| /* |
| * Copyright (C) 2016-2019 Apple Inc. All rights reserved. |
| * |
| * Redistribution and use in source and binary forms, with or without |
| * modification, are permitted provided that the following conditions |
| * are met: |
| * 1. Redistributions of source code must retain the above copyright |
| * notice, this list of conditions and the following disclaimer. |
| * 2. Redistributions in binary form must reproduce the above copyright |
| * notice, this list of conditions and the following disclaimer in the |
| * documentation and/or other materials provided with the distribution. |
| * |
| * THIS SOFTWARE IS PROVIDED BY APPLE INC. ``AS IS'' AND ANY |
| * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR |
| * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR |
| * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
| * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY |
| * OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| */ |
| |
| #include "config.h" |
| #include "SamplingProfiler.h" |
| |
| #if ENABLE(SAMPLING_PROFILER) |
| |
| #include "CallFrame.h" |
| #include "CatchScope.h" |
| #include "CodeBlock.h" |
| #include "CodeBlockSet.h" |
| #include "HeapIterationScope.h" |
| #include "HeapUtil.h" |
| #include "InlineCallFrame.h" |
| #include "Interpreter.h" |
| #include "JSCInlines.h" |
| #include "JSFunction.h" |
| #include "LLIntPCRanges.h" |
| #include "MachineContext.h" |
| #include "MarkedBlock.h" |
| #include "MarkedBlockSet.h" |
| #include "MarkedSpaceInlines.h" |
| #include "NativeExecutable.h" |
| #include "PCToCodeOriginMap.h" |
| #include "SlotVisitor.h" |
| #include "StrongInlines.h" |
| #include "VM.h" |
| #include "WasmCallee.h" |
| #include "WasmCalleeRegistry.h" |
| #include <thread> |
| #include <wtf/FilePrintStream.h> |
| #include <wtf/HashSet.h> |
| #include <wtf/RefPtr.h> |
| #include <wtf/StackTrace.h> |
| #include <wtf/text/StringBuilder.h> |
| #include <wtf/text/StringConcatenateNumbers.h> |
| |
| namespace JSC { |
| |
| static double sNumTotalStackTraces = 0; |
| static double sNumTotalWalks = 0; |
| static double sNumFailedWalks = 0; |
| static const uint32_t sNumWalkReportingFrequency = 50; |
| static const double sWalkErrorPercentage = .05; |
| static constexpr bool sReportStatsOnlyWhenTheyreAboveThreshold = false; |
| static constexpr bool sReportStats = false; |
| |
| using FrameType = SamplingProfiler::FrameType; |
| using UnprocessedStackFrame = SamplingProfiler::UnprocessedStackFrame; |
| |
| ALWAYS_INLINE static void reportStats() |
| { |
| if (sReportStats && sNumTotalWalks && static_cast<uint64_t>(sNumTotalWalks) % sNumWalkReportingFrequency == 0) { |
| if (!sReportStatsOnlyWhenTheyreAboveThreshold || (sNumFailedWalks / sNumTotalWalks > sWalkErrorPercentage)) { |
| dataLogF("Num total walks: %llu. Failed walks percent: %lf\n", |
| static_cast<unsigned long long>(sNumTotalWalks), sNumFailedWalks / sNumTotalWalks); |
| } |
| } |
| } |
| |
| class FrameWalker { |
| public: |
| FrameWalker(VM& vm, CallFrame* callFrame, const AbstractLocker& codeBlockSetLocker, const AbstractLocker& machineThreadsLocker, const AbstractLocker& wasmCalleeLocker) |
| : m_vm(vm) |
| , m_callFrame(callFrame) |
| , m_entryFrame(vm.topEntryFrame) |
| , m_codeBlockSetLocker(codeBlockSetLocker) |
| , m_machineThreadsLocker(machineThreadsLocker) |
| , m_wasmCalleeLocker(wasmCalleeLocker) |
| { |
| } |
| |
| SUPPRESS_ASAN |
| size_t walk(Vector<UnprocessedStackFrame>& stackTrace, bool& didRunOutOfSpace) |
| { |
| if (sReportStats) |
| sNumTotalWalks++; |
| resetAtMachineFrame(); |
| size_t maxStackTraceSize = stackTrace.size(); |
| while (!isAtTop() && !m_bailingOut && m_depth < maxStackTraceSize) { |
| recordJITFrame(stackTrace); |
| advanceToParentFrame(); |
| resetAtMachineFrame(); |
| } |
| didRunOutOfSpace = m_depth >= maxStackTraceSize && !isAtTop(); |
| reportStats(); |
| return m_depth; |
| } |
| |
| bool wasValidWalk() const |
| { |
| return !m_bailingOut; |
| } |
| |
| protected: |
| |
| SUPPRESS_ASAN |
| void recordJITFrame(Vector<UnprocessedStackFrame>& stackTrace) |
| { |
| CallSiteIndex callSiteIndex; |
| CalleeBits unsafeCallee = m_callFrame->unsafeCallee(); |
| CodeBlock* codeBlock = m_callFrame->unsafeCodeBlock(); |
| if (unsafeCallee.isWasm()) |
| codeBlock = nullptr; |
| if (codeBlock) { |
| ASSERT(isValidCodeBlock(codeBlock)); |
| callSiteIndex = m_callFrame->unsafeCallSiteIndex(); |
| } |
| stackTrace[m_depth] = UnprocessedStackFrame(codeBlock, unsafeCallee, callSiteIndex); |
| #if ENABLE(WEBASSEMBLY) |
| if (unsafeCallee.isWasm()) { |
| auto* wasmCallee = unsafeCallee.asWasmCallee(); |
| if (Wasm::CalleeRegistry::singleton().isValidCallee(m_wasmCalleeLocker, wasmCallee)) { |
