Perform CPU sampling by CPU sampling thread only iff processing thread is not running.

- perform CPU profiler sampling in the sampler thread as we used to;
- skip sampling in the sampling thread if processing thread is running;
- only install SIGPROF handler when CPU profiling is enabled.

BUG=v8:2364

Review URL: https://codereview.chromium.org/11231002
Patch from Sergey Rogulenko <rogulenko@google.com> and Andrey Kosyakov <caseq@chromium.org>.

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12985 ce2b1a6d-e550-0410-aec6-3dcde31c8c00

src/cpu-profiler-inl.h

@@ -31,7 +31,6 @@
 #include "cpu-profiler.h"
 
 #include <new>
-#include "circular-queue-inl.h"
 #include "profile-generator-inl.h"
 #include "unbound-queue-inl.h"
 
@@ -56,11 +55,18 @@ void SharedFunctionInfoMoveEventRecord::UpdateCodeMap(CodeMap* code_map) {
 }
 
 
-TickSample* ProfilerEventsProcessor::TickSampleEvent() {
+TickSample* ProfilerEventsProcessor::StartTickSampleEvent() {
+  if (!ticks_buffer_is_empty_ || ticks_buffer_is_initialized_) return NULL;
+  ticks_buffer_is_initialized_ = true;
   generator_->Tick();
-  TickSampleEventRecord* evt =
-      new(ticks_buffer_.Enqueue()) TickSampleEventRecord(enqueue_order_);
-  return &evt->sample;
+  ticks_buffer_ = TickSampleEventRecord(enqueue_order_);
+  return &ticks_buffer_.sample;
+}
+
+
+void ProfilerEventsProcessor::FinishTickSampleEvent() {
+  ASSERT(ticks_buffer_is_initialized_ && ticks_buffer_is_empty_);
+  ticks_buffer_is_empty_ = false;
 }
 
 

src/cpu-profiler.cc

@@ -39,19 +39,19 @@
 namespace v8 {
 namespace internal {
 
-static const int kEventsBufferSize = 256 * KB;
-static const int kTickSamplesBufferChunkSize = 64 * KB;
-static const int kTickSamplesBufferChunksCount = 16;
 static const int kProfilerStackSize = 64 * KB;
 
 
-ProfilerEventsProcessor::ProfilerEventsProcessor(ProfileGenerator* generator)
+ProfilerEventsProcessor::ProfilerEventsProcessor(ProfileGenerator* generator,
+                                                 Sampler* sampler,
+                                                 int period_in_useconds)
     : Thread(Thread::Options("v8:ProfEvntProc", kProfilerStackSize)),
       generator_(generator),
+      sampler_(sampler),
       running_(true),
-      ticks_buffer_(sizeof(TickSampleEventRecord),
-                    kTickSamplesBufferChunkSize,
-                    kTickSamplesBufferChunksCount),
+      period_in_useconds_(period_in_useconds),
+      ticks_buffer_is_empty_(true),
+      ticks_buffer_is_initialized_(false),
       enqueue_order_(0) {
 }
 
@@ -215,23 +215,17 @@ bool ProfilerEventsProcessor::ProcessTicks(unsigned dequeue_order) {
       generator_->RecordTickSample(record.sample);
     }
 
-    const TickSampleEventRecord* rec =
-        TickSampleEventRecord::cast(ticks_buffer_.StartDequeue());
-    if (rec == NULL) return !ticks_from_vm_buffer_.IsEmpty();
-    // Make a local copy of tick sample record to ensure that it won't
-    // be modified as we are processing it. This is possible as the
-    // sampler writes w/o any sync to the queue, so if the processor
-    // will get far behind, a record may be modified right under its
-    // feet.
-    TickSampleEventRecord record = *rec;
-    if (record.order == dequeue_order) {
+    if (ticks_buffer_is_empty_) return !ticks_from_vm_buffer_.IsEmpty();
+    if (ticks_buffer_.order == dequeue_order) {
       // A paranoid check to make sure that we don't get a memory overrun
       // in case of frames_count having a wild value.
-      if (record.sample.frames_count < 0
-          || record.sample.frames_count > TickSample::kMaxFramesCount)
-        record.sample.frames_count = 0;
-      generator_->RecordTickSample(record.sample);
-      ticks_buffer_.FinishDequeue();
+      if (ticks_buffer_.sample.frames_count < 0
+          || ticks_buffer_.sample.frames_count > TickSample::kMaxFramesCount) {
+        ticks_buffer_.sample.frames_count = 0;
+      }
+      generator_->RecordTickSample(ticks_buffer_.sample);
+      ticks_buffer_is_empty_ = true;
+      ticks_buffer_is_initialized_ = false;
     } else {
       return true;
     }
@@ -239,22 +233,29 @@ bool ProfilerEventsProcessor::ProcessTicks(unsigned dequeue_order) {
 }
 
