
+ private:
+  char pad_[64 - kFreeListSizeMod64];
+};
+
+template<>
+class CentralFreeListPaddedTo<0> : public CentralFreeList {
+};
+
+class CentralFreeListPadded : public CentralFreeListPaddedTo<
+  sizeof(CentralFreeList) % 64> {
+};
+
+}  // namespace TCMALLOC_NAMESPACE
+
+#endif  // TCMALLOC_CENTRAL_FREELIST_H_

BeefRT/gperftools/src/common.h

@@ -0,0 +1,271 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+//
+// Common definitions for tcmalloc code.
+
+#ifndef TCMALLOC_COMMON_H_
+#define TCMALLOC_COMMON_H_
+
+#include "tcconfig.h"
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_STDINT_H
+#include <stdint.h>                     // for uintptr_t, uint64_t
+#endif
+#include "internal_logging.h"  // for ASSERT, etc
+#include "base/basictypes.h"   // for LIKELY, etc
+
+#ifdef HAVE_BUILTIN_EXPECT
+#define LIKELY(x) __builtin_expect(!!(x), 1)
+#define UNLIKELY(x) __builtin_expect(!!(x), 0)
+#else
+#define LIKELY(x) (x)
+#define UNLIKELY(x) (x)
+#endif
+
+// Type that can hold a page number
+typedef uintptr_t PageID;
+
+// Type that can hold the length of a run of pages
+typedef uintptr_t Length;
+
+//-------------------------------------------------------------------
+// Configuration
+//-------------------------------------------------------------------
+
+// Using large pages speeds up the execution at a cost of larger memory use.
+// Deallocation may speed up by a factor as the page map gets 8x smaller, so
+// lookups in the page map result in fewer L2 cache misses, which translates to
+// speedup for application/platform combinations with high L2 cache pressure.
+// As the number of size classes increases with large pages, we increase
+// the thread cache allowance to avoid passing more free ranges to and from
+// central lists.  Also, larger pages are less likely to get freed.
+// These two factors cause a bounded increase in memory use.
+
+#if defined(TCMALLOC_LARGE_PAGES)
+static const size_t kPageShift  = 15;
+static const size_t kNumClasses = 78;
+static const size_t kMinAlign   = 16;
+#elif defined(TCMALLOC_LARGE_PAGES64K)
+static const size_t kPageShift  = 16;
+static const size_t kNumClasses = 82;
+static const size_t kMinAlign   = 16;
+#elif defined(TCMALLOC_ALIGN_8BYTES)
+static const size_t kPageShift  = 13;
+static const size_t kNumClasses = 95;
+// Unless we force to use 8 bytes alignment we use an alignment of
+// at least 16 bytes to statisfy requirements for some SSE types.
+// Keep in mind when using the 16 bytes alignment you can have a space
+// waste due alignment of 25%. (eg malloc of 24 bytes will get 32 bytes)
+static const size_t kMinAlign   = 8;
+#else
+static const size_t kPageShift  = 13;
+static const size_t kNumClasses = 88;
+static const size_t kMinAlign   = 16;
+#endif
+static const size_t kMaxThreadCacheSize = 4 << 20;
+
+static const size_t kPageSize   = 1 << kPageShift;
+static const size_t kMaxSize    = 256 * 1024;
+static const size_t kAlignment  = 8;
+static const size_t kLargeSizeClass = 0;
+// For all span-lengths < kMaxPages we keep an exact-size list.
+static const size_t kMaxPages = 1 << (20 - kPageShift);
+
+// Default bound on the total amount of thread caches.
+#ifdef TCMALLOC_SMALL_BUT_SLOW
+// Make the overall thread cache no bigger than that of a single thread
+// for the small memory footprint case.
+static const size_t kDefaultOverallThreadCacheSize = kMaxThreadCacheSize;
+#else
+static const size_t kDefaultOverallThreadCacheSize = 8u * kMaxThreadCacheSize;
+#endif
+
+// Lower bound on the per-thread cache sizes
+static const size_t kMinThreadCacheSize = kMaxSize * 2;
+
+// The number of bytes one ThreadCache will steal from another when
+// the first ThreadCache is forced to Scavenge(), delaying the
+// next call to Scavenge for this thread.
+static const size_t kStealAmount = 1 << 16;
+
+// The number of times that a deallocation can cause a freelist to
+// go over its max_length() before shrinking max_length().
+static const int kMaxOverages = 3;
+
+// Maximum length we allow a per-thread free-list to have before we
+// move objects from it into the corresponding central free-list.  We
+// want this big to avoid locking the central free-list too often.  It
+// should not hurt to make this list somewhat big because the
+// scavenging code will shrink it down when its contents are not in use.
+static const int kMaxDynamicFreeListLength = 8192;
+
+static const Length kMaxValidPages = (~static_cast<Length>(0)) >> kPageShift;
+
+#if defined __x86_64__
+// All current and planned x86_64 processors only look at the lower 48 bits
+// in virtual to physical address translation.  The top 16 are thus unused.
+// TODO(rus): Under what operating systems can we increase it safely to 17?
+// This lets us use smaller page maps.  On first allocation, a 36-bit page map
+// uses only 96 KB instead of the 4.5 MB used by a 52-bit page map.
+static const int kAddressBits = (sizeof(void*) < 8 ? (8 * sizeof(void*)) : 48);
+#else
+static const int kAddressBits = 8 * sizeof(void*);
+#endif
+
+namespace TCMALLOC_NAMESPACE {
+
+// Convert byte size into pages.  This won't overflow, but may return
+// an unreasonably large value if bytes is huge enough.
+inline Length pages(size_t bytes) {
+  return (bytes >> kPageShift) +
+      ((bytes & (kPageSize - 1)) > 0 ? 1 : 0);
+}
+
+// For larger allocation sizes, we use larger memory alignments to
+// reduce the number of size classes.
+int AlignmentForSize(size_t size);
+
+// Size-class information + mapping
+class SizeMap {
+ private:
+  // Number of objects to move between a per-thread list and a central
+  // list in one shot.  We want this to be not too small so we can
+  // amortize the lock overhead for accessing the central list.  Making
+  // it too big may temporarily cause unnecessary memory wastage in the
+  // per-thread free list until the scavenger cleans up the list.
+  int num_objects_to_move_[kNumClasses];
+
+  //-------------------------------------------------------------------
+  // Mapping from size to size_class and vice versa
+  //-------------------------------------------------------------------
+
+  // Sizes <= 1024 have an alignment >= 8.  So for such sizes we have an
+  // array indexed by ceil(size/8).  Sizes > 1024 have an alignment >= 128.
+  // So for these larger sizes we have an array indexed by ceil(size/128).
+  //
+  // We flatten both logical arrays into one physical array and use
+  // arithmetic to compute an appropriate index.  The constants used by
+  // ClassIndex() were selected to make the flattening work.
+  //
+  // Examples:
+  //   Size       Expression                      Index
+  //   -------------------------------------------------------
+  //   0          (0 + 7) / 8                     0
+  //   1          (1 + 7) / 8                     1
+  //   ...
+  //   1024       (1024 + 7) / 8                  128
+  //   1025       (1025 + 127 + (120<<7)) / 128   129
+  //   ...
+  //   32768      (32768 + 127 + (120<<7)) / 128  376
+  static const int kMaxSmallSize = 1024;
+  static const size_t kClassArraySize =
+      ((kMaxSize + 127 + (120 << 7)) >> 7) + 1;
+  unsigned char class_array_[kClassArraySize];
+
+  // Compute index of the class_array[] entry for a given size
+  static inline size_t ClassIndex(int s) {
+    // Use unsigned arithmetic to avoid unnecessary sign extensions.
+    ASSERT(0 <= s);
+    ASSERT(s <= kMaxSize);
+    if (LIKELY(s <= kMaxSmallSize)) {
+      return (static_cast<uint32_t>(s) + 7) >> 3;
+    } else {
+      return (static_cast<uint32_t>(s) + 127 + (120 << 7)) >> 7;
+    }
+  }
+
+  int NumMoveSize(size_t size);
+
+  // Mapping from size class to max size storable in that class
+  size_t class_to_size_[kNumClasses];
+
+  // Mapping from size class to number of pages to allocate at a time
+  size_t class_to_pages_[kNumClasses];
+
+ public:
+  // Constructor should do nothing since we rely on explicit Init()
+  // call, which may or may not be called before the constructor runs.
+  SizeMap() { }
+
+  // Initialize the mapping arrays
+  void Init();
+
+  inline int SizeClass(int size) {
+    return class_array_[ClassIndex(size)];
+  }
+
+  // Get the byte-size for a specified class
+  inline size_t ByteSizeForClass(size_t cl) {
+    return class_to_size_[cl];
+  }
+
+  // Mapping from size class to max size storable in that class
+  inline size_t class_to_size(size_t cl) {
+    return class_to_size_[cl];
+  }
+
+  // Mapping from size class to number of pages to allocate at a time
+  inline size_t class_to_pages(size_t cl) {
+    return class_to_pages_[cl];
+  }
+
+  // Number of objects to move between a per-thread list and a central
+  // list in one shot.  We want this to be not too small so we can
+  // amortize the lock overhead for accessing the central list.  Making
+  // it too big may temporarily cause unnecessary memory wastage in the
+  // per-thread free list until the scavenger cleans up the list.
+  inline int num_objects_to_move(size_t cl) {
+    return num_objects_to_move_[cl];
+  }
+};
+
+// Allocates "bytes" worth of memory and returns it.  Increments
+// metadata_system_bytes appropriately.  May return NULL if allocation
+// fails.  Requires pageheap_lock is held.
+void* MetaDataAlloc(size_t bytes);
+
+// Returns the total number of bytes allocated from the system.
+// Requires pageheap_lock is held.
+uint64_t metadata_system_bytes();
+
+// size/depth are made the same size as a pointer so that some generic
+// code below can conveniently cast them back and forth to void*.
+static const int kMaxStackDepth = 31;
+struct StackTrace {
+  uintptr_t size;          // Size of object
+  uintptr_t depth;         // Number of PC values stored in array below
+  void*     stack[kMaxStackDepth];
+};
+
+}  // namespace TCMALLOC_NAMESPACE
+
+#endif  // TCMALLOC_COMMON_H_

BeefRT/gperftools/src/config.h.in

@@ -0,0 +1,288 @@
+/* src/config.h.in.  Generated from configure.ac by autoheader.  */
+
+
+#ifndef GPERFTOOLS_CONFIG_H_
+#define GPERFTOOLS_CONFIG_H_
+
+
+/* Define to 1 if compiler supports __builtin_expect */
+#undef HAVE_BUILTIN_EXPECT
+
+/* Define to 1 if compiler supports __builtin_stack_pointer */
+#undef HAVE_BUILTIN_STACK_POINTER
+
+/* Define to 1 if you have the <conflict-signal.h> header file. */
+#undef HAVE_CONFLICT_SIGNAL_H
+
+/* Define to 1 if you have the <cygwin/signal.h> header file. */
+#undef HAVE_CYGWIN_SIGNAL_H
+
+/* Define to 1 if you have the declaration of `cfree', and to 0 if you don't.
+   */
+#undef HAVE_DECL_CFREE
+
+/* Define to 1 if you have the declaration of `memalign', and to 0 if you
+   don't. */
+#undef HAVE_DECL_MEMALIGN
+
+/* Define to 1 if you have the declaration of `nanosleep', and to 0 if you
+   don't. */
+#undef HAVE_DECL_NANOSLEEP
+
+/* Define to 1 if you have the declaration of `posix_memalign', and to 0 if
+   you don't. */
+#undef HAVE_DECL_POSIX_MEMALIGN
+
+/* Define to 1 if you have the declaration of `pvalloc', and to 0 if you
+   don't. */
+#undef HAVE_DECL_PVALLOC
+
+/* Define to 1 if you have the declaration of `sleep', and to 0 if you don't.
+   */
+#undef HAVE_DECL_SLEEP
+
+/* Define to 1 if you have the declaration of `uname', and to 0 if you don't.
+   */
+#undef HAVE_DECL_UNAME
+
+/* Define to 1 if you have the declaration of `valloc', and to 0 if you don't.
+   */
+#undef HAVE_DECL_VALLOC
+
+/* Define to 1 if you have the <dlfcn.h> header file. */
+#undef HAVE_DLFCN_H
+
+/* Define to 1 if the system has the type `Elf32_Versym'. */
+#undef HAVE_ELF32_VERSYM
+
+/* Define to 1 if you have the <execinfo.h> header file. */
+#undef HAVE_EXECINFO_H
+
+/* Define to 1 if you have the <fcntl.h> header file. */
+#undef HAVE_FCNTL_H
+
+/* Define to 1 if you have the <features.h> header file. */
+#undef HAVE_FEATURES_H
+
+/* Define to 1 if you have the `fork' function. */
+#undef HAVE_FORK
+
+/* Define to 1 if you have the `geteuid' function. */
+#undef HAVE_GETEUID
+
+/* Define to 1 if you have the `getpagesize' function. */
+#undef HAVE_GETPAGESIZE
+
+/* Define to 1 if you have the <glob.h> header file. */
+#undef HAVE_GLOB_H
+
+/* Define to 1 if you have the <grp.h> header file. */
+#undef HAVE_GRP_H
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#undef HAVE_INTTYPES_H
+
+/* Define to 1 if you have the <libunwind.h> header file. */
+#undef HAVE_LIBUNWIND_H
+
+/* Define to 1 if you have the <linux/ptrace.h> header file. */
+#undef HAVE_LINUX_PTRACE_H
+
+/* Define to 1 if you have the <malloc.h> header file. */
+#undef HAVE_MALLOC_H
+
+/* Define to 1 if you have the <malloc/malloc.h> header file. */
+#undef HAVE_MALLOC_MALLOC_H
+
+/* Define to 1 if you have the <memory.h> header file. */
+#undef HAVE_MEMORY_H
+
+/* Define to 1 if you have a working `mmap' system call. */
+#undef HAVE_MMAP
+
+/* define if the compiler implements namespaces */
+#undef HAVE_NAMESPACES
+
+/* Define to 1 if you have the <poll.h> header file. */
+#undef HAVE_POLL_H
+
+/* define if libc has program_invocation_name */
+#undef HAVE_PROGRAM_INVOCATION_NAME
+
+/* Define if you have POSIX threads libraries and header files. */
+#undef HAVE_PTHREAD
+
+/* Define to 1 if you have the <pwd.h> header file. */
+#undef HAVE_PWD_H
+
+/* Define to 1 if you have the `sbrk' function. */
+#undef HAVE_SBRK
+
+/* Define to 1 if you have the <sched.h> header file. */
+#undef HAVE_SCHED_H
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#undef HAVE_STDINT_H
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#undef HAVE_STDLIB_H
+
+/* Define to 1 if you have the <strings.h> header file. */
+#undef HAVE_STRINGS_H
+
+/* Define to 1 if you have the <string.h> header file. */
+#undef HAVE_STRING_H
+
+/* Define to 1 if the system has the type `struct mallinfo'. */
+#undef HAVE_STRUCT_MALLINFO
+
+/* Define to 1 if you have the <sys/cdefs.h> header file. */
+#undef HAVE_SYS_CDEFS_H
+
+/* Define to 1 if you have the <sys/malloc.h> header file. */
+#undef HAVE_SYS_MALLOC_H
+
+/* Define to 1 if you have the <sys/param.h> header file. */
+#undef HAVE_SYS_PARAM_H
+
+/* Define to 1 if you have the <sys/prctl.h> header file. */
+#undef HAVE_SYS_PRCTL_H
+
+/* Define to 1 if you have the <sys/resource.h> header file. */
+#undef HAVE_SYS_RESOURCE_H
+
+/* Define to 1 if you have the <sys/socket.h> header file. */
+#undef HAVE_SYS_SOCKET_H
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#undef HAVE_SYS_STAT_H
+
+/* Define to 1 if you have the <sys/syscall.h> header file. */
+#undef HAVE_SYS_SYSCALL_H
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#undef HAVE_SYS_TYPES_H
+
+/* <sys/ucontext.h> is broken on redhat 7 */
+#undef HAVE_SYS_UCONTEXT_H
+
+/* Define to 1 if you have the <sys/wait.h> header file. */
+#undef HAVE_SYS_WAIT_H
+
+/* Define to 1 if compiler supports __thread */
+#undef HAVE_TLS
+
+/* Define to 1 if you have the <ucontext.h> header file. */
+#undef HAVE_UCONTEXT_H
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#undef HAVE_UNISTD_H
+
+/* Define to 1 if you have the <unwind.h> header file. */
+#undef HAVE_UNWIND_H
+
+/* Define to 1 if you have the <valgrind.h> header file. */
+#undef HAVE_VALGRIND_H
+
+/* define if your compiler has __attribute__ */
+#undef HAVE___ATTRIBUTE__
+
+/* Define to 1 if compiler supports __environ */
+#undef HAVE___ENVIRON
+
+/* Define to 1 if the system has the type `__int64'. */
+#undef HAVE___INT64
+
+/* prefix where we look for installed files */
+#undef INSTALL_PREFIX
+
+/* Define to 1 if int32_t is equivalent to intptr_t */
+#undef INT32_EQUALS_INTPTR
+
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#undef LT_OBJDIR
+
+/* Define to 'volatile' if __malloc_hook is declared volatile */
+#undef MALLOC_HOOK_MAYBE_VOLATILE
+
+/* Define to 1 if your C compiler doesn't accept -c and -o together. */
+#undef NO_MINUS_C_MINUS_O
+
+/* Name of package */
+#undef PACKAGE
+
+/* Define to the address where bug reports for this package should be sent. */
+#undef PACKAGE_BUGREPORT
+
+/* Define to the full name of this package. */
+#undef PACKAGE_NAME
+
+/* Define to the full name and version of this package. */
+#undef PACKAGE_STRING
+
+/* Define to the one symbol short name of this package. */
+#undef PACKAGE_TARNAME
+
+/* Define to the home page for this package. */
+#undef PACKAGE_URL
+
+/* Define to the version of this package. */
+#undef PACKAGE_VERSION
+
+/* How to access the PC from a struct ucontext */
+#undef PC_FROM_UCONTEXT
+
+/* Always the empty-string on non-windows systems. On windows, should be
+   "__declspec(dllexport)". This way, when we compile the dll, we export our
+   functions/classes. It's safe to define this here because config.h is only
+   used internally, to compile the DLL, and every DLL source file #includes
+   "config.h" before anything else. */
+#undef PERFTOOLS_DLL_DECL
+
+/* printf format code for printing a size_t and ssize_t */
+#undef PRIdS
+
+/* printf format code for printing a size_t and ssize_t */
+#undef PRIuS
+
+/* printf format code for printing a size_t and ssize_t */
+#undef PRIxS
+
+/* Mark the systems where we know it's bad if pthreads runs too
+   early before main (before threads are initialized, presumably).  */
+#ifdef __FreeBSD__
+#define PTHREADS_CRASHES_IF_RUN_TOO_EARLY 1
+#endif
+
+/* Define to necessary symbol if this constant uses a non-standard name on
+   your system. */
+#undef PTHREAD_CREATE_JOINABLE
+
+/* Define to 1 if you have the ANSI C header files. */
+#undef STDC_HEADERS
+
+/* the namespace where STL code like vector<> is defined */
+#undef STL_NAMESPACE
+
+/* Version number of package */
+#undef VERSION
+
+/* C99 says: define this to get the PRI... macros from stdint.h */
+#ifndef __STDC_FORMAT_MACROS
+# define __STDC_FORMAT_MACROS 1
+#endif
+
+/* Define to `__inline__' or `__inline' if that's what the C compiler
+   calls it, or to nothing if 'inline' is not supported under any name.  */
+#ifndef __cplusplus
+#undef inline
+#endif
+
+
+#ifdef __MINGW32__
+#include "windows/mingw.h"
+#endif
+
+#endif  /* #ifndef GPERFTOOLS_CONFIG_H_ */
+

BeefRT/gperftools/src/config_for_unittests.h

@@ -0,0 +1,64 @@
+// Copyright (c) 2007, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// All Rights Reserved.
+//
+// Author: Craig Silverstein
+//
+// This file is needed for windows -- unittests are not part of the
+// perftools dll, but still want to include config.h just like the
+// dll does, so they can use internal tools and APIs for testing.
+//
+// The problem is that config.h declares PERFTOOLS_DLL_DECL to be
+// for exporting symbols, but the unittest needs to *import* symbols
+// (since it's not the dll).
+//
+// The solution is to have this file, which is just like config.h but
+// sets PERFTOOLS_DLL_DECL to do a dllimport instead of a dllexport.
+//
+// The reason we need this extra PERFTOOLS_DLL_DECL_FOR_UNITTESTS
+// variable is in case people want to set PERFTOOLS_DLL_DECL explicitly
+// to something other than __declspec(dllexport).  In that case, they
+// may want to use something other than __declspec(dllimport) for the
+// unittest case.  For that, we allow folks to define both
+// PERFTOOLS_DLL_DECL and PERFTOOLS_DLL_DECL_FOR_UNITTESTS explicitly.
+//
+// NOTE: This file is equivalent to config.h on non-windows systems,
+// which never defined PERFTOOLS_DLL_DECL_FOR_UNITTESTS and always
+// define PERFTOOLS_DLL_DECL to the empty string.
+
+#include "config.h"
+
+#undef PERFTOOLS_DLL_DECL
+#ifdef PERFTOOLS_DLL_DECL_FOR_UNITTESTS
+# define PERFTOOLS_DLL_DECL  PERFTOOLS_DLL_DECL_FOR_UNITTESTS
+#else
+# define PERFTOOLS_DLL_DECL  // if DLL_DECL_FOR_UNITTESTS isn't defined, use ""
+#endif

BeefRT/gperftools/src/getpc.h

@@ -0,0 +1,186 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein
+//
+// This is an internal header file used by profiler.cc.  It defines
+// the single (inline) function GetPC.  GetPC is used in a signal
+// handler to figure out the instruction that was being executed when
+// the signal-handler was triggered.
+//
+// To get this, we use the ucontext_t argument to the signal-handler
+// callback, which holds the full context of what was going on when
+// the signal triggered.  How to get from a ucontext_t to a Program
+// Counter is OS-dependent.
+
+#ifndef BASE_GETPC_H_
+#define BASE_GETPC_H_
+
+#include "config.h"
+
+// On many linux systems, we may need _GNU_SOURCE to get access to
+// the defined constants that define the register we want to see (eg
+// REG_EIP).  Note this #define must come first!
+#define _GNU_SOURCE 1
+// If #define _GNU_SOURCE causes problems, this might work instead.
+// It will cause problems for FreeBSD though!, because it turns off
+// the needed __BSD_VISIBLE.
+//#define _XOPEN_SOURCE 500
+
+#include <string.h>         // for memcmp
+#if defined(HAVE_SYS_UCONTEXT_H)
+#include <sys/ucontext.h>
+#elif defined(HAVE_UCONTEXT_H)
+#include <ucontext.h>       // for ucontext_t (and also mcontext_t)
+#elif defined(HAVE_CYGWIN_SIGNAL_H)
+#include <cygwin/signal.h>
+typedef ucontext ucontext_t;
+#endif
+
+
+// Take the example where function Foo() calls function Bar().  For
+// many architectures, Bar() is responsible for setting up and tearing
+// down its own stack frame.  In that case, it's possible for the
+// interrupt to happen when execution is in Bar(), but the stack frame
+// is not properly set up (either before it's done being set up, or
+// after it's been torn down but before Bar() returns).  In those
+// cases, the stack trace cannot see the caller function anymore.
+//
+// GetPC can try to identify this situation, on architectures where it
+// might occur, and unwind the current function call in that case to
+// avoid false edges in the profile graph (that is, edges that appear
+// to show a call skipping over a function).  To do this, we hard-code
+// in the asm instructions we might see when setting up or tearing
+// down a stack frame.
+//
+// This is difficult to get right: the instructions depend on the
+// processor, the compiler ABI, and even the optimization level.  This
+// is a best effort patch -- if we fail to detect such a situation, or
+// mess up the PC, nothing happens; the returned PC is not used for
+// any further processing.
+struct CallUnrollInfo {
+  // Offset from (e)ip register where this instruction sequence
+  // should be matched. Interpreted as bytes. Offset 0 is the next
+  // instruction to execute. Be extra careful with negative offsets in
+  // architectures of variable instruction length (like x86) - it is
+  // not that easy as taking an offset to step one instruction back!
+  int pc_offset;
+  // The actual instruction bytes. Feel free to make it larger if you
+  // need a longer sequence.
+  unsigned char ins[16];
+  // How many bytes to match from ins array?
+  int ins_size;
+  // The offset from the stack pointer (e)sp where to look for the
+  // call return address. Interpreted as bytes.
+  int return_sp_offset;
+};
+
+
+// The dereferences needed to get the PC from a struct ucontext were
+// determined at configure time, and stored in the macro
+// PC_FROM_UCONTEXT in config.h.  The only thing we need to do here,
+// then, is to do the magic call-unrolling for systems that support it.
+
+// -- Special case 1: linux x86, for which we have CallUnrollInfo
+#if defined(__linux) && defined(__i386) && defined(__GNUC__)
+static const CallUnrollInfo callunrollinfo[] = {
+  // Entry to a function:  push %ebp;  mov  %esp,%ebp
+  // Top-of-stack contains the caller IP.
+  { 0,
+    {0x55, 0x89, 0xe5}, 3,
+    0
+  },
+  // Entry to a function, second instruction:  push %ebp;  mov  %esp,%ebp
+  // Top-of-stack contains the old frame, caller IP is +4.
+  { -1,
+    {0x55, 0x89, 0xe5}, 3,
+    4
+  },
+  // Return from a function: RET.
+  // Top-of-stack contains the caller IP.
+  { 0,
+    {0xc3}, 1,
+    0
+  }
+};
+
+inline void* GetPC(const ucontext_t& signal_ucontext) {
+  // See comment above struct CallUnrollInfo.  Only try instruction
+  // flow matching if both eip and esp looks reasonable.
+  const int eip = signal_ucontext.uc_mcontext.gregs[REG_EIP];
+  const int esp = signal_ucontext.uc_mcontext.gregs[REG_ESP];
+  if ((eip & 0xffff0000) != 0 && (~eip & 0xffff0000) != 0 &&
+      (esp & 0xffff0000) != 0) {
+    char* eip_char = reinterpret_cast<char*>(eip);
+    for (int i = 0; i < sizeof(callunrollinfo)/sizeof(*callunrollinfo); ++i) {
+      if (!memcmp(eip_char + callunrollinfo[i].pc_offset,
+                  callunrollinfo[i].ins, callunrollinfo[i].ins_size)) {
+        // We have a match.
+        void **retaddr = (void**)(esp + callunrollinfo[i].return_sp_offset);
+        return *retaddr;
+      }
+    }
+  }
+  return (void*)eip;
+}
+
+// Special case #2: Windows, which has to do something totally different.
+#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__MINGW32__)
+// If this is ever implemented, probably the way to do it is to have
+// profiler.cc use a high-precision timer via timeSetEvent:
+//    http://msdn2.microsoft.com/en-us/library/ms712713.aspx
+// We'd use it in mode TIME_CALLBACK_FUNCTION/TIME_PERIODIC.
+// The callback function would be something like prof_handler, but
+// alas the arguments are different: no ucontext_t!  I don't know
+// how we'd get the PC (using StackWalk64?)
+//    http://msdn2.microsoft.com/en-us/library/ms680650.aspx
+
+#include "base/logging.h"   // for RAW_LOG
+#ifndef HAVE_CYGWIN_SIGNAL_H
+typedef int ucontext_t;
+#endif
+
+inline void* GetPC(const struct ucontext_t& signal_ucontext) {
+  RAW_LOG(ERROR, "GetPC is not yet implemented on Windows\n");
+  return NULL;
+}
+
+// Normal cases.  If this doesn't compile, it's probably because
+// PC_FROM_UCONTEXT is the empty string.  You need to figure out
+// the right value for your system, and add it to the list in
+// configure.ac (or set it manually in your config.h).
+#else
+inline void* GetPC(const ucontext_t& signal_ucontext) {
+  return (void*)signal_ucontext.PC_FROM_UCONTEXT;   // defined in config.h
+}
+
+#endif
+
+#endif  // BASE_GETPC_H_

BeefRT/gperftools/src/google/heap-checker.h

@@ -0,0 +1,33 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/heap-checker.h>

BeefRT/gperftools/src/google/heap-profiler.h

@@ -0,0 +1,34 @@
+/* Copyright (c) 2005, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/heap-profiler.h>

BeefRT/gperftools/src/google/malloc_extension.h

@@ -0,0 +1,33 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/malloc_extension.h>

BeefRT/gperftools/src/google/malloc_extension_c.h

@@ -0,0 +1,34 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/malloc_extension_c.h>

BeefRT/gperftools/src/google/malloc_hook.h

@@ -0,0 +1,33 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/malloc_hook.h>

BeefRT/gperftools/src/google/malloc_hook_c.h

@@ -0,0 +1,34 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/malloc_hook_c.h>

BeefRT/gperftools/src/google/profiler.h

@@ -0,0 +1,34 @@
+/* Copyright (c) 2005, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/profiler.h>

BeefRT/gperftools/src/google/stacktrace.h

@@ -0,0 +1,33 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/stacktrace.h>

BeefRT/gperftools/src/google/tcmalloc.h

@@ -0,0 +1,34 @@
+/* Copyright (c) 2003, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/* The code has moved to gperftools/.  Use that include-directory for
+ * new code.
+ */
+#include <gperftools/tcmalloc.h>

BeefRT/gperftools/src/gperftools/heap-checker.h

@@ -0,0 +1,424 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Maxim Lifantsev (with design ideas by Sanjay Ghemawat)
+//
+//
+// Module for detecing heap (memory) leaks.
+//
+// For full(er) information, see doc/heap_checker.html
+//
+// This module can be linked into programs with
+// no slowdown caused by this unless you activate the leak-checker:
+//
+//    1. Set the environment variable HEAPCHEK to _type_ before
+//       running the program.
+//
+// _type_ is usually "normal" but can also be "minimal", "strict", or
+// "draconian".  (See the html file for other options, like 'local'.)
+//
+// After that, just run your binary.  If the heap-checker detects
+// a memory leak at program-exit, it will print instructions on how
+// to track down the leak.
+
+#ifndef BASE_HEAP_CHECKER_H_
+#define BASE_HEAP_CHECKER_H_
+
+#include <sys/types.h>  // for size_t
+// I can't #include config.h in this public API file, but I should
+// really use configure (and make malloc_extension.h a .in file) to
+// figure out if the system has stdint.h or not.  But I'm lazy, so
+// for now I'm assuming it's a problem only with MSVC.
+#ifndef _MSC_VER
+#include <stdint.h>     // for uintptr_t
+#endif
+#include <stdarg.h>     // for va_list
+#include <vector>
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+
+// The class is thread-safe with respect to all the provided static methods,
+// as well as HeapLeakChecker objects: they can be accessed by multiple threads.
+class PERFTOOLS_DLL_DECL HeapLeakChecker {
+ public:
+
+  // ----------------------------------------------------------------------- //
+  // Static functions for working with (whole-program) leak checking.
+
+  // If heap leak checking is currently active in some mode
+  // e.g. if leak checking was started (and is still active now)
+  // due to HEAPCHECK=... defined in the environment.
+  // The return value reflects iff HeapLeakChecker objects manually
+  // constructed right now will be doing leak checking or nothing.
+  // Note that we can go from active to inactive state during InitGoogle()
+  // if FLAGS_heap_check gets set to "" by some code before/during InitGoogle().
+  static bool IsActive();
+
+  // Return pointer to the whole-program checker if it has been created
+  // and NULL otherwise.
+  // Once GlobalChecker() returns non-NULL that object will not disappear and
+  // will be returned by all later GlobalChecker calls.
+  // This is mainly to access BytesLeaked() and ObjectsLeaked() (see below)
+  // for the whole-program checker after one calls NoGlobalLeaks()
+  // or similar and gets false.
+  static HeapLeakChecker* GlobalChecker();
+
+  // Do whole-program leak check now (if it was activated for this binary);
+  // return false only if it was activated and has failed.
+  // The mode of the check is controlled by the command-line flags.
+  // This method can be called repeatedly.
+  // Things like GlobalChecker()->SameHeap() can also be called explicitly
+  // to do the desired flavor of the check.
+  static bool NoGlobalLeaks();
+
+  // If whole-program checker if active,
+  // cancel its automatic execution after main() exits.
+  // This requires that some leak check (e.g. NoGlobalLeaks())
+  // has been called at least once on the whole-program checker.
+  static void CancelGlobalCheck();
+
+  // ----------------------------------------------------------------------- //
+  // Non-static functions for starting and doing leak checking.
+
+  // Start checking and name the leak check performed.
+  // The name is used in naming dumped profiles
+  // and needs to be unique only within your binary.
+  // It must also be a string that can be a part of a file name,
+  // in particular not contain path expressions.
+  explicit HeapLeakChecker(const char *name);
+
+  // Destructor (verifies that some *NoLeaks or *SameHeap method
+  // has been called at least once).
+  ~HeapLeakChecker();
+
+  // These used to be different but are all the same now: they return
+  // true iff all memory allocated since this HeapLeakChecker object
+  // was constructor is still reachable from global state.
+  //
+  // Because we fork to convert addresses to symbol-names, and forking
+  // is not thread-safe, and we may be called in a threaded context,
+  // we do not try to symbolize addresses when called manually.
+  bool NoLeaks() { return DoNoLeaks(DO_NOT_SYMBOLIZE); }
+
+  // These forms are obsolete; use NoLeaks() instead.
+  // TODO(csilvers): mark as DEPRECATED.
+  bool QuickNoLeaks()  { return NoLeaks(); }
+  bool BriefNoLeaks()  { return NoLeaks(); }
+  bool SameHeap()      { return NoLeaks(); }
+  bool QuickSameHeap() { return NoLeaks(); }
+  bool BriefSameHeap() { return NoLeaks(); }
+
+  // Detailed information about the number of leaked bytes and objects
+  // (both of these can be negative as well).
+  // These are available only after a *SameHeap or *NoLeaks
+  // method has been called.
+  // Note that it's possible for both of these to be zero
+  // while SameHeap() or NoLeaks() returned false in case
+  // of a heap state change that is significant
+  // but preserves the byte and object counts.
+  ssize_t BytesLeaked() const;
+  ssize_t ObjectsLeaked() const;
+
+  // ----------------------------------------------------------------------- //
+  // Static helpers to make us ignore certain leaks.
+
+  // Scoped helper class.  Should be allocated on the stack inside a
+  // block of code.  Any heap allocations done in the code block
+  // covered by the scoped object (including in nested function calls
+  // done by the code block) will not be reported as leaks.  This is
+  // the recommended replacement for the GetDisableChecksStart() and
+  // DisableChecksToHereFrom() routines below.
+  //
+  // Example:
+  //   void Foo() {
+  //     HeapLeakChecker::Disabler disabler;
+  //     ... code that allocates objects whose leaks should be ignored ...
+  //   }
+  //
+  // REQUIRES: Destructor runs in same thread as constructor
+  class Disabler {
+   public:
+    Disabler();
+    ~Disabler();
+   private:
+    Disabler(const Disabler&);        // disallow copy
+    void operator=(const Disabler&);  // and assign
+  };
+
+  // Ignore an object located at 'ptr' (can go at the start or into the object)
+  // as well as all heap objects (transitively) referenced from it for the
+  // purposes of heap leak checking. Returns 'ptr' so that one can write
+  //   static T* obj = IgnoreObject(new T(...));
+  //
+  // If 'ptr' does not point to an active allocated object at the time of this
+  // call, it is ignored; but if it does, the object must not get deleted from
+  // the heap later on.
+  //
+  // See also HiddenPointer, below, if you need to prevent a pointer from
+  // being traversed by the heap checker but do not wish to transitively
+  // whitelist objects referenced through it.
+  template <typename T>
+  static T* IgnoreObject(T* ptr) {
+    DoIgnoreObject(static_cast<const void*>(const_cast<const T*>(ptr)));
+    return ptr;
+  }
+
+  // Undo what an earlier IgnoreObject() call promised and asked to do.
+  // At the time of this call 'ptr' must point at or inside of an active
+  // allocated object which was previously registered with IgnoreObject().
+  static void UnIgnoreObject(const void* ptr);
+
+  // ----------------------------------------------------------------------- //
+  // Internal types defined in .cc
+
+  class Allocator;
+  struct RangeValue;
+
+ private:
+
+  // ----------------------------------------------------------------------- //
+  // Various helpers
+
+  // Create the name of the heap profile file.
+  // Should be deleted via Allocator::Free().
+  char* MakeProfileNameLocked();
+
+  // Helper for constructors
+  void Create(const char *name, bool make_start_snapshot);
+
+  enum ShouldSymbolize { SYMBOLIZE, DO_NOT_SYMBOLIZE };
+
+  // Helper for *NoLeaks and *SameHeap
+  bool DoNoLeaks(ShouldSymbolize should_symbolize);
+
+  // Helper for NoGlobalLeaks, also called by the global destructor.
+  static bool NoGlobalLeaksMaybeSymbolize(ShouldSymbolize should_symbolize);
+
+  // These used to be public, but they are now deprecated.
+  // Will remove entirely when all internal uses are fixed.
+  // In the meantime, use friendship so the unittest can still test them.
+  static void* GetDisableChecksStart();
+  static void DisableChecksToHereFrom(const void* start_address);
+  static void DisableChecksIn(const char* pattern);
+  friend void RangeDisabledLeaks();
+  friend void NamedTwoDisabledLeaks();
+  friend void* RunNamedDisabledLeaks(void*);
+  friend void TestHeapLeakCheckerNamedDisabling();
+  // TODO(csilvers): remove this one, at least
+  friend int main(int, char**);
+
+
+  // Actually implements IgnoreObject().
+  static void DoIgnoreObject(const void* ptr);
+
+  // Disable checks based on stack trace entry at a depth <=
+  // max_depth.  Used to hide allocations done inside some special
+  // libraries.
+  static void DisableChecksFromToLocked(const void* start_address,
+                                        const void* end_address,
+                                        int max_depth);
+
+  // Helper for DoNoLeaks to ignore all objects reachable from all live data
+  static void IgnoreAllLiveObjectsLocked(const void* self_stack_top);
+
+  // Callback we pass to ListAllProcessThreads (see thread_lister.h)
+  // that is invoked when all threads of our process are found and stopped.
+  // The call back does the things needed to ignore live data reachable from
+  // thread stacks and registers for all our threads
+  // as well as do other global-live-data ignoring
+  // (via IgnoreNonThreadLiveObjectsLocked)
+  // during the quiet state of all threads being stopped.
+  // For the argument meaning see the comment by ListAllProcessThreads.
+  // Here we only use num_threads and thread_pids, that ListAllProcessThreads
+  // fills for us with the number and pids of all the threads of our process
+  // it found and attached to.
+  static int IgnoreLiveThreadsLocked(void* parameter,
+                                     int num_threads,
+                                     pid_t* thread_pids,
+                                     va_list ap);
+
+  // Helper for IgnoreAllLiveObjectsLocked and IgnoreLiveThreadsLocked
+  // that we prefer to execute from IgnoreLiveThreadsLocked
+  // while all threads are stopped.
+  // This helper does live object discovery and ignoring
+  // for all objects that are reachable from everything
+  // not related to thread stacks and registers.
+  static void IgnoreNonThreadLiveObjectsLocked();
+
+  // Helper for IgnoreNonThreadLiveObjectsLocked and IgnoreLiveThreadsLocked
+  // to discover and ignore all heap objects
+  // reachable from currently considered live objects
+  // (live_objects static global variable in out .cc file).
+  // "name", "name2" are two strings that we print one after another
+  // in a debug message to describe what kind of live object sources
+  // are being used.
+  static void IgnoreLiveObjectsLocked(const char* name, const char* name2);
+
+  // Do the overall whole-program heap leak check if needed;
+  // returns true when did the leak check.
+  static bool DoMainHeapCheck();
+
+  // Type of task for UseProcMapsLocked
+  enum ProcMapsTask {
+    RECORD_GLOBAL_DATA,
+    DISABLE_LIBRARY_ALLOCS
+  };
+
+  // Success/Error Return codes for UseProcMapsLocked.
+  enum ProcMapsResult {
+    PROC_MAPS_USED,
+    CANT_OPEN_PROC_MAPS,
+    NO_SHARED_LIBS_IN_PROC_MAPS
+  };
+
+  // Read /proc/self/maps, parse it, and do the 'proc_maps_task' for each line.
+  static ProcMapsResult UseProcMapsLocked(ProcMapsTask proc_maps_task);
+
+  // A ProcMapsTask to disable allocations from 'library'
+  // that is mapped to [start_address..end_address)
+  // (only if library is a certain system library).
+  static void DisableLibraryAllocsLocked(const char* library,
+                                         uintptr_t start_address,
+                                         uintptr_t end_address);
+
+  // Return true iff "*ptr" points to a heap object
+  // ("*ptr" can point at the start or inside of a heap object
+  //  so that this works e.g. for pointers to C++ arrays, C++ strings,
+  //  multiple-inherited objects, or pointers to members).
+  // We also fill *object_size for this object then
+  // and we move "*ptr" to point to the very start of the heap object.
+  static inline bool HaveOnHeapLocked(const void** ptr, size_t* object_size);
+
+  // Helper to shutdown heap leak checker when it's not needed
+  // or can't function properly.
+  static void TurnItselfOffLocked();
+
+  // Internally-used c-tor to start whole-executable checking.
+  HeapLeakChecker();
+
+  // ----------------------------------------------------------------------- //
+  // Friends and externally accessed helpers.
+
+  // Helper for VerifyHeapProfileTableStackGet in the unittest
+  // to get the recorded allocation caller for ptr,
+  // which must be a heap object.
+  static const void* GetAllocCaller(void* ptr);
+  friend void VerifyHeapProfileTableStackGet();
+
+  // This gets to execute before constructors for all global objects
+  static void BeforeConstructorsLocked();
+  friend void HeapLeakChecker_BeforeConstructors();
+
+  // This gets to execute after destructors for all global objects
+  friend void HeapLeakChecker_AfterDestructors();
+
+  // Full starting of recommended whole-program checking.
+  friend void HeapLeakChecker_InternalInitStart();
+
+  // Runs REGISTER_HEAPCHECK_CLEANUP cleanups and potentially
+  // calls DoMainHeapCheck
+  friend void HeapLeakChecker_RunHeapCleanups();
+
+  // ----------------------------------------------------------------------- //
+  // Member data.
+
+  class SpinLock* lock_;  // to make HeapLeakChecker objects thread-safe
+  const char* name_;  // our remembered name (we own it)
+                      // NULL means this leak checker is a noop
+
+  // Snapshot taken when the checker was created.  May be NULL
+  // for the global heap checker object.  We use void* instead of
+  // HeapProfileTable::Snapshot* to avoid including heap-profile-table.h.
+  void* start_snapshot_;
+
+  bool has_checked_;  // if we have done the leak check, so these are ready:
+  ssize_t inuse_bytes_increase_;  // bytes-in-use increase for this checker
+  ssize_t inuse_allocs_increase_;  // allocations-in-use increase
+                                   // for this checker
+  bool keep_profiles_;  // iff we should keep the heap profiles we've made
+
+  // ----------------------------------------------------------------------- //
+
+  // Disallow "evil" constructors.
+  HeapLeakChecker(const HeapLeakChecker&);
+  void operator=(const HeapLeakChecker&);
+};
+
+
+// Holds a pointer that will not be traversed by the heap checker.
+// Contrast with HeapLeakChecker::IgnoreObject(o), in which o and
+// all objects reachable from o are ignored by the heap checker.
+template <class T>
+class HiddenPointer {
+ public:
+  explicit HiddenPointer(T* t)
+      : masked_t_(reinterpret_cast<uintptr_t>(t) ^ kHideMask) {
+  }
+  // Returns unhidden pointer.  Be careful where you save the result.
+  T* get() const { return reinterpret_cast<T*>(masked_t_ ^ kHideMask); }
+
+ private:
+  // Arbitrary value, but not such that xor'ing with it is likely
+  // to map one valid pointer to another valid pointer:
+  static const uintptr_t kHideMask =
+      static_cast<uintptr_t>(0xF03A5F7BF03A5F7Bll);
+  uintptr_t masked_t_;
+};
+
+// A class that exists solely to run its destructor.  This class should not be
+// used directly, but instead by the REGISTER_HEAPCHECK_CLEANUP macro below.
+class PERFTOOLS_DLL_DECL HeapCleaner {
+ public:
+  typedef void (*void_function)(void);
+  HeapCleaner(void_function f);
+  static void RunHeapCleanups();
+ private:
+  static std::vector<void_function>* heap_cleanups_;
+};
+
+// A macro to declare module heap check cleanup tasks
+// (they run only if we are doing heap leak checking.)
+// 'body' should be the cleanup code to run.  'name' doesn't matter,
+// but must be unique amongst all REGISTER_HEAPCHECK_CLEANUP calls.
+#define REGISTER_HEAPCHECK_CLEANUP(name, body)  \
+  namespace { \
+  void heapcheck_cleanup_##name() { body; } \
+  static HeapCleaner heapcheck_cleaner_##name(&heapcheck_cleanup_##name); \
+  }
+
+#endif  // BASE_HEAP_CHECKER_H_

BeefRT/gperftools/src/gperftools/heap-profiler.h

@@ -0,0 +1,104 @@
+/* Copyright (c) 2005, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ *
+ * Module for heap-profiling.
+ *
+ * For full(er) information, see doc/heapprofile.html
+ *
+ * This module can be linked into your program with
+ * no slowdown caused by this unless you activate the profiler
+ * using one of the following methods:
+ *
+ *    1. Before starting the program, set the environment variable
+ *       "HEAPPROFILE" to be the name of the file to which the profile
+ *       data should be written.
+ *
+ *    2. Programmatically, start and stop the profiler using the
+ *       routines "HeapProfilerStart(filename)" and "HeapProfilerStop()".
+ *
+ */
+
+#ifndef BASE_HEAP_PROFILER_H_
+#define BASE_HEAP_PROFILER_H_
+
+#include <stddef.h>
+
+/* Annoying stuff for windows; makes sure clients can import these functions */
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+/* All this code should be usable from within C apps. */
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Start profiling and arrange to write profile data to file names
+ * of the form: "prefix.0000", "prefix.0001", ...
+ */
+PERFTOOLS_DLL_DECL void HeapProfilerStart(const char* prefix);
+
+/* Returns non-zero if we are currently profiling the heap.  (Returns
+ * an int rather than a bool so it's usable from C.)  This is true
+ * between calls to HeapProfilerStart() and HeapProfilerStop(), and
+ * also if the program has been run with HEAPPROFILER, or some other
+ * way to turn on whole-program profiling.
+ */
+int IsHeapProfilerRunning();
+
+/* Stop heap profiling.  Can be restarted again with HeapProfilerStart(),
+ * but the currently accumulated profiling information will be cleared.
+ */
+PERFTOOLS_DLL_DECL void HeapProfilerStop();
+
+/* Dump a profile now - can be used for dumping at a hopefully
+ * quiescent state in your program, in order to more easily track down
+ * memory leaks. Will include the reason in the logged message
+ */
+PERFTOOLS_DLL_DECL void HeapProfilerDump(const char *reason);
+
+/* Generate current heap profiling information.
+ * Returns an empty string when heap profiling is not active.
+ * The returned pointer is a '\0'-terminated string allocated using malloc()
+ * and should be free()-ed as soon as the caller does not need it anymore.
+ */
+PERFTOOLS_DLL_DECL char* GetHeapProfile();
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  /* BASE_HEAP_PROFILER_H_ */

