[profiler] Emit runtime call stats into sampling profile

These are added to the sampler stack trace when RCS are
enabled.
Resource name for a RCS frame is reported as "V8Runtime".
Counter names match ones from src/counters.h

BUG=chromium:660428

Review-Url: https://codereview.chromium.org/2461003002
Cr-Commit-Position: refs/heads/master@{#40658}

src/counters.h

@@ -499,6 +499,7 @@ class RuntimeCallTimer {
   RuntimeCallTimer() {}
   RuntimeCallCounter* counter() { return counter_; }
   base::ElapsedTimer timer() { return timer_; }
+  RuntimeCallTimer* parent() const { return parent_; }
 
  private:
   friend class RuntimeCallStats;

src/profiler/cpu-profiler.cc

@@ -305,7 +305,7 @@ void CpuProfiler::StartProcessorIfNotStarted() {
   // Disable logging when using the new implementation.
   saved_is_logging_ = logger->is_logging_;
   logger->is_logging_ = false;
-  generator_.reset(new ProfileGenerator(profiles_.get()));
+  generator_.reset(new ProfileGenerator(isolate_, profiles_.get()));
   processor_.reset(new ProfilerEventsProcessor(isolate_, generator_.get(),
                                                sampling_interval_));
   logger->SetUpProfilerListener();

src/profiler/profile-generator.cc

@@ -648,8 +648,9 @@ void CpuProfilesCollection::AddPathToCurrentProfiles(
   current_profiles_semaphore_.Signal();
 }
 
-ProfileGenerator::ProfileGenerator(CpuProfilesCollection* profiles)
-    : profiles_(profiles) {}
+ProfileGenerator::ProfileGenerator(Isolate* isolate,
+                                   CpuProfilesCollection* profiles)
+    : isolate_(isolate), profiles_(profiles) {}
 
 void ProfileGenerator::RecordTickSample(const TickSample& sample) {
   std::vector<CodeEntry*> entries;
@@ -670,16 +671,14 @@ void ProfileGenerator::RecordTickSample(const TickSample& sample) {
       // Don't use PC when in external callback code, as it can point
       // inside callback's code, and we will erroneously report
       // that a callback calls itself.
-      entries.push_back(code_map_.FindEntry(
-          reinterpret_cast<Address>(sample.external_callback_entry)));
+      entries.push_back(FindEntry(sample.external_callback_entry));
     } else {
-      CodeEntry* pc_entry =
-          code_map_.FindEntry(reinterpret_cast<Address>(sample.pc));
+      CodeEntry* pc_entry = FindEntry(sample.pc);
       // If there is no pc_entry we're likely in native code.
       // Find out, if top of stack was pointing inside a JS function
       // meaning that we have encountered a frameless invocation.
       if (!pc_entry && !sample.has_external_callback) {
-        pc_entry = code_map_.FindEntry(reinterpret_cast<Address>(sample.tos));
+        pc_entry = FindEntry(sample.tos);
       }
       // If pc is in the function code before it set up stack frame or after the
       // frame was destroyed SafeStackFrameIterator incorrectly thinks that
@@ -712,8 +711,7 @@ void ProfileGenerator::RecordTickSample(const TickSample& sample) {
 
     for (unsigned i = 0; i < sample.frames_count; ++i) {
       Address stack_pos = reinterpret_cast<Address>(sample.stack[i]);
-      CodeEntry* entry = code_map_.FindEntry(stack_pos);
-
+      CodeEntry* entry = FindEntry(stack_pos);
       if (entry) {
         // Find out if the entry has an inlining stack associated.
         int pc_offset =
@@ -756,6 +754,22 @@ void ProfileGenerator::RecordTickSample(const TickSample& sample) {
                                       sample.update_stats);
 }
 
+CodeEntry* ProfileGenerator::FindEntry(void* address) {
+  CodeEntry* entry = code_map_.FindEntry(reinterpret_cast<Address>(address));
+  if (!entry) {
+    RuntimeCallStats* rcs = isolate_->counters()->runtime_call_stats();
+    void* start = reinterpret_cast<void*>(rcs);
+    void* end = reinterpret_cast<void*>(rcs + 1);
+    if (start <= address && address < end) {
+      RuntimeCallCounter* counter =
+          reinterpret_cast<RuntimeCallCounter*>(address);
+      entry = new CodeEntry(CodeEventListener::FUNCTION_TAG, counter->name,
+                            CodeEntry::kEmptyNamePrefix, "native V8Runtime");
+      code_map_.AddCode(reinterpret_cast<Address>(address), entry, 1);
+    }
+  }
+  return entry;
+}
 