 
+void ProfilerEventsProcessor::ProcessEventsQueue(int64_t stop_time,
+                                                 unsigned* dequeue_order) {
+  while (OS::Ticks() < stop_time) {
+    if (ProcessTicks(*dequeue_order)) {
+      // All ticks of the current dequeue_order are processed,
+      // proceed to the next code event.
+      ProcessCodeEvent(dequeue_order);
+    }
+  }
+}
+
+
 void ProfilerEventsProcessor::Run() {
   unsigned dequeue_order = 0;
 
   while (running_) {
-    // Process ticks until we have any.
-    if (ProcessTicks(dequeue_order)) {
-      // All ticks of the current dequeue_order are processed,
-      // proceed to the next code event.
-      ProcessCodeEvent(&dequeue_order);
+    int64_t stop_time = OS::Ticks() + period_in_useconds_;
+    if (sampler_ != NULL) {
+      sampler_->DoSample();
     }
-    YieldCPU();
+    ProcessEventsQueue(stop_time, &dequeue_order);
   }
 
-  // Process remaining tick events.
-  ticks_buffer_.FlushResidualRecords();
-  // Perform processing until we have tick events, skip remaining code events.
   while (ProcessTicks(dequeue_order) && ProcessCodeEvent(&dequeue_order)) { }
 }
 
@@ -310,15 +311,22 @@ CpuProfile* CpuProfiler::FindProfile(Object* security_token, unsigned uid) {
 }
 
 
-TickSample* CpuProfiler::TickSampleEvent(Isolate* isolate) {
+TickSample* CpuProfiler::StartTickSampleEvent(Isolate* isolate) {
   if (CpuProfiler::is_profiling(isolate)) {
-    return isolate->cpu_profiler()->processor_->TickSampleEvent();
+    return isolate->cpu_profiler()->processor_->StartTickSampleEvent();
   } else {
     return NULL;
   }
 }
 
 
+void CpuProfiler::FinishTickSampleEvent(Isolate* isolate) {
+  if (CpuProfiler::is_profiling(isolate)) {
+    isolate->cpu_profiler()->processor_->FinishTickSampleEvent();
+  }
+}
+
+
 void CpuProfiler::DeleteAllProfiles() {
   Isolate* isolate = Isolate::Current();
   ASSERT(isolate->cpu_profiler() != NULL);
@@ -486,13 +494,15 @@ void CpuProfiler::StartProcessorIfNotStarted() {
   if (processor_ == NULL) {
     Isolate* isolate = Isolate::Current();
 
+    Sampler* sampler = isolate->logger()->sampler();
     // Disable logging when using the new implementation.
     saved_logging_nesting_ = isolate->logger()->logging_nesting_;
     isolate->logger()->logging_nesting_ = 0;
     generator_ = new ProfileGenerator(profiles_);
-    processor_ = new ProfilerEventsProcessor(generator_);
+    processor_ = new ProfilerEventsProcessor(generator_,
+                                             sampler,
+                                             FLAG_cpu_profiler_sampling_period);
     NoBarrier_Store(&is_profiling_, true);
-    processor_->Start();
     // Enumerate stuff we already have in the heap.
     if (isolate->heap()->HasBeenSetUp()) {
       if (!FLAG_prof_browser_mode) {
@@ -505,12 +515,13 @@ void CpuProfiler::StartProcessorIfNotStarted() {
       isolate->logger()->LogAccessorCallbacks();
     }
     // Enable stack sampling.
-    Sampler* sampler = reinterpret_cast<Sampler*>(isolate->logger()->ticker_);
     if (!sampler->IsActive()) {
       sampler->Start();
       need_to_stop_sampler_ = true;
     }
+    sampler->SetHasProcessingThread(true);
     sampler->IncreaseProfilingDepth();
+    processor_->Start();
   }
 }
 
