Add OOP memory logging allocator shim

chrome/common/profiling/memlog_allocator_shim.cc

+// Copyright 2017 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "chrome/common/profiling/memlog_allocator_shim.h"
+
+#include "base/allocator/allocator_shim.h"
+#include "base/debug/debugging_flags.h"
+#include "base/debug/stack_trace.h"
+#include "base/synchronization/lock.h"
+#include "base/trace_event/heap_profiler_allocation_register.h"
+#include "chrome/common/profiling/memlog_stream.h"
+
+namespace profiling {
+
+namespace {
+
+using base::allocator::AllocatorDispatch;
+
+MemlogSenderPipe* g_sender_pipe = nullptr;
+
+// This maximum number of stack entries to log. Long-term we will likely want
+// to raise this to avoid truncation. This matches the current value in the
+// in-process heap profiler (heap_profiler_allocation_context.h) so the two
+// systems performance and memory overhead can be compared consistently.
+const int kMaxStackEntries = 48;
+
+// Matches the native buffer size on the pipe.
+const int kSendBufferSize = 65536;
+
+// Prime since this is used like a hash table. Numbers of this magnitude seemed
+// to provide sufficient parallelism to avoid lock overhead in ad-hoc testing.
+const int kNumSendBuffers = 17;
+
+class SendBuffer {
+ public:
+  SendBuffer() : buffer_(new char[kSendBufferSize]) {}
+  ~SendBuffer() { delete[] buffer_; }
+
+  void Send(const void* data, size_t sz) {
+    base::AutoLock lock(lock_);
+
+    if (used_ + sz > kSendBufferSize)
+      SendCurrentBuffer();
+
+    memcpy(&buffer_[used_], data, sz);
+    used_ += sz;
+  }
+
+ private:
+  void SendCurrentBuffer() {
+    g_sender_pipe->Send(buffer_, used_);
+    used_ = 0;
+  }
+
+  base::Lock lock_;
+
+  char* buffer_;
+  size_t used_ = 0;
+
+  DISALLOW_COPY_AND_ASSIGN(SendBuffer);
+};
+
+SendBuffer* g_send_buffers = nullptr;
+
+// "address" is the address in question, which is used to select which send
+// buffer to use.
+void DoSend(const void* address, const void* data, size_t size) {
+  base::trace_event::AllocationRegister::AddressHasher hasher;
+  int bin_to_use = hasher(address) % kNumSendBuffers;
+  g_send_buffers[bin_to_use].Send(data, size);
+}
+
+void* HookAlloc(const AllocatorDispatch* self, size_t size, void* context) {
+  const AllocatorDispatch* const next = self->next;
+  void* ptr = next->alloc_function(next, size, context);
+  AllocatorShimLogAlloc(ptr, size);
+  return ptr;
+}
+
+void* HookZeroInitAlloc(const AllocatorDispatch* self,
+                        size_t n,
+                        size_t size,
+                        void* context) {
+  const AllocatorDispatch* const next = self->next;
+  void* ptr = next->alloc_zero_initialized_function(next, n, size, context);
+  AllocatorShimLogAlloc(ptr, n * size);
+  return ptr;
+}
+
+void* HookAllocAligned(const AllocatorDispatch* self,
+                       size_t alignment,
+                       size_t size,
+                       void* context) {
+  const AllocatorDispatch* const next = self->next;
+  void* ptr = next->alloc_aligned_function(next, alignment, size, context);
+  AllocatorShimLogAlloc(ptr, size);
+  return ptr;
+}
+
+void* HookRealloc(const AllocatorDispatch* self,
+                  void* address,
+                  size_t size,
+                  void* context) {
+  const AllocatorDispatch* const next = self->next;
+  void* ptr = next->realloc_function(next, address, size, context);
+  AllocatorShimLogFree(address);
+  if (size > 0)  // realloc(size == 0) means free()
+    AllocatorShimLogAlloc(ptr, size);
+  return ptr;
+}
+
+void HookFree(const AllocatorDispatch* self, void* address, void* context) {
+  AllocatorShimLogFree(address);
+  const AllocatorDispatch* const next = self->next;
+  next->free_function(next, address, context);
+}
+
+size_t HookGetSizeEstimate(const AllocatorDispatch* self,
+                           void* address,
+                           void* context) {
+  const AllocatorDispatch* const next = self->next;
+  return next->get_size_estimate_function(next, address, context);
+}
+
+unsigned HookBatchMalloc(const AllocatorDispatch* self,
+                         size_t size,
+                         void** results,
+                         unsigned num_requested,
+                         void* context) {
+  const AllocatorDispatch* const next = self->next;
+  unsigned count =
+      next->batch_malloc_function(next, size, results, num_requested, context);
+  for (unsigned i = 0; i < count; ++i)
+    AllocatorShimLogAlloc(results[i], size);
+  return count;
+}
+
+void HookBatchFree(const AllocatorDispatch* self,
+                   void** to_be_freed,
+                   unsigned num_to_be_freed,
+                   void* context) {
+  const AllocatorDispatch* const next = self->next;
+  for (unsigned i = 0; i < num_to_be_freed; ++i)
+    AllocatorShimLogFree(to_be_freed[i]);
+  next->batch_free_function(next, to_be_freed, num_to_be_freed, context);
+}
+
+void HookFreeDefiniteSize(const AllocatorDispatch* self,
+                          void* ptr,
+                          size_t size,
+                          void* context) {
+  AllocatorShimLogFree(ptr);
+  const AllocatorDispatch* const next = self->next;
+  next->free_definite_size_function(next, ptr, size, context);
+}
+
+AllocatorDispatch g_memlog_hooks = {
+    &HookAlloc,             // alloc_function
+    &HookZeroInitAlloc,     // alloc_zero_initialized_function
+    &HookAllocAligned,      // alloc_aligned_function
+    &HookRealloc,           // realloc_function
+    &HookFree,              // free_function
+    &HookGetSizeEstimate,   // get_size_estimate_function
+    &HookBatchMalloc,       // batch_malloc_function
+    &HookBatchFree,         // batch_free_function
+    &HookFreeDefiniteSize,  // free_definite_size_function
+    nullptr,                // next
+};
+
+}  // namespace
+
+void InitAllocatorShim(MemlogSenderPipe* sender_pipe) {
+  g_send_buffers = new SendBuffer[kNumSendBuffers];
+
+  g_sender_pipe = sender_pipe;
+#ifdef NDEBUG
+  base::allocator::InsertAllocatorDispatch(&g_memlog_hooks);
+#endif
+}
+
+void AllocatorShimLogAlloc(void* address, size_t sz) {
+  if (!g_send_buffers)
+    return;
+  if (address) {
+    constexpr size_t max_message_size =
+        sizeof(AllocPacket) + kMaxStackEntries * sizeof(uint64_t);
+    char message[max_message_size];