| // At this point, Wasm::Callee would be dying (ref count is 0), but its fields are still live. |
| // And we can safely copy Wasm::IndexOrName even when any lock is held by suspended threads. |
| stackTrace[m_depth].wasmIndexOrName = wasmCallee->indexOrName(); |
| stackTrace[m_depth].wasmCompilationMode = wasmCallee->compilationMode(); |
| } |
| } |
| #endif |
| m_depth++; |
| } |
| |
| SUPPRESS_ASAN |
| void advanceToParentFrame() |
| { |
| m_callFrame = m_callFrame->unsafeCallerFrame(m_entryFrame); |
| } |
| |
| bool isAtTop() const |
| { |
| return !m_callFrame; |
| } |
| |
| SUPPRESS_ASAN |
| void resetAtMachineFrame() |
| { |
| if (isAtTop()) |
| return; |
| |
| if (!isValidFramePointer(m_callFrame)) { |
| // Guard against pausing the process at weird program points. |
| m_bailingOut = true; |
| if (sReportStats) |
| sNumFailedWalks++; |
| return; |
| } |
| |
| CodeBlock* codeBlock = m_callFrame->unsafeCodeBlock(); |
| if (!codeBlock || m_callFrame->unsafeCallee().isWasm()) |
| return; |
| |
| if (!isValidCodeBlock(codeBlock)) { |
| m_bailingOut = true; |
| if (sReportStats) |
| sNumFailedWalks++; |
| return; |
| } |
| } |
| |
| bool isValidFramePointer(void* callFrame) |
| { |
| uint8_t* fpCast = bitwise_cast<uint8_t*>(callFrame); |
| for (auto& thread : m_vm.heap.machineThreads().threads(m_machineThreadsLocker)) { |
| uint8_t* stackBase = static_cast<uint8_t*>(thread->stack().origin()); |
| uint8_t* stackLimit = static_cast<uint8_t*>(thread->stack().end()); |
| RELEASE_ASSERT(stackBase); |
| RELEASE_ASSERT(stackLimit); |
| RELEASE_ASSERT(stackLimit <= stackBase); |
| if (fpCast < stackBase && fpCast >= stackLimit) |
| return true; |
| } |
| return false; |
| } |
| |
| bool isValidCodeBlock(CodeBlock* codeBlock) |
| { |
| if (!codeBlock) |
| return false; |
| bool result = m_vm.heap.codeBlockSet().contains(m_codeBlockSetLocker, codeBlock); |
| return result; |
| } |
| |
| VM& m_vm; |
| CallFrame* m_callFrame; |
| EntryFrame* m_entryFrame; |
| const AbstractLocker& m_codeBlockSetLocker; |
| const AbstractLocker& m_machineThreadsLocker; |
| const AbstractLocker& m_wasmCalleeLocker; |
| bool m_bailingOut { false }; |
| size_t m_depth { 0 }; |
| }; |
| |
| class CFrameWalker : public FrameWalker { |
| public: |
| typedef FrameWalker Base; |
| |
| CFrameWalker(VM& vm, void* machineFrame, CallFrame* callFrame, const AbstractLocker& codeBlockSetLocker, const AbstractLocker& machineThreadsLocker, const AbstractLocker& wasmCalleeLocker) |
| : Base(vm, callFrame, codeBlockSetLocker, machineThreadsLocker, wasmCalleeLocker) |
| , m_machineFrame(machineFrame) |
| { |
| } |
| |
| size_t walk(Vector<UnprocessedStackFrame>& stackTrace, bool& didRunOutOfSpace) |
| { |
| if (sReportStats) |
| sNumTotalWalks++; |
| resetAtMachineFrame(); |
| size_t maxStackTraceSize = stackTrace.size(); |
| // The way the C walker decides if a frame it is about to trace is C or JS is by |
| // ensuring m_callFrame points to some frame above the machineFrame. |
| if (!isAtTop() && !m_bailingOut && m_machineFrame == m_callFrame) { |
| recordJITFrame(stackTrace); |
| Base::advanceToParentFrame(); |
| resetAtMachineFrame(); |
| } |
| |
| while (!isAtTop() && !m_bailingOut && m_depth < maxStackTraceSize) { |
| if (m_machineFrame >= m_callFrame) { |
| // If we get to this state we probably have an invalid trace. |
| m_bailingOut = true; |
| break; |
| } |
| |
| if (isCFrame()) { |
| RELEASE_ASSERT(!LLInt::isLLIntPC(frame()->callerFrame)); |
| stackTrace[m_depth] = UnprocessedStackFrame(frame()->returnPC); |
| m_depth++; |
| } else |
| recordJITFrame(stackTrace); |
| advanceToParentFrame(); |
| resetAtMachineFrame(); |
| } |
| didRunOutOfSpace = m_depth >= maxStackTraceSize && !isAtTop(); |
| reportStats(); |
| return m_depth; |
| } |
| |
| private: |
| |
| bool isCFrame() |
| { |
| return frame()->callerFrame != m_callFrame; |
| } |
| |
| void advanceToParentFrame() |
| { |
| if (!isCFrame()) |
| Base::advanceToParentFrame(); |
| m_machineFrame = frame()->callerFrame; |
| } |
| |
| void resetAtMachineFrame() |
| { |
| if (!isValidFramePointer(m_machineFrame)) { |
| // Guard against pausing the process at weird program points. |
| m_bailingOut = true; |
| if (sReportStats) |
| sNumFailedWalks++; |
| return; |
| } |
| Base::resetAtMachineFrame(); |
| } |
| |
| CallerFrameAndPC* frame() |
| { |
| return reinterpret_cast<CallerFrameAndPC*>(m_machineFrame); |
| } |
| |
| void* m_machineFrame; |
| }; |
| |
| SamplingProfiler::SamplingProfiler(VM& vm, RefPtr<Stopwatch>&& stopwatch) |
| : m_isPaused(false) |
| , m_isShutDown(false) |
| , m_vm(vm) |
| , m_weakRandom() |
| , m_stopwatch(WTFMove(stopwatch)) |
| , m_timingInterval(Seconds::fromMicroseconds(Options::sampleInterval())) |
| { |
| if (sReportStats) { |