@@ -545,16 +556,17 @@ void CpuProfiler::StopProcessorIfLastProfile(const char* title) {
 
 
 void CpuProfiler::StopProcessor() {
+  NoBarrier_Store(&is_profiling_, false);
+  processor_->Stop();
+  processor_->Join();
   Logger* logger = Isolate::Current()->logger();
-  Sampler* sampler = reinterpret_cast<Sampler*>(logger->ticker_);
+  Sampler* sampler = logger->sampler();
   sampler->DecreaseProfilingDepth();
+  sampler->SetHasProcessingThread(false);
   if (need_to_stop_sampler_) {
     sampler->Stop();
     need_to_stop_sampler_ = false;
   }
-  NoBarrier_Store(&is_profiling_, false);
-  processor_->Stop();
-  processor_->Join();
   delete processor_;
   delete generator_;
   processor_ = NULL;

src/cpu-profiler.h

@@ -124,7 +124,9 @@ class TickSampleEventRecord {
 // methods called by event producers: VM and stack sampler threads.
 class ProfilerEventsProcessor : public Thread {
  public:
-  explicit ProfilerEventsProcessor(ProfileGenerator* generator);
+  explicit ProfilerEventsProcessor(ProfileGenerator* generator,
+                                   Sampler* sampler,
+                                   int period_in_useconds);
   virtual ~ProfilerEventsProcessor() {}
 
   // Thread control.
@@ -156,11 +158,12 @@ class ProfilerEventsProcessor : public Thread {
   // Puts current stack into tick sample events buffer.
   void AddCurrentStack();
 
-  // Tick sample events are filled directly in the buffer of the circular
-  // queue (because the structure is of fixed width, but usually not all
-  // stack frame entries are filled.) This method returns a pointer to the
-  // next record of the buffer.
-  INLINE(TickSample* TickSampleEvent());
+  // StartTickSampleEvent returns a pointer only if the ticks_buffer_ is empty,
+  // FinishTickSampleEvent marks the ticks_buffer_ as filled.
+  // Finish should be called only after successful Start (returning non-NULL
+  // pointer).
+  INLINE(TickSample* StartTickSampleEvent());
+  INLINE(void FinishTickSampleEvent());
 
  private:
   union CodeEventsContainer {
@@ -173,13 +176,19 @@ class ProfilerEventsProcessor : public Thread {
   // Called from events processing thread (Run() method.)
   bool ProcessCodeEvent(unsigned* dequeue_order);
   bool ProcessTicks(unsigned dequeue_order);
+  void ProcessEventsQueue(int64_t stop_time, unsigned* dequeue_order);
 
   INLINE(static bool FilterOutCodeCreateEvent(Logger::LogEventsAndTags tag));
 
   ProfileGenerator* generator_;
+  Sampler* sampler_;
   bool running_;
+  // Sampling period in microseconds.
+  const int period_in_useconds_;
   UnboundQueue<CodeEventsContainer> events_buffer_;
-  SamplingCircularQueue ticks_buffer_;
+  TickSampleEventRecord ticks_buffer_;
+  bool ticks_buffer_is_empty_;
+  bool ticks_buffer_is_initialized_;
   UnboundQueue<TickSampleEventRecord> ticks_from_vm_buffer_;
   unsigned enqueue_order_;
 };
@@ -218,7 +227,10 @@ class CpuProfiler {
   static bool HasDetachedProfiles();
 
   // Invoked from stack sampler (thread or signal handler.)
-  static TickSample* TickSampleEvent(Isolate* isolate);
+  // Finish should be called only after successful Start (returning non-NULL
+  // pointer).
+  static TickSample* StartTickSampleEvent(Isolate* isolate);
+  static void FinishTickSampleEvent(Isolate* isolate);
 
   // Must be called via PROFILE macro, otherwise will crash when
   // profiling is not enabled.

src/flag-definitions.h

@@ -346,6 +346,10 @@ DEFINE_bool(compilation_cache, true, "enable compilation cache")
 
 DEFINE_bool(cache_prototype_transitions, true, "cache prototype transitions")
 
+// cpu-profiler.cc
+DEFINE_int(cpu_profiler_sampling_period, 1000,
+           "CPU profiler sampling period in microseconds")
+
 // debug.cc
 DEFINE_bool(trace_debug_json, false, "trace debugging JSON request/response")
 DEFINE_bool(debugger_auto_break, true,

src/platform-cygwin.cc

@@ -699,7 +699,7 @@ class SamplerThread : public Thread {
     memset(&context, 0, sizeof(context));
 