BeefRT/gperftools/src/gperftools/malloc_extension.h

@@ -0,0 +1,420 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+//
+// Extra extensions exported by some malloc implementations.  These
+// extensions are accessed through a virtual base class so an
+// application can link against a malloc that does not implement these
+// extensions, and it will get default versions that do nothing.
+//
+// NOTE FOR C USERS: If you wish to use this functionality from within
+// a C program, see malloc_extension_c.h.
+
+#ifndef BASE_MALLOC_EXTENSION_H_
+#define BASE_MALLOC_EXTENSION_H_
+
+#include <stddef.h>
+// I can't #include config.h in this public API file, but I should
+// really use configure (and make malloc_extension.h a .in file) to
+// figure out if the system has stdint.h or not.  But I'm lazy, so
+// for now I'm assuming it's a problem only with MSVC.
+#ifndef _MSC_VER
+#include <stdint.h>
+#endif
+#include <string>
+#include <vector>
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+static const int kMallocHistogramSize = 64;
+
+// One day, we could support other types of writers (perhaps for C?)
+typedef std::string MallocExtensionWriter;
+
+namespace base {
+struct MallocRange;
+}
+
+// Interface to a pluggable system allocator.
+class SysAllocator {
+ public:
+  SysAllocator() {
+  }
+  virtual ~SysAllocator();
+
+  // Allocates "size"-byte of memory from system aligned with "alignment".
+  // Returns NULL if failed. Otherwise, the returned pointer p up to and
+  // including (p + actual_size -1) have been allocated.
+  virtual void* Alloc(size_t size, size_t *actual_size, size_t alignment) = 0;
+};
+
+// The default implementations of the following routines do nothing.
+// All implementations should be thread-safe; the current one
+// (TCMallocImplementation) is.
+class PERFTOOLS_DLL_DECL MallocExtension {
+ public:
+  virtual ~MallocExtension();
+
+  // Call this very early in the program execution -- say, in a global
+  // constructor -- to set up parameters and state needed by all
+  // instrumented malloc implemenatations.  One example: this routine
+  // sets environemnt variables to tell STL to use libc's malloc()
+  // instead of doing its own memory management.  This is safe to call
+  // multiple times, as long as each time is before threads start up.
+  static void Initialize();
+
+  // See "verify_memory.h" to see what these routines do
+  virtual bool VerifyAllMemory();
+  virtual bool VerifyNewMemory(const void* p);
+  virtual bool VerifyArrayNewMemory(const void* p);
+  virtual bool VerifyMallocMemory(const void* p);
+  virtual bool MallocMemoryStats(int* blocks, size_t* total,
+                                 int histogram[kMallocHistogramSize]);
+
+  // Get a human readable description of the current state of the malloc
+  // data structures.  The state is stored as a null-terminated string
+  // in a prefix of "buffer[0,buffer_length-1]".
+  // REQUIRES: buffer_length > 0.
+  virtual void GetStats(char* buffer, int buffer_length);
+
+  // Outputs to "writer" a sample of live objects and the stack traces
+  // that allocated these objects.  The format of the returned output
+  // is equivalent to the output of the heap profiler and can
+  // therefore be passed to "pprof". This function is equivalent to
+  // ReadStackTraces. The main difference is that this function returns
+  // serialized data appropriately formatted for use by the pprof tool.
+  // NOTE: by default, tcmalloc does not do any heap sampling, and this
+  //       function will always return an empty sample.  To get useful
+  //       data from GetHeapSample, you must also set the environment
+  //       variable TCMALLOC_SAMPLE_PARAMETER to a value such as 524288.
+  virtual void GetHeapSample(MallocExtensionWriter* writer);
+
+  // Outputs to "writer" the stack traces that caused growth in the
+  // address space size.  The format of the returned output is
+  // equivalent to the output of the heap profiler and can therefore
+  // be passed to "pprof". This function is equivalent to
+  // ReadHeapGrowthStackTraces. The main difference is that this function
+  // returns serialized data appropriately formatted for use by the
+  // pprof tool.  (This does not depend on, or require,
+  // TCMALLOC_SAMPLE_PARAMETER.)
+  virtual void GetHeapGrowthStacks(MallocExtensionWriter* writer);
+
+  // Invokes func(arg, range) for every controlled memory
+  // range.  *range is filled in with information about the range.
+  //
+  // This is a best-effort interface useful only for performance
+  // analysis.  The implementation may not call func at all.
+  typedef void (RangeFunction)(void*, const base::MallocRange*);
+  virtual void Ranges(void* arg, RangeFunction func);
+
+  // -------------------------------------------------------------------
+  // Control operations for getting and setting malloc implementation
+  // specific parameters.  Some currently useful properties:
+  //
+  // generic
+  // -------
+  // "generic.current_allocated_bytes"
+  //      Number of bytes currently allocated by application
+  //      This property is not writable.
+  //
+  // "generic.heap_size"
+  //      Number of bytes in the heap ==
+  //            current_allocated_bytes +
+  //            fragmentation +
+  //            freed memory regions
+  //      This property is not writable.
+  //
+  // tcmalloc
+  // --------
+  // "tcmalloc.max_total_thread_cache_bytes"
+  //      Upper limit on total number of bytes stored across all
+  //      per-thread caches.  Default: 16MB.
+  //
+  // "tcmalloc.current_total_thread_cache_bytes"
+  //      Number of bytes used across all thread caches.
+  //      This property is not writable.
+  //
+  // "tcmalloc.central_cache_free_bytes"
+  //      Number of free bytes in the central cache that have been
+  //      assigned to size classes. They always count towards virtual
+  //      memory usage, and unless the underlying memory is swapped out
+  //      by the OS, they also count towards physical memory usage.
+  //      This property is not writable.
+  //
+  // "tcmalloc.transfer_cache_free_bytes"
+  //      Number of free bytes that are waiting to be transfered between
+  //      the central cache and a thread cache. They always count
+  //      towards virtual memory usage, and unless the underlying memory
+  //      is swapped out by the OS, they also count towards physical
+  //      memory usage. This property is not writable.
+  //
+  // "tcmalloc.thread_cache_free_bytes"
+  //      Number of free bytes in thread caches. They always count
+  //      towards virtual memory usage, and unless the underlying memory
+  //      is swapped out by the OS, they also count towards physical
+  //      memory usage. This property is not writable.
+  //
+  // "tcmalloc.pageheap_free_bytes"
+  //      Number of bytes in free, mapped pages in page heap.  These
+  //      bytes can be used to fulfill allocation requests.  They
+  //      always count towards virtual memory usage, and unless the
+  //      underlying memory is swapped out by the OS, they also count
+  //      towards physical memory usage.  This property is not writable.
+  //
+  // "tcmalloc.pageheap_unmapped_bytes"
+  //        Number of bytes in free, unmapped pages in page heap.
+  //        These are bytes that have been released back to the OS,
+  //        possibly by one of the MallocExtension "Release" calls.
+  //        They can be used to fulfill allocation requests, but
+  //        typically incur a page fault.  They always count towards
+  //        virtual memory usage, and depending on the OS, typically
+  //        do not count towards physical memory usage.  This property
+  //        is not writable.
+  // -------------------------------------------------------------------
+
+  // Get the named "property"'s value.  Returns true if the property
+  // is known.  Returns false if the property is not a valid property
+  // name for the current malloc implementation.
+  // REQUIRES: property != NULL; value != NULL
+  virtual bool GetNumericProperty(const char* property, size_t* value);
+
+  // Set the named "property"'s value.  Returns true if the property
+  // is known and writable.  Returns false if the property is not a
+  // valid property name for the current malloc implementation, or
+  // is not writable.
+  // REQUIRES: property != NULL
+  virtual bool SetNumericProperty(const char* property, size_t value);
+
+  // Mark the current thread as "idle".  This routine may optionally
+  // be called by threads as a hint to the malloc implementation that
+  // any thread-specific resources should be released.  Note: this may
+  // be an expensive routine, so it should not be called too often.
+  //
+  // Also, if the code that calls this routine will go to sleep for
+  // a while, it should take care to not allocate anything between
+  // the call to this routine and the beginning of the sleep.
+  //
+  // Most malloc implementations ignore this routine.
+  virtual void MarkThreadIdle();
+
+  // Mark the current thread as "busy".  This routine should be
+  // called after MarkThreadIdle() if the thread will now do more
+  // work.  If this method is not called, performance may suffer.
+  //
+  // Most malloc implementations ignore this routine.
+  virtual void MarkThreadBusy();
+
+  // Gets the system allocator used by the malloc extension instance. Returns
+  // NULL for malloc implementations that do not support pluggable system
+  // allocators.
+  virtual SysAllocator* GetSystemAllocator();
+
+  // Sets the system allocator to the specified.
+  //
+  // Users could register their own system allocators for malloc implementation
+  // that supports pluggable system allocators, such as TCMalloc, by doing:
+  //   alloc = new MyOwnSysAllocator();
+  //   MallocExtension::instance()->SetSystemAllocator(alloc);
+  // It's up to users whether to fall back (recommended) to the default
+  // system allocator (use GetSystemAllocator() above) or not. The caller is
+  // responsible to any necessary locking.
+  // See tcmalloc/system-alloc.h for the interface and
+  //     tcmalloc/memfs_malloc.cc for the examples.
+  //
+  // It's a no-op for malloc implementations that do not support pluggable
+  // system allocators.
+  virtual void SetSystemAllocator(SysAllocator *a);
+
+  // Try to release num_bytes of free memory back to the operating
+  // system for reuse.  Use this extension with caution -- to get this
+  // memory back may require faulting pages back in by the OS, and
+  // that may be slow.  (Currently only implemented in tcmalloc.)
+  virtual void ReleaseToSystem(size_t num_bytes);
+
+  // Same as ReleaseToSystem() but release as much memory as possible.
+  virtual void ReleaseFreeMemory();
+
+  // Sets the rate at which we release unused memory to the system.
+  // Zero means we never release memory back to the system.  Increase
+  // this flag to return memory faster; decrease it to return memory
+  // slower.  Reasonable rates are in the range [0,10].  (Currently
+  // only implemented in tcmalloc).
+  virtual void SetMemoryReleaseRate(double rate);
+
+  // Gets the release rate.  Returns a value < 0 if unknown.
+  virtual double GetMemoryReleaseRate();
+
+  // Returns the estimated number of bytes that will be allocated for
+  // a request of "size" bytes.  This is an estimate: an allocation of
+  // SIZE bytes may reserve more bytes, but will never reserve less.
+  // (Currently only implemented in tcmalloc, other implementations
+  // always return SIZE.)
+  // This is equivalent to malloc_good_size() in OS X.
+  virtual size_t GetEstimatedAllocatedSize(size_t size);
+
+  // Returns the actual number N of bytes reserved by tcmalloc for the
+  // pointer p.  The client is allowed to use the range of bytes
+  // [p, p+N) in any way it wishes (i.e. N is the "usable size" of this
+  // allocation).  This number may be equal to or greater than the number
+  // of bytes requested when p was allocated.
+  // p must have been allocated by this malloc implementation,
+  // must not be an interior pointer -- that is, must be exactly
+  // the pointer returned to by malloc() et al., not some offset
+  // from that -- and should not have been freed yet.  p may be NULL.
+  // (Currently only implemented in tcmalloc; other implementations
+  // will return 0.)
+  // This is equivalent to malloc_size() in OS X, malloc_usable_size()
+  // in glibc, and _msize() for windows.
+  virtual size_t GetAllocatedSize(const void* p);
+
+  // Returns kOwned if this malloc implementation allocated the memory
+  // pointed to by p, or kNotOwned if some other malloc implementation
+  // allocated it or p is NULL.  May also return kUnknownOwnership if
+  // the malloc implementation does not keep track of ownership.
+  // REQUIRES: p must be a value returned from a previous call to
+  // malloc(), calloc(), realloc(), memalign(), posix_memalign(),
+  // valloc(), pvalloc(), new, or new[], and must refer to memory that
+  // is currently allocated (so, for instance, you should not pass in
+  // a pointer after having called free() on it).
+  enum Ownership {
+    // NOTE: Enum values MUST be kept in sync with the version in
+    // malloc_extension_c.h
+    kUnknownOwnership = 0,
+    kOwned,
+    kNotOwned
+  };
+  virtual Ownership GetOwnership(const void* p);
+
+  // The current malloc implementation.  Always non-NULL.
+  static MallocExtension* instance();
+
+  // Change the malloc implementation.  Typically called by the
+  // malloc implementation during initialization.
+  static void Register(MallocExtension* implementation);
+
+  // Returns detailed information about malloc's freelists. For each list,
+  // return a FreeListInfo:
+  struct FreeListInfo {
+    size_t min_object_size;
+    size_t max_object_size;
+    size_t total_bytes_free;
+    const char* type;
+  };
+  // Each item in the vector refers to a different freelist. The lists
+  // are identified by the range of allocations that objects in the
+  // list can satisfy ([min_object_size, max_object_size]) and the
+  // type of freelist (see below). The current size of the list is
+  // returned in total_bytes_free (which count against a processes
+  // resident and virtual size).
+  //
+  // Currently supported types are:
+  //
+  // "tcmalloc.page{_unmapped}" - tcmalloc's page heap. An entry for each size
+  //          class in the page heap is returned. Bytes in "page_unmapped"
+  //          are no longer backed by physical memory and do not count against
+  //          the resident size of a process.
+  //
+  // "tcmalloc.large{_unmapped}" - tcmalloc's list of objects larger
+  //          than the largest page heap size class. Only one "large"
+  //          entry is returned. There is no upper-bound on the size
+  //          of objects in the large free list; this call returns
+  //          kint64max for max_object_size.  Bytes in
+  //          "large_unmapped" are no longer backed by physical memory
+  //          and do not count against the resident size of a process.
+  //
+  // "tcmalloc.central" - tcmalloc's central free-list. One entry per
+  //          size-class is returned. Never unmapped.
+  //
+  // "debug.free_queue" - free objects queued by the debug allocator
+  //                      and not returned to tcmalloc.
+  //
+  // "tcmalloc.thread" - tcmalloc's per-thread caches. Never unmapped.
+  virtual void GetFreeListSizes(std::vector<FreeListInfo>* v);
+
+  // Get a list of stack traces of sampled allocation points.  Returns
+  // a pointer to a "new[]-ed" result array, and stores the sample
+  // period in "sample_period".
+  //
+  // The state is stored as a sequence of adjacent entries
+  // in the returned array.  Each entry has the following form:
+  //    uintptr_t count;        // Number of objects with following trace
+  //    uintptr_t size;         // Total size of objects with following trace
+  //    uintptr_t depth;        // Number of PC values in stack trace
+  //    void*     stack[depth]; // PC values that form the stack trace
+  //
+  // The list of entries is terminated by a "count" of 0.
+  //
+  // It is the responsibility of the caller to "delete[]" the returned array.
+  //
+  // May return NULL to indicate no results.
+  //
+  // This is an internal extension.  Callers should use the more
+  // convenient "GetHeapSample(string*)" method defined above.
+  virtual void** ReadStackTraces(int* sample_period);
+
+  // Like ReadStackTraces(), but returns stack traces that caused growth
+  // in the address space size.
+  virtual void** ReadHeapGrowthStackTraces();
+};
+
+namespace base {
+
+// Information passed per range.  More fields may be added later.
+struct MallocRange {
+  enum Type {
+    INUSE,                // Application is using this range
+    FREE,                 // Range is currently free
+    UNMAPPED,             // Backing physical memory has been returned to the OS
+    UNKNOWN
+    // More enum values may be added in the future
+  };
+
+  uintptr_t address;    // Address of range
+  size_t length;        // Byte length of range
+  Type type;            // Type of this range
+  double fraction;      // Fraction of range that is being used (0 if !INUSE)
+
+  // Perhaps add the following:
+  // - stack trace if this range was sampled
+  // - heap growth stack trace if applicable to this range
+  // - age when allocated (for inuse) or freed (if not in use)
+};
+
+} // namespace base
+
+#endif  // BASE_MALLOC_EXTENSION_H_

BeefRT/gperftools/src/gperftools/malloc_extension_c.h

@@ -0,0 +1,99 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * --
+ * Author: Craig Silverstein
+ *
+ * C shims for the C++ malloc_extension.h.  See malloc_extension.h for
+ * details.  Note these C shims always work on
+ * MallocExtension::instance(); it is not possible to have more than
+ * one MallocExtension object in C applications.
+ */
+
+#ifndef _MALLOC_EXTENSION_C_H_
+#define _MALLOC_EXTENSION_C_H_
+
+#include <stddef.h>
+#include <sys/types.h>
+
+/* Annoying stuff for windows -- makes sure clients can import these fns */
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define kMallocExtensionHistogramSize 64
+
+PERFTOOLS_DLL_DECL int MallocExtension_VerifyAllMemory(void);
+PERFTOOLS_DLL_DECL int MallocExtension_VerifyNewMemory(const void* p);
+PERFTOOLS_DLL_DECL int MallocExtension_VerifyArrayNewMemory(const void* p);
+PERFTOOLS_DLL_DECL int MallocExtension_VerifyMallocMemory(const void* p);
+PERFTOOLS_DLL_DECL int MallocExtension_MallocMemoryStats(int* blocks, size_t* total,
+                                      int histogram[kMallocExtensionHistogramSize]);
+PERFTOOLS_DLL_DECL void MallocExtension_GetStats(char* buffer, int buffer_length);
+
+/* TODO(csilvers): write a C version of these routines, that perhaps
+ * takes a function ptr and a void *.
+ */
+/* void MallocExtension_GetHeapSample(string* result); */
+/* void MallocExtension_GetHeapGrowthStacks(string* result); */
+
+PERFTOOLS_DLL_DECL int MallocExtension_GetNumericProperty(const char* property, size_t* value);
+PERFTOOLS_DLL_DECL int MallocExtension_SetNumericProperty(const char* property, size_t value);
+PERFTOOLS_DLL_DECL void MallocExtension_MarkThreadIdle(void);
+PERFTOOLS_DLL_DECL void MallocExtension_MarkThreadBusy(void);
+PERFTOOLS_DLL_DECL void MallocExtension_ReleaseToSystem(size_t num_bytes);
+PERFTOOLS_DLL_DECL void MallocExtension_ReleaseFreeMemory(void);
+PERFTOOLS_DLL_DECL size_t MallocExtension_GetEstimatedAllocatedSize(size_t size);
+PERFTOOLS_DLL_DECL size_t MallocExtension_GetAllocatedSize(const void* p);
+
+/*
+ * NOTE: These enum values MUST be kept in sync with the version in
+ *       malloc_extension.h
+ */
+typedef enum {
+  MallocExtension_kUnknownOwnership = 0,
+  MallocExtension_kOwned,
+  MallocExtension_kNotOwned
+} MallocExtension_Ownership;
+
+PERFTOOLS_DLL_DECL MallocExtension_Ownership MallocExtension_GetOwnership(const void* p);
+
+#ifdef __cplusplus
+}   // extern "C"
+#endif
+
+#endif /* _MALLOC_EXTENSION_C_H_ */

BeefRT/gperftools/src/gperftools/malloc_hook.h

@@ -0,0 +1,358 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//
+// Some of our malloc implementations can invoke the following hooks whenever
+// memory is allocated or deallocated.  MallocHook is thread-safe, and things
+// you do before calling AddFooHook(MyHook) are visible to any resulting calls
+// to MyHook.  Hooks must be thread-safe.  If you write:
+//
+//   CHECK(MallocHook::AddNewHook(&MyNewHook));
+//
+// MyNewHook will be invoked in subsequent calls in the current thread, but
+// there are no guarantees on when it might be invoked in other threads.
+//
+// There are a limited number of slots available for each hook type.  Add*Hook
+// will return false if there are no slots available.  Remove*Hook will return
+// false if the given hook was not already installed.
+//
+// The order in which individual hooks are called in Invoke*Hook is undefined.
+//
+// It is safe for a hook to remove itself within Invoke*Hook and add other
+// hooks.  Any hooks added inside a hook invocation (for the same hook type)
+// will not be invoked for the current invocation.
+//
+// One important user of these hooks is the heap profiler.
+//
+// CAVEAT: If you add new MallocHook::Invoke* calls then those calls must be
+// directly in the code of the (de)allocation function that is provided to the
+// user and that function must have an ATTRIBUTE_SECTION(malloc_hook) attribute.
+//
+// Note: the Invoke*Hook() functions are defined in malloc_hook-inl.h.  If you
+// need to invoke a hook (which you shouldn't unless you're part of tcmalloc),
+// be sure to #include malloc_hook-inl.h in addition to malloc_hook.h.
+//
+// NOTE FOR C USERS: If you want to use malloc_hook functionality from
+// a C program, #include malloc_hook_c.h instead of this file.
+
+#ifndef _MALLOC_HOOK_H_
+#define _MALLOC_HOOK_H_
+
+#include <stddef.h>
+#include <sys/types.h>
+extern "C" {
+#include "malloc_hook_c.h"  // a C version of the malloc_hook interface
+}
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+// The C++ methods below call the C version (MallocHook_*), and thus
+// convert between an int and a bool.  Windows complains about this
+// (a "performance warning") which we don't care about, so we suppress.
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable:4800)
+#endif
+
+// Note: malloc_hook_c.h defines MallocHook_*Hook and
+// MallocHook_{Add,Remove}*Hook.  The version of these inside the MallocHook
+// class are defined in terms of the malloc_hook_c version.  See malloc_hook_c.h
+// for details of these types/functions.
+
+class PERFTOOLS_DLL_DECL MallocHook {
+ public:
+  // The NewHook is invoked whenever an object is allocated.
+  // It may be passed NULL if the allocator returned NULL.
+  typedef MallocHook_NewHook NewHook;
+  inline static bool AddNewHook(NewHook hook) {
+    return MallocHook_AddNewHook(hook);
+  }
+  inline static bool RemoveNewHook(NewHook hook) {
+    return MallocHook_RemoveNewHook(hook);
+  }
+  inline static void InvokeNewHook(const void* p, size_t s);
+
+  // The DeleteHook is invoked whenever an object is deallocated.
+  // It may be passed NULL if the caller is trying to delete NULL.
+  typedef MallocHook_DeleteHook DeleteHook;
+  inline static bool AddDeleteHook(DeleteHook hook) {
+    return MallocHook_AddDeleteHook(hook);
+  }
+  inline static bool RemoveDeleteHook(DeleteHook hook) {
+    return MallocHook_RemoveDeleteHook(hook);
+  }
+  inline static void InvokeDeleteHook(const void* p);
+
+  // The PreMmapHook is invoked with mmap or mmap64 arguments just
+  // before the call is actually made.  Such a hook may be useful
+  // in memory limited contexts, to catch allocations that will exceed
+  // a memory limit, and take outside actions to increase that limit.
+  typedef MallocHook_PreMmapHook PreMmapHook;
+  inline static bool AddPreMmapHook(PreMmapHook hook) {
+    return MallocHook_AddPreMmapHook(hook);
+  }
+  inline static bool RemovePreMmapHook(PreMmapHook hook) {
+    return MallocHook_RemovePreMmapHook(hook);
+  }
+  inline static void InvokePreMmapHook(const void* start,
+                                       size_t size,
+                                       int protection,
+                                       int flags,
+                                       int fd,
+                                       off_t offset);
+
+  // The MmapReplacement is invoked after the PreMmapHook but before
+  // the call is actually made. The MmapReplacement should return true
+  // if it handled the call, or false if it is still necessary to
+  // call mmap/mmap64.
+  // This should be used only by experts, and users must be be
+  // extremely careful to avoid recursive calls to mmap. The replacement
+  // should be async signal safe.
+  // Only one MmapReplacement is supported. After setting an MmapReplacement
+  // you must call RemoveMmapReplacement before calling SetMmapReplacement
+  // again.
+  typedef MallocHook_MmapReplacement MmapReplacement;
+  inline static bool SetMmapReplacement(MmapReplacement hook) {
+    return MallocHook_SetMmapReplacement(hook);
+  }
+  inline static bool RemoveMmapReplacement(MmapReplacement hook) {
+    return MallocHook_RemoveMmapReplacement(hook);
+  }
+  inline static bool InvokeMmapReplacement(const void* start,
+                                           size_t size,
+                                           int protection,
+                                           int flags,
+                                           int fd,
+                                           off_t offset,
+                                           void** result);
+
+
+  // The MmapHook is invoked whenever a region of memory is mapped.
+  // It may be passed MAP_FAILED if the mmap failed.
+  typedef MallocHook_MmapHook MmapHook;
+  inline static bool AddMmapHook(MmapHook hook) {
+    return MallocHook_AddMmapHook(hook);
+  }
+  inline static bool RemoveMmapHook(MmapHook hook) {
+    return MallocHook_RemoveMmapHook(hook);
+  }
+  inline static void InvokeMmapHook(const void* result,
+                                    const void* start,
+                                    size_t size,
+                                    int protection,
+                                    int flags,
+                                    int fd,
+                                    off_t offset);
+
+  // The MunmapReplacement is invoked with munmap arguments just before
+  // the call is actually made. The MunmapReplacement should return true
+  // if it handled the call, or false if it is still necessary to
+  // call munmap.
+  // This should be used only by experts. The replacement should be
+  // async signal safe.
+  // Only one MunmapReplacement is supported. After setting an
+  // MunmapReplacement you must call RemoveMunmapReplacement before
+  // calling SetMunmapReplacement again.
+  typedef MallocHook_MunmapReplacement MunmapReplacement;
+  inline static bool SetMunmapReplacement(MunmapReplacement hook) {
+    return MallocHook_SetMunmapReplacement(hook);
+  }
+  inline static bool RemoveMunmapReplacement(MunmapReplacement hook) {
+    return MallocHook_RemoveMunmapReplacement(hook);
+  }
+  inline static bool InvokeMunmapReplacement(const void* p,
+                                             size_t size,
+                                             int* result);
+
+  // The MunmapHook is invoked whenever a region of memory is unmapped.
+  typedef MallocHook_MunmapHook MunmapHook;
+  inline static bool AddMunmapHook(MunmapHook hook) {
+    return MallocHook_AddMunmapHook(hook);
+  }
+  inline static bool RemoveMunmapHook(MunmapHook hook) {
+    return MallocHook_RemoveMunmapHook(hook);
+  }
+  inline static void InvokeMunmapHook(const void* p, size_t size);
+
+  // The MremapHook is invoked whenever a region of memory is remapped.
+  typedef MallocHook_MremapHook MremapHook;
+  inline static bool AddMremapHook(MremapHook hook) {
+    return MallocHook_AddMremapHook(hook);
+  }
+  inline static bool RemoveMremapHook(MremapHook hook) {
+    return MallocHook_RemoveMremapHook(hook);
+  }
+  inline static void InvokeMremapHook(const void* result,
+                                      const void* old_addr,
+                                      size_t old_size,
+                                      size_t new_size,
+                                      int flags,
+                                      const void* new_addr);
+
+  // The PreSbrkHook is invoked just before sbrk is called -- except when
+  // the increment is 0.  This is because sbrk(0) is often called
+  // to get the top of the memory stack, and is not actually a
+  // memory-allocation call.  It may be useful in memory-limited contexts,
+  // to catch allocations that will exceed the limit and take outside
+  // actions to increase such a limit.
+  typedef MallocHook_PreSbrkHook PreSbrkHook;
+  inline static bool AddPreSbrkHook(PreSbrkHook hook) {
+    return MallocHook_AddPreSbrkHook(hook);
+  }
+  inline static bool RemovePreSbrkHook(PreSbrkHook hook) {
+    return MallocHook_RemovePreSbrkHook(hook);
+  }
+  inline static void InvokePreSbrkHook(ptrdiff_t increment);
+
+  // The SbrkHook is invoked whenever sbrk is called -- except when
+  // the increment is 0.  This is because sbrk(0) is often called
+  // to get the top of the memory stack, and is not actually a
+  // memory-allocation call.
+  typedef MallocHook_SbrkHook SbrkHook;
+  inline static bool AddSbrkHook(SbrkHook hook) {
+    return MallocHook_AddSbrkHook(hook);
+  }
+  inline static bool RemoveSbrkHook(SbrkHook hook) {
+    return MallocHook_RemoveSbrkHook(hook);
+  }
+  inline static void InvokeSbrkHook(const void* result, ptrdiff_t increment);
+
+  // Get the current stack trace.  Try to skip all routines up to and
+  // and including the caller of MallocHook::Invoke*.
+  // Use "skip_count" (similarly to GetStackTrace from stacktrace.h)
+  // as a hint about how many routines to skip if better information
+  // is not available.
+  inline static int GetCallerStackTrace(void** result, int max_depth,
+                                        int skip_count) {
+    return MallocHook_GetCallerStackTrace(result, max_depth, skip_count);
+  }
+
+  // Unhooked versions of mmap() and munmap().   These should be used
+  // only by experts, since they bypass heapchecking, etc.
+  // Note: These do not run hooks, but they still use the MmapReplacement
+  // and MunmapReplacement.
+  static void* UnhookedMMap(void *start, size_t length, int prot, int flags,
+                            int fd, off_t offset);
+  static int UnhookedMUnmap(void *start, size_t length);
+
+  // The following are DEPRECATED.
+  inline static NewHook GetNewHook();
+  inline static NewHook SetNewHook(NewHook hook) {
+    return MallocHook_SetNewHook(hook);
+  }
+
+  inline static DeleteHook GetDeleteHook();
+  inline static DeleteHook SetDeleteHook(DeleteHook hook) {
+    return MallocHook_SetDeleteHook(hook);
+  }
+
+  inline static PreMmapHook GetPreMmapHook();
+  inline static PreMmapHook SetPreMmapHook(PreMmapHook hook) {
+    return MallocHook_SetPreMmapHook(hook);
+  }
+
+  inline static MmapHook GetMmapHook();
+  inline static MmapHook SetMmapHook(MmapHook hook) {
+    return MallocHook_SetMmapHook(hook);
+  }
+
+  inline static MunmapHook GetMunmapHook();
+  inline static MunmapHook SetMunmapHook(MunmapHook hook) {
+    return MallocHook_SetMunmapHook(hook);
+  }
+
+  inline static MremapHook GetMremapHook();
+  inline static MremapHook SetMremapHook(MremapHook hook) {
+    return MallocHook_SetMremapHook(hook);
+  }
+
+  inline static PreSbrkHook GetPreSbrkHook();
+  inline static PreSbrkHook SetPreSbrkHook(PreSbrkHook hook) {
+    return MallocHook_SetPreSbrkHook(hook);
+  }
+
+  inline static SbrkHook GetSbrkHook();
+  inline static SbrkHook SetSbrkHook(SbrkHook hook) {
+    return MallocHook_SetSbrkHook(hook);
+  }
+  // End of DEPRECATED methods.
+
+ private:
+  // Slow path versions of Invoke*Hook.
+  static void InvokeNewHookSlow(const void* p, size_t s);
+  static void InvokeDeleteHookSlow(const void* p);
+  static void InvokePreMmapHookSlow(const void* start,
+                                    size_t size,
+                                    int protection,
+                                    int flags,
+                                    int fd,
+                                    off_t offset);
+  static void InvokeMmapHookSlow(const void* result,
+                                 const void* start,
+                                 size_t size,
+                                 int protection,
+                                 int flags,
+                                 int fd,
+                                 off_t offset);
+  static bool InvokeMmapReplacementSlow(const void* start,
+                                        size_t size,
+                                        int protection,
+                                        int flags,
+                                        int fd,
+                                        off_t offset,
+                                        void** result);
+  static void InvokeMunmapHookSlow(const void* p, size_t size);
+  static bool InvokeMunmapReplacementSlow(const void* p,
+                                          size_t size,
+                                          int* result);
+  static void InvokeMremapHookSlow(const void* result,
+                                   const void* old_addr,
+                                   size_t old_size,
+                                   size_t new_size,
+                                   int flags,
+                                   const void* new_addr);
+  static void InvokePreSbrkHookSlow(ptrdiff_t increment);
+  static void InvokeSbrkHookSlow(const void* result, ptrdiff_t increment);
+};
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+
+#endif /* _MALLOC_HOOK_H_ */

BeefRT/gperftools/src/gperftools/malloc_hook_c.h

@@ -0,0 +1,173 @@
+/* Copyright (c) 2008, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * --
+ * Author: Craig Silverstein
+ *
+ * C shims for the C++ malloc_hook.h.  See malloc_hook.h for details
+ * on how to use these.
+ */
+
+#ifndef _MALLOC_HOOK_C_H_
+#define _MALLOC_HOOK_C_H_
+
+#include <stddef.h>
+#include <sys/types.h>
+
+/* Annoying stuff for windows; makes sure clients can import these functions */
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Get the current stack trace.  Try to skip all routines up to and
+ * and including the caller of MallocHook::Invoke*.
+ * Use "skip_count" (similarly to GetStackTrace from stacktrace.h)
+ * as a hint about how many routines to skip if better information
+ * is not available.
+ */
+PERFTOOLS_DLL_DECL
+int MallocHook_GetCallerStackTrace(void** result, int max_depth,
+                                   int skip_count);
+
+/* The MallocHook_{Add,Remove}*Hook functions return 1 on success and 0 on
+ * failure.
+ */
+
+typedef void (*MallocHook_NewHook)(const void* ptr, size_t size);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddNewHook(MallocHook_NewHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveNewHook(MallocHook_NewHook hook);
+
+typedef void (*MallocHook_DeleteHook)(const void* ptr);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddDeleteHook(MallocHook_DeleteHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveDeleteHook(MallocHook_DeleteHook hook);
+
+typedef void (*MallocHook_PreMmapHook)(const void *start,
+                                       size_t size,
+                                       int protection,
+                                       int flags,
+                                       int fd,
+                                       off_t offset);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddPreMmapHook(MallocHook_PreMmapHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemovePreMmapHook(MallocHook_PreMmapHook hook);
+
+typedef void (*MallocHook_MmapHook)(const void* result,
+                                    const void* start,
+                                    size_t size,
+                                    int protection,
+                                    int flags,
+                                    int fd,
+                                    off_t offset);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddMmapHook(MallocHook_MmapHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveMmapHook(MallocHook_MmapHook hook);
+
+typedef int (*MallocHook_MmapReplacement)(const void* start,
+                                          size_t size,
+                                          int protection,
+                                          int flags,
+                                          int fd,
+                                          off_t offset,
+                                          void** result);
+int MallocHook_SetMmapReplacement(MallocHook_MmapReplacement hook);
+int MallocHook_RemoveMmapReplacement(MallocHook_MmapReplacement hook);
+
+typedef void (*MallocHook_MunmapHook)(const void* ptr, size_t size);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddMunmapHook(MallocHook_MunmapHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveMunmapHook(MallocHook_MunmapHook hook);
+
+typedef int (*MallocHook_MunmapReplacement)(const void* ptr,
+                                            size_t size,
+                                            int* result);
+int MallocHook_SetMunmapReplacement(MallocHook_MunmapReplacement hook);
+int MallocHook_RemoveMunmapReplacement(MallocHook_MunmapReplacement hook);
+
+typedef void (*MallocHook_MremapHook)(const void* result,
+                                      const void* old_addr,
+                                      size_t old_size,
+                                      size_t new_size,
+                                      int flags,
+                                      const void* new_addr);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddMremapHook(MallocHook_MremapHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveMremapHook(MallocHook_MremapHook hook);
+
+typedef void (*MallocHook_PreSbrkHook)(ptrdiff_t increment);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddPreSbrkHook(MallocHook_PreSbrkHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemovePreSbrkHook(MallocHook_PreSbrkHook hook);
+
+typedef void (*MallocHook_SbrkHook)(const void* result, ptrdiff_t increment);
+PERFTOOLS_DLL_DECL
+int MallocHook_AddSbrkHook(MallocHook_SbrkHook hook);
+PERFTOOLS_DLL_DECL
+int MallocHook_RemoveSbrkHook(MallocHook_SbrkHook hook);
+
+/* The following are DEPRECATED. */
+PERFTOOLS_DLL_DECL
+MallocHook_NewHook MallocHook_SetNewHook(MallocHook_NewHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_DeleteHook MallocHook_SetDeleteHook(MallocHook_DeleteHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_PreMmapHook MallocHook_SetPreMmapHook(MallocHook_PreMmapHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_MmapHook MallocHook_SetMmapHook(MallocHook_MmapHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_MunmapHook MallocHook_SetMunmapHook(MallocHook_MunmapHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_MremapHook MallocHook_SetMremapHook(MallocHook_MremapHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_PreSbrkHook MallocHook_SetPreSbrkHook(MallocHook_PreSbrkHook hook);
+PERFTOOLS_DLL_DECL
+MallocHook_SbrkHook MallocHook_SetSbrkHook(MallocHook_SbrkHook hook);
+/* End of DEPRECATED functions. */
+
+#ifdef __cplusplus
+}   // extern "C"
+#endif
+
+#endif /* _MALLOC_HOOK_C_H_ */

BeefRT/gperftools/src/gperftools/profiler.h

@@ -0,0 +1,168 @@
+/* Copyright (c) 2005, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat
+ *
+ * Module for CPU profiling based on periodic pc-sampling.
+ *
+ * For full(er) information, see doc/cpuprofile.html
+ *
+ * This module is linked into your program with
+ * no slowdown caused by this unless you activate the profiler
+ * using one of the following methods:
+ *
+ *    1. Before starting the program, set the environment variable
+ *       "PROFILE" to be the name of the file to which the profile
+ *       data should be written.
+ *
+ *    2. Programmatically, start and stop the profiler using the
+ *       routines "ProfilerStart(filename)" and "ProfilerStop()".
+ *
+ *
+ * (Note: if using linux 2.4 or earlier, only the main thread may be
+ * profiled.)
+ *
+ * Use pprof to view the resulting profile output.
+ *    % pprof <path_to_executable> <profile_file_name>
+ *    % pprof --gv  <path_to_executable> <profile_file_name>
+ *
+ * These functions are thread-safe.
+ */
+
+#ifndef BASE_PROFILER_H_
+#define BASE_PROFILER_H_
+
+#include <time.h>       /* For time_t */
+
+/* Annoying stuff for windows; makes sure clients can import these functions */
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+/* All this code should be usable from within C apps. */
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* Profiler options, for use with ProfilerStartWithOptions.  To use:
+ *
+ *   struct ProfilerOptions options;
+ *   memset(&options, 0, sizeof options);
+ *
+ * then fill in fields as needed.
+ *
+ * This structure is intended to be usable from C code, so no constructor
+ * is provided to initialize it.  (Use memset as described above).
+ */
+struct ProfilerOptions {
+  /* Filter function and argument.
+   *
+   * If filter_in_thread is not NULL, when a profiling tick is delivered
+   * the profiler will call:
+   *
+   *   (*filter_in_thread)(filter_in_thread_arg)
+   *
+   * If it returns nonzero, the sample will be included in the profile.
+   * Note that filter_in_thread runs in a signal handler, so must be
+   * async-signal-safe.
+   *
+   * A typical use would be to set up filter results for each thread
+   * in the system before starting the profiler, then to make
+   * filter_in_thread be a very simple function which retrieves those
+   * results in an async-signal-safe way.  Retrieval could be done
+   * using thread-specific data, or using a shared data structure that
+   * supports async-signal-safe lookups.
+   */
+  int (*filter_in_thread)(void *arg);
+  void *filter_in_thread_arg;
+};
+
+/* Start profiling and write profile info into fname, discarding any
+ * existing profiling data in that file.
+ *
+ * This is equivalent to calling ProfilerStartWithOptions(fname, NULL).
+ */
+PERFTOOLS_DLL_DECL int ProfilerStart(const char* fname);
+
+/* Start profiling and write profile into fname, discarding any
+ * existing profiling data in that file.
+ *
+ * The profiler is configured using the options given by 'options'.
+ * Options which are not specified are given default values.
+ *
+ * 'options' may be NULL, in which case all are given default values.
+ *
+ * Returns nonzero if profiling was started successfully, or zero else.
+ */
+PERFTOOLS_DLL_DECL int ProfilerStartWithOptions(
+    const char *fname, const struct ProfilerOptions *options);
+
+/* Stop profiling. Can be started again with ProfilerStart(), but
+ * the currently accumulated profiling data will be cleared.
+ */
+PERFTOOLS_DLL_DECL void ProfilerStop();
+
+/* Flush any currently buffered profiling state to the profile file.
+ * Has no effect if the profiler has not been started.
+ */
+PERFTOOLS_DLL_DECL void ProfilerFlush();
+
+
+/* DEPRECATED: these functions were used to enable/disable profiling
+ * in the current thread, but no longer do anything.
+ */
+PERFTOOLS_DLL_DECL void ProfilerEnable();
+PERFTOOLS_DLL_DECL void ProfilerDisable();
+
+/* Returns nonzero if profile is currently enabled, zero if it's not. */
+PERFTOOLS_DLL_DECL int ProfilingIsEnabledForAllThreads();
+
+/* Routine for registering new threads with the profiler.
+ */
+PERFTOOLS_DLL_DECL void ProfilerRegisterThread();
+
+/* Stores state about profiler's current status into "*state". */
+struct ProfilerState {
+  int    enabled;             /* Is profiling currently enabled? */
+  time_t start_time;          /* If enabled, when was profiling started? */
+  char   profile_name[1024];  /* Name of profile file being written, or '\0' */
+  int    samples_gathered;    /* Number of samples gathered so far (or 0) */
+};
+PERFTOOLS_DLL_DECL void ProfilerGetCurrentState(struct ProfilerState* state);
+
+#ifdef __cplusplus
+}  // extern "C"
+#endif
+
+#endif  /* BASE_PROFILER_H_ */

BeefRT/gperftools/src/gperftools/stacktrace.h

@@ -0,0 +1,116 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//
+// Routines to extract the current stack trace.  These functions are
+// thread-safe.
+
+#ifndef GOOGLE_STACKTRACE_H_
+#define GOOGLE_STACKTRACE_H_
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+
+// Skips the most recent "skip_count" stack frames (also skips the
+// frame generated for the "GetStackFrames" routine itself), and then
+// records the pc values for up to the next "max_depth" frames in
+// "result", and the corresponding stack frame sizes in "sizes".
+// Returns the number of values recorded in "result"/"sizes".
+//
+// Example:
+//      main() { foo(); }
+//      foo() { bar(); }
+//      bar() {
+//        void* result[10];
+//        int sizes[10];
+//        int depth = GetStackFrames(result, sizes, 10, 1);
+//      }
+//
+// The GetStackFrames call will skip the frame for "bar".  It will
+// return 2 and will produce pc values that map to the following
+// procedures:
+//      result[0]       foo
+//      result[1]       main
+// (Actually, there may be a few more entries after "main" to account for
+// startup procedures.)
+// And corresponding stack frame sizes will also be recorded:
+//    sizes[0]       16
+//    sizes[1]       16
+// (Stack frame sizes of 16 above are just for illustration purposes.)
+// Stack frame sizes of 0 or less indicate that those frame sizes couldn't
+// be identified.
+//
+// This routine may return fewer stack frame entries than are
+// available. Also note that "result" and "sizes" must both be non-NULL.
+extern PERFTOOLS_DLL_DECL int GetStackFrames(void** result, int* sizes, int max_depth,
+                          int skip_count);
+
+// Same as above, but to be used from a signal handler. The "uc" parameter
+// should be the pointer to ucontext_t which was passed as the 3rd parameter
+// to sa_sigaction signal handler. It may help the unwinder to get a
+// better stack trace under certain conditions. The "uc" may safely be NULL.
+extern PERFTOOLS_DLL_DECL int GetStackFramesWithContext(void** result, int* sizes, int max_depth,
+                                     int skip_count, const void *uc);
+
+// This is similar to the GetStackFrames routine, except that it returns
+// the stack trace only, and not the stack frame sizes as well.
+// Example:
+//      main() { foo(); }
+//      foo() { bar(); }
+//      bar() {
+//        void* result[10];
+//        int depth = GetStackTrace(result, 10, 1);
+//      }
+//
+// This produces:
+//      result[0]       foo
+//      result[1]       main
+//           ....       ...
+//
+// "result" must not be NULL.
+extern PERFTOOLS_DLL_DECL int GetStackTrace(void** result, int max_depth,
+                                            int skip_count);
+
+// Same as above, but to be used from a signal handler. The "uc" parameter
+// should be the pointer to ucontext_t which was passed as the 3rd parameter
+// to sa_sigaction signal handler. It may help the unwinder to get a
+// better stack trace under certain conditions. The "uc" may safely be NULL.
+extern PERFTOOLS_DLL_DECL int GetStackTraceWithContext(void** result, int max_depth,
+                                    int skip_count, const void *uc);
+
+#endif /* GOOGLE_STACKTRACE_H_ */

BeefRT/gperftools/src/gperftools/tcmalloc.h.in

@@ -0,0 +1,133 @@
+/* Copyright (c) 2003, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat <opensource@google.com>
+ *         .h file by Craig Silverstein <opensource@google.com>
+ */
+
+#ifndef TCMALLOC_TCMALLOC_H_
+#define TCMALLOC_TCMALLOC_H_
+
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>   // where glibc defines __THROW
+#endif
+
+// __THROW is defined in glibc systems.  It means, counter-intuitively,
+// "This function will never throw an exception."  It's an optional
+// optimization tool, but we may need to use it to match glibc prototypes.
+#ifndef __THROW    /* I guess we're not on a glibc system */
+# define __THROW   /* __THROW is just an optimization, so ok to make it "" */
+#endif
+
+// Define the version number so folks can check against it
+#define TC_VERSION_MAJOR  @TC_VERSION_MAJOR@
+#define TC_VERSION_MINOR  @TC_VERSION_MINOR@
+#define TC_VERSION_PATCH  "@TC_VERSION_PATCH@"
+#define TC_VERSION_STRING "gperftools @TC_VERSION_MAJOR@.@TC_VERSION_MINOR@@TC_VERSION_PATCH@"
+
+// For struct mallinfo, if it's defined.
+#ifdef HAVE_STRUCT_MALLINFO
+// Malloc can be in several places on older versions of OS X.
+# if defined(HAVE_MALLOC_H)
+# include <malloc.h>
+# elif defined(HAVE_SYS_MALLOC_H)
+# include <sys/malloc.h>
+# elif defined(HAVE_MALLOC_MALLOC_H)
+# include <malloc/malloc.h>
+# endif
+#endif
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+#ifdef __cplusplus
+namespace std {
+struct nothrow_t;
+}
+
+extern "C" {
+#endif
+  // Returns a human-readable version string.  If major, minor,
+  // and/or patch are not NULL, they are set to the major version,
+  // minor version, and patch-code (a string, usually "").
+  PERFTOOLS_DLL_DECL const char* tc_version(int* major, int* minor,
+                                            const char** patch) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_free(void* ptr) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_calloc(size_t nmemb, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_memalign(size_t __alignment,
+                                       size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL int tc_posix_memalign(void** ptr,
+                                           size_t align, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_valloc(size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t __size) __THROW;
+
+  PERFTOOLS_DLL_DECL void tc_malloc_stats(void) __THROW;
+  PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) __THROW;
+#if @ac_cv_have_struct_mallinfo@
+  PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) __THROW;
+#endif
+
+  // This is an alias for MallocExtension::instance()->GetAllocatedSize().
+  // It is equivalent to
+  //    OS X: malloc_size()
+  //    glibc: malloc_usable_size()
+  //    Windows: _msize()
+  PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) __THROW;
+
+#ifdef __cplusplus
+  PERFTOOLS_DLL_DECL int tc_set_new_mode(int flag) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_new(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size,
+                                          const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p,
+                                            const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_newarray(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size,
+                                               const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p,
+                                                 const std::nothrow_t&) __THROW;
+}
+#endif
+
+#endif  // #ifndef TCMALLOC_TCMALLOC_H_

BeefRT/gperftools/src/heap-checker-bcad.cc

@@ -0,0 +1,92 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// All Rights Reserved.
+//
+// Author: Maxim Lifantsev
+//
+// A file to ensure that components of heap leak checker run before
+// all global object constructors and after all global object
+// destructors.
+//
+// This file must be the last library any binary links against.
+// Otherwise, the heap checker may not be able to run early enough to
+// catalog all the global objects in your program.  If this happens,
+// and later in the program you allocate memory and have one of these
+// "uncataloged" global objects point to it, the heap checker will
+// consider that allocation to be a leak, even though it's not (since
+// the allocated object is reachable from global data and hence "live").
+
+#include <stdlib.h>      // for abort()
+#include <gperftools/malloc_extension.h>
+
+// A dummy variable to refer from heap-checker.cc.  This is to make
+// sure this file is not optimized out by the linker.
+bool heap_leak_checker_bcad_variable;
+
+extern void HeapLeakChecker_AfterDestructors();  // in heap-checker.cc
+
+// A helper class to ensure that some components of heap leak checking
+// can happen before construction and after destruction
+// of all global/static objects.
+class HeapLeakCheckerGlobalPrePost {
+ public:
+  HeapLeakCheckerGlobalPrePost() {
+    if (count_ == 0) {
+      // The 'new int' will ensure that we have run an initial malloc
+      // hook, which will set up the heap checker via
+      // MallocHook_InitAtFirstAllocation_HeapLeakChecker.  See malloc_hook.cc.
+      // This is done in this roundabout fashion in order to avoid self-deadlock
+      // if we directly called HeapLeakChecker_BeforeConstructors here.
+      delete new int;
+      // This needs to be called before the first allocation of an STL
+      // object, but after libc is done setting up threads (because it
+      // calls setenv, which requires a thread-aware errno).  By
+      // putting it here, we hope it's the first bit of code executed
+      // after the libc global-constructor code.
+      MallocExtension::Initialize();
+    }
+    ++count_;
+  }
+  ~HeapLeakCheckerGlobalPrePost() {
+    if (count_ <= 0)  abort();
+    --count_;
+    if (count_ == 0)  HeapLeakChecker_AfterDestructors();
+  }
+ private:
+  // Counter of constructions/destructions of objects of this class
+  // (just in case there are more than one of them).
+  static int count_;
+};
+
+int HeapLeakCheckerGlobalPrePost::count_ = 0;
+
+// The early-construction/late-destruction global object.
+static const HeapLeakCheckerGlobalPrePost heap_leak_checker_global_pre_post;

