From 36d22d82aa202bb199967e9512281e9a53db42c9 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 7 Apr 2024 21:33:14 +0200 Subject: Adding upstream version 115.7.0esr. Signed-off-by: Daniel Baumann --- .../core/BaseAndGeckoProfilerDetail.cpp | 92 + mozglue/baseprofiler/core/EHABIStackWalk.cpp | 592 +++ mozglue/baseprofiler/core/EHABIStackWalk.h | 30 + mozglue/baseprofiler/core/PageInformation.cpp | 49 + mozglue/baseprofiler/core/PageInformation.h | 77 + mozglue/baseprofiler/core/PlatformMacros.h | 130 + mozglue/baseprofiler/core/ProfileBuffer.cpp | 217 + mozglue/baseprofiler/core/ProfileBuffer.h | 211 + mozglue/baseprofiler/core/ProfileBufferEntry.cpp | 1367 +++++++ mozglue/baseprofiler/core/ProfileBufferEntry.h | 364 ++ mozglue/baseprofiler/core/ProfileJSONWriter.cpp | 101 + mozglue/baseprofiler/core/ProfiledThreadData.cpp | 191 + mozglue/baseprofiler/core/ProfiledThreadData.h | 120 + mozglue/baseprofiler/core/ProfilerBacktrace.cpp | 125 + mozglue/baseprofiler/core/ProfilerBacktrace.h | 162 + mozglue/baseprofiler/core/ProfilerMarkers.cpp | 348 ++ mozglue/baseprofiler/core/ProfilerUtils.cpp | 162 + mozglue/baseprofiler/core/ProfilingCategory.cpp | 71 + mozglue/baseprofiler/core/ProfilingStack.cpp | 52 + mozglue/baseprofiler/core/RegisteredThread.cpp | 42 + mozglue/baseprofiler/core/RegisteredThread.h | 164 + mozglue/baseprofiler/core/ThreadInfo.h | 62 + mozglue/baseprofiler/core/VTuneProfiler.cpp | 92 + mozglue/baseprofiler/core/VTuneProfiler.h | 84 + .../baseprofiler/core/platform-linux-android.cpp | 507 +++ mozglue/baseprofiler/core/platform-macos.cpp | 220 ++ mozglue/baseprofiler/core/platform-win32.cpp | 348 ++ mozglue/baseprofiler/core/platform.cpp | 3824 ++++++++++++++++++ mozglue/baseprofiler/core/platform.h | 149 + .../baseprofiler/core/shared-libraries-linux.cc | 854 ++++ .../baseprofiler/core/shared-libraries-macos.cc | 206 + .../baseprofiler/core/shared-libraries-win32.cc | 222 ++ mozglue/baseprofiler/core/vtune/ittnotify.h | 4127 ++++++++++++++++++++ 33 files changed, 15362 insertions(+) create mode 100644 mozglue/baseprofiler/core/BaseAndGeckoProfilerDetail.cpp create mode 100644 mozglue/baseprofiler/core/EHABIStackWalk.cpp create mode 100644 mozglue/baseprofiler/core/EHABIStackWalk.h create mode 100644 mozglue/baseprofiler/core/PageInformation.cpp create mode 100644 mozglue/baseprofiler/core/PageInformation.h create mode 100644 mozglue/baseprofiler/core/PlatformMacros.h create mode 100644 mozglue/baseprofiler/core/ProfileBuffer.cpp create mode 100644 mozglue/baseprofiler/core/ProfileBuffer.h create mode 100644 mozglue/baseprofiler/core/ProfileBufferEntry.cpp create mode 100644 mozglue/baseprofiler/core/ProfileBufferEntry.h create mode 100644 mozglue/baseprofiler/core/ProfileJSONWriter.cpp create mode 100644 mozglue/baseprofiler/core/ProfiledThreadData.cpp create mode 100644 mozglue/baseprofiler/core/ProfiledThreadData.h create mode 100644 mozglue/baseprofiler/core/ProfilerBacktrace.cpp create mode 100644 mozglue/baseprofiler/core/ProfilerBacktrace.h create mode 100644 mozglue/baseprofiler/core/ProfilerMarkers.cpp create mode 100644 mozglue/baseprofiler/core/ProfilerUtils.cpp create mode 100644 mozglue/baseprofiler/core/ProfilingCategory.cpp create mode 100644 mozglue/baseprofiler/core/ProfilingStack.cpp create mode 100644 mozglue/baseprofiler/core/RegisteredThread.cpp create mode 100644 mozglue/baseprofiler/core/RegisteredThread.h create mode 100644 mozglue/baseprofiler/core/ThreadInfo.h create mode 100644 mozglue/baseprofiler/core/VTuneProfiler.cpp create mode 100644 mozglue/baseprofiler/core/VTuneProfiler.h create mode 100644 mozglue/baseprofiler/core/platform-linux-android.cpp create mode 100644 mozglue/baseprofiler/core/platform-macos.cpp create mode 100644 mozglue/baseprofiler/core/platform-win32.cpp create mode 100644 mozglue/baseprofiler/core/platform.cpp create mode 100644 mozglue/baseprofiler/core/platform.h create mode 100644 mozglue/baseprofiler/core/shared-libraries-linux.cc create mode 100644 mozglue/baseprofiler/core/shared-libraries-macos.cc create mode 100644 mozglue/baseprofiler/core/shared-libraries-win32.cc create mode 100644 mozglue/baseprofiler/core/vtune/ittnotify.h (limited to 'mozglue/baseprofiler/core') diff --git a/mozglue/baseprofiler/core/BaseAndGeckoProfilerDetail.cpp b/mozglue/baseprofiler/core/BaseAndGeckoProfilerDetail.cpp new file mode 100644 index 0000000000..213f25cf16 --- /dev/null +++ b/mozglue/baseprofiler/core/BaseAndGeckoProfilerDetail.cpp @@ -0,0 +1,92 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "mozilla/BaseAndGeckoProfilerDetail.h" + +#include +#include + +namespace mozilla::profiler::detail { + +constexpr std::string_view scPidPrefix = "pid:"; + +// Convert a C string to a BaseProfilerProcessId. Return unspecified +// BaseProfilerProcessId if the string is not exactly a valid pid. +static baseprofiler::BaseProfilerProcessId StringToPid(const char* aString) { + if (!aString || aString[0] == '\0') { + // Null or empty. + return baseprofiler::BaseProfilerProcessId{}; + } + + if (aString[0] == '0') { + if (aString[1] != '\0') { + // Don't accept leading zeroes. + return baseprofiler::BaseProfilerProcessId{}; + } + return baseprofiler::BaseProfilerProcessId::FromNumber(0); + } + + using PidNumber = baseprofiler::BaseProfilerProcessId::NumberType; + PidNumber pid = 0; + for (;;) { + const char c = *aString; + if (c == '\0') { + break; + } + if (c < '0' || c > '9') { + // Only accept decimal digits. + return baseprofiler::BaseProfilerProcessId{}; + } + static_assert(!std::numeric_limits::is_signed, + "The following relies on unsigned arithmetic"); + PidNumber newPid = pid * 10u + PidNumber(c - '0'); + if (newPid < pid) { + // Unsigned overflow. + return baseprofiler::BaseProfilerProcessId{}; + } + pid = newPid; + ++aString; + } + return baseprofiler::BaseProfilerProcessId::FromNumber(pid); +} + +[[nodiscard]] MFBT_API bool FilterHasPid( + const char* aFilter, baseprofiler::BaseProfilerProcessId aPid) { + if (strncmp(aFilter, scPidPrefix.data(), scPidPrefix.length()) != 0) { + // The filter is not starting with "pid:". + return false; + } + + return StringToPid(aFilter + scPidPrefix.length()) == aPid; +} + +[[nodiscard]] MFBT_API bool FiltersExcludePid( + Span aFilters, + baseprofiler::BaseProfilerProcessId aPid) { + if (aFilters.empty()) { + return false; + } + + // First, check if the list only contains "pid:..." strings. + for (const char* const filter : aFilters) { + if (strncmp(filter, scPidPrefix.data(), scPidPrefix.length()) != 0) { + // At least one filter is *not* a "pid:...", our pid is not excluded. + return false; + } + } + + // Here, all filters start with "pid:". Check if the given pid is included. + for (const char* const filter : aFilters) { + if (StringToPid(filter + scPidPrefix.length()) == aPid) { + // Our pid is present, so it's not excluded. + return false; + } + } + // Our pid was not in a list of only pids, so it's excluded. + return true; +} + +} // namespace mozilla::profiler::detail diff --git a/mozglue/baseprofiler/core/EHABIStackWalk.cpp b/mozglue/baseprofiler/core/EHABIStackWalk.cpp new file mode 100644 index 0000000000..0c2c855c9b --- /dev/null +++ b/mozglue/baseprofiler/core/EHABIStackWalk.cpp @@ -0,0 +1,592 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +/* + * This is an implementation of stack unwinding according to a subset + * of the ARM Exception Handling ABI, as described in: + * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0038a/IHI0038A_ehabi.pdf + * + * This handles only the ARM-defined "personality routines" (chapter + * 9), and don't track the value of FP registers, because profiling + * needs only chain of PC/SP values. + * + * Because the exception handling info may not be accurate for all + * possible places where an async signal could occur (e.g., in a + * prologue or epilogue), this bounds-checks all stack accesses. + * + * This file uses "struct" for structures in the exception tables and + * "class" otherwise. We should avoid violating the C++11 + * standard-layout rules in the former. + */ + +#include "BaseProfiler.h" + +#include "EHABIStackWalk.h" + +#include "BaseProfilerSharedLibraries.h" +#include "platform.h" + +#include "mozilla/Atomics.h" +#include "mozilla/DebugOnly.h" +#include "mozilla/EndianUtils.h" + +#include +#include +#include +#include +#include + +#ifndef PT_ARM_EXIDX +# define PT_ARM_EXIDX 0x70000001 +#endif + +namespace mozilla { +namespace baseprofiler { + +struct PRel31 { + uint32_t mBits; + bool topBit() const { return mBits & 0x80000000; } + uint32_t value() const { return mBits & 0x7fffffff; } + int32_t offset() const { return (static_cast(mBits) << 1) >> 1; } + const void* compute() const { + return reinterpret_cast(this) + offset(); + } + + private: + PRel31(const PRel31& copied) = delete; + PRel31() = delete; +}; + +struct EHEntry { + PRel31 startPC; + PRel31 exidx; + + private: + EHEntry(const EHEntry& copied) = delete; + EHEntry() = delete; +}; + +class EHState { + // Note that any core register can be used as a "frame pointer" to + // influence the unwinding process, so this must track all of them. + uint32_t mRegs[16]; + + public: + bool unwind(const EHEntry* aEntry, const void* stackBase); + uint32_t& operator[](int i) { return mRegs[i]; } + const uint32_t& operator[](int i) const { return mRegs[i]; } + explicit EHState(const mcontext_t&); +}; + +enum { R_SP = 13, R_LR = 14, R_PC = 15 }; + +class EHTable { + uint32_t mStartPC; + uint32_t mEndPC; + uint32_t mBaseAddress; + const EHEntry* mEntriesBegin; + const EHEntry* mEntriesEnd; + std::string mName; + + public: + EHTable(const void* aELF, size_t aSize, const std::string& aName); + const EHEntry* lookup(uint32_t aPC) const; + bool isValid() const { return mEntriesEnd != mEntriesBegin; } + const std::string& name() const { return mName; } + uint32_t startPC() const { return mStartPC; } + uint32_t endPC() const { return mEndPC; } + uint32_t baseAddress() const { return mBaseAddress; } +}; + +class EHAddrSpace { + std::vector mStarts; + std::vector mTables; + static Atomic sCurrent; + + public: + explicit EHAddrSpace(const std::vector& aTables); + const EHTable* lookup(uint32_t aPC) const; + static void Update(); + static const EHAddrSpace* Get(); +}; + +void EHABIStackWalkInit() { EHAddrSpace::Update(); } + +size_t EHABIStackWalk(const mcontext_t& aContext, void* stackBase, void** aSPs, + void** aPCs, const size_t aNumFrames) { + const EHAddrSpace* space = EHAddrSpace::Get(); + EHState state(aContext); + size_t count = 0; + + while (count < aNumFrames) { + uint32_t pc = state[R_PC], sp = state[R_SP]; + aPCs[count] = reinterpret_cast(pc); + aSPs[count] = reinterpret_cast(sp); + count++; + + if (!space) break; + // TODO: cache these lookups. Binary-searching libxul is + // expensive (possibly more expensive than doing the actual + // unwind), and even a small cache should help. + const EHTable* table = space->lookup(pc); + if (!table) break; + const EHEntry* entry = table->lookup(pc); + if (!entry) break; + if (!state.unwind(entry, stackBase)) break; + } + + return count; +} + +class EHInterp { + public: + // Note that stackLimit is exclusive and stackBase is inclusive + // (i.e, stackLimit < SP <= stackBase), following the convention + // set by the AAPCS spec. + EHInterp(EHState& aState, const EHEntry* aEntry, uint32_t aStackLimit, + uint32_t aStackBase) + : mState(aState), + mStackLimit(aStackLimit), + mStackBase(aStackBase), + mNextWord(0), + mWordsLeft(0), + mFailed(false) { + const PRel31& exidx = aEntry->exidx; + uint32_t firstWord; + + if (exidx.mBits == 1) { // EXIDX_CANTUNWIND + mFailed = true; + return; + } + if (exidx.topBit()) { + firstWord = exidx.mBits; + } else { + mNextWord = reinterpret_cast(exidx.compute()); + firstWord = *mNextWord++; + } + + switch (firstWord >> 24) { + case 0x80: // short + mWord = firstWord << 8; + mBytesLeft = 3; + break; + case 0x81: + case 0x82: // long; catch descriptor size ignored + mWord = firstWord << 16; + mBytesLeft = 2; + mWordsLeft = (firstWord >> 16) & 0xff; + break; + default: + // unknown personality + mFailed = true; + } + } + + bool unwind(); + + private: + // TODO: GCC has been observed not CSEing repeated reads of + // mState[R_SP] with writes to mFailed between them, suggesting that + // it hasn't determined that they can't alias and is thus missing + // optimization opportunities. So, we may want to flatten EHState + // into this class; this may also make the code simpler. + EHState& mState; + uint32_t mStackLimit; + uint32_t mStackBase; + const uint32_t* mNextWord; + uint32_t mWord; + uint8_t mWordsLeft; + uint8_t mBytesLeft; + bool mFailed; + + enum { + I_ADDSP = 0x00, // 0sxxxxxx (subtract if s) + M_ADDSP = 0x80, + I_POPMASK = 0x80, // 1000iiii iiiiiiii (if any i set) + M_POPMASK = 0xf0, + I_MOVSP = 0x90, // 1001nnnn + M_MOVSP = 0xf0, + I_POPN = 0xa0, // 1010lnnn + M_POPN = 0xf0, + I_FINISH = 0xb0, // 10110000 + I_POPLO = 0xb1, // 10110001 0000iiii (if any i set) + I_ADDSPBIG = 0xb2, // 10110010 uleb128 + I_POPFDX = 0xb3, // 10110011 sssscccc + I_POPFDX8 = 0xb8, // 10111nnn + M_POPFDX8 = 0xf8, + // "Intel Wireless MMX" extensions omitted. + I_POPFDD = 0xc8, // 1100100h sssscccc + M_POPFDD = 0xfe, + I_POPFDD8 = 0xd0, // 11010nnn + M_POPFDD8 = 0xf8 + }; + + uint8_t next() { + if (mBytesLeft == 0) { + if (mWordsLeft == 0) { + return I_FINISH; + } + mWordsLeft--; + mWord = *mNextWord++; + mBytesLeft = 4; + } + mBytesLeft--; + mWord = (mWord << 8) | (mWord >> 24); // rotate + return mWord; + } + + uint32_t& vSP() { return mState[R_SP]; } + uint32_t* ptrSP() { return reinterpret_cast(vSP()); } + + void checkStackBase() { + if (vSP() > mStackBase) mFailed = true; + } + void checkStackLimit() { + if (vSP() <= mStackLimit) mFailed = true; + } + void checkStackAlign() { + if ((vSP() & 3) != 0) mFailed = true; + } + void checkStack() { + checkStackBase(); + checkStackLimit(); + checkStackAlign(); + } + + void popRange(uint8_t first, uint8_t last, uint16_t mask) { + bool hasSP = false; + uint32_t tmpSP; + if (mask == 0) mFailed = true; + for (uint8_t r = first; r <= last; ++r) { + if (mask & 1) { + if (r == R_SP) { + hasSP = true; + tmpSP = *ptrSP(); + } else + mState[r] = *ptrSP(); + vSP() += 4; + checkStackBase(); + if (mFailed) return; + } + mask >>= 1; + } + if (hasSP) { + vSP() = tmpSP; + checkStack(); + } + } +}; + +bool EHState::unwind(const EHEntry* aEntry, const void* stackBasePtr) { + // The unwinding program cannot set SP to less than the initial value. + uint32_t stackLimit = mRegs[R_SP] - 4; + uint32_t stackBase = reinterpret_cast(stackBasePtr); + EHInterp interp(*this, aEntry, stackLimit, stackBase); + return interp.unwind(); +} + +bool EHInterp::unwind() { + mState[R_PC] = 0; + checkStack(); + while (!mFailed) { + uint8_t insn = next(); +#if DEBUG_EHABI_UNWIND + LOG("unwind insn = %02x", (unsigned)insn); +#endif + // Try to put the common cases first. + + // 00xxxxxx: vsp = vsp + (xxxxxx << 2) + 4 + // 01xxxxxx: vsp = vsp - (xxxxxx << 2) - 4 + if ((insn & M_ADDSP) == I_ADDSP) { + uint32_t offset = ((insn & 0x3f) << 2) + 4; + if (insn & 0x40) { + vSP() -= offset; + checkStackLimit(); + } else { + vSP() += offset; + checkStackBase(); + } + continue; + } + + // 10100nnn: Pop r4-r[4+nnn] + // 10101nnn: Pop r4-r[4+nnn], r14 + if ((insn & M_POPN) == I_POPN) { + uint8_t n = (insn & 0x07) + 1; + bool lr = insn & 0x08; + uint32_t* ptr = ptrSP(); + vSP() += (n + (lr ? 1 : 0)) * 4; + checkStackBase(); + for (uint8_t r = 4; r < 4 + n; ++r) mState[r] = *ptr++; + if (lr) mState[R_LR] = *ptr++; + continue; + } + + // 1011000: Finish + if (insn == I_FINISH) { + if (mState[R_PC] == 0) { + mState[R_PC] = mState[R_LR]; + // Non-standard change (bug 916106): Prevent the caller from + // re-using LR. Since the caller is by definition not a leaf + // routine, it will have to restore LR from somewhere to + // return to its own caller, so we can safely zero it here. + // This makes a difference only if an error in unwinding + // (e.g., caused by starting from within a prologue/epilogue) + // causes us to load a pointer to a leaf routine as LR; if we + // don't do something, we'll go into an infinite loop of + // "returning" to that same function. + mState[R_LR] = 0; + } + return true; + } + + // 1001nnnn: Set vsp = r[nnnn] + if ((insn & M_MOVSP) == I_MOVSP) { + vSP() = mState[insn & 0x0f]; + checkStack(); + continue; + } + + // 11001000 sssscccc: Pop VFP regs D[16+ssss]-D[16+ssss+cccc] (as FLDMFDD) + // 11001001 sssscccc: Pop VFP regs D[ssss]-D[ssss+cccc] (as FLDMFDD) + if ((insn & M_POPFDD) == I_POPFDD) { + uint8_t n = (next() & 0x0f) + 1; + // Note: if the 16+ssss+cccc > 31, the encoding is reserved. + // As the space is currently unused, we don't try to check. + vSP() += 8 * n; + checkStackBase(); + continue; + } + + // 11010nnn: Pop VFP regs D[8]-D[8+nnn] (as FLDMFDD) + if ((insn & M_POPFDD8) == I_POPFDD8) { + uint8_t n = (insn & 0x07) + 1; + vSP() += 8 * n; + checkStackBase(); + continue; + } + + // 10110010 uleb128: vsp = vsp + 0x204 + (uleb128 << 2) + if (insn == I_ADDSPBIG) { + uint32_t acc = 0; + uint8_t shift = 0; + uint8_t byte; + do { + if (shift >= 32) return false; + byte = next(); + acc |= (byte & 0x7f) << shift; + shift += 7; + } while (byte & 0x80); + uint32_t offset = 0x204 + (acc << 2); + // The calculations above could have overflowed. + // But the one we care about is this: + if (vSP() + offset < vSP()) mFailed = true; + vSP() += offset; + // ...so that this is the only other check needed: + checkStackBase(); + continue; + } + + // 1000iiii iiiiiiii (i not all 0): Pop under masks {r15-r12}, {r11-r4} + if ((insn & M_POPMASK) == I_POPMASK) { + popRange(4, 15, ((insn & 0x0f) << 8) | next()); + continue; + } + + // 1011001 0000iiii (i not all 0): Pop under mask {r3-r0} + if (insn == I_POPLO) { + popRange(0, 3, next() & 0x0f); + continue; + } + + // 10110011 sssscccc: Pop VFP regs D[ssss]-D[ssss+cccc] (as FLDMFDX) + if (insn == I_POPFDX) { + uint8_t n = (next() & 0x0f) + 1; + vSP() += 8 * n + 4; + checkStackBase(); + continue; + } + + // 10111nnn: Pop VFP regs D[8]-D[8+nnn] (as FLDMFDX) + if ((insn & M_POPFDX8) == I_POPFDX8) { + uint8_t n = (insn & 0x07) + 1; + vSP() += 8 * n + 4; + checkStackBase(); + continue; + } + + // unhandled instruction +#ifdef DEBUG_EHABI_UNWIND + LOG("Unhandled EHABI instruction 0x%02x", insn); +#endif + mFailed = true; + } + return false; +} + +bool operator<(const EHTable& lhs, const EHTable& rhs) { + return lhs.startPC() < rhs.startPC(); +} + +// Async signal unsafe. +EHAddrSpace::EHAddrSpace(const std::vector& aTables) + : mTables(aTables) { + std::sort(mTables.begin(), mTables.end()); + DebugOnly lastEnd = 0; + for (std::vector::iterator i = mTables.begin(); i != mTables.end(); + ++i) { + MOZ_ASSERT(i->startPC() >= lastEnd); + mStarts.push_back(i->startPC()); + lastEnd = i->endPC(); + } +} + +const EHTable* EHAddrSpace::lookup(uint32_t aPC) const { + ptrdiff_t i = (std::upper_bound(mStarts.begin(), mStarts.end(), aPC) - + mStarts.begin()) - + 1; + + if (i < 0 || aPC >= mTables[i].endPC()) return 0; + return &mTables[i]; +} + +const EHEntry* EHTable::lookup(uint32_t aPC) const { + MOZ_ASSERT(aPC >= mStartPC); + if (aPC >= mEndPC) return nullptr; + + const EHEntry* begin = mEntriesBegin; + const EHEntry* end = mEntriesEnd; + MOZ_ASSERT(begin < end); + if (aPC < reinterpret_cast(begin->startPC.compute())) + return nullptr; + + while (end - begin > 1) { +#ifdef EHABI_UNWIND_MORE_ASSERTS + if ((end - 1)->startPC.compute() < begin->startPC.compute()) { + MOZ_CRASH("unsorted exidx"); + } +#endif + const EHEntry* mid = begin + (end - begin) / 2; + if (aPC < reinterpret_cast(mid->startPC.compute())) + end = mid; + else + begin = mid; + } + return begin; +} + +#if MOZ_LITTLE_ENDIAN() +static const unsigned char hostEndian = ELFDATA2LSB; +#elif MOZ_BIG_ENDIAN() +static const unsigned char hostEndian = ELFDATA2MSB; +#else +# error "No endian?" +#endif + +// Async signal unsafe: std::vector::reserve, std::string copy ctor. +EHTable::EHTable(const void* aELF, size_t aSize, const std::string& aName) + : mStartPC(~0), // largest uint32_t + mEndPC(0), + mEntriesBegin(nullptr), + mEntriesEnd(nullptr), + mName(aName) { + const uint32_t fileHeaderAddr = reinterpret_cast(aELF); + + if (aSize < sizeof(Elf32_Ehdr)) return; + + const Elf32_Ehdr& file = *(reinterpret_cast(fileHeaderAddr)); + if (memcmp(&file.e_ident[EI_MAG0], ELFMAG, SELFMAG) != 0 || + file.e_ident[EI_CLASS] != ELFCLASS32 || + file.e_ident[EI_DATA] != hostEndian || + file.e_ident[EI_VERSION] != EV_CURRENT || file.e_machine != EM_ARM || + file.e_version != EV_CURRENT) + // e_flags? + return; + + MOZ_ASSERT(file.e_phoff + file.e_phnum * file.e_phentsize <= aSize); + const Elf32_Phdr *exidxHdr = 0, *zeroHdr = 0; + for (unsigned i = 0; i < file.e_phnum; ++i) { + const Elf32_Phdr& phdr = *(reinterpret_cast( + fileHeaderAddr + file.e_phoff + i * file.e_phentsize)); + if (phdr.p_type == PT_ARM_EXIDX) { + exidxHdr = &phdr; + } else if (phdr.p_type == PT_LOAD) { + if (phdr.p_offset == 0) { + zeroHdr = &phdr; + } + if (phdr.p_flags & PF_X) { + mStartPC = std::min(mStartPC, phdr.p_vaddr); + mEndPC = std::max(mEndPC, phdr.p_vaddr + phdr.p_memsz); + } + } + } + if (!exidxHdr) return; + if (!zeroHdr) return; + mBaseAddress = fileHeaderAddr - zeroHdr->p_vaddr; + mStartPC += mBaseAddress; + mEndPC += mBaseAddress; + mEntriesBegin = + reinterpret_cast(mBaseAddress + exidxHdr->p_vaddr); + mEntriesEnd = reinterpret_cast( + mBaseAddress + exidxHdr->p_vaddr + exidxHdr->p_memsz); +} + +Atomic EHAddrSpace::sCurrent(nullptr); + +// Async signal safe; can fail if Update() hasn't returned yet. +const EHAddrSpace* EHAddrSpace::Get() { return sCurrent; } + +// Collect unwinding information from loaded objects. Calls after the +// first have no effect. Async signal unsafe. +void EHAddrSpace::Update() { + const EHAddrSpace* space = sCurrent; + if (space) return; + + SharedLibraryInfo info = SharedLibraryInfo::GetInfoForSelf(); + std::vector tables; + + for (size_t i = 0; i < info.GetSize(); ++i) { + const SharedLibrary& lib = info.GetEntry(i); + // FIXME: This isn't correct if the start address isn't p_offset 0, because + // the start address will not point at the file header. But this is worked + // around by magic number checks in the EHTable constructor. + EHTable tab(reinterpret_cast(lib.GetStart()), + lib.GetEnd() - lib.GetStart(), lib.GetDebugPath()); + if (tab.isValid()) tables.push_back(tab); + } + space = new EHAddrSpace(tables); + + if (!sCurrent.compareExchange(nullptr, space)) { + delete space; + space = sCurrent; + } +} + +EHState::EHState(const mcontext_t& context) { +#ifdef linux + mRegs[0] = context.arm_r0; + mRegs[1] = context.arm_r1; + mRegs[2] = context.arm_r2; + mRegs[3] = context.arm_r3; + mRegs[4] = context.arm_r4; + mRegs[5] = context.arm_r5; + mRegs[6] = context.arm_r6; + mRegs[7] = context.arm_r7; + mRegs[8] = context.arm_r8; + mRegs[9] = context.arm_r9; + mRegs[10] = context.arm_r10; + mRegs[11] = context.arm_fp; + mRegs[12] = context.arm_ip; + mRegs[13] = context.arm_sp; + mRegs[14] = context.arm_lr; + mRegs[15] = context.arm_pc; +#else +# error "Unhandled OS for ARM EHABI unwinding" +#endif +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/EHABIStackWalk.h b/mozglue/baseprofiler/core/EHABIStackWalk.h new file mode 100644 index 0000000000..d5f4edc0d7 --- /dev/null +++ b/mozglue/baseprofiler/core/EHABIStackWalk.h @@ -0,0 +1,30 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +/* + * This is an implementation of stack unwinding according to a subset + * of the ARM Exception Handling ABI; see the comment at the top of + * the .cpp file for details. + */ + +#ifndef mozilla_EHABIStackWalk_h__ +#define mozilla_EHABIStackWalk_h__ + +#include +#include + +namespace mozilla { +namespace baseprofiler { + +void EHABIStackWalkInit(); + +size_t EHABIStackWalk(const mcontext_t& aContext, void* stackBase, void** aSPs, + void** aPCs, size_t aNumFrames); + +} // namespace baseprofiler +} // namespace mozilla + +#endif diff --git a/mozglue/baseprofiler/core/PageInformation.cpp b/mozglue/baseprofiler/core/PageInformation.cpp new file mode 100644 index 0000000000..bb7c7872aa --- /dev/null +++ b/mozglue/baseprofiler/core/PageInformation.cpp @@ -0,0 +1,49 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "PageInformation.h" + +#include "BaseProfiler.h" + +#include "mozilla/BaseProfileJSONWriter.h" + +namespace mozilla { +namespace baseprofiler { + +PageInformation::PageInformation(uint64_t aTabID, uint64_t aInnerWindowID, + const std::string& aUrl, + uint64_t aEmbedderInnerWindowID) + : mTabID(aTabID), + mInnerWindowID(aInnerWindowID), + mUrl(aUrl), + mEmbedderInnerWindowID(aEmbedderInnerWindowID), + mRefCnt(0) {} + +bool PageInformation::Equals(PageInformation* aOtherPageInfo) const { + // It's enough to check inner window IDs because they are unique for each + // page. Therefore, we don't have to check the tab ID or url. + return InnerWindowID() == aOtherPageInfo->InnerWindowID(); +} + +void PageInformation::StreamJSON(SpliceableJSONWriter& aWriter) const { + aWriter.StartObjectElement(); + // Here, we are converting uint64_t to double. Both tab and Inner + // Window IDs are created using `nsContentUtils::GenerateProcessSpecificId`, + // which is specifically designed to only use 53 of the 64 bits to be lossless + // when passed into and out of JS as a double. + aWriter.DoubleProperty("tabID", TabID()); + aWriter.DoubleProperty("innerWindowID", InnerWindowID()); + aWriter.StringProperty("url", Url()); + aWriter.DoubleProperty("embedderInnerWindowID", EmbedderInnerWindowID()); + aWriter.EndObject(); +} + +size_t PageInformation::SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + return aMallocSizeOf(this); +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/PageInformation.h b/mozglue/baseprofiler/core/PageInformation.h new file mode 100644 index 0000000000..41808877b3 --- /dev/null +++ b/mozglue/baseprofiler/core/PageInformation.h @@ -0,0 +1,77 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef PageInformation_h +#define PageInformation_h + +#include "mozilla/Atomics.h" +#include "mozilla/Maybe.h" +#include "mozilla/MemoryReporting.h" + +#include + +namespace mozilla { +namespace baseprofiler { + +class SpliceableJSONWriter; + +// This class contains information that's relevant to a single page only +// while the page information is important and registered with the profiler, +// but regardless of whether the profiler is running. All accesses to it are +// protected by the profiler state lock. +// When the page gets unregistered, we keep the profiler buffer position +// to determine if we are still using this page. If not, we unregister +// it in the next page registration. +class PageInformation final { + public: + PageInformation(uint64_t aTabID, uint64_t aInnerWindowID, + const std::string& aUrl, uint64_t aEmbedderInnerWindowID); + + // Using hand-rolled ref-counting, because RefCounted.h macros don't produce + // the same code between mozglue and libxul, see bug 1536656. + MFBT_API void AddRef() const { ++mRefCnt; } + MFBT_API void Release() const { + MOZ_ASSERT(int32_t(mRefCnt) > 0); + if (--mRefCnt) { + delete this; + } + } + + size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const; + bool Equals(PageInformation* aOtherPageInfo) const; + void StreamJSON(SpliceableJSONWriter& aWriter) const; + + uint64_t InnerWindowID() const { return mInnerWindowID; } + uint64_t TabID() const { return mTabID; } + const std::string& Url() const { return mUrl; } + uint64_t EmbedderInnerWindowID() const { return mEmbedderInnerWindowID; } + + Maybe BufferPositionWhenUnregistered() const { + return mBufferPositionWhenUnregistered; + } + + void NotifyUnregistered(uint64_t aBufferPosition) { + mBufferPositionWhenUnregistered = Some(aBufferPosition); + } + + private: + const uint64_t mTabID; + const uint64_t mInnerWindowID; + const std::string mUrl; + const uint64_t mEmbedderInnerWindowID; + + // Holds the buffer position when page is unregistered. + // It's used to determine if we still use this page in the profiler or + // not. + Maybe mBufferPositionWhenUnregistered; + + mutable Atomic mRefCnt; +}; + +} // namespace baseprofiler +} // namespace mozilla + +#endif // PageInformation_h diff --git a/mozglue/baseprofiler/core/PlatformMacros.h b/mozglue/baseprofiler/core/PlatformMacros.h new file mode 100644 index 0000000000..c72e94c128 --- /dev/null +++ b/mozglue/baseprofiler/core/PlatformMacros.h @@ -0,0 +1,130 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef PLATFORM_MACROS_H +#define PLATFORM_MACROS_H + +// Define platform selection macros in a consistent way. Don't add anything +// else to this file, so it can remain freestanding. The primary factorisation +// is on (ARCH,OS) pairs ("PLATforms") but ARCH_ and OS_ macros are defined +// too, since they are sometimes convenient. +// +// Note: "GP" is short for "Gecko Profiler". + +#undef GP_PLAT_x86_android +#undef GP_PLAT_amd64_android +#undef GP_PLAT_arm_android +#undef GP_PLAT_arm64_android +#undef GP_PLAT_x86_linux +#undef GP_PLAT_amd64_linux +#undef GP_PLAT_arm_linux +#undef GP_PLAT_mips64_linux +#undef GP_PLAT_amd64_darwin +#undef GP_PLAT_arm64_darwin +#undef GP_PLAT_x86_windows +#undef GP_PLAT_amd64_windows +#undef GP_PLAT_arm64_windows + +#undef GP_ARCH_x86 +#undef GP_ARCH_amd64 +#undef GP_ARCH_arm +#undef GP_ARCH_arm64 +#undef GP_ARCH_mips64 + +#undef GP_OS_android +#undef GP_OS_linux +#undef GP_OS_darwin +#undef GP_OS_windows + +// We test __ANDROID__ before __linux__ because __linux__ is defined on both +// Android and Linux, whereas GP_OS_android is not defined on vanilla Linux. + +#if defined(__ANDROID__) && defined(__i386__) +# define GP_PLAT_x86_android 1 +# define GP_ARCH_x86 1 +# define GP_OS_android 1 + +#elif defined(__ANDROID__) && defined(__x86_64__) +# define GP_PLAT_amd64_android 1 +# define GP_ARCH_amd64 1 +# define GP_OS_android 1 + +#elif defined(__ANDROID__) && defined(__arm__) +# define GP_PLAT_arm_android 1 +# define GP_ARCH_arm 1 +# define GP_OS_android 1 + +#elif defined(__ANDROID__) && defined(__aarch64__) +# define GP_PLAT_arm64_android 1 +# define GP_ARCH_arm64 1 +# define GP_OS_android 1 + +#elif defined(__linux__) && defined(__i386__) +# define