| sNumTotalWalks = 0; |
| sNumFailedWalks = 0; |
| } |
| |
| m_currentFrames.grow(256); |
| vm.heap.objectSpace().enablePreciseAllocationTracking(); |
| } |
| |
| SamplingProfiler::~SamplingProfiler() |
| { |
| } |
| |
| void SamplingProfiler::createThreadIfNecessary(const AbstractLocker&) |
| { |
| ASSERT(m_lock.isLocked()); |
| |
| if (m_thread) |
| return; |
| |
| RefPtr<SamplingProfiler> profiler = this; |
| m_thread = Thread::create("jsc.sampling-profiler.thread", [profiler] { |
| profiler->timerLoop(); |
| }); |
| } |
| |
| void SamplingProfiler::timerLoop() |
| { |
| while (true) { |
| Seconds stackTraceProcessingTime = 0_s; |
| { |
| LockHolder locker(m_lock); |
| if (UNLIKELY(m_isShutDown)) |
| return; |
| |
| if (!m_isPaused && m_jscExecutionThread) |
| takeSample(locker, stackTraceProcessingTime); |
| |
| m_lastTime = m_stopwatch->elapsedTime(); |
| } |
| |
| // Read section 6.2 of this paper for more elaboration of why we add a random |
| // fluctuation here. The main idea is to prevent our timer from being in sync |
| // with some system process such as a scheduled context switch. |
| // http://plv.colorado.edu/papers/mytkowicz-pldi10.pdf |
| double randomSignedNumber = (m_weakRandom.get() * 2.0) - 1.0; // A random number between [-1, 1). |
| Seconds randomFluctuation = m_timingInterval * 0.2 * randomSignedNumber; |
| WTF::sleep(m_timingInterval - std::min(m_timingInterval, stackTraceProcessingTime) + randomFluctuation); |
| } |
| } |
| |
| void SamplingProfiler::takeSample(const AbstractLocker&, Seconds& stackTraceProcessingTime) |
| { |
| ASSERT(m_lock.isLocked()); |
| if (m_vm.entryScope) { |
| Seconds nowTime = m_stopwatch->elapsedTime(); |
| |
| auto machineThreadsLocker = holdLock(m_vm.heap.machineThreads().getLock()); |
| auto codeBlockSetLocker = holdLock(m_vm.heap.codeBlockSet().getLock()); |
| auto executableAllocatorLocker = holdLock(ExecutableAllocator::singleton().getLock()); |
| #if ENABLE(WEBASSEMBLY) |
| auto wasmCalleesLocker = holdLock(Wasm::CalleeRegistry::singleton().getLock()); |
| #else |
| LockHolder wasmCalleesLocker(NoLockingNecessary); |
| #endif |
| |
| auto didSuspend = m_jscExecutionThread->suspend(); |
| if (didSuspend) { |
| // While the JSC thread is suspended, we can't do things like malloc because the JSC thread |
| // may be holding the malloc lock. |
| void* machineFrame; |
| CallFrame* callFrame; |
| void* machinePC; |
| bool topFrameIsLLInt = false; |
| void* llintPC; |
| { |
| PlatformRegisters registers; |
| m_jscExecutionThread->getRegisters(registers); |
| machineFrame = MachineContext::framePointer(registers); |
| callFrame = static_cast<CallFrame*>(machineFrame); |
| auto instructionPointer = MachineContext::instructionPointer(registers); |
| if (instructionPointer) |
| machinePC = instructionPointer->untaggedExecutableAddress(); |
| else |
| machinePC = nullptr; |
| llintPC = removeCodePtrTag(MachineContext::llintInstructionPointer(registers)); |
| assertIsNotTagged(machinePC); |
| } |
| // FIXME: Lets have a way of detecting when we're parsing code. |
| // https://bugs.webkit.org/show_bug.cgi?id=152761 |
| if (ExecutableAllocator::singleton().isValidExecutableMemory(executableAllocatorLocker, machinePC)) { |
| if (m_vm.isExecutingInRegExpJIT) { |
| // FIXME: We're executing a regexp. Lets gather more intersting data. |
| // https://bugs.webkit.org/show_bug.cgi?id=152729 |
| callFrame = m_vm.topCallFrame; // We need to do this or else we'd fail our backtrace validation b/c this isn't a JS frame. |
| } |
| } else if (LLInt::isLLIntPC(machinePC)) { |
| topFrameIsLLInt = true; |
| // We're okay to take a normal stack trace when the PC |
| // is in LLInt code. |
| } else { |
| // We resort to topCallFrame to see if we can get anything |
| // useful. We usually get here when we're executing C code. |
| callFrame = m_vm.topCallFrame; |
| } |
| |
| size_t walkSize; |
| bool wasValidWalk; |
| bool didRunOutOfVectorSpace; |
| if (Options::sampleCCode()) { |
| CFrameWalker walker(m_vm, machineFrame, callFrame, codeBlockSetLocker, machineThreadsLocker, wasmCalleesLocker); |
| walkSize = walker.walk(m_currentFrames, didRunOutOfVectorSpace); |
| wasValidWalk = walker.wasValidWalk(); |
| } else { |
| FrameWalker walker(m_vm, callFrame, codeBlockSetLocker, machineThreadsLocker, wasmCalleesLocker); |
| walkSize = walker.walk(m_currentFrames, didRunOutOfVectorSpace); |
| wasValidWalk = walker.wasValidWalk(); |
| } |
| |
| m_jscExecutionThread->resume(); |
| |
| auto startTime = MonotonicTime::now(); |
| // We can now use data structures that malloc, and do other interesting things, again. |
| |
| // FIXME: It'd be interesting to take data about the program's state when |