BeefRT/gperftools/src/heap-profile-table.cc

@@ -0,0 +1,700 @@
+#if 0
+
+// Copyright (c) 2006, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//         Maxim Lifantsev (refactoring)
+//
+
+#include <config.h>
+
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>   // for write()
+#endif
+#include <fcntl.h>    // for open()
+#ifdef HAVE_GLOB_H
+#include <glob.h>
+#ifndef GLOB_NOMATCH  // true on some old cygwins
+# define GLOB_NOMATCH 0
+#endif
+#endif
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h> // for PRIxPTR
+#endif
+#ifdef HAVE_POLL_H
+#include <poll.h>
+#endif
+#include <errno.h>
+#include <stdarg.h>
+#include <string>
+#include <map>
+#include <algorithm>  // for sort(), equal(), and copy()
+
+#include "heap-profile-table.h"
+
+#include "base/logging.h"
+#include "raw_printer.h"
+#include "symbolize.h"
+#include <gperftools/stacktrace.h>
+#include <gperftools/malloc_hook.h>
+#include "memory_region_map.h"
+#include "base/commandlineflags.h"
+#include "base/logging.h"    // for the RawFD I/O commands
+#include "base/sysinfo.h"
+
+using std::sort;
+using std::equal;
+using std::copy;
+using std::string;
+using std::map;
+
+using tcmalloc::FillProcSelfMaps;   // from sysinfo.h
+using tcmalloc::DumpProcSelfMaps;   // from sysinfo.h
+
+//----------------------------------------------------------------------
+
+DEFINE_bool(cleanup_old_heap_profiles,
+            EnvToBool("HEAP_PROFILE_CLEANUP", true),
+            "At initialization time, delete old heap profiles.");
+
+DEFINE_int32(heap_check_max_leaks,
+             EnvToInt("HEAP_CHECK_MAX_LEAKS", 20),
+             "The maximum number of leak reports to print.");
+
+//----------------------------------------------------------------------
+
+// header of the dumped heap profile
+static const char kProfileHeader[] = "heap profile: ";
+static const char kProcSelfMapsHeader[] = "\nMAPPED_LIBRARIES:\n";
+
+//----------------------------------------------------------------------
+
+const char HeapProfileTable::kFileExt[] = ".heap";
+
+//----------------------------------------------------------------------
+
+// Size for alloc_table_ and mmap_table_.
+static const int kHashTableSize = 179999;
+/*static*/ const int HeapProfileTable::kMaxStackDepth;
+
+//----------------------------------------------------------------------
+
+// We strip out different number of stack frames in debug mode
+// because less inlining happens in that case
+#ifdef NDEBUG
+static const int kStripFrames = 2;
+#else
+static const int kStripFrames = 3;
+#endif
+
+// For sorting Stats or Buckets by in-use space
+static bool ByAllocatedSpace(HeapProfileTable::Stats* a,
+                             HeapProfileTable::Stats* b) {
+  // Return true iff "a" has more allocated space than "b"
+  return (a->alloc_size - a->free_size) > (b->alloc_size - b->free_size);
+}
+
+//----------------------------------------------------------------------
+
+HeapProfileTable::HeapProfileTable(Allocator alloc, DeAllocator dealloc)
+    : alloc_(alloc), dealloc_(dealloc) {
+  // Initialize the overall profile stats.
+  memset(&total_, 0, sizeof(total_));
+
+  // Make the malloc table.
+  const int alloc_table_bytes = kHashTableSize * sizeof(*alloc_table_);
+  alloc_table_ = reinterpret_cast<Bucket**>(alloc_(alloc_table_bytes));
+  memset(alloc_table_, 0, alloc_table_bytes);
+  num_alloc_buckets_ = 0;
+
+  // Initialize the mmap table.
+  mmap_table_ = NULL;
+  num_available_mmap_buckets_ = 0;
+
+  // Make malloc and mmap allocation maps.
+  alloc_address_map_ =
+      new(alloc_(sizeof(AllocationMap))) AllocationMap(alloc_, dealloc_);
+  mmap_address_map_ = NULL;
+}
+
+HeapProfileTable::~HeapProfileTable() {
+  DeallocateBucketTable(alloc_table_);
+  alloc_table_ = NULL;
+  DeallocateBucketTable(mmap_table_);
+  mmap_table_ = NULL;
+  DeallocateAllocationMap(alloc_address_map_);
+  alloc_address_map_ = NULL;
+  DeallocateAllocationMap(mmap_address_map_);
+  mmap_address_map_ = NULL;
+}
+
+void HeapProfileTable::DeallocateAllocationMap(AllocationMap* allocation) {
+  if (allocation != NULL) {
+    alloc_address_map_->~AllocationMap();
+    dealloc_(allocation);
+  }
+}
+
+void HeapProfileTable::DeallocateBucketTable(Bucket** table) {
+  if (table != NULL) {
+    for (int b = 0; b < kHashTableSize; b++) {
+      for (Bucket* x = table[b]; x != 0; /**/) {
+        Bucket* b = x;
+        x = x->next;
+        dealloc_(b->stack);
+        dealloc_(b);
+      }
+    }
+    dealloc_(table);
+  }
+}
+
+HeapProfileTable::Bucket* HeapProfileTable::GetBucket(
+    int depth, const void* const key[], Bucket** table,
+    int* bucket_count) {
+  // Make hash-value
+  uintptr_t h = 0;
+  for (int i = 0; i < depth; i++) {
+    h += reinterpret_cast<uintptr_t>(key[i]);
+    h += h << 10;
+    h ^= h >> 6;
+  }
+  h += h << 3;
+  h ^= h >> 11;
+
+  // Lookup stack trace in table
+  unsigned int buck = ((unsigned int) h) % kHashTableSize;
+  for (Bucket* b = table[buck]; b != 0; b = b->next) {
+    if ((b->hash == h) &&
+        (b->depth == depth) &&
+        equal(key, key + depth, b->stack)) {
+      return b;
+    }
+  }
+
+  // Create new bucket
+  const size_t key_size = sizeof(key[0]) * depth;
+  const void** kcopy = reinterpret_cast<const void**>(alloc_(key_size));
+  copy(key, key + depth, kcopy);
+  Bucket* b = reinterpret_cast<Bucket*>(alloc_(sizeof(Bucket)));
+  memset(b, 0, sizeof(*b));
+  b->hash  = h;
+  b->depth = depth;
+  b->stack = kcopy;
+  b->next  = table[buck];
+  table[buck] = b;
+  if (bucket_count != NULL) {
+    ++(*bucket_count);
+  }
+  return b;
+}
+
+int HeapProfileTable::GetCallerStackTrace(
+    int skip_count, void* stack[kMaxStackDepth]) {
+  return MallocHook::GetCallerStackTrace(
+      stack, kMaxStackDepth, kStripFrames + skip_count + 1);
+}
+
+void HeapProfileTable::RecordAlloc(
+    const void* ptr, size_t bytes, int stack_depth,
+    const void* const call_stack[]) {
+  Bucket* b = GetBucket(stack_depth, call_stack, alloc_table_,
+                        &num_alloc_buckets_);
+  b->allocs++;
+  b->alloc_size += bytes;
+  total_.allocs++;
+  total_.alloc_size += bytes;
+
+  AllocValue v;
+  v.set_bucket(b);  // also did set_live(false); set_ignore(false)
+  v.bytes = bytes;
+  alloc_address_map_->Insert(ptr, v);
+}
+
+void HeapProfileTable::RecordFree(const void* ptr) {
+  AllocValue v;
+  if (alloc_address_map_->FindAndRemove(ptr, &v)) {
+    Bucket* b = v.bucket();
+    b->frees++;
+    b->free_size += v.bytes;
+    total_.frees++;
+    total_.free_size += v.bytes;
+  }
+}
+
+bool HeapProfileTable::FindAlloc(const void* ptr, size_t* object_size) const {
+  const AllocValue* alloc_value = alloc_address_map_->Find(ptr);
+  if (alloc_value != NULL) *object_size = alloc_value->bytes;
+  return alloc_value != NULL;
+}
+
+bool HeapProfileTable::FindAllocDetails(const void* ptr,
+                                        AllocInfo* info) const {
+  const AllocValue* alloc_value = alloc_address_map_->Find(ptr);
+  if (alloc_value != NULL) {
+    info->object_size = alloc_value->bytes;
+    info->call_stack = alloc_value->bucket()->stack;
+    info->stack_depth = alloc_value->bucket()->depth;
+  }
+  return alloc_value != NULL;
+}
+
+bool HeapProfileTable::FindInsideAlloc(const void* ptr,
+                                       size_t max_size,
+                                       const void** object_ptr,
+                                       size_t* object_size) const {
+  const AllocValue* alloc_value =
+    alloc_address_map_->FindInside(&AllocValueSize, max_size, ptr, object_ptr);
+  if (alloc_value != NULL) *object_size = alloc_value->bytes;
+  return alloc_value != NULL;
+}
+
+bool HeapProfileTable::MarkAsLive(const void* ptr) {
+  AllocValue* alloc = alloc_address_map_->FindMutable(ptr);
+  if (alloc && !alloc->live()) {
+    alloc->set_live(true);
+    return true;
+  }
+  return false;
+}
+
+void HeapProfileTable::MarkAsIgnored(const void* ptr) {
+  AllocValue* alloc = alloc_address_map_->FindMutable(ptr);
+  if (alloc) {
+    alloc->set_ignore(true);
+  }
+}
+
+// We'd be happier using snprintfer, but we don't to reduce dependencies.
+int HeapProfileTable::UnparseBucket(const Bucket& b,
+                                    char* buf, int buflen, int bufsize,
+                                    const char* extra,
+                                    Stats* profile_stats) {
+  if (profile_stats != NULL) {
+    profile_stats->allocs += b.allocs;
+    profile_stats->alloc_size += b.alloc_size;
+    profile_stats->frees += b.frees;
+    profile_stats->free_size += b.free_size;
+  }
+  int printed =
+    _snprintf(buf + buflen, bufsize - buflen, "%6d: %8" PRId64 " [%6d: %8" PRId64 "] @%s",
+             b.allocs - b.frees,
+             b.alloc_size - b.free_size,
+             b.allocs,
+             b.alloc_size,
+             extra);
+  // If it looks like the snprintf failed, ignore the fact we printed anything
+  if (printed < 0 || printed >= bufsize - buflen) return buflen;
+  buflen += printed;
+  for (int d = 0; d < b.depth; d++) {
+    printed = _snprintf(buf + buflen, bufsize - buflen, " 0x%08" PRIxPTR,
+                       reinterpret_cast<uintptr_t>(b.stack[d]));
+    if (printed < 0 || printed >= bufsize - buflen) return buflen;
+    buflen += printed;
+  }
+  printed = _snprintf(buf + buflen, bufsize - buflen, "\n");
+  if (printed < 0 || printed >= bufsize - buflen) return buflen;
+  buflen += printed;
+  return buflen;
+}
+
+HeapProfileTable::Bucket**
+HeapProfileTable::MakeSortedBucketList() const {
+  Bucket** list = reinterpret_cast<Bucket**>(alloc_(sizeof(Bucket) *
+      (num_alloc_buckets_ + num_available_mmap_buckets_)));
+
+  RAW_DCHECK(mmap_table_ != NULL || num_available_mmap_buckets_ == 0, "");
+
+  int n = 0;
+
+  for (int b = 0; b < kHashTableSize; b++) {
+    for (Bucket* x = alloc_table_[b]; x != 0; x = x->next) {
+      list[n++] = x;
+    }
+  }
+  RAW_DCHECK(n == num_alloc_buckets_, "");
+
+  if (mmap_table_ != NULL) {
+    for (int b = 0; b < kHashTableSize; b++) {
+      for (Bucket* x = mmap_table_[b]; x != 0; x = x->next) {
+        list[n++] = x;
+      }
+    }
+  }
+  RAW_DCHECK(n == num_alloc_buckets_ + num_available_mmap_buckets_, "");
+
+  sort(list, list + num_alloc_buckets_ + num_available_mmap_buckets_,
+       ByAllocatedSpace);
+
+  return list;
+}
+
+void HeapProfileTable::RefreshMMapData() {
+  // Make the table
+  static const int mmap_table_bytes = kHashTableSize * sizeof(*mmap_table_);
+  if (mmap_table_ == NULL) {
+    mmap_table_ = reinterpret_cast<Bucket**>(alloc_(mmap_table_bytes));
+    memset(mmap_table_, 0, mmap_table_bytes);
+  }
+  num_available_mmap_buckets_ = 0;
+
+  ClearMMapData();
+  mmap_address_map_ =
+      new(alloc_(sizeof(AllocationMap))) AllocationMap(alloc_, dealloc_);
+
+  MemoryRegionMap::LockHolder l;
+  for (MemoryRegionMap::RegionIterator r =
+           MemoryRegionMap::BeginRegionLocked();
+       r != MemoryRegionMap::EndRegionLocked(); ++r) {
+    Bucket* b =
+        GetBucket(r->call_stack_depth, r->call_stack, mmap_table_, NULL);
+    if (b->alloc_size == 0) {
+      num_available_mmap_buckets_ += 1;
+    }
+    b->allocs += 1;
+    b->alloc_size += r->end_addr - r->start_addr;
+
+    AllocValue v;
+    v.set_bucket(b);
+    v.bytes = r->end_addr - r->start_addr;
+    mmap_address_map_->Insert(reinterpret_cast<const void*>(r->start_addr), v);
+  }
+}
+
+void HeapProfileTable::ClearMMapData() {
+  if (mmap_address_map_ != NULL) {
+    mmap_address_map_->Iterate(ZeroBucketCountsIterator, this);
+    mmap_address_map_->~AllocationMap();
+    dealloc_(mmap_address_map_);
+    mmap_address_map_ = NULL;
+  }
+}
+
+void HeapProfileTable::IterateOrderedAllocContexts(
+    AllocContextIterator callback) const {
+  Bucket** list = MakeSortedBucketList();
+  AllocContextInfo info;
+  for (int i = 0; i < num_alloc_buckets_; ++i) {
+    *static_cast<Stats*>(&info) = *static_cast<Stats*>(list[i]);
+    info.stack_depth = list[i]->depth;
+    info.call_stack = list[i]->stack;
+    callback(info);
+  }
+  dealloc_(list);
+}
+
+int HeapProfileTable::FillOrderedProfile(char buf[], int size) const {
+  Bucket** list = MakeSortedBucketList();
+
+  // Our file format is "bucket, bucket, ..., bucket, proc_self_maps_info".
+  // In the cases buf is too small, we'd rather leave out the last
+  // buckets than leave out the /proc/self/maps info.  To ensure that,
+  // we actually print the /proc/self/maps info first, then move it to
+  // the end of the buffer, then write the bucket info into whatever
+  // is remaining, and then move the maps info one last time to close
+  // any gaps.  Whew!
+  int map_length = _snprintf(buf, size, "%s", kProcSelfMapsHeader);
+  if (map_length < 0 || map_length >= size) return 0;
+  bool dummy;   // "wrote_all" -- did /proc/self/maps fit in its entirety?
+  map_length += FillProcSelfMaps(buf + map_length, size - map_length, &dummy);
+  RAW_DCHECK(map_length <= size, "");
+  char* const map_start = buf + size - map_length;      // move to end
+  memmove(map_start, buf, map_length);
+  size -= map_length;
+
+  Stats stats;
+  memset(&stats, 0, sizeof(stats));
+  int bucket_length = _snprintf(buf, size, "%s", kProfileHeader);
+  if (bucket_length < 0 || bucket_length >= size) return 0;
+  Bucket total_with_mmap(total_);
+  if (mmap_table_ != NULL) {
+    total_with_mmap.alloc_size += MemoryRegionMap::MapSize();
+    total_with_mmap.free_size += MemoryRegionMap::UnmapSize();
+  }
+  bucket_length = UnparseBucket(total_with_mmap, buf, bucket_length, size,
+                                " heapprofile", &stats);
+  for (int i = 0; i < num_alloc_buckets_; i++) {
+    bucket_length = UnparseBucket(*list[i], buf, bucket_length, size, "",
+                                  &stats);
+  }
+  RAW_DCHECK(bucket_length < size, "");
+
+  dealloc_(list);
+
+  RAW_DCHECK(buf + bucket_length <= map_start, "");
+  memmove(buf + bucket_length, map_start, map_length);  // close the gap
+
+  return bucket_length + map_length;
+}
+
+inline
+void HeapProfileTable::DumpNonLiveIterator(const void* ptr, AllocValue* v,
+                                           const DumpArgs& args) {
+  if (v->live()) {
+    v->set_live(false);
+    return;
+  }
+  if (v->ignore()) {
+    return;
+  }
+  Bucket b;
+  memset(&b, 0, sizeof(b));
+  b.allocs = 1;
+  b.alloc_size = v->bytes;
+  b.depth = v->bucket()->depth;
+  b.stack = v->bucket()->stack;
+  char buf[1024];
+  int len = UnparseBucket(b, buf, 0, sizeof(buf), "", args.profile_stats);
+  RawWrite(args.fd, buf, len);
+}
+
+inline void HeapProfileTable::ZeroBucketCountsIterator(
+    const void* ptr, AllocValue* v, HeapProfileTable* heap_profile) {
+  Bucket* b = v->bucket();
+  if (b != NULL) {
+    b->allocs = 0;
+    b->alloc_size = 0;
+    b->free_size = 0;
+    b->frees = 0;
+  }
+}
+
+// Callback from NonLiveSnapshot; adds entry to arg->dest
+// if not the entry is not live and is not present in arg->base.
+void HeapProfileTable::AddIfNonLive(const void* ptr, AllocValue* v,
+                                    AddNonLiveArgs* arg) {
+  if (v->live()) {
+    v->set_live(false);
+  } else {
+    if (arg->base != NULL && arg->base->map_.Find(ptr) != NULL) {
+      // Present in arg->base, so do not save
+    } else {
+      arg->dest->Add(ptr, *v);
+    }
+  }
+}
+
+bool HeapProfileTable::WriteProfile(const char* file_name,
+                                    const Bucket& total,
+                                    AllocationMap* allocations) {
+  RAW_VLOG(1, "Dumping non-live heap profile to %s", file_name);
+  RawFD fd = RawOpenForWriting(file_name);
+  if (fd != kIllegalRawFD) {
+    RawWrite(fd, kProfileHeader, strlen(kProfileHeader));
+    char buf[512];
+    int len = UnparseBucket(total, buf, 0, sizeof(buf), " heapprofile",
+                            NULL);
+    RawWrite(fd, buf, len);
+    const DumpArgs args(fd, NULL);
+    allocations->Iterate<const DumpArgs&>(DumpNonLiveIterator, args);
+    RawWrite(fd, kProcSelfMapsHeader, strlen(kProcSelfMapsHeader));
+    DumpProcSelfMaps(fd);
+    RawClose(fd);
+    return true;
+  } else {
+    RAW_LOG(ERROR, "Failed dumping filtered heap profile to %s", file_name);
+    return false;
+  }
+}
+
+void HeapProfileTable::CleanupOldProfiles(const char* prefix) {
+  if (!FLAGS_cleanup_old_heap_profiles)
+    return;
+  string pattern = string(prefix) + ".*" + kFileExt;
+#if defined(HAVE_GLOB_H)
+  glob_t g;
+  const int r = glob(pattern.c_str(), GLOB_ERR, NULL, &g);
+  if (r == 0 || r == GLOB_NOMATCH) {
+    const int prefix_length = strlen(prefix);
+    for (int i = 0; i < g.gl_pathc; i++) {
+      const char* fname = g.gl_pathv[i];
+      if ((strlen(fname) >= prefix_length) &&
+          (memcmp(fname, prefix, prefix_length) == 0)) {
+        RAW_VLOG(1, "Removing old heap profile %s", fname);
+        unlink(fname);
+      }
+    }
+  }
+  globfree(&g);
+#else   /* HAVE_GLOB_H */
+  RAW_LOG(WARNING, "Unable to remove old heap profiles (can't run glob())");
+#endif
+}
+
+HeapProfileTable::Snapshot* HeapProfileTable::TakeSnapshot() {
+  Snapshot* s = new (alloc_(sizeof(Snapshot))) Snapshot(alloc_, dealloc_);
+  alloc_address_map_->Iterate(AddToSnapshot, s);
+  return s;
+}
+
+void HeapProfileTable::ReleaseSnapshot(Snapshot* s) {
+  s->~Snapshot();
+  dealloc_(s);
+}
+
+// Callback from TakeSnapshot; adds a single entry to snapshot
+void HeapProfileTable::AddToSnapshot(const void* ptr, AllocValue* v,
+                                     Snapshot* snapshot) {
+  snapshot->Add(ptr, *v);
+}
+
+HeapProfileTable::Snapshot* HeapProfileTable::NonLiveSnapshot(
+    Snapshot* base) {
+  RAW_VLOG(2, "NonLiveSnapshot input: %d %d\n",
+           int(total_.allocs - total_.frees),
+           int(total_.alloc_size - total_.free_size));
+
+  Snapshot* s = new (alloc_(sizeof(Snapshot))) Snapshot(alloc_, dealloc_);
+  AddNonLiveArgs args;
+  args.dest = s;
+  args.base = base;
+  alloc_address_map_->Iterate<AddNonLiveArgs*>(AddIfNonLive, &args);
+  RAW_VLOG(2, "NonLiveSnapshot output: %d %d\n",
+           int(s->total_.allocs - s->total_.frees),
+           int(s->total_.alloc_size - s->total_.free_size));
+  return s;
+}
+
+// Information kept per unique bucket seen
+struct HeapProfileTable::Snapshot::Entry {
+  int count;
+  int bytes;
+  Bucket* bucket;
+  Entry() : count(0), bytes(0) { }
+
+  // Order by decreasing bytes
+  bool operator<(const Entry& x) const {
+    return this->bytes > x.bytes;
+  }
+};
+
+// State used to generate leak report.  We keep a mapping from Bucket pointer
+// the collected stats for that bucket.
+struct HeapProfileTable::Snapshot::ReportState {
+  map<Bucket*, Entry> buckets_;
+};
+
+// Callback from ReportLeaks; updates ReportState.
+void HeapProfileTable::Snapshot::ReportCallback(const void* ptr,
+                                                AllocValue* v,
+                                                ReportState* state) {
+  Entry* e = &state->buckets_[v->bucket()]; // Creates empty Entry first time
+  e->bucket = v->bucket();
+  e->count++;
+  e->bytes += v->bytes;
+}
+
+void HeapProfileTable::Snapshot::ReportLeaks(const char* checker_name,
+                                             const char* filename,
+                                             bool should_symbolize) {
+  // This is only used by the heap leak checker, but is intimately
+  // tied to the allocation map that belongs in this module and is
+  // therefore placed here.
+  RAW_LOG(ERROR, "Leak check %s detected leaks of %" PRIuS " bytes "
+          "in %" PRIuS " objects",
+          checker_name,
+          size_t(total_.alloc_size),
+          size_t(total_.allocs));
+
+  // Group objects by Bucket
+  ReportState state;
+  map_.Iterate(&ReportCallback, &state);
+
+  // Sort buckets by decreasing leaked size
+  const int n = state.buckets_.size();
+  Entry* entries = new Entry[n];
+  int dst = 0;
+  for (map<Bucket*,Entry>::const_iterator iter = state.buckets_.begin();
+       iter != state.buckets_.end();
+       ++iter) {
+    entries[dst++] = iter->second;
+  }
+  sort(entries, entries + n);
+
+  // Report a bounded number of leaks to keep the leak report from
+  // growing too long.
+  const int to_report =
+      (FLAGS_heap_check_max_leaks > 0 &&
+       n > FLAGS_heap_check_max_leaks) ? FLAGS_heap_check_max_leaks : n;
+  RAW_LOG(ERROR, "The %d largest leaks:", to_report);
+
+  // Print
+  SymbolTable symbolization_table;
+  for (int i = 0; i < to_report; i++) {
+    const Entry& e = entries[i];
+    for (int j = 0; j < e.bucket->depth; j++) {
+      symbolization_table.Add(e.bucket->stack[j]);
+    }
+  }
+  static const int kBufSize = 2<<10;
+  char buffer[kBufSize];
+  if (should_symbolize)
+    symbolization_table.Symbolize();
+  for (int i = 0; i < to_report; i++) {
+    const Entry& e = entries[i];
+    base::RawPrinter printer(buffer, kBufSize);
+    printer.Printf("Leak of %d bytes in %d objects allocated from:\n",
+                   e.bytes, e.count);
+    for (int j = 0; j < e.bucket->depth; j++) {
+      const void* pc = e.bucket->stack[j];
+      printer.Printf("\t@ %" PRIxPTR " %s\n",
+          reinterpret_cast<uintptr_t>(pc), symbolization_table.GetSymbol(pc));
+    }
+    RAW_LOG(ERROR, "%s", buffer);
+  }
+
+  if (to_report < n) {
+    RAW_LOG(ERROR, "Skipping leaks numbered %d..%d",
+            to_report, n-1);
+  }
+  delete[] entries;
+
+  // TODO: Dump the sorted Entry list instead of dumping raw data?
+  // (should be much shorter)
+  if (!HeapProfileTable::WriteProfile(filename, total_, &map_)) {
+    RAW_LOG(ERROR, "Could not write pprof profile to %s", filename);
+  }
+}
+
+void HeapProfileTable::Snapshot::ReportObject(const void* ptr,
+                                              AllocValue* v,
+                                              char* unused) {
+  // Perhaps also log the allocation stack trace (unsymbolized)
+  // on this line in case somebody finds it useful.
+  RAW_LOG(ERROR, "leaked %" PRIuS " byte object %p", v->bytes, ptr);
+}
+
+void HeapProfileTable::Snapshot::ReportIndividualObjects() {
+  char unused;
+  map_.Iterate(ReportObject, &unused);
+}
+
+#endif

BeefRT/gperftools/src/heap-profile-table.h

@@ -0,0 +1,422 @@
+// Copyright (c) 2006, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//         Maxim Lifantsev (refactoring)
+//
+
+#ifndef BASE_HEAP_PROFILE_TABLE_H_
+#define BASE_HEAP_PROFILE_TABLE_H_
+
+#include "addressmap-inl.h"
+#include "base/basictypes.h"
+#include "base/logging.h"   // for RawFD
+
+// Table to maintain a heap profile data inside,
+// i.e. the set of currently active heap memory allocations.
+// thread-unsafe and non-reentrant code:
+// each instance object must be used by one thread
+// at a time w/o self-recursion.
+//
+// TODO(maxim): add a unittest for this class.
+class HeapProfileTable {
+ public:
+
+  // Extension to be used for heap pforile files.
+  static const char kFileExt[];
+
+  // Longest stack trace we record.
+  static const int kMaxStackDepth = 32;
+
+  // data types ----------------------------
+
+  // Profile stats.
+  struct Stats {
+    int32 allocs;      // Number of allocation calls
+    int32 frees;       // Number of free calls
+    int64 alloc_size;  // Total size of all allocated objects so far
+    int64 free_size;   // Total size of all freed objects so far
+
+    // semantic equality
+    bool Equivalent(const Stats& x) const {
+      return allocs - frees == x.allocs - x.frees  &&
+             alloc_size - free_size == x.alloc_size - x.free_size;
+    }
+  };
+
+  // Info we can return about an allocation.
+  struct AllocInfo {
+    size_t object_size;  // size of the allocation
+    const void* const* call_stack;  // call stack that made the allocation call
+    int stack_depth;  // depth of call_stack
+    bool live;
+    bool ignored;
+  };
+
+  // Info we return about an allocation context.
+  // An allocation context is a unique caller stack trace
+  // of an allocation operation.
+  struct AllocContextInfo : public Stats {
+    int stack_depth;                // Depth of stack trace
+    const void* const* call_stack;  // Stack trace
+  };
+
+  // Memory (de)allocator interface we'll use.
+  typedef void* (*Allocator)(size_t size);
+  typedef void  (*DeAllocator)(void* ptr);
+
+  // interface ---------------------------
+
+  HeapProfileTable(Allocator alloc, DeAllocator dealloc);
+  ~HeapProfileTable();
+
+  // Collect the stack trace for the function that asked to do the
+  // allocation for passing to RecordAlloc() below.
+  //
+  // The stack trace is stored in 'stack'. The stack depth is returned.
+  //
+  // 'skip_count' gives the number of stack frames between this call
+  // and the memory allocation function.
+  static int GetCallerStackTrace(int skip_count, void* stack[kMaxStackDepth]);
+
+  // Record an allocation at 'ptr' of 'bytes' bytes.  'stack_depth'
+  // and 'call_stack' identifying the function that requested the
+  // allocation. They can be generated using GetCallerStackTrace() above.
+  void RecordAlloc(const void* ptr, size_t bytes,
+                   int stack_depth, const void* const call_stack[]);
+
+  // Record the deallocation of memory at 'ptr'.
+  void RecordFree(const void* ptr);
+
+  // Return true iff we have recorded an allocation at 'ptr'.
+  // If yes, fill *object_size with the allocation byte size.
+  bool FindAlloc(const void* ptr, size_t* object_size) const;
+  // Same as FindAlloc, but fills all of *info.
+  bool FindAllocDetails(const void* ptr, AllocInfo* info) const;
+
+  // Return true iff "ptr" points into a recorded allocation
+  // If yes, fill *object_ptr with the actual allocation address
+  // and *object_size with the allocation byte size.
+  // max_size specifies largest currently possible allocation size.
+  bool FindInsideAlloc(const void* ptr, size_t max_size,
+                       const void** object_ptr, size_t* object_size) const;
+
+  // If "ptr" points to a recorded allocation and it's not marked as live
+  // mark it as live and return true. Else return false.
+  // All allocations start as non-live.
+  bool MarkAsLive(const void* ptr);
+
+  // If "ptr" points to a recorded allocation, mark it as "ignored".
+  // Ignored objects are treated like other objects, except that they
+  // are skipped in heap checking reports.
+  void MarkAsIgnored(const void* ptr);
+
+  // Return current total (de)allocation statistics.  It doesn't contain
+  // mmap'ed regions.
+  const Stats& total() const { return total_; }
+
+  // Allocation data iteration callback: gets passed object pointer and
+  // fully-filled AllocInfo.
+  typedef void (*AllocIterator)(const void* ptr, const AllocInfo& info);
+
+  // Iterate over the allocation profile data calling "callback"
+  // for every allocation.
+  void IterateAllocs(AllocIterator callback) const {
+    alloc_address_map_->Iterate(MapArgsAllocIterator, callback);
+  }
+
+  // Allocation context profile data iteration callback
+  typedef void (*AllocContextIterator)(const AllocContextInfo& info);
+
+  // Iterate over the allocation context profile data calling "callback"
+  // for every allocation context. Allocation contexts are ordered by the
+  // size of allocated space.
+  void IterateOrderedAllocContexts(AllocContextIterator callback) const;
+
+  // Fill profile data into buffer 'buf' of size 'size'
+  // and return the actual size occupied by the dump in 'buf'.
+  // The profile buckets are dumped in the decreasing order
+  // of currently allocated bytes.
+  // We do not provision for 0-terminating 'buf'.
+  int FillOrderedProfile(char buf[], int size) const;
+
+  // Cleanup any old profile files matching prefix + ".*" + kFileExt.
+  static void CleanupOldProfiles(const char* prefix);
+
+  // Return a snapshot of the current contents of *this.
+  // Caller must call ReleaseSnapshot() on result when no longer needed.
+  // The result is only valid while this exists and until
+  // the snapshot is discarded by calling ReleaseSnapshot().
+  class Snapshot;
+  Snapshot* TakeSnapshot();
+
+  // Release a previously taken snapshot.  snapshot must not
+  // be used after this call.
+  void ReleaseSnapshot(Snapshot* snapshot);
+
+  // Return a snapshot of every non-live, non-ignored object in *this.
+  // If "base" is non-NULL, skip any objects present in "base".
+  // As a side-effect, clears the "live" bit on every live object in *this.
+  // Caller must call ReleaseSnapshot() on result when no longer needed.
+  Snapshot* NonLiveSnapshot(Snapshot* base);
+
+  // Refresh the internal mmap information from MemoryRegionMap.  Results of
+  // FillOrderedProfile and IterateOrderedAllocContexts will contain mmap'ed
+  // memory regions as at calling RefreshMMapData.
+  void RefreshMMapData();
+
+  // Clear the internal mmap information.  Results of FillOrderedProfile and
+  // IterateOrderedAllocContexts won't contain mmap'ed memory regions after
+  // calling ClearMMapData.
+  void ClearMMapData();
+
+ private:
+
+  // data types ----------------------------
+
+  // Hash table bucket to hold (de)allocation stats
+  // for a given allocation call stack trace.
+  struct Bucket : public Stats {
+    uintptr_t    hash;   // Hash value of the stack trace
+    int          depth;  // Depth of stack trace
+    const void** stack;  // Stack trace
+    Bucket*      next;   // Next entry in hash-table
+  };
+
+  // Info stored in the address map
+  struct AllocValue {
+    // Access to the stack-trace bucket
+    Bucket* bucket() const {
+      return reinterpret_cast<Bucket*>(bucket_rep & ~uintptr_t(kMask));
+    }
+    // This also does set_live(false).
+    void set_bucket(Bucket* b) { bucket_rep = reinterpret_cast<uintptr_t>(b); }
+    size_t  bytes;   // Number of bytes in this allocation
+
+    // Access to the allocation liveness flag (for leak checking)
+    bool live() const { return bucket_rep & kLive; }
+    void set_live(bool l) {
+      bucket_rep = (bucket_rep & ~uintptr_t(kLive)) | (l ? kLive : 0);
+    }
+
+    // Should this allocation be ignored if it looks like a leak?
+    bool ignore() const { return (bucket_rep & kIgnore) != 0; }
+    void set_ignore(bool r) {
+      bucket_rep = (bucket_rep & ~uintptr_t(kIgnore)) | (r ? kIgnore : 0);
+    }
+
+   private:
+    // We store a few bits in the bottom bits of bucket_rep.
+    // (Alignment is at least four, so we have at least two bits.)
+    static const int kLive = 1;
+    static const int kIgnore = 2;
+    static const int kMask = kLive | kIgnore;
+
+    uintptr_t bucket_rep;
+  };
+
+  // helper for FindInsideAlloc
+  static size_t AllocValueSize(const AllocValue& v) { return v.bytes; }
+
+  typedef AddressMap<AllocValue> AllocationMap;
+
+  // Arguments that need to be passed DumpNonLiveIterator callback below.
+  struct DumpArgs {
+    RawFD fd;  // file to write to
+    Stats* profile_stats;  // stats to update (may be NULL)
+
+    DumpArgs(RawFD a, Stats* d)
+      : fd(a), profile_stats(d) { }
+  };
+
+  // helpers ----------------------------
+
+  // Unparse bucket b and print its portion of profile dump into buf.
+  // We return the amount of space in buf that we use.  We start printing
+  // at buf + buflen, and promise not to go beyond buf + bufsize.
+  // We do not provision for 0-terminating 'buf'.
+  //
+  // If profile_stats is non-NULL, we update *profile_stats by
+  // counting bucket b.
+  //
+  // "extra" is appended to the unparsed bucket.  Typically it is empty,
+  // but may be set to something like " heapprofile" for the total
+  // bucket to indicate the type of the profile.
+  static int UnparseBucket(const Bucket& b,
+                           char* buf, int buflen, int bufsize,
+                           const char* extra,
+                           Stats* profile_stats);
+
+  // Deallocate a given allocation map.
+  void DeallocateAllocationMap(AllocationMap* allocation);
+
+  // Deallocate a given bucket table.
+  void DeallocateBucketTable(Bucket** table);
+
+  // Get the bucket for the caller stack trace 'key' of depth 'depth' from a
+  // bucket hash map 'table' creating the bucket if needed.  '*bucket_count'
+  // is incremented both when 'bucket_count' is not NULL and when a new
+  // bucket object is created.
+  Bucket* GetBucket(int depth, const void* const key[], Bucket** table,
+                    int* bucket_count);
+
+  // Helper for IterateAllocs to do callback signature conversion
+  // from AllocationMap::Iterate to AllocIterator.
+  static void MapArgsAllocIterator(const void* ptr, AllocValue* v,
+                                   AllocIterator callback) {
+    AllocInfo info;
+    info.object_size = v->bytes;
+    info.call_stack = v->bucket()->stack;
+    info.stack_depth = v->bucket()->depth;
+    info.live = v->live();
+    info.ignored = v->ignore();
+    callback(ptr, info);
+  }
+
+  // Helper for DumpNonLiveProfile to do object-granularity
+  // heap profile dumping. It gets passed to AllocationMap::Iterate.
+  inline static void DumpNonLiveIterator(const void* ptr, AllocValue* v,
+                                         const DumpArgs& args);
+
+  // Helper for filling size variables in buckets by zero.
+  inline static void ZeroBucketCountsIterator(
+      const void* ptr, AllocValue* v, HeapProfileTable* heap_profile);
+
+  // Helper for IterateOrderedAllocContexts and FillOrderedProfile.
+  // Creates a sorted list of Buckets whose length is num_alloc_buckets_ +
+  // num_avaliable_mmap_buckets_.
+  // The caller is responsible for deallocating the returned list.
+  Bucket** MakeSortedBucketList() const;
+
+  // Helper for TakeSnapshot.  Saves object to snapshot.
+  static void AddToSnapshot(const void* ptr, AllocValue* v, Snapshot* s);
+
+  // Arguments passed to AddIfNonLive
+  struct AddNonLiveArgs {
+    Snapshot* dest;
+    Snapshot* base;
+  };
+
+  // Helper for NonLiveSnapshot.  Adds the object to the destination
+  // snapshot if it is non-live.
+  static void AddIfNonLive(const void* ptr, AllocValue* v,
+                           AddNonLiveArgs* arg);
+
+  // Write contents of "*allocations" as a heap profile to
+  // "file_name".  "total" must contain the total of all entries in
+  // "*allocations".
+  static bool WriteProfile(const char* file_name,
+                           const Bucket& total,
+                           AllocationMap* allocations);
+
+  // data ----------------------------
+
+  // Memory (de)allocator that we use.
+  Allocator alloc_;
+  DeAllocator dealloc_;
+
+  // Overall profile stats; we use only the Stats part,
+  // but make it a Bucket to pass to UnparseBucket.
+  // It doesn't contain mmap'ed regions.
+  Bucket total_;
+
+  // Bucket hash table for malloc.
+  // We hand-craft one instead of using one of the pre-written
+  // ones because we do not want to use malloc when operating on the table.
+  // It is only few lines of code, so no big deal.
+  Bucket** alloc_table_;
+  int num_alloc_buckets_;
+
+  // Bucket hash table for mmap.
+  // This table is filled with the information from MemoryRegionMap by calling
+  // RefreshMMapData.
+  Bucket** mmap_table_;
+  int num_available_mmap_buckets_;
+
+  // Map of all currently allocated objects and mapped regions we know about.
+  AllocationMap* alloc_address_map_;
+  AllocationMap* mmap_address_map_;
+
+  DISALLOW_COPY_AND_ASSIGN(HeapProfileTable);
+};
+
+class HeapProfileTable::Snapshot {
+ public:
+  const Stats& total() const { return total_; }
+
+  // Report anything in this snapshot as a leak.
+  // May use new/delete for temporary storage.
+  // If should_symbolize is true, will fork (which is not threadsafe)
+  // to turn addresses into symbol names.  Set to false for maximum safety.
+  // Also writes a heap profile to "filename" that contains
+  // all of the objects in this snapshot.
+  void ReportLeaks(const char* checker_name, const char* filename,
+                   bool should_symbolize);
+
+  // Report the addresses of all leaked objects.
+  // May use new/delete for temporary storage.
+  void ReportIndividualObjects();
+
+  bool Empty() const {
+    return (total_.allocs == 0) && (total_.alloc_size == 0);
+  }
+
+ private:
+  friend class HeapProfileTable;
+
+  // Total count/size are stored in a Bucket so we can reuse UnparseBucket
+  Bucket total_;
+
+  // We share the Buckets managed by the parent table, but have our
+  // own object->bucket map.
+  AllocationMap map_;
+
+  Snapshot(Allocator alloc, DeAllocator dealloc) : map_(alloc, dealloc) {
+    memset(&total_, 0, sizeof(total_));
+  }
+
+  // Callback used to populate a Snapshot object with entries found
+  // in another allocation map.
+  inline void Add(const void* ptr, const AllocValue& v) {
+    map_.Insert(ptr, v);
+    total_.allocs++;
+    total_.alloc_size += v.bytes;
+  }
+
+  // Helpers for sorting and generating leak reports
+  struct Entry;
+  struct ReportState;
+  static void ReportCallback(const void* ptr, AllocValue* v, ReportState*);
+  static void ReportObject(const void* ptr, AllocValue* v, char*);
+
+  DISALLOW_COPY_AND_ASSIGN(Snapshot);
+};
+
+#endif  // BASE_HEAP_PROFILE_TABLE_H_

BeefRT/gperftools/src/heap-profiler.cc

@@ -0,0 +1,569 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//
+// TODO: Log large allocations
+
+#include <config.h>
+#include <stddef.h>
+#include <stdio.h>
+#include <stdlib.h>
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h>
+#endif
+#ifdef HAVE_FCNTL_H
+#include <fcntl.h>    // for open()
+#endif
+#ifdef HAVE_MMAP
+#include <sys/mman.h>
+#endif
+#include <errno.h>
+#include <assert.h>
+#include <sys/types.h>
+
+#include <algorithm>
+#include <string>
+
+#include <gperftools/heap-profiler.h>
+
+#include "base/logging.h"
+#include "base/basictypes.h"   // for PRId64, among other things
+#include "base/googleinit.h"
+#include "base/commandlineflags.h"
+#include "malloc_hook-inl.h"
+#include "tcmalloc_guard.h"
+#include <gperftools/malloc_hook.h>
+#include <gperftools/malloc_extension.h>
+#include "base/spinlock.h"
+#include "base/low_level_alloc.h"
+#include "base/sysinfo.h"      // for GetUniquePathFromEnv()
+#include "heap-profile-table.h"
+#include "memory_region_map.h"
+
+
+#ifndef	PATH_MAX
+#ifdef MAXPATHLEN
+#define	PATH_MAX	MAXPATHLEN
+#else
+#define	PATH_MAX	4096         // seems conservative for max filename len!
+#endif
+#endif
+
+using STL_NAMESPACE::string;
+using STL_NAMESPACE::sort;
+
+//----------------------------------------------------------------------
+// Flags that control heap-profiling
+//
+// The thread-safety of the profiler depends on these being immutable
+// after main starts, so don't change them.
+//----------------------------------------------------------------------
+
+DEFINE_int64(heap_profile_allocation_interval,
+             EnvToInt64("HEAP_PROFILE_ALLOCATION_INTERVAL", 1 << 30 /*1GB*/),
+             "If non-zero, dump heap profiling information once every "
+             "specified number of bytes allocated by the program since "
+             "the last dump.");
+DEFINE_int64(heap_profile_deallocation_interval,
+             EnvToInt64("HEAP_PROFILE_DEALLOCATION_INTERVAL", 0),
+             "If non-zero, dump heap profiling information once every "
+             "specified number of bytes deallocated by the program "
+             "since the last dump.");
+// We could also add flags that report whenever inuse_bytes changes by
+// X or -X, but there hasn't been a need for that yet, so we haven't.
+DEFINE_int64(heap_profile_inuse_interval,
+             EnvToInt64("HEAP_PROFILE_INUSE_INTERVAL", 100 << 20 /*100MB*/),
+             "If non-zero, dump heap profiling information whenever "
+             "the high-water memory usage mark increases by the specified "
+             "number of bytes.");
+DEFINE_int64(heap_profile_time_interval,
+             EnvToInt64("HEAP_PROFILE_TIME_INTERVAL", 0),
+             "If non-zero, dump heap profiling information once every "
+             "specified number of seconds since the last dump.");
+DEFINE_bool(mmap_log,
+            EnvToBool("HEAP_PROFILE_MMAP_LOG", false),
+            "Should mmap/munmap calls be logged?");
+DEFINE_bool(mmap_profile,
+            EnvToBool("HEAP_PROFILE_MMAP", false),
+            "If heap-profiling is on, also profile mmap, mremap, and sbrk)");
+DEFINE_bool(only_mmap_profile,
+            EnvToBool("HEAP_PROFILE_ONLY_MMAP", false),
+            "If heap-profiling is on, only profile mmap, mremap, and sbrk; "
+            "do not profile malloc/new/etc");
+
+
+//----------------------------------------------------------------------
+// Locking
+//----------------------------------------------------------------------
+
+// A pthread_mutex has way too much lock contention to be used here.
+//
+// I would like to use Mutex, but it can call malloc(),
+// which can cause us to fall into an infinite recursion.
+//
+// So we use a simple spinlock.
+static SpinLock heap_lock(SpinLock::LINKER_INITIALIZED);
+
+//----------------------------------------------------------------------
+// Simple allocator for heap profiler's internal memory
+//----------------------------------------------------------------------
+
+static LowLevelAlloc::Arena *heap_profiler_memory;
+
+static void* ProfilerMalloc(size_t bytes) {
+  return LowLevelAlloc::AllocWithArena(bytes, heap_profiler_memory);
+}
+static void ProfilerFree(void* p) {
+  LowLevelAlloc::Free(p);
+}
+
+// We use buffers of this size in DoGetHeapProfile.
+static const int kProfileBufferSize = 1 << 20;
+
+// This is a last-ditch buffer we use in DumpProfileLocked in case we
+// can't allocate more memory from ProfilerMalloc.  We expect this
+// will be used by HeapProfileEndWriter when the application has to
+// exit due to out-of-memory.  This buffer is allocated in
+// HeapProfilerStart.  Access to this must be protected by heap_lock.
+static char* global_profiler_buffer = NULL;
+
+
+//----------------------------------------------------------------------
+// Profiling control/state data
+//----------------------------------------------------------------------
+
+// Access to all of these is protected by heap_lock.
+static bool  is_on = false;           // If are on as a subsytem.
+static bool  dumping = false;         // Dumping status to prevent recursion
+static char* filename_prefix = NULL;  // Prefix used for profile file names
+                                      // (NULL if no need for dumping yet)
+static int   dump_count = 0;          // How many dumps so far
+static int64 last_dump_alloc = 0;     // alloc_size when did we last dump
+static int64 last_dump_free = 0;      // free_size when did we last dump
+static int64 high_water_mark = 0;     // In-use-bytes at last high-water dump
+static int64 last_dump_time = 0;      // The time of the last dump
+
+static HeapProfileTable* heap_profile = NULL;  // the heap profile table
+
+//----------------------------------------------------------------------
+// Profile generation
+//----------------------------------------------------------------------
+
+// Input must be a buffer of size at least 1MB.
+static char* DoGetHeapProfileLocked(char* buf, int buflen) {
+  // We used to be smarter about estimating the required memory and
+  // then capping it to 1MB and generating the profile into that.
+  if (buf == NULL || buflen < 1)
+    return NULL;
+
+  RAW_DCHECK(heap_lock.IsHeld(), "");
+  int bytes_written = 0;
+  if (is_on) {
+    if (FLAGS_mmap_profile) {
+      heap_profile->RefreshMMapData();
+    }
+    bytes_written = heap_profile->FillOrderedProfile(buf, buflen - 1);
+    if (FLAGS_mmap_profile) {
+      heap_profile->ClearMMapData();
+    }
+  }
+  buf[bytes_written] = '\0';
+  RAW_DCHECK(bytes_written == strlen(buf), "");
+
+  return buf;
+}
+
+extern "C" char* GetHeapProfile() {
+  // Use normal malloc: we return the profile to the user to free it:
+  char* buffer = reinterpret_cast<char*>(malloc(kProfileBufferSize));
+  SpinLockHolder l(&heap_lock);
+  return DoGetHeapProfileLocked(buffer, kProfileBufferSize);
+}
+
+// defined below
+static void NewHook(const void* ptr, size_t size);
+static void DeleteHook(const void* ptr);
+
+// Helper for HeapProfilerDump.
+static void DumpProfileLocked(const char* reason) {
+  RAW_DCHECK(heap_lock.IsHeld(), "");
+  RAW_DCHECK(is_on, "");
+  RAW_DCHECK(!dumping, "");
+
+  if (filename_prefix == NULL) return;  // we do not yet need dumping
+
+  dumping = true;
+
+  // Make file name
+  char file_name[1000];
+  dump_count++;
+  snprintf(file_name, sizeof(file_name), "%s.%04d%s",
+           filename_prefix, dump_count, HeapProfileTable::kFileExt);
+
+  // Dump the profile
+  RAW_VLOG(0, "Dumping heap profile to %s (%s)", file_name, reason);
+  // We must use file routines that don't access memory, since we hold
+  // a memory lock now.
+  RawFD fd = RawOpenForWriting(file_name);
+  if (fd == kIllegalRawFD) {
+    RAW_LOG(ERROR, "Failed dumping heap profile to %s", file_name);
+    dumping = false;
+    return;
+  }
+
+  // This case may be impossible, but it's best to be safe.
+  // It's safe to use the global buffer: we're protected by heap_lock.
+  if (global_profiler_buffer == NULL) {
+    global_profiler_buffer =
+        reinterpret_cast<char*>(ProfilerMalloc(kProfileBufferSize));
+  }
+
+  char* profile = DoGetHeapProfileLocked(global_profiler_buffer,
+                                         kProfileBufferSize);
+  RawWrite(fd, profile, strlen(profile));
+  RawClose(fd);
+
+  dumping = false;
+}
+
+//----------------------------------------------------------------------
+// Profile collection
+//----------------------------------------------------------------------
+
+// Dump a profile after either an allocation or deallocation, if
+// the memory use has changed enough since the last dump.
+static void MaybeDumpProfileLocked() {
+  if (!dumping) {
+    const HeapProfileTable::Stats& total = heap_profile->total();
+    const int64 inuse_bytes = total.alloc_size - total.free_size;
+    bool need_to_dump = false;
+    char buf[128];
+    int64 current_time = time(NULL);
+    if (FLAGS_heap_profile_allocation_interval > 0 &&
+        total.alloc_size >=
+        last_dump_alloc + FLAGS_heap_profile_allocation_interval) {
+      snprintf(buf, sizeof(buf), ("%" PRId64 " MB allocated cumulatively, "
+                                  "%" PRId64 " MB currently in use"),
+               total.alloc_size >> 20, inuse_bytes >> 20);
+      need_to_dump = true;
+    } else if (FLAGS_heap_profile_deallocation_interval > 0 &&
+               total.free_size >=
+               last_dump_free + FLAGS_heap_profile_deallocation_interval) {
+      snprintf(buf, sizeof(buf), ("%" PRId64 " MB freed cumulatively, "
+                                  "%" PRId64 " MB currently in use"),
+               total.free_size >> 20, inuse_bytes >> 20);
+      need_to_dump = true;
+    } else if (FLAGS_heap_profile_inuse_interval > 0 &&
+               inuse_bytes >
+               high_water_mark + FLAGS_heap_profile_inuse_interval) {
+      snprintf(buf, sizeof(buf), "%" PRId64 " MB currently in use",
+               inuse_bytes >> 20);
+      need_to_dump = true;
+    } else if (FLAGS_heap_profile_time_interval > 0 &&
+               current_time - last_dump_time >=
+               FLAGS_heap_profile_time_interval) {
+      snprintf(buf, sizeof(buf), "%" PRId64 " sec since the last dump",
+               current_time - last_dump_time);
+      need_to_dump = true;
+      last_dump_time = current_time;
+    }
+    if (need_to_dump) {
+      DumpProfileLocked(buf);
+
+      last_dump_alloc = total.alloc_size;
+      last_dump_free = total.free_size;
+      if (inuse_bytes > high_water_mark)
+        high_water_mark = inuse_bytes;
+    }
+  }
+}
+
+// Record an allocation in the profile.
+static void RecordAlloc(const void* ptr, size_t bytes, int skip_count) {
+  // Take the stack trace outside the critical section.
+  void* stack[HeapProfileTable::kMaxStackDepth];
+  int depth = HeapProfileTable::GetCallerStackTrace(skip_count + 1, stack);
+  SpinLockHolder l(&heap_lock);
+  if (is_on) {
+    heap_profile->RecordAlloc(ptr, bytes, depth, stack);
+    MaybeDumpProfileLocked();
+  }
+}
+
+// Record a deallocation in the profile.
+static void RecordFree(const void* ptr) {
+  SpinLockHolder l(&heap_lock);
+  if (is_on) {
+    heap_profile->RecordFree(ptr);
+    MaybeDumpProfileLocked();
+  }
+}
+
+//----------------------------------------------------------------------
+// Allocation/deallocation hooks for MallocHook
+//----------------------------------------------------------------------
+
+// static
+void NewHook(const void* ptr, size_t size) {
+  if (ptr != NULL) RecordAlloc(ptr, size, 0);
+}
+
+// static
+void DeleteHook(const void* ptr) {
+  if (ptr != NULL) RecordFree(ptr);
+}
+
+// TODO(jandrews): Re-enable stack tracing
+#ifdef TODO_REENABLE_STACK_TRACING
+static void RawInfoStackDumper(const char* message, void*) {
+  RAW_LOG(INFO, "%.*s", static_cast<int>(strlen(message) - 1), message);
+  // -1 is to chop the \n which will be added by RAW_LOG
+}
+#endif
+
+static void MmapHook(const void* result, const void* start, size_t size,
+                     int prot, int flags, int fd, off_t offset) {
+  if (FLAGS_mmap_log) {  // log it
+    // We use PRIxS not just '%p' to avoid deadlocks
+    // in pretty-printing of NULL as "nil".
+    // TODO(maxim): instead should use a safe snprintf reimplementation
+    RAW_LOG(INFO,
+            "mmap(start=0x%" PRIxPTR ", len=%" PRIuS ", prot=0x%x, flags=0x%x, "
+            "fd=%d, offset=0x%x) = 0x%" PRIxPTR "",
+            (uintptr_t) start, size, prot, flags, fd, (unsigned int) offset,
+            (uintptr_t) result);
+#ifdef TODO_REENABLE_STACK_TRACING
+    DumpStackTrace(1, RawInfoStackDumper, NULL);
+#endif
+  }
+}
+
+static void MremapHook(const void* result, const void* old_addr,
+                       size_t old_size, size_t new_size,
+                       int flags, const void* new_addr) {
+  if (FLAGS_mmap_log) {  // log it
+    // We use PRIxS not just '%p' to avoid deadlocks
+    // in pretty-printing of NULL as "nil".
+    // TODO(maxim): instead should use a safe snprintf reimplementation
+    RAW_LOG(INFO,
+            "mremap(old_addr=0x%" PRIxPTR ", old_size=%" PRIuS ", "
+            "new_size=%" PRIuS ", flags=0x%x, new_addr=0x%" PRIxPTR ") = "
+            "0x%" PRIxPTR "",
+            (uintptr_t) old_addr, old_size, new_size, flags,
+            (uintptr_t) new_addr, (uintptr_t) result);
+#ifdef TODO_REENABLE_STACK_TRACING
+    DumpStackTrace(1, RawInfoStackDumper, NULL);
+#endif
+  }
+}
+
+static void MunmapHook(const void* ptr, size_t size) {
+  if (FLAGS_mmap_log) {  // log it
+    // We use PRIxS not just '%p' to avoid deadlocks
+    // in pretty-printing of NULL as "nil".
+    // TODO(maxim): instead should use a safe snprintf reimplementation
+    RAW_LOG(INFO, "munmap(start=0x%" PRIxPTR ", len=%" PRIuS ")",
+                  (uintptr_t) ptr, size);
+#ifdef TODO_REENABLE_STACK_TRACING
+    DumpStackTrace(1, RawInfoStackDumper, NULL);
+#endif
+  }
+}
+
+static void SbrkHook(const void* result, ptrdiff_t increment) {
+  if (FLAGS_mmap_log) {  // log it
+    RAW_LOG(INFO, "sbrk(inc=%" PRIdS ") = 0x%" PRIxPTR "",
+                  increment, (uintptr_t) result);
+#ifdef TODO_REENABLE_STACK_TRACING
+    DumpStackTrace(1, RawInfoStackDumper, NULL);
+#endif
+  }
+}
+
+//----------------------------------------------------------------------
+// Starting/stopping/dumping
+//----------------------------------------------------------------------
+
+extern "C" void HeapProfilerStart(const char* prefix) {
+  SpinLockHolder l(&heap_lock);
+
+  if (is_on) return;
+
+  is_on = true;
+
+  RAW_VLOG(0, "Starting tracking the heap");
+
+  // This should be done before the hooks are set up, since it should
+  // call new, and we want that to be accounted for correctly.
+  MallocExtension::Initialize();
+
+  if (FLAGS_only_mmap_profile) {
+    FLAGS_mmap_profile = true;
+  }
+
+  if (FLAGS_mmap_profile) {
+    // Ask MemoryRegionMap to record all mmap, mremap, and sbrk
+    // call stack traces of at least size kMaxStackDepth:
+    MemoryRegionMap::Init(HeapProfileTable::kMaxStackDepth);
+  }
+
+  if (FLAGS_mmap_log) {
+    // Install our hooks to do the logging:
+    RAW_CHECK(MallocHook::AddMmapHook(&MmapHook), "");
+    RAW_CHECK(MallocHook::AddMremapHook(&MremapHook), "");
+    RAW_CHECK(MallocHook::AddMunmapHook(&MunmapHook), "");
+    RAW_CHECK(MallocHook::AddSbrkHook(&SbrkHook), "");
+  }
+
+  heap_profiler_memory =
+    LowLevelAlloc::NewArena(0, LowLevelAlloc::DefaultArena());
+
+  // Reserve space now for the heap profiler, so we can still write a
+  // heap profile even if the application runs out of memory.
+  global_profiler_buffer =
+      reinterpret_cast<char*>(ProfilerMalloc(kProfileBufferSize));
+
+  heap_profile = new(ProfilerMalloc(sizeof(HeapProfileTable)))
+                   HeapProfileTable(ProfilerMalloc, ProfilerFree);
+
+  last_dump_alloc = 0;
+  last_dump_free = 0;
+  high_water_mark = 0;
+  last_dump_time = 0;
+
+  // We do not reset dump_count so if the user does a sequence of
+  // HeapProfilerStart/HeapProfileStop, we will get a continuous
+  // sequence of profiles.
+
+  if (FLAGS_only_mmap_profile == false) {
+    // Now set the hooks that capture new/delete and malloc/free.
+    RAW_CHECK(MallocHook::AddNewHook(&NewHook), "");
+    RAW_CHECK(MallocHook::AddDeleteHook(&DeleteHook), "");
+  }
+
+  // Copy filename prefix
+  RAW_DCHECK(filename_prefix == NULL, "");
+  const int prefix_length = strlen(prefix);
+  filename_prefix = reinterpret_cast<char*>(ProfilerMalloc(prefix_length + 1));
+  memcpy(filename_prefix, prefix, prefix_length);
+  filename_prefix[prefix_length] = '\0';
+}
+
+extern "C" int IsHeapProfilerRunning() {
+  SpinLockHolder l(&heap_lock);
+  return is_on ? 1 : 0;   // return an int, because C code doesn't have bool
+}
+
+extern "C" void HeapProfilerStop() {
+  SpinLockHolder l(&heap_lock);
+
+  if (!is_on) return;
+
+  if (FLAGS_only_mmap_profile == false) {
+    // Unset our new/delete hooks, checking they were set:
+    RAW_CHECK(MallocHook::RemoveNewHook(&NewHook), "");
+    RAW_CHECK(MallocHook::RemoveDeleteHook(&DeleteHook), "");
+  }
+  if (FLAGS_mmap_log) {
+    // Restore mmap/sbrk hooks, checking that our hooks were set:
+    RAW_CHECK(MallocHook::RemoveMmapHook(&MmapHook), "");
+    RAW_CHECK(MallocHook::RemoveMremapHook(&MremapHook), "");
+    RAW_CHECK(MallocHook::RemoveSbrkHook(&SbrkHook), "");
+    RAW_CHECK(MallocHook::RemoveMunmapHook(&MunmapHook), "");
+  }
+
+  // free profile
+  heap_profile->~HeapProfileTable();
+  ProfilerFree(heap_profile);
+  heap_profile = NULL;
+
+  // free output-buffer memory
+  ProfilerFree(global_profiler_buffer);
+
+  // free prefix
+  ProfilerFree(filename_prefix);
+  filename_prefix = NULL;
+
+  if (!LowLevelAlloc::DeleteArena(heap_profiler_memory)) {
+    RAW_LOG(FATAL, "Memory leak in HeapProfiler:");
+  }
+
+  if (FLAGS_mmap_profile) {
+    MemoryRegionMap::Shutdown();
+  }
+
+  is_on = false;
+}
+
+extern "C" void HeapProfilerDump(const char *reason) {
+  SpinLockHolder l(&heap_lock);
+  if (is_on && !dumping) {
+    DumpProfileLocked(reason);
+  }
+}
+
+//----------------------------------------------------------------------
+// Initialization/finalization code
+//----------------------------------------------------------------------
+
+// Initialization code
+static void HeapProfilerInit() {
+  // Everything after this point is for setting up the profiler based on envvar
+  char fname[PATH_MAX];
+  if (!GetUniquePathFromEnv("HEAPPROFILE", fname)) {
+    return;
+  }
+  // We do a uid check so we don't write out files in a setuid executable.
+#ifdef HAVE_GETEUID
+  if (getuid() != geteuid()) {
+    RAW_LOG(WARNING, ("HeapProfiler: ignoring HEAPPROFILE because "
+                      "program seems to be setuid\n"));
+    return;
+  }
+#endif
+
+  HeapProfileTable::CleanupOldProfiles(fname);
+
+  HeapProfilerStart(fname);
+}
+
+// class used for finalization -- dumps the heap-profile at program exit
+struct HeapProfileEndWriter {
+  ~HeapProfileEndWriter() { HeapProfilerDump("Exiting"); }
+};
+
+// We want to make sure tcmalloc is up and running before starting the profiler
+static const TCMallocGuard tcmalloc_initializer;
+REGISTER_MODULE_INITIALIZER(heapprofiler, HeapProfilerInit());
+static HeapProfileEndWriter heap_profile_end_writer;