GP_PLAT_x86_linux 1 +# define GP_ARCH_x86 1 +# define GP_OS_linux 1 + +#elif defined(__linux__) && defined(__x86_64__) +# define GP_PLAT_amd64_linux 1 +# define GP_ARCH_amd64 1 +# define GP_OS_linux 1 + +#elif defined(__linux__) && defined(__arm__) +# define GP_PLAT_arm_linux 1 +# define GP_ARCH_arm 1 +# define GP_OS_linux 1 + +#elif defined(__linux__) && defined(__aarch64__) +# define GP_PLAT_arm64_linux 1 +# define GP_ARCH_arm64 1 +# define GP_OS_linux 1 + +#elif defined(__linux__) && defined(__mips64) +# define GP_PLAT_mips64_linux 1 +# define GP_ARCH_mips64 1 +# define GP_OS_linux 1 + +#elif defined(__APPLE__) && defined(__aarch64__) +# define GP_PLAT_arm64_darwin 1 +# define GP_ARCH_arm64 1 +# define GP_OS_darwin 1 + +#elif defined(__APPLE__) && defined(__x86_64__) +# define GP_PLAT_amd64_darwin 1 +# define GP_ARCH_amd64 1 +# define GP_OS_darwin 1 + +#elif defined(__FreeBSD__) && defined(__x86_64__) +# define GP_PLAT_amd64_freebsd 1 +# define GP_ARCH_amd64 1 +# define GP_OS_freebsd 1 + +#elif defined(__FreeBSD__) && defined(__aarch64__) +# define GP_PLAT_arm64_freebsd 1 +# define GP_ARCH_arm64 1 +# define GP_OS_freebsd 1 + +#elif (defined(_MSC_VER) || defined(__MINGW32__)) && \ + (defined(_M_IX86) || defined(__i386__)) +# define GP_PLAT_x86_windows 1 +# define GP_ARCH_x86 1 +# define GP_OS_windows 1 + +#elif (defined(_MSC_VER) || defined(__MINGW32__)) && \ + (defined(_M_X64) || defined(__x86_64__)) +# define GP_PLAT_amd64_windows 1 +# define GP_ARCH_amd64 1 +# define GP_OS_windows 1 + +#elif defined(_MSC_VER) && defined(_M_ARM64) +# define GP_PLAT_arm64_windows 1 +# define GP_ARCH_arm64 1 +# define GP_OS_windows 1 + +#else +# error "Unsupported platform" +#endif + +#endif /* ndef PLATFORM_MACROS_H */ diff --git a/mozglue/baseprofiler/core/ProfileBuffer.cpp b/mozglue/baseprofiler/core/ProfileBuffer.cpp new file mode 100644 index 0000000000..dd5504274b --- /dev/null +++ b/mozglue/baseprofiler/core/ProfileBuffer.cpp @@ -0,0 +1,217 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "ProfileBuffer.h" + +#include "mozilla/MathAlgorithms.h" + +#include "BaseProfiler.h" + +namespace mozilla { +namespace baseprofiler { + +ProfileBuffer::ProfileBuffer(ProfileChunkedBuffer& aBuffer) + : mEntries(aBuffer) { + // Assume the given buffer is in-session. + MOZ_ASSERT(mEntries.IsInSession()); +} + +/* static */ +ProfileBufferBlockIndex ProfileBuffer::AddEntry( + ProfileChunkedBuffer& aProfileChunkedBuffer, + const ProfileBufferEntry& aEntry) { + switch (aEntry.GetKind()) { +#define SWITCH_KIND(KIND, TYPE, SIZE) \ + case ProfileBufferEntry::Kind::KIND: { \ + return aProfileChunkedBuffer.PutFrom(&aEntry, 1 + (SIZE)); \ + break; \ + } + + FOR_EACH_PROFILE_BUFFER_ENTRY_KIND(SWITCH_KIND) + +#undef SWITCH_KIND + default: + MOZ_ASSERT(false, "Unhandled baseprofiler::ProfilerBuffer entry KIND"); + return ProfileBufferBlockIndex{}; + } +} + +// Called from signal, call only reentrant functions +uint64_t ProfileBuffer::AddEntry(const ProfileBufferEntry& aEntry) { + return AddEntry(mEntries, aEntry).ConvertToProfileBufferIndex(); +} + +/* static */ +ProfileBufferBlockIndex ProfileBuffer::AddThreadIdEntry( + ProfileChunkedBuffer& aProfileChunkedBuffer, + BaseProfilerThreadId aThreadId) { + return AddEntry(aProfileChunkedBuffer, + ProfileBufferEntry::ThreadId(aThreadId)); +} + +uint64_t ProfileBuffer::AddThreadIdEntry(BaseProfilerThreadId aThreadId) { + return AddThreadIdEntry(mEntries, aThreadId).ConvertToProfileBufferIndex(); +} + +void ProfileBuffer::CollectCodeLocation( + const char* aLabel, const char* aStr, uint32_t aFrameFlags, + uint64_t aInnerWindowID, const Maybe& aLineNumber, + const Maybe& aColumnNumber, + const Maybe& aCategoryPair) { + AddEntry(ProfileBufferEntry::Label(aLabel)); + AddEntry(ProfileBufferEntry::FrameFlags(uint64_t(aFrameFlags))); + + if (aStr) { + // Store the string using one or more DynamicStringFragment entries. + size_t strLen = strlen(aStr) + 1; // +1 for the null terminator + // If larger than the prescribed limit, we will cut the string and end it + // with an ellipsis. + const bool tooBig = strLen > kMaxFrameKeyLength; + if (tooBig) { + strLen = kMaxFrameKeyLength; + } + char chars[ProfileBufferEntry::kNumChars]; + for (size_t j = 0;; j += ProfileBufferEntry::kNumChars) { + // Store up to kNumChars characters in the entry. + size_t len = ProfileBufferEntry::kNumChars; + const bool last = j + len >= strLen; + if (last) { + // Only the last entry may be smaller than kNumChars. + len = strLen - j; + if (tooBig) { + // That last entry is part of a too-big string, replace the end + // characters with an ellipsis "...". + len = std::max(len, size_t(4)); + chars[len - 4] = '.'; + chars[len - 3] = '.'; + chars[len - 2] = '.'; + chars[len - 1] = '\0'; + // Make sure the memcpy will not overwrite our ellipsis! + len -= 4; + } + } + memcpy(chars, &aStr[j], len); + AddEntry(ProfileBufferEntry::DynamicStringFragment(chars)); + if (last) { + break; + } + } + } + + if (aInnerWindowID) { + AddEntry(ProfileBufferEntry::InnerWindowID(aInnerWindowID)); + } + + if (aLineNumber) { + AddEntry(ProfileBufferEntry::LineNumber(*aLineNumber)); + } + + if (aColumnNumber) { + AddEntry(ProfileBufferEntry::ColumnNumber(*aColumnNumber)); + } + + if (aCategoryPair.isSome()) { + AddEntry(ProfileBufferEntry::CategoryPair(int(*aCategoryPair))); + } +} + +size_t ProfileBuffer::SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { + // Measurement of the following members may be added later if DMD finds it + // is worthwhile: + // - memory pointed to by the elements within mEntries + return mEntries.SizeOfExcludingThis(aMallocSizeOf); +} + +size_t ProfileBuffer::SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); +} + +void ProfileBuffer::CollectOverheadStats(TimeDuration aSamplingTime, + TimeDuration aLocking, + TimeDuration aCleaning, + TimeDuration aCounters, + TimeDuration aThreads) { + double timeUs = aSamplingTime.ToMilliseconds() * 1000.0; + if (mFirstSamplingTimeUs == 0.0) { + mFirstSamplingTimeUs = timeUs; + } else { + // Note that we'll have 1 fewer interval than other numbers (because + // we need both ends of an interval to know its duration). The final + // difference should be insignificant over the expected many thousands + // of iterations. + mIntervalsUs.Count(timeUs - mLastSamplingTimeUs); + } + mLastSamplingTimeUs = timeUs; + // Time to take the lock before sampling. + double lockingUs = aLocking.ToMilliseconds() * 1000.0; + // Time to discard expired data. + double cleaningUs = aCleaning.ToMilliseconds() * 1000.0; + // Time to gather all counters. + double countersUs = aCounters.ToMilliseconds() * 1000.0; + // Time to sample all threads. + double threadsUs = aThreads.ToMilliseconds() * 1000.0; + + // Add to our gathered stats. + mOverheadsUs.Count(lockingUs + cleaningUs + countersUs + threadsUs); + mLockingsUs.Count(lockingUs); + mCleaningsUs.Count(cleaningUs); + mCountersUs.Count(countersUs); + mThreadsUs.Count(threadsUs); + + // Record details in buffer, if requested. + static const bool sRecordSamplingOverhead = []() { + const char* recordOverheads = getenv("MOZ_PROFILER_RECORD_OVERHEADS"); + return recordOverheads && recordOverheads[0] != '\0'; + }(); + if (sRecordSamplingOverhead) { + AddEntry(ProfileBufferEntry::ProfilerOverheadTime(timeUs)); + AddEntry(ProfileBufferEntry::ProfilerOverheadDuration(lockingUs)); + AddEntry(ProfileBufferEntry::ProfilerOverheadDuration(cleaningUs)); + AddEntry(ProfileBufferEntry::ProfilerOverheadDuration(countersUs)); + AddEntry(ProfileBufferEntry::ProfilerOverheadDuration(threadsUs)); + } +} + +ProfilerBufferInfo ProfileBuffer::GetProfilerBufferInfo() const { + return {BufferRangeStart(), + BufferRangeEnd(), + static_cast(*mEntries.BufferLength() / + 8), // 8 bytes per entry. + mIntervalsUs, + mOverheadsUs, + mLockingsUs, + mCleaningsUs, + mCountersUs, + mThreadsUs}; +} + +/* ProfileBufferCollector */ + +void ProfileBufferCollector::CollectNativeLeafAddr(void* aAddr) { + mBuf.AddEntry(ProfileBufferEntry::NativeLeafAddr(aAddr)); +} + +void ProfileBufferCollector::CollectProfilingStackFrame( + const ProfilingStackFrame& aFrame) { + // WARNING: this function runs within the profiler's "critical section". + + MOZ_ASSERT(aFrame.isLabelFrame() || + (aFrame.isJsFrame() && !aFrame.isOSRFrame())); + + const char* label = aFrame.label(); + const char* dynamicString = aFrame.dynamicString(); + Maybe line; + Maybe column; + + MOZ_ASSERT(aFrame.isLabelFrame()); + + mBuf.CollectCodeLocation(label, dynamicString, aFrame.flags(), + aFrame.realmID(), line, column, + Some(aFrame.categoryPair())); +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/ProfileBuffer.h b/mozglue/baseprofiler/core/ProfileBuffer.h new file mode 100644 index 0000000000..f77b429df8 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfileBuffer.h @@ -0,0 +1,211 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef MOZ_PROFILE_BUFFER_H +#define MOZ_PROFILE_BUFFER_H + +#include "ProfileBufferEntry.h" + +#include "BaseProfiler.h" +#include "mozilla/Maybe.h" +#include "mozilla/PowerOfTwo.h" +#include "mozilla/ProfileBufferChunkManagerSingle.h" +#include "mozilla/ProfileChunkedBuffer.h" + +namespace mozilla { +namespace baseprofiler { + +// Class storing most profiling data in a ProfileChunkedBuffer. +// +// This class is used as a queue of entries which, after construction, never +// allocates. This makes it safe to use in the profiler's "critical section". +class ProfileBuffer final { + public: + // ProfileBuffer constructor + // @param aBuffer The in-session ProfileChunkedBuffer to use as buffer + // manager. + explicit ProfileBuffer(ProfileChunkedBuffer& aBuffer); + + ProfileChunkedBuffer& UnderlyingChunkedBuffer() const { return mEntries; } + + bool IsThreadSafe() const { return mEntries.IsThreadSafe(); } + + // Add |aEntry| to the buffer, ignoring what kind of entry it is. + // Returns the position of the entry. + uint64_t AddEntry(const ProfileBufferEntry& aEntry); + + // Add to the buffer a sample start (ThreadId) entry for aThreadId. + // Returns the position of the entry. + uint64_t AddThreadIdEntry(BaseProfilerThreadId aThreadId); + + void CollectCodeLocation(const char* aLabel, const char* aStr, + uint32_t aFrameFlags, uint64_t aInnerWindowID, + const Maybe& aLineNumber, + const Maybe& aColumnNumber, + const Maybe& aCategoryPair); + + // Maximum size of a frameKey string that we'll handle. + static const size_t kMaxFrameKeyLength = 512; + + // Stream JSON for samples in the buffer to aWriter, using the supplied + // UniqueStacks object. + // Only streams samples for the given thread ID and which were taken at or + // after aSinceTime. If ID is 0, ignore the stored thread ID; this should only + // be used when the buffer contains only one sample. + // Return the thread ID of the streamed sample(s), or 0. + BaseProfilerThreadId StreamSamplesToJSON(SpliceableJSONWriter& aWriter, + BaseProfilerThreadId aThreadId, + double aSinceTime, + UniqueStacks& aUniqueStacks) const; + + void StreamMarkersToJSON(SpliceableJSONWriter& aWriter, + BaseProfilerThreadId aThreadId, + const TimeStamp& aProcessStartTime, + double aSinceTime, + UniqueStacks& aUniqueStacks) const; + void StreamPausedRangesToJSON(SpliceableJSONWriter& aWriter, + double aSinceTime) const; + void StreamProfilerOverheadToJSON(SpliceableJSONWriter& aWriter, + const TimeStamp& aProcessStartTime, + double aSinceTime) const; + void StreamCountersToJSON(SpliceableJSONWriter& aWriter, + const TimeStamp& aProcessStartTime, + double aSinceTime) const; + + // Find (via |aLastSample|) the most recent sample for the thread denoted by + // |aThreadId| and clone it, patching in the current time as appropriate. + // Mutate |aLastSample| to point to the newly inserted sample. + // Returns whether duplication was successful. + bool DuplicateLastSample(BaseProfilerThreadId aThreadId, + const TimeStamp& aProcessStartTime, + Maybe& aLastSample); + + void DiscardSamplesBeforeTime(double aTime); + + size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const; + size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const; + + void CollectOverheadStats(TimeDuration aSamplingTime, TimeDuration aLocking, + TimeDuration aCleaning, TimeDuration aCounters, + TimeDuration aThreads); + + ProfilerBufferInfo GetProfilerBufferInfo() const; + + private: + // Add |aEntry| to the provider ProfileChunkedBuffer. + // `static` because it may be used to add an entry to a `ProfileChunkedBuffer` + // that is not attached to a `ProfileBuffer`. + static ProfileBufferBlockIndex AddEntry( + ProfileChunkedBuffer& aProfileChunkedBuffer, + const ProfileBufferEntry& aEntry); + + // Add a sample start (ThreadId) entry for aThreadId to the provided + // ProfileChunkedBuffer. Returns the position of the entry. + // `static` because it may be used to add an entry to a `ProfileChunkedBuffer` + // that is not attached to a `ProfileBuffer`. + static ProfileBufferBlockIndex AddThreadIdEntry( + ProfileChunkedBuffer& aProfileChunkedBuffer, + BaseProfilerThreadId aThreadId); + + // The storage in which this ProfileBuffer stores its entries. + ProfileChunkedBuffer& mEntries; + + public: + // `BufferRangeStart()` and `BufferRangeEnd()` return `uint64_t` values + // corresponding to the first entry and past the last entry stored in + // `mEntries`. + // + // The returned values are not guaranteed to be stable, because other threads + // may also be accessing the buffer concurrently. But they will always + // increase, and can therefore give an indication of how far these values have + // *at least* reached. In particular: + // - Entries whose index is strictly less that `BufferRangeStart()` have been + // discarded by now, so any related data may also be safely discarded. + // - It is safe to try and read entries at any index strictly less than + // `BufferRangeEnd()` -- but note that these reads may fail by the time you + // request them, as old entries get overwritten by new ones. + uint64_t BufferRangeStart() const { return mEntries.GetState().mRangeStart; } + uint64_t BufferRangeEnd() const { return mEntries.GetState().mRangeEnd; } + + private: + // Single pre-allocated chunk (to avoid spurious mallocs), used when: + // - Duplicating sleeping stacks (hence scExpectedMaximumStackSize). + // - Adding JIT info. + // - Streaming stacks to JSON. + // Mutable because it's accessed from non-multithreaded const methods. + mutable Maybe mMaybeWorkerChunkManager; + ProfileBufferChunkManagerSingle& WorkerChunkManager() const { + if (mMaybeWorkerChunkManager.isNothing()) { + // Only actually allocate it on first use. (Some ProfileBuffers are + // temporary and don't actually need this.) + mMaybeWorkerChunkManager.emplace( + ProfileBufferChunk::SizeofChunkMetadata() + + ProfileBufferChunkManager::scExpectedMaximumStackSize); + } + return *mMaybeWorkerChunkManager; + } + + // Time from launch (us) when first sampling was recorded. + double mFirstSamplingTimeUs = 0.0; + // Time from launch (us) when last sampling was recorded. + double mLastSamplingTimeUs = 0.0; + // Sampling stats: Interval (us) between successive samplings. + ProfilerStats mIntervalsUs; + // Sampling stats: Total duration (us) of each sampling. (Split detail below.) + ProfilerStats mOverheadsUs; + // Sampling stats: Time (us) to acquire the lock before sampling. + ProfilerStats mLockingsUs; + // Sampling stats: Time (us) to discard expired data. + ProfilerStats mCleaningsUs; + // Sampling stats: Time (us) to collect counter data. + ProfilerStats mCountersUs; + // Sampling stats: Time (us) to sample thread stacks. + ProfilerStats mThreadsUs; +}; + +/** + * Helper type used to implement ProfilerStackCollector. This type is used as + * the collector for MergeStacks by ProfileBuffer. It holds a reference to the + * buffer, as well as additional feature flags which are needed to control the + * data collection strategy + */ +class ProfileBufferCollector final : public ProfilerStackCollector { + public: + ProfileBufferCollector(ProfileBuffer& aBuf, uint64_t aSamplePos, + uint64_t aBufferRangeStart) + : mBuf(aBuf), + mSamplePositionInBuffer(aSamplePos), + mBufferRangeStart(aBufferRangeStart) { + MOZ_ASSERT( + mSamplePositionInBuffer >= mBufferRangeStart, + "The sample position should always be after the buffer range start"); + } + + // Position at which the sample starts in the profiler buffer (which may be + // different from the buffer in which the sample data is collected here). + Maybe SamplePositionInBuffer() override { + return Some(mSamplePositionInBuffer); + } + + // Profiler buffer's range start (which may be different from the buffer in + // which the sample data is collected here). + Maybe BufferRangeStart() override { + return Some(mBufferRangeStart); + } + + virtual void CollectNativeLeafAddr(void* aAddr) override; + virtual void CollectProfilingStackFrame( + const ProfilingStackFrame& aFrame) override; + + private: + ProfileBuffer& mBuf; + uint64_t mSamplePositionInBuffer; + uint64_t mBufferRangeStart; +}; + +} // namespace baseprofiler +} // namespace mozilla + +#endif diff --git a/mozglue/baseprofiler/core/ProfileBufferEntry.cpp b/mozglue/baseprofiler/core/ProfileBufferEntry.cpp new file mode 100644 index 0000000000..faa3826171 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfileBufferEntry.cpp @@ -0,0 +1,1367 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "ProfileBufferEntry.h" + +#include +#include + +#include "mozilla/Logging.h" +#include "mozilla/ScopeExit.h" +#include "mozilla/Sprintf.h" +#include "mozilla/StackWalk.h" + +#include "BaseProfiler.h" +#include "mozilla/BaseProfilerMarkers.h" +#include "platform.h" +#include "ProfileBuffer.h" +#include "ProfilerBacktrace.h" + +namespace mozilla { +namespace baseprofiler { + +//////////////////////////////////////////////////////////////////////// +// BEGIN ProfileBufferEntry + +ProfileBufferEntry::ProfileBufferEntry() + : mKind(Kind::INVALID), mStorage{0, 0, 0, 0, 0, 0, 0, 0} {} + +// aString must be a static string. +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, const char* aString) + : mKind(aKind) { + memcpy(mStorage, &aString, sizeof(aString)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, char aChars[kNumChars]) + : mKind(aKind) { + memcpy(mStorage, aChars, kNumChars); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, void* aPtr) : mKind(aKind) { + memcpy(mStorage, &aPtr, sizeof(aPtr)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, double aDouble) + : mKind(aKind) { + memcpy(mStorage, &aDouble, sizeof(aDouble)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, int aInt) : mKind(aKind) { + memcpy(mStorage, &aInt, sizeof(aInt)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, int64_t aInt64) + : mKind(aKind) { + memcpy(mStorage, &aInt64, sizeof(aInt64)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, uint64_t aUint64) + : mKind(aKind) { + memcpy(mStorage, &aUint64, sizeof(aUint64)); +} + +ProfileBufferEntry::ProfileBufferEntry(Kind aKind, + BaseProfilerThreadId aThreadId) + : mKind(aKind) { + static_assert(std::is_trivially_copyable_v); + static_assert(sizeof(aThreadId) <= sizeof(mStorage)); + memcpy(mStorage, &aThreadId, sizeof(aThreadId)); +} + +const char* ProfileBufferEntry::GetString() const { + const char* result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +void* ProfileBufferEntry::GetPtr() const { + void* result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +double ProfileBufferEntry::GetDouble() const { + double result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +int ProfileBufferEntry::GetInt() const { + int result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +int64_t ProfileBufferEntry::GetInt64() const { + int64_t result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +uint64_t ProfileBufferEntry::GetUint64() const { + uint64_t result; + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +BaseProfilerThreadId ProfileBufferEntry::GetThreadId() const { + BaseProfilerThreadId result; + static_assert(std::is_trivially_copyable_v); + memcpy(&result, mStorage, sizeof(result)); + return result; +} + +void ProfileBufferEntry::CopyCharsInto(char (&aOutArray)[kNumChars]) const { + memcpy(aOutArray, mStorage, kNumChars); +} + +// END ProfileBufferEntry +//////////////////////////////////////////////////////////////////////// + +// As mentioned in ProfileBufferEntry.h, the JSON format contains many +// arrays whose elements are laid out according to various schemas to help +// de-duplication. This RAII class helps write these arrays by keeping track of +// the last non-null element written and adding the appropriate number of null +// elements when writing new non-null elements. It also automatically opens and +// closes an array element on the given JSON writer. +// +// You grant the AutoArraySchemaWriter exclusive access to the JSONWriter and +// the UniqueJSONStrings objects for the lifetime of AutoArraySchemaWriter. Do +// not access them independently while the AutoArraySchemaWriter is alive. +// If you need to add complex objects, call FreeFormElement(), which will give +// you temporary access to the writer. +// +// Example usage: +// +// // Define the schema of elements in this type of array: [FOO, BAR, BAZ] +// enum Schema : uint32_t { +// FOO = 0, +// BAR = 1, +// BAZ = 2 +// }; +// +// AutoArraySchemaWriter writer(someJsonWriter, someUniqueStrings); +// if (shouldWriteFoo) { +// writer.IntElement(FOO, getFoo()); +// } +// ... etc ... +// +// The elements need to be added in-order. +class MOZ_RAII AutoArraySchemaWriter { + public: + explicit AutoArraySchemaWriter(SpliceableJSONWriter& aWriter) + : mJSONWriter(aWriter), mNextFreeIndex(0) { + mJSONWriter.StartArrayElement(); + } + + ~AutoArraySchemaWriter() { mJSONWriter.EndArray(); } + + template + void IntElement(uint32_t aIndex, T aValue) { + static_assert(!std::is_same_v, + "Narrowing uint64 -> int64 conversion not allowed"); + FillUpTo(aIndex); + mJSONWriter.IntElement(static_cast(aValue)); + } + + void DoubleElement(uint32_t aIndex, double aValue) { + FillUpTo(aIndex); + mJSONWriter.DoubleElement(aValue); + } + + void TimeMsElement(uint32_t aIndex, double aTime_ms) { + FillUpTo(aIndex); + mJSONWriter.TimeDoubleMsElement(aTime_ms); + } + + void BoolElement(uint32_t aIndex, bool aValue) { + FillUpTo(aIndex); + mJSONWriter.BoolElement(aValue); + } + + protected: + SpliceableJSONWriter& Writer() { return mJSONWriter; } + + void FillUpTo(uint32_t aIndex) { + MOZ_ASSERT(aIndex >= mNextFreeIndex); + mJSONWriter.NullElements(aIndex - mNextFreeIndex); + mNextFreeIndex = aIndex + 1; + } + + private: + SpliceableJSONWriter& mJSONWriter; + uint32_t mNextFreeIndex; +}; + +// Same as AutoArraySchemaWriter, but this can also write strings (output as +// indexes into the table of unique strings). +class MOZ_RAII AutoArraySchemaWithStringsWriter : public AutoArraySchemaWriter { + public: + AutoArraySchemaWithStringsWriter(SpliceableJSONWriter& aWriter, + UniqueJSONStrings& aStrings) + : AutoArraySchemaWriter(aWriter), mStrings(aStrings) {} + + void StringElement(uint32_t aIndex, const Span& aValue) { + FillUpTo(aIndex); + mStrings.WriteElement(Writer(), aValue); + } + + private: + UniqueJSONStrings& mStrings; +}; + +UniqueStacks::StackKey UniqueStacks::BeginStack(const FrameKey& aFrame) { + return StackKey(GetOrAddFrameIndex(aFrame)); +} + +UniqueStacks::StackKey UniqueStacks::AppendFrame(const StackKey& aStack, + const FrameKey& aFrame) { + return StackKey(aStack, GetOrAddStackIndex(aStack), + GetOrAddFrameIndex(aFrame)); +} + +bool UniqueStacks::FrameKey::NormalFrameData::operator==( + const NormalFrameData& aOther) const { + return mLocation == aOther.mLocation && + mRelevantForJS == aOther.mRelevantForJS && + mInnerWindowID == aOther.mInnerWindowID && mLine == aOther.mLine && + mColumn == aOther.mColumn && mCategoryPair == aOther.mCategoryPair; +} + +UniqueStacks::UniqueStacks() + : mUniqueStrings(MakeUnique( + FailureLatchInfallibleSource::Singleton())), + mFrameTableWriter(FailureLatchInfallibleSource::Singleton()), + mStackTableWriter(FailureLatchInfallibleSource::Singleton()) { + mFrameTableWriter.StartBareList(); + mStackTableWriter.StartBareList(); +} + +uint32_t UniqueStacks::GetOrAddStackIndex(const StackKey& aStack) { + uint32_t count = mStackToIndexMap.count(); + auto entry = mStackToIndexMap.lookupForAdd(aStack); + if (entry) { + MOZ_ASSERT(entry->value() < count); + return entry->value(); + } + + MOZ_RELEASE_ASSERT(mStackToIndexMap.add(entry, aStack, count)); + StreamStack(aStack); + return count; +} + +uint32_t UniqueStacks::GetOrAddFrameIndex(const FrameKey& aFrame) { + uint32_t count = mFrameToIndexMap.count(); + auto entry = mFrameToIndexMap.lookupForAdd(aFrame); + if (entry) { + MOZ_ASSERT(entry->value() < count); + return entry->value(); + } + + MOZ_RELEASE_ASSERT(mFrameToIndexMap.add(entry, aFrame, count)); + StreamNonJITFrame(aFrame); + return count; +} + +void UniqueStacks::SpliceFrameTableElements(SpliceableJSONWriter& aWriter) { + mFrameTableWriter.EndBareList(); + aWriter.TakeAndSplice(mFrameTableWriter.TakeChunkedWriteFunc()); +} + +void UniqueStacks::SpliceStackTableElements(SpliceableJSONWriter& aWriter) { + mStackTableWriter.EndBareList(); + aWriter.TakeAndSplice(mStackTableWriter.TakeChunkedWriteFunc()); +} + +void UniqueStacks::StreamStack(const StackKey& aStack) { + enum Schema : uint32_t { PREFIX = 0, FRAME = 1 }; + + AutoArraySchemaWriter writer(mStackTableWriter); + if (aStack.mPrefixStackIndex.isSome()) { + writer.IntElement(PREFIX, *aStack.mPrefixStackIndex); + } + writer.IntElement(FRAME, aStack.mFrameIndex); +} + +void UniqueStacks::StreamNonJITFrame(const FrameKey& aFrame) { + using NormalFrameData = FrameKey::NormalFrameData; + + enum Schema : uint32_t { + LOCATION = 0, + RELEVANT_FOR_JS = 1, + INNER_WINDOW_ID = 2, + IMPLEMENTATION = 3, + LINE = 4, + COLUMN = 5, + CATEGORY = 6, + SUBCATEGORY = 7 + }; + + AutoArraySchemaWithStringsWriter writer(mFrameTableWriter, *mUniqueStrings); + + const NormalFrameData& data = aFrame.mData.as(); + writer.StringElement(LOCATION, data.mLocation); + writer.BoolElement(RELEVANT_FOR_JS, data.mRelevantForJS); + + // It's okay to convert uint64_t to double here because DOM always creates IDs + // that are convertible to double. + writer.DoubleElement(INNER_WINDOW_ID, data.mInnerWindowID); + + if (data.mLine.isSome()) { + writer.IntElement(LINE, *data.mLine); + } + if (data.mColumn.isSome()) { + writer.IntElement(COLUMN, *data.mColumn); + } + if (data.mCategoryPair.isSome()) { + const ProfilingCategoryPairInfo& info = + GetProfilingCategoryPairInfo(*data.mCategoryPair); + writer.IntElement(CATEGORY, uint32_t(info.mCategory)); + writer.IntElement(SUBCATEGORY, info.mSubcategoryIndex); + } +} + +struct ProfileSample { + uint32_t mStack; + double mTime; + Maybe mResponsiveness; +}; + +static void WriteSample(SpliceableJSONWriter& aWriter, + const ProfileSample& aSample) { + enum Schema : uint32_t { + STACK = 0, + TIME = 1, + EVENT_DELAY = 2, + }; + + AutoArraySchemaWriter writer(aWriter); + + writer.IntElement(STACK, aSample.mStack); + + writer.TimeMsElement(TIME, aSample.mTime); + + if (aSample.mResponsiveness.isSome()) { + writer.DoubleElement(EVENT_DELAY, *aSample.mResponsiveness); + } +} + +class EntryGetter { + public: + explicit EntryGetter(ProfileChunkedBuffer::Reader& aReader, + uint64_t aInitialReadPos = 0) + : mBlockIt( + aReader.At(ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + aInitialReadPos))), + mBlockItEnd(aReader.end()) { + if (!ReadLegacyOrEnd()) { + // Find and read the next non-legacy entry. + Next(); + } + } + + bool Has() const { return mBlockIt != mBlockItEnd; } + + const ProfileBufferEntry& Get() const { + MOZ_ASSERT(Has(), "Caller should have checked `Has()` before `Get()`"); + return mEntry; + } + + void Next() { + MOZ_ASSERT(Has(), "Caller should have checked `Has()` before `Next()`"); + for (;;) { + ++mBlockIt; + if (ReadLegacyOrEnd()) { + // Either we're at the end, or we could read a legacy entry -> Done. + break; + } + // Otherwise loop around until we hit the end or a legacy entry. + } + } + + ProfileBufferBlockIndex CurBlockIndex() const { + return mBlockIt.CurrentBlockIndex(); + } + + uint64_t CurPos() const { + return CurBlockIndex().ConvertToProfileBufferIndex(); + } + + private: + // Try to read the entry at the current `mBlockIt` position. + // * If we're at the end of the buffer, just return `true`. + // * If there is a "legacy" entry (containing a real `ProfileBufferEntry`), + // read it into `mEntry`, and return `true` as well. + // * Otherwise the entry contains a "modern" type that cannot be read into + // `mEntry`, return `false` (so `EntryGetter` can skip to another entry). + bool ReadLegacyOrEnd() { + if (!Has()) { + return true; + } + // Read the entry "kind", which is always at the start of all entries. + ProfileBufferEntryReader aER = *mBlockIt; + auto type = static_cast( + aER.ReadObject()); + MOZ_ASSERT(static_cast(type) < + static_cast( + ProfileBufferEntry::Kind::MODERN_LIMIT)); + if (type >= ProfileBufferEntry::Kind::LEGACY_LIMIT) { + aER.SetRemainingBytes(0); + return false; + } + // Here, we have a legacy item, we need to read it from the start. + // Because the above `ReadObject` moved the reader, we ned to reset it to + // the start of the entry before reading the whole entry. + aER = *mBlockIt; + aER.ReadBytes(&mEntry, aER.RemainingBytes()); + return true; + } + + ProfileBufferEntry mEntry; + ProfileChunkedBuffer::BlockIterator mBlockIt; + const ProfileChunkedBuffer::BlockIterator mBlockItEnd; +}; + +// The following grammar shows legal sequences of profile buffer entries. +// The sequences beginning with a ThreadId entry are known as "samples". +// +// ( +// ( /* Samples */ +// ThreadId +// Time +// ( NativeLeafAddr +// | Label FrameFlags? DynamicStringFragment* LineNumber? CategoryPair? +// | JitReturnAddr +// )+ +// Responsiveness? +// ) +// | MarkerData +// | ( /* Counters */ +// CounterId +// Time +// ( +// CounterKey +// Count +// Number? +// )* +// ) +// | CollectionStart +// | CollectionEnd +// | Pause +// | Resume +// | ( ProfilerOverheadTime /* Sampling start timestamp */ +// ProfilerOverheadDuration /* Lock acquisition */ +// ProfilerOverheadDuration /* Expired data cleaning */ +// ProfilerOverheadDuration /* Counters */ +// ProfilerOverheadDuration /* Threads */ +// ) +// )* +// +// The most complicated part is the stack entry sequence that begins with +// Label. Here are some examples. +// +// - ProfilingStack frames without a dynamic string: +// +// Label("js::RunScript") +// CategoryPair(ProfilingCategoryPair::JS) +// +// Label("XREMain::XRE_main") +// LineNumber(4660) +// CategoryPair(ProfilingCategoryPair::OTHER) +// +// Label("ElementRestyler::ComputeStyleChangeFor") +// LineNumber(3003) +// CategoryPair(ProfilingCategoryPair::CSS) +// +// - ProfilingStack frames with a dynamic string: +// +// Label("nsObserverService::NotifyObservers") +// FrameFlags(uint64_t(ProfilingStackFrame::Flags::IS_LABEL_FRAME)) +// DynamicStringFragment("domwindo") +// DynamicStringFragment("wopened") +// LineNumber(291) +// CategoryPair(ProfilingCategoryPair::OTHER) +// +// Label("") +// FrameFlags(uint64_t(ProfilingStackFrame::Flags::IS_JS_FRAME)) +// DynamicStringFragment("closeWin") +// DynamicStringFragment("dow (chr") +// DynamicStringFragment("ome://gl") +// DynamicStringFragment("obal/con") +// DynamicStringFragment("tent/glo") +// DynamicStringFragment("balOverl") +// DynamicStringFragment("ay.js:5)") +// DynamicStringFragment("") # this string holds the closing '\0' +// LineNumber(25) +// CategoryPair(ProfilingCategoryPair::JS) +// +// Label("") +// FrameFlags(uint64_t(ProfilingStackFrame::Flags::IS_JS_FRAME)) +// DynamicStringFragment("bound (s") +// DynamicStringFragment("elf-host") +// DynamicStringFragment("ed:914)") +// LineNumber(945) +// CategoryPair(ProfilingCategoryPair::JS) +// +// - A profiling stack frame with an overly long dynamic string: +// +// Label("") +// FrameFlags(uint64_t(ProfilingStackFrame::Flags::IS_LABEL_FRAME)) +// DynamicStringFragment("(too lon") +// DynamicStringFragment("g)") +// LineNumber(100) +// CategoryPair(ProfilingCategoryPair::NETWORK) +// +// - A wasm JIT frame: +// +// Label("") +// FrameFlags(uint64_t(0)) +// DynamicStringFragment("wasm-fun") +// DynamicStringFragment("ction[87") +// DynamicStringFragment("36] (blo") +// DynamicStringFragment("b:http:/") +// DynamicStringFragment("/webasse") +// DynamicStringFragment("mbly.org") +// DynamicStringFragment("/3dc5759") +// DynamicStringFragment("4-ce58-4") +// DynamicStringFragment("626-975b") +// DynamicStringFragment("-08ad116") +// DynamicStringFragment("30bc1:38") +// DynamicStringFragment("29856)") +// +// - A JS frame in a synchronous sample: +// +// Label("") +// FrameFlags(uint64_t(ProfilingStackFrame::Flags::IS_LABEL_FRAME)) +// DynamicStringFragment("u (https") +// DynamicStringFragment("://perf-") +// DynamicStringFragment("html.io/") +// DynamicStringFragment("ac0da204") +// DynamicStringFragment("aaa44d75") +// DynamicStringFragment("a800.bun") +// DynamicStringFragment("dle.js:2") +// DynamicStringFragment("5)") + +// Because this is a format entirely internal to the Profiler, any parsing +// error indicates a bug in the ProfileBuffer writing or the parser itself, +// or possibly flaky hardware. +#define ERROR_AND_CONTINUE(msg) \ + { \ + fprintf(stderr, "ProfileBuffer parse error: %s", msg); \ + MOZ_ASSERT(false, msg); \ + continue; \ + } + +BaseProfilerThreadId ProfileBuffer::StreamSamplesToJSON( + SpliceableJSONWriter& aWriter, BaseProfilerThreadId aThreadId, + double aSinceTime, UniqueStacks& aUniqueStacks) const { + UniquePtr dynStrBuf = MakeUnique(kMaxFrameKeyLength); + + return mEntries.