| // we fail to take a stack trace: https://bugs.webkit.org/show_bug.cgi?id=152758 |
| if (wasValidWalk && walkSize) { |
| if (sReportStats) |
| sNumTotalStackTraces++; |
| Vector<UnprocessedStackFrame> stackTrace; |
| stackTrace.reserveInitialCapacity(walkSize); |
| for (size_t i = 0; i < walkSize; i++) { |
| UnprocessedStackFrame frame = m_currentFrames[i]; |
| stackTrace.uncheckedAppend(frame); |
| } |
| |
| m_unprocessedStackTraces.append(UnprocessedStackTrace { nowTime, machinePC, topFrameIsLLInt, llintPC, WTFMove(stackTrace) }); |
| |
| if (didRunOutOfVectorSpace) |
| m_currentFrames.grow(m_currentFrames.size() * 1.25); |
| } |
| |
| auto endTime = MonotonicTime::now(); |
| stackTraceProcessingTime = endTime - startTime; |
| } |
| } |
| } |
| |
| static ALWAYS_INLINE BytecodeIndex tryGetBytecodeIndex(unsigned llintPC, CodeBlock* codeBlock) |
| { |
| #if ENABLE(DFG_JIT) |
| RELEASE_ASSERT(!codeBlock->hasCodeOrigins()); |
| #endif |
| |
| unsigned bytecodeOffset = llintPC; |
| if (bytecodeOffset < codeBlock->instructionsSize()) |
| return BytecodeIndex(bytecodeOffset); |
| return BytecodeIndex(); |
| } |
| |
| void SamplingProfiler::processUnverifiedStackTraces(const AbstractLocker&) |
| { |
| // This function needs to be called from the JSC execution thread. |
| RELEASE_ASSERT(m_lock.isLocked()); |
| |
| TinyBloomFilter filter = m_vm.heap.objectSpace().blocks().filter(); |
| |
| for (UnprocessedStackTrace& unprocessedStackTrace : m_unprocessedStackTraces) { |
| m_stackTraces.append(StackTrace()); |
| StackTrace& stackTrace = m_stackTraces.last(); |
| stackTrace.timestamp = unprocessedStackTrace.timestamp; |
| |
| auto populateCodeLocation = [] (CodeBlock* codeBlock, BytecodeIndex bytecodeIndex, StackFrame::CodeLocation& location) { |
| if (bytecodeIndex.offset() < codeBlock->instructionsSize()) { |
| int divot; |
| int startOffset; |
| int endOffset; |
| codeBlock->expressionRangeForBytecodeIndex(bytecodeIndex, divot, startOffset, endOffset, |
| location.lineNumber, location.columnNumber); |
| location.bytecodeIndex = bytecodeIndex; |
| } |
| if (Options::collectSamplingProfilerDataForJSCShell()) { |
| location.codeBlockHash = codeBlock->hash(); |
| location.jitType = codeBlock->jitType(); |
| } |
| }; |
| |
| auto appendCodeBlock = [&] (CodeBlock* codeBlock, BytecodeIndex bytecodeIndex) { |
| stackTrace.frames.append(StackFrame(codeBlock->ownerExecutable())); |
| m_liveCellPointers.add(codeBlock->ownerExecutable()); |
| populateCodeLocation(codeBlock, bytecodeIndex, stackTrace.frames.last().semanticLocation); |
| }; |
| |
| auto appendEmptyFrame = [&] { |
| stackTrace.frames.append(StackFrame()); |
| }; |
| |
| auto storeCalleeIntoLastFrame = [&] (UnprocessedStackFrame& unprocessedStackFrame) { |
| // Set the callee if it's a valid GC object. |
| CalleeBits calleeBits = unprocessedStackFrame.unverifiedCallee; |
| StackFrame& stackFrame = stackTrace.frames.last(); |
| bool alreadyHasExecutable = !!stackFrame.executable; |
| #if ENABLE(WEBASSEMBLY) |
| if (calleeBits.isWasm()) { |
| stackFrame.frameType = FrameType::Wasm; |
| stackFrame.wasmIndexOrName = unprocessedStackFrame.wasmIndexOrName; |
| stackFrame.wasmCompilationMode = unprocessedStackFrame.wasmCompilationMode; |
| return; |
| } |
| #endif |
| |
| JSValue callee = calleeBits.asCell(); |
| if (!HeapUtil::isValueGCObject(m_vm.heap, filter, callee)) { |
| if (!alreadyHasExecutable) |
| stackFrame.frameType = FrameType::Unknown; |
| return; |
| } |
| |
| JSCell* calleeCell = callee.asCell(); |
| auto setFallbackFrameType = [&] { |
| ASSERT(!alreadyHasExecutable); |
| FrameType result = FrameType::Unknown; |
| CallData callData; |
| CallType callType; |
| callType = getCallData(m_vm, calleeCell, callData); |
| if (callType == CallType::Host) |
| result = FrameType::Host; |
| |
| stackFrame.frameType = result; |
| }; |
| |
| auto addCallee = [&] (JSObject* callee) { |
| stackFrame.callee = callee; |
| m_liveCellPointers.add(callee); |
| }; |
| |
| if (calleeCell->type() != JSFunctionType) { |
| if (JSObject* object = jsDynamicCast<JSObject*>(calleeCell->vm(), calleeCell)) |
| addCallee(object); |
| |
| if (!alreadyHasExecutable) |
| setFallbackFrameType(); |
| |
| return; |
| } |
| |
| addCallee(jsCast<JSFunction*>(calleeCell)); |
| |
| if (alreadyHasExecutable) |
| return; |
| |
| ExecutableBase* executable = jsCast<JSFunction*>(calleeCell)->executable(); |
| if (!executable) { |
| setFallbackFrameType(); |
| return; |
| } |
| |
| RELEASE_ASSERT(HeapUtil::isPointerGCObjectJSCell(m_vm.heap, filter, executable)); |
| stackFrame.frameType = FrameType::Executable; |
| stackFrame.executable = executable; |
| m_liveCellPointers.add(executable); |
| }; |
| |
| auto appendCodeOrigin = [&] (CodeBlock* machineCodeBlock, CodeOrigin origin) { |
