Record Chrome trace events in tcmalloc heap profiles

third_party/tcmalloc/chromium/src/heap-profiler.cc

Index: third_party/tcmalloc/chromium/src/heap-profiler.cc
diff --git a/third_party/tcmalloc/chromium/src/heap-profiler.cc b/third_party/tcmalloc/chromium/src/heap-profiler.cc
index 47130e8bc5fa8e593a9378a4671a89175ab83884..94eb548f81d49c12b6a8a2e7b7b295952197d340 100644
--- a/third_party/tcmalloc/chromium/src/heap-profiler.cc
+++ b/third_party/tcmalloc/chromium/src/heap-profiler.cc
@@ -216,6 +216,9 @@ static int64 last_dump_time = 0;      // The time of the last dump
 static HeapProfileTable* heap_profile = NULL;  // the heap profile table
 static DeepHeapProfile* deep_profile = NULL;  // deep memory profiler
 
+// Callback an appplication can use to generate its own "stacks".
+static PseudoStackGenerator pseudo_stack_generator = NULL;
+
 //----------------------------------------------------------------------
 // Profile generation
 //----------------------------------------------------------------------
@@ -232,7 +235,11 @@ static char* DoGetHeapProfileLocked(char* buf, int buflen) {
   if (is_on) {
     HeapProfileTable::Stats const stats = heap_profile->total();
     (void)stats;   // avoid an unused-variable warning in non-debug mode.
-    bytes_written = heap_profile->FillOrderedProfile(buf, buflen - 1);
+    if (pseudo_stack_generator) {
+      bytes_written = heap_profile->FillPseudoStackProfile(buf, buflen - 1);
+    } else {
+      bytes_written = heap_profile->FillOrderedProfile(buf, buflen - 1);
+    }
     // FillOrderedProfile should not reduce the set of active mmap-ed regions,
     // hence MemoryRegionMap will let us remove everything we've added above:
     RAW_DCHECK(stats.Equivalent(heap_profile->total()), "");
@@ -376,6 +383,21 @@ static void RecordAlloc(const void* ptr, size_t bytes, int skip_count) {
   }
 }
 
+// Record an allocation in the profile.
+static void PseudoStackRecordAlloc(const void* ptr,
+                                   size_t bytes,
+                                   int skip_count) {

[Dai Mikurube (NOT FULLTIME), 2013/06/19 04:34:03]

It's almost the same with RecordAlloc().  Why do we need this additional
function?

Also, don't we need RecordFree?

[James Cook, 2013/06/29 00:02:42]

I elected to copy the function because this is performance critical code inside
the allocator - I thought it was better to do this than to combine the common
parts of RecordAlloc and PseudoStackRecordAlloc and introduce an extra function
call or branch.

RecordFree is identical in both cases, the pseudo-stack push/pops are handled in
the Chrome trace-event level and don't need any information about frees.

+  // Take the stack trace outside the critical section.
+  void* stack[HeapProfileTable::kMaxStackDepth];
+  // Generate our pseudo-stack by via callback into the client code.
+  int depth = (*pseudo_stack_generator)(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);
@@ -399,6 +421,11 @@ void DeleteHook(const void* ptr) {
   if (ptr != NULL) RecordFree(ptr);
 }
 
+// static
+void PseudoStackNewHook(const void* ptr, size_t size) {

[Dai Mikurube (NOT FULLTIME), 2013/06/19 04:34:03]

ditto.

+  if (ptr != NULL) PseudoStackRecordAlloc(ptr, size, 0);
+}
+
 // TODO(jandrews): Re-enable stack tracing
 #ifdef TODO_REENABLE_STACK_TRACING
 static void RawInfoStackDumper(const char* message, void*) {
@@ -531,16 +558,22 @@ extern "C" void HeapProfilerStart(const char* prefix) {
 
   if (FLAGS_only_mmap_profile == false) {
     // Now set the hooks that capture new/delete and malloc/free.
-    RAW_CHECK(MallocHook::AddNewHook(&NewHook), "");
+    if (pseudo_stack_generator) {
+      RAW_CHECK(MallocHook::AddNewHook(&PseudoStackNewHook), "");
+    } else {
+      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';
+  // Copy filename prefix if provided.
+  if (prefix) {
+    RAW_DCHECK(filename_prefix == NULL, "");
+    const int prefix_length = strlen(prefix);
+    filename_prefix = reinterpret_cast<char*>(ProfilerMalloc(prefix_length + 1));

[Dai Mikurube (NOT FULLTIME), 2013/06/19 04:34:03]

Over 80?

+    memcpy(filename_prefix, prefix, prefix_length);
+    filename_prefix[prefix_length] = '\0';
+  }
 }
 
 extern "C" void IterateAllocatedObjects(AddressVisitor visitor, void* data) {
@@ -563,7 +596,11 @@ extern "C" void HeapProfilerStop() {
 
   if (FLAGS_only_mmap_profile == false) {
     // Unset our new/delete hooks, checking they were set:
-    RAW_CHECK(MallocHook::RemoveNewHook(&NewHook), "");
+    if (pseudo_stack_generator) {
+      RAW_CHECK(MallocHook::RemoveNewHook(&PseudoStackNewHook), "");
+    } else {
+      RAW_CHECK(MallocHook::RemoveNewHook(&NewHook), "");
+    }
     RAW_CHECK(MallocHook::RemoveDeleteHook(&DeleteHook), "");
   }
   if (FLAGS_mmap_log) {
@@ -635,6 +672,11 @@ extern "C" void HeapProfilerDumpAliveObjects(const char* filename) {
   heap_profile->DumpMarkedObjects(HeapProfileTable::MARK_TWO, filename);
 }
 
+extern "C" void SetPseudoStackGenerator(PseudoStackGenerator callback) {

[Dai Mikurube (NOT FULLTIME), 2013/06/19 04:34:03]

The function name should contain HeapProfiler or something like that since the
function is in the global namespace.

Also, what happens when this function is called after HeapProfilerStart()?  What
happens when this function is called twice?  I think we should hide it, and have
HeapProfilerStartWithPseudoStack() or something like that.  Wdyt?

[James Cook, 2013/06/29 00:02:42]

Good idea. Done.

+  SpinLockHolder l(&heap_lock);
+  pseudo_stack_generator = callback;
+}
+
 //----------------------------------------------------------------------
 // Initialization/finalization code
 //----------------------------------------------------------------------