     TickSample sample_obj;
-    TickSample* sample = CpuProfiler::TickSampleEvent(sampler->isolate());
+    TickSample* sample = CpuProfiler::StartTickSampleEvent(sampler->isolate());
     if (sample == NULL) sample = &sample_obj;
 
     static const DWORD kSuspendFailed = static_cast<DWORD>(-1);
@@ -720,6 +720,7 @@ class SamplerThread : public Thread {
       sampler->SampleStack(sample);
       sampler->Tick(sample);
     }
+    CpuProfiler::FinishTickSampleEvent(sampler->isolate());
     ResumeThread(profiled_thread);
   }
 
@@ -774,6 +775,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -788,4 +794,12 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-freebsd.cc

@@ -685,7 +685,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
   if (sampler == NULL || !sampler->IsActive()) return;
 
   TickSample sample_obj;
-  TickSample* sample = CpuProfiler::TickSampleEvent(isolate);
+  TickSample* sample = CpuProfiler::StartTickSampleEvent(isolate);
   if (sample == NULL) sample = &sample_obj;
 
   // Extracting the sample from the context is extremely machine dependent.
@@ -707,6 +707,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
 #endif
   sampler->SampleStack(sample);
   sampler->Tick(sample);
+  CpuProfiler::FinishTickSampleEvent(isolate);
 }
 
 
@@ -890,6 +891,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -904,4 +910,12 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-linux.cc

@@ -1025,6 +1025,7 @@ static int GetThreadID() {
 static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
   USE(info);
   if (signal != SIGPROF) return;
+
   Isolate* isolate = Isolate::UncheckedCurrent();
   if (isolate == NULL || !isolate->IsInitialized() || !isolate->IsInUse()) {
     // We require a fully initialized and entered isolate.
@@ -1039,7 +1040,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
   if (sampler == NULL || !sampler->IsActive()) return;
 
   TickSample sample_obj;
-  TickSample* sample = CpuProfiler::TickSampleEvent(isolate);
+  TickSample* sample = CpuProfiler::StartTickSampleEvent(isolate);
   if (sample == NULL) sample = &sample_obj;
 
   // Extracting the sample from the context is extremely machine dependent.
@@ -1075,16 +1076,74 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
 #endif  // V8_HOST_ARCH_*
   sampler->SampleStack(sample);
   sampler->Tick(sample);
+  CpuProfiler::FinishTickSampleEvent(isolate);
 }
 
 
+class CpuProfilerSignalHandler {
+ public:
+  static void SetUp() { if (!mutex_) mutex_ = OS::CreateMutex(); }
+  static void TearDown() { delete mutex_; }
+
+  static void InstallSignalHandler() {
+    struct sigaction sa;
+    ScopedLock lock(mutex_);
+    if (signal_handler_installed_counter_ > 0) {
+      signal_handler_installed_counter_++;
+      return;
+    }
+    sa.sa_sigaction = ProfilerSignalHandler;
+    sigemptyset(&sa.sa_mask);
+    sa.sa_flags = SA_RESTART | SA_SIGINFO;
+    if (sigaction(SIGPROF, &sa, &old_signal_handler_) == 0) {
+      signal_handler_installed_counter_++;
+    }
+  }
+
+  static void RestoreSignalHandler() {
+    ScopedLock lock(mutex_);
+    if (signal_handler_installed_counter_ == 0)
+      return;
+    if (signal_handler_installed_counter_ == 1) {
+      sigaction(SIGPROF, &old_signal_handler_, 0);
+    }
+    signal_handler_installed_counter_--;
+  }
+
+  static bool signal_handler_installed() {
+    return signal_handler_installed_counter_ > 0;
+  }
+
+ private:
+  static int signal_handler_installed_counter_;
+  static struct sigaction old_signal_handler_;
+  static Mutex* mutex_;
+};
+
+
+int CpuProfilerSignalHandler::signal_handler_installed_counter_ = 0;
+struct sigaction CpuProfilerSignalHandler::old_signal_handler_;
+Mutex* CpuProfilerSignalHandler::mutex_ = NULL;
+
+
 class Sampler::PlatformData : public Malloced {
  public:
-  PlatformData() : vm_tid_(GetThreadID()) {}
+  PlatformData()
+      : vm_tgid_(getpid()),
+        vm_tid_(GetThreadID()) {}
 