[Primiano Tucci (use gerrit), 2017/06/21 14:11:29]

Doesn't this have to be declared  ALIGNAS(sizeof(uint64_t)) ? 
If this is not aligned all the assignment operations below in lines 209-217 will
very silently be way slower on x86/64 and eventualy SIGBUS the day that you use
this on arm.

+    // TODO(ajwong) check that this is technically valid.
+    AllocPacket* alloc_packet = reinterpret_cast<AllocPacket*>(message);
+
+    uint64_t* stack =
+        reinterpret_cast<uint64_t*>(&message[sizeof(AllocPacket)]);
+
+#if BUILDFLAG(CAN_UNWIND_WITH_FRAME_POINTERS)
+    const void* frames[kMaxStackEntries];
+    size_t frame_count = base::debug::TraceStackFramePointers(
+        frames, kMaxStackEntries,
+        1);  // exclude this function from the trace.
+#else   // BUILDFLAG(CAN_UNWIND_WITH_FRAME_POINTERS)
+    // Fall-back to capturing the stack with base::debug::StackTrace,
+    // which is likely slower, but more reliable.
+    base::debug::StackTrace stack_trace(kMaxStackEntries);
+    size_t frame_count = 0u;
+    const void* const* frames = stack_trace.Addresses(&frame_count);
+#endif  // BUILDFLAG(CAN_UNWIND_WITH_FRAME_POINTERS)
+
+    // If there are too many frames, keep the ones furthest from main().
+    for (size_t i = 0; i < frame_count; i++)
+      stack[i] = (uint64_t)frames[i];
+
+    alloc_packet->op = kAllocPacketType;
+    alloc_packet->time = 0;  // TODO(brettw) add timestamp.
+    alloc_packet->address = (uint64_t)address;
+    alloc_packet->size = sz;
+    alloc_packet->stack_len = static_cast<uint32_t>(frame_count);
+
+    DoSend(address, message,
+           sizeof(AllocPacket) + alloc_packet->stack_len * sizeof(uint64_t));
+  }
+}
+
+void AllocatorShimLogFree(void* address) {
+  if (!g_send_buffers)
+    return;
+  if (address) {
+    FreePacket free_packet;
+    free_packet.op = kFreePacketType;
+    free_packet.time = 0;  // TODO(brettw) add timestamp.
+    free_packet.address = (uint64_t)address;
+
+    DoSend(address, &free_packet, sizeof(FreePacket));
+  }
+}
+
+}  // namespace profiling