BeefRT/gperftools/src/internal_logging.cc

@@ -0,0 +1,193 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Sanjay Ghemawat <opensource@google.com>
+
+#include <config.h>
+#include "internal_logging.h"
+#include <stdarg.h>                     // for va_end, va_start
+#include <stdio.h>                      // for vsnprintf, va_list, etc
+#include <stdlib.h>                     // for abort
+#include <string.h>                     // for strlen, memcpy
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>    // for write()
+#endif
+
+#include <gperftools/malloc_extension.h>
+#include "base/logging.h"   // for perftools_vsnprintf
+#include "base/spinlock.h"              // for SpinLockHolder, SpinLock
+
+static const int kLogBufSize = 800;
+
+// Variables for storing crash output.  Allocated statically since we
+// may not be able to heap-allocate while crashing.
+static SpinLock crash_lock(base::LINKER_INITIALIZED);
+static bool crashed = false;
+static const int kStatsBufferSize = 16 << 10;
+static char stats_buffer[kStatsBufferSize] = { 0 };
+
+namespace TCMALLOC_NAMESPACE {
+
+static void WriteMessage(const char* msg, int length) {
+  _write(STDERR_FILENO, msg, length);
+}
+
+void (*log_message_writer)(const char* msg, int length) = WriteMessage;
+
+
+class Logger {
+ public:
+  bool Add(const LogItem& item);
+  bool AddStr(const char* str, int n);
+  bool AddNum(uint64_t num, int base);  // base must be 10 or 16.
+
+  static const int kBufSize = 200;
+  char* p_;
+  char* end_;
+  char buf_[kBufSize];
+};
+
+void Log(LogMode mode, const char* filename, int line,
+         LogItem a, LogItem b, LogItem c, LogItem d) {
+  Logger state;
+  state.p_ = state.buf_;
+  state.end_ = state.buf_ + sizeof(state.buf_);
+  state.AddStr(filename, strlen(filename))
+      && state.AddStr(":", 1)
+      && state.AddNum(line, 10)
+      && state.AddStr("]", 1)
+      && state.Add(a)
+      && state.Add(b)
+      && state.Add(c)
+      && state.Add(d);
+
+  // Teminate with newline
+  if (state.p_ >= state.end_) {
+    state.p_ = state.end_ - 1;
+  }
+  *state.p_ = '\n';
+  state.p_++;
+
+  int msglen = state.p_ - state.buf_;
+  if (mode == kLog) {
+    (*log_message_writer)(state.buf_, msglen);
+    return;
+  }
+
+  bool first_crash = false;
+  {
+    SpinLockHolder l(&crash_lock);
+    if (!crashed) {
+      crashed = true;
+      first_crash = true;
+    }
+  }
+
+  (*log_message_writer)(state.buf_, msglen);
+  if (first_crash && mode == kCrashWithStats) {
+    MallocExtension::instance()->GetStats(stats_buffer, kStatsBufferSize);
+    (*log_message_writer)(stats_buffer, strlen(stats_buffer));
+  }
+
+  abort();
+}
+
+bool Logger::Add(const LogItem& item) {
+  // Separate items with spaces
+  if (p_ < end_) {
+    *p_ = ' ';
+    p_++;
+  }
+
+  switch (item.tag_) {
+    case LogItem::kStr:
+      return AddStr(item.u_.str, strlen(item.u_.str));
+    case LogItem::kUnsigned:
+      return AddNum(item.u_.unum, 10);
+    case LogItem::kSigned:
+      if (item.u_.snum < 0) {
+        // The cast to uint64_t is intentionally before the negation
+        // so that we do not attempt to negate -2^63.
+        return AddStr("-", 1)
+            && AddNum(- static_cast<uint64_t>(item.u_.snum), 10);
+      } else {
+        return AddNum(static_cast<uint64_t>(item.u_.snum), 10);
+      }
+    case LogItem::kPtr:
+      return AddStr("0x", 2)
+          && AddNum(reinterpret_cast<uintptr_t>(item.u_.ptr), 16);
+    default:
+      return false;
+  }
+}
+
+bool Logger::AddStr(const char* str, int n) {
+  if (end_ - p_ < n) {
+    return false;
+  } else {
+    memcpy(p_, str, n);
+    p_ += n;
+    return true;
+  }
+}
+
+bool Logger::AddNum(uint64_t num, int base) {
+  static const char kDigits[] = "0123456789abcdef";
+  char space[22];  // more than enough for 2^64 in smallest supported base (10)
+  char* end = space + sizeof(space);
+  char* pos = end;
+  do {
+    pos--;
+    *pos = kDigits[num % base];
+    num /= base;
+  } while (num > 0 && pos > space);
+  return AddStr(pos, end - pos);
+}
+
+}  // end tcmalloc namespace
+
+void TCMALLOC_NAMESPACE::TCMalloc_Printer::printf(const char* format, ...) {
+  if (left_ > 0) {
+    va_list ap;
+    va_start(ap, format);
+    const int r = perftools_vsnprintf(buf_, left_, format, ap);
+    va_end(ap);
+    if (r < 0) {
+      // Perhaps an old glibc that returns -1 on truncation?
+      left_ = 0;
+    } else if (r > left_) {
+      // Truncation
+      left_ = 0;
+    } else {
+      left_ -= r;
+      buf_ += r;
+    }
+  }
+}

BeefRT/gperftools/src/internal_logging.h

@@ -0,0 +1,146 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+//
+// Internal logging and related utility routines.
+
+#ifndef TCMALLOC_INTERNAL_LOGGING_H_
+#define TCMALLOC_INTERNAL_LOGGING_H_
+
+#include "tcconfig.h"
+#include <stddef.h>                     // for size_t
+#if defined HAVE_STDINT_H
+#include <stdint.h>
+#elif defined HAVE_INTTYPES_H
+#include <inttypes.h>
+#else
+#include <sys/types.h>
+#endif
+
+//-------------------------------------------------------------------
+// Utility routines
+//-------------------------------------------------------------------
+
+// Safe logging helper: we write directly to the stderr file
+// descriptor and avoid FILE buffering because that may invoke
+// malloc().
+//
+// Example:
+//   Log(kLog, __FILE__, __LINE__, "error", bytes);
+
+namespace TCMALLOC_NAMESPACE {
+enum LogMode {
+  kLog,                       // Just print the message
+  kCrash,                     // Print the message and crash
+  kCrashWithStats             // Print the message, some stats, and crash
+};
+
+class Logger;
+
+// A LogItem holds any of the argument types that can be passed to Log()
+class LogItem {
+ public:
+  LogItem()                     : tag_(kEnd)      { }
+  LogItem(const char* v)        : tag_(kStr)      { u_.str = v; }
+  LogItem(int v)                : tag_(kSigned)   { u_.snum = v; }
+  LogItem(long v)               : tag_(kSigned)   { u_.snum = v; }
+  LogItem(long long v)          : tag_(kSigned)   { u_.snum = v; }
+  LogItem(unsigned int v)       : tag_(kUnsigned) { u_.unum = v; }
+  LogItem(unsigned long v)      : tag_(kUnsigned) { u_.unum = v; }
+  LogItem(unsigned long long v) : tag_(kUnsigned) { u_.unum = v; }
+  LogItem(const void* v)        : tag_(kPtr)      { u_.ptr = v; }
+ private:
+  friend class Logger;
+  enum Tag {
+    kStr,
+    kSigned,
+    kUnsigned,
+    kPtr,
+    kEnd
+  };
+  Tag tag_;
+  union {
+    const char* str;
+    const void* ptr;
+    int64_t snum;
+    uint64_t unum;
+  } u_;
+};
+
+extern PERFTOOLS_DLL_DECL void Log(LogMode mode, const char* filename, int line,
+                LogItem a, LogItem b = LogItem(),
+                LogItem c = LogItem(), LogItem d = LogItem());
+
+// Tests can override this function to collect logging messages.
+extern PERFTOOLS_DLL_DECL void (*log_message_writer)(const char* msg, int length);
+
+}  // end tcmalloc namespace
+
+// Like assert(), but executed even in NDEBUG mode
+#undef CHECK_CONDITION
+#define CHECK_CONDITION(cond)                                            \
+do {                                                                     \
+  if (!(cond)) {                                                         \
+    ::TCMALLOC_NAMESPACE::Log(::TCMALLOC_NAMESPACE::kCrash, __FILE__, __LINE__, #cond);      \
+  }                                                                      \
+} while (0)
+
+// Our own version of assert() so we can avoid hanging by trying to do
+// all kinds of goofy printing while holding the malloc lock.
+#ifndef NDEBUG
+#define ASSERT(cond) CHECK_CONDITION(cond)
+#else
+#define ASSERT(cond) ((void) 0)
+#endif
+
+namespace TCMALLOC_NAMESPACE
+{
+	// Print into buffer
+	class TCMalloc_Printer {
+	private:
+		char* buf_;           // Where should we write next
+		int   left_;          // Space left in buffer (including space for \0)
+
+	public:
+		// REQUIRES: "length > 0"
+		TCMalloc_Printer(char* buf, int length) : buf_(buf), left_(length) {
+			buf[0] = '\0';
+		}
+
+		void printf(const char* format, ...)
+#ifdef HAVE___ATTRIBUTE__
+			__attribute__((__format__(__printf__, 2, 3)))
+#endif
+			;
+	};
+}
+
+#endif  // TCMALLOC_INTERNAL_LOGGING_H_

BeefRT/gperftools/src/ios/config.h

@@ -0,0 +1,291 @@
+/* src/config.h.  Generated from config.h.in by configure.  */
+/* src/config.h.in.  Generated from configure.ac by autoheader.  */
+
+#define NO_HEAP_CHECK
+
+
+#ifndef GPERFTOOLS_CONFIG_H_
+#define GPERFTOOLS_CONFIG_H_
+
+
+/* Define to 1 if compiler supports __builtin_expect */
+#define HAVE_BUILTIN_EXPECT 1
+
+/* Define to 1 if compiler supports __builtin_stack_pointer */
+/* #undef HAVE_BUILTIN_STACK_POINTER */
+
+/* Define to 1 if you have the <conflict-signal.h> header file. */
+/* #undef HAVE_CONFLICT_SIGNAL_H */
+
+/* Define to 1 if you have the <cygwin/signal.h> header file. */
+/* #undef HAVE_CYGWIN_SIGNAL_H */
+
+/* Define to 1 if you have the declaration of `cfree', and to 0 if you don't.
+   */
+#define HAVE_DECL_CFREE 0
+
+/* Define to 1 if you have the declaration of `memalign', and to 0 if you
+   don't. */
+#define HAVE_DECL_MEMALIGN 0
+
+/* Define to 1 if you have the declaration of `nanosleep', and to 0 if you
+   don't. */
+/* #undef HAVE_DECL_NANOSLEEP */
+
+/* Define to 1 if you have the declaration of `posix_memalign', and to 0 if
+   you don't. */
+#define HAVE_DECL_POSIX_MEMALIGN 0
+
+/* Define to 1 if you have the declaration of `pvalloc', and to 0 if you
+   don't. */
+#define HAVE_DECL_PVALLOC 0
+
+/* Define to 1 if you have the declaration of `sleep', and to 0 if you don't.
+   */
+/* #undef HAVE_DECL_SLEEP */
+
+/* Define to 1 if you have the declaration of `uname', and to 0 if you don't.
+   */
+#define HAVE_DECL_UNAME 1
+
+/* Define to 1 if you have the declaration of `valloc', and to 0 if you don't.
+   */
+#define HAVE_DECL_VALLOC 0
+
+/* Define to 1 if you have the <dlfcn.h> header file. */
+#define HAVE_DLFCN_H 1
+
+/* Define to 1 if the system has the type `Elf32_Versym'. */
+/* #undef HAVE_ELF32_VERSYM */
+
+/* Define to 1 if you have the <execinfo.h> header file. */
+#define HAVE_EXECINFO_H 1
+
+/* Define to 1 if you have the <fcntl.h> header file. */
+#define HAVE_FCNTL_H 1
+
+/* Define to 1 if you have the <features.h> header file. */
+/* #undef HAVE_FEATURES_H */
+
+/* Define to 1 if you have the `fork' function. */
+#define HAVE_FORK 1
+
+/* Define to 1 if you have the `geteuid' function. */
+#define HAVE_GETEUID 1
+
+/* Define to 1 if you have the `getpagesize' function. */
+#define HAVE_GETPAGESIZE 1
+
+/* Define to 1 if you have the <glob.h> header file. */
+#define HAVE_GLOB_H 1
+
+/* Define to 1 if you have the <grp.h> header file. */
+#define HAVE_GRP_H 1
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#define HAVE_INTTYPES_H 1
+
+/* Define to 1 if you have the <libunwind.h> header file. */
+#define HAVE_LIBUNWIND_H 1
+
+/* Define to 1 if you have the <linux/ptrace.h> header file. */
+/* #undef HAVE_LINUX_PTRACE_H */
+
+/* Define to 1 if you have the <malloc.h> header file. */
+/* #undef HAVE_MALLOC_H */
+
+/* Define to 1 if you have the <malloc/malloc.h> header file. */
+#define HAVE_MALLOC_MALLOC_H 1
+
+/* Define to 1 if you have the <memory.h> header file. */
+#define HAVE_MEMORY_H 1
+
+/* Define to 1 if you have a working `mmap' system call. */
+#define HAVE_MMAP 1
+
+/* define if the compiler implements namespaces */
+#define HAVE_NAMESPACES 1
+
+/* Define to 1 if you have the <poll.h> header file. */
+#define HAVE_POLL_H 1
+
+/* define if libc has program_invocation_name */
+/* #undef HAVE_PROGRAM_INVOCATION_NAME */
+
+/* Define if you have POSIX threads libraries and header files. */
+#define HAVE_PTHREAD 1
+
+/* Define to 1 if you have the <pwd.h> header file. */
+#define HAVE_PWD_H 1
+
+/* Define to 1 if you have the `sbrk' function. */
+#define HAVE_SBRK 1
+
+/* Define to 1 if you have the <sched.h> header file. */
+#define HAVE_SCHED_H 1
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#define HAVE_STDINT_H 1
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#define HAVE_STDLIB_H 1
+
+/* Define to 1 if you have the <strings.h> header file. */
+#define HAVE_STRINGS_H 1
+
+/* Define to 1 if you have the <string.h> header file. */
+#define HAVE_STRING_H 1
+
+/* Define to 1 if the system has the type `struct mallinfo'. */
+/* #undef HAVE_STRUCT_MALLINFO */
+
+/* Define to 1 if you have the <sys/cdefs.h> header file. */
+#define HAVE_SYS_CDEFS_H 1
+
+/* Define to 1 if you have the <sys/malloc.h> header file. */
+#define HAVE_SYS_MALLOC_H 1
+
+/* Define to 1 if you have the <sys/param.h> header file. */
+#define HAVE_SYS_PARAM_H 1
+
+/* Define to 1 if you have the <sys/prctl.h> header file. */
+/* #undef HAVE_SYS_PRCTL_H */
+
+/* Define to 1 if you have the <sys/resource.h> header file. */
+#define HAVE_SYS_RESOURCE_H 1
+
+/* Define to 1 if you have the <sys/socket.h> header file. */
+#define HAVE_SYS_SOCKET_H 1
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#define HAVE_SYS_STAT_H 1
+
+/* Define to 1 if you have the <sys/syscall.h> header file. */
+#define HAVE_SYS_SYSCALL_H 1
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#define HAVE_SYS_TYPES_H 1
+
+/* <sys/ucontext.h> is broken on redhat 7 */
+#define HAVE_SYS_UCONTEXT_H 1
+
+/* Define to 1 if you have the <sys/wait.h> header file. */
+#define HAVE_SYS_WAIT_H 1
+
+/* Define to 1 if compiler supports __thread */
+//#define HAVE_TLS 1
+
+/* Define to 1 if you have the <ucontext.h> header file. */
+/* #undef HAVE_UCONTEXT_H */
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#define HAVE_UNISTD_H 1
+
+/* Define to 1 if you have the <unwind.h> header file. */
+#define HAVE_UNWIND_H 1
+
+/* Define to 1 if you have the <valgrind.h> header file. */
+/* #undef HAVE_VALGRIND_H */
+
+/* define if your compiler has __attribute__ */
+#define HAVE___ATTRIBUTE__ 1
+
+/* Define to 1 if compiler supports __environ */
+/* #undef HAVE___ENVIRON */
+
+/* Define to 1 if the system has the type `__int64'. */
+/* #undef HAVE___INT64 */
+
+/* prefix where we look for installed files */
+#define INSTALL_PREFIX "/usr/local"
+
+/* Define to 1 if int32_t is equivalent to intptr_t */
+/* #undef INT32_EQUALS_INTPTR */
+
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#define LT_OBJDIR ".libs/"
+
+/* Define to 'volatile' if __malloc_hook is declared volatile */
+#define MALLOC_HOOK_MAYBE_VOLATILE /**/
+
+/* Define to 1 if your C compiler doesn't accept -c and -o together. */
+/* #undef NO_MINUS_C_MINUS_O */
+
+/* Name of package */
+#define PACKAGE "gperftools"
+
+/* Define to the address where bug reports for this package should be sent. */
+#define PACKAGE_BUGREPORT "google-perftools@googlegroups.com"
+
+/* Define to the full name of this package. */
+#define PACKAGE_NAME "gperftools"
+
+/* Define to the full name and version of this package. */
+#define PACKAGE_STRING "gperftools 2.1"
+
+/* Define to the one symbol short name of this package. */
+#define PACKAGE_TARNAME "gperftools"
+
+/* Define to the home page for this package. */
+#define PACKAGE_URL ""
+
+/* Define to the version of this package. */
+#define PACKAGE_VERSION "2.1"
+
+/* How to access the PC from a struct ucontext */
+#define PC_FROM_UCONTEXT uc_mcontext->__ss.__rip
+
+/* Always the empty-string on non-windows systems. On windows, should be
+   "__declspec(dllexport)". This way, when we compile the dll, we export our
+   functions/classes. It's safe to define this here because config.h is only
+   used internally, to compile the DLL, and every DLL source file #includes
+   "config.h" before anything else. */
+#define PERFTOOLS_DLL_DECL /**/
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIdS "ld"
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIuS "lu"
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIxS "lx"
+
+/* Mark the systems where we know it's bad if pthreads runs too
+   early before main (before threads are initialized, presumably).  */
+#ifdef __FreeBSD__
+#define PTHREADS_CRASHES_IF_RUN_TOO_EARLY 1
+#endif
+
+/* Define to necessary symbol if this constant uses a non-standard name on
+   your system. */
+/* #undef PTHREAD_CREATE_JOINABLE */
+
+/* Define to 1 if you have the ANSI C header files. */
+#define STDC_HEADERS 1
+
+/* the namespace where STL code like vector<> is defined */
+#define STL_NAMESPACE std
+
+/* Version number of package */
+#define VERSION "2.1"
+
+/* C99 says: define this to get the PRI... macros from stdint.h */
+#ifndef __STDC_FORMAT_MACROS
+# define __STDC_FORMAT_MACROS 1
+#endif
+
+/* Define to `__inline__' or `__inline' if that's what the C compiler
+   calls it, or to nothing if 'inline' is not supported under any name.  */
+#ifndef __cplusplus
+/* #undef inline */
+#endif
+
+
+#ifdef __MINGW32__
+#include "windows/mingw.h"
+#endif
+
+#endif  /* #ifndef GPERFTOOLS_CONFIG_H_ */
+

BeefRT/gperftools/src/ios/gperftools/tcmalloc.h

@@ -0,0 +1,133 @@
+/* Copyright (c) 2003, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat <opensource@google.com>
+ *         .h file by Craig Silverstein <opensource@google.com>
+ */
+
+#ifndef TCMALLOC_TCMALLOC_H_
+#define TCMALLOC_TCMALLOC_H_
+
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>   // where glibc defines __THROW
+#endif
+
+// __THROW is defined in glibc systems.  It means, counter-intuitively,
+// "This function will never throw an exception."  It's an optional
+// optimization tool, but we may need to use it to match glibc prototypes.
+#ifndef __THROW    /* I guess we're not on a glibc system */
+# define __THROW   /* __THROW is just an optimization, so ok to make it "" */
+#endif
+
+// Define the version number so folks can check against it
+#define TC_VERSION_MAJOR  2
+#define TC_VERSION_MINOR  1
+#define TC_VERSION_PATCH  ""
+#define TC_VERSION_STRING "gperftools 2.1"
+
+// For struct mallinfo, if it's defined.
+#ifdef HAVE_STRUCT_MALLINFO
+// Malloc can be in several places on older versions of OS X.
+# if defined(HAVE_MALLOC_H)
+# include <malloc.h>
+# elif defined(HAVE_SYS_MALLOC_H)
+# include <sys/malloc.h>
+# elif defined(HAVE_MALLOC_MALLOC_H)
+# include <malloc/malloc.h>
+# endif
+#endif
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+#ifdef __cplusplus
+namespace std {
+struct nothrow_t;
+}
+
+extern "C" {
+#endif
+  // Returns a human-readable version string.  If major, minor,
+  // and/or patch are not NULL, they are set to the major version,
+  // minor version, and patch-code (a string, usually "").
+  PERFTOOLS_DLL_DECL const char* tc_version(int* major, int* minor,
+                                            const char** patch) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_free(void* ptr) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_calloc(size_t nmemb, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_memalign(size_t __alignment,
+                                       size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL int tc_posix_memalign(void** ptr,
+                                           size_t align, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_valloc(size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t __size) __THROW;
+
+  PERFTOOLS_DLL_DECL void tc_malloc_stats(void) __THROW;
+  PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) __THROW;
+#if 0
+  PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) __THROW;
+#endif
+
+  // This is an alias for MallocExtension::instance()->GetAllocatedSize().
+  // It is equivalent to
+  //    OS X: malloc_size()
+  //    glibc: malloc_usable_size()
+  //    Windows: _msize()
+  PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) __THROW;
+
+#ifdef __cplusplus
+  PERFTOOLS_DLL_DECL int tc_set_new_mode(int flag) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_new(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size,
+                                          const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p,
+                                            const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_newarray(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size,
+                                               const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p,
+                                                 const std::nothrow_t&) __THROW;
+}
+#endif
+
+#endif  // #ifndef TCMALLOC_TCMALLOC_H_

BeefRT/gperftools/src/libc_override.h

@@ -0,0 +1,90 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein <opensource@google.com>
+//
+// This .h file imports the code that causes tcmalloc to override libc
+// versions of malloc/free/new/delete/etc.  That is, it provides the
+// logic that makes it so calls to malloc(10) go through tcmalloc,
+// rather than the default (libc) malloc.
+//
+// This file also provides a method: ReplaceSystemAlloc(), that every
+// libc_override_*.h file it #includes is required to provide.  This
+// is called when first setting up tcmalloc -- that is, when a global
+// constructor in tcmalloc.cc is executed -- to do any initialization
+// work that may be required for this OS.  (Note we cannot entirely
+// control when tcmalloc is initialized, and the system may do some
+// mallocs and frees before this routine is called.)  It may be a
+// noop.
+//
+// Every libc has its own way of doing this, and sometimes the compiler
+// matters too, so we have a different file for each libc, and often
+// for different compilers and OS's.
+
+#ifndef TCMALLOC_LIBC_OVERRIDE_INL_H_
+#define TCMALLOC_LIBC_OVERRIDE_INL_H_
+
+#include <config.h>
+#ifdef HAVE_FEATURES_H
+#include <features.h>   // for __GLIBC__
+#endif
+#include <gperftools/tcmalloc.h>
+
+static void ReplaceSystemAlloc();  // defined in the .h files below
+
+// For windows, there are two ways to get tcmalloc.  If we're
+// patching, then src/windows/patch_function.cc will do the necessary
+// overriding here.  Otherwise, we doing the 'redefine' trick, where
+// we remove malloc/new/etc from mscvcrt.dll, and just need to define
+// them now.
+#if defined(_WIN32) && defined(WIN32_DO_PATCHING)
+void PatchWindowsFunctions();   // in src/windows/patch_function.cc
+static void ReplaceSystemAlloc() { PatchWindowsFunctions(); }
+
+#elif defined(_WIN32) && !defined(WIN32_DO_PATCHING)
+#include "libc_override_redefine.h"
+
+#elif defined(__APPLE__)
+#include "libc_override_osx.h"
+
+#elif defined(__GLIBC__)
+#include "libc_override_glibc.h"
+
+// Not all gcc systems necessarily support weak symbols, but all the
+// ones I know of do, so for now just assume they all do.
+#elif defined(__GNUC__)
+#include "libc_override_gcc_and_weak.h"
+
+#else
+#error Need to add support for your libc/OS here
+
+#endif
+
+#endif  // TCMALLOC_LIBC_OVERRIDE_INL_H_

BeefRT/gperftools/src/libc_override_gcc_and_weak.h

@@ -0,0 +1,99 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein <opensource@google.com>
+//
+// Used to override malloc routines on systems that define the
+// memory allocation routines to be weak symbols in their libc
+// (almost all unix-based systems are like this), on gcc, which
+// suppports the 'alias' attribute.
+
+#ifndef TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_
+#define TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_
+
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>    // for __THROW
+#endif
+#include <gperftools/tcmalloc.h>
+
+#ifndef __THROW    // I guess we're not on a glibc-like system
+# define __THROW   // __THROW is just an optimization, so ok to make it ""
+#endif
+
+#ifndef __GNUC__
+# error libc_override_gcc_and_weak.h is for gcc distributions only.
+#endif
+
+#define ALIAS(tc_fn)   __attribute__ ((alias (#tc_fn)))
+
+void* operator new(size_t size) throw (std::bad_alloc)
+    ALIAS(tc_new);
+void operator delete(void* p) __THROW
+    ALIAS(tc_delete);
+void* operator new[](size_t size) throw (std::bad_alloc)
+    ALIAS(tc_newarray);
+void operator delete[](void* p) __THROW
+    ALIAS(tc_deletearray);
+void* operator new(size_t size, const std::nothrow_t& nt) __THROW
+    ALIAS(tc_new_nothrow);
+void* operator new[](size_t size, const std::nothrow_t& nt) __THROW
+    ALIAS(tc_newarray_nothrow);
+void operator delete(void* p, const std::nothrow_t& nt) __THROW
+    ALIAS(tc_delete_nothrow);
+void operator delete[](void* p, const std::nothrow_t& nt) __THROW
+    ALIAS(tc_deletearray_nothrow);
+
+extern "C" {
+  void* malloc(size_t size) __THROW               ALIAS(tc_malloc);
+  void free(void* ptr) __THROW                    ALIAS(tc_free);
+  void* realloc(void* ptr, size_t size) __THROW   ALIAS(tc_realloc);
+  void* calloc(size_t n, size_t size) __THROW     ALIAS(tc_calloc);
+  void cfree(void* ptr) __THROW                   ALIAS(tc_cfree);
+  void* memalign(size_t align, size_t s) __THROW  ALIAS(tc_memalign);
+  void* valloc(size_t size) __THROW               ALIAS(tc_valloc);
+  void* pvalloc(size_t size) __THROW              ALIAS(tc_pvalloc);
+  int posix_memalign(void** r, size_t a, size_t s) __THROW
+      ALIAS(tc_posix_memalign);
+  void malloc_stats(void) __THROW                 ALIAS(tc_malloc_stats);
+  int mallopt(int cmd, int value) __THROW         ALIAS(tc_mallopt);
+#ifdef HAVE_STRUCT_MALLINFO
+  struct mallinfo mallinfo(void) __THROW          ALIAS(tc_mallinfo);
+#endif
+  size_t malloc_size(void* p) __THROW             ALIAS(tc_malloc_size);
+  size_t malloc_usable_size(void* p) __THROW      ALIAS(tc_malloc_size);
+}   // extern "C"
+
+#undef ALIAS
+
+// No need to do anything at tcmalloc-registration time: we do it all
+// via overriding weak symbols (at link time).
+static void ReplaceSystemAlloc() { }
+
+#endif  // TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_

BeefRT/gperftools/src/libc_override_glibc.h

@@ -0,0 +1,148 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein <opensource@google.com>
+//
+// Used to override malloc routines on systems that are using glibc.
+
+#ifndef TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_
+#define TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_
+
+#include <config.h>
+#include <features.h>     // for __GLIBC__
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>    // for __THROW
+#endif
+#include <gperftools/tcmalloc.h>
+
+#ifndef __GLIBC__
+# error libc_override_glibc.h is for glibc distributions only.
+#endif
+
+// In glibc, the memory-allocation methods are weak symbols, so we can
+// just override them with our own.  If we're using gcc, we can use
+// __attribute__((alias)) to do the overriding easily (exception:
+// Mach-O, which doesn't support aliases).  Otherwise we have to use a
+// function call.
+#if !defined(__GNUC__) || defined(__MACH__)
+
+// This also defines ReplaceSystemAlloc().
+# include "libc_override_redefine.h"  // defines functions malloc()/etc
+
+#else  // #if !defined(__GNUC__) || defined(__MACH__)
+
+// If we get here, we're a gcc system, so do all the overriding we do
+// with gcc.  This does the overriding of all the 'normal' memory
+// allocation.  This also defines ReplaceSystemAlloc().
+# include "libc_override_gcc_and_weak.h"
+
+// We also have to do some glibc-specific overriding.  Some library
+// routines on RedHat 9 allocate memory using malloc() and free it
+// using __libc_free() (or vice-versa).  Since we provide our own
+// implementations of malloc/free, we need to make sure that the
+// __libc_XXX variants (defined as part of glibc) also point to the
+// same implementations.  Since it only matters for redhat, we
+// do it inside the gcc #ifdef, since redhat uses gcc.
+// TODO(csilvers): only do this if we detect we're an old enough glibc?
+
+#define ALIAS(tc_fn)   __attribute__ ((alias (#tc_fn)))
+extern "C" {
+  void* __libc_malloc(size_t size)                ALIAS(tc_malloc);
+  void __libc_free(void* ptr)                     ALIAS(tc_free);
+  void* __libc_realloc(void* ptr, size_t size)    ALIAS(tc_realloc);
+  void* __libc_calloc(size_t n, size_t size)      ALIAS(tc_calloc);
+  void __libc_cfree(void* ptr)                    ALIAS(tc_cfree);
+  void* __libc_memalign(size_t align, size_t s)   ALIAS(tc_memalign);
+  void* __libc_valloc(size_t size)                ALIAS(tc_valloc);
+  void* __libc_pvalloc(size_t size)               ALIAS(tc_pvalloc);
+  int __posix_memalign(void** r, size_t a, size_t s)  ALIAS(tc_posix_memalign);
+}   // extern "C"
+#undef ALIAS
+
+#endif  // #if defined(__GNUC__) && !defined(__MACH__)
+
+
+// We also have to hook libc malloc.  While our work with weak symbols
+// should make sure libc malloc is never called in most situations, it
+// can be worked around by shared libraries with the DEEPBIND
+// environment variable set.  The below hooks libc to call our malloc
+// routines even in that situation.  In other situations, this hook
+// should never be called.
+extern "C" {
+static void* glibc_override_malloc(size_t size, const void *caller) {
+  return tc_malloc(size);
+}
+static void* glibc_override_realloc(void *ptr, size_t size,
+                                    const void *caller) {
+  return tc_realloc(ptr, size);
+}
+static void glibc_override_free(void *ptr, const void *caller) {
+  tc_free(ptr);
+}
+static void* glibc_override_memalign(size_t align, size_t size,
+                                     const void *caller) {
+  return tc_memalign(align, size);
+}
+
+// We should be using __malloc_initialize_hook here, like the #if 0
+// code below.  (See http://swoolley.org/man.cgi/3/malloc_hook.)
+// However, this causes weird linker errors with programs that link
+// with -static, so instead we just assign the vars directly at
+// static-constructor time.  That should serve the same effect of
+// making sure the hooks are set before the first malloc call the
+// program makes.
+#if 0
+#include <malloc.h>  // for __malloc_hook, etc.
+void glibc_override_malloc_init_hook(void) {
+  __malloc_hook = glibc_override_malloc;
+  __realloc_hook = glibc_override_realloc;
+  __free_hook = glibc_override_free;
+  __memalign_hook = glibc_override_memalign;
+}
+
+void (* MALLOC_HOOK_MAYBE_VOLATILE __malloc_initialize_hook)(void)
+    = &glibc_override_malloc_init_hook;
+#endif
+
+void* (* MALLOC_HOOK_MAYBE_VOLATILE __malloc_hook)(size_t, const void*)
+    = &glibc_override_malloc;
+void* (* MALLOC_HOOK_MAYBE_VOLATILE __realloc_hook)(void*, size_t, const void*)
+    = &glibc_override_realloc;
+void (* MALLOC_HOOK_MAYBE_VOLATILE __free_hook)(void*, const void*)
+    = &glibc_override_free;
+void* (* MALLOC_HOOK_MAYBE_VOLATILE __memalign_hook)(size_t,size_t, const void*)
+    = &glibc_override_memalign;
+
+}   // extern "C"
+
+// No need to write ReplaceSystemAlloc(); one of the #includes above
+// did it for us.
+
+#endif  // TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_

BeefRT/gperftools/src/libc_override_osx.h

@@ -0,0 +1,275 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein <opensource@google.com>
+//
+// Used to override malloc routines on OS X systems.  We use the
+// malloc-zone functionality built into OS X to register our malloc
+// routine.
+//
+// 1) We used to use the normal 'override weak libc malloc/etc'
+// technique for OS X.  This is not optimal because mach does not
+// support the 'alias' attribute, so we had to have forwarding
+// functions.  It also does not work very well with OS X shared
+// libraries (dylibs) -- in general, the shared libs don't use
+// tcmalloc unless run with the DYLD_FORCE_FLAT_NAMESPACE envvar.
+//
+// 2) Another approach would be to use an interposition array:
+//      static const interpose_t interposers[] __attribute__((section("__DATA, __interpose"))) = {
+//        { (void *)tc_malloc, (void *)malloc },
+//        { (void *)tc_free, (void *)free },
+//      };
+// This requires the user to set the DYLD_INSERT_LIBRARIES envvar, so
+// is not much better.
+//
+// 3) Registering a new malloc zone avoids all these issues:
+//  http://www.opensource.apple.com/source/Libc/Libc-583/include/malloc/malloc.h
+//  http://www.opensource.apple.com/source/Libc/Libc-583/gen/malloc.c
+// If we make tcmalloc the default malloc zone (undocumented but
+// possible) then all new allocs use it, even those in shared
+// libraries.  Allocs done before tcmalloc was installed, or in libs
+// that aren't using tcmalloc for some reason, will correctly go
+// through the malloc-zone interface when free-ing, and will pick up
+// the libc free rather than tcmalloc free.  So it should "never"
+// cause a crash (famous last words).
+//
+// 4) The routines one must define for one's own malloc have changed
+// between OS X versions.  This requires some hoops on our part, but
+// is only really annoying when it comes to posix_memalign.  The right
+// behavior there depends on what OS version tcmalloc was compiled on,
+// but also what OS version the program is running on.  For now, we
+// punt and don't implement our own posix_memalign.  Apps that really
+// care can use tc_posix_memalign directly.
+
+#ifndef TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_
+#define TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_
+
+#include <config.h>
+#ifdef HAVE_FEATURES_H
+#include <features.h>
+#endif
+#include <gperftools/tcmalloc.h>
+
+#if !defined(__APPLE__)
+# error libc_override_glibc-osx.h is for OS X distributions only.
+#endif
+
+#include <AvailabilityMacros.h>
+#include <malloc/malloc.h>
+
+// from AvailabilityMacros.h
+#if defined(MAC_OS_X_VERSION_10_6) && \
+    MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6
+extern "C" {
+  // This function is only available on 10.6 (and later) but the
+  // LibSystem headers do not use AvailabilityMacros.h to handle weak
+  // importing automatically.  This prototype is a copy of the one in
+  // <malloc/malloc.h> with the WEAK_IMPORT_ATTRBIUTE added.
+  extern malloc_zone_t *malloc_default_purgeable_zone(void)
+      WEAK_IMPORT_ATTRIBUTE;
+}
+#endif
+
+// We need to provide wrappers around all the libc functions.
+namespace {
+size_t mz_size(malloc_zone_t* zone, const void* ptr) {
+  if (MallocExtension::instance()->GetOwnership(ptr) != MallocExtension::kOwned)
+    return 0;  // malloc_zone semantics: return 0 if we don't own the memory
+
+  // TODO(csilvers): change this method to take a const void*, one day.
+  return MallocExtension::instance()->GetAllocatedSize(const_cast<void*>(ptr));
+}
+
+void* mz_malloc(malloc_zone_t* zone, size_t size) {
+  return tc_malloc(size);
+}
+
+void* mz_calloc(malloc_zone_t* zone, size_t num_items, size_t size) {
+  return tc_calloc(num_items, size);
+}
+
+void* mz_valloc(malloc_zone_t* zone, size_t size) {
+  return tc_valloc(size);
+}
+
+void mz_free(malloc_zone_t* zone, void* ptr) {
+  return tc_free(ptr);
+}
+
+void* mz_realloc(malloc_zone_t* zone, void* ptr, size_t size) {
+  return tc_realloc(ptr, size);
+}
+
+void* mz_memalign(malloc_zone_t* zone, size_t align, size_t size) {
+  return tc_memalign(align, size);
+}
+
+void mz_destroy(malloc_zone_t* zone) {
+  // A no-op -- we will not be destroyed!
+}
+
+// malloc_introspection callbacks.  I'm not clear on what all of these do.
+kern_return_t mi_enumerator(task_t task, void *,
+                            unsigned type_mask, vm_address_t zone_address,
+                            memory_reader_t reader,
+                            vm_range_recorder_t recorder) {
+  // Should enumerate all the pointers we have.  Seems like a lot of work.
+  return KERN_FAILURE;
+}
+
+size_t mi_good_size(malloc_zone_t *zone, size_t size) {
+  // I think it's always safe to return size, but we maybe could do better.
+  return size;
+}
+
+boolean_t mi_check(malloc_zone_t *zone) {
+  return MallocExtension::instance()->VerifyAllMemory();
+}
+
+void mi_print(malloc_zone_t *zone, boolean_t verbose) {
+  int bufsize = 8192;
+  if (verbose)
+    bufsize = 102400;   // I picked this size arbitrarily
+  char* buffer = new char[bufsize];
+  MallocExtension::instance()->GetStats(buffer, bufsize);
+  fprintf(stdout, "%s", buffer);
+  delete[] buffer;
+}
+
+void mi_log(malloc_zone_t *zone, void *address) {
+  // I don't think we support anything like this
+}
+
+void mi_force_lock(malloc_zone_t *zone) {
+  // Hopefully unneeded by us!
+}
+
+void mi_force_unlock(malloc_zone_t *zone) {
+  // Hopefully unneeded by us!
+}
+
+void mi_statistics(malloc_zone_t *zone, malloc_statistics_t *stats) {
+  // TODO(csilvers): figure out how to fill these out
+  stats->blocks_in_use = 0;
+  stats->size_in_use = 0;
+  stats->max_size_in_use = 0;
+  stats->size_allocated = 0;
+}
+
+boolean_t mi_zone_locked(malloc_zone_t *zone) {
+  return false;  // Hopefully unneeded by us!
+}
+
+}  // unnamed namespace
+
+// OS X doesn't have pvalloc, cfree, malloc_statc, etc, so we can just
+// define our own. :-)  OS X supplies posix_memalign in some versions
+// but not others, either strongly or weakly linked, in a way that's
+// difficult enough to code to correctly, that I just don't try to
+// support either memalign() or posix_memalign().  If you need them
+// and are willing to code to tcmalloc, you can use tc_posix_memalign().
+extern "C" {
+  void  cfree(void* p)                   { tc_cfree(p);               }
+  void* pvalloc(size_t s)                { return tc_pvalloc(s);      }
+  void malloc_stats(void)                { tc_malloc_stats();         }
+  int mallopt(int cmd, int v)            { return tc_mallopt(cmd, v); }
+  // No struct mallinfo on OS X, so don't define mallinfo().
+  // An alias for malloc_size(), which OS X defines.
+  size_t malloc_usable_size(void* p)     { return tc_malloc_size(p); }
+}  // extern "C"
+
+static void ReplaceSystemAlloc() {
+  static malloc_introspection_t tcmalloc_introspection;
+  memset(&tcmalloc_introspection, 0, sizeof(tcmalloc_introspection));
+
+  tcmalloc_introspection.enumerator = &mi_enumerator;
+  tcmalloc_introspection.good_size = &mi_good_size;
+  tcmalloc_introspection.check = &mi_check;
+  tcmalloc_introspection.print = &mi_print;
+  tcmalloc_introspection.log = &mi_log;
+  tcmalloc_introspection.force_lock = &mi_force_lock;
+  tcmalloc_introspection.force_unlock = &mi_force_unlock;
+
+  static malloc_zone_t tcmalloc_zone;
+  memset(&tcmalloc_zone, 0, sizeof(malloc_zone_t));
+
+  // Start with a version 4 zone which is used for OS X 10.4 and 10.5.
+  tcmalloc_zone.version = 4;
+  tcmalloc_zone.zone_name = "tcmalloc";
+  tcmalloc_zone.size = &mz_size;
+  tcmalloc_zone.malloc = &mz_malloc;
+  tcmalloc_zone.calloc = &mz_calloc;
+  tcmalloc_zone.valloc = &mz_valloc;
+  tcmalloc_zone.free = &mz_free;
+  tcmalloc_zone.realloc = &mz_realloc;
+  tcmalloc_zone.destroy = &mz_destroy;
+  tcmalloc_zone.batch_malloc = NULL;
+  tcmalloc_zone.batch_free = NULL;
+  tcmalloc_zone.introspect = &tcmalloc_introspection;
+
+  // from AvailabilityMacros.h
+#if defined(MAC_OS_X_VERSION_10_6) && \
+    MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6
+  // Switch to version 6 on OSX 10.6 to support memalign.
+  tcmalloc_zone.version = 6;
+  tcmalloc_zone.free_definite_size = NULL;
+  tcmalloc_zone.memalign = &mz_memalign;
+  tcmalloc_introspection.zone_locked = &mi_zone_locked;
+
+  // Request the default purgable zone to force its creation. The
+  // current default zone is registered with the purgable zone for
+  // doing tiny and small allocs.  Sadly, it assumes that the default
+  // zone is the szone implementation from OS X and will crash if it
+  // isn't.  By creating the zone now, this will be true and changing
+  // the default zone won't cause a problem.  This only needs to
+  // happen when actually running on OS X 10.6 and higher (note the
+  // ifdef above only checks if we were *compiled* with 10.6 or
+  // higher; at runtime we have to check if this symbol is defined.)
+  if (malloc_default_purgeable_zone) {
+    malloc_default_purgeable_zone();
+  }
+#endif
+
+  // Register the tcmalloc zone. At this point, it will not be the
+  // default zone.
+  malloc_zone_register(&tcmalloc_zone);
+
+  // Unregister and reregister the default zone.  Unregistering swaps
+  // the specified zone with the last one registered which for the
+  // default zone makes the more recently registered zone the default
+  // zone.  The default zone is then re-registered to ensure that
+  // allocations made from it earlier will be handled correctly.
+  // Things are not guaranteed to work that way, but it's how they work now.
+  malloc_zone_t *default_zone = malloc_default_zone();
+  malloc_zone_unregister(default_zone);
+  malloc_zone_register(default_zone);
+}
+
+#endif  // TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_

