v8/src/cpu-profiler.h

276 lines
8.3 KiB
C
Raw Normal View History

// Copyright 2012 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#ifndef V8_CPU_PROFILER_H_
#define V8_CPU_PROFILER_H_
#include "src/allocation.h"
#include "src/atomicops.h"
#include "src/circular-queue.h"
#include "src/platform/time.h"
#include "src/sampler.h"
#include "src/unbound-queue.h"
namespace v8 {
namespace internal {
// Forward declarations.
class CodeEntry;
class CodeMap;
class CompilationInfo;
class CpuProfile;
class CpuProfilesCollection;
class ProfileGenerator;
#define CODE_EVENTS_TYPE_LIST(V) \
V(CODE_CREATION, CodeCreateEventRecord) \
V(CODE_MOVE, CodeMoveEventRecord) \
V(CODE_DISABLE_OPT, CodeDisableOptEventRecord) \
V(SHARED_FUNC_MOVE, SharedFunctionInfoMoveEventRecord) \
V(REPORT_BUILTIN, ReportBuiltinEventRecord)
class CodeEventRecord {
public:
#define DECLARE_TYPE(type, ignore) type,
enum Type {
NONE = 0,
CODE_EVENTS_TYPE_LIST(DECLARE_TYPE)
NUMBER_OF_TYPES
};
#undef DECLARE_TYPE
Type type;
mutable unsigned order;
};
class CodeCreateEventRecord : public CodeEventRecord {
public:
Address start;
CodeEntry* entry;
unsigned size;
Address shared;
INLINE(void UpdateCodeMap(CodeMap* code_map));
};
class CodeMoveEventRecord : public CodeEventRecord {
public:
Address from;
Address to;
INLINE(void UpdateCodeMap(CodeMap* code_map));
};
class CodeDisableOptEventRecord : public CodeEventRecord {
public:
Address start;
const char* bailout_reason;
INLINE(void UpdateCodeMap(CodeMap* code_map));
};
class SharedFunctionInfoMoveEventRecord : public CodeEventRecord {
public:
Address from;
Address to;
INLINE(void UpdateCodeMap(CodeMap* code_map));
};
class ReportBuiltinEventRecord : public CodeEventRecord {
public:
Address start;
Builtins::Name builtin_id;
INLINE(void UpdateCodeMap(CodeMap* code_map));
};
class TickSampleEventRecord {
public:
// The parameterless constructor is used when we dequeue data from
// the ticks buffer.
TickSampleEventRecord() { }
explicit TickSampleEventRecord(unsigned order) : order(order) { }
unsigned order;
TickSample sample;
};
class CodeEventsContainer {
public:
explicit CodeEventsContainer(
CodeEventRecord::Type type = CodeEventRecord::NONE) {
generic.type = type;
}
union {
CodeEventRecord generic;
#define DECLARE_CLASS(ignore, type) type type##_;
CODE_EVENTS_TYPE_LIST(DECLARE_CLASS)
#undef DECLARE_TYPE
};
};
// This class implements both the profile events processor thread and
// methods called by event producers: VM and stack sampler threads.
class ProfilerEventsProcessor : public Thread {
public:
Support higher CPU profiler sampling rate on posix systems New flag is added that allows to specify CPU profiler sampling rate in microseconds as command line argument. It was tested to work fine with 100us interval(currently it is 1ms). Default values are kept the same as in the current implementation. The new implementation is enabled only on POSIX platforms which use signals to collect samples. Other platforms that pause thread being sampled are to follow. SIGPROF signals are now sent on the profiler event processor thread to make sure that the processing thread does fall far behind the sampling. The patch is based on the previous one that was rolled out in r13851. The main difference is that the circular queue is not modified for now. On Linux sampling for CPU profiler is initiated on the profiler event processor thread, other platforms to follow. CPU profiler continues to use SamplingCircularQueue, we will probably replace it with a single sample buffer when Mac and Win ports support profiling on the event processing thread. When --prof option is specified profiling is initiated either on the profiler event processor thread if CPU profiler is on or on the SignalSender thread as it used to be if no CPU profiles are being collected. ProfilerEventsProcessor::ProcessEventsAndDoSample now waits in a tight loop, processing collected samples until sampling interval expires. To save CPU resources I'm planning to change that to use nanosleep as only one sample is expected in the queue at any point. BUG=v8:2814 R=bmeurer@chromium.org Review URL: https://codereview.chromium.org/21101002 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16310 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-08-26 07:17:12 +00:00
ProfilerEventsProcessor(ProfileGenerator* generator,
Sampler* sampler,
TimeDelta period);
virtual ~ProfilerEventsProcessor() {}
// Thread control.