| size_t startIndex = stackTrace.frames.size(); // We want to change stack traces that we're about to append. |
| |
| CodeOrigin machineOrigin; |
| origin.walkUpInlineStack([&] (const CodeOrigin& codeOrigin) { |
| machineOrigin = codeOrigin; |
| auto* inlineCallFrame = codeOrigin.inlineCallFrame(); |
| appendCodeBlock(inlineCallFrame ? inlineCallFrame->baselineCodeBlock.get() : machineCodeBlock, codeOrigin.bytecodeIndex()); |
| }); |
| |
| if (Options::collectSamplingProfilerDataForJSCShell()) { |
| RELEASE_ASSERT(machineOrigin.isSet()); |
| RELEASE_ASSERT(!machineOrigin.inlineCallFrame()); |
| |
| StackFrame::CodeLocation machineLocation = stackTrace.frames.last().semanticLocation; |
| |
| // We want to tell each inlined frame about the machine frame |
| // they were inlined into. Currently, we only use this for dumping |
| // output on the command line, but we could extend it to the web |
| // inspector in the future if we find a need for it there. |
| RELEASE_ASSERT(stackTrace.frames.size()); |
| m_liveCellPointers.add(machineCodeBlock); |
| for (size_t i = startIndex; i < stackTrace.frames.size() - 1; i++) |
| stackTrace.frames[i].machineLocation = std::make_pair(machineLocation, machineCodeBlock); |
| } |
| }; |
| |
| // Prepend the top-most inlined frame if needed and gather |
| // location information about where the top frame is executing. |
| size_t startIndex = 0; |
| if (unprocessedStackTrace.frames.size() && !!unprocessedStackTrace.frames[0].verifiedCodeBlock) { |
| CodeBlock* topCodeBlock = unprocessedStackTrace.frames[0].verifiedCodeBlock; |
| if (unprocessedStackTrace.topFrameIsLLInt) { |
| // We reuse LLInt CodeBlocks for the baseline JIT, so we need to check for both jit types. |
| // This might also be false for various reasons (known and unknown), even though |
| // it's super unlikely. One reason that this can be false is when we throw from a DFG frame, |
| // and we end up having to unwind past an EntryFrame, we will end up executing |
| // inside the LLInt's handleUncaughtException. So we just protect against this |
| // by ignoring it. |
| BytecodeIndex bytecodeIndex = BytecodeIndex(0); |
| if (topCodeBlock->jitType() == JITType::InterpreterThunk || topCodeBlock->jitType() == JITType::BaselineJIT) { |
| unsigned bits = static_cast<unsigned>(bitwise_cast<uintptr_t>(unprocessedStackTrace.llintPC)); |
| bytecodeIndex = tryGetBytecodeIndex(bits, topCodeBlock); |
| |
| UNUSED_PARAM(bytecodeIndex); // FIXME: do something with this info for the web inspector: https://bugs.webkit.org/show_bug.cgi?id=153455 |
| |
| appendCodeBlock(topCodeBlock, bytecodeIndex); |
| storeCalleeIntoLastFrame(unprocessedStackTrace.frames[0]); |
| startIndex = 1; |
| } |
| } else { |
| #if ENABLE(JIT) |
| if (Optional<CodeOrigin> codeOrigin = topCodeBlock->findPC(unprocessedStackTrace.topPC)) { |
| appendCodeOrigin(topCodeBlock, *codeOrigin); |
| storeCalleeIntoLastFrame(unprocessedStackTrace.frames[0]); |
| startIndex = 1; |
| } |
| #endif |
| UNUSED_PARAM(appendCodeOrigin); |
| } |
| } |
| |
| for (size_t i = startIndex; i < unprocessedStackTrace.frames.size(); i++) { |
| UnprocessedStackFrame& unprocessedStackFrame = unprocessedStackTrace.frames[i]; |
| if (CodeBlock* codeBlock = unprocessedStackFrame.verifiedCodeBlock) { |
| CallSiteIndex callSiteIndex = unprocessedStackFrame.callSiteIndex; |
| |
| auto appendCodeBlockNoInlining = [&] { |
| appendCodeBlock(codeBlock, tryGetBytecodeIndex(callSiteIndex.bits(), codeBlock)); |
| }; |
| |
| #if ENABLE(DFG_JIT) |
| if (codeBlock->hasCodeOrigins()) { |
| if (codeBlock->canGetCodeOrigin(callSiteIndex)) |
| appendCodeOrigin(codeBlock, codeBlock->codeOrigin(callSiteIndex)); |
| else |
| appendCodeBlock(codeBlock, BytecodeIndex()); |
| } else |
| appendCodeBlockNoInlining(); |
| #else |
| appendCodeBlockNoInlining(); |
| #endif |
| } else if (unprocessedStackFrame.cCodePC) { |
| appendEmptyFrame(); |
| stackTrace.frames.last().cCodePC = unprocessedStackFrame.cCodePC; |
| stackTrace.frames.last().frameType = FrameType::C; |
| } else |
| appendEmptyFrame(); |
| |
| // Note that this is okay to do if we walked the inline stack because |
| // the machine frame will be at the top of the processed stack trace. |
| if (!unprocessedStackFrame.cCodePC) |
| storeCalleeIntoLastFrame(unprocessedStackFrame); |
| } |
| } |
| |
| m_unprocessedStackTraces.clear(); |
| } |
| |
| void SamplingProfiler::visit(SlotVisitor& slotVisitor) |
| { |
| RELEASE_ASSERT(m_lock.isLocked()); |
| for (JSCell* cell : m_liveCellPointers) |
| slotVisitor.appendUnbarriered(cell); |
| } |
| |
| void SamplingProfiler::shutdown() |
| { |
| LockHolder locker(m_lock); |
| m_isShutDown = true; |
| } |
| |
| void SamplingProfiler::start() |
| { |