BeefRT/gperftools/src/libc_override_redefine.h

@@ -0,0 +1,95 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Craig Silverstein <opensource@google.com>
+//
+// Used on systems that don't have their own definition of
+// malloc/new/etc.  (Typically this will be a windows msvcrt.dll that
+// has been edited to remove the definitions.)  We can just define our
+// own as normal functions.
+//
+// This should also work on systems were all the malloc routines are
+// defined as weak symbols, and there's no support for aliasing.
+
+#ifndef TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_
+#define TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_
+
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>    // for __THROW
+#endif
+
+#ifndef __THROW    // I guess we're not on a glibc-like system
+# define __THROW   // __THROW is just an optimization, so ok to make it ""
+#endif
+
+FAIL
+
+void* operator new(size_t size)                  { return tc_new(size);       }
+void operator delete(void* p) __THROW            { tc_delete(p);              }
+void* operator new[](size_t size)                { return tc_newarray(size);  }
+void operator delete[](void* p) __THROW          { tc_deletearray(p);         }
+void* operator new(size_t size, const std::nothrow_t& nt) __THROW {
+  return tc_new_nothrow(size, nt);
+}
+void* operator new[](size_t size, const std::nothrow_t& nt) __THROW {
+  return tc_newarray_nothrow(size, nt);
+}
+void operator delete(void* ptr, const std::nothrow_t& nt) __THROW {
+  return tc_delete_nothrow(ptr, nt);
+}
+void operator delete[](void* ptr, const std::nothrow_t& nt) __THROW {
+  return tc_deletearray_nothrow(ptr, nt);
+}
+extern "C" {
+  void* malloc(size_t s) __THROW                 { return tc_malloc(s);       }
+  void  free(void* p) __THROW                    { tc_free(p);                }
+  void* realloc(void* p, size_t s) __THROW       { return tc_realloc(p, s);   }
+  void* calloc(size_t n, size_t s) __THROW       { return tc_calloc(n, s);    }
+  void  cfree(void* p) __THROW                   { tc_cfree(p);               }
+  void* memalign(size_t a, size_t s) __THROW     { return tc_memalign(a, s);  }
+  void* valloc(size_t s) __THROW                 { return tc_valloc(s);       }
+  void* pvalloc(size_t s) __THROW                { return tc_pvalloc(s);      }
+  int posix_memalign(void** r, size_t a, size_t s) __THROW {
+    return tc_posix_memalign(r, a, s);
+  }
+  void malloc_stats(void) __THROW                { tc_malloc_stats();         }
+  int mallopt(int cmd, int v) __THROW            { return tc_mallopt(cmd, v); }
+#ifdef HAVE_STRUCT_MALLINFO
+  struct mallinfo mallinfo(void) __THROW         { return tc_mallinfo();      }
+#endif
+  size_t malloc_size(void* p) __THROW            { return tc_malloc_size(p); }
+  size_t malloc_usable_size(void* p) __THROW     { return tc_malloc_size(p); }
+}  // extern "C"
+
+// No need to do anything at tcmalloc-registration time: we do it all
+// via overriding weak symbols (at link time).
+static void ReplaceSystemAlloc() { }
+
+#endif  // TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_

BeefRT/gperftools/src/linked_list.h

@@ -0,0 +1,102 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+//
+// Some very basic linked list functions for dealing with using void * as
+// storage.
+
+#ifndef TCMALLOC_LINKED_LIST_H_
+#define TCMALLOC_LINKED_LIST_H_
+
+#include <stddef.h>
+
+namespace TCMALLOC_NAMESPACE {
+
+inline void *SLL_Next(void *t) {
+  return *(reinterpret_cast<void**>(t));
+}
+
+inline void SLL_SetNext(void *t, void *n) {
+  *(reinterpret_cast<void**>(t)) = n;
+}
+
+inline void SLL_Push(void **list, void *element) {
+  SLL_SetNext(element, *list);
+  *list = element;
+}
+
+inline void *SLL_Pop(void **list) {
+  void *result = *list;
+  *list = SLL_Next(*list);
+  return result;
+}
+
+// Remove N elements from a linked list to which head points.  head will be
+// modified to point to the new head.  start and end will point to the first
+// and last nodes of the range.  Note that end will point to NULL after this
+// function is called.
+inline void SLL_PopRange(void **head, int N, void **start, void **end) {
+  if (N == 0) {
+    *start = NULL;
+    *end = NULL;
+    return;
+  }
+
+  void *tmp = *head;
+  for (int i = 1; i < N; ++i) {
+    tmp = SLL_Next(tmp);
+  }
+
+  *start = *head;
+  *end = tmp;
+  *head = SLL_Next(tmp);
+  // Unlink range from list.
+  SLL_SetNext(tmp, NULL);
+}
+
+inline void SLL_PushRange(void **head, void *start, void *end) {
+  if (!start) return;
+  SLL_SetNext(end, *head);
+  *head = start;
+}
+
+inline size_t SLL_Size(void *head) {
+  int count = 0;
+  while (head) {
+    count++;
+    head = SLL_Next(head);
+  }
+  return count;
+}
+
+}  // namespace tcmalloc
+
+#endif  // TCMALLOC_LINKED_LIST_H_

BeefRT/gperftools/src/malloc_extension.cc

@@ -0,0 +1,372 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+#include <config.h>
+#include <assert.h>
+#include <string.h>
+#include <stdio.h>
+#if defined HAVE_STDINT_H
+#include <stdint.h>
+#elif defined HAVE_INTTYPES_H
+#include <inttypes.h>
+#else
+#include <sys/types.h>
+#endif
+#include <string>
+#include "base/dynamic_annotations.h"
+#include "base/sysinfo.h"    // for FillProcSelfMaps
+#ifndef NO_HEAP_CHECK
+#include "gperftools/heap-checker.h"
+#endif
+#include "gperftools/malloc_extension.h"
+#include "gperftools/malloc_extension_c.h"
+#include "maybe_threads.h"
+
+using STL_NAMESPACE::string;
+using STL_NAMESPACE::vector;
+
+static void DumpAddressMap(string* result) {
+//   *result += "\nMAPPED_LIBRARIES:\n";
+//   // We keep doubling until we get a fit
+//   const size_t old_resultlen = result->size();
+//   for (int amap_size = 10240; amap_size < 10000000; amap_size *= 2) {
+//     result->resize(old_resultlen + amap_size);
+//     bool wrote_all = false;
+//     const int bytes_written =
+//         TCMALLOC_NAMESPACE::FillProcSelfMaps(&((*result)[old_resultlen]), amap_size,
+//                                    &wrote_all);
+//     if (wrote_all) {   // we fit!
+//       (*result)[old_resultlen + bytes_written] = '\0';
+//       result->resize(old_resultlen + bytes_written);
+//       return;
+//     }
+//   }
+//   result->reserve(old_resultlen);   // just don't print anything
+}
+
+// Note: this routine is meant to be called before threads are spawned.
+void MallocExtension::Initialize() {
+  static bool initialize_called = false;
+
+  if (initialize_called) return;
+  initialize_called = true;
+
+#ifdef __GLIBC__
+  // GNU libc++ versions 3.3 and 3.4 obey the environment variables
+  // GLIBCPP_FORCE_NEW and GLIBCXX_FORCE_NEW respectively.  Setting
+  // one of these variables forces the STL default allocator to call
+  // new() or delete() for each allocation or deletion.  Otherwise
+  // the STL allocator tries to avoid the high cost of doing
+  // allocations by pooling memory internally.  However, tcmalloc
+  // does allocations really fast, especially for the types of small
+  // items one sees in STL, so it's better off just using us.
+  // TODO: control whether we do this via an environment variable?
+  setenv("GLIBCPP_FORCE_NEW", "1", false /* no overwrite*/);
+  setenv("GLIBCXX_FORCE_NEW", "1", false /* no overwrite*/);
+
+  // Now we need to make the setenv 'stick', which it may not do since
+  // the env is flakey before main() is called.  But luckily stl only
+  // looks at this env var the first time it tries to do an alloc, and
+  // caches what it finds.  So we just cause an stl alloc here.
+  string dummy("I need to be allocated");
+  dummy += "!";         // so the definition of dummy isn't optimized out
+#endif  /* __GLIBC__ */
+}
+
+// SysAllocator implementation
+SysAllocator::~SysAllocator() {}
+
+// Default implementation -- does nothing
+MallocExtension::~MallocExtension() { }
+bool MallocExtension::VerifyAllMemory() { return true; }
+bool MallocExtension::VerifyNewMemory(const void* p) { return true; }
+bool MallocExtension::VerifyArrayNewMemory(const void* p) { return true; }
+bool MallocExtension::VerifyMallocMemory(const void* p) { return true; }
+
+bool MallocExtension::GetNumericProperty(const char* property, size_t* value) {
+  return false;
+}
+
+bool MallocExtension::SetNumericProperty(const char* property, size_t value) {
+  return false;
+}
+
+void MallocExtension::GetStats(char* buffer, int length) {
+  assert(length > 0);
+  buffer[0] = '\0';
+}
+
+bool MallocExtension::MallocMemoryStats(int* blocks, size_t* total,
+                                       int histogram[kMallocHistogramSize]) {
+  *blocks = 0;
+  *total = 0;
+  memset(histogram, 0, sizeof(*histogram) * kMallocHistogramSize);
+  return true;
+}
+
+void** MallocExtension::ReadStackTraces(int* sample_period) {
+  return NULL;
+}
+
+void** MallocExtension::ReadHeapGrowthStackTraces() {
+  return NULL;
+}
+
+void MallocExtension::MarkThreadIdle() {
+  // Default implementation does nothing
+}
+
+void MallocExtension::MarkThreadBusy() {
+  // Default implementation does nothing
+}
+
+SysAllocator* MallocExtension::GetSystemAllocator() {
+  return NULL;
+}
+
+void MallocExtension::SetSystemAllocator(SysAllocator *a) {
+  // Default implementation does nothing
+}
+
+void MallocExtension::ReleaseToSystem(size_t num_bytes) {
+  // Default implementation does nothing
+}
+
+void MallocExtension::ReleaseFreeMemory() {
+  ReleaseToSystem(static_cast<size_t>(-1));   // SIZE_T_MAX
+}
+
+void MallocExtension::SetMemoryReleaseRate(double rate) {
+  // Default implementation does nothing
+}
+
+double MallocExtension::GetMemoryReleaseRate() {
+  return -1.0;
+}
+
+size_t MallocExtension::GetEstimatedAllocatedSize(size_t size) {
+  return size;
+}
+
+size_t MallocExtension::GetAllocatedSize(const void* p) {
+  assert(GetOwnership(p) != kNotOwned);
+  return 0;
+}
+
+MallocExtension::Ownership MallocExtension::GetOwnership(const void* p) {
+  return kUnknownOwnership;
+}
+
+void MallocExtension::GetFreeListSizes(
+    vector<MallocExtension::FreeListInfo>* v) {
+  v->clear();
+}
+
+// The current malloc extension object.
+
+static pthread_once_t module_init = PTHREAD_ONCE_INIT;
+static MallocExtension* current_instance = NULL;
+
+static void InitModule() {
+  current_instance = new MallocExtension;
+#ifndef NO_HEAP_CHECK
+  HeapLeakChecker::IgnoreObject(current_instance);
+#endif
+}
+
+MallocExtension* MallocExtension::instance() {
+  perftools_pthread_once(&module_init, InitModule);
+  return current_instance;
+}
+
+void MallocExtension::Register(MallocExtension* implementation) {
+  perftools_pthread_once(&module_init, InitModule);
+  // When running under valgrind, our custom malloc is replaced with
+  // valgrind's one and malloc extensions will not work.  (Note:
+  // callers should be responsible for checking that they are the
+  // malloc that is really being run, before calling Register.  This
+  // is just here as an extra sanity check.)
+  if (!RunningOnValgrind()) {
+    current_instance = implementation;
+  }
+}
+
+// -----------------------------------------------------------------------
+// Heap sampling support
+// -----------------------------------------------------------------------
+
+namespace {
+
+// Accessors
+uintptr_t Count(void** entry) {
+  return reinterpret_cast<uintptr_t>(entry[0]);
+}
+uintptr_t Size(void** entry) {
+  return reinterpret_cast<uintptr_t>(entry[1]);
+}
+uintptr_t Depth(void** entry) {
+  return reinterpret_cast<uintptr_t>(entry[2]);
+}
+void* PC(void** entry, int i) {
+  return entry[3+i];
+}
+
+void PrintCountAndSize(MallocExtensionWriter* writer,
+                       uintptr_t count, uintptr_t size) {
+  char buf[100];
+  _snprintf(buf, sizeof(buf),
+           "%6" PRIu64 ": %8" PRIu64 " [%6" PRIu64 ": %8" PRIu64 "] @",
+           static_cast<uint64>(count),
+           static_cast<uint64>(size),
+           static_cast<uint64>(count),
+           static_cast<uint64>(size));
+  writer->append(buf, strlen(buf));
+}
+
+void PrintHeader(MallocExtensionWriter* writer,
+                 const char* label, void** entries) {
+  // Compute the total count and total size
+  uintptr_t total_count = 0;
+  uintptr_t total_size = 0;
+  for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) {
+    total_count += Count(entry);
+    total_size += Size(entry);
+  }
+
+  const char* const kTitle = "heap profile: ";
+  writer->append(kTitle, strlen(kTitle));
+  PrintCountAndSize(writer, total_count, total_size);
+  writer->append(" ", 1);
+  writer->append(label, strlen(label));
+  writer->append("\n", 1);
+}
+
+void PrintStackEntry(MallocExtensionWriter* writer, void** entry) {
+  PrintCountAndSize(writer, Count(entry), Size(entry));
+
+  for (int i = 0; i < Depth(entry); i++) {
+    char buf[32];
+    _snprintf(buf, sizeof(buf), " %p", PC(entry, i));
+    writer->append(buf, strlen(buf));
+  }
+  writer->append("\n", 1);
+}
+
+}
+
+void MallocExtension::GetHeapSample(MallocExtensionWriter* writer) {
+  int sample_period = 0;
+  void** entries = ReadStackTraces(&sample_period);
+  if (entries == NULL) {
+    const char* const kErrorMsg =
+        "This malloc implementation does not support sampling.\n"
+        "As of 2005/01/26, only tcmalloc supports sampling, and\n"
+        "you are probably running a binary that does not use\n"
+        "tcmalloc.\n";
+    writer->append(kErrorMsg, strlen(kErrorMsg));
+    return;
+  }
+
+  char label[32];
+  sprintf(label, "heap_v2/%d", sample_period);
+  PrintHeader(writer, label, entries);
+  for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) {
+    PrintStackEntry(writer, entry);
+  }
+  delete[] entries;
+
+  DumpAddressMap(writer);
+}
+
+void MallocExtension::GetHeapGrowthStacks(MallocExtensionWriter* writer) {
+  void** entries = ReadHeapGrowthStackTraces();
+  if (entries == NULL) {
+    const char* const kErrorMsg =
+        "This malloc implementation does not support "
+        "ReadHeapGrowthStackTraces().\n"
+        "As of 2005/09/27, only tcmalloc supports this, and you\n"
+        "are probably running a binary that does not use tcmalloc.\n";
+    writer->append(kErrorMsg, strlen(kErrorMsg));
+    return;
+  }
+
+  // Do not canonicalize the stack entries, so that we get a
+  // time-ordered list of stack traces, which may be useful if the
+  // client wants to focus on the latest stack traces.
+  PrintHeader(writer, "growth", entries);
+  for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) {
+    PrintStackEntry(writer, entry);
+  }
+  delete[] entries;
+
+  DumpAddressMap(writer);
+}
+
+void MallocExtension::Ranges(void* arg, RangeFunction func) {
+  // No callbacks by default
+}
+
+// These are C shims that work on the current instance.
+
+#define C_SHIM(fn, retval, paramlist, arglist)          \
+  extern "C" PERFTOOLS_DLL_DECL retval MallocExtension_##fn paramlist {    \
+    return MallocExtension::instance()->fn arglist;     \
+  }
+
+C_SHIM(VerifyAllMemory, int, (void), ());
+C_SHIM(VerifyNewMemory, int, (const void* p), (p));
+C_SHIM(VerifyArrayNewMemory, int, (const void* p), (p));
+C_SHIM(VerifyMallocMemory, int, (const void* p), (p));
+C_SHIM(MallocMemoryStats, int,
+       (int* blocks, size_t* total, int histogram[kMallocHistogramSize]),
+       (blocks, total, histogram));
+
+C_SHIM(GetStats, void,
+       (char* buffer, int buffer_length), (buffer, buffer_length));
+C_SHIM(GetNumericProperty, int,
+       (const char* property, size_t* value), (property, value));
+C_SHIM(SetNumericProperty, int,
+       (const char* property, size_t value), (property, value));
+
+C_SHIM(MarkThreadIdle, void, (void), ());
+C_SHIM(MarkThreadBusy, void, (void), ());
+C_SHIM(ReleaseFreeMemory, void, (void), ());
+C_SHIM(ReleaseToSystem, void, (size_t num_bytes), (num_bytes));
+C_SHIM(GetEstimatedAllocatedSize, size_t, (size_t size), (size));
+C_SHIM(GetAllocatedSize, size_t, (const void* p), (p));
+
+// Can't use the shim here because of the need to translate the enums.
+extern "C"
+MallocExtension_Ownership MallocExtension_GetOwnership(const void* p) {
+  return static_cast<MallocExtension_Ownership>(
+      MallocExtension::instance()->GetOwnership(p));
+}

BeefRT/gperftools/src/malloc_hook-inl.h

@@ -0,0 +1,322 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat
+//
+// This has the implementation details of malloc_hook that are needed
+// to use malloc-hook inside the tcmalloc system.  It does not hold
+// any of the client-facing calls that are used to add new hooks.
+
+#ifndef _MALLOC_HOOK_INL_H_
+#define _MALLOC_HOOK_INL_H_
+
+#include <stddef.h>
+#include <sys/types.h>
+#include "base/atomicops.h"
+#include "base/basictypes.h"
+#include "gperftools/malloc_hook.h"
+
+namespace base { namespace internal {
+
+// The following (implementation) code is DEPRECATED.
+// A simple atomic pointer class that can be initialized by the linker
+// when you define a namespace-scope variable as:
+//
+//   AtomicPtr<Foo*> my_global = { &initial_value };
+//
+// This isn't suitable for a general atomic<> class because of the
+// public access to data_.
+template<typename PtrT>
+class AtomicPtr {
+ public:
+  COMPILE_ASSERT(sizeof(PtrT) <= sizeof(AtomicWord),
+                 PtrT_should_fit_in_AtomicWord);
+
+  PtrT Get() const {
+    // Depending on the system, Acquire_Load(AtomicWord*) may have
+    // been defined to return an AtomicWord, Atomic32, or Atomic64.
+    // We hide that implementation detail here with an explicit cast.
+    // This prevents MSVC 2005, at least, from complaining (it has to
+    // do with __wp64; AtomicWord is __wp64, but Atomic32/64 aren't).
+    return reinterpret_cast<PtrT>(static_cast<AtomicWord>(
+      base::subtle::NoBarrier_Load(&data_)));
+  }
+
+  // Sets the contained value to new_val and returns the old value,
+  // atomically, with acquire and release semantics.
+  // This is a full-barrier instruction.
+  PtrT Exchange(PtrT new_val);
+
+  // Atomically executes:
+  //      result = data_
+  //      if (data_ == old_val)
+  //        data_ = new_val;
+  //      return result;
+  // This is a full-barrier instruction.
+  PtrT CompareAndSwap(PtrT old_val, PtrT new_val);
+
+  // Not private so that the class is an aggregate and can be
+  // initialized by the linker. Don't access this directly.
+  AtomicWord data_;
+};
+
+// These are initialized in malloc_hook.cc
+extern AtomicPtr<MallocHook::NewHook>     new_hook_;
+extern AtomicPtr<MallocHook::DeleteHook>  delete_hook_;
+extern AtomicPtr<MallocHook::PreMmapHook> premmap_hook_;
+extern AtomicPtr<MallocHook::MmapHook>    mmap_hook_;
+extern AtomicPtr<MallocHook::MunmapHook>  munmap_hook_;
+extern AtomicPtr<MallocHook::MremapHook>  mremap_hook_;
+extern AtomicPtr<MallocHook::PreSbrkHook> presbrk_hook_;
+extern AtomicPtr<MallocHook::SbrkHook>    sbrk_hook_;
+// End DEPRECATED code.
+
+// Maximum of 7 hooks means that HookList is 8 words.
+static const int kHookListMaxValues = 7;
+
+// HookList: a class that provides synchronized insertions and removals and
+// lockless traversal.  Most of the implementation is in malloc_hook.cc.
+template <typename T>
+struct PERFTOOLS_DLL_DECL HookList {
+  COMPILE_ASSERT(sizeof(T) <= sizeof(AtomicWord), T_should_fit_in_AtomicWord);
+
+  // Adds value to the list.  Note that duplicates are allowed.  Thread-safe and
+  // blocking (acquires hooklist_spinlock).  Returns true on success; false
+  // otherwise (failures include invalid value and no space left).
+  bool Add(T value);
+
+  // Removes the first entry matching value from the list.  Thread-safe and
+  // blocking (acquires hooklist_spinlock).  Returns true on success; false
+  // otherwise (failures include invalid value and no value found).
+  bool Remove(T value);
+
+  // Store up to n values of the list in output_array, and return the number of
+  // elements stored.  Thread-safe and non-blocking.  This is fast (one memory
+  // access) if the list is empty.
+  int Traverse(T* output_array, int n) const;
+
+  // Fast inline implementation for fast path of Invoke*Hook.
+  bool empty() const {
+    return base::subtle::NoBarrier_Load(&priv_end) == 0;
+  }
+
+  // This internal data is not private so that the class is an aggregate and can
+  // be initialized by the linker.  Don't access this directly.  Use the
+  // INIT_HOOK_LIST macro in malloc_hook.cc.
+
+  // One more than the index of the last valid element in priv_data.  During
+  // 'Remove' this may be past the last valid element in priv_data, but
+  // subsequent values will be 0.
+  AtomicWord priv_end;
+  AtomicWord priv_data[kHookListMaxValues];
+};
+
+extern HookList<MallocHook::NewHook> new_hooks_;
+extern HookList<MallocHook::DeleteHook> delete_hooks_;
+extern HookList<MallocHook::PreMmapHook> premmap_hooks_;
+extern HookList<MallocHook::MmapHook> mmap_hooks_;
+extern HookList<MallocHook::MmapReplacement> mmap_replacement_;
+extern HookList<MallocHook::MunmapHook> munmap_hooks_;
+extern HookList<MallocHook::MunmapReplacement> munmap_replacement_;
+extern HookList<MallocHook::MremapHook> mremap_hooks_;
+extern HookList<MallocHook::PreSbrkHook> presbrk_hooks_;
+extern HookList<MallocHook::SbrkHook> sbrk_hooks_;
+
+} }  // namespace base::internal
+
+// The following method is DEPRECATED
+inline MallocHook::NewHook MallocHook::GetNewHook() {
+  return base::internal::new_hook_.Get();
+}
+
+inline void MallocHook::InvokeNewHook(const void* p, size_t s) {
+  if (!base::internal::new_hooks_.empty()) {
+    InvokeNewHookSlow(p, s);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::NewHook hook = MallocHook::GetNewHook();
+  if (hook != NULL) (*hook)(p, s);
+  // End DEPRECATED code.
+}
+
+// The following method is DEPRECATED
+inline MallocHook::DeleteHook MallocHook::GetDeleteHook() {
+  return base::internal::delete_hook_.Get();
+}
+
+inline void MallocHook::InvokeDeleteHook(const void* p) {
+  if (!base::internal::delete_hooks_.empty()) {
+    InvokeDeleteHookSlow(p);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::DeleteHook hook = MallocHook::GetDeleteHook();
+  if (hook != NULL) (*hook)(p);
+  // End DEPRECATED code.
+}
+
+// The following method is DEPRECATED
+inline MallocHook::PreMmapHook MallocHook::GetPreMmapHook() {
+  return base::internal::premmap_hook_.Get();
+}
+
+inline void MallocHook::InvokePreMmapHook(const void* start,
+                                          size_t size,
+                                          int protection,
+                                          int flags,
+                                          int fd,
+                                          off_t offset) {
+  if (!base::internal::premmap_hooks_.empty()) {
+    InvokePreMmapHookSlow(start, size, protection, flags, fd, offset);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::PreMmapHook hook = MallocHook::GetPreMmapHook();
+  if (hook != NULL) (*hook)(start, size,
+                            protection, flags,
+                            fd, offset);
+  // End DEPRECATED code.
+}
+
+// The following method is DEPRECATED
+inline MallocHook::MmapHook MallocHook::GetMmapHook() {
+  return base::internal::mmap_hook_.Get();
+}
+
+inline void MallocHook::InvokeMmapHook(const void* result,
+                                       const void* start,
+                                       size_t size,
+                                       int protection,
+                                       int flags,
+                                       int fd,
+                                       off_t offset) {
+  if (!base::internal::mmap_hooks_.empty()) {
+    InvokeMmapHookSlow(result, start, size, protection, flags, fd, offset);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::MmapHook hook = MallocHook::GetMmapHook();
+  if (hook != NULL) (*hook)(result,
+                            start, size,
+                            protection, flags,
+                            fd, offset);
+  // End DEPRECATED code.
+}
+
+inline bool MallocHook::InvokeMmapReplacement(const void* start,
+                                              size_t size,
+                                              int protection,
+                                              int flags,
+                                              int fd,
+                                              off_t offset,
+                                              void** result) {
+  if (!base::internal::mmap_replacement_.empty()) {
+    return InvokeMmapReplacementSlow(start, size,
+                                     protection, flags,
+                                     fd, offset,
+                                     result);
+  }
+  return false;
+}
+
+// The following method is DEPRECATED
+inline MallocHook::MunmapHook MallocHook::GetMunmapHook() {
+  return base::internal::munmap_hook_.Get();
+}
+
+inline void MallocHook::InvokeMunmapHook(const void* p, size_t size) {
+  if (!base::internal::munmap_hooks_.empty()) {
+    InvokeMunmapHookSlow(p, size);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::MunmapHook hook = MallocHook::GetMunmapHook();
+  if (hook != NULL) (*hook)(p, size);
+  // End DEPRECATED code.
+}
+
+inline bool MallocHook::InvokeMunmapReplacement(
+    const void* p, size_t size, int* result) {
+  if (!base::internal::mmap_replacement_.empty()) {
+    return InvokeMunmapReplacementSlow(p, size, result);
+  }
+  return false;
+}
+
+// The following method is DEPRECATED
+inline MallocHook::MremapHook MallocHook::GetMremapHook() {
+  return base::internal::mremap_hook_.Get();
+}
+
+inline void MallocHook::InvokeMremapHook(const void* result,
+                                         const void* old_addr,
+                                         size_t old_size,
+                                         size_t new_size,
+                                         int flags,
+                                         const void* new_addr) {
+  if (!base::internal::mremap_hooks_.empty()) {
+    InvokeMremapHookSlow(result, old_addr, old_size, new_size, flags, new_addr);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::MremapHook hook = MallocHook::GetMremapHook();
+  if (hook != NULL) (*hook)(result,
+                            old_addr, old_size,
+                            new_size, flags, new_addr);
+  // End DEPRECATED code.
+}
+
+// The following method is DEPRECATED
+inline MallocHook::PreSbrkHook MallocHook::GetPreSbrkHook() {
+  return base::internal::presbrk_hook_.Get();
+}
+
+inline void MallocHook::InvokePreSbrkHook(ptrdiff_t increment) {
+  if (!base::internal::presbrk_hooks_.empty() && increment != 0) {
+    InvokePreSbrkHookSlow(increment);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::PreSbrkHook hook = MallocHook::GetPreSbrkHook();
+  if (hook != NULL && increment != 0) (*hook)(increment);
+  // End DEPRECATED code.
+}
+
+// The following method is DEPRECATED
+inline MallocHook::SbrkHook MallocHook::GetSbrkHook() {
+  return base::internal::sbrk_hook_.Get();
+}
+
+inline void MallocHook::InvokeSbrkHook(const void* result,
+                                       ptrdiff_t increment) {
+  if (!base::internal::sbrk_hooks_.empty() && increment != 0) {
+    InvokeSbrkHookSlow(result, increment);
+  }
+  // The following code is DEPRECATED.
+  MallocHook::SbrkHook hook = MallocHook::GetSbrkHook();
+  if (hook != NULL && increment != 0) (*hook)(result, increment);
+  // End DEPRECATED code.
+}
+
+#endif /* _MALLOC_HOOK_INL_H_ */

BeefRT/gperftools/src/malloc_hook.cc

@@ -0,0 +1,723 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+#include "tcconfig.h"
+
+// Disable the glibc prototype of mremap(), as older versions of the
+// system headers define this function with only four arguments,
+// whereas newer versions allow an optional fifth argument:
+#ifdef HAVE_MMAP
+# define mremap glibc_mremap
+# include <sys/mman.h>
+# undef mremap
+#endif
+
+#include <stddef.h>
+#ifdef HAVE_STDINT_H
+#include <stdint.h>
+#endif
+#include <algorithm>
+#include "base/logging.h"
+#include "base/spinlock.h"
+#include "maybe_threads.h"
+#include "malloc_hook-inl.h"
+#include "gperftools/malloc_hook.h"
+
+// This #ifdef should almost never be set.  Set NO_TCMALLOC_SAMPLES if
+// you're porting to a system where you really can't get a stacktrace.
+#ifdef NO_TCMALLOC_SAMPLES
+  // We use #define so code compiles even if you #include stacktrace.h somehow.
+# define GetStackTrace(stack, depth, skip)  (0)
+#else
+# include "gperftools/stacktrace.h"
+#endif
+
+// __THROW is defined in glibc systems.  It means, counter-intuitively,
+// "This function will never throw an exception."  It's an optional
+// optimization tool, but we may need to use it to match glibc prototypes.
+#ifndef __THROW    // I guess we're not on a glibc system
+# define __THROW   // __THROW is just an optimization, so ok to make it ""
+#endif
+
+using std::copy;
+
+
+// Declaration of default weak initialization function, that can be overridden
+// by linking-in a strong definition (as heap-checker.cc does).  This is
+// extern "C" so that it doesn't trigger gold's --detect-odr-violations warning,
+// which only looks at C++ symbols.
+//
+// This function is declared here as weak, and defined later, rather than a more
+// straightforward simple weak definition, as a workround for an icc compiler
+// issue ((Intel reference 290819).  This issue causes icc to resolve weak
+// symbols too early, at compile rather than link time.  By declaring it (weak)
+// here, then defining it below after its use, we can avoid the problem.
+extern "C" {
+ATTRIBUTE_WEAK void MallocHook_InitAtFirstAllocation_HeapLeakChecker();
+}
+
+namespace {
+
+void RemoveInitialHooksAndCallInitializers();  // below.
+
+pthread_once_t once = PTHREAD_ONCE_INIT;
+
+// These hooks are installed in MallocHook as the only initial hooks.  The first
+// hook that is called will run RemoveInitialHooksAndCallInitializers (see the
+// definition below) and then redispatch to any malloc hooks installed by
+// RemoveInitialHooksAndCallInitializers.
+//
+// Note(llib): there is a possibility of a race in the event that there are
+// multiple threads running before the first allocation.  This is pretty
+// difficult to achieve, but if it is then multiple threads may concurrently do
+// allocations.  The first caller will call
+// RemoveInitialHooksAndCallInitializers via one of the initial hooks.  A
+// concurrent allocation may, depending on timing either:
+// * still have its initial malloc hook installed, run that and block on waiting
+//   for the first caller to finish its call to
+//   RemoveInitialHooksAndCallInitializers, and proceed normally.
+// * occur some time during the RemoveInitialHooksAndCallInitializers call, at
+//   which point there could be no initial hooks and the subsequent hooks that
+//   are about to be set up by RemoveInitialHooksAndCallInitializers haven't
+//   been installed yet.  I think the worst we can get is that some allocations
+//   will not get reported to some hooks set by the initializers called from
+//   RemoveInitialHooksAndCallInitializers.
+
+void InitialNewHook(const void* ptr, size_t size) {
+  perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers);
+  MallocHook::InvokeNewHook(ptr, size);
+}
+
+void InitialPreMMapHook(const void* start,
+                               size_t size,
+                               int protection,
+                               int flags,
+                               int fd,
+                               off_t offset) {
+  perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers);
+  MallocHook::InvokePreMmapHook(start, size, protection, flags, fd, offset);
+}
+
+void InitialPreSbrkHook(ptrdiff_t increment) {
+  perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers);
+  MallocHook::InvokePreSbrkHook(increment);
+}
+
+// This function is called at most once by one of the above initial malloc
+// hooks.  It removes all initial hooks and initializes all other clients that
+// want to get control at the very first memory allocation.  The initializers
+// may assume that the initial malloc hooks have been removed.  The initializers
+// may set up malloc hooks and allocate memory.
+void RemoveInitialHooksAndCallInitializers() {
+  RAW_CHECK(MallocHook::RemoveNewHook(&InitialNewHook), "");
+  RAW_CHECK(MallocHook::RemovePreMmapHook(&InitialPreMMapHook), "");
+  RAW_CHECK(MallocHook::RemovePreSbrkHook(&InitialPreSbrkHook), "");
+
+  // HeapLeakChecker is currently the only module that needs to get control on
+  // the first memory allocation, but one can add other modules by following the
+  // same weak/strong function pattern.
+  MallocHook_InitAtFirstAllocation_HeapLeakChecker();
+}
+
+}  // namespace
+
+// Weak default initialization function that must go after its use.
+extern "C" void MallocHook_InitAtFirstAllocation_HeapLeakChecker() {
+  // Do nothing.
+}
+
+namespace base { namespace internal {
+
+// The code below is DEPRECATED.
+template<typename PtrT>
+PtrT AtomicPtr<PtrT>::Exchange(PtrT new_val) {
+  base::subtle::GMemoryBarrier();  // Release semantics.
+  // Depending on the system, NoBarrier_AtomicExchange(AtomicWord*)
+  // may have been defined to return an AtomicWord, Atomic32, or
+  // Atomic64.  We hide that implementation detail here with an
+  // explicit cast.  This prevents MSVC 2005, at least, from complaining.
+  PtrT old_val = reinterpret_cast<PtrT>(static_cast<AtomicWord>(
+      base::subtle::NoBarrier_AtomicExchange(
+          &data_,
+          reinterpret_cast<AtomicWord>(new_val))));
+  base::subtle::GMemoryBarrier();  // And acquire semantics.
+  return old_val;
+}
+
+template<typename PtrT>
+PtrT AtomicPtr<PtrT>::CompareAndSwap(PtrT old_val, PtrT new_val) {
+  base::subtle::GMemoryBarrier();  // Release semantics.
+  PtrT retval = reinterpret_cast<PtrT>(static_cast<AtomicWord>(
+      base::subtle::NoBarrier_CompareAndSwap(
+          &data_,
+          reinterpret_cast<AtomicWord>(old_val),
+          reinterpret_cast<AtomicWord>(new_val))));
+  base::subtle::GMemoryBarrier();  // And acquire semantics.
+  return retval;
+}
+
+AtomicPtr<MallocHook::NewHook>    new_hook_ = { 0 };
+AtomicPtr<MallocHook::DeleteHook> delete_hook_ = { 0 };
+AtomicPtr<MallocHook::PreMmapHook> premmap_hook_ = { 0 };
+AtomicPtr<MallocHook::MmapHook>   mmap_hook_ = { 0 };
+AtomicPtr<MallocHook::MunmapHook> munmap_hook_ = { 0 };
+AtomicPtr<MallocHook::MremapHook> mremap_hook_ = { 0 };
+AtomicPtr<MallocHook::PreSbrkHook> presbrk_hook_ = { 0 };
+AtomicPtr<MallocHook::SbrkHook>   sbrk_hook_ = { 0 };
+// End of DEPRECATED code section.
+
+// This lock is shared between all implementations of HookList::Add & Remove.
+// The potential for contention is very small.  This needs to be a SpinLock and
+// not a Mutex since it's possible for Mutex locking to allocate memory (e.g.,
+// per-thread allocation in debug builds), which could cause infinite recursion.
+static SpinLock hooklist_spinlock(base::LINKER_INITIALIZED);
+
+template <typename T>
+bool HookList<T>::Add(T value_as_t) {
+  AtomicWord value = bit_cast<AtomicWord>(value_as_t);
+  if (value == 0) {
+    return false;
+  }
+  SpinLockHolder l(&hooklist_spinlock);
+  // Find the first slot in data that is 0.
+  int index = 0;
+  while ((index < kHookListMaxValues) &&
+         (base::subtle::NoBarrier_Load(&priv_data[index]) != 0)) {
+    ++index;
+  }
+  if (index == kHookListMaxValues) {
+    return false;
+  }
+  AtomicWord prev_num_hooks = base::subtle::Acquire_Load(&priv_end);
+  base::subtle::Release_Store(&priv_data[index], value);
+  if (prev_num_hooks <= index) {
+    base::subtle::Release_Store(&priv_end, index + 1);
+  }
+  return true;
+}
+
+template <typename T>
+bool HookList<T>::Remove(T value_as_t) {
+  if (value_as_t == 0) {
+    return false;
+  }
+  SpinLockHolder l(&hooklist_spinlock);
+  AtomicWord hooks_end = base::subtle::Acquire_Load(&priv_end);
+  int index = 0;
+  while (index < hooks_end && value_as_t != bit_cast<T>(
+             base::subtle::Acquire_Load(&priv_data[index]))) {
+    ++index;
+  }
+  if (index == hooks_end) {
+    return false;
+  }
+  base::subtle::Release_Store(&priv_data[index], 0);
+  if (hooks_end == index + 1) {
+    // Adjust hooks_end down to the lowest possible value.
+    hooks_end = index;
+    while ((hooks_end > 0) &&
+           (base::subtle::Acquire_Load(&priv_data[hooks_end - 1]) == 0)) {
+      --hooks_end;
+    }
+    base::subtle::Release_Store(&priv_end, hooks_end);
+  }
+  return true;
+}
+
+template <typename T>
+int HookList<T>::Traverse(T* output_array, int n) const {
+  AtomicWord hooks_end = base::subtle::Acquire_Load(&priv_end);
+  int actual_hooks_end = 0;
+  for (int i = 0; i < hooks_end && n > 0; ++i) {
+    AtomicWord data = base::subtle::Acquire_Load(&priv_data[i]);
+    if (data != 0) {
+      *output_array++ = bit_cast<T>(data);
+      ++actual_hooks_end;
+      --n;
+    }
+  }
+  return actual_hooks_end;
+}
+
+// Initialize a HookList (optionally with the given initial_value in index 0).
+#define INIT_HOOK_LIST { 0 }
+#define INIT_HOOK_LIST_WITH_VALUE(initial_value)                \
+  { 1, { reinterpret_cast<AtomicWord>(initial_value) } }
+
+// Explicit instantiation for malloc_hook_test.cc.  This ensures all the methods
+// are instantiated.
+template struct HookList<MallocHook::NewHook>;
+
+HookList<MallocHook::NewHook> new_hooks_ =
+    INIT_HOOK_LIST_WITH_VALUE(&InitialNewHook);
+HookList<MallocHook::DeleteHook> delete_hooks_ = INIT_HOOK_LIST;
+HookList<MallocHook::PreMmapHook> premmap_hooks_ =
+    INIT_HOOK_LIST_WITH_VALUE(&InitialPreMMapHook);
+HookList<MallocHook::MmapHook> mmap_hooks_ = INIT_HOOK_LIST;
+HookList<MallocHook::MunmapHook> munmap_hooks_ = INIT_HOOK_LIST;
+HookList<MallocHook::MremapHook> mremap_hooks_ = INIT_HOOK_LIST;
+HookList<MallocHook::PreSbrkHook> presbrk_hooks_ =
+    INIT_HOOK_LIST_WITH_VALUE(InitialPreSbrkHook);
+HookList<MallocHook::SbrkHook> sbrk_hooks_ = INIT_HOOK_LIST;
+
+// These lists contain either 0 or 1 hooks.
+HookList<MallocHook::MmapReplacement> mmap_replacement_ = { 0 };
+HookList<MallocHook::MunmapReplacement> munmap_replacement_ = { 0 };
+
+#undef INIT_HOOK_LIST_WITH_VALUE
+#undef INIT_HOOK_LIST
+
+} }  // namespace base::internal
+
+// The code below is DEPRECATED.
+using base::internal::new_hook_;
+using base::internal::delete_hook_;
+using base::internal::premmap_hook_;
+using base::internal::mmap_hook_;
+using base::internal::munmap_hook_;
+using base::internal::mremap_hook_;
+using base::internal::presbrk_hook_;
+using base::internal::sbrk_hook_;
+// End of DEPRECATED code section.
+
+using base::internal::kHookListMaxValues;
+using base::internal::new_hooks_;
+using base::internal::delete_hooks_;
+using base::internal::premmap_hooks_;
+using base::internal::mmap_hooks_;
+using base::internal::mmap_replacement_;
+using base::internal::munmap_hooks_;
+using base::internal::munmap_replacement_;
+using base::internal::mremap_hooks_;
+using base::internal::presbrk_hooks_;
+using base::internal::sbrk_hooks_;
+
+// These are available as C bindings as well as C++, hence their
+// definition outside the MallocHook class.
+extern "C"
+int MallocHook_AddNewHook(MallocHook_NewHook hook) {
+  RAW_VLOG(10, "AddNewHook(%p)", hook);
+  return new_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveNewHook(MallocHook_NewHook hook) {
+  RAW_VLOG(10, "RemoveNewHook(%p)", hook);
+  return new_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddDeleteHook(MallocHook_DeleteHook hook) {
+  RAW_VLOG(10, "AddDeleteHook(%p)", hook);
+  return delete_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveDeleteHook(MallocHook_DeleteHook hook) {
+  RAW_VLOG(10, "RemoveDeleteHook(%p)", hook);
+  return delete_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddPreMmapHook(MallocHook_PreMmapHook hook) {
+  RAW_VLOG(10, "AddPreMmapHook(%p)", hook);
+  return premmap_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemovePreMmapHook(MallocHook_PreMmapHook hook) {
+  RAW_VLOG(10, "RemovePreMmapHook(%p)", hook);
+  return premmap_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_SetMmapReplacement(MallocHook_MmapReplacement hook) {
+  RAW_VLOG(10, "SetMmapReplacement(%p)", hook);
+  // NOTE this is a best effort CHECK. Concurrent sets could succeed since
+  // this test is outside of the Add spin lock.
+  RAW_CHECK(mmap_replacement_.empty(), "Only one MMapReplacement is allowed.");
+  return mmap_replacement_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveMmapReplacement(MallocHook_MmapReplacement hook) {
+  RAW_VLOG(10, "RemoveMmapReplacement(%p)", hook);
+  return mmap_replacement_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddMmapHook(MallocHook_MmapHook hook) {
+  RAW_VLOG(10, "AddMmapHook(%p)", hook);
+  return mmap_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveMmapHook(MallocHook_MmapHook hook) {
+  RAW_VLOG(10, "RemoveMmapHook(%p)", hook);
+  return mmap_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddMunmapHook(MallocHook_MunmapHook hook) {
+  RAW_VLOG(10, "AddMunmapHook(%p)", hook);
+  return munmap_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveMunmapHook(MallocHook_MunmapHook hook) {
+  RAW_VLOG(10, "RemoveMunmapHook(%p)", hook);
+  return munmap_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_SetMunmapReplacement(MallocHook_MunmapReplacement hook) {
+  RAW_VLOG(10, "SetMunmapReplacement(%p)", hook);
+  // NOTE this is a best effort CHECK. Concurrent sets could succeed since
+  // this test is outside of the Add spin lock.
+  RAW_CHECK(munmap_replacement_.empty(),
+            "Only one MunmapReplacement is allowed.");
+  return munmap_replacement_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveMunmapReplacement(MallocHook_MunmapReplacement hook) {
+  RAW_VLOG(10, "RemoveMunmapReplacement(%p)", hook);
+  return munmap_replacement_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddMremapHook(MallocHook_MremapHook hook) {
+  RAW_VLOG(10, "AddMremapHook(%p)", hook);
+  return mremap_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveMremapHook(MallocHook_MremapHook hook) {
+  RAW_VLOG(10, "RemoveMremapHook(%p)", hook);
+  return mremap_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddPreSbrkHook(MallocHook_PreSbrkHook hook) {
+  RAW_VLOG(10, "AddPreSbrkHook(%p)", hook);
+  return presbrk_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemovePreSbrkHook(MallocHook_PreSbrkHook hook) {
+  RAW_VLOG(10, "RemovePreSbrkHook(%p)", hook);
+  return presbrk_hooks_.Remove(hook);
+}
+
+extern "C"
+int MallocHook_AddSbrkHook(MallocHook_SbrkHook hook) {
+  RAW_VLOG(10, "AddSbrkHook(%p)", hook);
+  return sbrk_hooks_.Add(hook);
+}
+
+extern "C"
+int MallocHook_RemoveSbrkHook(MallocHook_SbrkHook hook) {
+  RAW_VLOG(10, "RemoveSbrkHook(%p)", hook);
+  return sbrk_hooks_.Remove(hook);
+}
+
+// The code below is DEPRECATED.
+extern "C"
+MallocHook_NewHook MallocHook_SetNewHook(MallocHook_NewHook hook) {
+  RAW_VLOG(10, "SetNewHook(%p)", hook);
+  return new_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_DeleteHook MallocHook_SetDeleteHook(MallocHook_DeleteHook hook) {
+  RAW_VLOG(10, "SetDeleteHook(%p)", hook);
+  return delete_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_PreMmapHook MallocHook_SetPreMmapHook(MallocHook_PreMmapHook hook) {
+  RAW_VLOG(10, "SetPreMmapHook(%p)", hook);
+  return premmap_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_MmapHook MallocHook_SetMmapHook(MallocHook_MmapHook hook) {
+  RAW_VLOG(10, "SetMmapHook(%p)", hook);
+  return mmap_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_MunmapHook MallocHook_SetMunmapHook(MallocHook_MunmapHook hook) {
+  RAW_VLOG(10, "SetMunmapHook(%p)", hook);
+  return munmap_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_MremapHook MallocHook_SetMremapHook(MallocHook_MremapHook hook) {
+  RAW_VLOG(10, "SetMremapHook(%p)", hook);
+  return mremap_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_PreSbrkHook MallocHook_SetPreSbrkHook(MallocHook_PreSbrkHook hook) {
+  RAW_VLOG(10, "SetPreSbrkHook(%p)", hook);
+  return presbrk_hook_.Exchange(hook);
+}
+
+extern "C"
+MallocHook_SbrkHook MallocHook_SetSbrkHook(MallocHook_SbrkHook hook) {
+  RAW_VLOG(10, "SetSbrkHook(%p)", hook);
+  return sbrk_hook_.Exchange(hook);
+}
+// End of DEPRECATED code section.
+
+// Note: embedding the function calls inside the traversal of HookList would be
+// very confusing, as it is legal for a hook to remove itself and add other
+// hooks.  Doing traversal first, and then calling the hooks ensures we only
+// call the hooks registered at the start.
+#define INVOKE_HOOKS(HookType, hook_list, args) do {                    \
+    HookType hooks[kHookListMaxValues];                                 \
+    int num_hooks = hook_list.Traverse(hooks, kHookListMaxValues);      \
+    for (int i = 0; i < num_hooks; ++i) {                               \
+      (*hooks[i])args;                                                  \
+    }                                                                   \
+  } while (0)
+
+// There should only be one replacement. Return the result of the first
+// one, or false if there is none.
+#define INVOKE_REPLACEMENT(HookType, hook_list, args) do {              \
+    HookType hooks[kHookListMaxValues];                                 \
+    int num_hooks = hook_list.Traverse(hooks, kHookListMaxValues);      \
+    return (num_hooks > 0 && (*hooks[0])args);                          \
+  } while (0)
+
+
+void MallocHook::InvokeNewHookSlow(const void* p, size_t s) {
+  INVOKE_HOOKS(NewHook, new_hooks_, (p, s));
+}
+
+void MallocHook::InvokeDeleteHookSlow(const void* p) {
+  INVOKE_HOOKS(DeleteHook, delete_hooks_, (p));
+}
+
+void MallocHook::InvokePreMmapHookSlow(const void* start,
+                                       size_t size,
+                                       int protection,
+                                       int flags,
+                                       int fd,
+                                       off_t offset) {
+  INVOKE_HOOKS(PreMmapHook, premmap_hooks_, (start, size, protection, flags, fd,
+                                            offset));
+}
+
+void MallocHook::InvokeMmapHookSlow(const void* result,
+                                    const void* start,
+                                    size_t size,
+                                    int protection,
+                                    int flags,
+                                    int fd,
+                                    off_t offset) {
+  INVOKE_HOOKS(MmapHook, mmap_hooks_, (result, start, size, protection, flags,
+                                       fd, offset));
+}
+
+bool MallocHook::InvokeMmapReplacementSlow(const void* start,
+                                           size_t size,
+                                           int protection,
+                                           int flags,
+                                           int fd,
+                                           off_t offset,
+                                           void** result) {
+  INVOKE_REPLACEMENT(MmapReplacement, mmap_replacement_,
+                      (start, size, protection, flags, fd, offset, result));
+}
+
+void MallocHook::InvokeMunmapHookSlow(const void* p, size_t s) {
+  INVOKE_HOOKS(MunmapHook, munmap_hooks_, (p, s));
+}
+
+bool MallocHook::InvokeMunmapReplacementSlow(const void* p,
+                                             size_t s,
+                                             int* result) {
+  INVOKE_REPLACEMENT(MunmapReplacement, munmap_replacement_, (p, s, result));
+}
+
+void MallocHook::InvokeMremapHookSlow(const void* result,
+                                      const void* old_addr,
+                                      size_t old_size,
+                                      size_t new_size,
+                                      int flags,
+                                      const void* new_addr) {
+  INVOKE_HOOKS(MremapHook, mremap_hooks_, (result, old_addr, old_size, new_size,
+                                           flags, new_addr));
+}
+
+void MallocHook::InvokePreSbrkHookSlow(ptrdiff_t increment) {
+  INVOKE_HOOKS(PreSbrkHook, presbrk_hooks_, (increment));
+}
+
+void MallocHook::InvokeSbrkHookSlow(const void* result, ptrdiff_t increment) {
+  INVOKE_HOOKS(SbrkHook, sbrk_hooks_, (result, increment));
+}
+
+#undef INVOKE_HOOKS
+
+DEFINE_ATTRIBUTE_SECTION_VARS(google_malloc);
+DECLARE_ATTRIBUTE_SECTION_VARS(google_malloc);
+  // actual functions are in debugallocation.cc or tcmalloc.cc
+DEFINE_ATTRIBUTE_SECTION_VARS(malloc_hook);
+DECLARE_ATTRIBUTE_SECTION_VARS(malloc_hook);
+  // actual functions are in this file, malloc_hook.cc, and low_level_alloc.cc
+
+#define ADDR_IN_ATTRIBUTE_SECTION(addr, name) \
+  (reinterpret_cast<uintptr_t>(ATTRIBUTE_SECTION_START(name)) <= \
+     reinterpret_cast<uintptr_t>(addr) && \
+   reinterpret_cast<uintptr_t>(addr) < \
+     reinterpret_cast<uintptr_t>(ATTRIBUTE_SECTION_STOP(name)))
+
+// Return true iff 'caller' is a return address within a function
+// that calls one of our hooks via MallocHook:Invoke*.
+// A helper for GetCallerStackTrace.
+static inline bool InHookCaller(const void* caller) {
+  return ADDR_IN_ATTRIBUTE_SECTION(caller, google_malloc) ||
+         ADDR_IN_ATTRIBUTE_SECTION(caller, malloc_hook);
+  // We can use one section for everything except tcmalloc_or_debug
+  // due to its special linkage mode, which prevents merging of the sections.
+}
+
+#undef ADDR_IN_ATTRIBUTE_SECTION
+
+static bool checked_sections = false;
+
+static inline void CheckInHookCaller() {
+  if (!checked_sections) {
+    INIT_ATTRIBUTE_SECTION_VARS(google_malloc);
+    if (ATTRIBUTE_SECTION_START(google_malloc) ==
+        ATTRIBUTE_SECTION_STOP(google_malloc)) {
+      RAW_LOG(ERROR, "google_malloc section is missing, "
+                     "thus InHookCaller is broken!");
+    }
+    INIT_ATTRIBUTE_SECTION_VARS(malloc_hook);
+    if (ATTRIBUTE_SECTION_START(malloc_hook) ==
+        ATTRIBUTE_SECTION_STOP(malloc_hook)) {
+      RAW_LOG(ERROR, "malloc_hook section is missing, "
+                     "thus InHookCaller is broken!");
+    }
+    checked_sections = true;
+  }
+}
+
+// We can improve behavior/compactness of this function
+// if we pass a generic test function (with a generic arg)
+// into the implementations for GetStackTrace instead of the skip_count.
+extern "C" int MallocHook_GetCallerStackTrace(void** result, int max_depth,
+                                              int skip_count) {
+#if defined(NO_TCMALLOC_SAMPLES)
+  return 0;
+#elif !defined(HAVE_ATTRIBUTE_SECTION_START)
+  // Fall back to GetStackTrace and good old but fragile frame skip counts.
+  // Note: this path is inaccurate when a hook is not called directly by an
+  // allocation function but is daisy-chained through another hook,
+  // search for MallocHook::(Get|Set|Invoke)* to find such cases.
+  return GetStackTrace(result, max_depth, skip_count + int(DEBUG_MODE));
+  // due to -foptimize-sibling-calls in opt mode
+  // there's no need for extra frame skip here then
+#else
+  CheckInHookCaller();
+  // MallocHook caller determination via InHookCaller works, use it:
+  static const int kMaxSkip = 32 + 6 + 3;
+    // Constant tuned to do just one GetStackTrace call below in practice
+    // and not get many frames that we don't actually need:
+    // currently max passsed max_depth is 32,
+    // max passed/needed skip_count is 6
+    // and 3 is to account for some hook daisy chaining.
+  static const int kStackSize = kMaxSkip + 1;
+  void* stack[kStackSize];
+  int depth = GetStackTrace(stack, kStackSize, 1);  // skip this function frame
+  if (depth == 0)   // silenty propagate cases when GetStackTrace does not work
+    return 0;
+  for (int i = 0; i < depth; ++i) {  // stack[0] is our immediate caller
+    if (InHookCaller(stack[i])) {
+      RAW_VLOG(10, "Found hooked allocator at %d: %p <- %p",
+                   i, stack[i], stack[i+1]);
+      i += 1;  // skip hook caller frame
+      depth -= i;  // correct depth
+      if (depth > max_depth) depth = max_depth;
+      copy(stack + i, stack + i + depth, result);
+      if (depth < max_depth  &&  depth + i == kStackSize) {
+        // get frames for the missing depth
+        depth +=
+          GetStackTrace(result + depth, max_depth - depth, 1 + kStackSize);
+      }
+      return depth;
+    }
+  }
+  RAW_LOG(WARNING, "Hooked allocator frame not found, returning empty trace");
+    // If this happens try increasing kMaxSkip
+    // or else something must be wrong with InHookCaller,
+    // e.g. for every section used in InHookCaller
+    // all functions in that section must be inside the same library.
+  return 0;
+#endif
+}
+
+// On systems where we know how, we override mmap/munmap/mremap/sbrk
+// to provide support for calling the related hooks (in addition,
+// of course, to doing what these functions normally do).
+
+#if defined(__linux)
+# include "malloc_hook_mmap_linux.h"
+
+#elif defined(__FreeBSD__)
+# include "malloc_hook_mmap_freebsd.h"
+
+#else
+
+/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot,
+                                         int flags, int fd, off_t offset) {
+  void* result;
+  if (!MallocHook::InvokeMmapReplacement(
+          start, length, prot, flags, fd, offset, &result)) {
+    result = mmap(start, length, prot, flags, fd, offset);
+  }
+  return result;
+}
+
+/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) {
+  int result;
+  if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) {
+    result = munmap(start, length);
+  }
+  return result;
+}
+
+#endif