 CodeEntry* ProfileGenerator::EntryForVMState(StateTag tag) {
   switch (tag) {

src/profiler/profile-generator.h

@@ -379,15 +379,17 @@ class CpuProfilesCollection {
 
 class ProfileGenerator {
  public:
-  explicit ProfileGenerator(CpuProfilesCollection* profiles);
+  ProfileGenerator(Isolate* isolate, CpuProfilesCollection* profiles);
 
   void RecordTickSample(const TickSample& sample);
 
   CodeMap* code_map() { return &code_map_; }
 
  private:
+  CodeEntry* FindEntry(void* address);
   CodeEntry* EntryForVMState(StateTag tag);
 
+  Isolate* isolate_;
   CpuProfilesCollection* profiles_;
   CodeMap code_map_;
 

src/profiler/tick-sample.cc

@@ -5,6 +5,7 @@
 #include "src/profiler/tick-sample.h"
 
 #include "include/v8-profiler.h"
+#include "src/counters.h"
 #include "src/frames-inl.h"
 #include "src/msan.h"
 #include "src/simulator.h"
@@ -237,7 +238,15 @@ bool TickSample::GetStackSample(Isolate* v8_isolate, RegisterState* regs,
        it.top_frame_type() == internal::StackFrame::BUILTIN_EXIT)) {
     frames[i++] = isolate->c_function();
   }
+  i::RuntimeCallTimer* timer =
+      isolate->counters()->runtime_call_stats()->current_timer();
   for (; !it.done() && i < frames_limit; it.Advance()) {
+    while (timer && reinterpret_cast<i::Address>(timer) < it.frame()->fp() &&
+           i < frames_limit) {
+      frames[i++] = reinterpret_cast<i::Address>(timer->counter());
+      timer = timer->parent();
+    }
+    if (i == frames_limit) break;
     if (!it.frame()->is_interpreted()) {
       frames[i++] = it.frame()->pc();
       continue;

test/cctest/test-cpu-profiler.cc

@@ -71,10 +71,11 @@ static const char* reason(const i::DeoptimizeReason reason) {
 }
 
 TEST(StartStop) {
-  CpuProfilesCollection profiles(CcTest::i_isolate());
-  ProfileGenerator generator(&profiles);
+  i::Isolate* isolate = CcTest::i_isolate();
+  CpuProfilesCollection profiles(isolate);
+  ProfileGenerator generator(isolate, &profiles);
   std::unique_ptr<ProfilerEventsProcessor> processor(
-      new ProfilerEventsProcessor(CcTest::i_isolate(), &generator,
+      new ProfilerEventsProcessor(isolate, &generator,
                                   v8::base::TimeDelta::FromMicroseconds(100)));
   processor->Start();
   processor->StopSynchronously();
@@ -157,10 +158,9 @@ TEST(CodeEvents) {
   i::AbstractCode* args4_code = CreateCode(&env);
 
   CpuProfilesCollection* profiles = new CpuProfilesCollection(isolate);
-  ProfileGenerator* generator = new ProfileGenerator(profiles);
-  ProfilerEventsProcessor* processor =
-      new ProfilerEventsProcessor(CcTest::i_isolate(), generator,
-                                  v8::base::TimeDelta::FromMicroseconds(100));
+  ProfileGenerator* generator = new ProfileGenerator(isolate, profiles);
+  ProfilerEventsProcessor* processor = new ProfilerEventsProcessor(
+      isolate, generator, v8::base::TimeDelta::FromMicroseconds(100));
   CpuProfiler profiler(isolate, profiles, generator, processor);
   profiles->StartProfiling("", false);
   processor->Start();
@@ -226,7 +226,7 @@ TEST(TickEvents) {
   i::AbstractCode* frame3_code = CreateCode(&env);
 