virtual void Run();
void StopSynchronously();
INLINE(bool running()) { return running_; }
void Enqueue(const CodeEventsContainer& event);
// Puts current stack into tick sample events buffer.
void AddCurrentStack(Isolate* isolate);
// 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* StartTickSample();
inline void FinishTickSample();
// SamplingCircularQueue has stricter alignment requirements than a normal new
// can fulfil, so we need to provide our own new/delete here.
void* operator new(size_t size);
void operator delete(void* ptr);
private:
// Called from events processing thread (Run() method.)
bool ProcessCodeEvent();
enum SampleProcessingResult {
OneSampleProcessed,
FoundSampleForNextCodeEvent,
NoSamplesInQueue
};
SampleProcessingResult ProcessOneSample();
Support higher CPU profiler sampling rate on posix systems New flag is added that allows to specify CPU profiler sampling rate in microseconds as command line argument. It was tested to work fine with 100us interval(currently it is 1ms). Default values are kept the same as in the current implementation. The new implementation is enabled only on POSIX platforms which use signals to collect samples. Other platforms that pause thread being sampled are to follow. SIGPROF signals are now sent on the profiler event processor thread to make sure that the processing thread does fall far behind the sampling. The patch is based on the previous one that was rolled out in r13851. The main difference is that the circular queue is not modified for now. On Linux sampling for CPU profiler is initiated on the profiler event processor thread, other platforms to follow. CPU profiler continues to use SamplingCircularQueue, we will probably replace it with a single sample buffer when Mac and Win ports support profiling on the event processing thread. When --prof option is specified profiling is initiated either on the profiler event processor thread if CPU profiler is on or on the SignalSender thread as it used to be if no CPU profiles are being collected. ProfilerEventsProcessor::ProcessEventsAndDoSample now waits in a tight loop, processing collected samples until sampling interval expires. To save CPU resources I'm planning to change that to use nanosleep as only one sample is expected in the queue at any point. BUG=v8:2814 R=bmeurer@chromium.org Review URL: https://codereview.chromium.org/21101002 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16310 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-08-26 07:17:12 +00:00
ProfileGenerator* generator_;
Support higher CPU profiler sampling rate on posix systems New flag is added that allows to specify CPU profiler sampling rate in microseconds as command line argument. It was tested to work fine with 100us interval(currently it is 1ms). Default values are kept the same as in the current implementation. The new implementation is enabled only on POSIX platforms which use signals to collect samples. Other platforms that pause thread being sampled are to follow. SIGPROF signals are now sent on the profiler event processor thread to make sure that the processing thread does fall far behind the sampling. The patch is based on the previous one that was rolled out in r13851. The main difference is that the circular queue is not modified for now. On Linux sampling for CPU profiler is initiated on the profiler event processor thread, other platforms to follow. CPU profiler continues to use SamplingCircularQueue, we will probably replace it with a single sample buffer when Mac and Win ports support profiling on the event processing thread. When --prof option is specified profiling is initiated either on the profiler event processor thread if CPU profiler is on or on the SignalSender thread as it used to be if no CPU profiles are being collected. ProfilerEventsProcessor::ProcessEventsAndDoSample now waits in a tight loop, processing collected samples until sampling interval expires. To save CPU resources I'm planning to change that to use nanosleep as only one sample is expected in the queue at any point. BUG=v8:2814 R=bmeurer@chromium.org Review URL: https://codereview.chromium.org/21101002 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16310 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-08-26 07:17:12 +00:00
Sampler* sampler_;
bool running_;
Support higher CPU profiler sampling rate on posix systems New flag is added that allows to specify CPU profiler sampling rate in microseconds as command line argument. It was tested to work fine with 100us interval(currently it is 1ms). Default values are kept the same as in the current implementation. The new implementation is enabled only on POSIX platforms which use signals to collect samples. Other platforms that pause thread being sampled are to follow. SIGPROF signals are now sent on the profiler event processor thread to make sure that the processing thread does fall far behind the sampling. The patch is based on the previous one that was rolled out in r13851. The main difference is that the circular queue is not modified for now. On Linux sampling for CPU profiler is initiated on the profiler event processor thread, other platforms to follow. CPU profiler continues to use SamplingCircularQueue, we will probably replace it with a single sample buffer when Mac and Win ports support profiling on the event processing thread. When --prof option is specified profiling is initiated either on the profiler event processor thread if CPU profiler is on or on the SignalSender thread as it used to be if no CPU profiles are being collected. ProfilerEventsProcessor::ProcessEventsAndDoSample now waits in a tight loop, processing collected samples until sampling interval expires. To save CPU resources I'm planning to change that to use nanosleep as only one sample is expected in the queue at any point. BUG=v8:2814 R=bmeurer@chromium.org Review URL: https://codereview.chromium.org/21101002 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@16310 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2013-08-26 07:17:12 +00:00
// Sampling period in microseconds.