BeefRT/gperftools/src/malloc_hook_mmap_freebsd.h

@@ -0,0 +1,165 @@
+// Copyright (c) 2011, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// Override mmap/munmap/mremap/sbrk to provide support for calling the
+// related hooks (in addition, of course, to doing what these
+// functions normally do).
+
+#ifndef __FreeBSD__
+# error Should only be including malloc_hook_mmap_freebsd.h on FreeBSD systems.
+#endif
+
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <sys/mman.h>
+#include <errno.h>
+
+// Make sure mmap doesn't get #define'd away by <sys/mman.h>
+#undef mmap
+
+// According to the FreeBSD documentation, use syscall if you do not
+// need 64-bit alignment otherwise use __syscall. Indeed, syscall
+// doesn't work correctly in most situations on 64-bit. It's return
+// type is 'int' so for things like SYS_mmap, it actually truncates
+// the returned address to 32-bits.
+#if defined(__amd64__) || defined(__x86_64__)
+# define MALLOC_HOOK_SYSCALL __syscall
+#else
+# define MALLOC_HOOK_SYSCALL syscall
+#endif
+
+
+extern "C" {
+  void* mmap(void *start, size_t length,int prot, int flags,
+             int fd, off_t offset) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  int munmap(void* start, size_t length) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  void* sbrk(intptr_t increment) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+}
+
+static inline void* do_mmap(void *start, size_t length,
+                            int prot, int flags,
+                            int fd, off_t offset) __THROW {
+  return (void *)MALLOC_HOOK_SYSCALL(SYS_mmap,
+                                     start, length, prot, flags, fd, offset);
+}
+
+static inline void* do_sbrk(intptr_t increment) {
+  void* curbrk = 0;
+
+#if defined(__x86_64__) || defined(__amd64__)
+# ifdef PIC
+  __asm__ __volatile__(
+      "movq .curbrk@GOTPCREL(%%rip), %%rdx;"
+      "movq (%%rdx), %%rax;"
+      "movq %%rax, %0;"
+      : "=r" (curbrk)
+      :: "%rdx", "%rax");
+# else
+  __asm__ __volatile__(
+      "movq .curbrk(%%rip), %%rax;"
+      "movq %%rax, %0;"
+      : "=r" (curbrk)
+      :: "%rax");
+# endif
+#else
+  __asm__ __volatile__(
+      "movl .curbrk, %%eax;"
+      "movl %%eax, %0;"
+      : "=r" (curbrk)
+      :: "%eax");
+#endif
+
+  if (increment == 0) {
+    return curbrk;
+  }
+
+  char* prevbrk = static_cast<char*>(curbrk);
+  void* newbrk = prevbrk + increment;
+
+  if (brk(newbrk) == -1) {
+    return reinterpret_cast<void*>(static_cast<intptr_t>(-1));
+  }
+
+  return prevbrk;
+}
+
+
+extern "C" void* mmap(void *start, size_t length, int prot, int flags,
+                      int fd, off_t offset) __THROW {
+  MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset);
+  void *result;
+  if (!MallocHook::InvokeMmapReplacement(
+          start, length, prot, flags, fd, offset, &result)) {
+    result = do_mmap(start, length, prot, flags, fd,
+                       static_cast<size_t>(offset)); // avoid sign extension
+  }
+  MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset);
+  return result;
+}
+
+extern "C" int munmap(void* start, size_t length) __THROW {
+  MallocHook::InvokeMunmapHook(start, length);
+  int result;
+  if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) {
+    result = MALLOC_HOOK_SYSCALL(SYS_munmap, start, length);
+  }
+
+  return result;
+}
+
+extern "C" void* sbrk(intptr_t increment) __THROW {
+  MallocHook::InvokePreSbrkHook(increment);
+  void *result = do_sbrk(increment);
+  MallocHook::InvokeSbrkHook(result, increment);
+  return result;
+}
+
+/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot,
+                                         int flags, int fd, off_t offset) {
+  void* result;
+  if (!MallocHook::InvokeMmapReplacement(
+	  start, length, prot, flags, fd, offset, &result)) {
+    result = do_mmap(start, length, prot, flags, fd, offset);
+  }
+
+  return result;
+}
+
+/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) {
+  int result;
+  if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) {
+    result = MALLOC_HOOK_SYSCALL(SYS_munmap, start, length);
+  }
+  return result;
+}
+
+#undef MALLOC_HOOK_SYSCALL

BeefRT/gperftools/src/malloc_hook_mmap_linux.h

@@ -0,0 +1,234 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+// We define mmap() and mmap64(), which somewhat reimplements libc's mmap
+// syscall stubs.  Unfortunately libc only exports the stubs via weak symbols
+// (which we're overriding with our mmap64() and mmap() wrappers) so we can't
+// just call through to them.
+
+#ifndef __linux
+# error Should only be including malloc_hook_mmap_linux.h on linux systems.
+#endif
+
+#include <unistd.h>
+#include <syscall.h>
+#include <sys/mman.h>
+#include <errno.h>
+#include "base/linux_syscall_support.h"
+
+// The x86-32 case and the x86-64 case differ:
+// 32b has a mmap2() syscall, 64b does not.
+// 64b and 32b have different calling conventions for mmap().
+
+// I test for 64-bit first so I don't have to do things like
+// '#if (defined(__mips__) && !defined(__MIPS64__))' as a mips32 check.
+#if defined(__x86_64__) || defined(__PPC64__) || (defined(_MIPS_SIM) && _MIPS_SIM == _ABI64)
+
+static inline void* do_mmap64(void *start, size_t length,
+                              int prot, int flags,
+                              int fd, __off64_t offset) __THROW {
+  return sys_mmap(start, length, prot, flags, fd, offset);
+}
+
+#define MALLOC_HOOK_HAVE_DO_MMAP64 1
+
+#elif defined(__i386__) || defined(__PPC__) || defined(__mips__) || \
+      defined(__arm__)
+
+static inline void* do_mmap64(void *start, size_t length,
+                              int prot, int flags,
+                              int fd, __off64_t offset) __THROW {
+  void *result;
+
+  // Try mmap2() unless it's not supported
+  static bool have_mmap2 = true;
+  if (have_mmap2) {
+    static int pagesize = 0;
+    if (!pagesize) pagesize = getpagesize();
+
+    // Check that the offset is page aligned
+    if (offset & (pagesize - 1)) {
+      result = MAP_FAILED;
+      errno = EINVAL;
+      goto out;
+    }
+
+    result = (void *)syscall(SYS_mmap2,
+                             start, length, prot, flags, fd,
+                             (off_t) (offset / pagesize));
+    if (result != MAP_FAILED || errno != ENOSYS)  goto out;
+
+    // We don't have mmap2() after all - don't bother trying it in future
+    have_mmap2 = false;
+  }
+
+  if (((off_t)offset) != offset) {
+    // If we're trying to map a 64-bit offset, fail now since we don't
+    // have 64-bit mmap() support.
+    result = MAP_FAILED;
+    errno = EINVAL;
+    goto out;
+  }
+
+#ifdef __NR_mmap
+  {
+    // Fall back to old 32-bit offset mmap() call
+    // Old syscall interface cannot handle six args, so pass in an array
+    int32 args[6] = { (int32) start, (int32) length, prot, flags, fd,
+                      (off_t) offset };
+    result = (void *)syscall(SYS_mmap, args);
+  }
+#else
+  // Some Linux ports like ARM EABI Linux has no mmap, just mmap2.
+  result = MAP_FAILED;
+#endif
+
+ out:
+  return result;
+}
+
+#define MALLOC_HOOK_HAVE_DO_MMAP64 1
+
+#endif  // #if defined(__x86_64__)
+
+
+#ifdef MALLOC_HOOK_HAVE_DO_MMAP64
+
+// We use do_mmap64 abstraction to put MallocHook::InvokeMmapHook
+// calls right into mmap and mmap64, so that the stack frames in the caller's
+// stack are at the same offsets for all the calls of memory allocating
+// functions.
+
+// Put all callers of MallocHook::Invoke* in this module into
+// malloc_hook section,
+// so that MallocHook::GetCallerStackTrace can function accurately:
+
+// Make sure mmap doesn't get #define'd away by <sys/mman.h>
+# undef mmap
+
+extern "C" {
+  void* mmap64(void *start, size_t length, int prot, int flags,
+               int fd, __off64_t offset  ) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  void* mmap(void *start, size_t length,int prot, int flags,
+             int fd, off_t offset) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  int munmap(void* start, size_t length) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  void* mremap(void* old_addr, size_t old_size, size_t new_size,
+               int flags, ...) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+  void* sbrk(ptrdiff_t increment) __THROW
+    ATTRIBUTE_SECTION(malloc_hook);
+}
+
+extern "C" void* mmap64(void *start, size_t length, int prot, int flags,
+                        int fd, __off64_t offset) __THROW {
+  MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset);
+  void *result;
+  if (!MallocHook::InvokeMmapReplacement(
+          start, length, prot, flags, fd, offset, &result)) {
+    result = do_mmap64(start, length, prot, flags, fd, offset);
+  }
+  MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset);
+  return result;
+}
+
+# if !defined(__USE_FILE_OFFSET64) || !defined(__REDIRECT_NTH)
+
+extern "C" void* mmap(void *start, size_t length, int prot, int flags,
+                      int fd, off_t offset) __THROW {
+  MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset);
+  void *result;
+  if (!MallocHook::InvokeMmapReplacement(
+          start, length, prot, flags, fd, offset, &result)) {
+    result = do_mmap64(start, length, prot, flags, fd,
+                       static_cast<size_t>(offset)); // avoid sign extension
+  }
+  MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset);
+  return result;
+}
+
+# endif  // !defined(__USE_FILE_OFFSET64) || !defined(__REDIRECT_NTH)
+
+extern "C" int munmap(void* start, size_t length) __THROW {
+  MallocHook::InvokeMunmapHook(start, length);
+  int result;
+  if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) {
+    result = sys_munmap(start, length);
+  }
+  return result;
+}
+
+extern "C" void* mremap(void* old_addr, size_t old_size, size_t new_size,
+                        int flags, ...) __THROW {
+  va_list ap;
+  va_start(ap, flags);
+  void *new_address = va_arg(ap, void *);
+  va_end(ap);
+  void* result = sys_mremap(old_addr, old_size, new_size, flags, new_address);
+  MallocHook::InvokeMremapHook(result, old_addr, old_size, new_size, flags,
+                               new_address);
+  return result;
+}
+
+// libc's version:
+extern "C" void* __sbrk(ptrdiff_t increment);
+
+extern "C" void* sbrk(ptrdiff_t increment) __THROW {
+  MallocHook::InvokePreSbrkHook(increment);
+  void *result = __sbrk(increment);
+  MallocHook::InvokeSbrkHook(result, increment);
+  return result;
+}
+
+/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot,
+                                         int flags, int fd, off_t offset) {
+  void* result;
+  if (!MallocHook::InvokeMmapReplacement(
+          start, length, prot, flags, fd, offset, &result)) {
+    result = do_mmap64(start, length, prot, flags, fd, offset);
+  }
+  return result;
+}
+
+/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) {
+  int result;
+  if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) {
+    result = syscall(SYS_munmap, start, length);
+  }
+  return result;
+}
+
+#undef MALLOC_HOOK_HAVE_DO_MMAP64
+
+#endif  // #ifdef MALLOC_HOOK_HAVE_DO_MMAP64

BeefRT/gperftools/src/maybe_threads.cc

@@ -0,0 +1,133 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Menage <opensource@google.com>
+//
+// Some wrappers for pthread functions so that we can be LD_PRELOADed
+// against non-pthreads apps.
+//
+// This module will behave very strangely if some pthreads functions
+// exist and others don't.
+
+#include "config.h"
+#include <assert.h>
+#include <string.h>    // for memcmp
+#include <stdio.h>     // for __isthreaded on FreeBSD
+// We don't actually need strings. But including this header seems to
+// stop the compiler trying to short-circuit our pthreads existence
+// tests and claiming that the address of a function is always
+// non-zero. I have no idea why ...
+#include <string>
+#include "maybe_threads.h"
+#include "base/basictypes.h"
+
+// __THROW is defined in glibc systems.  It means, counter-intuitively,
+// "This function will never throw an exception."  It's an optional
+// optimization tool, but we may need to use it to match glibc prototypes.
+#ifndef __THROW    // I guess we're not on a glibc system
+# define __THROW   // __THROW is just an optimization, so ok to make it ""
+#endif
+
+// These are the methods we're going to conditionally include.
+extern "C" {
+  int pthread_key_create (pthread_key_t*, void (*)(void*))
+      __THROW ATTRIBUTE_WEAK;
+  void *pthread_getspecific(pthread_key_t)
+      __THROW ATTRIBUTE_WEAK;
+  int pthread_setspecific(pthread_key_t, const void*)
+      __THROW ATTRIBUTE_WEAK;
+  int pthread_once(pthread_once_t *, void (*)(void))
+      ATTRIBUTE_WEAK;
+}
+
+#define MAX_PERTHREAD_VALS 16
+static void *perftools_pthread_specific_vals[MAX_PERTHREAD_VALS];
+static int next_key;
+
+int perftools_pthread_key_create(pthread_key_t *key,
+                                 void (*destr_function) (void *)) {
+  if (pthread_key_create) {
+    return pthread_key_create(key, destr_function);
+  } else {
+    assert(next_key < MAX_PERTHREAD_VALS);
+    *key = (pthread_key_t)(next_key++);
+    return 0;
+  }
+}
+
+void *perftools_pthread_getspecific(pthread_key_t key) {
+  if (pthread_getspecific) {
+    return pthread_getspecific(key);
+  } else {
+    return perftools_pthread_specific_vals[(int)key];
+  }
+}
+
+int perftools_pthread_setspecific(pthread_key_t key, void *val) {
+  if (pthread_setspecific) {
+    return pthread_setspecific(key, val);
+  } else {
+    perftools_pthread_specific_vals[(int)key] = val;
+    return 0;
+  }
+}
+
+
+static pthread_once_t pthread_once_init = PTHREAD_ONCE_INIT;
+int perftools_pthread_once(pthread_once_t *ctl,
+                           void  (*init_routine) (void)) {
+#ifdef __FreeBSD__
+  // On __FreeBSD__, calling pthread_once on a system that is not
+  // linked with -pthread is silently a noop. :-( Luckily, we have a
+  // workaround: FreeBSD exposes __isthreaded in <stdio.h>, which is
+  // set to 1 when the first thread is spawned.  So on those systems,
+  // we can use our own separate pthreads-once mechanism, which is
+  // used until __isthreaded is 1 (which will never be true if the app
+  // is not linked with -pthread).
+  static bool pthread_once_ran_before_threads = false;
+  if (pthread_once_ran_before_threads) {
+    return 0;
+  }
+  if (!__isthreaded) {
+    init_routine();
+    pthread_once_ran_before_threads = true;
+    return 0;
+  }
+#endif
+  if (pthread_once) {
+    return pthread_once(ctl, init_routine);
+  } else {
+    if (memcmp(ctl, &pthread_once_init, sizeof(*ctl)) == 0) {
+      init_routine();
+      ++*(char*)(ctl);        // make it so it's no longer equal to init
+    }
+    return 0;
+  }
+}

BeefRT/gperftools/src/maybe_threads.h

@@ -0,0 +1,52 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Paul Menage <opensource@google.com>
+
+//-------------------------------------------------------------------
+// Some wrappers for pthread functions so that we can be LD_PRELOADed
+// against non-pthreads apps.
+//-------------------------------------------------------------------
+
+#ifndef GOOGLE_MAYBE_THREADS_H_
+#define GOOGLE_MAYBE_THREADS_H_
+
+#ifdef HAVE_PTHREAD
+#include <pthread.h>
+#endif
+
+int perftools_pthread_key_create(pthread_key_t *key,
+                                 void (*destr_function) (void *));
+void *perftools_pthread_getspecific(pthread_key_t key);
+int perftools_pthread_setspecific(pthread_key_t key, void *val);
+int perftools_pthread_once(pthread_once_t *ctl,
+                           void  (*init_routine) (void));
+
+#endif  /* GOOGLE_MAYBE_THREADS_H_ */

BeefRT/gperftools/src/memfs_malloc.cc

@@ -0,0 +1,267 @@
+// Copyright (c) 2007, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Arun Sharma
+//
+// A tcmalloc system allocator that uses a memory based filesystem such as
+// tmpfs or hugetlbfs
+//
+// Since these only exist on linux, we only register this allocator there.
+
+#ifdef __linux
+
+#include <config.h>
+#include <errno.h>                      // for errno, EINVAL
+#include <inttypes.h>                   // for PRId64
+#include <limits.h>                     // for PATH_MAX
+#include <stddef.h>                     // for size_t, NULL
+#ifdef HAVE_STDINT_H
+#include <stdint.h>                     // for int64_t, uintptr_t
+#endif
+#include <stdio.h>                      // for snprintf
+#include <stdlib.h>                     // for mkstemp
+#include <string.h>                     // for strerror
+#include <sys/mman.h>                   // for mmap, MAP_FAILED, etc
+#include <sys/statfs.h>                 // for fstatfs, statfs
+#include <unistd.h>                     // for ftruncate, off_t, unlink
+#include <new>                          // for operator new
+#include <string>
+
+#include <gperftools/malloc_extension.h>
+#include "base/basictypes.h"
+#include "base/googleinit.h"
+#include "base/sysinfo.h"
+#include "internal_logging.h"
+
+// TODO(sanjay): Move the code below into the tcmalloc namespace
+using tcmalloc::kLog;
+using tcmalloc::kCrash;
+using tcmalloc::Log;
+using std::string;
+
+DEFINE_string(memfs_malloc_path, EnvToString("TCMALLOC_MEMFS_MALLOC_PATH", ""),
+              "Path where hugetlbfs or tmpfs is mounted. The caller is "
+              "responsible for ensuring that the path is unique and does "
+              "not conflict with another process");
+DEFINE_int64(memfs_malloc_limit_mb,
+             EnvToInt("TCMALLOC_MEMFS_LIMIT_MB", 0),
+             "Limit total allocation size to the "
+             "specified number of MiB.  0 == no limit.");
+DEFINE_bool(memfs_malloc_abort_on_fail,
+            EnvToBool("TCMALLOC_MEMFS_ABORT_ON_FAIL", false),
+            "abort() whenever memfs_malloc fails to satisfy an allocation "
+            "for any reason.");
+DEFINE_bool(memfs_malloc_ignore_mmap_fail,
+            EnvToBool("TCMALLOC_MEMFS_IGNORE_MMAP_FAIL", false),
+            "Ignore failures from mmap");
+DEFINE_bool(memfs_malloc_map_private,
+            EnvToBool("TCMALLOC_MEMFS_MAP_PRIVATE", false),
+	    "Use MAP_PRIVATE with mmap");
+
+// Hugetlbfs based allocator for tcmalloc
+class HugetlbSysAllocator: public SysAllocator {
+public:
+  explicit HugetlbSysAllocator(SysAllocator* fallback)
+    : failed_(true),  // To disable allocator until Initialize() is called.
+      big_page_size_(0),
+      hugetlb_fd_(-1),
+      hugetlb_base_(0),
+      fallback_(fallback) {
+  }
+
+  void* Alloc(size_t size, size_t *actual_size, size_t alignment);
+  bool Initialize();
+
+  bool failed_;          // Whether failed to allocate memory.
+
+private:
+  void* AllocInternal(size_t size, size_t *actual_size, size_t alignment);
+
+  int64 big_page_size_;
+  int hugetlb_fd_;       // file descriptor for hugetlb
+  off_t hugetlb_base_;
+
+  SysAllocator* fallback_;  // Default system allocator to fall back to.
+};
+static char hugetlb_space[sizeof(HugetlbSysAllocator)];
+
+// No locking needed here since we assume that tcmalloc calls
+// us with an internal lock held (see tcmalloc/system-alloc.cc).
+void* HugetlbSysAllocator::Alloc(size_t size, size_t *actual_size,
+                                 size_t alignment) {
+  if (failed_) {
+    return fallback_->Alloc(size, actual_size, alignment);
+  }
+
+  // We don't respond to allocation requests smaller than big_page_size_ unless
+  // the caller is ok to take more than they asked for. Used by MetaDataAlloc.
+  if (actual_size == NULL && size < big_page_size_) {
+    return fallback_->Alloc(size, actual_size, alignment);
+  }
+
+  // Enforce huge page alignment.  Be careful to deal with overflow.
+  size_t new_alignment = alignment;
+  if (new_alignment < big_page_size_) new_alignment = big_page_size_;
+  size_t aligned_size = ((size + new_alignment - 1) /
+                         new_alignment) * new_alignment;
+  if (aligned_size < size) {
+    return fallback_->Alloc(size, actual_size, alignment);
+  }
+
+  void* result = AllocInternal(aligned_size, actual_size, new_alignment);
+  if (result != NULL) {
+    return result;
+  }
+  Log(kLog, __FILE__, __LINE__,
+      "HugetlbSysAllocator: (failed, allocated)", failed_, hugetlb_base_);
+  if (FLAGS_memfs_malloc_abort_on_fail) {
+    Log(kCrash, __FILE__, __LINE__,
+        "memfs_malloc_abort_on_fail is set");
+  }
+  return fallback_->Alloc(size, actual_size, alignment);
+}
+
+void* HugetlbSysAllocator::AllocInternal(size_t size, size_t* actual_size,
+                                         size_t alignment) {
+  // Ask for extra memory if alignment > pagesize
+  size_t extra = 0;
+  if (alignment > big_page_size_) {
+    extra = alignment - big_page_size_;
+  }
+
+  // Test if this allocation would put us over the limit.
+  off_t limit = FLAGS_memfs_malloc_limit_mb*1024*1024;
+  if (limit > 0 && hugetlb_base_ + size + extra > limit) {
+    // Disable the allocator when there's less than one page left.
+    if (limit - hugetlb_base_ < big_page_size_) {
+      Log(kLog, __FILE__, __LINE__, "reached memfs_malloc_limit_mb");
+      failed_ = true;
+    }
+    else {
+      Log(kLog, __FILE__, __LINE__,
+          "alloc too large (size, bytes left)", size, limit-hugetlb_base_);
+    }
+    return NULL;
+  }
+
+  // This is not needed for hugetlbfs, but needed for tmpfs.  Annoyingly
+  // hugetlbfs returns EINVAL for ftruncate.
+  int ret = ftruncate(hugetlb_fd_, hugetlb_base_ + size + extra);
+  if (ret != 0 && errno != EINVAL) {
+    Log(kLog, __FILE__, __LINE__,
+        "ftruncate failed", strerror(errno));
+    failed_ = true;
+    return NULL;
+  }
+
+  // Note: size + extra does not overflow since:
+  //            size + alignment < (1<<NBITS).
+  // and        extra <= alignment
+  // therefore  size + extra < (1<<NBITS)
+  void *result;
+  result = mmap(0, size + extra, PROT_WRITE|PROT_READ,
+                FLAGS_memfs_malloc_map_private ? MAP_PRIVATE : MAP_SHARED,
+                hugetlb_fd_, hugetlb_base_);
+  if (result == reinterpret_cast<void*>(MAP_FAILED)) {
+    if (!FLAGS_memfs_malloc_ignore_mmap_fail) {
+      Log(kLog, __FILE__, __LINE__,
+          "mmap failed (size, error)", size + extra, strerror(errno));
+      failed_ = true;
+    }
+    return NULL;
+  }
+  uintptr_t ptr = reinterpret_cast<uintptr_t>(result);
+
+  // Adjust the return memory so it is aligned
+  size_t adjust = 0;
+  if ((ptr & (alignment - 1)) != 0) {
+    adjust = alignment - (ptr & (alignment - 1));
+  }
+  ptr += adjust;
+  hugetlb_base_ += (size + extra);
+
+  if (actual_size) {
+    *actual_size = size + extra - adjust;
+  }
+
+  return reinterpret_cast<void*>(ptr);
+}
+
+bool HugetlbSysAllocator::Initialize() {
+  char path[PATH_MAX];
+  const int pathlen = FLAGS_memfs_malloc_path.size();
+  if (pathlen + 8 > sizeof(path)) {
+    Log(kCrash, __FILE__, __LINE__, "XX fatal: memfs_malloc_path too long");
+    return false;
+  }
+  memcpy(path, FLAGS_memfs_malloc_path.data(), pathlen);
+  memcpy(path + pathlen, ".XXXXXX", 8);  // Also copies terminating \0
+
+  int hugetlb_fd = mkstemp(path);
+  if (hugetlb_fd == -1) {
+    Log(kLog, __FILE__, __LINE__,
+        "warning: unable to create memfs_malloc_path",
+        path, strerror(errno));
+    return false;
+  }
+
+  // Cleanup memory on process exit
+  if (unlink(path) == -1) {
+    Log(kCrash, __FILE__, __LINE__,
+        "fatal: error unlinking memfs_malloc_path", path, strerror(errno));
+    return false;
+  }
+
+  // Use fstatfs to figure out the default page size for memfs
+  struct statfs sfs;
+  if (fstatfs(hugetlb_fd, &sfs) == -1) {
+    Log(kCrash, __FILE__, __LINE__,
+        "fatal: error fstatfs of memfs_malloc_path", strerror(errno));
+    return false;
+  }
+  int64 page_size = sfs.f_bsize;
+
+  hugetlb_fd_ = hugetlb_fd;
+  big_page_size_ = page_size;
+  failed_ = false;
+  return true;
+}
+
+REGISTER_MODULE_INITIALIZER(memfs_malloc, {
+  if (FLAGS_memfs_malloc_path.length()) {
+    SysAllocator* alloc = MallocExtension::instance()->GetSystemAllocator();
+    HugetlbSysAllocator* hp = new (hugetlb_space) HugetlbSysAllocator(alloc);
+    if (hp->Initialize()) {
+      MallocExtension::instance()->SetSystemAllocator(hp);
+    }
+  }
+});
+
+#endif   /* ifdef __linux */

BeefRT/gperftools/src/memory_region_map.cc

@@ -0,0 +1,656 @@
+#if 0
+
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Maxim Lifantsev
+ */
+
+//
+// Background and key design points of MemoryRegionMap.
+//
+// MemoryRegionMap is a low-level module with quite atypical requirements that
+// result in some degree of non-triviality of the implementation and design.
+//
+// MemoryRegionMap collects info about *all* memory regions created with
+// mmap, munmap, mremap, sbrk.
+// They key word above is 'all': all that are happening in a process
+// during its lifetime frequently starting even before global object
+// constructor execution.
+//
+// This is needed by the primary client of MemoryRegionMap:
+// HeapLeakChecker uses the regions and the associated stack traces
+// to figure out what part of the memory is the heap:
+// if MemoryRegionMap were to miss some (early) regions, leak checking would
+// stop working correctly.
+//
+// To accomplish the goal of functioning before/during global object
+// constructor execution MemoryRegionMap is done as a singleton service
+// that relies on own on-demand initialized static constructor-less data,
+// and only relies on other low-level modules that can also function properly
+// even before global object constructors run.
+//
+// Accomplishing the goal of collecting data about all mmap, munmap, mremap,
+// sbrk occurrences is a more involved: conceptually to do this one needs to
+// record some bits of data in particular about any mmap or sbrk call,
+// but to do that one needs to allocate memory for that data at some point,
+// but all memory allocations in the end themselves come from an mmap
+// or sbrk call (that's how the address space of the process grows).
+//
+// Also note that we need to do all the above recording from
+// within an mmap/sbrk hook which is sometimes/frequently is made by a memory
+// allocator, including the allocator MemoryRegionMap itself must rely on.
+// In the case of heap-checker usage this includes even the very first
+// mmap/sbrk call happening in the program: heap-checker gets activated due to
+// a link-time installed mmap/sbrk hook and it initializes MemoryRegionMap
+// and asks it to record info about this very first call right from that
+// very first hook invocation.
+//
+// MemoryRegionMap is doing its memory allocations via LowLevelAlloc:
+// unlike more complex standard memory allocator, LowLevelAlloc cooperates with
+// MemoryRegionMap by not holding any of its own locks while it calls mmap
+// to get memory, thus we are able to call LowLevelAlloc from
+// our mmap/sbrk hooks without causing a deadlock in it.
+// For the same reason of deadlock prevention the locking in MemoryRegionMap
+// itself is write-recursive which is an exception to Google's mutex usage.
+//
+// We still need to break the infinite cycle of mmap calling our hook,
+// which asks LowLevelAlloc for memory to record this mmap,
+// which (sometimes) causes mmap, which calls our hook, and so on.
+// We do this as follows: on a recursive call of MemoryRegionMap's
+// mmap/sbrk/mremap hook we record the data about the allocation in a
+// static fixed-sized stack (saved_regions), when the recursion unwinds
+// but before returning from the outer hook call we unwind this stack and
+// move the data from saved_regions to its permanent place in the RegionSet,
+// which can cause more allocations and mmap-s and recursion and unwinding,
+// but the whole process ends eventually due to the fact that for the small
+// allocations we are doing LowLevelAlloc reuses one mmap call and parcels out
+// the memory it created to satisfy several of our allocation requests.
+//
+
+// ========================================================================= //
+
+#include <config.h>
+
+#ifdef HAVE_UNISTD_H
+#include <unistd.h>
+#endif
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h>
+#endif
+#ifdef HAVE_MMAP
+#include <sys/mman.h>
+#elif !defined(MAP_FAILED)
+#define MAP_FAILED -1  // the only thing we need from mman.h
+#endif
+#ifdef HAVE_PTHREAD
+#include <pthread.h>   // for pthread_t, pthread_self()
+#endif
+#include <stddef.h>
+
+#include <algorithm>
+#include <set>
+
+#include "memory_region_map.h"
+
+#include "base/logging.h"
+#include "base/low_level_alloc.h"
+#include "malloc_hook-inl.h"
+
+#include <gperftools/stacktrace.h>
+#include <gperftools/malloc_hook.h>
+
+// MREMAP_FIXED is a linux extension.  How it's used in this file,
+// setting it to 0 is equivalent to saying, "This feature isn't
+// supported", which is right.
+#ifndef MREMAP_FIXED
+# define MREMAP_FIXED  0
+#endif
+
+using std::max;
+
+// ========================================================================= //
+
+int MemoryRegionMap::client_count_ = 0;
+int MemoryRegionMap::max_stack_depth_ = 0;
+MemoryRegionMap::RegionSet* MemoryRegionMap::regions_ = NULL;
+LowLevelAlloc::Arena* MemoryRegionMap::arena_ = NULL;
+SpinLock MemoryRegionMap::lock_(SpinLock::LINKER_INITIALIZED);
+SpinLock MemoryRegionMap::owner_lock_(  // ACQUIRED_AFTER(lock_)
+    SpinLock::LINKER_INITIALIZED);
+int MemoryRegionMap::recursion_count_ = 0;  // GUARDED_BY(owner_lock_)
+pthread_t MemoryRegionMap::lock_owner_tid_;  // GUARDED_BY(owner_lock_)
+int64 MemoryRegionMap::map_size_ = 0;
+int64 MemoryRegionMap::unmap_size_ = 0;
+
+// ========================================================================= //
+
+// Simple hook into execution of global object constructors,
+// so that we do not call pthread_self() when it does not yet work.
+static bool libpthread_initialized = false;
+static bool initializer = (libpthread_initialized = true, true);
+
+static inline bool current_thread_is(pthread_t should_be) {
+  // Before main() runs, there's only one thread, so we're always that thread
+  if (!libpthread_initialized) return true;
+  // this starts working only sometime well into global constructor execution:
+  return pthread_equal(pthread_self(), should_be);
+}
+
+// ========================================================================= //
+
+// Constructor-less place-holder to store a RegionSet in.
+union MemoryRegionMap::RegionSetRep {
+  char rep[sizeof(RegionSet)];
+  void* align_it;  // do not need a better alignment for 'rep' than this
+  RegionSet* region_set() { return reinterpret_cast<RegionSet*>(rep); }
+};
+
+// The bytes where MemoryRegionMap::regions_ will point to.
+// We use RegionSetRep with noop c-tor so that global construction
+// does not interfere.
+static MemoryRegionMap::RegionSetRep regions_rep;
+
+// ========================================================================= //
+
+// Has InsertRegionLocked been called recursively
+// (or rather should we *not* use regions_ to record a hooked mmap).
+static bool recursive_insert = false;
+
+void MemoryRegionMap::Init(int max_stack_depth) {
+  RAW_VLOG(10, "MemoryRegionMap Init");
+  RAW_CHECK(max_stack_depth >= 0, "");
+  // Make sure we don't overflow the memory in region stacks:
+  RAW_CHECK(max_stack_depth <= kMaxStackDepth,
+            "need to increase kMaxStackDepth?");
+  Lock();
+  client_count_ += 1;
+  max_stack_depth_ = max(max_stack_depth_, max_stack_depth);
+  if (client_count_ > 1) {
+    // not first client: already did initialization-proper
+    Unlock();
+    RAW_VLOG(10, "MemoryRegionMap Init increment done");
+    return;
+  }
+  // Set our hooks and make sure they were installed:
+  RAW_CHECK(MallocHook::AddMmapHook(&MmapHook), "");
+  RAW_CHECK(MallocHook::AddMremapHook(&MremapHook), "");
+  RAW_CHECK(MallocHook::AddSbrkHook(&SbrkHook), "");
+  RAW_CHECK(MallocHook::AddMunmapHook(&MunmapHook), "");
+  // We need to set recursive_insert since the NewArena call itself
+  // will already do some allocations with mmap which our hooks will catch
+  // recursive_insert allows us to buffer info about these mmap calls.
+  // Note that Init() can be (and is) sometimes called
+  // already from within an mmap/sbrk hook.
+  recursive_insert = true;
+  arena_ = LowLevelAlloc::NewArena(0, LowLevelAlloc::DefaultArena());
+  recursive_insert = false;
+  HandleSavedRegionsLocked(&InsertRegionLocked);  // flush the buffered ones
+    // Can't instead use HandleSavedRegionsLocked(&DoInsertRegionLocked) before
+    // recursive_insert = false; as InsertRegionLocked will also construct
+    // regions_ on demand for us.
+  Unlock();
+  RAW_VLOG(10, "MemoryRegionMap Init done");
+}
+
+bool MemoryRegionMap::Shutdown() {
+  RAW_VLOG(10, "MemoryRegionMap Shutdown");
+  Lock();
+  RAW_CHECK(client_count_ > 0, "");
+  client_count_ -= 1;
+  if (client_count_ != 0) {  // not last client; need not really shutdown
+    Unlock();
+    RAW_VLOG(10, "MemoryRegionMap Shutdown decrement done");
+    return true;
+  }
+  RAW_CHECK(MallocHook::RemoveMmapHook(&MmapHook), "");
+  RAW_CHECK(MallocHook::RemoveMremapHook(&MremapHook), "");
+  RAW_CHECK(MallocHook::RemoveSbrkHook(&SbrkHook), "");
+  RAW_CHECK(MallocHook::RemoveMunmapHook(&MunmapHook), "");
+  if (regions_) regions_->~RegionSet();
+  regions_ = NULL;
+  bool deleted_arena = LowLevelAlloc::DeleteArena(arena_);
+  if (deleted_arena) {
+    arena_ = 0;
+  } else {
+    RAW_LOG(WARNING, "Can't delete LowLevelAlloc arena: it's being used");
+  }
+  Unlock();
+  RAW_VLOG(10, "MemoryRegionMap Shutdown done");
+  return deleted_arena;
+}
+
+// Invariants (once libpthread_initialized is true):
+//   * While lock_ is not held, recursion_count_ is 0 (and
+//     lock_owner_tid_ is the previous owner, but we don't rely on
+//     that).
+//   * recursion_count_ and lock_owner_tid_ are only written while
+//     both lock_ and owner_lock_ are held. They may be read under
+//     just owner_lock_.
+//   * At entry and exit of Lock() and Unlock(), the current thread
+//     owns lock_ iff pthread_equal(lock_owner_tid_, pthread_self())
+//     && recursion_count_ > 0.
+void MemoryRegionMap::Lock() {
+  {
+    SpinLockHolder l(&owner_lock_);
+    if (recursion_count_ > 0 && current_thread_is(lock_owner_tid_)) {
+      RAW_CHECK(lock_.IsHeld(), "Invariants violated");
+      recursion_count_++;
+      RAW_CHECK(recursion_count_ <= 5,
+                "recursive lock nesting unexpectedly deep");
+      return;
+    }
+  }
+  lock_.Lock();
+  {
+    SpinLockHolder l(&owner_lock_);
+    RAW_CHECK(recursion_count_ == 0,
+              "Last Unlock didn't reset recursion_count_");
+    if (libpthread_initialized)
+      lock_owner_tid_ = pthread_self();
+    recursion_count_ = 1;
+  }
+}
+
+void MemoryRegionMap::Unlock() {
+  SpinLockHolder l(&owner_lock_);
+  RAW_CHECK(recursion_count_ >  0, "unlock when not held");
+  RAW_CHECK(lock_.IsHeld(),
+            "unlock when not held, and recursion_count_ is wrong");
+  RAW_CHECK(current_thread_is(lock_owner_tid_), "unlock by non-holder");
+  recursion_count_--;
+  if (recursion_count_ == 0) {
+    lock_.Unlock();
+  }
+}
+
+bool MemoryRegionMap::LockIsHeld() {
+  SpinLockHolder l(&owner_lock_);
+  return lock_.IsHeld()  &&  current_thread_is(lock_owner_tid_);
+}
+
+const MemoryRegionMap::Region*
+MemoryRegionMap::DoFindRegionLocked(uintptr_t addr) {
+  RAW_CHECK(LockIsHeld(), "should be held (by this thread)");
+  if (regions_ != NULL) {
+    Region sample;
+    sample.SetRegionSetKey(addr);
+    RegionSet::iterator region = regions_->lower_bound(sample);
+    if (region != regions_->end()) {
+      RAW_CHECK(addr <= region->end_addr, "");
+      if (region->start_addr <= addr  &&  addr < region->end_addr) {
+        return &(*region);
+      }
+    }
+  }
+  return NULL;
+}
+
+bool MemoryRegionMap::FindRegion(uintptr_t addr, Region* result) {
+  Lock();
+  const Region* region = DoFindRegionLocked(addr);
+  if (region != NULL) *result = *region;  // create it as an independent copy
+  Unlock();
+  return region != NULL;
+}
+
+bool MemoryRegionMap::FindAndMarkStackRegion(uintptr_t stack_top,
+                                             Region* result) {
+  Lock();
+  const Region* region = DoFindRegionLocked(stack_top);
+  if (region != NULL) {
+    RAW_VLOG(10, "Stack at %p is inside region %p..%p",
+                reinterpret_cast<void*>(stack_top),
+                reinterpret_cast<void*>(region->start_addr),
+                reinterpret_cast<void*>(region->end_addr));
+    const_cast<Region*>(region)->set_is_stack();  // now we know
+      // cast is safe (set_is_stack does not change the set ordering key)
+    *result = *region;  // create *result as an independent copy
+  }
+  Unlock();
+  return region != NULL;
+}
+
+MemoryRegionMap::RegionIterator MemoryRegionMap::BeginRegionLocked() {
+  RAW_CHECK(LockIsHeld(), "should be held (by this thread)");
+  RAW_CHECK(regions_ != NULL, "");
+  return regions_->begin();
+}
+
+MemoryRegionMap::RegionIterator MemoryRegionMap::EndRegionLocked() {
+  RAW_CHECK(LockIsHeld(), "should be held (by this thread)");
+  RAW_CHECK(regions_ != NULL, "");
+  return regions_->end();
+}
+
+inline void MemoryRegionMap::DoInsertRegionLocked(const Region& region) {
+  RAW_VLOG(12, "Inserting region %p..%p from %p",
+              reinterpret_cast<void*>(region.start_addr),
+              reinterpret_cast<void*>(region.end_addr),
+              reinterpret_cast<void*>(region.caller()));
+  RegionSet::const_iterator i = regions_->lower_bound(region);
+  if (i != regions_->end() && i->start_addr <= region.start_addr) {
+    RAW_DCHECK(region.end_addr <= i->end_addr, "");  // lower_bound ensures this
+    return;  // 'region' is a subset of an already recorded region; do nothing
+    // We can be stricter and allow this only when *i has been created via
+    // an mmap with MAP_NORESERVE flag set.
+  }
+  if (DEBUG_MODE) {
+    RAW_CHECK(i == regions_->end()  ||  !region.Overlaps(*i),
+              "Wow, overlapping memory regions");
+    Region sample;
+    sample.SetRegionSetKey(region.start_addr);
+    i = regions_->lower_bound(sample);
+    RAW_CHECK(i == regions_->end()  ||  !region.Overlaps(*i),
+              "Wow, overlapping memory regions");
+  }
+  region.AssertIsConsistent();  // just making sure
+  // This inserts and allocates permanent storage for region
+  // and its call stack data: it's safe to do it now:
+  regions_->insert(region);
+  RAW_VLOG(12, "Inserted region %p..%p :",
+              reinterpret_cast<void*>(region.start_addr),
+              reinterpret_cast<void*>(region.end_addr));
+  if (VLOG_IS_ON(12))  LogAllLocked();
+}
+
+// These variables are local to MemoryRegionMap::InsertRegionLocked()
+// and MemoryRegionMap::HandleSavedRegionsLocked()
+// and are file-level to ensure that they are initialized at load time.
+
+// Number of unprocessed region inserts.
+static int saved_regions_count = 0;
+
+// Unprocessed inserts (must be big enough to hold all allocations that can
+// be caused by a InsertRegionLocked call).
+// Region has no constructor, so that c-tor execution does not interfere
+// with the any-time use of the static memory behind saved_regions.
+static MemoryRegionMap::Region saved_regions[20];
+
+inline void MemoryRegionMap::HandleSavedRegionsLocked(
+              void (*insert_func)(const Region& region)) {
+  while (saved_regions_count > 0) {
+    // Making a local-var copy of the region argument to insert_func
+    // including its stack (w/o doing any memory allocations) is important:
+    // in many cases the memory in saved_regions
+    // will get written-to during the (*insert_func)(r) call below.
+    Region r = saved_regions[--saved_regions_count];
+    (*insert_func)(r);
+  }
+}
+
+inline void MemoryRegionMap::InsertRegionLocked(const Region& region) {
+  RAW_CHECK(LockIsHeld(), "should be held (by this thread)");
+  // We can be called recursively, because RegionSet constructor
+  // and DoInsertRegionLocked() (called below) can call the allocator.
+  // recursive_insert tells us if that's the case. When this happens,
+  // region insertion information is recorded in saved_regions[],
+  // and taken into account when the recursion unwinds.
+  // Do the insert:
+  if (recursive_insert) {  // recursion: save in saved_regions
+    RAW_VLOG(12, "Saving recursive insert of region %p..%p from %p",
+                reinterpret_cast<void*>(region.start_addr),
+                reinterpret_cast<void*>(region.end_addr),
+                reinterpret_cast<void*>(region.caller()));
+    RAW_CHECK(saved_regions_count < arraysize(saved_regions), "");
+    // Copy 'region' to saved_regions[saved_regions_count]
+    // together with the contents of its call_stack,
+    // then increment saved_regions_count.
+    saved_regions[saved_regions_count++] = region;
+  } else {  // not a recusrive call
+    if (regions_ == NULL) {  // init regions_
+      RAW_VLOG(12, "Initializing region set");
+      regions_ = regions_rep.region_set();
+      recursive_insert = true;
+      new(regions_) RegionSet();
+      HandleSavedRegionsLocked(&DoInsertRegionLocked);
+      recursive_insert = false;
+    }
+    recursive_insert = true;
+    // Do the actual insertion work to put new regions into regions_:
+    DoInsertRegionLocked(region);
+    HandleSavedRegionsLocked(&DoInsertRegionLocked);
+    recursive_insert = false;
+  }
+}
+
+// We strip out different number of stack frames in debug mode
+// because less inlining happens in that case
+#ifdef NDEBUG
+static const int kStripFrames = 1;
+#else
+static const int kStripFrames = 3;
+#endif
+
+void MemoryRegionMap::RecordRegionAddition(const void* start, size_t size) {
+  // Record start/end info about this memory acquisition call in a new region:
+  Region region;
+  region.Create(start, size);
+  // First get the call stack info into the local varible 'region':
+  const int depth =
+    max_stack_depth_ > 0
+    ? MallocHook::GetCallerStackTrace(const_cast<void**>(region.call_stack),
+                                      max_stack_depth_, kStripFrames + 1)
+    : 0;
+  region.set_call_stack_depth(depth);  // record stack info fully
+  RAW_VLOG(10, "New global region %p..%p from %p",
+              reinterpret_cast<void*>(region.start_addr),
+              reinterpret_cast<void*>(region.end_addr),
+              reinterpret_cast<void*>(region.caller()));
+  // Note: none of the above allocates memory.
+  Lock();  // recursively lock
+  map_size_ += size;
+  InsertRegionLocked(region);
+    // This will (eventually) allocate storage for and copy over the stack data
+    // from region.call_stack_data_ that is pointed by region.call_stack().
+  Unlock();
+}
+
+void MemoryRegionMap::RecordRegionRemoval(const void* start, size_t size) {
+  Lock();
+  if (recursive_insert) {
+    // First remove the removed region from saved_regions, if it's
+    // there, to prevent overrunning saved_regions in recursive
+    // map/unmap call sequences, and also from later inserting regions
+    // which have already been unmapped.
+    uintptr_t start_addr = reinterpret_cast<uintptr_t>(start);
+    uintptr_t end_addr = start_addr + size;
+    int put_pos = 0;
+    int old_count = saved_regions_count;
+    for (int i = 0; i < old_count; ++i, ++put_pos) {
+      Region& r = saved_regions[i];
+      if (r.start_addr == start_addr && r.end_addr == end_addr) {
+        // An exact match, so it's safe to remove.
+        --saved_regions_count;
+        --put_pos;
+        RAW_VLOG(10, ("Insta-Removing saved region %p..%p; "
+                     "now have %d saved regions"),
+                 reinterpret_cast<void*>(start_addr),
+                 reinterpret_cast<void*>(end_addr),
+                 saved_regions_count);
+      } else {
+        if (put_pos < i) {
+          saved_regions[put_pos] = saved_regions[i];
+        }
+      }
+    }
+  }
+  if (regions_ == NULL) {  // We must have just unset the hooks,
+                           // but this thread was already inside the hook.
+    Unlock();
+    return;
+  }
+  if (!recursive_insert) {
+    HandleSavedRegionsLocked(&InsertRegionLocked);
+  }
+    // first handle adding saved regions if any
+  uintptr_t start_addr = reinterpret_cast<uintptr_t>(start);
+  uintptr_t end_addr = start_addr + size;
+  // subtract start_addr, end_addr from all the regions
+  RAW_VLOG(10, "Removing global region %p..%p; have %" PRIuS " regions",
+              reinterpret_cast<void*>(start_addr),
+              reinterpret_cast<void*>(end_addr),
+              regions_->size());
+  Region sample;
+  sample.SetRegionSetKey(start_addr);
+  // Only iterate over the regions that might overlap start_addr..end_addr:
+  for (RegionSet::iterator region = regions_->lower_bound(sample);
+       region != regions_->end()  &&  region->start_addr < end_addr;
+       /*noop*/) {
+    RAW_VLOG(13, "Looking at region %p..%p",
+                reinterpret_cast<void*>(region->start_addr),
+                reinterpret_cast<void*>(region->end_addr));
+    if (start_addr <= region->start_addr  &&
+        region->end_addr <= end_addr) {  // full deletion
+      RAW_VLOG(12, "Deleting region %p..%p",
+                  reinterpret_cast<void*>(region->start_addr),
+                  reinterpret_cast<void*>(region->end_addr));
+      RegionSet::iterator d = region;
+      ++region;
+      regions_->erase(d);
+      continue;
+    } else if (region->start_addr < start_addr  &&
+               end_addr < region->end_addr) {  // cutting-out split
+      RAW_VLOG(12, "Splitting region %p..%p in two",
+                  reinterpret_cast<void*>(region->start_addr),
+                  reinterpret_cast<void*>(region->end_addr));
+      // Make another region for the start portion:
+      // The new region has to be the start portion because we can't
+      // just modify region->end_addr as it's the sorting key.
+      Region r = *region;
+      r.set_end_addr(start_addr);
+      InsertRegionLocked(r);
+      // cut *region from start:
+      const_cast<Region&>(*region).set_start_addr(end_addr);
+    } else if (end_addr > region->start_addr  &&
+               start_addr <= region->start_addr) {  // cut from start
+      RAW_VLOG(12, "Start-chopping region %p..%p",
+                  reinterpret_cast<void*>(region->start_addr),
+                  reinterpret_cast<void*>(region->end_addr));
+      const_cast<Region&>(*region).set_start_addr(end_addr);
+    } else if (start_addr > region->start_addr  &&
+               start_addr < region->end_addr) {  // cut from end
+      RAW_VLOG(12, "End-chopping region %p..%p",
+                  reinterpret_cast<void*>(region->start_addr),
+                  reinterpret_cast<void*>(region->end_addr));
+      // Can't just modify region->end_addr (it's the sorting key):
+      Region r = *region;
+      r.set_end_addr(start_addr);
+      RegionSet::iterator d = region;
+      ++region;
+      // It's safe to erase before inserting since r is independent of *d:
+      // r contains an own copy of the call stack:
+      regions_->erase(d);
+      InsertRegionLocked(r);
+      continue;
+    }
+    ++region;
+  }
+  RAW_VLOG(12, "Removed region %p..%p; have %" PRIuS " regions",
+              reinterpret_cast<void*>(start_addr),
+              reinterpret_cast<void*>(end_addr),
+              regions_->size());
+  if (VLOG_IS_ON(12))  LogAllLocked();
+  unmap_size_ += size;
+  Unlock();
+}
+
+void MemoryRegionMap::MmapHook(const void* result,
+                               const void* start, size_t size,
+                               int prot, int flags,
+                               int fd, off_t offset) {
+  // TODO(maxim): replace all 0x%" PRIxS " by %p when RAW_VLOG uses a safe
+  // snprintf reimplementation that does not malloc to pretty-print NULL
+  RAW_VLOG(10, "MMap = 0x%" PRIxPTR " of %" PRIuS " at %" PRIu64 " "
+              "prot %d flags %d fd %d offs %" PRId64,
+              reinterpret_cast<uintptr_t>(result), size,
+              reinterpret_cast<uint64>(start), prot, flags, fd,
+              static_cast<int64>(offset));
+  if (result != reinterpret_cast<void*>(MAP_FAILED)  &&  size != 0) {
+    RecordRegionAddition(result, size);
+  }
+}
+
+void MemoryRegionMap::MunmapHook(const void* ptr, size_t size) {
+  RAW_VLOG(10, "MUnmap of %p %" PRIuS "", ptr, size);
+  if (size != 0) {
+    RecordRegionRemoval(ptr, size);
+  }
+}
+
+void MemoryRegionMap::MremapHook(const void* result,
+                                 const void* old_addr, size_t old_size,
+                                 size_t new_size, int flags,
+                                 const void* new_addr) {
+  RAW_VLOG(10, "MRemap = 0x%" PRIxPTR " of 0x%" PRIxPTR " %" PRIuS " "
+              "to %" PRIuS " flags %d new_addr=0x%" PRIxPTR,
+              (uintptr_t)result, (uintptr_t)old_addr,
+               old_size, new_size, flags,
+               flags & MREMAP_FIXED ? (uintptr_t)new_addr : 0);
+  if (result != reinterpret_cast<void*>(-1)) {
+    RecordRegionRemoval(old_addr, old_size);
+    RecordRegionAddition(result, new_size);
+  }
+}
+
+extern "C" void* __sbrk(ptrdiff_t increment);  // defined in libc
+
+void MemoryRegionMap::SbrkHook(const void* result, ptrdiff_t increment) {
+  RAW_VLOG(10, "Sbrk = 0x%" PRIxPTR " of %" PRIdS "", (uintptr_t)result, increment);
+  if (result != reinterpret_cast<void*>(-1)) {
+    if (increment > 0) {
+      void* new_end = sbrk(0);
+      RecordRegionAddition(result, reinterpret_cast<uintptr_t>(new_end) -
+                                   reinterpret_cast<uintptr_t>(result));
+    } else if (increment < 0) {
+      void* new_end = sbrk(0);
+      RecordRegionRemoval(new_end, reinterpret_cast<uintptr_t>(result) -
+                                   reinterpret_cast<uintptr_t>(new_end));
+    }
+  }
+}
+
+void MemoryRegionMap::LogAllLocked() {
+  RAW_CHECK(LockIsHeld(), "should be held (by this thread)");
+  RAW_LOG(INFO, "List of regions:");
+  uintptr_t previous = 0;
+  for (RegionSet::const_iterator r = regions_->begin();
+       r != regions_->end(); ++r) {
+    RAW_LOG(INFO, "Memory region 0x%" PRIxPTR "..0x%" PRIxPTR " "
+                  "from 0x%" PRIxPTR " stack=%d",
+                  r->start_addr, r->end_addr, r->caller(), r->is_stack);
+    RAW_CHECK(previous < r->end_addr, "wow, we messed up the set order");
+      // this must be caused by uncontrolled recursive operations on regions_
+    previous = r->end_addr;
+  }
+  RAW_LOG(INFO, "End of regions list");
+}
+
+#endif