| LockHolder locker(m_lock); |
| start(locker); |
| } |
| |
| void SamplingProfiler::start(const AbstractLocker& locker) |
| { |
| ASSERT(m_lock.isLocked()); |
| m_isPaused = false; |
| createThreadIfNecessary(locker); |
| } |
| |
| void SamplingProfiler::pause(const AbstractLocker&) |
| { |
| ASSERT(m_lock.isLocked()); |
| m_isPaused = true; |
| reportStats(); |
| } |
| |
| void SamplingProfiler::noticeCurrentThreadAsJSCExecutionThread(const AbstractLocker&) |
| { |
| ASSERT(m_lock.isLocked()); |
| m_jscExecutionThread = &Thread::current(); |
| } |
| |
| void SamplingProfiler::noticeCurrentThreadAsJSCExecutionThread() |
| { |
| LockHolder locker(m_lock); |
| noticeCurrentThreadAsJSCExecutionThread(locker); |
| } |
| |
| void SamplingProfiler::noticeJSLockAcquisition() |
| { |
| LockHolder locker(m_lock); |
| noticeCurrentThreadAsJSCExecutionThread(locker); |
| } |
| |
| void SamplingProfiler::noticeVMEntry() |
| { |
| LockHolder locker(m_lock); |
| ASSERT(m_vm.entryScope); |
| noticeCurrentThreadAsJSCExecutionThread(locker); |
| m_lastTime = m_stopwatch->elapsedTime(); |
| createThreadIfNecessary(locker); |
| } |
| |
| void SamplingProfiler::clearData(const AbstractLocker&) |
| { |
| ASSERT(m_lock.isLocked()); |
| m_stackTraces.clear(); |
| m_liveCellPointers.clear(); |
| m_unprocessedStackTraces.clear(); |
| } |
| |
| String SamplingProfiler::StackFrame::nameFromCallee(VM& vm) |
| { |
| if (!callee) |
| return String(); |
| |
| auto scope = DECLARE_CATCH_SCOPE(vm); |
| JSGlobalObject* globalObject = callee->globalObject(vm); |
| auto getPropertyIfPureOperation = [&] (const Identifier& ident) -> String { |
| PropertySlot slot(callee, PropertySlot::InternalMethodType::VMInquiry); |
| PropertyName propertyName(ident); |
| bool hasProperty = callee->getPropertySlot(globalObject, propertyName, slot); |
| scope.assertNoException(); |
| if (hasProperty) { |
| if (slot.isValue()) { |
| JSValue nameValue = slot.getValue(globalObject, propertyName); |
| if (isJSString(nameValue)) |
| return asString(nameValue)->tryGetValue(); |
| } |
| } |
| return String(); |
| }; |
| |
| String name = getPropertyIfPureOperation(vm.propertyNames->displayName); |
| if (!name.isEmpty()) |
| return name; |
| |
| return getPropertyIfPureOperation(vm.propertyNames->name); |
| } |
| |
| String SamplingProfiler::StackFrame::displayName(VM& vm) |
| { |
| { |
| String name = nameFromCallee(vm); |
| if (!name.isEmpty()) |
| return name; |
| } |
| |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::C: |
| #if HAVE(DLADDR) |
| if (frameType == FrameType::C) { |
| auto demangled = WTF::StackTrace::demangle(const_cast<void*>(cCodePC)); |
| if (demangled) |
| return String(demangled->demangledName() ? demangled->demangledName() : demangled->mangledName()); |
| WTF::dataLog("couldn't get a name"); |
| } |
| #endif |
| return "(unknown)"_s; |
| |
| case FrameType::Host: |
| return "(host)"_s; |
| |
| case FrameType::Wasm: |
| #if ENABLE(WEBASSEMBLY) |
| if (wasmIndexOrName) |
| return makeString(wasmIndexOrName.value()); |
| #endif |
| return "(wasm)"_s; |
| |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return static_cast<NativeExecutable*>(executable)->name(); |
| |
| if (executable->isFunctionExecutable()) |
| return static_cast<FunctionExecutable*>(executable)->ecmaName().string(); |
| if (executable->isProgramExecutable() || executable->isEvalExecutable()) |
| return "(program)"_s; |
| if (executable->isModuleProgramExecutable()) |
| return "(module)"_s; |
| |
| RELEASE_ASSERT_NOT_REACHED(); |
| return String(); |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return String(); |
| } |
| |
| String SamplingProfiler::StackFrame::displayNameForJSONTests(VM& vm) |
| { |
| { |
| String name = nameFromCallee(vm); |
| if (!name.isEmpty()) |
| return name; |
| } |
| |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::C: |
| return "(unknown)"_s; |
| |
| case FrameType::Host: |
| return "(host)"_s; |
| |
| case FrameType::Wasm: { |
| #if ENABLE(WEBASSEMBLY) |
| if (wasmIndexOrName) |
| return makeString(wasmIndexOrName.value()); |
| #endif |
| return "(wasm)"_s; |
| } |
| |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return static_cast<NativeExecutable*>(executable)->name(); |
| |
| if (executable->isFunctionExecutable()) { |
| String result = static_cast<FunctionExecutable*>(executable)->ecmaName().string(); |
| if (result.isEmpty()) |
| return "(anonymous function)"_s; |
| return result; |
| } |
| if (executable->isEvalExecutable()) |
| return "(eval)"_s; |
| if (executable->isProgramExecutable()) |
| return "(program)"_s; |
| if (executable->isModuleProgramExecutable()) |
| return "(module)"_s; |
| |
| RELEASE_ASSERT_NOT_REACHED(); |