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, + "ProfileChunkedBuffer cannot be out-of-session when sampler is " + "running"); + + BaseProfilerThreadId processedThreadId; + + EntryGetter e(*aReader); + + for (;;) { + // This block skips entries until we find the start of the next sample. + // This is useful in three situations. + // + // - The circular buffer overwrites old entries, so when we start parsing + // we might be in the middle of a sample, and we must skip forward to + // the start of the next sample. + // + // - We skip samples that don't have an appropriate ThreadId or Time. + // + // - We skip range Pause, Resume, CollectionStart, Counter and + // CollectionEnd entries between samples. + while (e.Has()) { + if (e.Get().IsThreadId()) { + break; + } + e.Next(); + } + + if (!e.Has()) { + break; + } + + // Due to the skip_to_next_sample block above, if we have an entry here it + // must be a ThreadId entry. + MOZ_ASSERT(e.Get().IsThreadId()); + + BaseProfilerThreadId threadId = e.Get().GetThreadId(); + e.Next(); + + // Ignore samples that are for the wrong thread. + if (threadId != aThreadId && aThreadId.IsSpecified()) { + continue; + } + + MOZ_ASSERT( + aThreadId.IsSpecified() || !processedThreadId.IsSpecified(), + "Unspecified aThreadId should only be used with 1-sample buffer"); + + ProfileSample sample; + + if (e.Has() && e.Get().IsTime()) { + sample.mTime = e.Get().GetDouble(); + e.Next(); + + // Ignore samples that are too old. + if (sample.mTime < aSinceTime) { + continue; + } + } else { + ERROR_AND_CONTINUE("expected a Time entry"); + } + + UniqueStacks::StackKey stack = + aUniqueStacks.BeginStack(UniqueStacks::FrameKey("(root)")); + + int numFrames = 0; + while (e.Has()) { + if (e.Get().IsNativeLeafAddr()) { + numFrames++; + + void* pc = e.Get().GetPtr(); + e.Next(); + + static const uint32_t BUF_SIZE = 256; + char buf[BUF_SIZE]; + + // Bug 753041: We need a double cast here to tell GCC that we don't + // want to sign extend 32-bit addresses starting with 0xFXXXXXX. + unsigned long long pcULL = (unsigned long long)(uintptr_t)pc; + SprintfLiteral(buf, "0x%llx", pcULL); + + // If the "MOZ_PROFILER_SYMBOLICATE" env-var is set, we add a local + // symbolication description to the PC address. This is off by + // default, and mainly intended for local development. + static const bool preSymbolicate = []() { + const char* symbolicate = getenv("MOZ_PROFILER_SYMBOLICATE"); + return symbolicate && symbolicate[0] != '\0'; + }(); + if (preSymbolicate) { + MozCodeAddressDetails details; + if (MozDescribeCodeAddress(pc, &details)) { + // Replace \0 terminator with space. + const uint32_t pcLen = strlen(buf); + buf[pcLen] = ' '; + // Add description after space. Note: Using a frame number of 0, + // as using `numFrames` wouldn't help here, and would prevent + // combining same function calls that happen at different depths. + // TODO: Remove unsightly "#00: " if too annoying. :-) + MozFormatCodeAddressDetails( + buf + pcLen + 1, BUF_SIZE - (pcLen + 1), 0, pc, &details); + } + } + + stack = aUniqueStacks.AppendFrame(stack, UniqueStacks::FrameKey(buf)); + + } else if (e.Get().IsLabel()) { + numFrames++; + + const char* label = e.Get().GetString(); + e.Next(); + + using FrameFlags = ProfilingStackFrame::Flags; + uint32_t frameFlags = 0; + if (e.Has() && e.Get().IsFrameFlags()) { + frameFlags = uint32_t(e.Get().GetUint64()); + e.Next(); + } + + bool relevantForJS = + frameFlags & uint32_t(FrameFlags::RELEVANT_FOR_JS); + + // Copy potential dynamic string fragments into dynStrBuf, so that + // dynStrBuf will then contain the entire dynamic string. + size_t i = 0; + dynStrBuf[0] = '\0'; + while (e.Has()) { + if (e.Get().IsDynamicStringFragment()) { + char chars[ProfileBufferEntry::kNumChars]; + e.Get().CopyCharsInto(chars); + for (char c : chars) { + if (i < kMaxFrameKeyLength) { + dynStrBuf[i] = c; + i++; + } + } + e.Next(); + } else { + break; + } + } + dynStrBuf[kMaxFrameKeyLength - 1] = '\0'; + bool hasDynamicString = (i != 0); + + std::string frameLabel; + if (label[0] != '\0' && hasDynamicString) { + if (frameFlags & uint32_t(FrameFlags::STRING_TEMPLATE_METHOD)) { + frameLabel += label; + frameLabel += '.'; + frameLabel += dynStrBuf.get(); + } else if (frameFlags & + uint32_t(FrameFlags::STRING_TEMPLATE_GETTER)) { + frameLabel += "get "; + frameLabel += label; + frameLabel += '.'; + frameLabel += dynStrBuf.get(); + } else if (frameFlags & + uint32_t(FrameFlags::STRING_TEMPLATE_SETTER)) { + frameLabel += "set "; + frameLabel += label; + frameLabel += '.'; + frameLabel += dynStrBuf.get(); + } else { + frameLabel += label; + frameLabel += ' '; + frameLabel += dynStrBuf.get(); + } + } else if (hasDynamicString) { + frameLabel += dynStrBuf.get(); + } else { + frameLabel += label; + } + + uint64_t innerWindowID = 0; + if (e.Has() && e.Get().IsInnerWindowID()) { + innerWindowID = uint64_t(e.Get().GetUint64()); + e.Next(); + } + + Maybe line; + if (e.Has() && e.Get().IsLineNumber()) { + line = Some(unsigned(e.Get().GetInt())); + e.Next(); + } + + Maybe column; + if (e.Has() && e.Get().IsColumnNumber()) { + column = Some(unsigned(e.Get().GetInt())); + e.Next(); + } + + Maybe categoryPair; + if (e.Has() && e.Get().IsCategoryPair()) { + categoryPair = + Some(ProfilingCategoryPair(uint32_t(e.Get().GetInt()))); + e.Next(); + } + + stack = aUniqueStacks.AppendFrame( + stack, UniqueStacks::FrameKey(std::move(frameLabel), + relevantForJS, innerWindowID, line, + column, categoryPair)); + + } else { + break; + } + } + + if (numFrames == 0) { + // It is possible to have empty stacks if native stackwalking is + // disabled. Skip samples with empty stacks. (See Bug 1497985). + // Thus, don't use ERROR_AND_CONTINUE, but just continue. + continue; + } + + sample.mStack = aUniqueStacks.GetOrAddStackIndex(stack); + + if (e.Has() && e.Get().IsResponsiveness()) { + sample.mResponsiveness = Some(e.Get().GetDouble()); + e.Next(); + } + + WriteSample(aWriter, sample); + + processedThreadId = threadId; + } + + return processedThreadId; + }); +} + +void ProfileBuffer::StreamMarkersToJSON(SpliceableJSONWriter& aWriter, + BaseProfilerThreadId aThreadId, + const TimeStamp& aProcessStartTime, + double aSinceTime, + UniqueStacks& aUniqueStacks) const { + mEntries.ReadEach([&](ProfileBufferEntryReader& aER) { + auto type = static_cast( + aER.ReadObject()); + MOZ_ASSERT(static_cast(type) < + static_cast( + ProfileBufferEntry::Kind::MODERN_LIMIT)); + if (type == ProfileBufferEntry::Kind::Marker) { + ::mozilla::base_profiler_markers_detail::DeserializeAfterKindAndStream( + aER, + [&](const BaseProfilerThreadId& aMarkerThreadId) { + return (aMarkerThreadId == aThreadId) ? &aWriter : nullptr; + }, + [&](ProfileChunkedBuffer& aChunkedBuffer) { + ProfilerBacktrace backtrace("", &aChunkedBuffer); + backtrace.StreamJSON(aWriter, TimeStamp::ProcessCreation(), + aUniqueStacks); + }, + // We don't have Rust markers in the mozglue. + [&](mozilla::base_profiler_markers_detail::Streaming:: + DeserializerTag) { + MOZ_ASSERT_UNREACHABLE("No Rust markers in mozglue."); + }); + } else { + // The entry was not a marker, we need to skip to the end. + aER.SetRemainingBytes(0); + } + }); +} + +void ProfileBuffer::StreamProfilerOverheadToJSON( + SpliceableJSONWriter& aWriter, const TimeStamp& aProcessStartTime, + double aSinceTime) const { + mEntries.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, + "ProfileChunkedBuffer cannot be out-of-session when sampler is " + "running"); + + EntryGetter e(*aReader); + + enum Schema : uint32_t { + TIME = 0, + LOCKING = 1, + MARKER_CLEANING = 2, + COUNTERS = 3, + THREADS = 4 + }; + + aWriter.StartObjectProperty("profilerOverhead"); + aWriter.StartObjectProperty("samples"); + // Stream all sampling overhead data. We skip other entries, because we + // process them in StreamSamplesToJSON()/etc. + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("time"); + schema.WriteField("locking"); + schema.WriteField("expiredMarkerCleaning"); + schema.WriteField("counters"); + schema.WriteField("threads"); + } + + aWriter.StartArrayProperty("data"); + double firstTime = 0.0; + double lastTime = 0.0; + ProfilerStats intervals, overheads, lockings, cleanings, counters, threads; + while (e.Has()) { + // valid sequence: ProfilerOverheadTime, ProfilerOverheadDuration * 4 + if (e.Get().IsProfilerOverheadTime()) { + double time = e.Get().GetDouble(); + if (time >= aSinceTime) { + e.Next(); + if (!e.Has() || !e.Get().IsProfilerOverheadDuration()) { + ERROR_AND_CONTINUE( + "expected a ProfilerOverheadDuration entry after " + "ProfilerOverheadTime"); + } + double locking = e.Get().GetDouble(); + e.Next(); + if (!e.Has() || !e.Get().IsProfilerOverheadDuration()) { + ERROR_AND_CONTINUE( + "expected a ProfilerOverheadDuration entry after " + "ProfilerOverheadTime,ProfilerOverheadDuration"); + } + double cleaning = e.Get().GetDouble(); + e.Next(); + if (!e.Has() || !e.Get().IsProfilerOverheadDuration()) { + ERROR_AND_CONTINUE( + "expected a ProfilerOverheadDuration entry after " + "ProfilerOverheadTime,ProfilerOverheadDuration*2"); + } + double counter = e.Get().GetDouble(); + e.Next(); + if (!e.Has() || !e.Get().IsProfilerOverheadDuration()) { + ERROR_AND_CONTINUE( + "expected a ProfilerOverheadDuration entry after " + "ProfilerOverheadTime,ProfilerOverheadDuration*3"); + } + double thread = e.Get().GetDouble(); + + if (firstTime == 0.0) { + firstTime = time; + } else { + // Note that we'll have 1 fewer interval than other numbers (because + // we need both ends of an interval to know its duration). The final + // difference should be insignificant over the expected many + // thousands of iterations. + intervals.Count(time - lastTime); + } + lastTime = time; + overheads.Count(locking + cleaning + counter + thread); + lockings.Count(locking); + cleanings.Count(cleaning); + counters.Count(counter); + threads.Count(thread); + + AutoArraySchemaWriter writer(aWriter); + writer.TimeMsElement(TIME, time); + writer.DoubleElement(LOCKING, locking); + writer.DoubleElement(MARKER_CLEANING, cleaning); + writer.DoubleElement(COUNTERS, counter); + writer.DoubleElement(THREADS, thread); + } + } + e.Next(); + } + aWriter.EndArray(); // data + aWriter.EndObject(); // samples + + // Only output statistics if there is at least one full interval (and + // therefore at least two samplings.) + if (intervals.n > 0) { + aWriter.StartObjectProperty("statistics"); + aWriter.DoubleProperty("profiledDuration", lastTime - firstTime); + aWriter.IntProperty("samplingCount", overheads.n); + aWriter.DoubleProperty("overheadDurations", overheads.sum); + aWriter.DoubleProperty("overheadPercentage", + overheads.sum / (lastTime - firstTime)); +#define PROFILER_STATS(name, var) \ + aWriter.DoubleProperty("mean" name, (var).sum / (var).n); \ + aWriter.DoubleProperty("min" name, (var).min); \ + aWriter.DoubleProperty("max" name, (var).max); + PROFILER_STATS("Interval", intervals); + PROFILER_STATS("Overhead", overheads); + PROFILER_STATS("Lockings", lockings); + PROFILER_STATS("Cleaning", cleanings); + PROFILER_STATS("Counter", counters); + PROFILER_STATS("Thread", threads); +#undef PROFILER_STATS + aWriter.EndObject(); // statistics + } + aWriter.EndObject(); // profilerOverhead + }); +} + +struct CounterKeyedSample { + double mTime; + uint64_t mNumber; + int64_t mCount; +}; + +using CounterKeyedSamples = Vector; + +using CounterMap = HashMap; + +// HashMap lookup, if not found, a default value is inserted. +// Returns reference to (existing or new) value inside the HashMap. +template +static auto& LookupOrAdd(HashM& aMap, Key&& aKey) { + auto addPtr = aMap.lookupForAdd(aKey); + if (!addPtr) { + MOZ_RELEASE_ASSERT(aMap.add(addPtr, std::forward(aKey), + typename HashM::Entry::ValueType{})); + MOZ_ASSERT(!!addPtr); + } + return addPtr->value(); +} + +void ProfileBuffer::StreamCountersToJSON(SpliceableJSONWriter& aWriter, + const TimeStamp& aProcessStartTime, + double aSinceTime) const { + // Because this is a format entirely internal to the Profiler, any parsing + // error indicates a bug in the ProfileBuffer writing or the parser itself, + // or possibly flaky hardware. + + mEntries.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, + "ProfileChunkedBuffer cannot be out-of-session when sampler is " + "running"); + + EntryGetter e(*aReader); + + enum Schema : uint32_t { TIME = 0, COUNT = 1, NUMBER = 2 }; + + // Stream all counters. We skip other entries, because we process them in + // StreamSamplesToJSON()/etc. + // + // Valid sequence in the buffer: + // CounterID + // Time + // ( CounterKey Count Number? )* + // + // And the JSON (example): + // "counters": { + // "name": "malloc", + // "category": "Memory", + // "description": "Amount of allocated memory", + // "sample_groups": { + // "id": 0, + // "samples": { + // "schema": {"time": 0, "number": 1, "count": 2}, + // "data": [ + // [ + // 16117.033968000002, + // 2446216, + // 6801320 + // ], + // [ + // 16118.037638, + // 2446216, + // 6801320 + // ], + // ], + // } + // } + // }, + + // Build the map of counters and populate it + HashMap counters; + + while (e.Has()) { + // skip all non-Counters, including if we start in the middle of a counter + if (e.Get().IsCounterId()) { + void* id = e.Get().GetPtr(); + CounterMap& counter = LookupOrAdd(counters, id); + e.Next(); + if (!e.Has() || !e.Get().IsTime()) { + ERROR_AND_CONTINUE("expected a Time entry"); + } + double time = e.Get().GetDouble(); + e.Next(); + if (time >= aSinceTime) { + while (e.Has() && e.Get().IsCounterKey()) { + uint64_t key = e.Get().GetUint64(); + CounterKeyedSamples& data = LookupOrAdd(counter, key); + e.Next(); + if (!e.Has() || !e.Get().IsCount()) { + ERROR_AND_CONTINUE("expected a Count entry"); + } + int64_t count = e.Get().GetUint64(); + e.Next(); + uint64_t number; + if (!e.Has() || !e.Get().IsNumber()) { + number = 0; + } else { + number = e.Get().GetInt64(); + e.Next(); + } + CounterKeyedSample sample = {time, number, count}; + MOZ_RELEASE_ASSERT(data.append(sample)); + } + } else { + // skip counter sample - only need to skip the initial counter + // id, then let the loop at the top skip the rest + } + } else { + e.Next(); + } + } + // we have a map of a map of counter entries; dump them to JSON + if (counters.count() == 0) { + return; + } + + aWriter.StartArrayProperty("counters"); + for (auto iter = counters.iter(); !iter.done(); iter.next()) { + CounterMap& counter = iter.get().value(); + const BaseProfilerCount* base_counter = + static_cast(iter.get().key()); + + aWriter.Start(); + aWriter.StringProperty("name", MakeStringSpan(base_counter->mLabel)); + aWriter.StringProperty("category", + MakeStringSpan(base_counter->mCategory)); + aWriter.StringProperty("description", + MakeStringSpan(base_counter->mDescription)); + + aWriter.StartArrayProperty("sample_groups"); + for (auto counter_iter = counter.iter(); !counter_iter.done(); + counter_iter.next()) { + CounterKeyedSamples& samples = counter_iter.get().value(); + uint64_t key = counter_iter.get().key(); + + size_t size = samples.length(); + if (size == 0) { + continue; + } + + bool hasNumber = false; + for (size_t i = 0; i < size; i++) { + if (samples[i].mNumber != 0) { + hasNumber = true; + break; + } + } + + aWriter.StartObjectElement(); + { + aWriter.IntProperty("id", static_cast(key)); + aWriter.StartObjectProperty("samples"); + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("time"); + schema.WriteField("count"); + if (hasNumber) { + schema.WriteField("number"); + } + } + + aWriter.StartArrayProperty("data"); + uint64_t previousNumber = 0; + int64_t previousCount = 0; + for (size_t i = 0; i < size; i++) { + // Encode as deltas, and only encode if different than the last + // sample + if (i == 0 || samples[i].mNumber != previousNumber || + samples[i].mCount != previousCount) { + MOZ_ASSERT(i == 0 || samples[i].mTime >= samples[i - 1].mTime); + MOZ_ASSERT(samples[i].mNumber >= previousNumber); + MOZ_ASSERT(samples[i].mNumber - previousNumber <= + uint64_t(std::numeric_limits::max())); + + AutoArraySchemaWriter writer(aWriter); + writer.TimeMsElement(TIME, samples[i].mTime); + writer.IntElement(COUNT, samples[i].mCount - previousCount); + if (hasNumber) { + writer.IntElement( + NUMBER, + static_cast(samples[i].mNumber - previousNumber)); + } + previousNumber = samples[i].mNumber; + previousCount = samples[i].mCount; + } + } + aWriter.EndArray(); // data + aWriter.EndObject(); // samples + } + aWriter.EndObject(); // sample_groups item + } + aWriter.EndArray(); // sample groups + aWriter.End(); // for each counter + } + aWriter.EndArray(); // counters + }); +} + +#undef ERROR_AND_CONTINUE + +static void AddPausedRange(SpliceableJSONWriter& aWriter, const char* aReason, + const Maybe& aStartTime, + const Maybe& aEndTime) { + aWriter.Start(); + if (aStartTime) { + aWriter.TimeDoubleMsProperty("startTime", *aStartTime); + } else { + aWriter.NullProperty("startTime"); + } + if (aEndTime) { + aWriter.TimeDoubleMsProperty("endTime", *aEndTime); + } else { + aWriter.NullProperty("endTime"); + } + aWriter.StringProperty("reason", MakeStringSpan(aReason)); + aWriter.End(); +} + +void ProfileBuffer::StreamPausedRangesToJSON(SpliceableJSONWriter& aWriter, + double aSinceTime) const { + mEntries.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, + "ProfileChunkedBuffer cannot be out-of-session when sampler is " + "running"); + + EntryGetter e(*aReader); + + Maybe currentPauseStartTime; + Maybe currentCollectionStartTime; + + while (e.Has()) { + if (e.Get().IsPause()) { + currentPauseStartTime = Some(e.Get().GetDouble()); + } else if (e.Get().IsResume()) { + AddPausedRange(aWriter, "profiler-paused", currentPauseStartTime, + Some(e.Get().GetDouble())); + currentPauseStartTime = Nothing(); + } else if (e.Get().IsCollectionStart()) { + currentCollectionStartTime = Some(e.Get().GetDouble()); + } else if (e.Get().IsCollectionEnd()) { + AddPausedRange(aWriter, "collecting", currentCollectionStartTime, + Some(e.Get().GetDouble())); + currentCollectionStartTime = Nothing(); + } + e.Next(); + } + + if (currentPauseStartTime) { + AddPausedRange(aWriter, "profiler-paused", currentPauseStartTime, + Nothing()); + } + if (currentCollectionStartTime) { + AddPausedRange(aWriter, "collecting", currentCollectionStartTime, + Nothing()); + } + }); +} + +bool ProfileBuffer::DuplicateLastSample(BaseProfilerThreadId aThreadId, + const TimeStamp& aProcessStartTime, + Maybe& aLastSample) { + if (!aLastSample) { + return false; + } + + ProfileChunkedBuffer tempBuffer( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex, WorkerChunkManager()); + + auto retrieveWorkerChunk = MakeScopeExit( + [&]() { WorkerChunkManager().Reset(tempBuffer.GetAllChunks()); }); + + const bool ok = mEntries.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, + "ProfileChunkedBuffer cannot be out-of-session when sampler is " + "running"); + + EntryGetter e(*aReader, *aLastSample); + + if (e.CurPos() != *aLastSample) { + // The last sample is no longer within the buffer range, so we cannot + // use it. Reset the stored buffer position to Nothing(). + aLastSample.reset(); + return false; + } + + MOZ_RELEASE_ASSERT(e.Has() && e.Get().IsThreadId() && + e.Get().GetThreadId() == aThreadId); + + e.Next(); + + // Go through the whole entry and duplicate it, until we find the next + // one. + while (e.Has()) { + switch (e.Get().GetKind()) { + case ProfileBufferEntry::Kind::Pause: + case ProfileBufferEntry::Kind::Resume: + case ProfileBufferEntry::Kind::PauseSampling: + case ProfileBufferEntry::Kind::ResumeSampling: + case ProfileBufferEntry::Kind::CollectionStart: + case ProfileBufferEntry::Kind::CollectionEnd: + case ProfileBufferEntry::Kind::ThreadId: + // We're done. + return true; + case ProfileBufferEntry::Kind::Time: + // Copy with new time + AddEntry( + tempBuffer, + ProfileBufferEntry::Time( + (TimeStamp::Now() - aProcessStartTime).ToMilliseconds())); + break; + case ProfileBufferEntry::Kind::CounterKey: + case ProfileBufferEntry::Kind::Number: + case ProfileBufferEntry::Kind::Count: + case ProfileBufferEntry::Kind::Responsiveness: + // Don't copy anything not part of a thread's stack sample + break; + case ProfileBufferEntry::Kind::CounterId: + // CounterId is normally followed by Time - if so, we'd like + // to skip it. If we duplicate Time, it won't hurt anything, just + // waste buffer space (and this can happen if the CounterId has + // fallen off the end of the buffer, but Time (and Number/Count) + // are still in the buffer). + e.Next(); + if (e.Has() && e.Get().GetKind() != ProfileBufferEntry::Kind::Time) { + // this would only happen if there was an invalid sequence + // in the buffer. Don't skip it. + continue; + } + // we've skipped Time + break; + case ProfileBufferEntry::Kind::ProfilerOverheadTime: + // ProfilerOverheadTime is normally followed by + // ProfilerOverheadDuration*4 - if so, we'd like to skip it. Don't + // duplicate, as we are in the middle of a sampling and will soon + // capture its own overhead. + e.Next(); + // A missing Time would only happen if there was an invalid + // sequence in the buffer. Don't skip unexpected entry. + if (e.Has() && + e.Get().GetKind() != + ProfileBufferEntry::Kind::ProfilerOverheadDuration) { + continue; + } + e.Next(); + if (e.Has() && + e.Get().GetKind() != + ProfileBufferEntry::Kind::ProfilerOverheadDuration) { + continue; + } + e.Next(); + if (e.Has() && + e.Get().GetKind() != + ProfileBufferEntry::Kind::ProfilerOverheadDuration) { + continue; + } + e.Next(); + if (e.Has() && + e.Get().GetKind() != + ProfileBufferEntry::Kind::ProfilerOverheadDuration) { + continue; + } + // we've skipped ProfilerOverheadTime and + // ProfilerOverheadDuration*4. + break; + default: { + // Copy anything else we don't know about. + AddEntry(tempBuffer, e.Get()); + break; + } + } + e.Next(); + } + return true; + }); + + if (!ok) { + return false; + } + + // If the buffer was big enough, there won't be any cleared blocks. + if (tempBuffer.GetState().mClearedBlockCount != 0) { + // No need to try to read stack again as it won't fit. Reset the stored + // buffer position to Nothing(). + aLastSample.reset(); + return false; + } + + aLastSample = Some(AddThreadIdEntry(aThreadId)); + + tempBuffer.Read([&](ProfileChunkedBuffer::Reader* aReader) { + MOZ_ASSERT(aReader, "tempBuffer cannot be out-of-session"); + + EntryGetter e(*aReader); + + while (e.Has()) { + AddEntry(e.Get()); + e.Next(); + } + }); + + return true; +} + +void ProfileBuffer::DiscardSamplesBeforeTime(double aTime) { + // This function does nothing! + // The duration limit will be removed from Firefox, see bug 1632365. + Unused << aTime; +} + +// END ProfileBuffer +//////////////////////////////////////////////////////////////////////// + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/ProfileBufferEntry.h b/mozglue/baseprofiler/core/ProfileBufferEntry.h new file mode 100644 index 0000000000..ee6e401bd8 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfileBufferEntry.h @@ -0,0 +1,364 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef ProfileBufferEntry_h +#define ProfileBufferEntry_h + +#include "BaseProfilingCategory.h" +#include "gtest/MozGtestFriend.h" +#include "mozilla/BaseProfileJSONWriter.h" +#include "mozilla/HashFunctions.h" +#include "mozilla/HashTable.h" +#include "mozilla/Maybe.h" +#include "mozilla/ProfileBufferEntryKinds.h" +#include "mozilla/UniquePtr.h" +#include "mozilla/Variant.h" +#include "mozilla/Vector.h" + +#include +#include + +namespace mozilla { +namespace baseprofiler { + +class ProfileBufferEntry { + public: + using KindUnderlyingType = + std::underlying_type_t<::mozilla::ProfileBufferEntryKind>; + using Kind = ::mozilla::ProfileBufferEntryKind; + + ProfileBufferEntry(); + + static constexpr size_t kNumChars = ::mozilla::ProfileBufferEntryNumChars; + + private: + // aString must be a static string. + ProfileBufferEntry(Kind aKind, const char* aString); + ProfileBufferEntry(Kind aKind, char aChars[kNumChars]); + ProfileBufferEntry(Kind aKind, void* aPtr); + ProfileBufferEntry(Kind aKind, double aDouble); + ProfileBufferEntry(Kind aKind, int64_t aInt64); + ProfileBufferEntry(Kind aKind, uint64_t aUint64); + ProfileBufferEntry(Kind aKind, int aInt); + ProfileBufferEntry(Kind aKind, BaseProfilerThreadId aThreadId); + + public: +#define CTOR(KIND, TYPE, SIZE) \ + static ProfileBufferEntry KIND(TYPE aVal) { \ + return ProfileBufferEntry(Kind::KIND, aVal); \ + } + FOR_EACH_PROFILE_BUFFER_ENTRY_KIND(CTOR) +#undef CTOR + + Kind GetKind() const { return mKind; } + +#define IS_KIND(KIND, TYPE, SIZE) \ + bool Is##KIND() const { return mKind == Kind::KIND; } + FOR_EACH_PROFILE_BUFFER_ENTRY_KIND(IS_KIND) +#undef IS_KIND + + private: + FRIEND_TEST(ThreadProfile, InsertOneEntry); + FRIEND_TEST(ThreadProfile, InsertOneEntryWithTinyBuffer); + FRIEND_TEST(ThreadProfile, InsertEntriesNoWrap); + FRIEND_TEST(ThreadProfile, InsertEntriesWrap); + FRIEND_TEST(ThreadProfile, MemoryMeasure); + friend class ProfileBuffer; + + Kind mKind; + uint8_t mStorage[kNumChars]; + + const char* GetString() const; + void* GetPtr() const; + double GetDouble() const; + int GetInt() const; + int64_t GetInt64() const; + uint64_t GetUint64() const; + BaseProfilerThreadId GetThreadId() const; + void CopyCharsInto(char (&aOutArray)[kNumChars]) const; +}; + +// Packed layout: 1 byte for the tag + 8 bytes for the value. +static_assert(sizeof(ProfileBufferEntry) == 9, "bad ProfileBufferEntry size"); + +class UniqueStacks { + public: + struct FrameKey { + explicit FrameKey(const char* aLocation) + : mData(NormalFrameData{std::string(aLocation), false, 0, Nothing(), + Nothing()}) {} + + FrameKey(std::string&& aLocation, bool aRelevantForJS, + uint64_t aInnerWindowID, const Maybe& aLine, + const Maybe& aColumn, + const Maybe& aCategoryPair) + : mData(NormalFrameData{aLocation, aRelevantForJS, aInnerWindowID, + aLine, aColumn, aCategoryPair}) {} + + FrameKey(const FrameKey& aToCopy) = default; + + uint32_t Hash() const; + bool operator==(const FrameKey& aOther) const { + return mData == aOther.mData; + } + + struct NormalFrameData { + bool operator==(const NormalFrameData& aOther) const; + + std::string mLocation; + bool mRelevantForJS; + uint64_t mInnerWindowID; + Maybe mLine; + Maybe mColumn; + Maybe mCategoryPair; + }; + Variant mData; + }; + + struct FrameKeyHasher { + using Lookup = FrameKey; + + static HashNumber hash(const FrameKey& aLookup) { + HashNumber hash = 0; + if (aLookup.mData.is()) { + const FrameKey::NormalFrameData& data = + aLookup.mData.as(); + if (!data.mLocation.empty()) { + hash = AddToHash(hash, HashString(data.mLocation.c_str())); + } + hash = AddToHash(hash, data.mRelevantForJS); + hash = mozilla::AddToHash(hash, data.mInnerWindowID); + if (data.mLine.isSome()) { + hash = AddToHash(hash, *data.mLine); + } + if (data.mColumn.isSome()) { + hash = AddToHash(hash, *data.mColumn); + } + if (data.mCategoryPair.isSome()) { + hash = AddToHash(hash, static_cast(*data.mCategoryPair)); + } + } + return hash; + } + + static bool match(const FrameKey& aKey, const FrameKey& aLookup) { + return aKey == aLookup; + } + + static void rekey(FrameKey& aKey, const FrameKey& aNewKey) { + aKey = aNewKey; + } + }; + + struct StackKey { + Maybe mPrefixStackIndex; + uint32_t mFrameIndex; + + explicit StackKey(uint32_t aFrame) + : mFrameIndex(aFrame), mHash(HashGeneric(aFrame)) {} + + StackKey(const StackKey& aPrefix, uint32_t aPrefixStackIndex, + uint32_t aFrame) + : mPrefixStackIndex(Some(aPrefixStackIndex)), + mFrameIndex(aFrame), + mHash(AddToHash(aPrefix.mHash, aFrame)) {} + + HashNumber Hash() const { return mHash; } + + bool operator==(const StackKey& aOther) const { + return mPrefixStackIndex == aOther.mPrefixStackIndex && + mFrameIndex == aOther.mFrameIndex; + } + + private: + HashNumber mHash; + }; + + struct StackKeyHasher { + using Lookup = StackKey; + + static HashNumber hash(const StackKey& aLookup) { return aLookup.Hash(); } + + static bool match(const StackKey& aKey, const StackKey& aLookup) { + return aKey == aLookup; + } + + static void rekey(StackKey& aKey, const StackKey& aNewKey) { + aKey = aNewKey; + } + }; + + UniqueStacks(); + + // Return a StackKey for aFrame as the stack's root frame (no prefix). + [[nodiscard]] StackKey BeginStack(const FrameKey& aFrame); + + // Return a new StackKey that is obtained by appending aFrame to aStack. + [[nodiscard]] StackKey AppendFrame(const StackKey& aStack, + const FrameKey& aFrame); + + [[nodiscard]] uint32_t GetOrAddFrameIndex(const FrameKey& aFrame); + [[nodiscard]] uint32_t GetOrAddStackIndex(const StackKey& aStack); + + void SpliceFrameTableElements(SpliceableJSONWriter& aWriter); + void SpliceStackTableElements(SpliceableJSONWriter& aWriter); + + UniqueJSONStrings& UniqueStrings() { + MOZ_RELEASE_ASSERT(mUniqueStrings.get()); + return *mUniqueStrings; + } + + private: + void StreamNonJITFrame(const FrameKey& aFrame); + void StreamStack(const StackKey& aStack); + + UniquePtr mUniqueStrings; + + SpliceableChunkedJSONWriter mFrameTableWriter; + HashMap mFrameToIndexMap; + + SpliceableChunkedJSONWriter mStackTableWriter; + HashMap mStackToIndexMap; +}; + +// +// Thread profile JSON Format +// -------------------------- +// +// The profile contains much duplicate information. The output JSON of the +// profile attempts to deduplicate strings, frames, and stack prefixes, to cut +// down on size and to increase JSON streaming speed. Deduplicated values are +// streamed as indices into their respective tables. +// +// Further, arrays of objects with the same set of properties (e.g., samples, +// frames) are output as arrays according to a schema instead of an object +// with property names. A property that is not present is represented in the +// array as null or undefined. +// +// The format of the thread profile JSON is shown by the following example +// with 1 sample and 1 marker: +// +// { +// "name": "Foo", +// "tid": 42, +// "samples": +// { +// "schema": +// { +// "stack": 0, /* index into stackTable */ +// "time": 1, /* number */ +// "eventDelay": 2, /* number */ +// }, +// "data": +// [ +// [ 1, 0.0, 0.0 ] /* { stack: 1, time: 0.0, eventDelay: 0.0 } */ +// ] +// }, +// +// "markers": +// { +// "schema": +// { +// "name": 0, /* index into stringTable */ +// "time": 1, /* number */ +// "data": 2 /* arbitrary JSON */ +// }, +// "data": +// [ +// [ 3, 0.1 ] /* { name: 'example marker', time: 0.1 } */ +// ] +// }, +// +// "stackTable": +// { +// "schema": +// { +// "prefix": 0, /* index into stackTable */ +// "frame": 1 /* index into frameTable */ +// }, +// "data": +// [ +// [ null, 0 ], /* (root) */ +// [ 0, 1 ] /* (root) > foo.js */ +// ] +// }, +// +// "frameTable": +// { +// "schema": +// { +// "location": 0, /* index into stringTable */ +// "relevantForJS": 1, /* bool */ +// "innerWindowID": 2, /* inner window ID of global JS `window` object */ +// "implementation": 3, /* index into stringTable */ +// "line": 4, /* number */ +// "column": 5, /* number */ +// "category": 6, /* index into profile.meta.categories */ +// "subcategory": 7 /* index into +// profile.meta.categories[category].subcategories */ +// }, +// "data": +// [ +// [ 0 ], /* { location: '(root)' } */ +// [ 1, 2 ] /* { location: 'foo.js', +// implementation: 'baseline' } */ +// ] +// }, +// +// "stringTable": +// [ +// "(root)", +// "foo.js", +// "baseline", +// "example marker" +// ] +// } +// +// Process: +// { +// "name": "Bar", +// "pid": 24, +// "threads": +// [ +// <0-N threads from above> +// ], +// "counters": /* includes the memory counter */ +// [ +// { +// "name": "qwerty", +// "category": "uiop", +// "description": "this is qwerty uiop", +// "sample_groups: +// [ +// { +// "id": 42, /* number (thread id, or object identifier (tab), etc) */ +// "samples: +// { +// "schema": +// { +// "time": 1, /* number */ +// "number": 2, /* number (of times the counter was touched) */ +// "count": 3 /* number (total for the counter) */ +// }, +// "data": +// [ +// [ 0.1, 1824, +// 454622 ] /* { time: 0.1, number: 1824, count: 454622 } */ +// ] +// }, +// }, +// /* more sample-group objects with different id's */ +// ] +// }, +// /* more counters */ +// ], +// } +// + +} // namespace baseprofiler +} // namespace mozilla + +#endif /* ndef ProfileBufferEntry_h */ diff --git a/mozglue/baseprofiler/core/ProfileJSONWriter.cpp b/mozglue/baseprofiler/core/ProfileJSONWriter.cpp new file mode 100644 index 0000000000..dc51f6958e --- /dev/null +++ b/mozglue/baseprofiler/core/ProfileJSONWriter.cpp @@ -0,0 +1,101 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "mozilla/BaseProfileJSONWriter.h" + +namespace mozilla::baseprofiler { + +UniqueJSONStrings::UniqueJSONStrings(FailureLatch& aFailureLatch) + : mStringTableWriter(aFailureLatch) { + mStringTableWriter.StartBareList(); + if (const char* failure = mStringTableWriter.GetFailure(); failure) { + ClearAndSetFailure(failure); + return; + } +} + +UniqueJSONStrings::UniqueJSONStrings(FailureLatch& aFailureLatch, + const UniqueJSONStrings& aOther, + ProgressLogger aProgressLogger) + : mStringTableWriter(aFailureLatch) { + using namespace mozilla::literals::ProportionValue_literals; // For `10_pc`. + + if (mStringTableWriter.Failed()) { + return; + } + + if (const char* failure = aOther.GetFailure(); failure) { + ClearAndSetFailure(failure); + return; + } + + mStringTableWriter.StartBareList(); + uint32_t count = aOther.mStringHashToIndexMap.count(); + if (count != 0) { + if (!mStringHashToIndexMap.reserve(count)) { + ClearAndSetFailure("Cannot reserve UniqueJSONStrings map storage"); + return; + } + auto iter = aOther.mStringHashToIndexMap.iter(); + for (auto&& [unusedIndex, progressLogger] : + aProgressLogger.CreateLoopSubLoggersFromTo( + 10_pc, 90_pc, count, "Copying unique strings...")) { + (void)unusedIndex; + if (iter.done()) { + break; + } + mStringHashToIndexMap.putNewInfallible(iter.get().key(), + iter.get().value()); + iter.next(); + } + aProgressLogger.SetLocalProgress(90_pc, "Copied unique strings"); + mStringTableWriter.CopyAndSplice( + aOther.mStringTableWriter.ChunkedWriteFunc()); + if (const char* failure = aOther.GetFailure(); failure) { + ClearAndSetFailure(failure); + } + aProgressLogger.SetLocalProgress(100_pc, "Spliced unique strings"); + } +} + +UniqueJSONStrings::~UniqueJSONStrings() = default; + +void UniqueJSONStrings::SpliceStringTableElements( + SpliceableJSONWriter& aWriter) { + aWriter.TakeAndSplice(mStringTableWriter.TakeChunkedWriteFunc()); +} + +void UniqueJSONStrings::ClearAndSetFailure(std::string aFailure) { + mStringTableWriter.SetFailure(std::move(aFailure)); + mStringHashToIndexMap.clear(); +} + +Maybe UniqueJSONStrings::GetOrAddIndex(const Span& aStr) { + if (Failed()) { + return Nothing{}; + } + + uint32_t count = mStringHashToIndexMap.count(); + HashNumber hash = HashString(aStr.data(), aStr.size()); + auto entry = mStringHashToIndexMap.lookupForAdd(hash); + if (entry) { + MOZ_ASSERT(entry->value() < count); + return Some(entry->value()); + } + + if (!mStringHashToIndexMap.add(entry, hash, count)) { + ClearAndSetFailure("OOM in UniqueJSONStrings::GetOrAddIndex adding a map"); + return Nothing{}; + } + mStringTableWriter.StringElement(aStr); + if (const char* failure = mStringTableWriter.GetFailure(); failure) { + ClearAndSetFailure(failure); + return Nothing{}; + } + return Some(count); +} + +} // namespace mozilla::baseprofiler diff --git a/mozglue/baseprofiler/core/ProfiledThreadData.cpp b/mozglue/baseprofiler/core/ProfiledThreadData.cpp new file mode 100644 index 0000000000..62cb994ae2 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfiledThreadData.cpp @@ -0,0 +1,191 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "ProfiledThreadData.h" + +#include "BaseProfiler.h" +#include "ProfileBuffer.h" + +#include "mozilla/BaseProfileJSONWriter.h" + +#if defined(GP_OS_darwin) +# include +#endif + +namespace mozilla { +namespace baseprofiler { + +ProfiledThreadData::ProfiledThreadData(ThreadInfo* aThreadInfo) + : mThreadInfo(aThreadInfo) {} + +ProfiledThreadData::~ProfiledThreadData() {} + +void ProfiledThreadData::StreamJSON(const ProfileBuffer& aBuffer, + SpliceableJSONWriter& aWriter, + const std::string& aProcessName, + const std::string& aETLDplus1, + const TimeStamp& aProcessStartTime, + double aSinceTime) { + UniqueStacks uniqueStacks; + + aWriter.SetUniqueStrings(uniqueStacks.UniqueStrings()); + + aWriter.Start(); + { + StreamSamplesAndMarkers(mThreadInfo->Name(), mThreadInfo->ThreadId(), + aBuffer, aWriter, aProcessName, aETLDplus1, + aProcessStartTime, mThreadInfo->RegisterTime(), + mUnregisterTime, aSinceTime, uniqueStacks); + + aWriter.StartObjectProperty("stackTable"); + { + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("prefix"); + schema.WriteField("frame"); + } + + aWriter.StartArrayProperty("data"); + { uniqueStacks.SpliceStackTableElements(aWriter); } + aWriter.EndArray(); + } + aWriter.EndObject(); + + aWriter.StartObjectProperty("frameTable"); + { + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("location"); + schema.WriteField("relevantForJS"); + schema.WriteField("innerWindowID"); + schema.WriteField("implementation"); + schema.WriteField("line"); + schema.WriteField("column"); + schema.WriteField("category"); + schema.WriteField("subcategory"); + } + + aWriter.StartArrayProperty("data"); + { uniqueStacks.SpliceFrameTableElements(aWriter); } + aWriter.EndArray(); + } + aWriter.EndObject(); + + aWriter.StartArrayProperty("stringTable"); + { + std::move(uniqueStacks.UniqueStrings()) + .SpliceStringTableElements(aWriter); + } + aWriter.EndArray(); + } + aWriter.End(); + + aWriter.ResetUniqueStrings(); +} + +BaseProfilerThreadId StreamSamplesAndMarkers( + const char* aName, BaseProfilerThreadId aThreadId, + const ProfileBuffer& aBuffer, SpliceableJSONWriter& aWriter, + const std::string& aProcessName, const std::string& aETLDplus1, + const TimeStamp& aProcessStartTime, const TimeStamp& aRegisterTime, + const TimeStamp& aUnregisterTime, double aSinceTime, + UniqueStacks& aUniqueStacks) { + BaseProfilerThreadId processedThreadId; + + aWriter.StringProperty( + "processType", + "(unknown)" /* XRE_GeckoProcessTypeToString(XRE_GetProcessType()) */); + + { + std::string name = aName; + // We currently need to distinguish threads output by Base Profiler from + // those in Gecko Profiler, as the frontend could get confused and lose + // tracks with the same name. + // TODO: As part of the profilers de-duplication, thread data from both + // profilers should end up in the same track, at which point this won't be + // necessary anymore. See meta bug 1557566. + name += " (pre-xul)"; + aWriter.StringProperty("name", name); + } + + // Use given process name (if any). + if (!aProcessName.empty()) { + aWriter.StringProperty("processName", aProcessName); + } + if (!aETLDplus1.empty()) { + aWriter.StringProperty("eTLD+1", aETLDplus1); + } + + if (aRegisterTime) { + aWriter.DoubleProperty( + "registerTime", (aRegisterTime - aProcessStartTime).ToMilliseconds()); + } else { + aWriter.NullProperty("registerTime"); + } + + if (aUnregisterTime) { + aWriter.DoubleProperty( + "unregisterTime", + (aUnregisterTime - aProcessStartTime).ToMilliseconds()); + } else { + aWriter.NullProperty("unregisterTime"); + } + + aWriter.StartObjectProperty("samples"); + { + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("stack"); + schema.WriteField("time"); + schema.WriteField("eventDelay"); + } + + aWriter.StartArrayProperty("data"); + { + processedThreadId = aBuffer.StreamSamplesToJSON( + aWriter, aThreadId, aSinceTime, aUniqueStacks); + } + aWriter.EndArray(); + } + aWriter.EndObject(); + + aWriter.StartObjectProperty("markers"); + { + { + JSONSchemaWriter schema(aWriter); + schema.WriteField("name"); + schema.WriteField("startTime"); + schema.WriteField("endTime"); + schema.WriteField("phase"); + schema.WriteField("category"); + schema.WriteField("data"); + } + + aWriter.StartArrayProperty("data"); + { + aBuffer.StreamMarkersToJSON(aWriter, aThreadId, aProcessStartTime, + aSinceTime, aUniqueStacks); + } + aWriter.EndArray(); + } + aWriter.EndObject(); + + // Tech note: If `ToNumber()` returns a uint64_t, the conversion to int64_t is + // "implementation-defined" before C++20. This is acceptable here, because + // this is a one-way conversion to a unique identifier that's used to visually + // separate data by thread on the front-end. + aWriter.IntProperty( + "pid", static_cast(profiler_current_process_id().ToNumber())); + aWriter.IntProperty("tid", + static_cast(aThreadId.IsSpecified() + ? aThreadId.ToNumber() + : processedThreadId.ToNumber())); + + return processedThreadId; +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/ProfiledThreadData.h b/mozglue/baseprofiler/core/ProfiledThreadData.h new file mode 100644 index 0000000000..0590b03326 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfiledThreadData.h @@ -0,0 +1,120 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef ProfiledThreadData_h +#define ProfiledThreadData_h + +#include "BaseProfilingStack.h" +#include "platform.h" +#include "ProfileBufferEntry.h" +#include "ThreadInfo.h" + +#include "mozilla/RefPtr.h" +#include "mozilla/TimeStamp.h" +#include "mozilla/UniquePtr.h" + +#include + +namespace mozilla { +namespace baseprofiler { + +class ProfileBuffer; + +// This class contains information about a thread that is only relevant while +// the profiler is running, for any threads (both alive and dead) whose thread +// name matches the "thread filter" in the current profiler run. +// ProfiledThreadData objects may be kept alive even after the thread is +// unregistered, as long as there is still data for that thread in the profiler +// buffer. +// +// Accesses to this class are protected by the profiler state lock. +// +// Created as soon as the following are true for the thread: +// - The profiler is running, and +// - the thread matches the profiler's thread filter, and +// - the thread is registered with the profiler. +// So it gets created in response to either (1) the profiler being started (for +// an existing registered thread) or (2) the thread being registered (if the +// profiler is already running). +// +// The thread may be unregistered during the lifetime of ProfiledThreadData. +// If that happens, NotifyUnregistered() is called. +// +// This class is the right place to store buffer positions. Profiler buffer +// positions become invalid if the profiler buffer is destroyed, which happens +// when the profiler is stopped. +class ProfiledThreadData final { + public: + explicit ProfiledThreadData(ThreadInfo* aThreadInfo); + ~ProfiledThreadData(); + + void NotifyUnregistered(uint64_t aBufferPosition) { + mLastSample = Nothing(); + MOZ_ASSERT(!mBufferPositionWhenReceivedJSContext, + "JSContext should have been cleared before the thread was " + "unregistered"); + mUnregisterTime = TimeStamp::Now(); + mBufferPositionWhenUnregistered = Some(aBufferPosition); + } + Maybe BufferPositionWhenUnregistered() { + return mBufferPositionWhenUnregistered; + } + + Maybe& LastSample() { return mLastSample; } + + void StreamJSON(const ProfileBuffer& aBuffer, SpliceableJSONWriter& aWriter, + const std::string& aProcessName, + const std::string& aETLDplus1, + const TimeStamp& aProcessStartTime, double aSinceTime); + + const RefPtr Info() const { return mThreadInfo; } + + void NotifyReceivedJSContext(uint64_t aCurrentBufferPosition) { + mBufferPositionWhenReceivedJSContext = Some(aCurrentBufferPosition); + } + + private: + // Group A: + // The following fields are interesting for the entire lifetime of a + // ProfiledThreadData object. + + // This thread's thread info. + const RefPtr mThreadInfo; + + // Group B: + // The following fields are only used while this thread is alive and + // registered. They become Nothing() once the thread is unregistered. + + // When sampling, this holds the position in ActivePS::mBuffer of the most + // recent sample for this thread, or Nothing() if there is no sample for this + // thread in the buffer. + Maybe mLastSample; + + // Only non-Nothing() if the thread currently has a JSContext. + Maybe mBufferPositionWhenReceivedJSContext; + + // Group C: + // The following fields are only used once this thread has been unregistered. + + Maybe mBufferPositionWhenUnregistered; + TimeStamp mUnregisterTime; +}; + +// Stream all samples and markers from aBuffer with the given aThreadId (or 0 +// for everything, which is assumed to be a single backtrace sample.) +// Returns the thread id of the output sample(s), or 0 if none was present. +BaseProfilerThreadId StreamSamplesAndMarkers( + const char* aName, BaseProfilerThreadId aThreadId, + const ProfileBuffer& aBuffer, SpliceableJSONWriter& aWriter, + const std::string& aProcessName, const std::string& aETLDplus1, + const TimeStamp& aProcessStartTime, const TimeStamp& aRegisterTime, + const TimeStamp& aUnregisterTime, double aSinceTime, + UniqueStacks& aUniqueStacks); + +} // namespace baseprofiler +} // namespace mozilla + +#endif // ProfiledThreadData_h diff --git a/mozglue/baseprofiler/core/ProfilerBacktrace.cpp b/mozglue/baseprofiler/core/ProfilerBacktrace.cpp new file mode 100644 index 0000000000..8343888351 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilerBacktrace.cpp @@ -0,0 +1,125 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "ProfilerBacktrace.h" + +#include "BaseProfiler.h" +#include "ProfileBuffer.h" +#include "ProfiledThreadData.h" +#include "ThreadInfo.h" + +#include "mozilla/BaseProfileJSONWriter.h" + +namespace mozilla { +namespace baseprofiler { + +ProfilerBacktrace::ProfilerBacktrace( + const char* aName, + UniquePtr aProfileChunkedBufferStorage, + UniquePtr aProfileBufferStorageOrNull /* = nullptr */) + : mName(aName), + mOptionalProfileChunkedBufferStorage( + std::move(aProfileChunkedBufferStorage)), + mProfileChunkedBuffer(mOptionalProfileChunkedBufferStorage.get()), + mOptionalProfileBufferStorage(std::move(aProfileBufferStorageOrNull)), + mProfileBuffer(mOptionalProfileBufferStorage.get()) { + if (mProfileBuffer) { + MOZ_RELEASE_ASSERT(mProfileChunkedBuffer, + "If we take ownership of a ProfileBuffer, we must also " + "receive ownership of a ProfileChunkedBuffer"); + MOZ_RELEASE_ASSERT( + mProfileChunkedBuffer == &mProfileBuffer->UnderlyingChunkedBuffer(), + "If we take ownership of a ProfileBuffer, we must also receive " + "ownership of its ProfileChunkedBuffer"); + } + MOZ_ASSERT( + !mProfileChunkedBuffer || !mProfileChunkedBuffer->IsThreadSafe(), + "ProfilerBacktrace only takes a non-thread-safe ProfileChunkedBuffer"); +} + +ProfilerBacktrace::ProfilerBacktrace( + const char* aName, + ProfileChunkedBuffer* aExternalProfileChunkedBufferOrNull /* = nullptr */, + ProfileBuffer* aExternalProfileBufferOrNull /* = nullptr */) + : mName(aName), + mProfileChunkedBuffer(aExternalProfileChunkedBufferOrNull), + mProfileBuffer(aExternalProfileBufferOrNull) { + if (!mProfileChunkedBuffer) { + if (mProfileBuffer) { + // We don't have a ProfileChunkedBuffer but we have a ProfileBuffer, use + // the latter's ProfileChunkedBuffer. + mProfileChunkedBuffer = &mProfileBuffer->UnderlyingChunkedBuffer(); + MOZ_ASSERT(!mProfileChunkedBuffer->IsThreadSafe(), + "ProfilerBacktrace only takes a non-thread-safe " + "ProfileChunkedBuffer"); + } + } else { + if (mProfileBuffer) { + MOZ_RELEASE_ASSERT( + mProfileChunkedBuffer == &mProfileBuffer->UnderlyingChunkedBuffer(), + "If we reference both ProfileChunkedBuffer and ProfileBuffer, they " + "must already be connected"); + } + MOZ_ASSERT(!mProfileChunkedBuffer->IsThreadSafe(), + "ProfilerBacktrace only takes a non-thread-safe " + "ProfileChunkedBuffer"); + } +} + +ProfilerBacktrace::~ProfilerBacktrace() {} + +BaseProfilerThreadId ProfilerBacktrace::StreamJSON( + SpliceableJSONWriter& aWriter, const TimeStamp& aProcessStartTime, + UniqueStacks& aUniqueStacks) { + BaseProfilerThreadId processedThreadId; + + // Unlike ProfiledThreadData::StreamJSON, we don't need to call + // ProfileBuffer::AddJITInfoForRange because ProfileBuffer does not contain + // any JitReturnAddr entries. For synchronous samples, JIT frames get expanded + // at sample time. + if (mProfileBuffer) { + processedThreadId = StreamSamplesAndMarkers( + mName.c_str(), BaseProfilerThreadId{}, *mProfileBuffer, aWriter, "", "", + aProcessStartTime, + /* aRegisterTime */ TimeStamp(), + /* aUnregisterTime */ TimeStamp(), + /* aSinceTime */ 0, aUniqueStacks); + } else if (mProfileChunkedBuffer) { + ProfileBuffer profileBuffer(*mProfileChunkedBuffer); + processedThreadId = StreamSamplesAndMarkers( + mName.c_str(), BaseProfilerThreadId{}, profileBuffer, aWriter, "", "", + aProcessStartTime, + /* aRegisterTime */ TimeStamp(), + /* aUnregisterTime */ TimeStamp(), + /* aSinceTime */ 0, aUniqueStacks); + } + // If there are no buffers, the backtrace is empty and nothing is streamed. + + return processedThreadId; +} + +} // namespace baseprofiler + +// static +template +UniquePtr +ProfileBufferEntryReader:: + Deserializer>::Read( + ProfileBufferEntryReader& aER) { + auto profileChunkedBuffer = aER.ReadObject>(); + if (!profileChunkedBuffer) { + return nullptr; + } + MOZ_ASSERT( + !profileChunkedBuffer->IsThreadSafe(), + "ProfilerBacktrace only stores non-thread-safe ProfileChunkedBuffers"); + std::string name = aER.ReadObject(); + return UniquePtr{ + new baseprofiler::ProfilerBacktrace(name.c_str(), + std::move(profileChunkedBuffer))}; +}; + +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/ProfilerBacktrace.h b/mozglue/baseprofiler/core/ProfilerBacktrace.h new file mode 100644 index 0000000000..0b5b69e738 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilerBacktrace.h @@ -0,0 +1,162 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef __PROFILER_BACKTRACE_H +#define __PROFILER_BACKTRACE_H + +#include "mozilla/ProfileChunkedBuffer.h" +#include "mozilla/UniquePtr.h" + +#include + +namespace mozilla { + +class TimeStamp; + +namespace baseprofiler { + +class ProfileBuffer; +class SpliceableJSONWriter; +class ThreadInfo; +class UniqueStacks; + +// ProfilerBacktrace encapsulates a synchronous sample. +// It can work with a ProfileBuffer and/or a ProfileChunkedBuffer (if both, they +// must already be linked together). The ProfileChunkedBuffer contains all the +// data; the ProfileBuffer is not strictly needed, only provide it if it is +// already available at the call site. +// And these buffers can either be: +// - owned here, so that the ProfilerBacktrace object can be kept for later +// use), OR +// - referenced through pointers (in cases where the backtrace is immediately +// streamed out, so we only need temporary references to external buffers); +// these pointers may be null for empty backtraces. +class ProfilerBacktrace { + public: + // Take ownership of external buffers and use them to keep, and to stream a + // backtrace. If a ProfileBuffer is given, its underlying chunked buffer must + // be provided as well. + explicit ProfilerBacktrace( + const char* aName, + UniquePtr aProfileChunkedBufferStorage, + UniquePtr aProfileBufferStorageOrNull = nullptr); + + // Take pointers to external buffers and use them to stream a backtrace. + // If null, the backtrace is effectively empty. + // If both are provided, they must already be connected. + explicit ProfilerBacktrace( + const char* aName, + ProfileChunkedBuffer* aExternalProfileChunkedBufferOrNull = nullptr, + ProfileBuffer* aExternalProfileBufferOrNull = nullptr); + + ~ProfilerBacktrace(); + + [[nodiscard]] bool IsEmpty() const { + return !mProfileChunkedBuffer || + ProfileBufferEntryWriter::Serializer::Bytes( + *mProfileChunkedBuffer) <= ULEB128Size(0u); + } + + // ProfilerBacktraces' stacks are deduplicated in the context of the + // profile that contains the backtrace as a marker payload. + // + // That is, markers that contain backtraces should not need their own stack, + // frame, and string tables. They should instead reuse their parent + // profile's tables. + BaseProfilerThreadId StreamJSON(SpliceableJSONWriter& aWriter, + const TimeStamp& aProcessStartTime, + UniqueStacks& aUniqueStacks); + + private: + // Used to de/serialize a ProfilerBacktrace. + friend ProfileBufferEntryWriter::Serializer; + friend ProfileBufferEntryReader::Deserializer; + + std::string mName; + + // `ProfileChunkedBuffer` in which `mProfileBuffer` stores its data; must be + // located before `mProfileBuffer` so that it's destroyed after. + UniquePtr mOptionalProfileChunkedBufferStorage; + // If null, there is no need to check mProfileBuffer's (if present) underlying + // buffer because this is done when constructed. + ProfileChunkedBuffer* mProfileChunkedBuffer; + + UniquePtr mOptionalProfileBufferStorage; + ProfileBuffer* mProfileBuffer; +}; + +} // namespace baseprofiler + +// Format: [ UniquePtr | name ] +// Initial len==0 marks a nullptr or empty backtrace. +template <> +struct ProfileBufferEntryWriter::Serializer { + static Length Bytes(const baseprofiler::ProfilerBacktrace& aBacktrace) { + if (!aBacktrace.mProfileChunkedBuffer) { + // No buffer. + return ULEB128Size(0u); + } + auto bufferBytes = SumBytes(*aBacktrace.mProfileChunkedBuffer); + if (bufferBytes <= ULEB128Size(0u)) { + // Empty buffer. + return ULEB128Size(0u); + } + return bufferBytes + SumBytes(aBacktrace.mName); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const baseprofiler::ProfilerBacktrace& aBacktrace) { + if (!aBacktrace.mProfileChunkedBuffer || + SumBytes(*aBacktrace.mProfileChunkedBuffer) <= ULEB128Size(0u)) { + // No buffer, or empty buffer. + aEW.WriteULEB128(0u); + return; + } + aEW.WriteObject(*aBacktrace.mProfileChunkedBuffer); + aEW.WriteObject(aBacktrace.mName); + } +}; + +template +struct ProfileBufferEntryWriter::Serializer< + UniquePtr> { + static Length Bytes(const UniquePtr& aBacktrace) { + if (!aBacktrace) { + // Null backtrace pointer (treated like an empty backtrace). + return ULEB128Size(0u); + } + return SumBytes(*aBacktrace); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const UniquePtr& aBacktrace) { + if (!aBacktrace) { + // Null backtrace pointer (treated like an empty backtrace). + aEW.WriteULEB128(0u); + return; + } + aEW.WriteObject(*aBacktrace); + } +}; + +template +struct ProfileBufferEntryReader::Deserializer< + UniquePtr> { + static void ReadInto( + ProfileBufferEntryReader& aER, + UniquePtr& aBacktrace) { + aBacktrace = Read(aER); + } + + static UniquePtr Read( + ProfileBufferEntryReader& aER); +}; + +} // namespace mozilla + +#endif // __PROFILER_BACKTRACE_H diff --git a/mozglue/baseprofiler/core/ProfilerMarkers.cpp b/mozglue/baseprofiler/core/ProfilerMarkers.cpp new file mode 100644 index 0000000000..eff6c7e364 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilerMarkers.cpp @@ -0,0 +1,348 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "mozilla/BaseProfilerMarkers.h" + +#include "mozilla/BaseProfilerUtils.h" + +#include + +namespace mozilla { +namespace base_profiler_markers_detail { + +// We need an atomic type that can hold a `DeserializerTag`. (Atomic doesn't +// work with too-small types.) +using DeserializerTagAtomic = unsigned; + +// The atomic sDeserializerCount still also include bits that act as a "RWLock": +// Whoever can set this bit gets exclusive access to the count and the whole +// sMarkerTypeFunctions1Based array, guaranteeing that it cannot be modified. +static constexpr DeserializerTagAtomic scExclusiveLock = 0x80'00'00'00u; +// Code that wants shared access can add this value, then ensure there is no +// exclusive lock, after which it's guaranteed that no exclusive lock can be +// taken until the shared lock count goes back to zero. +static constexpr DeserializerTagAtomic scSharedLockUnit = 0x00'01'00'00u; +// This mask isolates the actual count value from the lock bits. +static constexpr DeserializerTagAtomic scTagMask = 0x00'00'FF'FFu; + +// Number of currently-registered deserializers and other marker type functions. +// The high bits contain lock bits, see above. +static Atomic + sDeserializerCount{0}; + +// This needs to be big enough to handle all possible marker types. If one day +// this needs to be higher, the underlying DeserializerTag type will have to be +// changed. +static constexpr DeserializerTagAtomic DeserializerMax = 250; +static_assert(DeserializerMax <= scTagMask, + "DeserializerMax doesn't fit in scTagMask"); + +static_assert( + DeserializerMax <= std::numeric_limits::max(), + "The maximum number of deserializers must fit in the DeserializerTag type"); + +// Array of marker type functions. +// 1-based, i.e.: [0] -> tag 1, [DeserializerMax - 1] -> tag DeserializerMax. +// Elements are added at the next available atomically-incremented +// `sDeserializerCount` (minus 1) whenever a new marker type is used in a +// Firefox session; the content is kept between profiler runs in that session. +// There is theoretically a race between the increment and the time the entry is +// fully written, but in practice all new elements are written (during +// profiling, using a marker type for