BeefRT/gperftools/src/memory_region_map.h

@@ -0,0 +1,340 @@
+/* Copyright (c) 2006, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Maxim Lifantsev
+ */
+
+#ifndef BASE_MEMORY_REGION_MAP_H_
+#define BASE_MEMORY_REGION_MAP_H_
+
+#include <config.h>
+
+#ifdef HAVE_PTHREAD
+#include <pthread.h>
+#endif
+#include <stddef.h>
+#include <set>
+#include "base/stl_allocator.h"
+#include "base/spinlock.h"
+#include "base/thread_annotations.h"
+#include "base/low_level_alloc.h"
+
+// TODO(maxim): add a unittest:
+//  execute a bunch of mmaps and compare memory map what strace logs
+//  execute a bunch of mmap/munmup and compare memory map with
+//  own accounting of what those mmaps generated
+
+// Thread-safe class to collect and query the map of all memory regions
+// in a process that have been created with mmap, munmap, mremap, sbrk.
+// For each memory region, we keep track of (and provide to users)
+// the stack trace that allocated that memory region.
+// The recorded stack trace depth is bounded by
+// a user-supplied max_stack_depth parameter of Init().
+// After initialization with Init()
+// (which can happened even before global object constructor execution)
+// we collect the map by installing and monitoring MallocHook-s
+// to mmap, munmap, mremap, sbrk.
+// At any time one can query this map via provided interface.
+// For more details on the design of MemoryRegionMap
+// see the comment at the top of our .cc file.
+class MemoryRegionMap {
+ private:
+  // Max call stack recording depth supported by Init().  Set it to be
+  // high enough for all our clients.  Note: we do not define storage
+  // for this (doing that requires special handling in windows), so
+  // don't take the address of it!
+  static const int kMaxStackDepth = 32;
+
+ public:
+  // interface ================================================================
+
+  // Every client of MemoryRegionMap must call Init() before first use,
+  // and Shutdown() after last use.  This allows us to reference count
+  // this (singleton) class properly.  MemoryRegionMap assumes it's the
+  // only client of MallocHooks, so a client can only register other
+  // MallocHooks after calling Init() and must unregister them before
+  // calling Shutdown().
+
+  // Initialize this module to record memory allocation stack traces.
+  // Stack traces that have more than "max_stack_depth" frames
+  // are automatically shrunk to "max_stack_depth" when they are recorded.
+  // Init() can be called more than once w/o harm, largest max_stack_depth
+  // will be the effective one.
+  // It will install mmap, munmap, mremap, sbrk hooks
+  // and initialize arena_ and our hook and locks, hence one can use
+  // MemoryRegionMap::Lock()/Unlock() to manage the locks.
+  // Uses Lock/Unlock inside.
+  static void Init(int max_stack_depth);
+
+  // Try to shutdown this module undoing what Init() did.
+  // Returns true iff could do full shutdown (or it was not attempted).
+  // Full shutdown is attempted when the number of Shutdown() calls equals
+  // the number of Init() calls.
+  static bool Shutdown();
+
+  // Locks to protect our internal data structures.
+  // These also protect use of arena_ if our Init() has been done.
+  // The lock is recursive.
+  static void Lock() EXCLUSIVE_LOCK_FUNCTION(lock_);
+  static void Unlock() UNLOCK_FUNCTION(lock_);
+
+  // Returns true when the lock is held by this thread (for use in RAW_CHECK-s).
+  static bool LockIsHeld();
+
+  // Locker object that acquires the MemoryRegionMap::Lock
+  // for the duration of its lifetime (a C++ scope).
+  class LockHolder {
+   public:
+    LockHolder() { Lock(); }
+    ~LockHolder() { Unlock(); }
+   private:
+    DISALLOW_COPY_AND_ASSIGN(LockHolder);
+  };
+
+  // A memory region that we know about through malloc_hook-s.
+  // This is essentially an interface through which MemoryRegionMap
+  // exports the collected data to its clients.  Thread-compatible.
+  struct Region {
+    uintptr_t start_addr;  // region start address
+    uintptr_t end_addr;  // region end address
+    int call_stack_depth;  // number of caller stack frames that we saved
+    const void* call_stack[kMaxStackDepth];  // caller address stack array
+                                             // filled to call_stack_depth size
+    bool is_stack;  // does this region contain a thread's stack:
+                    // a user of MemoryRegionMap supplies this info
+
+    // Convenience accessor for call_stack[0],
+    // i.e. (the program counter of) the immediate caller
+    // of this region's allocation function,
+    // but it also returns NULL when call_stack_depth is 0,
+    // i.e whe we weren't able to get the call stack.
+    // This usually happens in recursive calls, when the stack-unwinder
+    // calls mmap() which in turn calls the stack-unwinder.
+    uintptr_t caller() const {
+      return reinterpret_cast<uintptr_t>(call_stack_depth >= 1
+                                         ? call_stack[0] : NULL);
+    }
+
+    // Return true iff this region overlaps region x.
+    bool Overlaps(const Region& x) const {
+      return start_addr < x.end_addr  &&  end_addr > x.start_addr;
+    }
+
+   private:  // helpers for MemoryRegionMap
+    friend class MemoryRegionMap;
+
+    // The ways we create Region-s:
+    void Create(const void* start, size_t size) {
+      start_addr = reinterpret_cast<uintptr_t>(start);
+      end_addr = start_addr + size;
+      is_stack = false;  // not a stack till marked such
+      call_stack_depth = 0;
+      AssertIsConsistent();
+    }
+    void set_call_stack_depth(int depth) {
+      RAW_DCHECK(call_stack_depth == 0, "");  // only one such set is allowed
+      call_stack_depth = depth;
+      AssertIsConsistent();
+    }
+
+    // The ways we modify Region-s:
+    void set_is_stack() { is_stack = true; }
+    void set_start_addr(uintptr_t addr) {
+      start_addr = addr;
+      AssertIsConsistent();
+    }
+    void set_end_addr(uintptr_t addr) {
+      end_addr = addr;
+      AssertIsConsistent();
+    }
+
+    // Verifies that *this contains consistent data, crashes if not the case.
+    void AssertIsConsistent() const {
+      RAW_DCHECK(start_addr < end_addr, "");
+      RAW_DCHECK(call_stack_depth >= 0  &&
+                 call_stack_depth <= kMaxStackDepth, "");
+    }
+
+    // Post-default construction helper to make a Region suitable
+    // for searching in RegionSet regions_.
+    void SetRegionSetKey(uintptr_t addr) {
+      // make sure *this has no usable data:
+      if (DEBUG_MODE) memset(this, 0xFF, sizeof(*this));
+      end_addr = addr;
+    }
+
+    // Note: call_stack[kMaxStackDepth] as a member lets us make Region
+    // a simple self-contained struct with correctly behaving bit-vise copying.
+    // This simplifies the code of this module but wastes some memory:
+    // in most-often use case of this module (leak checking)
+    // only one call_stack element out of kMaxStackDepth is actually needed.
+    // Making the storage for call_stack variable-sized,
+    // substantially complicates memory management for the Region-s:
+    // as they need to be created and manipulated for some time
+    // w/o any memory allocations, yet are also given out to the users.
+  };
+
+  // Find the region that covers addr and write its data into *result if found,
+  // in which case *result gets filled so that it stays fully functional
+  // even when the underlying region gets removed from MemoryRegionMap.
+  // Returns success. Uses Lock/Unlock inside.
+  static bool FindRegion(uintptr_t addr, Region* result);
+
+  // Find the region that contains stack_top, mark that region as
+  // a stack region, and write its data into *result if found,
+  // in which case *result gets filled so that it stays fully functional
+  // even when the underlying region gets removed from MemoryRegionMap.
+  // Returns success. Uses Lock/Unlock inside.
+  static bool FindAndMarkStackRegion(uintptr_t stack_top, Region* result);
+
+ private:  // our internal types ==============================================
+
+  // Region comparator for sorting with STL
+  struct RegionCmp {
+    bool operator()(const Region& x, const Region& y) const {
+      return x.end_addr < y.end_addr;
+    }
+  };
+
+  // We allocate STL objects in our own arena.
+  struct MyAllocator {
+    static void *Allocate(size_t n) {
+      return LowLevelAlloc::AllocWithArena(n, arena_);
+    }
+    static void Free(const void *p, size_t /* n */) {
+      LowLevelAlloc::Free(const_cast<void*>(p));
+    }
+  };
+
+  // Set of the memory regions
+  typedef std::set<Region, RegionCmp,
+              STL_Allocator<Region, MyAllocator> > RegionSet;
+
+ public:  // more in-depth interface ==========================================
+
+  // STL iterator with values of Region
+  typedef RegionSet::const_iterator RegionIterator;
+
+  // Return the begin/end iterators to all the regions.
+  // These need Lock/Unlock protection around their whole usage (loop).
+  // Even when the same thread causes modifications during such a loop
+  // (which are permitted due to recursive locking)
+  // the loop iterator will still be valid as long as its region
+  // has not been deleted, but EndRegionLocked should be
+  // re-evaluated whenever the set of regions has changed.
+  static RegionIterator BeginRegionLocked();
+  static RegionIterator EndRegionLocked();
+
+  // Return the accumulated sizes of mapped and unmapped regions.
+  static int64 MapSize() { return map_size_; }
+  static int64 UnmapSize() { return unmap_size_; }
+
+  // Effectively private type from our .cc =================================
+  // public to let us declare global objects:
+  union RegionSetRep;
+
+ private:
+  // representation ===========================================================
+
+  // Counter of clients of this module that have called Init().
+  static int client_count_;
+
+  // Maximal number of caller stack frames to save (>= 0).
+  static int max_stack_depth_;
+
+  // Arena used for our allocations in regions_.
+  static LowLevelAlloc::Arena* arena_;
+
+  // Set of the mmap/sbrk/mremap-ed memory regions
+  // To be accessed *only* when Lock() is held.
+  // Hence we protect the non-recursive lock used inside of arena_
+  // with our recursive Lock(). This lets a user prevent deadlocks
+  // when threads are stopped by ListAllProcessThreads at random spots
+  // simply by acquiring our recursive Lock() before that.
+  static RegionSet* regions_;
+
+  // Lock to protect regions_ variable and the data behind.
+  static SpinLock lock_;
+  // Lock to protect the recursive lock itself.
+  static SpinLock owner_lock_;
+
+  // Recursion count for the recursive lock.
+  static int recursion_count_;
+  // The thread id of the thread that's inside the recursive lock.
+  static pthread_t lock_owner_tid_;
+
+  // Total size of all mapped pages so far
+  static int64 map_size_;
+  // Total size of all unmapped pages so far
+  static int64 unmap_size_;
+
+  // helpers ==================================================================
+
+  // Helper for FindRegion and FindAndMarkStackRegion:
+  // returns the region covering 'addr' or NULL; assumes our lock_ is held.
+  static const Region* DoFindRegionLocked(uintptr_t addr);
+
+  // Verifying wrapper around regions_->insert(region)
+  // To be called to do InsertRegionLocked's work only!
+  inline static void DoInsertRegionLocked(const Region& region);
+  // Handle regions saved by InsertRegionLocked into a tmp static array
+  // by calling insert_func on them.
+  inline static void HandleSavedRegionsLocked(
+                       void (*insert_func)(const Region& region));
+  // Wrapper around DoInsertRegionLocked
+  // that handles the case of recursive allocator calls.
+  inline static void InsertRegionLocked(const Region& region);
+
+  // Record addition of a memory region at address "start" of size "size"
+  // (called from our mmap/mremap/sbrk hooks).
+  static void RecordRegionAddition(const void* start, size_t size);
+  // Record deletion of a memory region at address "start" of size "size"
+  // (called from our munmap/mremap/sbrk hooks).
+  static void RecordRegionRemoval(const void* start, size_t size);
+
+  // Hooks for MallocHook
+  static void MmapHook(const void* result,
+                       const void* start, size_t size,
+                       int prot, int flags,
+                       int fd, off_t offset);
+  static void MunmapHook(const void* ptr, size_t size);
+  static void MremapHook(const void* result, const void* old_addr,
+                         size_t old_size, size_t new_size, int flags,
+                         const void* new_addr);
+  static void SbrkHook(const void* result, ptrdiff_t increment);
+
+  // Log all memory regions; Useful for debugging only.
+  // Assumes Lock() is held
+  static void LogAllLocked();
+
+  DISALLOW_COPY_AND_ASSIGN(MemoryRegionMap);
+};
+
+#endif  // BASE_MEMORY_REGION_MAP_H_

BeefRT/gperftools/src/osx/config.h

@@ -0,0 +1,291 @@
+/* src/config.h.  Generated from config.h.in by configure.  */
+/* src/config.h.in.  Generated from configure.ac by autoheader.  */
+
+#define NO_HEAP_CHECK
+
+
+#ifndef GPERFTOOLS_CONFIG_H_
+#define GPERFTOOLS_CONFIG_H_
+
+
+/* Define to 1 if compiler supports __builtin_expect */
+#define HAVE_BUILTIN_EXPECT 1
+
+/* Define to 1 if compiler supports __builtin_stack_pointer */
+/* #undef HAVE_BUILTIN_STACK_POINTER */
+
+/* Define to 1 if you have the <conflict-signal.h> header file. */
+/* #undef HAVE_CONFLICT_SIGNAL_H */
+
+/* Define to 1 if you have the <cygwin/signal.h> header file. */
+/* #undef HAVE_CYGWIN_SIGNAL_H */
+
+/* Define to 1 if you have the declaration of `cfree', and to 0 if you don't.
+   */
+#define HAVE_DECL_CFREE 0
+
+/* Define to 1 if you have the declaration of `memalign', and to 0 if you
+   don't. */
+#define HAVE_DECL_MEMALIGN 0
+
+/* Define to 1 if you have the declaration of `nanosleep', and to 0 if you
+   don't. */
+/* #undef HAVE_DECL_NANOSLEEP */
+
+/* Define to 1 if you have the declaration of `posix_memalign', and to 0 if
+   you don't. */
+#define HAVE_DECL_POSIX_MEMALIGN 0
+
+/* Define to 1 if you have the declaration of `pvalloc', and to 0 if you
+   don't. */
+#define HAVE_DECL_PVALLOC 0
+
+/* Define to 1 if you have the declaration of `sleep', and to 0 if you don't.
+   */
+/* #undef HAVE_DECL_SLEEP */
+
+/* Define to 1 if you have the declaration of `uname', and to 0 if you don't.
+   */
+#define HAVE_DECL_UNAME 1
+
+/* Define to 1 if you have the declaration of `valloc', and to 0 if you don't.
+   */
+#define HAVE_DECL_VALLOC 0
+
+/* Define to 1 if you have the <dlfcn.h> header file. */
+#define HAVE_DLFCN_H 1
+
+/* Define to 1 if the system has the type `Elf32_Versym'. */
+/* #undef HAVE_ELF32_VERSYM */
+
+/* Define to 1 if you have the <execinfo.h> header file. */
+#define HAVE_EXECINFO_H 1
+
+/* Define to 1 if you have the <fcntl.h> header file. */
+#define HAVE_FCNTL_H 1
+
+/* Define to 1 if you have the <features.h> header file. */
+/* #undef HAVE_FEATURES_H */
+
+/* Define to 1 if you have the `fork' function. */
+#define HAVE_FORK 1
+
+/* Define to 1 if you have the `geteuid' function. */
+#define HAVE_GETEUID 1
+
+/* Define to 1 if you have the `getpagesize' function. */
+#define HAVE_GETPAGESIZE 1
+
+/* Define to 1 if you have the <glob.h> header file. */
+#define HAVE_GLOB_H 1
+
+/* Define to 1 if you have the <grp.h> header file. */
+#define HAVE_GRP_H 1
+
+/* Define to 1 if you have the <inttypes.h> header file. */
+#define HAVE_INTTYPES_H 1
+
+/* Define to 1 if you have the <libunwind.h> header file. */
+#define HAVE_LIBUNWIND_H 1
+
+/* Define to 1 if you have the <linux/ptrace.h> header file. */
+/* #undef HAVE_LINUX_PTRACE_H */
+
+/* Define to 1 if you have the <malloc.h> header file. */
+/* #undef HAVE_MALLOC_H */
+
+/* Define to 1 if you have the <malloc/malloc.h> header file. */
+#define HAVE_MALLOC_MALLOC_H 1
+
+/* Define to 1 if you have the <memory.h> header file. */
+#define HAVE_MEMORY_H 1
+
+/* Define to 1 if you have a working `mmap' system call. */
+#define HAVE_MMAP 1
+
+/* define if the compiler implements namespaces */
+#define HAVE_NAMESPACES 1
+
+/* Define to 1 if you have the <poll.h> header file. */
+#define HAVE_POLL_H 1
+
+/* define if libc has program_invocation_name */
+/* #undef HAVE_PROGRAM_INVOCATION_NAME */
+
+/* Define if you have POSIX threads libraries and header files. */
+#define HAVE_PTHREAD 1
+
+/* Define to 1 if you have the <pwd.h> header file. */
+#define HAVE_PWD_H 1
+
+/* Define to 1 if you have the `sbrk' function. */
+#define HAVE_SBRK 1
+
+/* Define to 1 if you have the <sched.h> header file. */
+#define HAVE_SCHED_H 1
+
+/* Define to 1 if you have the <stdint.h> header file. */
+#define HAVE_STDINT_H 1
+
+/* Define to 1 if you have the <stdlib.h> header file. */
+#define HAVE_STDLIB_H 1
+
+/* Define to 1 if you have the <strings.h> header file. */
+#define HAVE_STRINGS_H 1
+
+/* Define to 1 if you have the <string.h> header file. */
+#define HAVE_STRING_H 1
+
+/* Define to 1 if the system has the type `struct mallinfo'. */
+/* #undef HAVE_STRUCT_MALLINFO */
+
+/* Define to 1 if you have the <sys/cdefs.h> header file. */
+#define HAVE_SYS_CDEFS_H 1
+
+/* Define to 1 if you have the <sys/malloc.h> header file. */
+#define HAVE_SYS_MALLOC_H 1
+
+/* Define to 1 if you have the <sys/param.h> header file. */
+#define HAVE_SYS_PARAM_H 1
+
+/* Define to 1 if you have the <sys/prctl.h> header file. */
+/* #undef HAVE_SYS_PRCTL_H */
+
+/* Define to 1 if you have the <sys/resource.h> header file. */
+#define HAVE_SYS_RESOURCE_H 1
+
+/* Define to 1 if you have the <sys/socket.h> header file. */
+#define HAVE_SYS_SOCKET_H 1
+
+/* Define to 1 if you have the <sys/stat.h> header file. */
+#define HAVE_SYS_STAT_H 1
+
+/* Define to 1 if you have the <sys/syscall.h> header file. */
+#define HAVE_SYS_SYSCALL_H 1
+
+/* Define to 1 if you have the <sys/types.h> header file. */
+#define HAVE_SYS_TYPES_H 1
+
+/* <sys/ucontext.h> is broken on redhat 7 */
+#define HAVE_SYS_UCONTEXT_H 1
+
+/* Define to 1 if you have the <sys/wait.h> header file. */
+#define HAVE_SYS_WAIT_H 1
+
+/* Define to 1 if compiler supports __thread */
+#define HAVE_TLS 1
+
+/* Define to 1 if you have the <ucontext.h> header file. */
+/* #undef HAVE_UCONTEXT_H */
+
+/* Define to 1 if you have the <unistd.h> header file. */
+#define HAVE_UNISTD_H 1
+
+/* Define to 1 if you have the <unwind.h> header file. */
+#define HAVE_UNWIND_H 1
+
+/* Define to 1 if you have the <valgrind.h> header file. */
+/* #undef HAVE_VALGRIND_H */
+
+/* define if your compiler has __attribute__ */
+#define HAVE___ATTRIBUTE__ 1
+
+/* Define to 1 if compiler supports __environ */
+/* #undef HAVE___ENVIRON */
+
+/* Define to 1 if the system has the type `__int64'. */
+/* #undef HAVE___INT64 */
+
+/* prefix where we look for installed files */
+#define INSTALL_PREFIX "/usr/local"
+
+/* Define to 1 if int32_t is equivalent to intptr_t */
+/* #undef INT32_EQUALS_INTPTR */
+
+/* Define to the sub-directory in which libtool stores uninstalled libraries.
+   */
+#define LT_OBJDIR ".libs/"
+
+/* Define to 'volatile' if __malloc_hook is declared volatile */
+#define MALLOC_HOOK_MAYBE_VOLATILE /**/
+
+/* Define to 1 if your C compiler doesn't accept -c and -o together. */
+/* #undef NO_MINUS_C_MINUS_O */
+
+/* Name of package */
+#define PACKAGE "gperftools"
+
+/* Define to the address where bug reports for this package should be sent. */
+#define PACKAGE_BUGREPORT "google-perftools@googlegroups.com"
+
+/* Define to the full name of this package. */
+#define PACKAGE_NAME "gperftools"
+
+/* Define to the full name and version of this package. */
+#define PACKAGE_STRING "gperftools 2.1"
+
+/* Define to the one symbol short name of this package. */
+#define PACKAGE_TARNAME "gperftools"
+
+/* Define to the home page for this package. */
+#define PACKAGE_URL ""
+
+/* Define to the version of this package. */
+#define PACKAGE_VERSION "2.1"
+
+/* How to access the PC from a struct ucontext */
+#define PC_FROM_UCONTEXT uc_mcontext->__ss.__rip
+
+/* Always the empty-string on non-windows systems. On windows, should be
+   "__declspec(dllexport)". This way, when we compile the dll, we export our
+   functions/classes. It's safe to define this here because config.h is only
+   used internally, to compile the DLL, and every DLL source file #includes
+   "config.h" before anything else. */
+#define PERFTOOLS_DLL_DECL /**/
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIdS "ld"
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIuS "lu"
+
+/* printf format code for printing a size_t and ssize_t */
+#define PRIxS "lx"
+
+/* Mark the systems where we know it's bad if pthreads runs too
+   early before main (before threads are initialized, presumably).  */
+#ifdef __FreeBSD__
+#define PTHREADS_CRASHES_IF_RUN_TOO_EARLY 1
+#endif
+
+/* Define to necessary symbol if this constant uses a non-standard name on
+   your system. */
+/* #undef PTHREAD_CREATE_JOINABLE */
+
+/* Define to 1 if you have the ANSI C header files. */
+#define STDC_HEADERS 1
+
+/* the namespace where STL code like vector<> is defined */
+#define STL_NAMESPACE std
+
+/* Version number of package */
+#define VERSION "2.1"
+
+/* C99 says: define this to get the PRI... macros from stdint.h */
+#ifndef __STDC_FORMAT_MACROS
+# define __STDC_FORMAT_MACROS 1
+#endif
+
+/* Define to `__inline__' or `__inline' if that's what the C compiler
+   calls it, or to nothing if 'inline' is not supported under any name.  */
+#ifndef __cplusplus
+/* #undef inline */
+#endif
+
+
+#ifdef __MINGW32__
+#include "windows/mingw.h"
+#endif
+
+#endif  /* #ifndef GPERFTOOLS_CONFIG_H_ */
+

BeefRT/gperftools/src/osx/gperftools/tcmalloc.h

@@ -0,0 +1,133 @@
+/* Copyright (c) 2003, Google Inc.
+ * All rights reserved.
+ * 
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ * 
+ *     * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above
+ * copyright notice, this list of conditions and the following disclaimer
+ * in the documentation and/or other materials provided with the
+ * distribution.
+ *     * Neither the name of Google Inc. nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ---
+ * Author: Sanjay Ghemawat <opensource@google.com>
+ *         .h file by Craig Silverstein <opensource@google.com>
+ */
+
+#ifndef TCMALLOC_TCMALLOC_H_
+#define TCMALLOC_TCMALLOC_H_
+
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_SYS_CDEFS_H
+#include <sys/cdefs.h>   // where glibc defines __THROW
+#endif
+
+// __THROW is defined in glibc systems.  It means, counter-intuitively,
+// "This function will never throw an exception."  It's an optional
+// optimization tool, but we may need to use it to match glibc prototypes.
+#ifndef __THROW    /* I guess we're not on a glibc system */
+# define __THROW   /* __THROW is just an optimization, so ok to make it "" */
+#endif
+
+// Define the version number so folks can check against it
+#define TC_VERSION_MAJOR  2
+#define TC_VERSION_MINOR  1
+#define TC_VERSION_PATCH  ""
+#define TC_VERSION_STRING "gperftools 2.1"
+
+// For struct mallinfo, if it's defined.
+#ifdef HAVE_STRUCT_MALLINFO
+// Malloc can be in several places on older versions of OS X.
+# if defined(HAVE_MALLOC_H)
+# include <malloc.h>
+# elif defined(HAVE_SYS_MALLOC_H)
+# include <sys/malloc.h>
+# elif defined(HAVE_MALLOC_MALLOC_H)
+# include <malloc/malloc.h>
+# endif
+#endif
+
+// Annoying stuff for windows -- makes sure clients can import these functions
+#ifndef PERFTOOLS_DLL_DECL
+# ifdef _WIN32
+#   define PERFTOOLS_DLL_DECL  __declspec(dllimport)
+# else
+#   define PERFTOOLS_DLL_DECL
+# endif
+#endif
+
+#ifdef __cplusplus
+namespace std {
+struct nothrow_t;
+}
+
+extern "C" {
+#endif
+  // Returns a human-readable version string.  If major, minor,
+  // and/or patch are not NULL, they are set to the major version,
+  // minor version, and patch-code (a string, usually "").
+  PERFTOOLS_DLL_DECL const char* tc_version(int* major, int* minor,
+                                            const char** patch) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_free(void* ptr) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_calloc(size_t nmemb, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) __THROW;
+
+  PERFTOOLS_DLL_DECL void* tc_memalign(size_t __alignment,
+                                       size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL int tc_posix_memalign(void** ptr,
+                                           size_t align, size_t size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_valloc(size_t __size) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t __size) __THROW;
+
+  PERFTOOLS_DLL_DECL void tc_malloc_stats(void) __THROW;
+  PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) __THROW;
+#if 0
+  PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) __THROW;
+#endif
+
+  // This is an alias for MallocExtension::instance()->GetAllocatedSize().
+  // It is equivalent to
+  //    OS X: malloc_size()
+  //    glibc: malloc_usable_size()
+  //    Windows: _msize()
+  PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) __THROW;
+
+#ifdef __cplusplus
+  PERFTOOLS_DLL_DECL int tc_set_new_mode(int flag) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_new(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size,
+                                          const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p,
+                                            const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void* tc_newarray(size_t size);
+  PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size,
+                                               const std::nothrow_t&) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray(void* p) __THROW;
+  PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p,
+                                                 const std::nothrow_t&) __THROW;
+}
+#endif
+
+#endif  // #ifndef TCMALLOC_TCMALLOC_H_

BeefRT/gperftools/src/packed-cache-inl.h

@@ -0,0 +1,238 @@
+// Copyright (c) 2007, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Geoff Pike
+//
+// This file provides a minimal cache that can hold a <key, value> pair
+// with little if any wasted space.  The types of the key and value
+// must be unsigned integral types or at least have unsigned semantics
+// for >>, casting, and similar operations.
+//
+// Synchronization is not provided.  However, the cache is implemented
+// as an array of cache entries whose type is chosen at compile time.
+// If a[i] is atomic on your hardware for the chosen array type then
+// raciness will not necessarily lead to bugginess.  The cache entries
+// must be large enough to hold a partial key and a value packed
+// together.  The partial keys are bit strings of length
+// kKeybits - kHashbits, and the values are bit strings of length kValuebits.
+//
+// In an effort to use minimal space, every cache entry represents
+// some <key, value> pair; the class provides no way to mark a cache
+// entry as empty or uninitialized.  In practice, you may want to have
+// reserved keys or values to get around this limitation.  For example, in
+// tcmalloc's PageID-to-sizeclass cache, a value of 0 is used as
+// "unknown sizeclass."
+//
+// Usage Considerations
+// --------------------
+//
+// kHashbits controls the size of the cache.  The best value for
+// kHashbits will of course depend on the application.  Perhaps try
+// tuning the value of kHashbits by measuring different values on your
+// favorite benchmark.  Also remember not to be a pig; other
+// programs that need resources may suffer if you are.
+//
+// The main uses for this class will be when performance is
+// critical and there's a convenient type to hold the cache's
+// entries.  As described above, the number of bits required
+// for a cache entry is (kKeybits - kHashbits) + kValuebits.  Suppose
+// kKeybits + kValuebits is 43.  Then it probably makes sense to
+// chose kHashbits >= 11 so that cache entries fit in a uint32.
+//
+// On the other hand, suppose kKeybits = kValuebits = 64.  Then
+// using this class may be less worthwhile.  You'll probably
+// be using 128 bits for each entry anyway, so maybe just pick
+// a hash function, H, and use an array indexed by H(key):
+//    void Put(K key, V value) { a_[H(key)] = pair<K, V>(key, value); }
+//    V GetOrDefault(K key, V default) { const pair<K, V> &p = a_[H(key)]; ... }
+//    etc.
+//
+// Further Details
+// ---------------
+//
+// For caches used only by one thread, the following is true:
+// 1. For a cache c,
+//      (c.Put(key, value), c.GetOrDefault(key, 0)) == value
+//    and
+//      (c.Put(key, value), <...>, c.GetOrDefault(key, 0)) == value
+//    if the elided code contains no c.Put calls.
+//
+// 2. Has(key) will return false if no <key, value> pair with that key
+//    has ever been Put.  However, a newly initialized cache will have
+//    some <key, value> pairs already present.  When you create a new
+//    cache, you must specify an "initial value."  The initialization
+//    procedure is equivalent to Clear(initial_value), which is
+//    equivalent to Put(k, initial_value) for all keys k from 0 to
+//    2^kHashbits - 1.
+//
+// 3. If key and key' differ then the only way Put(key, value) may
+//    cause Has(key') to change is that Has(key') may change from true to
+//    false. Furthermore, a Put() call that doesn't change Has(key')
+//    doesn't change GetOrDefault(key', ...) either.
+//
+// Implementation details:
+//
+// This is a direct-mapped cache with 2^kHashbits entries; the hash
+// function simply takes the low bits of the key.  We store whole keys
+// if a whole key plus a whole value fits in an entry.  Otherwise, an
+// entry is the high bits of a key and a value, packed together.
+// E.g., a 20 bit key and a 7 bit value only require a uint16 for each
+// entry if kHashbits >= 11.
+//
+// Alternatives to this scheme will be added as needed.
+
+#ifndef TCMALLOC_PACKED_CACHE_INL_H_
+#define TCMALLOC_PACKED_CACHE_INL_H_
+
+#include "tcconfig.h"
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_STDINT_H
+#include <stdint.h>                     // for uintptr_t
+#endif
+#include "base/basictypes.h"
+#include "internal_logging.h"
+
+// A safe way of doing "(1 << n) - 1" -- without worrying about overflow
+// Note this will all be resolved to a constant expression at compile-time
+#define N_ONES_(IntType, N)                                     \
+  ( (N) == 0 ? 0 : ((static_cast<IntType>(1) << ((N)-1))-1 +    \
+                    (static_cast<IntType>(1) << ((N)-1))) )
+
+// The types K and V provide upper bounds on the number of valid keys
+// and values, but we explicitly require the keys to be less than
+// 2^kKeybits and the values to be less than 2^kValuebits.  The size of
+// the table is controlled by kHashbits, and the type of each entry in
+// the cache is T.  See also the big comment at the top of the file.
+template <int kKeybits, typename T>
+class PackedCache {
+ public:
+  typedef uintptr_t K;
+  typedef size_t V;
+#ifdef TCMALLOC_SMALL_BUT_SLOW
+  // Decrease the size map cache if running in the small memory mode.
+  static const int kHashbits = 12;
+#else
+  static const int kHashbits = 16;
+#endif
+  static const int kValuebits = 7;
+  static const bool kUseWholeKeys = kKeybits + kValuebits <= 8 * sizeof(T);
+
+  explicit PackedCache(V initial_value) {
+    COMPILE_ASSERT(kKeybits <= sizeof(K) * 8, key_size);
+    COMPILE_ASSERT(kValuebits <= sizeof(V) * 8, value_size);
+    COMPILE_ASSERT(kHashbits <= kKeybits, hash_function);
+    COMPILE_ASSERT(kKeybits - kHashbits + kValuebits <= kTbits,
+                   entry_size_must_be_big_enough);
+    Clear(initial_value);
+  }
+
+  void Put(K key, V value) {
+    ASSERT(key == (key & kKeyMask));
+    ASSERT(value == (value & kValueMask));
+    array_[Hash(key)] = KeyToUpper(key) | value;
+  }
+
+  bool Has(K key) const {
+    ASSERT(key == (key & kKeyMask));
+    return KeyMatch(array_[Hash(key)], key);
+  }
+
+  V GetOrDefault(K key, V default_value) const {
+    // As with other code in this class, we touch array_ as few times
+    // as we can.  Assuming entries are read atomically (e.g., their
+    // type is uintptr_t on most hardware) then certain races are
+    // harmless.
+    ASSERT(key == (key & kKeyMask));
+    T entry = array_[Hash(key)];
+    return KeyMatch(entry, key) ? EntryToValue(entry) : default_value;
+  }
+
+  void Clear(V value) {
+    ASSERT(value == (value & kValueMask));
+    for (int i = 0; i < 1 << kHashbits; i++) {
+      ASSERT(kUseWholeKeys || KeyToUpper(i) == 0);
+      array_[i] = kUseWholeKeys ? (value | KeyToUpper(i)) : value;
+    }
+  }
+
+ private:
+  // We are going to pack a value and the upper part of a key (or a
+  // whole key) into an entry of type T.  The UPPER type is for the
+  // upper part of a key, after the key has been masked and shifted
+  // for inclusion in an entry.
+  typedef T UPPER;
+
+  static V EntryToValue(T t) { return t & kValueMask; }
+
+  // If we have space for a whole key, we just shift it left.
+  // Otherwise kHashbits determines where in a K to find the upper
+  // part of the key, and kValuebits determines where in the entry to
+  // put it.
+  static UPPER KeyToUpper(K k) {
+    if (kUseWholeKeys) {
+      return static_cast<T>(k) << kValuebits;
+    } else {
+      const int shift = kHashbits - kValuebits;
+      // Assume kHashbits >= kValuebits.  It'd be easy to lift this assumption.
+      return static_cast<T>(k >> shift) & kUpperMask;
+    }
+  }
+
+  static size_t Hash(K key) {
+    return static_cast<size_t>(key) & N_ONES_(size_t, kHashbits);
+  }
+
+  // Does the entry match the relevant part of the given key?
+  static bool KeyMatch(T entry, K key) {
+    return kUseWholeKeys ?
+        (entry >> kValuebits == key) :
+        ((KeyToUpper(key) ^ entry) & kUpperMask) == 0;
+  }
+
+  static const int kTbits = 8 * sizeof(T);
+  static const int kUpperbits = kUseWholeKeys ? kKeybits : kKeybits - kHashbits;
+
+  // For masking a K.
+  static const K kKeyMask = N_ONES_(K, kKeybits);
+
+  // For masking a T.
+  static const T kUpperMask = N_ONES_(T, kUpperbits) << kValuebits;
+
+  // For masking a V or a T.
+  static const V kValueMask = N_ONES_(V, kValuebits);
+
+  // array_ is the cache.  Its elements are volatile because any
+  // thread can write any array element at any time.
+  volatile T array_[1 << kHashbits];
+};
+
+#undef N_ONES_
+
+#endif  // TCMALLOC_PACKED_CACHE_INL_H_