| return String(); |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return String(); |
| } |
| |
| int SamplingProfiler::StackFrame::functionStartLine() |
| { |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::Host: |
| case FrameType::C: |
| case FrameType::Wasm: |
| return -1; |
| |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return -1; |
| return static_cast<ScriptExecutable*>(executable)->firstLine(); |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return -1; |
| } |
| |
| unsigned SamplingProfiler::StackFrame::functionStartColumn() |
| { |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::Host: |
| case FrameType::C: |
| case FrameType::Wasm: |
| return std::numeric_limits<unsigned>::max(); |
| |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return std::numeric_limits<unsigned>::max(); |
| |
| return static_cast<ScriptExecutable*>(executable)->startColumn(); |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return std::numeric_limits<unsigned>::max(); |
| } |
| |
| intptr_t SamplingProfiler::StackFrame::sourceID() |
| { |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::Host: |
| case FrameType::C: |
| case FrameType::Wasm: |
| return -1; |
| |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return -1; |
| |
| return static_cast<ScriptExecutable*>(executable)->sourceID(); |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return -1; |
| } |
| |
| String SamplingProfiler::StackFrame::url() |
| { |
| switch (frameType) { |
| case FrameType::Unknown: |
| case FrameType::Host: |
| case FrameType::C: |
| case FrameType::Wasm: |
| return emptyString(); |
| case FrameType::Executable: |
| if (executable->isHostFunction()) |
| return emptyString(); |
| |
| String url = static_cast<ScriptExecutable*>(executable)->sourceURL(); |
| if (url.isEmpty()) |
| return static_cast<ScriptExecutable*>(executable)->source().provider()->sourceURLDirective(); // Fall back to sourceURL directive. |
| return url; |
| } |
| RELEASE_ASSERT_NOT_REACHED(); |
| return String(); |
| } |
| |
| Vector<SamplingProfiler::StackTrace> SamplingProfiler::releaseStackTraces(const AbstractLocker& locker) |
| { |
| ASSERT(m_lock.isLocked()); |
| { |
| HeapIterationScope heapIterationScope(m_vm.heap); |
| processUnverifiedStackTraces(locker); |
| } |
| |
| Vector<StackTrace> result(WTFMove(m_stackTraces)); |
| clearData(locker); |
| return result; |
| } |
| |
| String SamplingProfiler::stackTracesAsJSON() |
| { |
| DeferGC deferGC(m_vm.heap); |
| auto locker = holdLock(m_lock); |
| |
| { |
| HeapIterationScope heapIterationScope(m_vm.heap); |
| processUnverifiedStackTraces(locker); |
| } |
| |
| StringBuilder json; |
| json.append('['); |
| |
| bool loopedOnce = false; |
| auto comma = [&] { |
| if (loopedOnce) |
| json.append(','); |
| }; |
| for (StackTrace& stackTrace : m_stackTraces) { |
| comma(); |
| json.append('['); |
| loopedOnce = false; |
| for (StackFrame& stackFrame : stackTrace.frames) { |
| comma(); |
| json.appendQuotedJSONString(stackFrame.displayNameForJSONTests(m_vm)); |
| loopedOnce = true; |
| } |
| json.append(']'); |
| loopedOnce = true; |
| } |
| |
| json.append(']'); |
| |
| clearData(locker); |
| |
| return json.toString(); |
| } |
| |
| void SamplingProfiler::registerForReportAtExit() |
| { |
| static Lock registrationLock; |
| static HashSet<RefPtr<SamplingProfiler>>* profilesToReport; |
| |
| LockHolder holder(registrationLock); |
| |
| if (!profilesToReport) { |
| profilesToReport = new HashSet<RefPtr<SamplingProfiler>>(); |
| atexit([]() { |
| for (const auto& profile : *profilesToReport) |
| profile->reportDataToOptionFile(); |
| }); |
| } |
| |
| profilesToReport->add(adoptRef(this)); |
| m_needsReportAtExit = true; |
| } |
| |
| void SamplingProfiler::reportDataToOptionFile() |
| { |
| if (m_needsReportAtExit) { |
| m_needsReportAtExit = false; |
| JSLockHolder holder(m_vm); |
| const char* path = Options::samplingProfilerPath(); |
| StringPrintStream pathOut; |
| pathOut.print(path, "/"); |
| pathOut.print("JSCSampilingProfile-", reinterpret_cast<uintptr_t>(this), ".txt"); |
| auto out = FilePrintStream::open(pathOut.toCString().data(), "w"); |
| reportTopFunctions(*out); |
| reportTopBytecodes(*out); |
| } |
| } |
| |
| void SamplingProfiler::reportTopFunctions() |
| { |
| reportTopFunctions(WTF::dataFile()); |
| } |
| |
| void SamplingProfiler::reportTopFunctions(PrintStream& out) |
| { |
| auto locker = holdLock(m_lock); |
| DeferGCForAWhile deferGC(m_vm.heap); |
| |
| { |
| HeapIterationScope heapIterationScope(m_vm.heap); |
| processUnverifiedStackTraces(locker); |
| } |
| |
| |
| HashMap<String, size_t> functionCounts; |
| for (StackTrace& stackTrace : m_stackTraces) { |
| if (!stackTrace.frames.size()) |
| continue; |
| |