the first time) long before they are read +// (after profiling is paused). +static Streaming::MarkerTypeFunctions + sMarkerTypeFunctions1Based[DeserializerMax]; + +/* static */ Streaming::DeserializerTag Streaming::TagForMarkerTypeFunctions( + Streaming::MarkerDataDeserializer aDeserializer, + Streaming::MarkerTypeNameFunction aMarkerTypeNameFunction, + Streaming::MarkerSchemaFunction aMarkerSchemaFunction) { + MOZ_RELEASE_ASSERT(!!aDeserializer); + MOZ_RELEASE_ASSERT(!!aMarkerTypeNameFunction); + MOZ_RELEASE_ASSERT(!!aMarkerSchemaFunction); + + // Add a shared lock request, which will prevent future exclusive locking. + DeserializerTagAtomic tagWithLock = (sDeserializerCount += scSharedLockUnit); + + // An exclusive locker may have arrived before us, just wait for it to finish. + while ((tagWithLock & scExclusiveLock) != 0u) { + tagWithLock = sDeserializerCount; + } + + MOZ_ASSERT( + // This is equivalent to shifting right to only keep the lock counts. + tagWithLock / scSharedLockUnit < + // This is effectively half of the permissible shared lock range, + // that would mean way too many threads doing this work here! + scExclusiveLock / scSharedLockUnit / 2, + "The shared lock count is getting unexpectedly high, verify the " + "algorithm, and tweak constants if needed"); + + // Reserve a tag. Even if there are multiple shared-lock holders here, each + // one will get a different value, and therefore will access a different part + // of the sMarkerTypeFunctions1Based array. + const DeserializerTagAtomic tag = ++sDeserializerCount & scTagMask; + + MOZ_RELEASE_ASSERT( + tag <= DeserializerMax, + "Too many deserializers, consider increasing DeserializerMax. " + "Or is a deserializer stored again and again?"); + sMarkerTypeFunctions1Based[tag - 1] = {aDeserializer, aMarkerTypeNameFunction, + aMarkerSchemaFunction}; + + // And release our shared lock, to allow exclusive readers. + sDeserializerCount -= scSharedLockUnit; + + return static_cast(tag); +} + +/* static */ Streaming::MarkerDataDeserializer Streaming::DeserializerForTag( + Streaming::DeserializerTag aTag) { + MOZ_RELEASE_ASSERT( + aTag > 0 && static_cast(aTag) <= + static_cast(sDeserializerCount), + "Out-of-range tag value"); + return sMarkerTypeFunctions1Based[aTag - 1].mMarkerDataDeserializer; +} + +Streaming::LockedMarkerTypeFunctionsList::LockedMarkerTypeFunctionsList() { + for (;;) { + const DeserializerTagAtomic count = sDeserializerCount; + if ((count & scTagMask) != count) { + // Someone already has a lock, loop around. + continue; + } + + // There are currently no locks, try to add our exclusive lock. + if (!sDeserializerCount.compareExchange(count, count | scExclusiveLock)) { + // Someone else modified sDeserializerCount since our read, loop around. + continue; + } + + // We applied our exclusive lock, we can now read the list of functions, + // without interference until ~LockedMarkerTypeFunctionsList(). + // (Note that sDeserializerCount may receive shared lock requests, but the + // count won't change.) + mMarkerTypeFunctionsSpan = {sMarkerTypeFunctions1Based, count}; + break; + } +} + +Streaming::LockedMarkerTypeFunctionsList::~LockedMarkerTypeFunctionsList() { + MOZ_ASSERT( + (sDeserializerCount & scExclusiveLock) == scExclusiveLock, + "sDeserializerCount should still have the the exclusive lock bit set"); + MOZ_ASSERT( + (sDeserializerCount & scTagMask) == + DeserializerTagAtomic(mMarkerTypeFunctionsSpan.size()), + "sDeserializerCount should have the same count since construction"); + sDeserializerCount &= ~scExclusiveLock; +} + +// Only accessed on the main thread. +// Both profilers (Base and Gecko) could be active at the same time, so keep a +// ref-count to only allocate at most one buffer at any time. +static int sBufferForMainThreadAddMarkerRefCount = 0; +static ProfileChunkedBuffer* sBufferForMainThreadAddMarker = nullptr; + +ProfileChunkedBuffer* GetClearedBufferForMainThreadAddMarker() { + if (!mozilla::baseprofiler::profiler_is_main_thread()) { + return nullptr; + } + + if (sBufferForMainThreadAddMarker) { + MOZ_ASSERT(sBufferForMainThreadAddMarker->IsInSession(), + "sBufferForMainThreadAddMarker should always be in-session"); + sBufferForMainThreadAddMarker->Clear(); + MOZ_ASSERT( + sBufferForMainThreadAddMarker->IsInSession(), + "Cleared sBufferForMainThreadAddMarker should still be in-session"); + } + + return sBufferForMainThreadAddMarker; +} + +MFBT_API void EnsureBufferForMainThreadAddMarker() { + if (!mozilla::baseprofiler::profiler_is_main_thread()) { + return; + } + + if (sBufferForMainThreadAddMarkerRefCount++ == 0) { + // First `Ensure`, allocate the buffer. + MOZ_ASSERT(!sBufferForMainThreadAddMarker); + sBufferForMainThreadAddMarker = new ProfileChunkedBuffer( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex, + MakeUnique( + ProfileBufferChunkManager::scExpectedMaximumStackSize)); + MOZ_ASSERT(sBufferForMainThreadAddMarker); + MOZ_ASSERT(sBufferForMainThreadAddMarker->IsInSession()); + } +} + +MFBT_API void ReleaseBufferForMainThreadAddMarker() { + if (!mozilla::baseprofiler::profiler_is_main_thread()) { + return; + } + + if (sBufferForMainThreadAddMarkerRefCount == 0) { + // Unexpected Release! This should not normally happen, but it's harmless in + // practice, it means the buffer is not alive anyway. + return; + } + + MOZ_ASSERT(sBufferForMainThreadAddMarker); + MOZ_ASSERT(sBufferForMainThreadAddMarker->IsInSession()); + if (--sBufferForMainThreadAddMarkerRefCount == 0) { + // Last `Release`, destroy the buffer. + delete sBufferForMainThreadAddMarker; + sBufferForMainThreadAddMarker = nullptr; + } +} + +} // namespace base_profiler_markers_detail + +void MarkerSchema::Stream(JSONWriter& aWriter, + const Span& aName) && { + // The caller should have started a JSON array, in which we can add an object + // that defines a marker schema. + + if (mLocations.empty()) { + // SpecialFrontendLocation case, don't output anything for this type. + return; + } + + aWriter.StartObjectElement(); + { + aWriter.StringProperty("name", aName); + + if (!mChartLabel.empty()) { + aWriter.StringProperty("chartLabel", mChartLabel); + } + + if (!mTooltipLabel.empty()) { + aWriter.StringProperty("tooltipLabel", mTooltipLabel); + } + + if (!mTableLabel.empty()) { + aWriter.StringProperty("tableLabel", mTableLabel); + } + + aWriter.StartArrayProperty("display"); + { + for (Location location : mLocations) { + aWriter.StringElement(LocationToStringSpan(location)); + } + } + aWriter.EndArray(); + + aWriter.StartArrayProperty("data"); + { + for (const DataRow& row : mData) { + aWriter.StartObjectElement(); + { + row.match( + [&aWriter](const DynamicData& aData) { + aWriter.StringProperty("key", aData.mKey); + if (aData.mLabel) { + aWriter.StringProperty("label", *aData.mLabel); + } + aWriter.StringProperty("format", + FormatToStringSpan(aData.mFormat)); + if (aData.mSearchable) { + aWriter.BoolProperty( + "searchable", + *aData.mSearchable == Searchable::Searchable); + } + }, + [&aWriter](const StaticData& aStaticData) { + aWriter.StringProperty("label", aStaticData.mLabel); + aWriter.StringProperty("value", aStaticData.mValue); + }); + } + aWriter.EndObject(); + } + } + aWriter.EndArray(); + } + aWriter.EndObject(); +} + +/* static */ +Span MarkerSchema::LocationToStringSpan( + MarkerSchema::Location aLocation) { + switch (aLocation) { + case Location::MarkerChart: + return mozilla::MakeStringSpan("marker-chart"); + case Location::MarkerTable: + return mozilla::MakeStringSpan("marker-table"); + case Location::TimelineOverview: + return mozilla::MakeStringSpan("timeline-overview"); + case Location::TimelineMemory: + return mozilla::MakeStringSpan("timeline-memory"); + case Location::TimelineIPC: + return mozilla::MakeStringSpan("timeline-ipc"); + case Location::TimelineFileIO: + return mozilla::MakeStringSpan("timeline-fileio"); + case Location::StackChart: + return mozilla::MakeStringSpan("stack-chart"); + default: + MOZ_CRASH("Unexpected Location enum"); + return {}; + } +} + +/* static */ +Span MarkerSchema::FormatToStringSpan( + MarkerSchema::Format aFormat) { + switch (aFormat) { + case Format::Url: + return mozilla::MakeStringSpan("url"); + case Format::FilePath: + return mozilla::MakeStringSpan("file-path"); + case Format::String: + return mozilla::MakeStringSpan("string"); + case Format::Duration: + return mozilla::MakeStringSpan("duration"); + case Format::Time: + return mozilla::MakeStringSpan("time"); + case Format::Seconds: + return mozilla::MakeStringSpan("seconds"); + case Format::Milliseconds: + return mozilla::MakeStringSpan("milliseconds"); + case Format::Microseconds: + return mozilla::MakeStringSpan("microseconds"); + case Format::Nanoseconds: + return mozilla::MakeStringSpan("nanoseconds"); + case Format::Bytes: + return mozilla::MakeStringSpan("bytes"); + case Format::Percentage: + return mozilla::MakeStringSpan("percentage"); + case Format::Integer: + return mozilla::MakeStringSpan("integer"); + case Format::Decimal: + return mozilla::MakeStringSpan("decimal"); + default: + MOZ_CRASH("Unexpected Format enum"); + return {}; + } +} + +} // namespace mozilla + +namespace mozilla::baseprofiler { +template MFBT_API ProfileBufferBlockIndex AddMarker(const ProfilerString8View&, + const MarkerCategory&, + MarkerOptions&&, + markers::TextMarker, + const std::string&); + +template MFBT_API ProfileBufferBlockIndex +AddMarkerToBuffer(ProfileChunkedBuffer&, const ProfilerString8View&, + const MarkerCategory&, MarkerOptions&&, markers::NoPayload); + +template MFBT_API ProfileBufferBlockIndex AddMarkerToBuffer( + ProfileChunkedBuffer&, const ProfilerString8View&, const MarkerCategory&, + MarkerOptions&&, markers::TextMarker, const std::string&); +} // namespace mozilla::baseprofiler diff --git a/mozglue/baseprofiler/core/ProfilerUtils.cpp b/mozglue/baseprofiler/core/ProfilerUtils.cpp new file mode 100644 index 0000000000..5f53910774 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilerUtils.cpp @@ -0,0 +1,162 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// This file implements functions from BaseProfilerUtils.h on all platforms. +// Functions with platform-specific implementations are separated in #if blocks +// below, with each block being self-contained with all the #includes and +// definitions it needs, to keep platform code easier to maintain in isolation. + +#include "mozilla/BaseProfilerUtils.h" + +// --------------------------------------------- WASI process & thread ids +#if defined(__wasi__) + +namespace mozilla::baseprofiler { + +// WASI is single-process and single-thread for now. + +BaseProfilerProcessId profiler_current_process_id() { + return BaseProfilerProcessId::FromNativeId(1u); +} + +BaseProfilerThreadId profiler_current_thread_id() { + return BaseProfilerThreadId::FromNativeId(1u); +} + +} // namespace mozilla::baseprofiler + +// --------------------------------------------- Windows process & thread ids +#elif defined(XP_WIN) + +# include +# include + +namespace mozilla::baseprofiler { + +BaseProfilerProcessId profiler_current_process_id() { + return BaseProfilerProcessId::FromNativeId(_getpid()); +} + +BaseProfilerThreadId profiler_current_thread_id() { + static_assert(std::is_same_v, + "BaseProfilerThreadId::NativeType must be exactly the type " + "returned by GetCurrentThreadId()"); + return BaseProfilerThreadId::FromNativeId(GetCurrentThreadId()); +} + +} // namespace mozilla::baseprofiler + +// --------------------------------------------- Non-Windows process id +#else +// All non-Windows platforms are assumed to be POSIX, which has getpid(). + +# include + +namespace mozilla::baseprofiler { + +BaseProfilerProcessId profiler_current_process_id() { + return BaseProfilerProcessId::FromNativeId(getpid()); +} + +} // namespace mozilla::baseprofiler + +// --------------------------------------------- Non-Windows thread id +// ------------------------------------------------------- macOS +# if defined(XP_MACOSX) + +# include + +namespace mozilla::baseprofiler { + +BaseProfilerThreadId profiler_current_thread_id() { + uint64_t tid; + if (pthread_threadid_np(nullptr, &tid) != 0) { + return BaseProfilerThreadId{}; + } + return BaseProfilerThreadId::FromNativeId(tid); +} + +} // namespace mozilla::baseprofiler + +// ------------------------------------------------------- Android +// Test Android before Linux, because Linux includes Android. +# elif defined(__ANDROID__) || defined(ANDROID) + +namespace mozilla::baseprofiler { + +BaseProfilerThreadId profiler_current_thread_id() { + return BaseProfilerThreadId::FromNativeId(gettid()); +} + +} // namespace mozilla::baseprofiler + +// ------------------------------------------------------- Linux +# elif defined(XP_LINUX) + +# include + +namespace mozilla::baseprofiler { + +BaseProfilerThreadId profiler_current_thread_id() { + // glibc doesn't provide a wrapper for gettid() until 2.30 + return BaseProfilerThreadId::FromNativeId(syscall(SYS_gettid)); +} + +} // namespace mozilla::baseprofiler + +// ------------------------------------------------------- FreeBSD +# elif defined(XP_FREEBSD) + +# include + +namespace mozilla::baseprofiler { + +BaseProfilerThreadId profiler_current_thread_id() { + long id; + if (thr_self(&id) != 0) { + return BaseProfilerThreadId{}; + } + return BaseProfilerThreadId::FromNativeId(id); +} + +} // namespace mozilla::baseprofiler + +// ------------------------------------------------------- Others +# else + +namespace mozilla::baseprofiler { + +BaseProfilerThreadId profiler_current_thread_id() { + return BaseProfilerThreadId::FromNativeId(std::this_thread::get_id()); +} + +} // namespace mozilla::baseprofiler + +# endif +#endif // End of non-XP_WIN. + +// --------------------------------------------- Platform-agnostic definitions + +namespace mozilla::baseprofiler { + +static BaseProfilerThreadId scBaseProfilerMainThreadId{}; + +void profiler_init_main_thread_id() { + if (!scBaseProfilerMainThreadId.IsSpecified()) { + scBaseProfilerMainThreadId = profiler_current_thread_id(); + } +} + +BaseProfilerThreadId profiler_main_thread_id() { + return scBaseProfilerMainThreadId; +} + +bool profiler_is_main_thread() { + return profiler_current_thread_id() == scBaseProfilerMainThreadId; +} + +} // namespace mozilla::baseprofiler diff --git a/mozglue/baseprofiler/core/ProfilingCategory.cpp b/mozglue/baseprofiler/core/ProfilingCategory.cpp new file mode 100644 index 0000000000..8ff2b15555 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilingCategory.cpp @@ -0,0 +1,71 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "BaseProfilingCategory.h" + +#include "mozilla/ArrayUtils.h" +#include "mozilla/Assertions.h" + +#include "BaseProfiler.h" + +namespace mozilla { +namespace baseprofiler { + +// clang-format off + +// ProfilingSubcategory_X: +// One enum for each category X, listing that category's subcategories. This +// allows the sProfilingCategoryInfo macro construction below to look up a +// per-category index for a subcategory. +#define SUBCATEGORY_ENUMS_BEGIN_CATEGORY(name, labelAsString, color) \ + enum class ProfilingSubcategory_##name : uint32_t { +#define SUBCATEGORY_ENUMS_SUBCATEGORY(category, name, labelAsString) \ + name, +#define SUBCATEGORY_ENUMS_END_CATEGORY \ + }; +MOZ_PROFILING_CATEGORY_LIST(SUBCATEGORY_ENUMS_BEGIN_CATEGORY, + SUBCATEGORY_ENUMS_SUBCATEGORY, + SUBCATEGORY_ENUMS_END_CATEGORY) +#undef SUBCATEGORY_ENUMS_BEGIN_CATEGORY +#undef SUBCATEGORY_ENUMS_SUBCATEGORY +#undef SUBCATEGORY_ENUMS_END_CATEGORY + +// sProfilingCategoryPairInfo: +// A list of ProfilingCategoryPairInfos with the same order as +// ProfilingCategoryPair, which can be used to map a ProfilingCategoryPair to +// its information. +#define CATEGORY_INFO_BEGIN_CATEGORY(name, labelAsString, color) +#define CATEGORY_INFO_SUBCATEGORY(category, name, labelAsString) \ + {ProfilingCategory::category, \ + uint32_t(ProfilingSubcategory_##category::name), labelAsString}, +#define CATEGORY_INFO_END_CATEGORY +const ProfilingCategoryPairInfo sProfilingCategoryPairInfo[] = { + MOZ_PROFILING_CATEGORY_LIST(CATEGORY_INFO_BEGIN_CATEGORY, + CATEGORY_INFO_SUBCATEGORY, + CATEGORY_INFO_END_CATEGORY) +}; +#undef CATEGORY_INFO_BEGIN_CATEGORY +#undef CATEGORY_INFO_SUBCATEGORY +#undef CATEGORY_INFO_END_CATEGORY + +// clang-format on + +const ProfilingCategoryPairInfo& GetProfilingCategoryPairInfo( + ProfilingCategoryPair aCategoryPair) { + static_assert( + MOZ_ARRAY_LENGTH(sProfilingCategoryPairInfo) == + uint32_t(ProfilingCategoryPair::COUNT), + "sProfilingCategoryPairInfo and ProfilingCategory need to have the " + "same order and the same length"); + + uint32_t categoryPairIndex = uint32_t(aCategoryPair); + MOZ_RELEASE_ASSERT(categoryPairIndex <= + uint32_t(ProfilingCategoryPair::LAST)); + return sProfilingCategoryPairInfo[categoryPairIndex]; +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/ProfilingStack.cpp b/mozglue/baseprofiler/core/ProfilingStack.cpp new file mode 100644 index 0000000000..f5cd2ddd04 --- /dev/null +++ b/mozglue/baseprofiler/core/ProfilingStack.cpp @@ -0,0 +1,52 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=80: + * This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "BaseProfilingStack.h" + +#include + +#include "mozilla/IntegerRange.h" +#include "mozilla/UniquePtr.h" +#include "mozilla/UniquePtrExtensions.h" + +#include "BaseProfiler.h" + +namespace mozilla { +namespace baseprofiler { + +ProfilingStack::~ProfilingStack() { + // The label macros keep a reference to the ProfilingStack to avoid a TLS + // access. If these are somehow not all cleared we will get a + // use-after-free so better to crash now. + MOZ_RELEASE_ASSERT(stackPointer == 0); + + delete[] frames; +} + +void ProfilingStack::ensureCapacitySlow() { + MOZ_ASSERT(stackPointer >= capacity); + const uint32_t kInitialCapacity = 128; + + uint32_t sp = stackPointer; + auto newCapacity = + std::max(sp + 1, capacity ? capacity * 2 : kInitialCapacity); + + auto* newFrames = new ProfilingStackFrame[newCapacity]; + + // It's important that `frames` / `capacity` / `stackPointer` remain + // consistent here at all times. + for (auto i : IntegerRange(capacity)) { + newFrames[i] = frames[i]; + } + + ProfilingStackFrame* oldFrames = frames; + frames = newFrames; + capacity = newCapacity; + delete[] oldFrames; +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/RegisteredThread.cpp b/mozglue/baseprofiler/core/RegisteredThread.cpp new file mode 100644 index 0000000000..85a7fc2c6d --- /dev/null +++ b/mozglue/baseprofiler/core/RegisteredThread.cpp @@ -0,0 +1,42 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "RegisteredThread.h" + +#include "BaseProfiler.h" + +namespace mozilla { +namespace baseprofiler { + +RegisteredThread::RegisteredThread(ThreadInfo* aInfo, void* aStackTop) + : mRacyRegisteredThread(aInfo->ThreadId()), + mPlatformData(AllocPlatformData(aInfo->ThreadId())), + mStackTop(aStackTop), + mThreadInfo(aInfo) { + // We don't have to guess on mac +#if defined(GP_OS_darwin) + pthread_t self = pthread_self(); + mStackTop = pthread_get_stackaddr_np(self); +#endif +} + +RegisteredThread::~RegisteredThread() {} + +size_t RegisteredThread::SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + size_t n = aMallocSizeOf(this); + + // Measurement of the following members may be added later if DMD finds it + // is worthwhile: + // - mPlatformData + // + // The following members are not measured: + // - mThreadInfo: because it is non-owning + + return n; +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/RegisteredThread.h b/mozglue/baseprofiler/core/RegisteredThread.h new file mode 100644 index 0000000000..baaf7dced6 --- /dev/null +++ b/mozglue/baseprofiler/core/RegisteredThread.h @@ -0,0 +1,164 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef RegisteredThread_h +#define RegisteredThread_h + +#include "platform.h" +#include "ThreadInfo.h" + +#include "mozilla/UniquePtr.h" + +namespace mozilla { +namespace baseprofiler { + +// This class contains the state for a single thread that is accessible without +// protection from gPSMutex in platform.cpp. Because there is no external +// protection against data races, it must provide internal protection. Hence +// the "Racy" prefix. +// +class RacyRegisteredThread final { + public: + explicit RacyRegisteredThread(BaseProfilerThreadId aThreadId) + : mThreadId(aThreadId), mSleep(AWAKE), mIsBeingProfiled(false) {} + + ~RacyRegisteredThread() {} + + void SetIsBeingProfiled(bool aIsBeingProfiled) { + mIsBeingProfiled = aIsBeingProfiled; + } + + bool IsBeingProfiled() const { return mIsBeingProfiled; } + + // This is called on every profiler restart. Put things that should happen at + // that time here. + void ReinitializeOnResume() { + // This is needed to cause an initial sample to be taken from sleeping + // threads that had been observed prior to the profiler stopping and + // restarting. Otherwise sleeping threads would not have any samples to + // copy forward while sleeping. + (void)mSleep.compareExchange(SLEEPING_OBSERVED, SLEEPING_NOT_OBSERVED); + } + + // This returns true for the second and subsequent calls in each sleep cycle. + bool CanDuplicateLastSampleDueToSleep() { + if (mSleep == AWAKE) { + return false; + } + + if (mSleep.compareExchange(SLEEPING_NOT_OBSERVED, SLEEPING_OBSERVED)) { + return false; + } + + return true; + } + + // Call this whenever the current thread sleeps. Calling it twice in a row + // without an intervening setAwake() call is an error. + void SetSleeping() { + MOZ_ASSERT(mSleep == AWAKE); + mSleep = SLEEPING_NOT_OBSERVED; + } + + // Call this whenever the current thread wakes. Calling it twice in a row + // without an intervening setSleeping() call is an error. + void SetAwake() { + MOZ_ASSERT(mSleep != AWAKE); + mSleep = AWAKE; + } + + bool IsSleeping() { return mSleep != AWAKE; } + + BaseProfilerThreadId ThreadId() const { return mThreadId; } + + class ProfilingStack& ProfilingStack() { return mProfilingStack; } + const class ProfilingStack& ProfilingStack() const { return mProfilingStack; } + + private: + class ProfilingStack mProfilingStack; + + // mThreadId contains the thread ID of the current thread. It is safe to read + // this from multiple threads concurrently, as it will never be mutated. + const BaseProfilerThreadId mThreadId; + + // mSleep tracks whether the thread is sleeping, and if so, whether it has + // been previously observed. This is used for an optimization: in some cases, + // when a thread is asleep, we duplicate the previous sample, which is + // cheaper than taking a new sample. + // + // mSleep is atomic because it is accessed from multiple threads. + // + // - It is written only by this thread, via setSleeping() and setAwake(). + // + // - It is read by SamplerThread::Run(). + // + // There are two cases where racing between threads can cause an issue. + // + // - If CanDuplicateLastSampleDueToSleep() returns false but that result is + // invalidated before being acted upon, we will take a full sample + // unnecessarily. This is additional work but won't cause any correctness + // issues. (In actual fact, this case is impossible. In order to go from + // CanDuplicateLastSampleDueToSleep() returning false to it returning true + // requires an intermediate call to it in order for mSleep to go from + // SLEEPING_NOT_OBSERVED to SLEEPING_OBSERVED.) + // + // - If CanDuplicateLastSampleDueToSleep() returns true but that result is + // invalidated before being acted upon -- i.e. the thread wakes up before + // DuplicateLastSample() is called -- we will duplicate the previous + // sample. This is inaccurate, but only slightly... we will effectively + // treat the thread as having slept a tiny bit longer than it really did. + // + // This latter inaccuracy could be avoided by moving the + // CanDuplicateLastSampleDueToSleep() check within the thread-freezing code, + // e.g. the section where Tick() is called. But that would reduce the + // effectiveness of the optimization because more code would have to be run + // before we can tell that duplication is allowed. + // + static const int AWAKE = 0; + static const int SLEEPING_NOT_OBSERVED = 1; + static const int SLEEPING_OBSERVED = 2; + Atomic mSleep; + + // Is this thread being profiled? (e.g., should markers be recorded?) + Atomic mIsBeingProfiled; +}; + +// This class contains information that's relevant to a single thread only +// while that thread is running and registered with the profiler, but +// regardless of whether the profiler is running. All accesses to it are +// protected by the profiler state lock. +class RegisteredThread final { + public: + RegisteredThread(ThreadInfo* aInfo, void* aStackTop); + ~RegisteredThread(); + + class RacyRegisteredThread& RacyRegisteredThread() { + return mRacyRegisteredThread; + } + const class RacyRegisteredThread& RacyRegisteredThread() const { + return mRacyRegisteredThread; + } + + PlatformData* GetPlatformData() const { return mPlatformData.get(); } + const void* StackTop() const { return mStackTop; } + + size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const; + + const RefPtr Info() const { return mThreadInfo; } + + private: + class RacyRegisteredThread mRacyRegisteredThread; + + const UniquePlatformData mPlatformData; + const void* mStackTop; + + const RefPtr mThreadInfo; +}; + +} // namespace baseprofiler +} // namespace mozilla + +#endif // RegisteredThread_h diff --git a/mozglue/baseprofiler/core/ThreadInfo.h b/mozglue/baseprofiler/core/ThreadInfo.h new file mode 100644 index 0000000000..80211396ac --- /dev/null +++ b/mozglue/baseprofiler/core/ThreadInfo.h @@ -0,0 +1,62 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef ThreadInfo_h +#define ThreadInfo_h + +#include "mozilla/Atomics.h" +#include "mozilla/BaseProfilerUtils.h" +#include "mozilla/TimeStamp.h" + +namespace mozilla { +namespace baseprofiler { + +// This class contains information about a thread which needs to be stored +// across restarts of the profiler and which can be useful even after the +// thread has stopped running. +// It uses threadsafe refcounting and only contains immutable data. +class ThreadInfo final { + public: + ThreadInfo(const char* aName, BaseProfilerThreadId aThreadId, + bool aIsMainThread, + const TimeStamp& aRegisterTime = TimeStamp::Now()) + : mName(aName), + mRegisterTime(aRegisterTime), + mThreadId(aThreadId), + mIsMainThread(aIsMainThread), + mRefCnt(0) { + MOZ_ASSERT(aThreadId.IsSpecified(), + "Given aThreadId should not be unspecified"); + } + + // Using hand-rolled ref-counting, because RefCounted.h macros don't produce + // the same code between mozglue and libxul, see bug 1536656. + MFBT_API void AddRef() const { ++mRefCnt; } + MFBT_API void Release() const { + MOZ_ASSERT(int32_t(mRefCnt) > 0); + if (--mRefCnt == 0) { + delete this; + } + } + + const char* Name() const { return mName.c_str(); } + TimeStamp RegisterTime() const { return mRegisterTime; } + BaseProfilerThreadId ThreadId() const { return mThreadId; } + bool IsMainThread() const { return mIsMainThread; } + + private: + const std::string mName; + const TimeStamp mRegisterTime; + const BaseProfilerThreadId mThreadId; + const bool mIsMainThread; + + mutable Atomic mRefCnt; +}; + +} // namespace baseprofiler +} // namespace mozilla + +#endif // ThreadInfo_h diff --git a/mozglue/baseprofiler/core/VTuneProfiler.cpp b/mozglue/baseprofiler/core/VTuneProfiler.cpp new file mode 100644 index 0000000000..2911c39f08 --- /dev/null +++ b/mozglue/baseprofiler/core/VTuneProfiler.cpp @@ -0,0 +1,92 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifdef XP_WIN +# undef UNICODE +# undef _UNICODE +#endif + +#include "VTuneProfiler.h" + +#include + +#include "BaseProfiler.h" + +namespace mozilla { +namespace baseprofiler { + +VTuneProfiler* VTuneProfiler::mInstance = nullptr; + +void VTuneProfiler::Initialize() { + // This is just a 'dirty trick' to find out if the ittnotify DLL was found. + // If it wasn't this function always returns 0, otherwise it returns + // incrementing numbers, if the library was found this wastes 2 events but + // that should be okay. + // TODO re-implement here if vtune is needed + // __itt_event testEvent = + // __itt_event_create("Test event", strlen("Test event")); + // testEvent = __itt_event_create("Test event 2", strlen("Test event 2")); + + // if (testEvent) { + // mInstance = new VTuneProfiler(); + // } +} + +void VTuneProfiler::Shutdown() {} + +void VTuneProfiler::TraceInternal(const char* aName, TracingKind aKind) { + // TODO re-implement here if vtune is needed + // std::string str(aName); + + // auto iter = mStrings.find(str); + + // __itt_event event; + // if (iter != mStrings.end()) { + // event = iter->second; + // } else { + // event = __itt_event_create(aName, str.length()); + // mStrings.insert({str, event}); + // } + + // if (aKind == TRACING_INTERVAL_START || aKind == TRACING_EVENT) { + // // VTune will consider starts not matched with an end to be single point + // in + // // time events. + // __itt_event_start(event); + // } else { + // __itt_event_end(event); + // } +} + +void VTuneProfiler::RegisterThreadInternal(const char* aName) { + // TODO re-implement here if vtune is needed + // std::string str(aName); + + // if (!str.compare("Main Thread (Base Profiler)")) { + // // Process main thread. + // switch (XRE_GetProcessType()) { + // case GeckoProcessType::GeckoProcessType_Default: + // __itt_thread_set_name("Main Process"); + // break; + // case GeckoProcessType::GeckoProcessType_Content: + // __itt_thread_set_name("Content Process"); + // break; + // case GeckoProcessType::GeckoProcessType_GMPlugin: + // __itt_thread_set_name("Plugin Process"); + // break; + // case GeckoProcessType::GeckoProcessType_GPU: + // __itt_thread_set_name("GPU Process"); + // break; + // default: + // __itt_thread_set_name("Unknown Process"); + // } + // return; + // } + // __itt_thread_set_name(aName); +} + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/VTuneProfiler.h b/mozglue/baseprofiler/core/VTuneProfiler.h new file mode 100644 index 0000000000..cf94ab7242 --- /dev/null +++ b/mozglue/baseprofiler/core/VTuneProfiler.h @@ -0,0 +1,84 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#ifndef VTuneProfiler_h +#define VTuneProfiler_h + +// The intent here is to add 0 overhead for regular users. In order to build +// the VTune profiler code at all --enable-vtune-instrumentation needs to be +// set as a build option. Even then, when none of the environment variables +// is specified that allow us to find the ittnotify DLL, these functions +// should be minimal overhead. When starting Firefox under VTune, these +// env vars will be automatically defined, otherwise INTEL_LIBITTNOTIFY32/64 +// should be set to point at the ittnotify DLL. +#ifndef MOZ_VTUNE_INSTRUMENTATION + +# define VTUNE_INIT() +# define VTUNE_SHUTDOWN() + +# define VTUNE_TRACING(name, kind) +# define VTUNE_REGISTER_THREAD(name) + +#else + +# include "BaseProfiler.h" + +// This is the regular Intel header, these functions are actually defined for +// us inside js/src/vtune by an intel C file which actually dynamically resolves +// them to the correct DLL. Through libxul these will 'magically' resolve. +# include "vtune/ittnotify.h" + +# include +# include +# include + +namespace mozilla { +namespace baseprofiler { + +class VTuneProfiler { + public: + static void Initialize(); + static void Shutdown(); + + enum TracingKind { + TRACING_EVENT, + TRACING_INTERVAL_START, + TRACING_INTERVAL_END, + }; + + static void Trace(const char* aName, TracingKind aKind) { + if (mInstance) { + mInstance->TraceInternal(aName, aKind); + } + } + static void RegisterThread(const char* aName) { + if (mInstance) { + mInstance->RegisterThreadInternal(aName); + } + } + + private: + void TraceInternal(const char* aName, TracingKind aKind); + void RegisterThreadInternal(const char* aName); + + // This is null when the ittnotify DLL could not be found. + static VTuneProfiler* mInstance; + + std::unordered_map mStrings; +}; + +# define VTUNE_INIT() VTuneProfiler::Initialize() +# define VTUNE_SHUTDOWN() VTuneProfiler::Shutdown() + +# define VTUNE_TRACING(name, kind) VTuneProfiler::Trace(name, kind) +# define VTUNE_REGISTER_THREAD(name) VTuneProfiler::RegisterThread(name) + +} // namespace baseprofiler +} // namespace mozilla + +#endif + +#endif /* VTuneProfiler_h */ diff --git a/mozglue/baseprofiler/core/platform-linux-android.cpp b/mozglue/baseprofiler/core/platform-linux-android.cpp new file mode 100644 index 0000000000..7921d9f4fd --- /dev/null +++ b/mozglue/baseprofiler/core/platform-linux-android.cpp @@ -0,0 +1,507 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +// Copyright (c) 2006-2011 The Chromium Authors. All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions +// are met: +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * 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. +// * Neither the name of Google, Inc. nor the names of its contributors +// may be used to endorse or promote products derived from this +// software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "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 THE +// COPYRIGHT OWNER 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. + +// This file is used for both Linux and Android. + +#include +#include + +#include +#if defined(GP_OS_freebsd) +# include +#endif +#include +#include +#include +#include +#include +#include +#include +#include +#include +// Ubuntu Dapper requires memory pages to be marked as +// executable. Otherwise, OS raises an exception when executing code +// in that page. +#include // mmap & munmap +#include // mmap & munmap +#include // open +#include // open +#include // sysconf +#include +#ifdef __GLIBC__ +# include // backtrace, backtrace_symbols +#endif // def __GLIBC__ +#include // index +#include +#include + +#include "prenv.h" +#include "mozilla/PodOperations.h" +#include "mozilla/DebugOnly.h" + +#include +#include + +using namespace mozilla; + +namespace mozilla { +namespace baseprofiler { + +static int64_t MicrosecondsSince1970() { + struct timeval tv; + gettimeofday(&tv, NULL); + return int64_t(tv.tv_sec) * 1000000 + int64_t(tv.tv_usec); +} + +void* GetStackTop(void* aGuess) { return aGuess; } + +static void PopulateRegsFromContext(Registers& aRegs, ucontext_t* aContext) { + aRegs.mContext = aContext; + mcontext_t& mcontext = aContext->uc_mcontext; + + // Extracting the sample from the context is extremely machine dependent. +#if defined(GP_PLAT_x86_linux) || defined(GP_PLAT_x86_android) + aRegs.mPC = reinterpret_cast