-  int vm_tid() const { return vm_tid_; }
+  void SendProfilingSignal() {
+    if (!CpuProfilerSignalHandler::signal_handler_installed()) return;
+    // Glibc doesn't provide a wrapper for tgkill(2).
+#if defined(ANDROID)
+    syscall(__NR_tgkill, vm_tgid_, vm_tid_, SIGPROF);
+#else
+    syscall(SYS_tgkill, vm_tgid_, vm_tid_, SIGPROF);
+#endif
+  }
 
  private:
+  const int vm_tgid_;
   const int vm_tid_;
 };
 
@@ -1100,28 +1159,11 @@ class SignalSender : public Thread {
 
   explicit SignalSender(int interval)
       : Thread(Thread::Options("SignalSender", kSignalSenderStackSize)),
-        vm_tgid_(getpid()),
         interval_(interval) {}
 
   static void SetUp() { if (!mutex_) mutex_ = OS::CreateMutex(); }
   static void TearDown() { delete mutex_; }
 
-  static void InstallSignalHandler() {
-    struct sigaction sa;
-    sa.sa_sigaction = ProfilerSignalHandler;
-    sigemptyset(&sa.sa_mask);
-    sa.sa_flags = SA_RESTART | SA_SIGINFO;
-    signal_handler_installed_ =
-        (sigaction(SIGPROF, &sa, &old_signal_handler_) == 0);
-  }
-
-  static void RestoreSignalHandler() {
-    if (signal_handler_installed_) {
-      sigaction(SIGPROF, &old_signal_handler_, 0);
-      signal_handler_installed_ = false;
-    }
-  }
-
   static void AddActiveSampler(Sampler* sampler) {
     ScopedLock lock(mutex_);
     SamplerRegistry::AddActiveSampler(sampler);
@@ -1142,7 +1184,6 @@ class SignalSender : public Thread {
       RuntimeProfiler::StopRuntimeProfilerThreadBeforeShutdown(instance_);
       delete instance_;
       instance_ = NULL;
-      RestoreSignalHandler();
     }
   }
 
@@ -1154,18 +1195,13 @@ class SignalSender : public Thread {
       bool cpu_profiling_enabled =
           (state == SamplerRegistry::HAS_CPU_PROFILING_SAMPLERS);
       bool runtime_profiler_enabled = RuntimeProfiler::IsEnabled();
-      if (cpu_profiling_enabled && !signal_handler_installed_) {
-        InstallSignalHandler();
-      } else if (!cpu_profiling_enabled && signal_handler_installed_) {
-        RestoreSignalHandler();
-      }
       // When CPU profiling is enabled both JavaScript and C++ code is
       // profiled. We must not suspend.
       if (!cpu_profiling_enabled) {
         if (rate_limiter_.SuspendIfNecessary()) continue;
       }
       if (cpu_profiling_enabled && runtime_profiler_enabled) {
-        if (!SamplerRegistry::IterateActiveSamplers(&DoCpuProfile, this)) {
+        if (!SamplerRegistry::IterateActiveSamplers(&DoCpuProfile, NULL)) {
           return;
         }
         Sleep(HALF_INTERVAL);
@@ -1175,8 +1211,7 @@ class SignalSender : public Thread {
         Sleep(HALF_INTERVAL);
       } else {
         if (cpu_profiling_enabled) {
-          if (!SamplerRegistry::IterateActiveSamplers(&DoCpuProfile,
-                                                      this)) {
+          if (!SamplerRegistry::IterateActiveSamplers(&DoCpuProfile, NULL)) {
             return;
           }
         }
@@ -1191,10 +1226,9 @@ class SignalSender : public Thread {
     }
   }
 
-  static void DoCpuProfile(Sampler* sampler, void* raw_sender) {
+  static void DoCpuProfile(Sampler* sampler, void*) {
     if (!sampler->IsProfiling()) return;
-    SignalSender* sender = reinterpret_cast<SignalSender*>(raw_sender);
-    sender->SendProfilingSignal(sampler->platform_data()->vm_tid());
+    sampler->platform_data()->SendProfilingSignal();
   }
 
   static void DoRuntimeProfile(Sampler* sampler, void* ignored) {
@@ -1202,16 +1236,6 @@ class SignalSender : public Thread {
     sampler->isolate()->runtime_profiler()->NotifyTick();
   }
 
-  void SendProfilingSignal(int tid) {
-    if (!signal_handler_installed_) return;
-    // Glibc doesn't provide a wrapper for tgkill(2).
-#if defined(ANDROID)
-    syscall(__NR_tgkill, vm_tgid_, tid, SIGPROF);
-#else
-    syscall(SYS_tgkill, vm_tgid_, tid, SIGPROF);
-#endif
-  }
-
   void Sleep(SleepInterval full_or_half) {
     // Convert ms to us and subtract 100 us to compensate delays
     // occuring during signal delivery.
@@ -1234,15 +1258,12 @@ class SignalSender : public Thread {
 #endif  // ANDROID
   }
 