const TimeDelta period_;
UnboundQueue<CodeEventsContainer> events_buffer_;
static const size_t kTickSampleBufferSize = 1 * MB;
static const size_t kTickSampleQueueLength =
kTickSampleBufferSize / sizeof(TickSampleEventRecord);
SamplingCircularQueue<TickSampleEventRecord,
kTickSampleQueueLength> ticks_buffer_;
UnboundQueue<TickSampleEventRecord> ticks_from_vm_buffer_;
unsigned last_code_event_id_;
unsigned last_processed_code_event_id_;
};
#define PROFILE(IsolateGetter, Call) \
do { \
Isolate* cpu_profiler_isolate = (IsolateGetter); \
v8::internal::Logger* logger = cpu_profiler_isolate->logger(); \
CpuProfiler* cpu_profiler = cpu_profiler_isolate->cpu_profiler(); \
if (logger->is_logging_code_events() || cpu_profiler->is_profiling()) { \
logger->Call; \
} \
} while (false)
class CpuProfiler : public CodeEventListener {
public:
explicit CpuProfiler(Isolate* isolate);
CpuProfiler(Isolate* isolate,
CpuProfilesCollection* test_collection,
ProfileGenerator* test_generator,
ProfilerEventsProcessor* test_processor);
virtual ~CpuProfiler();
void set_sampling_interval(TimeDelta value);
void StartProfiling(const char* title, bool record_samples = false);
void StartProfiling(String* title, bool record_samples);
CpuProfile* StopProfiling(const char* title);
CpuProfile* StopProfiling(String* title);
int GetProfilesCount();
CpuProfile* GetProfile(int index);
void DeleteAllProfiles();
void DeleteProfile(CpuProfile* profile);
// Invoked from stack sampler (thread or signal handler.)
inline TickSample* StartTickSample();
inline void FinishTickSample();
// Must be called via PROFILE macro, otherwise will crash when
// profiling is not enabled.
virtual void CallbackEvent(Name* name, Address entry_point);
virtual void CodeCreateEvent(Logger::LogEventsAndTags tag,
Code* code, const char* comment);
virtual void CodeCreateEvent(Logger::LogEventsAndTags tag,
Code* code, Name* name);
virtual void CodeCreateEvent(Logger::LogEventsAndTags tag,
Code* code,
SharedFunctionInfo* shared,
CompilationInfo* info,
Name* name);
virtual void CodeCreateEvent(Logger::LogEventsAndTags tag,
Code* code,
SharedFunctionInfo* shared,
CompilationInfo* info,
Name* source, int line, int column);
virtual void CodeCreateEvent(Logger::LogEventsAndTags tag,
Code* code, int args_count);
virtual void CodeMovingGCEvent() {}
virtual void CodeMoveEvent(Address from, Address to);
virtual void CodeDisableOptEvent(Code* code, SharedFunctionInfo* shared);
virtual void CodeDeleteEvent(Address from);
virtual void GetterCallbackEvent(Name* name, Address entry_point);
virtual void RegExpCodeCreateEvent(Code* code, String* source);
virtual void SetterCallbackEvent(Name* name, Address entry_point);
virtual void SharedFunctionInfoMoveEvent(Address from, Address to);
INLINE(bool is_profiling() const) { return is_profiling_; }
bool* is_profiling_address() {
return &is_profiling_;
}
ProfileGenerator* generator() const { return generator_; }
ProfilerEventsProcessor* processor() const { return processor_; }
Isolate* isolate() const { return isolate_; }
private:
void StartProcessorIfNotStarted();
void StopProcessorIfLastProfile(const char* title);
void StopProcessor();
void ResetProfiles();
void LogBuiltins();
Isolate* isolate_;
TimeDelta sampling_interval_;
CpuProfilesCollection* profiles_;
ProfileGenerator* generator_;
ProfilerEventsProcessor* processor_;
bool saved_is_logging_;
bool is_profiling_;
DISALLOW_COPY_AND_ASSIGN(CpuProfiler);
};
} } // namespace v8::internal
#endif // V8_CPU_PROFILER_H_