(mcontext.gregs[REG_EIP]); + aRegs.mSP = reinterpret_cast
(mcontext.gregs[REG_ESP]); + aRegs.mFP = reinterpret_cast
(mcontext.gregs[REG_EBP]); + aRegs.mLR = 0; +#elif defined(GP_PLAT_amd64_linux) || defined(GP_PLAT_amd64_android) + aRegs.mPC = reinterpret_cast
(mcontext.gregs[REG_RIP]); + aRegs.mSP = reinterpret_cast
(mcontext.gregs[REG_RSP]); + aRegs.mFP = reinterpret_cast
(mcontext.gregs[REG_RBP]); + aRegs.mLR = 0; +#elif defined(GP_PLAT_amd64_freebsd) + aRegs.mPC = reinterpret_cast
(mcontext.mc_rip); + aRegs.mSP = reinterpret_cast
(mcontext.mc_rsp); + aRegs.mFP = reinterpret_cast
(mcontext.mc_rbp); + aRegs.mLR = 0; +#elif defined(GP_PLAT_arm_linux) || defined(GP_PLAT_arm_android) + aRegs.mPC = reinterpret_cast
(mcontext.arm_pc); + aRegs.mSP = reinterpret_cast
(mcontext.arm_sp); + aRegs.mFP = reinterpret_cast
(mcontext.arm_fp); + aRegs.mLR = reinterpret_cast
(mcontext.arm_lr); +#elif defined(GP_PLAT_arm64_linux) || defined(GP_PLAT_arm64_android) + aRegs.mPC = reinterpret_cast
(mcontext.pc); + aRegs.mSP = reinterpret_cast
(mcontext.sp); + aRegs.mFP = reinterpret_cast
(mcontext.regs[29]); + aRegs.mLR = reinterpret_cast
(mcontext.regs[30]); +#elif defined(GP_PLAT_arm64_freebsd) + aRegs.mPC = reinterpret_cast
(mcontext.mc_gpregs.gp_elr); + aRegs.mSP = reinterpret_cast
(mcontext.mc_gpregs.gp_sp); + aRegs.mFP = reinterpret_cast
(mcontext.mc_gpregs.gp_x[29]); + aRegs.mLR = reinterpret_cast
(mcontext.mc_gpregs.gp_lr); +#elif defined(GP_PLAT_mips64_linux) || defined(GP_PLAT_mips64_android) + aRegs.mPC = reinterpret_cast
(mcontext.pc); + aRegs.mSP = reinterpret_cast
(mcontext.gregs[29]); + aRegs.mFP = reinterpret_cast
(mcontext.gregs[30]); + +#else +# error "bad platform" +#endif +} + +#if defined(GP_OS_android) +# define SYS_tgkill __NR_tgkill +#endif + +#if defined(GP_OS_linux) || defined(GP_OS_android) +int tgkill(pid_t tgid, pid_t tid, int signalno) { + return syscall(SYS_tgkill, tgid, tid, signalno); +} +#endif + +#if defined(GP_OS_freebsd) +# define tgkill thr_kill2 +#endif + +class PlatformData { + public: + explicit PlatformData(BaseProfilerThreadId aThreadId) {} + + ~PlatformData() {} +}; + +//////////////////////////////////////////////////////////////////////// +// BEGIN Sampler target specifics + +// The only way to reliably interrupt a Linux thread and inspect its register +// and stack state is by sending a signal to it, and doing the work inside the +// signal handler. But we don't want to run much code inside the signal +// handler, since POSIX severely restricts what we can do in signal handlers. +// So we use a system of semaphores to suspend the thread and allow the +// sampler thread to do all the work of unwinding and copying out whatever +// data it wants. +// +// A four-message protocol is used to reliably suspend and later resume the +// thread to be sampled (the samplee): +// +// Sampler (signal sender) thread Samplee (thread to be sampled) +// +// Prepare the SigHandlerCoordinator +// and point sSigHandlerCoordinator at it +// +// send SIGPROF to samplee ------- MSG 1 ----> (enter signal handler) +// wait(mMessage2) Copy register state +// into sSigHandlerCoordinator +// <------ MSG 2 ----- post(mMessage2) +// Samplee is now suspended. wait(mMessage3) +// Examine its stack/register +// state at leisure +// +// Release samplee: +// post(mMessage3) ------- MSG 3 -----> +// wait(mMessage4) Samplee now resumes. Tell +// the sampler that we are done. +// <------ MSG 4 ------ post(mMessage4) +// Now we know the samplee's signal (leave signal handler) +// handler has finished using +// sSigHandlerCoordinator. We can +// safely reuse it for some other thread. +// + +// A type used to coordinate between the sampler (signal sending) thread and +// the thread currently being sampled (the samplee, which receives the +// signals). +// +// The first message is sent using a SIGPROF signal delivery. The subsequent +// three are sent using sem_wait/sem_post pairs. They are named accordingly +// in the following struct. +struct SigHandlerCoordinator { + SigHandlerCoordinator() { + PodZero(&mUContext); + int r = sem_init(&mMessage2, /* pshared */ 0, 0); + r |= sem_init(&mMessage3, /* pshared */ 0, 0); + r |= sem_init(&mMessage4, /* pshared */ 0, 0); + MOZ_ASSERT(r == 0); + (void)r; + } + + ~SigHandlerCoordinator() { + int r = sem_destroy(&mMessage2); + r |= sem_destroy(&mMessage3); + r |= sem_destroy(&mMessage4); + MOZ_ASSERT(r == 0); + (void)r; + } + + sem_t mMessage2; // To sampler: "context is in sSigHandlerCoordinator" + sem_t mMessage3; // To samplee: "resume" + sem_t mMessage4; // To sampler: "finished with sSigHandlerCoordinator" + ucontext_t mUContext; // Context at signal +}; + +struct SigHandlerCoordinator* Sampler::sSigHandlerCoordinator = nullptr; + +static void SigprofHandler(int aSignal, siginfo_t* aInfo, void* aContext) { + // Avoid TSan warning about clobbering errno. + int savedErrno = errno; + + MOZ_ASSERT(aSignal == SIGPROF); + MOZ_ASSERT(Sampler::sSigHandlerCoordinator); + + // By sending us this signal, the sampler thread has sent us message 1 in + // the comment above, with the meaning "|sSigHandlerCoordinator| is ready + // for use, please copy your register context into it." + Sampler::sSigHandlerCoordinator->mUContext = + *static_cast(aContext); + + // Send message 2: tell the sampler thread that the context has been copied + // into |sSigHandlerCoordinator->mUContext|. sem_post can never fail by + // being interrupted by a signal, so there's no loop around this call. + int r = sem_post(&Sampler::sSigHandlerCoordinator->mMessage2); + MOZ_ASSERT(r == 0); + + // At this point, the sampler thread assumes we are suspended, so we must + // not touch any global state here. + + // Wait for message 3: the sampler thread tells us to resume. + while (true) { + r = sem_wait(&Sampler::sSigHandlerCoordinator->mMessage3); + if (r == -1 && errno == EINTR) { + // Interrupted by a signal. Try again. + continue; + } + // We don't expect any other kind of failure + MOZ_ASSERT(r == 0); + break; + } + + // Send message 4: tell the sampler thread that we are finished accessing + // |sSigHandlerCoordinator|. After this point it is not safe to touch + // |sSigHandlerCoordinator|. + r = sem_post(&Sampler::sSigHandlerCoordinator->mMessage4); + MOZ_ASSERT(r == 0); + + errno = savedErrno; +} + +Sampler::Sampler(PSLockRef aLock) : mMyPid(profiler_current_process_id()) { +#if defined(USE_EHABI_STACKWALK) + EHABIStackWalkInit(); +#endif + + // NOTE: We don't initialize LUL here, instead initializing it in + // SamplerThread's constructor. This is because with the + // profiler_suspend_and_sample_thread entry point, we want to be able to + // sample without waiting for LUL to be initialized. + + // Request profiling signals. + struct sigaction sa; + sa.sa_sigaction = SigprofHandler; + sigemptyset(&sa.sa_mask); + sa.sa_flags = SA_RESTART | SA_SIGINFO; + if (sigaction(SIGPROF, &sa, &mOldSigprofHandler) != 0) { + MOZ_CRASH("Error installing SIGPROF handler in the profiler"); + } +} + +void Sampler::Disable(PSLockRef aLock) { + // Restore old signal handler. This is global state so it's important that + // we do it now, while gPSMutex is locked. + sigaction(SIGPROF, &mOldSigprofHandler, 0); +} + +template +void Sampler::SuspendAndSampleAndResumeThread( + PSLockRef aLock, const RegisteredThread& aRegisteredThread, + const TimeStamp& aNow, const Func& aProcessRegs) { + // Only one sampler thread can be sampling at once. So we expect to have + // complete control over |sSigHandlerCoordinator|. + MOZ_ASSERT(!sSigHandlerCoordinator); + + if (!mSamplerTid.IsSpecified()) { + mSamplerTid = profiler_current_thread_id(); + } + BaseProfilerThreadId sampleeTid = aRegisteredThread.Info()->ThreadId(); + MOZ_RELEASE_ASSERT(sampleeTid != mSamplerTid); + + //----------------------------------------------------------------// + // Suspend the samplee thread and get its context. + + SigHandlerCoordinator coord; // on sampler thread's stack + sSigHandlerCoordinator = &coord; + + // Send message 1 to the samplee (the thread to be sampled), by + // signalling at it. + // This could fail if the thread doesn't exist anymore. + int r = tgkill(mMyPid.ToNumber(), sampleeTid.ToNumber(), SIGPROF); + if (r == 0) { + // Wait for message 2 from the samplee, indicating that the context + // is available and that the thread is suspended. + while (true) { + r = sem_wait(&sSigHandlerCoordinator->mMessage2); + if (r == -1 && errno == EINTR) { + // Interrupted by a signal. Try again. + continue; + } + // We don't expect any other kind of failure. + MOZ_ASSERT(r == 0); + break; + } + + //----------------------------------------------------------------// + // Sample the target thread. + + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING + // + // The profiler's "critical section" begins here. In the critical section, + // we must not do any dynamic memory allocation, nor try to acquire any lock + // or any other unshareable resource. This is because the thread to be + // sampled has been suspended at some entirely arbitrary point, and we have + // no idea which unsharable resources (locks, essentially) it holds. So any + // attempt to acquire any lock, including the implied locks used by the + // malloc implementation, risks deadlock. This includes TimeStamp::Now(), + // which gets a lock on Windows. + + // The samplee thread is now frozen and sSigHandlerCoordinator->mUContext is + // valid. We can poke around in it and unwind its stack as we like. + + // Extract the current register values. + Registers regs; + PopulateRegsFromContext(regs, &sSigHandlerCoordinator->mUContext); + aProcessRegs(regs, aNow); + + //----------------------------------------------------------------// + // Resume the target thread. + + // Send message 3 to the samplee, which tells it to resume. + r = sem_post(&sSigHandlerCoordinator->mMessage3); + MOZ_ASSERT(r == 0); + + // Wait for message 4 from the samplee, which tells us that it has + // finished with |sSigHandlerCoordinator|. + while (true) { + r = sem_wait(&sSigHandlerCoordinator->mMessage4); + if (r == -1 && errno == EINTR) { + continue; + } + MOZ_ASSERT(r == 0); + break; + } + + // The profiler's critical section ends here. After this point, none of the + // critical section limitations documented above apply. + // + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING + } + + // This isn't strictly necessary, but doing so does help pick up anomalies + // in which the signal handler is running when it shouldn't be. + sSigHandlerCoordinator = nullptr; +} + +// END Sampler target specifics +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN SamplerThread target specifics + +static void* ThreadEntry(void* aArg) { + auto thread = static_cast(aArg); + thread->Run(); + return nullptr; +} + +SamplerThread::SamplerThread(PSLockRef aLock, uint32_t aActivityGeneration, + double aIntervalMilliseconds, uint32_t aFeatures) + : mSampler(aLock), + mActivityGeneration(aActivityGeneration), + mIntervalMicroseconds( + std::max(1, int(floor(aIntervalMilliseconds * 1000 + 0.5)))) { +#if defined(USE_LUL_STACKWALK) + lul::LUL* lul = CorePS::Lul(aLock); + if (!lul && ProfilerFeature::HasStackWalkEnabled(aFeatures)) { + CorePS::SetLul(aLock, MakeUnique(logging_sink_for_LUL)); + // Read all the unwind info currently available. + lul = CorePS::Lul(aLock); + read_procmaps(lul); + + // Switch into unwind mode. After this point, we can't add or remove any + // unwind info to/from this LUL instance. The only thing we can do with + // it is Unwind() calls. + lul->EnableUnwinding(); + + // Has a test been requested? + if (getenv("MOZ_PROFILER_LUL_TEST")) { + int nTests = 0, nTestsPassed = 0; + RunLulUnitTests(&nTests, &nTestsPassed, lul); + } + } +#endif + + // Start the sampling thread. It repeatedly sends a SIGPROF signal. Sending + // the signal ourselves instead of relying on itimer provides much better + // accuracy. + if (pthread_create(&mThread, nullptr, ThreadEntry, this) != 0) { + MOZ_CRASH("pthread_create failed"); + } +} + +SamplerThread::~SamplerThread() { pthread_join(mThread, nullptr); } + +void SamplerThread::SleepMicro(uint32_t aMicroseconds) { + if (aMicroseconds >= 1000000) { + // Use usleep for larger intervals, because the nanosleep + // code below only supports intervals < 1 second. + MOZ_ALWAYS_TRUE(!::usleep(aMicroseconds)); + return; + } + + struct timespec ts; + ts.tv_sec = 0; + ts.tv_nsec = aMicroseconds * 1000UL; + + int rv = ::nanosleep(&ts, &ts); + + while (rv != 0 && errno == EINTR) { + // Keep waiting in case of interrupt. + // nanosleep puts the remaining time back into ts. + rv = ::nanosleep(&ts, &ts); + } + + MOZ_ASSERT(!rv, "nanosleep call failed"); +} + +void SamplerThread::Stop(PSLockRef aLock) { + // Restore old signal handler. This is global state so it's important that + // we do it now, while gPSMutex is locked. It's safe to do this now even + // though this SamplerThread is still alive, because the next time the main + // loop of Run() iterates it won't get past the mActivityGeneration check, + // and so won't send any signals. + mSampler.Disable(aLock); +} + +// END SamplerThread target specifics +//////////////////////////////////////////////////////////////////////// + +#if defined(GP_OS_linux) || defined(GP_OS_freebsd) + +// We use pthread_atfork() to temporarily disable signal delivery during any +// fork() call. Without that, fork() can be repeatedly interrupted by signal +// delivery, requiring it to be repeatedly restarted, which can lead to *long* +// delays. See bug 837390. +// +// We provide no paf_child() function to run in the child after forking. This +// is fine because we always immediately exec() after fork(), and exec() +// clobbers all process state. Also, we don't want the sampler to resume in the +// child process between fork() and exec(), it would be wasteful. +// +// Unfortunately all this is only doable on non-Android because Bionic doesn't +// have pthread_atfork. + +// In the parent, before the fork, increase gSkipSampling to ensure that +// profiler sampling loops will be skipped. There could be one in progress now, +// causing a small delay, but further sampling will be skipped, allowing `fork` +// to complete. +static void paf_prepare() { ++gSkipSampling; } + +// In the parent, after the fork, decrease gSkipSampling to let the sampler +// resume sampling (unless other places have made it non-zero as well). +static void paf_parent() { --gSkipSampling; } + +static void PlatformInit(PSLockRef aLock) { + // Set up the fork handlers. + pthread_atfork(paf_prepare, paf_parent, nullptr); +} + +#else + +static void PlatformInit(PSLockRef aLock) {} + +#endif + +#if defined(HAVE_NATIVE_UNWIND) +// TODO port getcontext from breakpad, if profiler_get_backtrace is needed. +# define REGISTERS_SYNC_POPULATE(regs) \ + MOZ_CRASH("profiler_get_backtrace() unsupported"); +#endif + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/platform-macos.cpp b/mozglue/baseprofiler/core/platform-macos.cpp new file mode 100644 index 0000000000..6e6e801f1f --- /dev/null +++ b/mozglue/baseprofiler/core/platform-macos.cpp @@ -0,0 +1,220 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include +#include +#include +#include + +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +// this port is based off of v8 svn revision 9837 + +namespace mozilla { +namespace baseprofiler { + +static int64_t MicrosecondsSince1970() { + struct timeval tv; + gettimeofday(&tv, NULL); + return int64_t(tv.tv_sec) * 1000000 + int64_t(tv.tv_usec); +} + +void* GetStackTop(void* aGuess) { + pthread_t thread = pthread_self(); + return pthread_get_stackaddr_np(thread); +} + +class PlatformData { + public: + explicit PlatformData(BaseProfilerThreadId aThreadId) + : mProfiledThread(mach_thread_self()) {} + + ~PlatformData() { + // Deallocate Mach port for thread. + mach_port_deallocate(mach_task_self(), mProfiledThread); + } + + thread_act_t ProfiledThread() { return mProfiledThread; } + + private: + // Note: for mProfiledThread Mach primitives are used instead of pthread's + // because the latter doesn't provide thread manipulation primitives required. + // For details, consult "Mac OS X Internals" book, Section 7.3. + thread_act_t mProfiledThread; +}; + +//////////////////////////////////////////////////////////////////////// +// BEGIN Sampler target specifics + +Sampler::Sampler(PSLockRef aLock) {} + +void Sampler::Disable(PSLockRef aLock) {} + +template +void Sampler::SuspendAndSampleAndResumeThread( + PSLockRef aLock, const RegisteredThread& aRegisteredThread, + const TimeStamp& aNow, const Func& aProcessRegs) { + thread_act_t samplee_thread = + aRegisteredThread.GetPlatformData()->ProfiledThread(); + + //----------------------------------------------------------------// + // Suspend the samplee thread and get its context. + + // We're using thread_suspend on OS X because pthread_kill (which is what we + // at one time used on Linux) has less consistent performance and causes + // strange crashes, see bug 1166778 and bug 1166808. thread_suspend + // is also just a lot simpler to use. + + if (KERN_SUCCESS != thread_suspend(samplee_thread)) { + return; + } + + //----------------------------------------------------------------// + // Sample the target thread. + + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING + // + // The profiler's "critical section" begins here. We must be very careful + // what we do here, or risk deadlock. See the corresponding comment in + // platform-linux-android.cpp for details. + +#if defined(__x86_64__) + thread_state_flavor_t flavor = x86_THREAD_STATE64; + x86_thread_state64_t state; + mach_msg_type_number_t count = x86_THREAD_STATE64_COUNT; +# if __DARWIN_UNIX03 +# define REGISTER_FIELD(name) __r##name +# else +# define REGISTER_FIELD(name) r##name +# endif // __DARWIN_UNIX03 +#elif defined(__aarch64__) + thread_state_flavor_t flavor = ARM_THREAD_STATE64; + arm_thread_state64_t state; + mach_msg_type_number_t count = ARM_THREAD_STATE64_COUNT; +# if __DARWIN_UNIX03 +# define REGISTER_FIELD(name) __##name +# else +# define REGISTER_FIELD(name) name +# endif // __DARWIN_UNIX03 +#else +# error "unknown architecture" +#endif + + if (thread_get_state(samplee_thread, flavor, + reinterpret_cast(&state), + &count) == KERN_SUCCESS) { + Registers regs; +#if defined(__x86_64__) + regs.mPC = reinterpret_cast
(state.REGISTER_FIELD(ip)); + regs.mSP = reinterpret_cast
(state.REGISTER_FIELD(sp)); + regs.mFP = reinterpret_cast
(state.REGISTER_FIELD(bp)); + regs.mLR = 0; +#elif defined(__aarch64__) + regs.mPC = reinterpret_cast
(state.REGISTER_FIELD(pc)); + regs.mSP = reinterpret_cast
(state.REGISTER_FIELD(sp)); + regs.mFP = reinterpret_cast
(state.REGISTER_FIELD(fp)); + regs.mLR = reinterpret_cast
(state.REGISTER_FIELD(lr)); +#else +# error "unknown architecture" +#endif + + aProcessRegs(regs, aNow); + } + +#undef REGISTER_FIELD + + //----------------------------------------------------------------// + // Resume the target thread. + + thread_resume(samplee_thread); + + // The profiler's critical section ends here. + // + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING +} + +// END Sampler target specifics +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN SamplerThread target specifics + +static void* ThreadEntry(void* aArg) { + auto thread = static_cast(aArg); + thread->Run(); + return nullptr; +} + +SamplerThread::SamplerThread(PSLockRef aLock, uint32_t aActivityGeneration, + double aIntervalMilliseconds, uint32_t aFeatures) + : mSampler(aLock), + mActivityGeneration(aActivityGeneration), + mIntervalMicroseconds( + std::max(1, int(floor(aIntervalMilliseconds * 1000 + 0.5)))), + mThread{nullptr} { + pthread_attr_t* attr_ptr = nullptr; + if (pthread_create(&mThread, attr_ptr, ThreadEntry, this) != 0) { + MOZ_CRASH("pthread_create failed"); + } +} + +SamplerThread::~SamplerThread() { pthread_join(mThread, nullptr); } + +void SamplerThread::SleepMicro(uint32_t aMicroseconds) { + usleep(aMicroseconds); + // FIXME: the OSX 10.12 page for usleep says "The usleep() function is + // obsolescent. Use nanosleep(2) instead." This implementation could be + // merged with the linux-android version. Also, this doesn't handle the + // case where the usleep call is interrupted by a signal. +} + +void SamplerThread::Stop(PSLockRef aLock) { mSampler.Disable(aLock); } + +// END SamplerThread target specifics +//////////////////////////////////////////////////////////////////////// + +static void PlatformInit(PSLockRef aLock) {} + +// clang-format off +#if defined(HAVE_NATIVE_UNWIND) +// Derive the stack pointer from the frame pointer. The 0x10 offset is +// 8 bytes for the previous frame pointer and 8 bytes for the return +// address both stored on the stack after at the beginning of the current +// frame. +# define REGISTERS_SYNC_POPULATE(regs) \ + regs.mSP = reinterpret_cast
(__builtin_frame_address(0)) + 0x10; \ + _Pragma("GCC diagnostic push") \ + _Pragma("GCC diagnostic ignored \"-Wframe-address\"") \ + regs.mFP = reinterpret_cast
(__builtin_frame_address(1)); \ + _Pragma("GCC diagnostic pop") \ + regs.mPC = reinterpret_cast
( \ + __builtin_extract_return_addr(__builtin_return_address(0))); \ + regs.mLR = 0; +#endif +// clang-format on + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/platform-win32.cpp b/mozglue/baseprofiler/core/platform-win32.cpp new file mode 100644 index 0000000000..808bd9d618 --- /dev/null +++ b/mozglue/baseprofiler/core/platform-win32.cpp @@ -0,0 +1,348 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +// Copyright (c) 2006-2011 The Chromium Authors. All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions +// are met: +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * 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. +// * Neither the name of Google, Inc. nor the names of its contributors +// may be used to endorse or promote products derived from this +// software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "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 THE +// COPYRIGHT OWNER 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 +#include +#include + +#include "nsWindowsDllInterceptor.h" +#include "mozilla/StackWalk_windows.h" +#include "mozilla/WindowsVersion.h" + +namespace mozilla { +namespace baseprofiler { + +static int64_t MicrosecondsSince1970() { + int64_t prt; + FILETIME ft; + SYSTEMTIME st; + + GetSystemTime(&st); + SystemTimeToFileTime(&st, &ft); + static_assert(sizeof(ft) == sizeof(prt), "Expect FILETIME to be 64 bits"); + memcpy(&prt, &ft, sizeof(prt)); + const int64_t epochBias = 116444736000000000LL; + prt = (prt - epochBias) / 10; + + return prt; +} + +void* GetStackTop(void* aGuess) { + PNT_TIB pTib = reinterpret_cast(NtCurrentTeb()); + return reinterpret_cast(pTib->StackBase); +} + +static void PopulateRegsFromContext(Registers& aRegs, CONTEXT* aContext) { +#if defined(GP_ARCH_amd64) + aRegs.mPC = reinterpret_cast
(aContext->Rip); + aRegs.mSP = reinterpret_cast
(aContext->Rsp); + aRegs.mFP = reinterpret_cast
(aContext->Rbp); + aRegs.mLR = 0; +#elif defined(GP_ARCH_x86) + aRegs.mPC = reinterpret_cast
(aContext->Eip); + aRegs.mSP = reinterpret_cast
(aContext->Esp); + aRegs.mFP = reinterpret_cast
(aContext->Ebp); + aRegs.mLR = 0; +#elif defined(GP_ARCH_arm64) + aRegs.mPC = reinterpret_cast
(aContext->Pc); + aRegs.mSP = reinterpret_cast
(aContext->Sp); + aRegs.mFP = reinterpret_cast
(aContext->Fp); + aRegs.mLR = reinterpret_cast