-  const int vm_tgid_;
   const int interval_;
   RuntimeProfilerRateLimiter rate_limiter_;
 
   // Protects the process wide state below.
   static Mutex* mutex_;
   static SignalSender* instance_;
-  static bool signal_handler_installed_;
-  static struct sigaction old_signal_handler_;
 
  private:
   DISALLOW_COPY_AND_ASSIGN(SignalSender);
@@ -1251,8 +1272,6 @@ class SignalSender : public Thread {
 
 Mutex* SignalSender::mutex_ = NULL;
 SignalSender* SignalSender::instance_ = NULL;
-struct sigaction SignalSender::old_signal_handler_;
-bool SignalSender::signal_handler_installed_ = false;
 
 
 void OS::SetUp() {
@@ -1280,11 +1299,13 @@ void OS::SetUp() {
   }
 #endif
   SignalSender::SetUp();
+  CpuProfilerSignalHandler::SetUp();
 }
 
 
 void OS::TearDown() {
   SignalSender::TearDown();
+  CpuProfilerSignalHandler::TearDown();
   delete limit_mutex;
 }
 
@@ -1294,6 +1315,7 @@ Sampler::Sampler(Isolate* isolate, int interval)
       interval_(interval),
       profiling_(false),
       active_(false),
+      has_processing_thread_(false),
       samples_taken_(0) {
   data_ = new PlatformData;
 }
@@ -1305,6 +1327,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  platform_data()->SendProfilingSignal();
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -1319,4 +1346,14 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+  CpuProfilerSignalHandler::InstallSignalHandler();
+}
+
+
+void Sampler::StopSampling() {
+  CpuProfilerSignalHandler::RestoreSignalHandler();
+}
+
+
 } }  // namespace v8::internal

src/platform-macos.cc

@@ -825,7 +825,7 @@ class SamplerThread : public Thread {
   void SampleContext(Sampler* sampler) {
     thread_act_t profiled_thread = sampler->platform_data()->profiled_thread();
     TickSample sample_obj;
-    TickSample* sample = CpuProfiler::TickSampleEvent(sampler->isolate());
+    TickSample* sample = CpuProfiler::StartTickSampleEvent(sampler->isolate());
     if (sample == NULL) sample = &sample_obj;
 
     if (KERN_SUCCESS != thread_suspend(profiled_thread)) return;
@@ -863,6 +863,7 @@ class SamplerThread : public Thread {
       sampler->SampleStack(sample);
       sampler->Tick(sample);
     }
+    CpuProfiler::FinishTickSampleEvent(sampler->isolate());
     thread_resume(profiled_thread);
   }
 
@@ -915,6 +916,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -929,4 +935,12 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-nullos.cc

@@ -514,4 +514,12 @@ void ProfileSampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-openbsd.cc

@@ -738,7 +738,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
   if (sampler == NULL || !sampler->IsActive()) return;
 
   TickSample sample_obj;
-  TickSample* sample = CpuProfiler::TickSampleEvent(isolate);
+  TickSample* sample = CpuProfiler::StartTickSampleEvent(isolate);
   if (sample == NULL) sample = &sample_obj;
 
   // Extracting the sample from the context is extremely machine dependent.
@@ -768,6 +768,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
 #endif  // __NetBSD__
   sampler->SampleStack(sample);
   sampler->Tick(sample);
+  CpuProfiler::FinishTickSampleEvent(isolate);
 }
 
 
@@ -970,6 +971,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -984,4 +990,12 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-solaris.cc

@@ -672,7 +672,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
   if (sampler == NULL || !sampler->IsActive()) return;
 
   TickSample sample_obj;
-  TickSample* sample = CpuProfiler::TickSampleEvent(isolate);
+  TickSample* sample = CpuProfiler::StartTickSampleEvent(isolate);
   if (sample == NULL) sample = &sample_obj;
 
   // Extracting the sample from the context is extremely machine dependent.
@@ -686,6 +686,7 @@ static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
 
   sampler->SampleStack(sample);
   sampler->Tick(sample);
+  CpuProfiler::FinishTickSampleEvent(isolate);
 }
 
 class Sampler::PlatformData : public Malloced {
@@ -889,6 +890,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -902,4 +908,13 @@ void Sampler::Stop() {
   SetActive(false);
 }
 