| StackFrame& frame = stackTrace.frames.first(); |
| String frameDescription = makeString(frame.displayName(m_vm), ':', frame.sourceID()); |
| functionCounts.add(frameDescription, 0).iterator->value++; |
| } |
| |
| auto takeMax = [&] () -> std::pair<String, size_t> { |
| String maxFrameDescription; |
| size_t maxFrameCount = 0; |
| for (const auto& entry : functionCounts) { |
| if (entry.value > maxFrameCount) { |
| maxFrameCount = entry.value; |
| maxFrameDescription = entry.key; |
| } |
| } |
| if (!maxFrameDescription.isEmpty()) |
| functionCounts.remove(maxFrameDescription); |
| return std::make_pair(maxFrameDescription, maxFrameCount); |
| }; |
| |
| if (Options::samplingProfilerTopFunctionsCount()) { |
| out.print("\n\nSampling rate: ", m_timingInterval.microseconds(), " microseconds\n"); |
| out.print("Top functions as <numSamples 'functionName:sourceID'>\n"); |
| for (size_t i = 0; i < Options::samplingProfilerTopFunctionsCount(); i++) { |
| auto pair = takeMax(); |
| if (pair.first.isEmpty()) |
| break; |
| out.printf("%6zu ", pair.second); |
| out.print(" '", pair.first, "'\n"); |
| } |
| } |
| } |
| |
| void SamplingProfiler::reportTopBytecodes() |
| { |
| reportTopBytecodes(WTF::dataFile()); |
| } |
| |
| void SamplingProfiler::reportTopBytecodes(PrintStream& out) |
| { |
| auto locker = holdLock(m_lock); |
| DeferGCForAWhile deferGC(m_vm.heap); |
| |
| { |
| HeapIterationScope heapIterationScope(m_vm.heap); |
| processUnverifiedStackTraces(locker); |
| } |
| |
| HashMap<String, size_t> bytecodeCounts; |
| for (StackTrace& stackTrace : m_stackTraces) { |
| if (!stackTrace.frames.size()) |
| continue; |
| |
| auto descriptionForLocation = [&] (StackFrame::CodeLocation location, Optional<Wasm::CompilationMode> wasmCompilationMode) -> String { |
| String bytecodeIndex; |
| String codeBlockHash; |
| String jitType; |
| if (location.hasBytecodeIndex()) |
| bytecodeIndex = toString(location.bytecodeIndex); |
| else |
| bytecodeIndex = "<nil>"; |
| |
| if (location.hasCodeBlockHash()) { |
| StringPrintStream stream; |
| location.codeBlockHash.dump(stream); |
| codeBlockHash = stream.toString(); |
| } else |
| codeBlockHash = "<nil>"; |
| |
| if (wasmCompilationMode) |
| jitType = Wasm::makeString(wasmCompilationMode.value()); |
| else |
| jitType = JITCode::typeName(location.jitType); |
| |
| return makeString("#", codeBlockHash, ":", jitType, ":", bytecodeIndex); |
| }; |
| |
| StackFrame& frame = stackTrace.frames.first(); |
| String frameDescription = makeString(frame.displayName(m_vm), descriptionForLocation(frame.semanticLocation, frame.wasmCompilationMode)); |
| if (Optional<std::pair<StackFrame::CodeLocation, CodeBlock*>> machineLocation = frame.machineLocation) { |
| frameDescription = makeString(frameDescription, " <-- ", |
| machineLocation->second->inferredName().data(), descriptionForLocation(machineLocation->first, WTF::nullopt)); |
| } |
| bytecodeCounts.add(frameDescription, 0).iterator->value++; |
| } |
| |
| auto takeMax = [&] () -> std::pair<String, size_t> { |
| String maxFrameDescription; |
| size_t maxFrameCount = 0; |
| for (const auto& entry : bytecodeCounts) { |
| if (entry.value > maxFrameCount) { |
| maxFrameCount = entry.value; |
| maxFrameDescription = entry.key; |
| } |
| } |
| if (!maxFrameDescription.isEmpty()) |
| bytecodeCounts.remove(maxFrameDescription); |
| return std::make_pair(maxFrameDescription, maxFrameCount); |
| }; |
| |
| if (Options::samplingProfilerTopBytecodesCount()) { |
| out.print("\n\nSampling rate: ", m_timingInterval.microseconds(), " microseconds\n"); |
| out.print("Hottest bytecodes as <numSamples 'functionName#hash:JITType:bytecodeIndex'>\n"); |
| for (size_t i = 0; i < Options::samplingProfilerTopBytecodesCount(); i++) { |
| auto pair = takeMax(); |
| if (pair.first.isEmpty()) |
| break; |
| out.printf("%6zu ", pair.second); |
| out.print(" '", pair.first, "'\n"); |
| } |
| } |
| } |
| |
| #if OS(DARWIN) |
| mach_port_t SamplingProfiler::machThread() |
| { |
| if (!m_thread) |
| return MACH_PORT_NULL; |
| |
| return m_thread->machThread(); |
| } |
| #endif |
| |
| } // namespace JSC |
| |
| namespace WTF { |
| |
| using namespace JSC; |
| |
| void printInternal(PrintStream& out, SamplingProfiler::FrameType frameType) |
| { |
| switch (frameType) { |
| case SamplingProfiler::FrameType::Executable: |
| out.print("Executable"); |
| break; |
| case SamplingProfiler::FrameType::Wasm: |
| out.print("Wasm"); |
| break; |
| case SamplingProfiler::FrameType::Host: |
| out.print("Host"); |
| break; |
| case SamplingProfiler::FrameType::C: |
| case SamplingProfiler::FrameType::Unknown: |
| out.print("Unknown"); |
| break; |
| } |
| } |
| |
| } // namespace WTF |
| |
| #endif // ENABLE(SAMPLING_PROFILER) |