BeefRT/gperftools/src/page_heap.cc

@@ -0,0 +1,809 @@
+// Copyright (c) 2008, Google Inc. 
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.  
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+#include "BFPlatform.h"
+#include "BeefySysLib/Common.h"
+#include "BeefySysLib/BFApp.h"
+#include <algorithm>
+
+#include "config.h"
+#ifdef HAVE_INTTYPES_H
+#include <inttypes.h>                   // for PRIuPTR
+#endif
+#include "gperftools/malloc_extension.h"      // for MallocRange, etc
+#include "base/basictypes.h"
+#include "base/commandlineflags.h"
+#include "internal_logging.h"  // for ASSERT, TCMalloc_Printer, etc
+#include "page_heap_allocator.h"  // for PageHeapAllocator
+#include "static_vars.h"       // for Static
+#include "system-alloc.h"      // for TCMalloc_SystemAlloc, etc
+
+#include <fstream>
+
+namespace Beefy
+{
+	String vformat(const char* fmt, va_list argPtr);
+}
+
+namespace TCMALLOC_NAMESPACE
+{
+	void TCLogStrF(const char* fmt ...)
+	{
+		/*va_list argList;
+		va_start(argList, fmt);
+		std::string str = Beefy::vformat(fmt, argList);
+		va_end(argList);
+
+		static FILE* fp = NULL;
+		if (fp == NULL)
+			fp = fopen("tclog.txt", "w");
+
+		fwrite(str.c_str(), str.length(), 1, fp);
+		fflush(fp);*/
+	}
+
+	DEFINE_double(tcmalloc_release_rate,
+		10.0f,//EnvToDouble("TCMALLOC_RELEASE_RATE", 1.0),
+		"Rate at which we release unused memory to the system.  "
+		"Zero means we never release memory back to the system.  "
+		"Increase this flag to return memory faster; decrease it "
+		"to return memory slower.  Reasonable rates are in the "
+		"range [0,10]");
+
+	DEFINE_int64(tcmalloc_heap_limit_mb,
+		0, //EnvToInt("TCMALLOC_HEAP_LIMIT_MB", 0),
+		"Limit total size of the process heap to the "
+		"specified number of MiB. "
+		"When we approach the limit the memory is released "
+		"to the system more aggressively (more minor page faults). "
+		"Zero means to allocate as long as system allows.");
+}
+
+namespace TCMALLOC_NAMESPACE {
+
+void SetSpanLength(Span* span, int length)
+{
+#ifdef TCMALLOC_CLEAR_SPAN_END
+	intptr pageSize = (intptr)1 << kPageShift;
+	intptr spanSize = (intptr)pageSize * (intptr)length;
+	void* spanStart = (void*)((intptr)span->start << kPageShift);
+	void* spanEnd = (void*)((intptr)spanStart + spanSize);
+	*(intptr*)((uint8*)spanEnd - sizeof(intptr)) = 0;
+	BF_FULL_MEMORY_FENCE();
+#endif
+	span->length = length;
+}
+
+PageHeap::PageHeap()
+    : pagemap_(MetaDataAlloc),
+      pagemap_cache_(0),
+      scavenge_counter_(0),
+      // Start scavenging at kMaxPages list
+      release_index_(kMaxPages) {
+  COMPILE_ASSERT(kNumClasses <= (1 << PageMapCache::kValuebits), valuebits);
+  DLL_Init(&large_.normal);
+  DLL_Init(&large_.returned);
+  for (int i = 0; i < kMaxPages; i++) {
+    DLL_Init(&free_[i].normal);
+    DLL_Init(&free_[i].returned);
+  }
+}
+
+Span* PageHeap::SearchFreeAndLargeLists(Length n) {
+  ASSERT(Check());
+  ASSERT(n > 0);
+
+  // Find first size >= n that has a non-empty list
+  for (Length s = n; s < kMaxPages; s++) {
+    Span* ll = &free_[s].normal;
+    // If we're lucky, ll is non-empty, meaning it has a suitable span.
+    if (!DLL_IsEmpty(ll)) {
+      ASSERT(ll->next->location == Span::ON_NORMAL_FREELIST);
+      return Carve(ll->next, n);
+    }
+    // Alternatively, maybe there's a usable returned span.
+    ll = &free_[s].returned;
+    if (!DLL_IsEmpty(ll)) {
+      // We did not call EnsureLimit before, to avoid releasing the span
+      // that will be taken immediately back.
+      // Calling EnsureLimit here is not very expensive, as it fails only if
+      // there is no more normal spans (and it fails efficiently)
+      // or SystemRelease does not work (there is probably no returned spans).
+      if (EnsureLimit(n)) {
+        // ll may have became empty due to coalescing
+        if (!DLL_IsEmpty(ll)) {
+          ASSERT(ll->next->location == Span::ON_RETURNED_FREELIST);
+          return Carve(ll->next, n);
+        }
+      }
+    }
+  }
+  // No luck in free lists, our last chance is in a larger class.
+  return AllocLarge(n);  // May be NULL
+}
+
+static const size_t kForcedCoalesceInterval = 128*1024*1024;
+
+Span* PageHeap::New(Length n) {
+  ASSERT(Check());
+  ASSERT(n > 0);
+
+  Span* result = SearchFreeAndLargeLists(n);
+  if (result != NULL)
+    return result;
+
+  if (stats_.free_bytes != 0 && stats_.unmapped_bytes != 0
+      && stats_.free_bytes + stats_.unmapped_bytes >= stats_.system_bytes / 4
+      && (stats_.system_bytes / kForcedCoalesceInterval
+          != (stats_.system_bytes + (n << kPageShift)) / kForcedCoalesceInterval)) {
+    // We're about to grow heap, but there are lots of free pages.
+    // tcmalloc's design decision to keep unmapped and free spans
+    // separately and never coalesce them means that sometimes there
+    // can be free pages span of sufficient size, but it consists of
+    // "segments" of different type so page heap search cannot find
+    // it. In order to prevent growing heap and wasting memory in such
+    // case we're going to unmap all free pages. So that all free
+    // spans are maximally coalesced.
+    //
+    // We're also limiting 'rate' of going into this path to be at
+    // most once per 128 megs of heap growth. Otherwise programs that
+    // grow heap frequently (and that means by small amount) could be
+    // penalized with higher count of minor page faults.
+    //
+    // See also large_heap_fragmentation_unittest.cc and
+    // https://code.google.com/p/gperftools/issues/detail?id=368
+    ReleaseAtLeastNPages(static_cast<Length>(0x7fffffff));
+
+    // then try again. If we are forced to grow heap because of large
+    // spans fragmentation and not because of problem described above,
+    // then at the very least we've just unmapped free but
+    // insufficiently big large spans back to OS. So in case of really
+    // unlucky memory fragmentation we'll be consuming virtual address
+    // space, but not real memory
+    result = SearchFreeAndLargeLists(n);
+    if (result != NULL) return result;
+  }
+
+  // Grow the heap and try again.
+  if (!GrowHeap(n)) {
+    ASSERT(Check());
+    return NULL;
+  }
+  return SearchFreeAndLargeLists(n);
+}
+
+Span* PageHeap::AllocLarge(Length n) {
+  // find the best span (closest to n in size).
+  // The following loops implements address-ordered best-fit.
+  Span *best = NULL;
+
+  // Search through normal list
+  for (Span* span = large_.normal.next;
+       span != &large_.normal;
+       span = span->next) {
+    if (span->length >= n) {
+      if ((best == NULL)
+          || (span->length < best->length)
+          || ((span->length == best->length) && (span->start < best->start))) {
+        best = span;
+        ASSERT(best->location == Span::ON_NORMAL_FREELIST);
+      }
+    }
+  }
+
+  Span *bestNormal = best;
+
+  // Search through released list in case it has a better fit
+  for (Span* span = large_.returned.next;
+       span != &large_.returned;
+       span = span->next) {
+    if (span->length >= n) {
+      if ((best == NULL)
+          || (span->length < best->length)
+          || ((span->length == best->length) && (span->start < best->start))) {
+        best = span;
+        ASSERT(best->location == Span::ON_RETURNED_FREELIST);
+      }
+    }
+  }
+
+  if (best == bestNormal) {
+    return best == NULL ? NULL : Carve(best, n);
+  }
+
+  // best comes from returned list.
+
+  if (EnsureLimit(n, false)) {
+    return Carve(best, n);
+  }
+
+  if (EnsureLimit(n, true)) {
+    // best could have been destroyed by coalescing.
+    // bestNormal is not a best-fit, and it could be destroyed as well.
+    // We retry, the limit is already ensured:
+    return AllocLarge(n);
+  }
+
+  // If bestNormal existed, EnsureLimit would succeeded:
+  ASSERT(bestNormal == NULL);
+  // We are not allowed to take best from returned list.
+  return NULL;
+}
+
+Span* PageHeap::Split(Span* span, Length n) {
+  ASSERT(0 < n);
+  ASSERT(n < span->length);
+  ASSERT(span->location == Span::IN_USE);
+  ASSERT(span->sizeclass == 0);
+  Event(span, 'T', n);
+
+  const int extra = span->length - n;
+  Span* leftover = NewSpan(span->start + n, extra);
+  ASSERT(leftover->location == Span::IN_USE);
+  Event(leftover, 'U', extra);
+  RecordSpan(leftover);
+  pagemap_.set(span->start + n - 1, span); // Update map from pageid to span
+  SetSpanLength(span, n);  
+
+  return leftover;
+}
+
+void PageHeap::CommitSpan(Span* span) {
+	TCMalloc_SystemCommit(reinterpret_cast<void*>(span->start << kPageShift),
+		static_cast<size_t>(span->length << kPageShift));
+	stats_.committed_bytes += span->length << kPageShift;
+
+	TCLogStrF("CommitSpan Span:%08X Addr:%08X\n", span, span->start << kPageShift);
+}
+
+void PageHeap::DecommitSpan(Span* span) {
+	TCMalloc_SystemRelease(reinterpret_cast<void*>(span->start << kPageShift),
+		static_cast<size_t>(span->length << kPageShift));
+	stats_.committed_bytes -= span->length << kPageShift;
+
+	TCLogStrF("DecommitSpan Span:%08X Addr:%08X\n", span, span->start << kPageShift);
+}
+
+Span* PageHeap::Carve(Span* span, Length n) {
+	ASSERT(n > 0);
+	ASSERT(span->location != Span::IN_USE);
+	const int old_location = span->location;
+	const int old_decommitDelay = span->decommitDelay;
+
+	RemoveFromFreeList(span);	
+	Event(span, 'A', n);	
+
+	const int extra = span->length - n;
+	ASSERT(extra >= 0);
+	if (extra > 0) {
+		Span* leftover = NewSpan(span->start + n, extra);
+		leftover->location = old_location;
+		leftover->decommitDelay = old_decommitDelay;
+		Event(leftover, 'S', extra);
+		RecordSpan(leftover);
+
+		// The previous span of |leftover| was just splitted -- no need to
+		// coalesce them. The next span of |leftover| was not previously coalesced
+		// with |span|, i.e. is NULL or has got location other than |old_location|.
+		const PageID p = leftover->start;
+		const Length len = leftover->length;
+		Span* next = GetDescriptor(p + len);
+		/*ASSERT(next == NULL ||
+			next->location == Span::IN_USE ||
+			next->location != leftover->location);*/
+
+		PrependToFreeList(leftover);  // Skip coalescing - no candidates possible
+		SetSpanLength(span, n);		
+		pagemap_.set(span->start + n - 1, span);
+	}
+	ASSERT(Check());
+	if (old_location == Span::ON_RETURNED_FREELIST) {
+		// We need to recommit this address space.
+		if (span->decommitDelay == 0)
+			CommitSpan(span);		
+	}
+
+	BF_FULL_MEMORY_FENCE();
+
+	// Don't set to inuse until after recommiting
+	span->location = Span::IN_USE;
+
+	ASSERT(span->location == Span::IN_USE);
+	ASSERT(span->length == n);
+	ASSERT(stats_.unmapped_bytes - stats_.awaiting_decommit_bytes + stats_.committed_bytes == stats_.system_bytes);
+
+	//BCF
+	//TODO: Remove
+	int* data = (int*) ((intptr_t) span->start << kPageShift);
+	*data = 0;
+
+	return span;
+}
+
+int itrIdx = 0;
+
+void PageHeap::Delete(Span* span, bool allowScavenge) {
+
+	itrIdx++;
+
+	ASSERT(Check());
+	ASSERT(span->location == Span::IN_USE);
+	ASSERT(span->length > 0);
+	ASSERT(GetDescriptor(span->start) == span);
+	ASSERT(GetDescriptor(span->start + span->length - 1) == span);
+	const Length n = span->length;
+	span->sizeclass = 0;
+	span->sample = 0;
+	span->location = Span::ON_NORMAL_FREELIST;
+	Event(span, 'D', span->length);
+	MergeIntoFreeList(span);  // Coalesces if possible
+	if (allowScavenge)
+		IncrementalScavenge(n);
+	ASSERT(stats_.unmapped_bytes - stats_.awaiting_decommit_bytes + stats_.committed_bytes == stats_.system_bytes);	
+	ASSERT(Check());
+}
+
+void PageHeap::MergeIntoFreeList(Span* span) {
+	ASSERT(span->location != Span::IN_USE);
+
+	// Coalesce -- we guarantee that "p" != 0, so no bounds checking
+	// necessary.  We do not bother resetting the stale pagemap
+	// entries for the pieces we are merging together because we only
+	// care about the pagemap entries for the boundaries.
+	//
+	// Note that the adjacent spans we merge into "span" may come out of a
+	// "normal" (committed) list, and cleanly merge with our IN_USE span, which
+	// is implicitly committed.  If the adjacents spans are on the "returned"
+	// (decommitted) list, then we must get both spans into the same state before
+	// or after we coalesce them.  The current code always decomits. This is
+	// achieved by blindly decommitting the entire coalesced region, which  may
+	// include any combination of committed and decommitted spans, at the end of
+	// the method.
+
+	// TODO(jar): "Always decommit" causes some extra calls to commit when we are
+	// called in GrowHeap() during an allocation :-/.  We need to eval the cost of
+	// that oscillation, and possibly do something to reduce it.
+
+	// TODO(jar): We need a better strategy for deciding to commit, or decommit,
+	// based on memory usage and free heap sizes.
+
+	const PageID p = span->start;
+	const Length n = span->length;
+	Span* prev = GetDescriptor(p - 1);
+	//BCF: Only merge together spans if they aren't decommited yet
+	if (prev != NULL && (prev->location != Span::IN_USE)) {
+		// Merge preceding span into this span
+		ASSERT(prev->start + prev->length == p);
+		const Length len = prev->length;
+		bool isValidSpan = true;
+		if (prev->location == Span::ON_RETURNED_FREELIST) {
+			// We're about to put the merge span into the returned freelist and call
+			// DecommitSpan() on it, which will mark the entire span including this
+			// one as released and decrease stats_.committed_bytes by the size of the
+			// merged span.  To make the math work out we temporarily increase the
+			// stats_.committed_bytes amount.
+			if (prev->decommitDelay == 0)
+				isValidSpan = false;			
+				//stats_.committed_bytes += prev->length << kPageShift;						
+		}
+		if (isValidSpan)			
+		{			
+			RemoveFromFreeList(prev);
+			DeleteSpan(prev);
+			span->start -= len;
+			SetSpanLength(span, span->length + len);			
+			pagemap_.set(span->start, span);
+			Event(span, 'L', len);
+		}
+	}
+	Span* next = GetDescriptor(p + n);
+	if (next != NULL && next->location != Span::IN_USE) {
+		// Merge next span into this span
+		ASSERT(next->start == p + n);
+		const Length len = next->length;
+		bool isValidSpan = true;
+		if (next->location == Span::ON_RETURNED_FREELIST) {
+			// See the comment below 'if (prev->location ...' for explanation.
+			if (next->decommitDelay == 0)
+				isValidSpan = false;
+				//stats_.committed_bytes += next->length << kPageShift;						
+		}
+		if (isValidSpan)
+		{			
+			RemoveFromFreeList(next);
+			DeleteSpan(next);
+			SetSpanLength(span, span->length + len);			
+			pagemap_.set(span->start + span->length - 1, span);
+			Event(span, 'R', len);
+		}
+	}
+
+	Event(span, 'D', span->length);
+	span->location = Span::ON_RETURNED_FREELIST;
+	span->decommitDelay = 7;
+	//DecommitSpan(span);
+	PrependToFreeList(span);
+}
+
+void PageHeap::PrependToFreeList(Span* span) {
+  ASSERT(span->location != Span::IN_USE);
+  SpanList* list = (span->length < kMaxPages) ? &free_[span->length] : &large_;
+  if (span->location == Span::ON_NORMAL_FREELIST) {
+    stats_.free_bytes += (span->length << kPageShift);
+    DLL_Prepend(&list->normal, span);
+  } else {
+	TCLogStrF("PrependToFreeList loc:Released Span:%08X Addr:%08X\n", span, span->start << kPageShift);
+    stats_.unmapped_bytes += (span->length << kPageShift);
+	if (span->decommitDelay != 0)
+		stats_.awaiting_decommit_bytes += (span->length << kPageShift);
+    DLL_Prepend(&list->returned, span);
+  }
+}
+
+//BCF - Decommits from old freelist items
+//  We do this once per GC cycle right before releasing memory from the new deleted objects.
+//  This keeps us from thrashing Allocs/Frees
+//  'immediate' causes decommit of all pending items, otherwise it takes several cycles
+void PageHeap::DecommitFromReleasedList(bool immediate)
+{
+	SpinLockHolder h(Static::pageheap_lock());
+	
+	for (int i = 0; i < kMaxPages + 1; i++)
+	{		
+		SpanList* slist = (i == kMaxPages) ? &large_ : &free_[i];	
+
+		int count = 0;
+		int awaitingDecommits = 0;
+		int decommits = 0;
+
+		if (!DLL_IsEmpty(&slist->returned))
+		{
+			Span* list = &slist->returned;
+			Span* span = list->next;
+			while (span != list)
+			{
+				count++;
+
+				if (span->decommitDelay > 0)
+				{					
+					if ((span->decommitDelay == 1) || (immediate))
+					{
+						decommits++;
+
+						span->decommitDelay = 0;
+						DecommitSpan(span);						
+						stats_.awaiting_decommit_bytes -= (span->length << kPageShift);
+					}
+					else
+					{
+						span->decommitDelay--;
+						awaitingDecommits++;
+					}
+				}
+				span = span->next;
+			}
+		}
+	}
+}
+
+void PageHeap::RemoveFromFreeList(Span* span) {
+  ASSERT(span->location != Span::IN_USE);
+  if (span->location == Span::ON_NORMAL_FREELIST) {
+    stats_.free_bytes -= (span->length << kPageShift);
+  } else {
+	  TCLogStrF("RemoveFromFreeList loc:Released Span:%08X Addr:%08X\n", span, span->start << kPageShift);
+	  stats_.unmapped_bytes -= (span->length << kPageShift);
+	  if (span->decommitDelay != 0)
+		  stats_.awaiting_decommit_bytes -= (span->length << kPageShift);	  
+  }
+  DLL_Remove(span);
+}
+
+void PageHeap::IncrementalScavenge(Length n) {
+  // Fast path; not yet time to release memory
+  scavenge_counter_ -= n;
+  if (scavenge_counter_ >= 0) return;  // Not yet time to scavenge
+
+  const double rate = FLAGS_tcmalloc_release_rate;
+  if (rate <= 1e-6) {
+    // Tiny release rate means that releasing is disabled.
+    scavenge_counter_ = kDefaultReleaseDelay;
+    return;
+  }
+
+  Length released_pages = ReleaseAtLeastNPages(1);
+
+  if (released_pages == 0) {
+    // Nothing to scavenge, delay for a while.
+    scavenge_counter_ = kDefaultReleaseDelay;
+  } else {
+    // Compute how long to wait until we return memory.
+    // FLAGS_tcmalloc_release_rate==1 means wait for 1000 pages
+    // after releasing one page.
+    const double mult = 1000.0 / rate;
+    double wait = mult * static_cast<double>(released_pages);
+    if (wait > kMaxReleaseDelay) {
+      // Avoid overflow and bound to reasonable range.
+      wait = kMaxReleaseDelay;
+    }
+    scavenge_counter_ = static_cast<int64_t>(wait);
+  }
+}
+
+Length PageHeap::ReleaseLastNormalSpan(SpanList* slist) {
+  Span* s = slist->normal.prev;
+  ASSERT(s->location == Span::ON_NORMAL_FREELIST);
+  
+  //if (TCMalloc_SystemRelease(reinterpret_cast<void*>(s->start << kPageShift),
+                         //static_cast<size_t>(s->length << kPageShift))) 
+
+  if (TCMalloc_CanRelease())
+  {
+    RemoveFromFreeList(s);
+    const Length n = s->length;
+    s->location = Span::ON_RETURNED_FREELIST;
+	s->decommitDelay = 3;
+    MergeIntoFreeList(s);  // Coalesces if possible.
+    return n;
+  }
+
+  return 0;
+}
+
+Length PageHeap::ReleaseAtLeastNPages(Length num_pages) {
+  Length released_pages = 0;
+
+  // Round robin through the lists of free spans, releasing the last
+  // span in each list.  Stop after releasing at least num_pages
+  // or when there is nothing more to release.
+  while (released_pages < num_pages && stats_.free_bytes > 0) {
+    for (int i = 0; i < kMaxPages+1 && released_pages < num_pages;
+         i++, release_index_++) {
+      if (release_index_ > kMaxPages) release_index_ = 0;
+      SpanList* slist = (release_index_ == kMaxPages) ?
+          &large_ : &free_[release_index_];
+      if (!DLL_IsEmpty(&slist->normal)) {
+        Length released_len = ReleaseLastNormalSpan(slist);
+        // Some systems do not support release
+        if (released_len == 0) return released_pages;
+        released_pages += released_len;
+      }
+    }
+  }
+  return released_pages;
+}
+
+bool PageHeap::EnsureLimit(Length n, bool withRelease)
+{
+  Length limit = (FLAGS_tcmalloc_heap_limit_mb*1024*1024) >> kPageShift;
+  if (limit == 0) return true; //there is no limit
+
+  // We do not use stats_.system_bytes because it does not take
+  // MetaDataAllocs into account.
+  Length takenPages = TCMalloc_SystemTaken >> kPageShift;
+  //XXX takenPages may be slightly bigger than limit for two reasons:
+  //* MetaDataAllocs ignore the limit (it is not easy to handle
+  //  out of memory there)
+  //* sys_alloc may round allocation up to huge page size,
+  //  although smaller limit was ensured
+
+  ASSERT(takenPages >= stats_.unmapped_bytes >> kPageShift);
+  takenPages -= stats_.unmapped_bytes >> kPageShift;
+
+  if (takenPages + n > limit && withRelease) {
+    takenPages -= ReleaseAtLeastNPages(takenPages + n - limit);
+  }
+
+  return takenPages + n <= limit;
+}
+
+void BFMemoryBarrier();
+    
+void PageHeap::RegisterSizeClass(Span* span, size_t sc) {
+  // Associate span object with all interior pages as well
+  ASSERT(span->location == Span::IN_USE);
+  ASSERT(GetDescriptor(span->start) == span);
+  ASSERT(GetDescriptor(span->start+span->length-1) == span);
+  Event(span, 'C', sc);
+
+  //ASSERT(sc > 1);
+
+  //BCF- zero out data.  When size changes we can't count on mBFMonitorPtrAndObjFlags being zero'd out, as the GC would normally do
+  //  when the size class isn't changing
+  if (span->sizeclass != sc)
+  {
+	  int* data = (int*)((intptr_t)span->start << kPageShift);
+	  *data = 0;
+	memset((void*)((intptr_t)span->start << kPageShift), 0, (1 << kPageShift) * span->length);
+  }
+
+    BF_FULL_MEMORY_FENCE();
+
+  span->sizeclass = sc;
+  for (Length i = 1; i < span->length-1; i++) {
+    pagemap_.set(span->start+i, span);
+  }
+}
+
+void PageHeap::GetSmallSpanStats(SmallSpanStats* result) {
+  for (int s = 0; s < kMaxPages; s++) {
+    result->normal_length[s] = DLL_Length(&free_[s].normal);
+    result->returned_length[s] = DLL_Length(&free_[s].returned);
+  }
+}
+
+void PageHeap::GetLargeSpanStats(LargeSpanStats* result) {
+  result->spans = 0;
+  result->normal_pages = 0;
+  result->returned_pages = 0;
+  for (Span* s = large_.normal.next; s != &large_.normal; s = s->next) {
+    result->normal_pages += s->length;;
+    result->spans++;
+  }
+  for (Span* s = large_.returned.next; s != &large_.returned; s = s->next) {
+    result->returned_pages += s->length;
+    result->spans++;
+  }
+}
+
+bool PageHeap::GetNextRange(PageID start, base::MallocRange* r) {
+  Span* span = reinterpret_cast<Span*>(pagemap_.Next(start));
+  if (span == NULL) {
+    return false;
+  }
+  r->address = span->start << kPageShift;
+  r->length = span->length << kPageShift;
+  r->fraction = 0;
+  switch (span->location) {
+    case Span::IN_USE:
+      r->type = base::MallocRange::INUSE;
+      r->fraction = 1;
+      if (span->sizeclass > 0) {
+        // Only some of the objects in this span may be in use.
+        const size_t osize = Static::sizemap()->class_to_size(span->sizeclass);
+        r->fraction = (1.0 * osize * span->refcount) / r->length;
+      }
+      break;
+    case Span::ON_NORMAL_FREELIST:
+      r->type = base::MallocRange::FREE;
+      break;
+    case Span::ON_RETURNED_FREELIST:	
+      r->type = base::MallocRange::UNMAPPED;
+      break;
+    default:
+      r->type = base::MallocRange::UNKNOWN;
+      break;
+  }
+  return true;
+}
+
+static void RecordGrowth(size_t growth) {
+  StackTrace* t = Static::stacktrace_allocator()->New();
+  t->depth = GetStackTrace(t->stack, kMaxStackDepth-1, 3);
+  t->size = growth;
+  t->stack[kMaxStackDepth-1] = reinterpret_cast<void*>(Static::growth_stacks());
+  Static::set_growth_stacks(t);
+}
+
+void* PageHeap::sAddressStart = (void*)((uintptr_t)-1 >> 1);
+void* PageHeap::sAddressEnd = 0;
+
+bool PageHeap::GrowHeap(Length n) {
+  ASSERT(kMaxPages >= kMinSystemAlloc);
+  if (n > kMaxValidPages) return false;
+  Length ask = (n>kMinSystemAlloc) ? n : static_cast<Length>(kMinSystemAlloc);
+  size_t actual_size;
+  void* ptr = NULL;
+  if (EnsureLimit(ask)) {
+      ptr = TCMalloc_SystemAlloc(ask << kPageShift, &actual_size, kPageSize);
+  }
+  if (ptr == NULL) {
+    if (n < ask) {
+      // Try growing just "n" pages
+      ask = n;
+      if (EnsureLimit(ask)) {
+        ptr = TCMalloc_SystemAlloc(ask << kPageShift, &actual_size, kPageSize);
+      }
+    }
+    if (ptr == NULL) return false;
+  }
+  ask = actual_size >> kPageShift;
+  RecordGrowth(ask << kPageShift);
+
+  //BCF
+  sAddressStart = BF_MIN(sAddressStart, ptr);
+  sAddressEnd = BF_MAX(sAddressEnd, (void*)((char*)ptr + actual_size));
+
+  uint64_t old_system_bytes = stats_.system_bytes;
+  stats_.system_bytes += (ask << kPageShift);
+  stats_.committed_bytes += (ask << kPageShift);
+  const PageID p = reinterpret_cast<uintptr_t>(ptr) >> kPageShift;
+  ASSERT(p > 0);
+
+  // If we have already a lot of pages allocated, just pre allocate a bunch of
+  // memory for the page map. This prevents fragmentation by pagemap metadata
+  // when a program keeps allocating and freeing large blocks.
+
+  if (old_system_bytes < kPageMapBigAllocationThreshold
+      && stats_.system_bytes >= kPageMapBigAllocationThreshold) {
+    pagemap_.PreallocateMoreMemory();
+  }
+
+  // Make sure pagemap_ has entries for all of the new pages.
+  // Plus ensure one before and one after so coalescing code
+  // does not need bounds-checking.
+  if (pagemap_.Ensure(p-1, ask+2)) {
+    // Pretend the new area is allocated and then Delete() it to cause
+    // any necessary coalescing to occur.
+    Span* span = NewSpan(p, ask);
+    RecordSpan(span);
+    Delete(span);
+	ASSERT(stats_.unmapped_bytes - stats_.awaiting_decommit_bytes + stats_.committed_bytes == stats_.system_bytes);
+    ASSERT(Check());
+    return true;
+  } else {
+    // We could not allocate memory within "pagemap_"
+    // TODO: Once we can return memory to the system, return the new span
+    return false;
+  }
+}
+
+bool PageHeap::Check() {
+  ASSERT(free_[0].normal.next == &free_[0].normal);
+  ASSERT(free_[0].returned.next == &free_[0].returned);
+  return true;
+}
+
+bool PageHeap::CheckExpensive() {
+  bool result = Check();
+  CheckList(&large_.normal, kMaxPages, 1000000000, Span::ON_NORMAL_FREELIST);
+  CheckList(&large_.returned, kMaxPages, 1000000000, Span::ON_RETURNED_FREELIST);
+  for (Length s = 1; s < kMaxPages; s++) {
+    CheckList(&free_[s].normal, s, s, Span::ON_NORMAL_FREELIST);
+    CheckList(&free_[s].returned, s, s, Span::ON_RETURNED_FREELIST);
+  }
+  return result;
+}
+
+bool PageHeap::CheckList(Span* list, Length min_pages, Length max_pages,
+                         int freelist) {
+  for (Span* s = list->next; s != list; s = s->next) {
+    CHECK_CONDITION(s->location == freelist);  // NORMAL or RETURNED
+    CHECK_CONDITION(s->length >= min_pages);
+    CHECK_CONDITION(s->length <= max_pages);
+    CHECK_CONDITION(GetDescriptor(s->start) == s);
+    CHECK_CONDITION(GetDescriptor(s->start+s->length-1) == s);
+  }
+  return true;
+}
+
+}  // namespace TCMALLOC_NAMESPACE

BeefRT/gperftools/src/page_heap.h

@@ -0,0 +1,307 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+#ifndef TCMALLOC_PAGE_HEAP_H_
+#define TCMALLOC_PAGE_HEAP_H_
+
+#include "tcconfig.h"
+#include <stddef.h>                     // for size_t
+#ifdef HAVE_STDINT_H
+#include <stdint.h>                     // for uint64_t, int64_t, uint16_t
+#endif
+#include "gperftools/malloc_extension.h"
+#include "base/basictypes.h"
+#include "common.h"
+#include "packed-cache-inl.h"
+#include "pagemap.h"
+#include "span.h"
+
+// We need to dllexport PageHeap just for the unittest.  MSVC complains
+// that we don't dllexport the PageHeap members, but we don't need to
+// test those, so I just suppress this warning.
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable:4251)
+#endif
+
+// This #ifdef should almost never be set.  Set NO_TCMALLOC_SAMPLES if
+// you're porting to a system where you really can't get a stacktrace.
+// Because we control the definition of GetStackTrace, all clients of
+// GetStackTrace should #include us rather than stacktrace.h.
+#ifdef NO_TCMALLOC_SAMPLES
+  // We use #define so code compiles even if you #include stacktrace.h somehow.
+# define GetStackTrace(stack, depth, skip)  (0)
+#else
+# include "gperftools/stacktrace.h"
+#endif
+
+namespace base {
+struct MallocRange;
+}
+
+namespace TCMALLOC_NAMESPACE {
+
+// -------------------------------------------------------------------------
+// Map from page-id to per-page data
+// -------------------------------------------------------------------------
+
+// We use PageMap2<> for 32-bit and PageMap3<> for 64-bit machines.
+// We also use a simple one-level cache for hot PageID-to-sizeclass mappings,
+// because sometimes the sizeclass is all the information we need.
+
+// Selector class -- general selector uses 3-level map
+template <int BITS> class MapSelector {
+ public:
+  typedef TCMalloc_PageMap3<BITS-kPageShift> Type;
+  typedef PackedCache<BITS-kPageShift, uint64_t> CacheType;
+};
+
+// A two-level map for 32-bit machines
+template <> class MapSelector<32> {
+ public:
+  typedef TCMalloc_PageMap2<32-kPageShift> Type;
+  typedef PackedCache<32-kPageShift, uint16_t> CacheType;
+};
+
+// -------------------------------------------------------------------------
+// Page-level allocator
+//  * Eager coalescing
+//
+// Heap for page-level allocation.  We allow allocating and freeing a
+// contiguous runs of pages (called a "span").
+// -------------------------------------------------------------------------
+
+class PERFTOOLS_DLL_DECL PageHeap {
+ public:
+  PageHeap();
+
+  // Allocate a run of "n" pages.  Returns zero if out of memory.
+  // Caller should not pass "n == 0" -- instead, n should have
+  // been rounded up already.
+  Span* New(Length n);
+
+  // Delete the span "[p, p+n-1]".
+  // REQUIRES: span was returned by earlier call to New() and
+  //           has not yet been deleted.
+  void Delete(Span* span, bool allowScavenge = false);  
+
+  // Mark an allocated span as being used for small objects of the
+  // specified size-class.
+  // REQUIRES: span was returned by an earlier call to New()
+  //           and has not yet been deleted.
+  void RegisterSizeClass(Span* span, size_t sc);
+
+  // Split an allocated span into two spans: one of length "n" pages
+  // followed by another span of length "span->length - n" pages.
+  // Modifies "*span" to point to the first span of length "n" pages.
+  // Returns a pointer to the second span.
+  //
+  // REQUIRES: "0 < n < span->length"
+  // REQUIRES: span->location == IN_USE
+  // REQUIRES: span->sizeclass == 0
+  Span* Split(Span* span, Length n);
+  void CommitSpan(Span* span);
+  void DecommitSpan(Span* span);
+  void DecommitFromReleasedList(bool immediate);
+
+  // Return the descriptor for the specified page.  Returns NULL if
+  // this PageID was not allocated previously.
+  inline Span* GetDescriptor(PageID p) const {
+    return reinterpret_cast<Span*>(pagemap_.get(p));
+  }
+
+  // If this page heap is managing a range with starting page # >= start,
+  // store info about the range in *r and return true.  Else return false.
+  bool GetNextRange(PageID start, base::MallocRange* r);
+
+  // Page heap statistics
+  struct Stats {
+	  Stats() : system_bytes(0), free_bytes(0), unmapped_bytes(0), committed_bytes(0), awaiting_decommit_bytes(0) {}
+	  uint64_t system_bytes;    // Total bytes allocated from system
+	  uint64_t free_bytes;      // Total bytes on normal freelists
+	  uint64_t unmapped_bytes;  // Total bytes on returned freelists
+	  uint64_t awaiting_decommit_bytes;  // Total bytes on returned freelists awaiting uncommit
+	  uint64_t committed_bytes;  // Bytes committed, always <= system_bytes_.
+
+  };
+  inline Stats stats() const { return stats_; }
+
+  struct SmallSpanStats {
+    // For each free list of small spans, the length (in spans) of the
+    // normal and returned free lists for that size.
+    int64 normal_length[kMaxPages];
+    int64 returned_length[kMaxPages];
+  };
+  void GetSmallSpanStats(SmallSpanStats* result);
+
+  // Stats for free large spans (i.e., spans with more than kMaxPages pages).
+  struct LargeSpanStats {
+    int64 spans;           // Number of such spans
+    int64 normal_pages;    // Combined page length of normal large spans
+    int64 returned_pages;  // Combined page length of unmapped spans
+  };
+  void GetLargeSpanStats(LargeSpanStats* result);
+
+  bool Check();
+  // Like Check() but does some more comprehensive checking.
+  bool CheckExpensive();
+  bool CheckList(Span* list, Length min_pages, Length max_pages,
+                 int freelist);  // ON_NORMAL_FREELIST or ON_RETURNED_FREELIST
+
+  // Try to release at least num_pages for reuse by the OS.  Returns
+  // the actual number of pages released, which may be less than
+  // num_pages if there weren't enough pages to release. The result
+  // may also be larger than num_pages since page_heap might decide to
+  // release one large range instead of fragmenting it into two
+  // smaller released and unreleased ranges.
+  Length ReleaseAtLeastNPages(Length num_pages);
+
+  // Return 0 if we have no information, or else the correct sizeclass for p.
+  // Reads and writes to pagemap_cache_ do not require locking.
+  // The entries are 64 bits on 64-bit hardware and 16 bits on
+  // 32-bit hardware, and we don't mind raciness as long as each read of
+  // an entry yields a valid entry, not a partially updated entry.
+  size_t GetSizeClassIfCached(PageID p) const {
+    return pagemap_cache_.GetOrDefault(p, 0);
+  }
+  void CacheSizeClass(PageID p, size_t cl) const { pagemap_cache_.Put(p, cl); }
+
+ public:
+  // Allocates a big block of memory for the pagemap once we reach more than
+  // 128MB
+  static const size_t kPageMapBigAllocationThreshold = 128 << 20;
+
+  // Minimum number of pages to fetch from system at a time.  Must be
+  // significantly bigger than kBlockSize to amortize system-call
+  // overhead, and also to reduce external fragementation.  Also, we
+  // should keep this value big because various incarnations of Linux
+  // have small limits on the number of mmap() regions per
+  // address-space.
+  // REQUIRED: kMinSystemAlloc <= kMaxPages;
+  static const int kMinSystemAlloc = kMaxPages;
+
+  // Never delay scavenging for more than the following number of
+  // deallocated pages.  With 4K pages, this comes to 4GB of
+  // deallocation.
+  static const int kMaxReleaseDelay = 1 << 20;
+
+  // If there is nothing to release, wait for so many pages before
+  // scavenging again.  With 4K pages, this comes to 1GB of memory.
+  static const int kDefaultReleaseDelay = 1 << 18;
+
+  static void* sAddressStart;
+  static void* sAddressEnd;
+
+  // Pick the appropriate map and cache types based on pointer size
+  typedef MapSelector<kAddressBits>::Type PageMap;
+  typedef MapSelector<kAddressBits>::CacheType PageMapCache;
+  PageMap pagemap_;
+  mutable PageMapCache pagemap_cache_;
+
+  // We segregate spans of a given size into two circular linked
+  // lists: one for normal spans, and one for spans whose memory
+  // has been returned to the system.
+  struct SpanList {
+    Span        normal;
+    Span        returned;
+  };
+
+  // List of free spans of length >= kMaxPages
+  SpanList large_;
+
+  // Array mapping from span length to a doubly linked list of free spans
+  SpanList free_[kMaxPages];
+
+  // Statistics on system, free, and unmapped bytes
+  Stats stats_;
+
+  Span* SearchFreeAndLargeLists(Length n);
+
+  bool GrowHeap(Length n);
+
+  // REQUIRES: span->length >= n
+  // REQUIRES: span->location != IN_USE
+  // Remove span from its free list, and move any leftover part of
+  // span into appropriate free lists.  Also update "span" to have
+  // length exactly "n" and mark it as non-free so it can be returned
+  // to the client.  After all that, decrease free_pages_ by n and
+  // return span.
+  Span* Carve(Span* span, Length n);
+
+  void RecordSpan(Span* span) {
+    pagemap_.set(span->start, span);
+    if (span->length > 1) {
+      pagemap_.set(span->start + span->length - 1, span);
+    }
+  }
+
+  // Allocate a large span of length == n.  If successful, returns a
+  // span of exactly the specified length.  Else, returns NULL.
+  Span* AllocLarge(Length n);
+
+  // Coalesce span with neighboring spans if possible, prepend to
+  // appropriate free list, and adjust stats.
+  void MergeIntoFreeList(Span* span);
+
+  // Prepends span to appropriate free list, and adjusts stats.
+  void PrependToFreeList(Span* span);
+
+  // Removes span from its free list, and adjust stats.
+  void RemoveFromFreeList(Span* span);
+
+  // Incrementally release some memory to the system.
+  // IncrementalScavenge(n) is called whenever n pages are freed.
+  void IncrementalScavenge(Length n);
+
+  // Release the last span on the normal portion of this list.
+  // Return the length of that span or zero if release failed.
+  Length ReleaseLastNormalSpan(SpanList* slist);
+
+  // Checks if we are allowed to take more memory from the system.
+  // If limit is reached and allowRelease is true, tries to release
+  // some unused spans.
+  bool EnsureLimit(Length n, bool allowRelease = true);
+
+  // Number of pages to deallocate before doing more scavenging
+  int64_t scavenge_counter_;
+
+  // Index of last free list where we released memory to the OS.
+  int release_index_;
+};
+
+}  // namespace TCMALLOC_NAMESPACE
+
+#ifdef _MSC_VER
+#pragma warning(pop)
+#endif
+
+#endif  // TCMALLOC_PAGE_HEAP_H_

BeefRT/gperftools/src/page_heap_allocator.h

@@ -0,0 +1,113 @@
+// Copyright (c) 2008, Google Inc.
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+
+#ifndef TCMALLOC_PAGE_HEAP_ALLOCATOR_H_
+#define TCMALLOC_PAGE_HEAP_ALLOCATOR_H_
+
+#include <stddef.h>                     // for NULL, size_t
+
+#include "common.h"            // for MetaDataAlloc
+#include "internal_logging.h"  // for ASSERT
+
+namespace TCMALLOC_NAMESPACE {
+
+// Simple allocator for objects of a specified type.  External locking
+// is required before accessing one of these objects.
+template <class T>
+class PageHeapAllocator {
+ public:
+  // We use an explicit Init function because these variables are statically
+  // allocated and their constructors might not have run by the time some
+  // other static variable tries to allocate memory.
+  void Init() {
+    ASSERT(sizeof(T) <= kAllocIncrement);
+    inuse_ = 0;
+    free_area_ = NULL;
+    free_avail_ = 0;
+    free_list_ = NULL;
+    // Reserve some space at the beginning to avoid fragmentation.
+    Delete(New());
+  }
+
+  T* New() {
+    // Consult free list
+    void* result;
+    if (free_list_ != NULL) {
+      result = free_list_;
+      free_list_ = *(reinterpret_cast<void**>(result));
+    } else {
+      if (free_avail_ < sizeof(T)) {
+        // Need more room. We assume that MetaDataAlloc returns
+        // suitably aligned memory.
+        free_area_ = reinterpret_cast<char*>(MetaDataAlloc(kAllocIncrement));
+        if (free_area_ == NULL) {
+          Log(kCrash, __FILE__, __LINE__,
+              "FATAL ERROR: Out of memory trying to allocate internal "
+              "tcmalloc data (bytes, object-size)",
+              kAllocIncrement, sizeof(T));
+        }
+        free_avail_ = kAllocIncrement;
+      }
+      result = free_area_;
+      free_area_ += sizeof(T);
+      free_avail_ -= sizeof(T);
+    }
+    inuse_++;
+    return reinterpret_cast<T*>(result);
+  }
+
+  void Delete(T* p) {
+    *(reinterpret_cast<void**>(p)) = free_list_;
+    free_list_ = p;
+    inuse_--;
+  }
+
+  int inuse() const { return inuse_; }
+
+ private:
+  // How much to allocate from system at a time
+  static const int kAllocIncrement = 128 << 10;
+
+  // Free area from which to carve new objects
+  char* free_area_;
+  size_t free_avail_;
+
+  // Free list of already carved objects
+  void* free_list_;
+
+  // Number of allocated but unfreed objects
+  int inuse_;
+};
+
+}  // namespace TCMALLOC_NAMESPACE
+
+#endif  // TCMALLOC_PAGE_HEAP_ALLOCATOR_H_

BeefRT/gperftools/src/pagemap.h

@@ -0,0 +1,324 @@
+// Copyright (c) 2005, Google Inc.
+// All rights reserved.
+// 
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+// 
+//     * Redistributions of source code must retain the above copyright
+// notice, this list of conditions and the following disclaimer.
+//     * Redistributions in binary form must reproduce the above
+// copyright notice, this list of conditions and the following disclaimer
+// in the documentation and/or other materials provided with the
+// distribution.
+//     * Neither the name of Google Inc. nor the names of its
+// contributors may be used to endorse or promote products derived from
+// this software without specific prior written permission.
+// 
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+// ---
+// Author: Sanjay Ghemawat <opensource@google.com>
+//
+// A data structure used by the caching malloc.  It maps from page# to
+// a pointer that contains info about that page.  We use two
+// representations: one for 32-bit addresses, and another for 64 bit
+// addresses.  Both representations provide the same interface.  The
+// first representation is implemented as a flat array, the seconds as
+// a three-level radix tree that strips away approximately 1/3rd of
+// the bits every time.
+//
+// The BITS parameter should be the number of bits required to hold
+// a page number.  E.g., with 32 bit pointers and 4K pages (i.e.,
+// page offset fits in lower 12 bits), BITS == 20.
+
+#ifndef TCMALLOC_PAGEMAP_H_
+#define TCMALLOC_PAGEMAP_H_
+
+#include "tcconfig.h"
+
+#include <stddef.h>                     // for NULL, size_t
+#include <string.h>                     // for memset
+#if defined HAVE_STDINT_H
+#include <stdint.h>
+#elif defined HAVE_INTTYPES_H
+#include <inttypes.h>
+#else
+#include <sys/types.h>
+#endif
+#include "internal_logging.h"  // for ASSERT
+
+// Single-level array
+template <int BITS>
+class TCMalloc_PageMap1 {
+ private:
+  static const int LENGTH = 1 << BITS;
+
+  void** array_;
+
+ public:
+  typedef uintptr_t Number;
+
+  explicit TCMalloc_PageMap1(void* (*allocator)(size_t)) {
+    array_ = reinterpret_cast<void**>((*allocator)(sizeof(void*) << BITS));
+    memset(array_, 0, sizeof(void*) << BITS);
+  }
+
+  // Ensure that the map contains initialized entries "x .. x+n-1".
+  // Returns true if successful, false if we could not allocate memory.
+  bool Ensure(Number x, size_t n) {
+    // Nothing to do since flat array was allocated at start.  All
+    // that's left is to check for overflow (that is, we don't want to
+    // ensure a number y where array_[y] would be an out-of-bounds
+    // access).
+    return n <= LENGTH - x;   // an overflow-free way to do "x + n <= LENGTH"
+  }
+
+  void PreallocateMoreMemory() {}
+
+  // Return the current value for KEY.  Returns NULL if not yet set,
+  // or if k is out of range.
+  void* get(Number k) const {
+    if ((k >> BITS) > 0) {
+      return NULL;
+    }
+    return array_[k];
+  }
+
+  // REQUIRES "k" is in range "[0,2^BITS-1]".
+  // REQUIRES "k" has been ensured before.
+  //
+  // Sets the value 'v' for key 'k'.
+  void set(Number k, void* v) {
+    array_[k] = v;
+  }
+
+  // Return the first non-NULL pointer found in this map for
+  // a page number >= k.  Returns NULL if no such number is found.
+  void* Next(Number k) const {
+    while (k < (1 << BITS)) {
+      if (array_[k] != NULL) return array_[k];
+      k++;
+    }
+    return NULL;
+  }
+};
+
+// Two-level radix tree
+template <int BITS>
+class TCMalloc_PageMap2 {
+ public:
+  // Put 32 entries in the root and (2^BITS)/32 entries in each leaf.
+  static const int ROOT_BITS = 5;
+  static const int ROOT_LENGTH = 1 << ROOT_BITS;
+
+  static const int LEAF_BITS = BITS - ROOT_BITS;
+  static const int LEAF_LENGTH = 1 << LEAF_BITS;
+
+  // Leaf node
+  struct Leaf {
+    void* values[LEAF_LENGTH];
+  };
+
+  Leaf* root_[ROOT_LENGTH];             // Pointers to 32 child nodes
+  void* (*allocator_)(size_t);          // Memory allocator
+
+ public:
+  typedef uintptr_t Number;
+
+  explicit TCMalloc_PageMap2(void* (*allocator)(size_t)) {
+    allocator_ = allocator;
+    memset(root_, 0, sizeof(root_));
+  }
+
+  void* get(Number k) const {
+    const Number i1 = k >> LEAF_BITS;
+    const Number i2 = k & (LEAF_LENGTH-1);
+    if ((k >> BITS) > 0 || root_[i1] == NULL) {
+      return NULL;
+    }
+    return root_[i1]->values[i2];
+  }
+
+  void set(Number k, void* v) {
+    ASSERT(k >> BITS == 0);
+
+    const Number i1 = k >> LEAF_BITS;
+    const Number i2 = k & (LEAF_LENGTH-1);
+    root_[i1]->values[i2] = v;
+  }
+
+  bool Ensure(Number start, size_t n) {
+    for (Number key = start; key <= start + n - 1; ) {
+      const Number i1 = key >> LEAF_BITS;
+
+      // Check for overflow
+      if (i1 >= ROOT_LENGTH)
+        return false;
+
+      // Make 2nd level node if necessary
+      if (root_[i1] == NULL) {
+        Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)(sizeof(Leaf)));
+        if (leaf == NULL) return false;
+        memset(leaf, 0, sizeof(*leaf));
+        root_[i1] = leaf;
+      }
+
+      // Advance key past whatever is covered by this leaf node
+      key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
+    }
+    return true;
+  }
+
+  void PreallocateMoreMemory() {
+    // Allocate enough to keep track of all possible pages
+    Ensure(0, 1 << BITS);
+  }
+
+  void* Next(Number k) const {
+    while (k < (1 << BITS)) {
+      const Number i1 = k >> LEAF_BITS;
+      Leaf* leaf = root_[i1];
+      if (leaf != NULL) {
+        // Scan forward in leaf
+        for (Number i2 = k & (LEAF_LENGTH - 1); i2 < LEAF_LENGTH; i2++) {
+          if (leaf->values[i2] != NULL) {
+            return leaf->values[i2];
+          }
+        }
+      }
+      // Skip to next top-level entry
+      k = (i1 + 1) << LEAF_BITS;
+    }
+    return NULL;
+  }
+};
+
+// Three-level radix tree
+template <int BITS>
+class TCMalloc_PageMap3 {
+ public:
+  // How many bits should we consume at each interior level
+  static const int INTERIOR_BITS = (BITS + 2) / 3; // Round-up
+  static const int INTERIOR_LENGTH = 1 << INTERIOR_BITS;
+
+  // How many bits should we consume at leaf level
+  static const int LEAF_BITS = BITS - 2*INTERIOR_BITS;
+  static const int LEAF_LENGTH = 1 << LEAF_BITS;
+
+  // Interior node
+  struct Node {
+    Node* ptrs[INTERIOR_LENGTH];
+  };
+
+  // Leaf node
+  struct Leaf {
+    void* values[LEAF_LENGTH];
+  };
+
+  Node* root_;                          // Root of radix tree
+  void* (*allocator_)(size_t);          // Memory allocator
+
+  Node* NewNode() {
+    Node* result = reinterpret_cast<Node*>((*allocator_)(sizeof(Node)));
+    if (result != NULL) {
+      memset(result, 0, sizeof(*result));
+    }
+    return result;
+  }
+
+ public:
+  typedef uintptr_t Number;
+
+  explicit TCMalloc_PageMap3(void* (*allocator)(size_t)) {
+    allocator_ = allocator;
+    root_ = NewNode();
+  }
+
+  void* get(Number k) const {
+    const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
+    const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1);
+    const Number i3 = k & (LEAF_LENGTH-1);
+    if ((k >> BITS) > 0 ||
+        root_->ptrs[i1] == NULL || root_->ptrs[i1]->ptrs[i2] == NULL) {
+      return NULL;
+    }
+    return reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3];
+  }
+
+  void set(Number k, void* v) {
+    ASSERT(k >> BITS == 0);
+    const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
+    const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1);
+    const Number i3 = k & (LEAF_LENGTH-1);
+    reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2])->values[i3] = v;
+  }
+
+  bool Ensure(Number start, size_t n) {
+    for (Number key = start; key <= start + n - 1; ) {
+      const Number i1 = key >> (LEAF_BITS + INTERIOR_BITS);
+      const Number i2 = (key >> LEAF_BITS) & (INTERIOR_LENGTH-1);
+
+      // Check for overflow
+      if (i1 >= INTERIOR_LENGTH || i2 >= INTERIOR_LENGTH)
+        return false;
+
+      // Make 2nd level node if necessary
+      if (root_->ptrs[i1] == NULL) {
+        Node* n = NewNode();
+        if (n == NULL) return false;
+        root_->ptrs[i1] = n;
+      }
+
+      // Make leaf node if necessary
+      if (root_->ptrs[i1]->ptrs[i2] == NULL) {
+        Leaf* leaf = reinterpret_cast<Leaf*>((*allocator_)(sizeof(Leaf)));
+        if (leaf == NULL) return false;
+        memset(leaf, 0, sizeof(*leaf));
+        root_->ptrs[i1]->ptrs[i2] = reinterpret_cast<Node*>(leaf);
+      }
+
+      // Advance key past whatever is covered by this leaf node
+      key = ((key >> LEAF_BITS) + 1) << LEAF_BITS;
+    }
+    return true;
+  }
+
+  void PreallocateMoreMemory() {
+  }
+
+  void* Next(Number k) const {
+    while (k < (Number(1) << BITS)) {
+      const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS);
+      const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1);
+      if (root_->ptrs[i1] == NULL) {
+        // Advance to next top-level entry
+        k = (i1 + 1) << (LEAF_BITS + INTERIOR_BITS);
+      } else {
+        Leaf* leaf = reinterpret_cast<Leaf*>(root_->ptrs[i1]->ptrs[i2]);
+        if (leaf != NULL) {
+          for (Number i3 = (k & (LEAF_LENGTH-1)); i3 < LEAF_LENGTH; i3++) {
+            if (leaf->values[i3] != NULL) {
+              return leaf->values[i3];
+            }
+          }
+        }
+        // Advance to next interior entry
+        k = ((k >> LEAF_BITS) + 1) << LEAF_BITS;
+      }
+    }
+    return NULL;
+  }
+};
+
+#endif  // TCMALLOC_PAGEMAP_H_