+
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform-win32.cc

@@ -2045,7 +2045,7 @@ class SamplerThread : public Thread {
     memset(&context, 0, sizeof(context));
 
     TickSample sample_obj;
-    TickSample* sample = CpuProfiler::TickSampleEvent(sampler->isolate());
+    TickSample* sample = CpuProfiler::StartTickSampleEvent(sampler->isolate());
     if (sample == NULL) sample = &sample_obj;
 
     static const DWORD kSuspendFailed = static_cast<DWORD>(-1);
@@ -2066,6 +2066,7 @@ class SamplerThread : public Thread {
       sampler->SampleStack(sample);
       sampler->Tick(sample);
     }
+    CpuProfiler::FinishTickSampleEvent(sampler->isolate());
     ResumeThread(profiled_thread);
   }
 
@@ -2120,6 +2121,11 @@ Sampler::~Sampler() {
 }
 
 
+void Sampler::DoSample() {
+  // TODO(rogulenko): implement
+}
+
+
 void Sampler::Start() {
   ASSERT(!IsActive());
   SetActive(true);
@@ -2134,4 +2140,12 @@ void Sampler::Stop() {
 }
 
 
+void Sampler::StartSampling() {
+}
+
+
+void Sampler::StopSampling() {
+}
+
+
 } }  // namespace v8::internal

src/platform.h

@@ -749,6 +749,9 @@ class Sampler {
     IncSamplesTaken();
   }
 
+  // Performs platform-specific stack sampling.
+  void DoSample();
+
   // This method is called for each sampling period with the current
   // program counter.
   virtual void Tick(TickSample* sample) = 0;
@@ -757,10 +760,28 @@ class Sampler {
   void Start();
   void Stop();
 
-  // Is the sampler used for profiling?
-  bool IsProfiling() const { return NoBarrier_Load(&profiling_) > 0; }
-  void IncreaseProfilingDepth() { NoBarrier_AtomicIncrement(&profiling_, 1); }
-  void DecreaseProfilingDepth() { NoBarrier_AtomicIncrement(&profiling_, -1); }
+  // Whether the sampling thread should use this Sampler for CPU profiling?
+  bool IsProfiling() const {
+    return NoBarrier_Load(&profiling_) > 0 &&
+        !NoBarrier_Load(&has_processing_thread_);
+  }
+  // Perform platform-specific initialization before DoSample() may be invoked.
+  void StartSampling();
+  // Perform platform-specific cleanup after samping.
+  void StopSampling();
+  void IncreaseProfilingDepth() {
+    if (NoBarrier_AtomicIncrement(&profiling_, 1) == 1) {
+      StartSampling();
+    }
+  }
+  void DecreaseProfilingDepth() {
+    if (!NoBarrier_AtomicIncrement(&profiling_, -1)) {
+      StopSampling();
+    }
+  }
+  void SetHasProcessingThread(bool value) {
+    NoBarrier_Store(&has_processing_thread_, value);
+  }
 
   // Whether the sampler is running (that is, consumes resources).
   bool IsActive() const { return NoBarrier_Load(&active_); }
@@ -787,6 +808,7 @@ class Sampler {
   const int interval_;
   Atomic32 profiling_;
   Atomic32 active_;
+  Atomic32 has_processing_thread_;
   PlatformData* data_;  // Platform specific data.
   int samples_taken_;  // Counts stack samples taken.
   DISALLOW_IMPLICIT_CONSTRUCTORS(Sampler);

test/cctest/test-cpu-profiler.cc

@@ -5,6 +5,7 @@
 #include "v8.h"
 #include "cpu-profiler-inl.h"
 #include "cctest.h"
+#include "platform.h"
 #include "../include/v8-profiler.h"
 
 using i::CodeEntry;
@@ -20,7 +21,7 @@ using i::TokenEnumerator;
 TEST(StartStop) {
   CpuProfilesCollection profiles;
   ProfileGenerator generator(&profiles);
-  ProfilerEventsProcessor processor(&generator);
+  ProfilerEventsProcessor processor(&generator, NULL, 1000);
   processor.Start();
   processor.Stop();
   processor.Join();
@@ -38,11 +39,13 @@ static inline i::Address ToAddress(int n) {
   return reinterpret_cast<i::Address>(n);
 }
 