   CpuProfilesCollection* profiles = new CpuProfilesCollection(isolate);
-  ProfileGenerator* generator = new ProfileGenerator(profiles);
+  ProfileGenerator* generator = new ProfileGenerator(isolate, profiles);
   ProfilerEventsProcessor* processor =
       new ProfilerEventsProcessor(CcTest::i_isolate(), generator,
                                   v8::base::TimeDelta::FromMicroseconds(100));
@@ -299,7 +299,7 @@ TEST(Issue1398) {
   i::AbstractCode* code = CreateCode(&env);
 
   CpuProfilesCollection* profiles = new CpuProfilesCollection(isolate);
-  ProfileGenerator* generator = new ProfileGenerator(profiles);
+  ProfileGenerator* generator = new ProfileGenerator(isolate, profiles);
   ProfilerEventsProcessor* processor =
       new ProfilerEventsProcessor(CcTest::i_isolate(), generator,
                                   v8::base::TimeDelta::FromMicroseconds(100));
@@ -1073,7 +1073,7 @@ static void TickLines(bool optimize) {
   CHECK(code_address);
 
   CpuProfilesCollection* profiles = new CpuProfilesCollection(isolate);
-  ProfileGenerator* generator = new ProfileGenerator(profiles);
+  ProfileGenerator* generator = new ProfileGenerator(isolate, profiles);
   ProfilerEventsProcessor* processor =
       new ProfilerEventsProcessor(CcTest::i_isolate(), generator,
                                   v8::base::TimeDelta::FromMicroseconds(100));

test/cctest/test-profile-generator.cc

@@ -344,11 +344,12 @@ class TestSetup {
 
 TEST(RecordTickSample) {
   TestSetup test_setup;
-  CpuProfilesCollection profiles(CcTest::i_isolate());
-  CpuProfiler profiler(CcTest::i_isolate());
+  i::Isolate* isolate = CcTest::i_isolate();
+  CpuProfilesCollection profiles(isolate);
+  CpuProfiler profiler(isolate);
   profiles.set_cpu_profiler(&profiler);
   profiles.StartProfiling("", false);
-  ProfileGenerator generator(&profiles);
+  ProfileGenerator generator(isolate, &profiles);
   CodeEntry* entry1 = new CodeEntry(i::Logger::FUNCTION_TAG, "aaa");
   CodeEntry* entry2 = new CodeEntry(i::Logger::FUNCTION_TAG, "bbb");
   CodeEntry* entry3 = new CodeEntry(i::Logger::FUNCTION_TAG, "ccc");
@@ -416,11 +417,12 @@ static void CheckNodeIds(ProfileNode* node, unsigned* expectedId) {
 
 TEST(SampleIds) {
   TestSetup test_setup;
-  CpuProfilesCollection profiles(CcTest::i_isolate());
-  CpuProfiler profiler(CcTest::i_isolate());
+  i::Isolate* isolate = CcTest::i_isolate();
+  CpuProfilesCollection profiles(isolate);
+  CpuProfiler profiler(isolate);
   profiles.set_cpu_profiler(&profiler);
   profiles.StartProfiling("", true);
-  ProfileGenerator generator(&profiles);
+  ProfileGenerator generator(isolate, &profiles);
   CodeEntry* entry1 = new CodeEntry(i::Logger::FUNCTION_TAG, "aaa");
   CodeEntry* entry2 = new CodeEntry(i::Logger::FUNCTION_TAG, "bbb");
   CodeEntry* entry3 = new CodeEntry(i::Logger::FUNCTION_TAG, "ccc");
@@ -473,11 +475,12 @@ TEST(SampleIds) {
 
 TEST(NoSamples) {
   TestSetup test_setup;
-  CpuProfilesCollection profiles(CcTest::i_isolate());
-  CpuProfiler profiler(CcTest::i_isolate());
+  i::Isolate* isolate = CcTest::i_isolate();
+  CpuProfilesCollection profiles(isolate);
+  CpuProfiler profiler(isolate);
   profiles.set_cpu_profiler(&profiler);
   profiles.StartProfiling("", false);
-  ProfileGenerator generator(&profiles);
+  ProfileGenerator generator(isolate, &profiles);
   CodeEntry* entry1 = new CodeEntry(i::Logger::FUNCTION_TAG, "aaa");
   generator.code_map()->AddCode(ToAddress(0x1500), entry1, 0x200);