Record Chrome trace events in tcmalloc heap profiles

base/debug/trace_memory.cc

+// Copyright 2013 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 "base/debug/trace_memory.h"
+
+#include "base/debug/trace_event.h"
+#include "base/lazy_instance.h"
+#include "base/logging.h"
+#include "base/memory/scoped_ptr.h"
+#include "base/message_loop.h"
+#include "base/threading/thread_local.h"
+
+#if !defined(NO_TCMALLOC) && !defined(OS_NACL) && defined(OS_LINUX)
+#include "third_party/tcmalloc/chromium/src/gperftools/heap-profiler.h"
+#endif
+
+namespace base {
+
+namespace {
+
+// Maximum number of nested TRACE_MEMORY macros.
+const int kMaxStackSize = 32;
+
+/////////////////////////////////////////////////////////////////////////////
+// Holds a memory dump until the tracing system needs to serialize it.
+class MemoryDumpHolder : public base::debug::ConvertableToTraceFormat {
+ public:
+  // Takes ownership of dump, which must be a JSON string, allocated with
+  // malloc() and NULL terminated.
+  explicit MemoryDumpHolder(char* dump) : dump_(dump) {}
+  virtual ~MemoryDumpHolder() { free(dump_); }
+
+  // base::debug::ConvertableToTraceFormat overrides:
+  virtual void AppendAsTraceFormat(std::string* out) const OVERRIDE {
+    out->append(dump_);
+  }
+
+ private:
+  char* dump_;
+
+  DISALLOW_COPY_AND_ASSIGN(MemoryDumpHolder);
+};
+
+/////////////////////////////////////////////////////////////////////////////
+// Records a stack of TRACE_MEMORY events. One per thread is required.
+struct TraceMemoryStack {
+  TraceMemoryStack() : index_(0) {
+    memset(category_stack_, 0, kMaxStackSize * sizeof(const char*));
+  }
+
+  // Points to the next free entry.
+  int index_;
+  const char* category_stack_[kMaxStackSize];
+};
+
+// A stack of TRACE_MEMORY events, per thread.
+LazyInstance<ThreadLocalPointer<TraceMemoryStack> >::Leaky trace_memory_stack =
+    LAZY_INSTANCE_INITIALIZER;
+
+// Returns a "pseudo-stack" of pointers to trace events.
+// TODO(jamescook): Record both category and name, perhaps in a pair for speed.
+int GetPseudoStack(void** stack_out) {
+  // Is this a fast enough lookup? Should we avoid LazyInstance somehow?
+  TraceMemoryStack* stack = trace_memory_stack.Get().Get();
+  // TODO - figure out how to initialize this on every thread so we don't need
+  // this if().
+  if (!stack)
+    return 0;  // could also init the stack here
+  const int count = stack->index_;
+  // memcpy works for zero bytes.
+  memcpy(stack_out, stack->category_stack_, count * sizeof(void*));
+  return count;
+}
+
+// Polls for memory tracing being enabled. If so, dumps a memory profile every
+// few seconds.
+// TODO(jamescook): Don't poll. Get a signal when tracing is enabled.
+void DumpMemoryProfile() {
+  // Don't trace allocations here in the memory tracing system.
+  TRACE_MEMORY(TRACE_DISABLED_BY_DEFAULT("memory"), TRACE_MEMORY_IGNORE);
+  // Check to see if tracing is enabled for the memory category.
+  bool enabled;
+  TRACE_EVENT_CATEGORY_GROUP_ENABLED(TRACE_DISABLED_BY_DEFAULT("memory"),
+                                     &enabled);
+  if (enabled) {
+    // We take ownership of this string.
+    char* dump = base::TraceMemoryDumpAsString();
+    scoped_ptr<MemoryDumpHolder> dump_holder(new MemoryDumpHolder(dump));
+    const int kSnapshotId = 1;
+    TRACE_EVENT_OBJECT_SNAPSHOT_WITH_ID(
+        "memory",
+        "memory::Heap",
+        kSnapshotId,
+        dump_holder.PassAs<base::debug::ConvertableToTraceFormat>());
+  }
+  base::MessageLoop::current()->PostDelayedTask(
+      FROM_HERE,
+      base::Bind(&DumpMemoryProfile),
+      base::TimeDelta::FromSeconds(5));
+}
+
+}  // namespace
+
+ScopedTraceMemory::ScopedTraceMemory(const char* category) {
+  // Get our thread's copy of the stack.
+  TraceMemoryStack* stack = trace_memory_stack.Get().Get();
+  if (!stack) {
+    trace_memory_stack.Get().Set(new TraceMemoryStack);
+    stack = trace_memory_stack.Get().Get();
+  }
+  // Bounds checking is for the weak.
+  const int index = stack->index_;
+  stack->category_stack_[index] = category;
+  stack->index_++;
+}
+
+ScopedTraceMemory::~ScopedTraceMemory() {
+  // Get our thread's copy of the stack.
+  TraceMemoryStack* stack = trace_memory_stack.Get().Get();
+  stack->index_--;
+}
+
+// static
+int ScopedTraceMemory::GetStackIndexForTest() {
+  return trace_memory_stack.Get().Get()->index_;
+}
+
+void TraceMemoryStart() {
+#if !defined(NO_TCMALLOC) && !defined(OS_NACL) && defined(OS_LINUX)
+  DVLOG(1) << "Starting trace memory";
+  // Ensure thread-local-storage is initialized by creating a dummy event.
+  ScopedTraceMemory initialize(TRACE_MEMORY_IGNORE);
+  ::SetPseudoStackGenerator(&GetPseudoStack);
+  ::HeapProfilerStart("/tmp/trace-memory");
+  // This won't do anything initially, but will get the polling timer going.
+  DumpMemoryProfile();
+#endif
+}
+
+void TraceMemoryDump() {
+#if !defined(NO_TCMALLOC) && !defined(OS_NACL) && defined(OS_LINUX)
+  DVLOG(1) << "Dumping trace memory";
+  ::HeapProfilerDump("dump-for-trace-memory");
+#endif
+}
+
+char* TraceMemoryDumpAsString() {
+#if !defined(NO_TCMALLOC) && !defined(OS_NACL) && defined(OS_LINUX)
+  DVLOG(1) << "Getting memory dump as string";
+  return ::GetHeapProfile();
+#endif
+  return NULL;
+}
+
+void TraceMemoryStop() {
+#if !defined(NO_TCMALLOC) && !defined(OS_NACL) && defined(OS_LINUX)
+  DVLOG(1) << "Stopping trace memory";
+  ::HeapProfilerStop();
+#endif
+}
+
+}  // namespace base