-static void EnqueueTickSampleEvent(ProfilerEventsProcessor* proc,
-                                   i::Address frame1,
-                                   i::Address frame2 = NULL,
-                                   i::Address frame3 = NULL) {
-  i::TickSample* sample = proc->TickSampleEvent();
+static void AddTickSampleEvent(ProfilerEventsProcessor* processor,
+                               i::Address frame1,
+                               i::Address frame2 = NULL,
+                               i::Address frame3 = NULL) {
+  i::TickSample* sample;
+  i::OS::Sleep(20);
+  while ((sample = processor->StartTickSampleEvent()) == NULL) i::OS::Sleep(20);
   sample->pc = frame1;
   sample->tos = frame1;
   sample->frames_count = 0;
@@ -54,6 +57,7 @@ static void EnqueueTickSampleEvent(ProfilerEventsProcessor* proc,
     sample->stack[1] = frame3;
     sample->frames_count = 2;
   }
+  processor->FinishTickSampleEvent();
 }
 
 namespace {
@@ -81,7 +85,7 @@ TEST(CodeEvents) {
   CpuProfilesCollection profiles;
   profiles.StartProfiling("", 1);
   ProfileGenerator generator(&profiles);
-  ProfilerEventsProcessor processor(&generator);
+  ProfilerEventsProcessor processor(&generator, NULL, 1000);
   processor.Start();
 
   // Enqueue code creation events.
@@ -108,8 +112,8 @@ TEST(CodeEvents) {
   processor.CodeMoveEvent(ToAddress(0x1400), ToAddress(0x1500));
   processor.CodeCreateEvent(i::Logger::STUB_TAG, 3, ToAddress(0x1600), 0x10);
   processor.CodeCreateEvent(i::Logger::STUB_TAG, 4, ToAddress(0x1605), 0x10);
-  // Enqueue a tick event to enable code events processing.
-  EnqueueTickSampleEvent(&processor, ToAddress(0x1000));
+  // Add a tick event to enable code events processing.
+  AddTickSampleEvent(&processor, ToAddress(0x1000));
 
   processor.Stop();
   processor.Join();
@@ -142,7 +146,7 @@ TEST(TickEvents) {
   CpuProfilesCollection profiles;
   profiles.StartProfiling("", 1);
   ProfileGenerator generator(&profiles);
-  ProfilerEventsProcessor processor(&generator);
+  ProfilerEventsProcessor processor(&generator, NULL, 1000);
   processor.Start();
 
   processor.CodeCreateEvent(i::Logger::BUILTIN_TAG,
@@ -154,12 +158,12 @@ TEST(TickEvents) {
                             "ddd",
                             ToAddress(0x1400),
                             0x80);
-  EnqueueTickSampleEvent(&processor, ToAddress(0x1210));
-  EnqueueTickSampleEvent(&processor, ToAddress(0x1305), ToAddress(0x1220));
-  EnqueueTickSampleEvent(&processor,
-                         ToAddress(0x1404),
-                         ToAddress(0x1305),
-                         ToAddress(0x1230));
+  AddTickSampleEvent(&processor, ToAddress(0x1210));
+  AddTickSampleEvent(&processor, ToAddress(0x1305), ToAddress(0x1220));
+  AddTickSampleEvent(&processor,
+                     ToAddress(0x1404),
+                     ToAddress(0x1305),
+                     ToAddress(0x1230));
 
   processor.Stop();
   processor.Join();
@@ -232,7 +236,7 @@ TEST(Issue1398) {
   CpuProfilesCollection profiles;
   profiles.StartProfiling("", 1);
   ProfileGenerator generator(&profiles);
-  ProfilerEventsProcessor processor(&generator);
+  ProfilerEventsProcessor processor(&generator, NULL, 1000);
   processor.Start();
 
   processor.CodeCreateEvent(i::Logger::BUILTIN_TAG,
@@ -240,13 +244,14 @@ TEST(Issue1398) {
                             ToAddress(0x1200),
                             0x80);
 
-  i::TickSample* sample = processor.TickSampleEvent();
+  i::TickSample* sample = processor.StartTickSampleEvent();
   sample->pc = ToAddress(0x1200);
   sample->tos = 0;
   sample->frames_count = i::TickSample::kMaxFramesCount;
   for (int i = 0; i < sample->frames_count; ++i) {
     sample->stack[i] = ToAddress(0x1200);
   }
+  processor.FinishTickSampleEvent();
 
   processor.Stop();
   processor.Join();