(aContext->Lr); +#else +# error "bad arch" +#endif +} + +// Gets a real (i.e. not pseudo) handle for the current thread, with the +// permissions needed for profiling. +// @return a real HANDLE for the current thread. +static HANDLE GetRealCurrentThreadHandleForProfiling() { + HANDLE realCurrentThreadHandle; + if (!::DuplicateHandle( + ::GetCurrentProcess(), ::GetCurrentThread(), ::GetCurrentProcess(), + &realCurrentThreadHandle, + THREAD_GET_CONTEXT | THREAD_SUSPEND_RESUME | THREAD_QUERY_INFORMATION, + FALSE, 0)) { + return nullptr; + } + + return realCurrentThreadHandle; +} + +class PlatformData { + public: + // Get a handle to the calling thread. This is the thread that we are + // going to profile. We need a real handle because we are going to use it in + // the sampler thread. + explicit PlatformData(BaseProfilerThreadId aThreadId) + : mProfiledThread(GetRealCurrentThreadHandleForProfiling()) { + MOZ_ASSERT(DWORD(aThreadId.ToNumber()) == ::GetCurrentThreadId()); + } + + ~PlatformData() { + if (mProfiledThread != nullptr) { + CloseHandle(mProfiledThread); + mProfiledThread = nullptr; + } + } + + HANDLE ProfiledThread() { return mProfiledThread; } + + private: + HANDLE mProfiledThread; +}; + +#if defined(USE_MOZ_STACK_WALK) +HANDLE +GetThreadHandle(PlatformData* aData) { return aData->ProfiledThread(); } +#endif + +static const HANDLE kNoThread = INVALID_HANDLE_VALUE; + +//////////////////////////////////////////////////////////////////////// +// BEGIN Sampler target specifics + +Sampler::Sampler(PSLockRef aLock) {} + +void Sampler::Disable(PSLockRef aLock) {} + +template +void Sampler::SuspendAndSampleAndResumeThread( + PSLockRef aLock, const RegisteredThread& aRegisteredThread, + const TimeStamp& aNow, const Func& aProcessRegs) { + HANDLE profiled_thread = + aRegisteredThread.GetPlatformData()->ProfiledThread(); + if (profiled_thread == nullptr) { + return; + } + + // Context used for sampling the register state of the profiled thread. + CONTEXT context; + memset(&context, 0, sizeof(context)); + + //----------------------------------------------------------------// + // Suspend the samplee thread and get its context. + + static const DWORD kSuspendFailed = static_cast(-1); + if (SuspendThread(profiled_thread) == kSuspendFailed) { + return; + } + + // SuspendThread is asynchronous, so the thread may still be running. + // Call GetThreadContext first to ensure the thread is really suspended. + // See https://blogs.msdn.microsoft.com/oldnewthing/20150205-00/?p=44743. + + // Using only CONTEXT_CONTROL is faster but on 64-bit it causes crashes in + // RtlVirtualUnwind (see bug 1120126) so we set all the flags. +#if defined(GP_ARCH_amd64) + context.ContextFlags = CONTEXT_FULL; +#else + context.ContextFlags = CONTEXT_CONTROL; +#endif + if (!GetThreadContext(profiled_thread, &context)) { + ResumeThread(profiled_thread); + return; + } + + //----------------------------------------------------------------// + // Sample the target thread. + + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING + // + // The profiler's "critical section" begins here. We must be very careful + // what we do here, or risk deadlock. See the corresponding comment in + // platform-linux-android.cpp for details. + + Registers regs; + PopulateRegsFromContext(regs, &context); + aProcessRegs(regs, aNow); + + //----------------------------------------------------------------// + // Resume the target thread. + + ResumeThread(profiled_thread); + + // The profiler's critical section ends here. + // + // WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING +} + +// END Sampler target specifics +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN SamplerThread target specifics + +static unsigned int __stdcall ThreadEntry(void* aArg) { + auto thread = static_cast(aArg); + thread->Run(); + return 0; +} + +SamplerThread::SamplerThread(PSLockRef aLock, uint32_t aActivityGeneration, + double aIntervalMilliseconds, uint32_t aFeatures) + : mSampler(aLock), + mActivityGeneration(aActivityGeneration), + mIntervalMicroseconds( + std::max(1, int(floor(aIntervalMilliseconds * 1000 + 0.5)))), + mNoTimerResolutionChange( + ProfilerFeature::HasNoTimerResolutionChange(aFeatures)) { + if ((!mNoTimerResolutionChange) && (mIntervalMicroseconds < 10 * 1000)) { + // By default the timer resolution (which tends to be 1/64Hz, around 16ms) + // is not changed. However, if the requested interval is sufficiently low, + // the resolution will be adjusted to match. Note that this affects all + // timers in Firefox, and could therefore hide issues while profiling. This + // change may be prevented with the "notimerresolutionchange" feature. + ::timeBeginPeriod(mIntervalMicroseconds / 1000); + } + + // Create a new thread. It is important to use _beginthreadex() instead of + // the Win32 function CreateThread(), because the CreateThread() does not + // initialize thread-specific structures in the C runtime library. + mThread = reinterpret_cast(_beginthreadex(nullptr, + /* stack_size */ 0, + ThreadEntry, this, + /* initflag */ 0, nullptr)); + if (mThread == 0) { + MOZ_CRASH("_beginthreadex failed"); + } +} + +SamplerThread::~SamplerThread() { + WaitForSingleObject(mThread, INFINITE); + + // Close our own handle for the thread. + if (mThread != kNoThread) { + CloseHandle(mThread); + } +} + +void SamplerThread::SleepMicro(uint32_t aMicroseconds) { + // For now, keep the old behaviour of minimum Sleep(1), even for + // smaller-than-usual sleeps after an overshoot, unless the user has + // explicitly opted into a sub-millisecond profiler interval. + if (mIntervalMicroseconds >= 1000) { + ::Sleep(std::max(1u, aMicroseconds / 1000)); + } else { + TimeStamp start = TimeStamp::Now(); + TimeStamp end = start + TimeDuration::FromMicroseconds(aMicroseconds); + + // First, sleep for as many whole milliseconds as possible. + if (aMicroseconds >= 1000) { + ::Sleep(aMicroseconds / 1000); + } + + // Then, spin until enough time has passed. + while (TimeStamp::Now() < end) { + YieldProcessor(); + } + } +} + +void SamplerThread::Stop(PSLockRef aLock) { + if ((!mNoTimerResolutionChange) && (mIntervalMicroseconds < 10 * 1000)) { + // Disable any timer resolution changes we've made. Do it now while + // gPSMutex is locked, i.e. before any other SamplerThread can be created + // and call ::timeBeginPeriod(). + // + // It's safe to do this now even though this SamplerThread is still alive, + // because the next time the main loop of Run() iterates it won't get past + // the mActivityGeneration check, and so it won't make any more ::Sleep() + // calls. + ::timeEndPeriod(mIntervalMicroseconds / 1000); + } + + mSampler.Disable(aLock); +} + +// END SamplerThread target specifics +//////////////////////////////////////////////////////////////////////// + +static void PlatformInit(PSLockRef aLock) {} + +#if defined(HAVE_NATIVE_UNWIND) +# define REGISTERS_SYNC_POPULATE(regs) \ + CONTEXT context; \ + RtlCaptureContext(&context); \ + PopulateRegsFromContext(regs, &context); +#endif + +#if defined(GP_PLAT_amd64_windows) +static WindowsDllInterceptor NtDllIntercept; + +typedef NTSTATUS(NTAPI* LdrUnloadDll_func)(HMODULE module); +static WindowsDllInterceptor::FuncHookType stub_LdrUnloadDll; + +static NTSTATUS NTAPI patched_LdrUnloadDll(HMODULE module) { + // Prevent the stack walker from suspending this thread when LdrUnloadDll + // holds the RtlLookupFunctionEntry lock. + AutoSuppressStackWalking suppress; + return stub_LdrUnloadDll(module); +} + +// These pointers are disguised as PVOID to avoid pulling in obscure headers +typedef PVOID(WINAPI* LdrResolveDelayLoadedAPI_func)( + PVOID ParentModuleBase, PVOID DelayloadDescriptor, PVOID FailureDllHook, + PVOID FailureSystemHook, PVOID ThunkAddress, ULONG Flags); +static WindowsDllInterceptor::FuncHookType + stub_LdrResolveDelayLoadedAPI; + +static PVOID WINAPI patched_LdrResolveDelayLoadedAPI( + PVOID ParentModuleBase, PVOID DelayloadDescriptor, PVOID FailureDllHook, + PVOID FailureSystemHook, PVOID ThunkAddress, ULONG Flags) { + // Prevent the stack walker from suspending this thread when + // LdrResolveDelayLoadAPI holds the RtlLookupFunctionEntry lock. + AutoSuppressStackWalking suppress; + return stub_LdrResolveDelayLoadedAPI(ParentModuleBase, DelayloadDescriptor, + FailureDllHook, FailureSystemHook, + ThunkAddress, Flags); +} + +MFBT_API void InitializeWin64ProfilerHooks() { + // This function could be called by both profilers, but we only want to run + // it once. + static bool ran = false; + if (ran) { + return; + } + ran = true; + + NtDllIntercept.Init("ntdll.dll"); + stub_LdrUnloadDll.Set(NtDllIntercept, "LdrUnloadDll", &patched_LdrUnloadDll); + if (IsWin8OrLater()) { // LdrResolveDelayLoadedAPI was introduced in Win8 + stub_LdrResolveDelayLoadedAPI.Set(NtDllIntercept, + "LdrResolveDelayLoadedAPI", + &patched_LdrResolveDelayLoadedAPI); + } +} +#endif // defined(GP_PLAT_amd64_windows) + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/platform.cpp b/mozglue/baseprofiler/core/platform.cpp new file mode 100644 index 0000000000..dd3bfe6674 --- /dev/null +++ b/mozglue/baseprofiler/core/platform.cpp @@ -0,0 +1,3824 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +// There are three kinds of samples done by the profiler. +// +// - A "periodic" sample is the most complex kind. It is done in response to a +// timer while the profiler is active. It involves writing a stack trace plus +// a variety of other values (memory measurements, responsiveness +// measurements, etc.) into the main ProfileBuffer. The sampling is done from +// off-thread, and so SuspendAndSampleAndResumeThread() is used to get the +// register values. +// +// - A "synchronous" sample is a simpler kind. It is done in response to an API +// call (profiler_get_backtrace()). It involves writing a stack trace and +// little else into a temporary ProfileBuffer, and wrapping that up in a +// ProfilerBacktrace that can be subsequently used in a marker. The sampling +// is done on-thread, and so REGISTERS_SYNC_POPULATE() is used to get the +// register values. +// +// - A "backtrace" sample is the simplest kind. It is done in response to an +// API call (profiler_suspend_and_sample_thread()). It involves getting a +// stack trace via a ProfilerStackCollector; it does not write to a +// ProfileBuffer. The sampling is done from off-thread, and so uses +// SuspendAndSampleAndResumeThread() to get the register values. + +#include "platform.h" + +#include +#include +#include +#include +#include +#include +#include + +// #include "memory_hooks.h" +#include "mozilla/ArrayUtils.h" +#include "mozilla/AutoProfilerLabel.h" +#include "mozilla/BaseAndGeckoProfilerDetail.h" +#include "mozilla/BaseProfilerDetail.h" +#include "mozilla/DoubleConversion.h" +#include "mozilla/Printf.h" +#include "mozilla/ProfileBufferChunkManagerSingle.h" +#include "mozilla/ProfileBufferChunkManagerWithLocalLimit.h" +#include "mozilla/ProfileChunkedBuffer.h" +#include "mozilla/Services.h" +#include "mozilla/Span.h" +#include "mozilla/StackWalk.h" +#ifdef XP_WIN +# include "mozilla/StackWalkThread.h" +#endif +#include "mozilla/StaticPtr.h" +#include "mozilla/ThreadLocal.h" +#include "mozilla/TimeStamp.h" + +#include "mozilla/UniquePtr.h" +#include "mozilla/Vector.h" +#include "prdtoa.h" +#include "prtime.h" + +#include "BaseProfiler.h" +#include "BaseProfilingCategory.h" +#include "PageInformation.h" +#include "ProfiledThreadData.h" +#include "ProfilerBacktrace.h" +#include "ProfileBuffer.h" +#include "RegisteredThread.h" +#include "BaseProfilerSharedLibraries.h" +#include "ThreadInfo.h" +#include "VTuneProfiler.h" + +// Win32 builds always have frame pointers, so FramePointerStackWalk() always +// works. +#if defined(GP_PLAT_x86_windows) +# define HAVE_NATIVE_UNWIND +# define USE_FRAME_POINTER_STACK_WALK +#endif + +// Win64 builds always omit frame pointers, so we use the slower +// MozStackWalk(), which works in that case. +#if defined(GP_PLAT_amd64_windows) +# define HAVE_NATIVE_UNWIND +# define USE_MOZ_STACK_WALK +#endif + +// AArch64 Win64 doesn't seem to use frame pointers, so we use the slower +// MozStackWalk(). +#if defined(GP_PLAT_arm64_windows) +# define HAVE_NATIVE_UNWIND +# define USE_MOZ_STACK_WALK +#endif + +// Mac builds use FramePointerStackWalk(). Even if we build without +// frame pointers, we'll still get useful stacks in system libraries +// because those always have frame pointers. +// We don't use MozStackWalk() on Mac. +#if defined(GP_OS_darwin) +# define HAVE_NATIVE_UNWIND +# define USE_FRAME_POINTER_STACK_WALK +#endif + +// No stack-walking in baseprofiler on linux, android, bsd. +// APIs now make it easier to capture backtraces from the Base Profiler, which +// is currently not supported on these platform, and would lead to a MOZ_CRASH +// in REGISTERS_SYNC_POPULATE(). `#if 0` added in bug 1658232, follow-up bugs +// should be referenced in meta bug 1557568. +#if 0 +// Android builds use the ARM Exception Handling ABI to unwind. +# if defined(GP_PLAT_arm_linux) || defined(GP_PLAT_arm_android) +# define HAVE_NATIVE_UNWIND +# define USE_EHABI_STACKWALK +# include "EHABIStackWalk.h" +# endif + +// Linux/BSD builds use LUL, which uses DWARF info to unwind stacks. +# if defined(GP_PLAT_amd64_linux) || defined(GP_PLAT_x86_linux) || \ + defined(GP_PLAT_amd64_android) || defined(GP_PLAT_x86_android) || \ + defined(GP_PLAT_mips64_linux) || defined(GP_PLAT_arm64_linux) || \ + defined(GP_PLAT_arm64_android) || defined(GP_PLAT_amd64_freebsd) || \ + defined(GP_PLAT_arm64_freebsd) +# define HAVE_NATIVE_UNWIND +# define USE_LUL_STACKWALK +# include "lul/LulMain.h" +# include "lul/platform-linux-lul.h" + +// On linux we use LUL for periodic samples and synchronous samples, but we use +// FramePointerStackWalk for backtrace samples when MOZ_PROFILING is enabled. +// (See the comment at the top of the file for a definition of +// periodic/synchronous/backtrace.). +// +// FramePointerStackWalk can produce incomplete stacks when the current entry is +// in a shared library without framepointers, however LUL can take a long time +// to initialize, which is undesirable for consumers of +// profiler_suspend_and_sample_thread like the Background Hang Reporter. +# if defined(MOZ_PROFILING) +# define USE_FRAME_POINTER_STACK_WALK +# endif +# endif +#endif + +// We can only stackwalk without expensive initialization on platforms which +// support FramePointerStackWalk or MozStackWalk. LUL Stackwalking requires +// initializing LUL, and EHABIStackWalk requires initializing EHABI, both of +// which can be expensive. +#if defined(USE_FRAME_POINTER_STACK_WALK) || defined(USE_MOZ_STACK_WALK) +# define HAVE_FASTINIT_NATIVE_UNWIND +#endif + +#ifdef MOZ_VALGRIND +# include +#else +# define VALGRIND_MAKE_MEM_DEFINED(_addr, _len) ((void)0) +#endif + +#if defined(GP_OS_linux) || defined(GP_OS_android) || defined(GP_OS_freebsd) +# include +#endif + +namespace mozilla { +namespace baseprofiler { + +using detail::RacyFeatures; + +bool LogTest(int aLevelToTest) { + static const int maxLevel = getenv("MOZ_BASE_PROFILER_VERBOSE_LOGGING") ? 5 + : getenv("MOZ_BASE_PROFILER_DEBUG_LOGGING") ? 4 + : getenv("MOZ_BASE_PROFILER_LOGGING") ? 3 + : 0; + return aLevelToTest <= maxLevel; +} + +void PrintToConsole(const char* aFmt, ...) { + va_list args; + va_start(args, aFmt); +#if defined(ANDROID) + __android_log_vprint(ANDROID_LOG_INFO, "Gecko", aFmt, args); +#else + vfprintf(stderr, aFmt, args); +#endif + va_end(args); +} + +ProfileChunkedBuffer& profiler_get_core_buffer() { + // This needs its own mutex, because it is used concurrently from functions + // guarded by gPSMutex as well as others without safety (e.g., + // profiler_add_marker). It is *not* used inside the critical section of the + // sampler, because mutexes cannot be used there. + static ProfileChunkedBuffer sProfileChunkedBuffer{ + ProfileChunkedBuffer::ThreadSafety::WithMutex}; + return sProfileChunkedBuffer; +} + +Atomic gSkipSampling; + +constexpr static bool ValidateFeatures() { + int expectedFeatureNumber = 0; + + // Feature numbers should start at 0 and increase by 1 each. +#define CHECK_FEATURE(n_, str_, Name_, desc_) \ + if ((n_) != expectedFeatureNumber) { \ + return false; \ + } \ + ++expectedFeatureNumber; + + BASE_PROFILER_FOR_EACH_FEATURE(CHECK_FEATURE) + +#undef CHECK_FEATURE + + return true; +} + +static_assert(ValidateFeatures(), "Feature list is invalid"); + +// Return all features that are available on this platform. +static uint32_t AvailableFeatures() { + uint32_t features = 0; + +#define ADD_FEATURE(n_, str_, Name_, desc_) \ + ProfilerFeature::Set##Name_(features); + + // Add all the possible features. + BASE_PROFILER_FOR_EACH_FEATURE(ADD_FEATURE) + +#undef ADD_FEATURE + + // Now remove features not supported on this platform/configuration. + ProfilerFeature::ClearJava(features); + ProfilerFeature::ClearJS(features); + ProfilerFeature::ClearScreenshots(features); +#if !defined(HAVE_NATIVE_UNWIND) + ProfilerFeature::ClearStackWalk(features); +#endif +#if !defined(GP_OS_windows) + ProfilerFeature::ClearNoTimerResolutionChange(features); +#endif + + return features; +} + +// Default features common to all contexts (even if not available). +static constexpr uint32_t DefaultFeatures() { + return ProfilerFeature::Java | ProfilerFeature::JS | + ProfilerFeature::StackWalk | ProfilerFeature::CPUUtilization | + ProfilerFeature::ProcessCPU; +} + +// Extra default features when MOZ_PROFILER_STARTUP is set (even if not +// available). +static constexpr uint32_t StartupExtraDefaultFeatures() { + // Enable mainthreadio by default for startup profiles as startup is heavy on + // I/O operations, and main thread I/O is really important to see there. + return ProfilerFeature::MainThreadIO | ProfilerFeature::IPCMessages; +} + +// The auto-lock/unlock mutex that guards accesses to CorePS and ActivePS. +// Use `PSAutoLock lock;` to take the lock until the end of the enclosing block. +// External profilers may use this same lock for their own data, but as the lock +// is non-recursive, *only* `f(PSLockRef, ...)` functions below should be +// called, to avoid double-locking. +class MOZ_RAII PSAutoLock { + public: + PSAutoLock() : mLock(gPSMutex) {} + + PSAutoLock(const PSAutoLock&) = delete; + void operator=(const PSAutoLock&) = delete; + + [[nodiscard]] static bool IsLockedOnCurrentThread() { + return gPSMutex.IsLockedOnCurrentThread(); + } + + private: + static detail::BaseProfilerMutex gPSMutex; + detail::BaseProfilerAutoLock mLock; +}; + +detail::BaseProfilerMutex PSAutoLock::gPSMutex{"Base Profiler mutex"}; + +// Only functions that take a PSLockRef arg can access CorePS's and ActivePS's +// fields. +typedef const PSAutoLock& PSLockRef; + +#define PS_GET(type_, name_) \ + static type_ name_(PSLockRef) { \ + MOZ_ASSERT(sInstance); \ + return sInstance->m##name_; \ + } + +#define PS_GET_LOCKLESS(type_, name_) \ + static type_ name_() { \ + MOZ_ASSERT(sInstance); \ + return sInstance->m##name_; \ + } + +#define PS_GET_AND_SET(type_, name_) \ + PS_GET(type_, name_) \ + static void Set##name_(PSLockRef, type_ a##name_) { \ + MOZ_ASSERT(sInstance); \ + sInstance->m##name_ = a##name_; \ + } + +// All functions in this file can run on multiple threads unless they have an +// NS_IsMainThread() assertion. + +// This class contains the profiler's core global state, i.e. that which is +// valid even when the profiler is not active. Most profile operations can't do +// anything useful when this class is not instantiated, so we release-assert +// its non-nullness in all such operations. +// +// Accesses to CorePS are guarded by gPSMutex. Getters and setters take a +// PSAutoLock reference as an argument as proof that the gPSMutex is currently +// locked. This makes it clear when gPSMutex is locked and helps avoid +// accidental unlocked accesses to global state. There are ways to circumvent +// this mechanism, but please don't do so without *very* good reason and a +// detailed explanation. +// +// The exceptions to this rule: +// +// - mProcessStartTime, because it's immutable; +// +// - each thread's RacyRegisteredThread object is accessible without locking via +// TLSRegisteredThread::RacyRegisteredThread(). +class CorePS { + private: + CorePS() + : mProcessStartTime(TimeStamp::ProcessCreation()) +#ifdef USE_LUL_STACKWALK + , + mLul(nullptr) +#endif + { + } + + ~CorePS() {} + + public: + static void Create(PSLockRef aLock) { + MOZ_ASSERT(!sInstance); + sInstance = new CorePS(); + } + + static void Destroy(PSLockRef aLock) { + MOZ_ASSERT(sInstance); + delete sInstance; + sInstance = nullptr; + } + + // Unlike ActivePS::Exists(), CorePS::Exists() can be called without gPSMutex + // being locked. This is because CorePS is instantiated so early on the main + // thread that we don't have to worry about it being racy. + static bool Exists() { return !!sInstance; } + + static void AddSizeOf(PSLockRef, MallocSizeOf aMallocSizeOf, + size_t& aProfSize, size_t& aLulSize) { + MOZ_ASSERT(sInstance); + + aProfSize += aMallocSizeOf(sInstance); + + for (auto& registeredThread : sInstance->mRegisteredThreads) { + aProfSize += registeredThread->SizeOfIncludingThis(aMallocSizeOf); + } + + for (auto& registeredPage : sInstance->mRegisteredPages) { + aProfSize += registeredPage->SizeOfIncludingThis(aMallocSizeOf); + } + + // Measurement of the following things may be added later if DMD finds it + // is worthwhile: + // - CorePS::mRegisteredThreads itself (its elements' children are + // measured above) + // - CorePS::mRegisteredPages itself (its elements' children are + // measured above) + // - CorePS::mInterposeObserver + +#if defined(USE_LUL_STACKWALK) + if (sInstance->mLul) { + aLulSize += sInstance->mLul->SizeOfIncludingThis(aMallocSizeOf); + } +#endif + } + + // No PSLockRef is needed for this field because it's immutable. + PS_GET_LOCKLESS(const TimeStamp&, ProcessStartTime) + + PS_GET(const Vector>&, RegisteredThreads) + + static void AppendRegisteredThread( + PSLockRef, UniquePtr&& aRegisteredThread) { + MOZ_ASSERT(sInstance); + MOZ_RELEASE_ASSERT( + sInstance->mRegisteredThreads.append(std::move(aRegisteredThread))); + } + + static void RemoveRegisteredThread(PSLockRef, + RegisteredThread* aRegisteredThread) { + MOZ_ASSERT(sInstance); + // Remove aRegisteredThread from mRegisteredThreads. + for (UniquePtr& rt : sInstance->mRegisteredThreads) { + if (rt.get() == aRegisteredThread) { + sInstance->mRegisteredThreads.erase(&rt); + return; + } + } + } + + PS_GET(Vector>&, RegisteredPages) + + static void AppendRegisteredPage(PSLockRef, + RefPtr&& aRegisteredPage) { + MOZ_ASSERT(sInstance); + struct RegisteredPageComparator { + PageInformation* aA; + bool operator()(PageInformation* aB) const { return aA->Equals(aB); } + }; + + auto foundPageIter = std::find_if( + sInstance->mRegisteredPages.begin(), sInstance->mRegisteredPages.end(), + RegisteredPageComparator{aRegisteredPage.get()}); + + if (foundPageIter != sInstance->mRegisteredPages.end()) { + if ((*foundPageIter)->Url() == "about:blank") { + // When a BrowsingContext is loaded, the first url loaded in it will be + // about:blank, and if the principal matches, the first document loaded + // in it will share an inner window. That's why we should delete the + // intermittent about:blank if they share the inner window. + sInstance->mRegisteredPages.erase(foundPageIter); + } else { + // Do not register the same page again. + return; + } + } + MOZ_RELEASE_ASSERT( + sInstance->mRegisteredPages.append(std::move(aRegisteredPage))); + } + + static void RemoveRegisteredPage(PSLockRef, + uint64_t aRegisteredInnerWindowID) { + MOZ_ASSERT(sInstance); + // Remove RegisteredPage from mRegisteredPages by given inner window ID. + sInstance->mRegisteredPages.eraseIf([&](const RefPtr& rd) { + return rd->InnerWindowID() == aRegisteredInnerWindowID; + }); + } + + static void ClearRegisteredPages(PSLockRef) { + MOZ_ASSERT(sInstance); + sInstance->mRegisteredPages.clear(); + } + + PS_GET(const Vector&, Counters) + + static void AppendCounter(PSLockRef, BaseProfilerCount* aCounter) { + MOZ_ASSERT(sInstance); + // we don't own the counter; they may be stored in static objects + MOZ_RELEASE_ASSERT(sInstance->mCounters.append(aCounter)); + } + + static void RemoveCounter(PSLockRef, BaseProfilerCount* aCounter) { + // we may be called to remove a counter after the profiler is stopped or + // late in shutdown. + if (sInstance) { + auto* counter = std::find(sInstance->mCounters.begin(), + sInstance->mCounters.end(), aCounter); + MOZ_RELEASE_ASSERT(counter != sInstance->mCounters.end()); + sInstance->mCounters.erase(counter); + } + } + +#ifdef USE_LUL_STACKWALK + static lul::LUL* Lul(PSLockRef) { + MOZ_ASSERT(sInstance); + return sInstance->mLul.get(); + } + static void SetLul(PSLockRef, UniquePtr aLul) { + MOZ_ASSERT(sInstance); + sInstance->mLul = std::move(aLul); + } +#endif + + PS_GET_AND_SET(const std::string&, ProcessName) + PS_GET_AND_SET(const std::string&, ETLDplus1) + + private: + // The singleton instance + static CorePS* sInstance; + + // The time that the process started. + const TimeStamp mProcessStartTime; + + // Info on all the registered threads. + // ThreadIds in mRegisteredThreads are unique. + Vector> mRegisteredThreads; + + // Info on all the registered pages. + // InnerWindowIDs in mRegisteredPages are unique. + Vector> mRegisteredPages; + + // Non-owning pointers to all active counters + Vector mCounters; + +#ifdef USE_LUL_STACKWALK + // LUL's state. Null prior to the first activation, non-null thereafter. + UniquePtr mLul; +#endif + + // Process name, provided by child process initialization code. + std::string mProcessName; + // Private name, provided by child process initialization code (eTLD+1 in + // fission) + std::string mETLDplus1; +}; + +CorePS* CorePS::sInstance = nullptr; + +class SamplerThread; + +static SamplerThread* NewSamplerThread(PSLockRef aLock, uint32_t aGeneration, + double aInterval, uint32_t aFeatures); + +struct LiveProfiledThreadData { + RegisteredThread* mRegisteredThread; + UniquePtr mProfiledThreadData; +}; + +// The buffer size is provided as a number of "entries", this is their size in +// bytes. +constexpr static uint32_t scBytesPerEntry = 8; + +// This class contains the profiler's global state that is valid only when the +// profiler is active. When not instantiated, the profiler is inactive. +// +// Accesses to ActivePS are guarded by gPSMutex, in much the same fashion as +// CorePS. +// +class ActivePS { + private: + // We need to decide how many chunks of what size we want to fit in the given + // total maximum capacity for this process, in the (likely) context of + // multiple processes doing the same choice and having an inter-process + // mechanism to control the overal memory limit. + + // Minimum chunk size allowed, enough for at least one stack. + constexpr static uint32_t scMinimumChunkSize = + 2 * ProfileBufferChunkManager::scExpectedMaximumStackSize; + + // Ideally we want at least 2 unreleased chunks to work with (1 current and 1 + // next), and 2 released chunks (so that one can be recycled when old, leaving + // one with some data). + constexpr static uint32_t scMinimumNumberOfChunks = 4; + + // And we want to limit chunks to a maximum size, which is a compromise + // between: + // - A big size, which helps with reducing the rate of allocations and IPCs. + // - A small size, which helps with equalizing the duration of recorded data + // (as the inter-process controller will discard the oldest chunks in all + // Firefox processes). + constexpr static uint32_t scMaximumChunkSize = 1024 * 1024; + + public: + // We should be able to store at least the minimum number of the smallest- + // possible chunks. + constexpr static uint32_t scMinimumBufferSize = + scMinimumNumberOfChunks * scMinimumChunkSize; + constexpr static uint32_t scMinimumBufferEntries = + scMinimumBufferSize / scBytesPerEntry; + + // Limit to 2GiB. + constexpr static uint32_t scMaximumBufferSize = 2u * 1024u * 1024u * 1024u; + constexpr static uint32_t scMaximumBufferEntries = + scMaximumBufferSize / scBytesPerEntry; + + constexpr static uint32_t ClampToAllowedEntries(uint32_t aEntries) { + if (aEntries <= scMinimumBufferEntries) { + return scMinimumBufferEntries; + } + if (aEntries >= scMaximumBufferEntries) { + return scMaximumBufferEntries; + } + return aEntries; + } + + private: + constexpr static uint32_t ChunkSizeForEntries(uint32_t aEntries) { + return uint32_t(std::min(size_t(ClampToAllowedEntries(aEntries)) * + scBytesPerEntry / scMinimumNumberOfChunks, + size_t(scMaximumChunkSize))); + } + + static uint32_t AdjustFeatures(uint32_t aFeatures, uint32_t aFilterCount) { + // Filter out any features unavailable in this platform/configuration. + aFeatures &= AvailableFeatures(); + + // Some features imply others. + if (aFeatures & ProfilerFeature::FileIOAll) { + aFeatures |= ProfilerFeature::MainThreadIO | ProfilerFeature::FileIO; + } else if (aFeatures & ProfilerFeature::FileIO) { + aFeatures |= ProfilerFeature::MainThreadIO; + } + + return aFeatures; + } + + ActivePS(PSLockRef aLock, const TimeStamp& aProfilingStartTime, + PowerOfTwo32 aCapacity, double aInterval, uint32_t aFeatures, + const char** aFilters, uint32_t aFilterCount, + const Maybe& aDuration) + : mProfilingStartTime(aProfilingStartTime), + mGeneration(sNextGeneration++), + mCapacity(aCapacity), + mDuration(aDuration), + mInterval(aInterval), + mFeatures(AdjustFeatures(aFeatures, aFilterCount)), + mProfileBufferChunkManager( + MakeUnique( + size_t(ClampToAllowedEntries(aCapacity.Value())) * + scBytesPerEntry, + ChunkSizeForEntries(aCapacity.Value()))), + mProfileBuffer([this]() -> ProfileChunkedBuffer& { + ProfileChunkedBuffer& buffer = profiler_get_core_buffer(); + buffer.SetChunkManager(*mProfileBufferChunkManager); + return buffer; + }()), + // The new sampler thread doesn't start sampling immediately because the + // main loop within Run() is blocked until this function's caller + // unlocks gPSMutex. + mSamplerThread( + NewSamplerThread(aLock, mGeneration, aInterval, aFeatures)), + mIsPaused(false), + mIsSamplingPaused(false) { + // Deep copy and lower-case aFilters. + MOZ_ALWAYS_TRUE(mFilters.resize(aFilterCount)); + MOZ_ALWAYS_TRUE(mFiltersLowered.resize(aFilterCount)); + for (uint32_t i = 0; i < aFilterCount; ++i) { + mFilters[i] = aFilters[i]; + mFiltersLowered[i].reserve(mFilters[i].size()); + std::transform(mFilters[i].cbegin(), mFilters[i].cend(), + std::back_inserter(mFiltersLowered[i]), ::tolower); + } + } + + ~ActivePS() { + if (mProfileBufferChunkManager) { + // We still control the chunk manager, remove it from the core buffer. + profiler_get_core_buffer().ResetChunkManager(); + } + } + + bool ThreadSelected(const char* aThreadName) { + if (mFiltersLowered.empty()) { + return true; + } + + std::string name = aThreadName; + std::transform(name.begin(), name.end(), name.begin(), ::tolower); + + for (const auto& filter : mFiltersLowered) { + if (filter == "*") { + return true; + } + + // Crude, non UTF-8 compatible, case insensitive substring search + if (name.find(filter) != std::string::npos) { + return true; + } + + // If the filter is "pid:", profile all threads. + if (mozilla::profiler::detail::FilterHasPid(filter.c_str())) { + return true; + } + } + + return false; + } + + public: + static void Create(PSLockRef aLock, const TimeStamp& aProfilingStartTime, + PowerOfTwo32 aCapacity, double aInterval, + uint32_t aFeatures, const char** aFilters, + uint32_t aFilterCount, const Maybe& aDuration) { + MOZ_ASSERT(!sInstance); + sInstance = new ActivePS(aLock, aProfilingStartTime, aCapacity, aInterval, + aFeatures, aFilters, aFilterCount, aDuration); + } + + [[nodiscard]] static SamplerThread* Destroy(PSLockRef aLock) { + MOZ_ASSERT(sInstance); + auto samplerThread = sInstance->mSamplerThread; + delete sInstance; + sInstance = nullptr; + + return samplerThread; + } + + static bool Exists(PSLockRef) { return !!sInstance; } + + static bool Equals(PSLockRef, PowerOfTwo32 aCapacity, + const Maybe& aDuration, double aInterval, + uint32_t aFeatures, const char** aFilters, + uint32_t aFilterCount) { + MOZ_ASSERT(sInstance); + if (sInstance->mCapacity != aCapacity || + sInstance->mDuration != aDuration || + sInstance->mInterval != aInterval || + sInstance->mFeatures != aFeatures || + sInstance->mFilters.length() != aFilterCount) { + return false; + } + + for (uint32_t i = 0; i < sInstance->mFilters.length(); ++i) { + if (strcmp(sInstance->mFilters[i].c_str(), aFilters[i]) != 0) { + return false; + } + } + return true; + } + + static size_t SizeOf(PSLockRef, MallocSizeOf aMallocSizeOf) { + MOZ_ASSERT(sInstance); + + size_t n = aMallocSizeOf(sInstance); + + n += sInstance->mProfileBuffer.SizeOfExcludingThis(aMallocSizeOf); + + // Measurement of the following members may be added later if DMD finds it + // is worthwhile: + // - mLiveProfiledThreads (both the array itself, and the contents) + // - mDeadProfiledThreads (both the array itself, and the contents) + // + + return n; + } + + static UniquePtr + ExtractBaseProfilerChunkManager(PSLockRef) { + MOZ_ASSERT(sInstance); + return std::move(sInstance->mProfileBufferChunkManager); + } + + static bool ShouldProfileThread(PSLockRef aLock, ThreadInfo* aInfo) { + MOZ_ASSERT(sInstance); + return sInstance->ThreadSelected(aInfo->Name()); + } + + PS_GET_LOCKLESS(TimeStamp, ProfilingStartTime) + + PS_GET(uint32_t, Generation) + + PS_GET(PowerOfTwo32, Capacity) + + PS_GET(Maybe, Duration) + + PS_GET(double, Interval) + + PS_GET(uint32_t, Features) + +#define PS_GET_FEATURE(n_, str_, Name_, desc_) \ + static bool Feature##Name_(PSLockRef) { \ + MOZ_ASSERT(sInstance); \ + return ProfilerFeature::Has##Name_(sInstance->mFeatures); \ + } + + BASE_PROFILER_FOR_EACH_FEATURE(PS_GET_FEATURE) + +#undef PS_GET_FEATURE + + PS_GET(const Vector&, Filters) + PS_GET(const Vector&, FiltersLowered) + + static void FulfillChunkRequests(PSLockRef) { + MOZ_ASSERT(sInstance); + if (sInstance->mProfileBufferChunkManager) { + sInstance->mProfileBufferChunkManager->FulfillChunkRequests(); + } + } + + static ProfileBuffer& Buffer(PSLockRef) { + MOZ_ASSERT(sInstance); + return sInstance->mProfileBuffer; + } + + static const Vector& LiveProfiledThreads(PSLockRef) { + MOZ_ASSERT(sInstance); + return sInstance->mLiveProfiledThreads; + } + + // Returns an array containing (RegisteredThread*, ProfiledThreadData*) pairs + // for all threads that should be included in a profile, both for threads + // that are still registered, and for threads that have been unregistered but + // still have data in the buffer. + // For threads that have already been unregistered, the RegisteredThread + // pointer will be null. + // The returned array is sorted by thread register time. + // Do not hold on to the return value across thread registration or profiler + // restarts. + static Vector> + ProfiledThreads(PSLockRef) { + MOZ_ASSERT(sInstance); + Vector> array; + MOZ_RELEASE_ASSERT( + array.initCapacity(sInstance->mLiveProfiledThreads.length() + + sInstance->mDeadProfiledThreads.length())); + for (auto& t : sInstance->mLiveProfiledThreads) { + MOZ_RELEASE_ASSERT(array.append( + std::make_pair(t.mRegisteredThread, t.mProfiledThreadData.get()))); + } + for (auto& t : sInstance->mDeadProfiledThreads) { + MOZ_RELEASE_ASSERT( + array.append(std::make_pair((RegisteredThread*)nullptr, t.get()))); + } + + std::sort(array.begin(), array.end(), + [](const std::pair& a, + const std::pair& b) { + return a.second->Info()->RegisterTime() < + b.second->Info()->RegisterTime(); + }); + return array; + } + + static Vector> ProfiledPages(PSLockRef aLock) { + MOZ_ASSERT(sInstance); + Vector> array; + for (auto& d : CorePS::RegisteredPages(aLock)) { + MOZ_RELEASE_ASSERT(array.append(d)); + } + for (auto& d : sInstance->mDeadProfiledPages) { + MOZ_RELEASE_ASSERT(array.append(d)); + } + // We don't need to sort the pages like threads since we won't show them + // as a list. + return array; + } + + // Do a linear search through mLiveProfiledThreads to find the + // ProfiledThreadData object for a RegisteredThread. + static ProfiledThreadData* GetProfiledThreadData( + PSLockRef, RegisteredThread* aRegisteredThread) { + MOZ_ASSERT(sInstance); + for (const LiveProfiledThreadData& thread : + sInstance->mLiveProfiledThreads) { + if (thread.mRegisteredThread == aRegisteredThread) { + return thread.mProfiledThreadData.get(); + } + } + return nullptr; + } + + static ProfiledThreadData* AddLiveProfiledThread( + PSLockRef, RegisteredThread* aRegisteredThread, + UniquePtr&& aProfiledThreadData) { + MOZ_ASSERT(sInstance); + MOZ_RELEASE_ASSERT( + sInstance->mLiveProfiledThreads.append(LiveProfiledThreadData{ + aRegisteredThread, std::move(aProfiledThreadData)})); + + // Return a weak pointer to the ProfiledThreadData object. + return sInstance->mLiveProfiledThreads.back().mProfiledThreadData.get(); + } + + static void UnregisterThread(PSLockRef aLockRef, + RegisteredThread* aRegisteredThread) { + MOZ_ASSERT(sInstance); + + DiscardExpiredDeadProfiledThreads(aLockRef); + + // Find the right entry in the mLiveProfiledThreads array and remove the + // element, moving the ProfiledThreadData object for the thread into the + // mDeadProfiledThreads array. + // The thread's RegisteredThread object gets destroyed here. + for (size_t i = 0; i < sInstance->mLiveProfiledThreads.length(); i++) { + LiveProfiledThreadData& thread = sInstance->mLiveProfiledThreads[i]; + if (thread.mRegisteredThread == aRegisteredThread) { + thread.mProfiledThreadData->NotifyUnregistered( + sInstance->mProfileBuffer.BufferRangeEnd()); + MOZ_RELEASE_ASSERT(sInstance->mDeadProfiledThreads.append( + std::move(thread.mProfiledThreadData))); + sInstance->mLiveProfiledThreads.erase( + &sInstance->mLiveProfiledThreads[i]); + return; + } + } + } + + PS_GET_AND_SET(bool, IsPaused) + + // True if sampling is paused (though generic `SetIsPaused()` or specific + // `SetIsSamplingPaused()`). + static bool IsSamplingPaused(PSLockRef lock) { + MOZ_ASSERT(sInstance); + return IsPaused(lock) || sInstance->mIsSamplingPaused; + } + + static void SetIsSamplingPaused(PSLockRef, bool aIsSamplingPaused) { + MOZ_ASSERT(sInstance); + sInstance->mIsSamplingPaused = aIsSamplingPaused; + } + + static void DiscardExpiredDeadProfiledThreads(PSLockRef) { + MOZ_ASSERT(sInstance); + uint64_t bufferRangeStart = sInstance->mProfileBuffer.BufferRangeStart(); + // Discard any dead threads that were unregistered before bufferRangeStart. + sInstance->mDeadProfiledThreads.eraseIf( + [bufferRangeStart]( + const UniquePtr& aProfiledThreadData) { + Maybe bufferPosition = + aProfiledThreadData->BufferPositionWhenUnregistered(); + MOZ_RELEASE_ASSERT(bufferPosition, + "should have unregistered this thread"); + return *bufferPosition < bufferRangeStart; + }); + } + + static void UnregisterPage(PSLockRef aLock, + uint64_t aRegisteredInnerWindowID) { + MOZ_ASSERT(sInstance); + auto& registeredPages = CorePS::RegisteredPages(aLock); + for (size_t i = 0; i < registeredPages.length(); i++) { + RefPtr& page = registeredPages[i]; + if (page->InnerWindowID() == aRegisteredInnerWindowID) { + page->NotifyUnregistered(sInstance->mProfileBuffer.BufferRangeEnd()); + MOZ_RELEASE_ASSERT( + sInstance->mDeadProfiledPages.append(std::move(page))); + registeredPages.erase(®isteredPages[i--]); + } + } + } + + static void DiscardExpiredPages(PSLockRef) { + MOZ_ASSERT(sInstance); + uint64_t bufferRangeStart = sInstance->mProfileBuffer.BufferRangeStart(); + // Discard any dead pages that were unregistered before + // bufferRangeStart. + sInstance->mDeadProfiledPages.eraseIf( + [bufferRangeStart](const RefPtr& aProfiledPage) { + Maybe bufferPosition = + aProfiledPage->BufferPositionWhenUnregistered(); + MOZ_RELEASE_ASSERT(bufferPosition, + "should have unregistered this page"); + return *bufferPosition < bufferRangeStart; + }); + } + + static void ClearUnregisteredPages(PSLockRef) { + MOZ_ASSERT(sInstance); + sInstance->mDeadProfiledPages.clear(); + } + + static void ClearExpiredExitProfiles(PSLockRef) { + MOZ_ASSERT(sInstance); + uint64_t bufferRangeStart = sInstance->mProfileBuffer.BufferRangeStart(); + // Discard exit profiles that were gathered before our buffer RangeStart. + sInstance->mExitProfiles.eraseIf( + [bufferRangeStart](const ExitProfile& aExitProfile) { + return aExitProfile.mBufferPositionAtGatherTime < bufferRangeStart; + }); + } + + static void AddExitProfile(PSLockRef aLock, const std::string& aExitProfile) { + MOZ_ASSERT(sInstance); + + ClearExpiredExitProfiles(aLock); + + MOZ_RELEASE_ASSERT(sInstance->mExitProfiles.append( + ExitProfile{aExitProfile, sInstance->mProfileBuffer.BufferRangeEnd()})); + } + + static Vector MoveExitProfiles(PSLockRef aLock) { + MOZ_ASSERT(sInstance); + + ClearExpiredExitProfiles(aLock); + + Vector profiles; + MOZ_RELEASE_ASSERT( + profiles.initCapacity(sInstance->mExitProfiles.length())); + for (auto& profile : sInstance->mExitProfiles) { + MOZ_RELEASE_ASSERT(profiles.append(std::move(profile.mJSON))); + } + sInstance->mExitProfiles.clear(); + return profiles; + } + + private: + // The singleton instance. + static ActivePS* sInstance; + + const TimeStamp mProfilingStartTime; + + // We need to track activity generations. If we didn't we could have the + // following scenario. + // + // - profiler_stop() locks gPSMutex, de-instantiates ActivePS, unlocks + // gPSMutex, deletes the SamplerThread (which does a join). + // + // - profiler_start() runs on a different thread, locks gPSMutex, + // re-instantiates ActivePS, unlocks gPSMutex -- all before the join + // completes. + // + // - SamplerThread::Run() locks gPSMutex, sees that ActivePS is instantiated, + // and continues as if the start/stop pair didn't occur. Also + // profiler_stop() is stuck, unable to finish. + // + // By checking ActivePS *and* the generation, we can avoid this scenario. + // sNextGeneration is used to track the next generation number; it is static + // because it must persist across different ActivePS instantiations. + const uint32_t mGeneration; + static uint32_t sNextGeneration; + + // The maximum number of 8-byte entries in mProfileBuffer. + const PowerOfTwo32 mCapacity; + + // The maximum duration of entries in mProfileBuffer, in seconds. + const Maybe mDuration; + + // The interval between samples, measured in milliseconds. + const double mInterval; + + // The profile features that are enabled. + const uint32_t mFeatures; + + // Substrings of names of threads we want to profile. + Vector mFilters; + Vector mFiltersLowered; + + // The chunk manager used by `mProfileBuffer` below. + // May become null if it gets transferred to the Gecko Profiler. + UniquePtr mProfileBufferChunkManager; + + // The buffer into which all samples are recorded. + ProfileBuffer mProfileBuffer; + + // ProfiledThreadData objects for any threads that were profiled at any point + // during this run of the profiler: + // - mLiveProfiledThreads contains all threads that are still registered, and + // - mDeadProfiledThreads contains all threads that have already been + // unregistered but for which there is still data in the profile buffer. + Vector mLiveProfiledThreads; + Vector> mDeadProfiledThreads; + + // Info on all the dead pages. + // Registered pages are being moved to this array after unregistration. + // We are keeping them in case we need them in the profile data. + // We are removing them when we ensure that we won't need them anymore. + Vector> mDeadProfiledPages; + + // The current sampler thread. This class is not responsible for destroying + // the SamplerThread object; the Destroy() method returns it so the caller + // can destroy it. + SamplerThread* const mSamplerThread; + + // Is the profiler fully paused? + bool mIsPaused; + + // Is the profiler periodic sampling paused? + bool mIsSamplingPaused; + + struct ExitProfile { + std::string mJSON; + uint64_t mBufferPositionAtGatherTime; + }; + Vector mExitProfiles; +}; + +ActivePS* ActivePS::sInstance = nullptr; +uint32_t ActivePS::sNextGeneration = 0; + +#undef PS_GET +#undef PS_GET_LOCKLESS +#undef PS_GET_AND_SET + +namespace detail { + +TimeStamp GetProfilingStartTime() { + if (!CorePS::Exists()) { + return {}; + } + PSAutoLock lock; + if (!ActivePS::Exists(lock)) { + return {}; + } + return ActivePS::ProfilingStartTime(); +} + +[[nodiscard]] MFBT_API UniquePtr +ExtractBaseProfilerChunkManager() { + PSAutoLock lock; + if (MOZ_UNLIKELY(!ActivePS::Exists(lock))) { + return nullptr; + } + return ActivePS::ExtractBaseProfilerChunkManager(lock); +} + +} // namespace detail + +Atomic RacyFeatures::sActiveAndFeatures(0); + +/* static */ +void RacyFeatures::SetActive(uint32_t aFeatures) { + sActiveAndFeatures = Active | aFeatures; +} + +/* static */ +void RacyFeatures::SetInactive() { sActiveAndFeatures = 0; } + +/* static */ +bool RacyFeatures::IsActive() { return uint32_t(sActiveAndFeatures) & Active; } + +/* static */ +void RacyFeatures::SetPaused() { sActiveAndFeatures |= Paused; } + +/* static */ +void RacyFeatures::SetUnpaused() { sActiveAndFeatures &= ~Paused; } + +/* static */ +void RacyFeatures::SetSamplingPaused() { sActiveAndFeatures |= SamplingPaused; } + +/* static */ +void RacyFeatures::SetSamplingUnpaused() { + sActiveAndFeatures &= ~SamplingPaused; +} + +/* static */ +bool RacyFeatures::IsActiveWithFeature(uint32_t aFeature) { + uint32_t af = sActiveAndFeatures; // copy it first + return (af & Active) && (af & aFeature); +} + +/* static */ +bool RacyFeatures::IsActiveWithoutFeature(uint32_t aFeature) { + uint32_t af = sActiveAndFeatures; // copy it first + return (af & Active) && !(af & aFeature); +} + +/* static */ +bool RacyFeatures::IsActiveAndUnpaused() { + uint32_t af = sActiveAndFeatures; // copy it first + return (af & Active) && !(af & Paused); +} + +/* static */ +bool RacyFeatures::IsActiveAndSamplingUnpaused() { + uint32_t af = sActiveAndFeatures; // copy it first + return (af & Active) && !(af & (Paused | SamplingPaused)); +} + +// Each live thread has a RegisteredThread, and we store a reference to it in +// TLS. This class encapsulates that TLS. +class TLSRegisteredThread { + public: + static bool Init(PSLockRef) { + bool ok1 = sRegisteredThread.init(); + bool ok2 = AutoProfilerLabel::sProfilingStack.init(); + return ok1 && ok2; + } + + // Get the entire RegisteredThread. Accesses are guarded by gPSMutex. + static class RegisteredThread* RegisteredThread(PSLockRef) { + return sRegisteredThread.get(); + } + + // Get only the RacyRegisteredThread. Accesses are not guarded by gPSMutex. + static class RacyRegisteredThread* RacyRegisteredThread() { + class RegisteredThread* registeredThread = sRegisteredThread.get(); + return registeredThread ? ®isteredThread->RacyRegisteredThread() + : nullptr; + } + + // Get only the ProfilingStack. Accesses are not guarded by gPSMutex. + // RacyRegisteredThread() can also be used to get the ProfilingStack, but that + // is marginally slower because it requires an extra pointer indirection. + static ProfilingStack* Stack() { + return AutoProfilerLabel::sProfilingStack.get(); + } + + static void SetRegisteredThread(PSLockRef, + class RegisteredThread* aRegisteredThread) { + sRegisteredThread.set(aRegisteredThread); + AutoProfilerLabel::sProfilingStack.set( + aRegisteredThread + ? &aRegisteredThread->RacyRegisteredThread().ProfilingStack() + : nullptr); + } + + private: + // This is a non-owning reference to the RegisteredThread; + // CorePS::mRegisteredThreads is the owning reference. On thread + // deregistration, this reference is cleared and the RegisteredThread is + // destroyed. + static MOZ_THREAD_LOCAL(class RegisteredThread*) sRegisteredThread; +}; + +MOZ_THREAD_LOCAL(RegisteredThread*) TLSRegisteredThread::sRegisteredThread; + +/* static */ +ProfilingStack* AutoProfilerLabel::GetProfilingStack() { + return sProfilingStack.get(); +} + +// Although you can access a thread's ProfilingStack via +// TLSRegisteredThread::sRegisteredThread, we also have a second TLS pointer +// directly to the ProfilingStack. Here's why. +// +// - We need to be able to push to and pop from the ProfilingStack in +// AutoProfilerLabel. +// +// - The class functions are hot and must be defined in BaseProfiler.h so they +// can be inlined. +// +// - We don't want to expose TLSRegisteredThread (and RegisteredThread) in +// BaseProfiler.h. +// +// This second pointer isn't ideal, but does provide a way to satisfy those +// constraints. TLSRegisteredThread is responsible for updating it. +MOZ_THREAD_LOCAL(ProfilingStack*) AutoProfilerLabel::sProfilingStack; + +namespace detail { + +[[nodiscard]] MFBT_API TimeStamp GetThreadRegistrationTime() { + if (!CorePS::Exists()) { + return {}; + } + + PSAutoLock lock; + + RegisteredThread* registeredThread = + TLSRegisteredThread::RegisteredThread(lock); + if (!registeredThread) { + return {}; + } + + return registeredThread->Info()->RegisterTime(); +} + +} // namespace detail + +// The name of the main thread. +static const char* const kMainThreadName = "GeckoMain"; + +//////////////////////////////////////////////////////////////////////// +// BEGIN sampling/unwinding code + +// The registers used for stack unwinding and a few other sampling purposes. +// The ctor does nothing; users are responsible for filling in the fields. +class Registers { + public: + Registers() : mPC{nullptr}, mSP{nullptr}, mFP{nullptr}, mLR{nullptr} {} + + void Clear() { memset(this, 0, sizeof(*this)); } + + // These fields are filled in by + // Sampler::SuspendAndSampleAndResumeThread() for periodic and backtrace + // samples, and by REGISTERS_SYNC_POPULATE for synchronous samples. + Address mPC; // Instruction pointer. + Address mSP; // Stack pointer. + Address mFP; // Frame pointer. + Address mLR; // ARM link register. +#if defined(GP_OS_linux) || defined(GP_OS_android) || defined(GP_OS_freebsd) + // This contains all the registers, which means it duplicates the four fields + // above. This is ok. + ucontext_t* mContext; // The context from the signal handler. +#endif +}; + +// Setting MAX_NATIVE_FRAMES too high risks the unwinder wasting a lot of time +// looping on corrupted stacks. +static const size_t MAX_NATIVE_FRAMES = 1024; + +struct NativeStack { + void* mPCs[MAX_NATIVE_FRAMES]; + void* mSPs[MAX_NATIVE_FRAMES]; + size_t mCount; // Number of frames filled. + + NativeStack() : mPCs(), mSPs(), mCount(0) {} +}; + +// Merges the profiling stack and native stack, outputting the details to +// aCollector. +static void MergeStacks(uint32_t aFeatures, bool aIsSynchronous, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, const NativeStack& aNativeStack, + ProfilerStackCollector& aCollector) { + // WARNING: this function runs within the profiler's "critical section". + // WARNING: this function might be called while the profiler is inactive, and + // cannot rely on ActivePS. + + const ProfilingStack& profilingStack = + aRegisteredThread.RacyRegisteredThread().ProfilingStack(); + const ProfilingStackFrame* profilingStackFrames = profilingStack.frames; + uint32_t profilingStackFrameCount = profilingStack.stackSize(); + + Maybe samplePosInBuffer; + if (!aIsSynchronous) { + // aCollector.SamplePositionInBuffer() will return Nothing() when + // profiler_suspend_and_sample_thread is called from the background hang + // reporter. + samplePosInBuffer = aCollector.SamplePositionInBuffer(); + } + // While the profiling stack array is ordered oldest-to-youngest, the JS and + // native arrays are ordered youngest-to-oldest. We must add frames to aInfo + // oldest-to-youngest. Thus, iterate over the profiling stack forwards and JS + // and native arrays backwards. Note: this means the terminating condition + // jsIndex and nativeIndex is being < 0. + uint32_t profilingStackIndex = 0; + int32_t nativeIndex = aNativeStack.mCount - 1; + + uint8_t* lastLabelFrameStackAddr = nullptr; + + // Iterate as long as there is at least one frame remaining. + while (profilingStackIndex != profilingStackFrameCount || nativeIndex >= 0) { + // There are 1 to 3 frames available. Find and add the oldest. + uint8_t* profilingStackAddr = nullptr; + uint8_t* nativeStackAddr = nullptr; + + if (profilingStackIndex != profilingStackFrameCount) { + const ProfilingStackFrame& profilingStackFrame = + profilingStackFrames[profilingStackIndex]; + + if (profilingStackFrame.isLabelFrame() || + profilingStackFrame.isSpMarkerFrame()) { + lastLabelFrameStackAddr = (uint8_t*)profilingStackFrame.stackAddress(); + } + + // Skip any JS_OSR frames. Such frames are used when the JS interpreter + // enters a jit frame on a loop edge (via on-stack-replacement, or OSR). + // To avoid both the profiling stack frame and jit frame being recorded + // (and showing up twice), the interpreter marks the interpreter + // profiling stack frame as JS_OSR to ensure that it doesn't get counted. + if (profilingStackFrame.isOSRFrame()) { + profilingStackIndex++; + continue; + } + + MOZ_ASSERT(lastLabelFrameStackAddr); + profilingStackAddr = lastLabelFrameStackAddr; + } + + if (nativeIndex >= 0) { + nativeStackAddr = (uint8_t*)aNativeStack.mSPs[nativeIndex]; + } + + // If there's a native stack frame which has the same SP as a profiling + // stack frame, pretend we didn't see the native stack frame. Ditto for a + // native stack frame which has the same SP as a JS stack frame. In effect + // this means profiling stack frames or JS frames trump conflicting native + // frames. + if (nativeStackAddr && (profilingStackAddr == nativeStackAddr)) { + nativeStackAddr = nullptr; + nativeIndex--; + MOZ_ASSERT(profilingStackAddr); + } + + // Sanity checks. + MOZ_ASSERT_IF(profilingStackAddr, profilingStackAddr != nativeStackAddr); + MOZ_ASSERT_IF(nativeStackAddr, nativeStackAddr != profilingStackAddr); + + // Check to see if profiling stack frame is top-most. + if (profilingStackAddr > nativeStackAddr) { + MOZ_ASSERT(profilingStackIndex < profilingStackFrameCount); + const ProfilingStackFrame& profilingStackFrame = + profilingStackFrames[profilingStackIndex]; + + // Sp marker frames are just annotations and should not be recorded in + // the profile. + if (!profilingStackFrame.isSpMarkerFrame()) { + if (aIsSynchronous && profilingStackFrame.categoryPair() == + ProfilingCategoryPair::PROFILER) { + // For stacks captured synchronously (ie. marker stacks), stop + // walking the stack as soon as we enter the profiler category, + // to avoid showing profiler internal code in marker stacks. + return; + } + aCollector.CollectProfilingStackFrame(profilingStackFrame); + } + profilingStackIndex++; + continue; + } + + // If we reach here, there must be a native stack frame and it must be the + // greatest frame. + if (nativeStackAddr) { + MOZ_ASSERT(nativeIndex >= 0); + void* addr = (void*)aNativeStack.mPCs[nativeIndex]; + aCollector.CollectNativeLeafAddr(addr); + } + if (nativeIndex >= 0) { + nativeIndex--; + } + } +} + +#if defined(GP_OS_windows) && defined(USE_MOZ_STACK_WALK) +static HANDLE GetThreadHandle(PlatformData* aData); +#endif + +#if defined(USE_FRAME_POINTER_STACK_WALK) || defined(USE_MOZ_STACK_WALK) +static void StackWalkCallback(uint32_t aFrameNumber, void* aPC, void* aSP, + void* aClosure) { + NativeStack* nativeStack = static_cast(aClosure); + MOZ_ASSERT(nativeStack->mCount < MAX_NATIVE_FRAMES); + nativeStack->mSPs[nativeStack->mCount] = aSP; + nativeStack->mPCs[nativeStack->mCount] = aPC; + nativeStack->mCount++; +} +#endif + +#if defined(USE_FRAME_POINTER_STACK_WALK) +static void DoFramePointerBacktrace(PSLockRef aLock, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, + NativeStack& aNativeStack) { + // WARNING: this function runs within the profiler's "critical section". + // WARNING: this function might be called while the profiler is inactive, and + // cannot rely on ActivePS. + + // Start with the current function. We use 0 as the frame number here because + // the FramePointerStackWalk() call below will use 1..N. This is a bit weird + // but it doesn't matter because StackWalkCallback() doesn't use the frame + // number argument. + StackWalkCallback(/* frameNum */ 0, aRegs.mPC, aRegs.mSP, &aNativeStack); + + uint32_t maxFrames = uint32_t(MAX_NATIVE_FRAMES - aNativeStack.mCount); + + const void* stackEnd = aRegisteredThread.StackTop(); + if (aRegs.mFP >= aRegs.mSP && aRegs.mFP <= stackEnd) { + FramePointerStackWalk(StackWalkCallback, maxFrames, &aNativeStack, + reinterpret_cast(aRegs.mFP), + const_cast(stackEnd)); + } +} +#endif + +#if defined(USE_MOZ_STACK_WALK) +static void DoMozStackWalkBacktrace(PSLockRef aLock, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, + NativeStack& aNativeStack) { + // WARNING: this function runs within the profiler's "critical section". + // WARNING: this function might be called while the profiler is inactive, and + // cannot rely on ActivePS. + + // Start with the current function. We use 0 as the frame number here because + // the MozStackWalkThread() call below will use 1..N. This is a bit weird but + // it doesn't matter because StackWalkCallback() doesn't use the frame number + // argument. + StackWalkCallback(/* frameNum */ 0, aRegs.mPC, aRegs.mSP, &aNativeStack); + + uint32_t maxFrames = uint32_t(MAX_NATIVE_FRAMES - aNativeStack.mCount); + + HANDLE thread = GetThreadHandle(aRegisteredThread.GetPlatformData()); + MOZ_ASSERT(thread); + MozStackWalkThread(StackWalkCallback, maxFrames, &aNativeStack, thread, + /* context */ nullptr); +} +#endif + +#ifdef USE_EHABI_STACKWALK +static void DoEHABIBacktrace(PSLockRef aLock, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, + NativeStack& aNativeStack) { + // WARNING: this function runs within the profiler's "critical section". + // WARNING: this function might be called while the profiler is inactive, and + // cannot rely on ActivePS. + + aNativeStack.mCount = + EHABIStackWalk(aRegs.mContext->uc_mcontext, + const_cast(aRegisteredThread.StackTop()), + aNativeStack.mSPs, aNativeStack.mPCs, MAX_NATIVE_FRAMES); +} +#endif + +#ifdef USE_LUL_STACKWALK + +// See the comment at the callsite for why this function is necessary. +# if defined(MOZ_HAVE_ASAN_IGNORE) +MOZ_ASAN_IGNORE static void ASAN_memcpy(void* aDst, const void* aSrc, + size_t aLen) { + // The obvious thing to do here is call memcpy(). However, although + // ASAN_memcpy() is not instrumented by ASAN, memcpy() still is, and the + // false positive still manifests! So we must implement memcpy() ourselves + // within this function. + char* dst = static_cast(aDst); + const char* src = static_cast(aSrc); + + for (size_t i = 0; i < aLen; i++) { + dst[i] = src[i]; + } +} +# endif + +static void DoLULBacktrace(PSLockRef aLock, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, NativeStack& aNativeStack) { + // WARNING: this function runs within the profiler's "critical section". + // WARNING: this function might be called while the profiler is inactive, and + // cannot rely on ActivePS. + + const mcontext_t* mc = &aRegs.mContext->uc_mcontext; + + lul::UnwindRegs startRegs; + memset(&startRegs, 0, sizeof(startRegs)); + +# if defined(GP_PLAT_amd64_linux) || defined(GP_PLAT_amd64_android) + startRegs.xip = lul::TaggedUWord(mc->gregs[REG_RIP]); + startRegs.xsp = lul::TaggedUWord(mc->gregs[REG_RSP]); + startRegs.xbp = lul::TaggedUWord(mc->gregs[REG_RBP]); +# elif defined(GP_PLAT_amd64_freebsd) + startRegs.xip = lul::TaggedUWord(mc->mc_rip); + startRegs.xsp = lul::TaggedUWord(mc->mc_rsp); + startRegs.xbp = lul::TaggedUWord(mc->mc_rbp); +# elif defined(GP_PLAT_arm_linux) || defined(GP_PLAT_arm_android) + startRegs.r15 = lul::TaggedUWord(mc->arm_pc); + startRegs.r14 = lul::TaggedUWord(mc->arm_lr); + startRegs.r13 = lul::TaggedUWord(mc->arm_sp); + startRegs.r12 = lul::TaggedUWord(mc->arm_ip); + startRegs.r11 = lul::TaggedUWord(mc->arm_fp); + startRegs.r7 = lul::TaggedUWord(mc->arm_r7); +# elif defined(GP_PLAT_arm64_linux) || defined(GP_PLAT_arm64_android) + startRegs.pc = lul::TaggedUWord(mc->pc); + startRegs.x29 = lul::TaggedUWord(mc->regs[29]); + startRegs.x30 = lul::TaggedUWord(mc->regs[30]); + startRegs.sp = lul::TaggedUWord(mc->sp); +# elif defined(GP_PLAT_arm64_freebsd) + startRegs.pc = lul::TaggedUWord(mc->mc_gpregs.gp_elr); + startRegs.x29 = lul::TaggedUWord(mc->mc_gpregs.gp_x[29]); + startRegs.x30 = lul::TaggedUWord(mc->mc_gpregs.gp_lr); + startRegs.sp = lul::TaggedUWord(mc->mc_gpregs.gp_sp); +# elif defined(GP_PLAT_x86_linux) || defined(GP_PLAT_x86_android) + startRegs.xip = lul::TaggedUWord(mc->gregs[REG_EIP]); + startRegs.xsp = lul::TaggedUWord(mc->gregs[REG_ESP]); + startRegs.xbp = lul::TaggedUWord(mc->gregs[REG_EBP]); +# elif defined(GP_PLAT_mips64_linux) + startRegs.pc = lul::TaggedUWord(mc->pc); + startRegs.sp = lul::TaggedUWord(mc->gregs[29]); + startRegs.fp = lul::TaggedUWord(mc->gregs[30]); +# else +# error "Unknown plat" +# endif + + // Copy up to N_STACK_BYTES from rsp-REDZONE upwards, but not going past the + // stack's registered top point. Do some basic sanity checks too. This + // assumes that the TaggedUWord holding the stack pointer value is valid, but + // it should be, since it was constructed that way in the code just above. + + // We could construct |stackImg| so that LUL reads directly from the stack in + // question, rather than from a copy of it. That would reduce overhead and + // space use a bit. However, it gives a problem with dynamic analysis tools + // (ASan, TSan, Valgrind) which is that such tools will report invalid or + // racing memory accesses, and such accesses will be reported deep inside LUL. + // By taking a copy here, we can either sanitise the copy (for Valgrind) or + // copy it using an unchecked memcpy (for ASan, TSan). That way we don't have + // to try and suppress errors inside LUL. + // + // N_STACK_BYTES is set to 160KB. This is big enough to hold all stacks + // observed in some minutes of testing, whilst keeping the size of this + // function (DoNativeBacktrace)'s frame reasonable. Most stacks observed in + // practice are small, 4KB or less, and so the copy costs are insignificant + // compared to other profiler overhead. + // + // |stackImg| is allocated on this (the sampling thread's) stack. That + // implies that the frame for this function is at least N_STACK_BYTES large. + // In general it would be considered unacceptable to have such a large frame + // on a stack, but it only exists for the unwinder thread, and so is not + // expected to be a problem. Allocating it on the heap is troublesome because + // this function runs whilst the sampled thread is suspended, so any heap + // allocation risks deadlock. Allocating it as a global variable is not + // thread safe, which would be a problem if we ever allow multiple sampler + // threads. Hence allocating it on the stack seems to be the least-worst + // option. + + lul::StackImage stackImg; + + { +# if defined(GP_PLAT_amd64_linux) || defined(GP_PLAT_amd64_android) || \ + defined(GP_PLAT_amd64_freebsd) + uintptr_t rEDZONE_SIZE = 128; + uintptr_t start = startRegs.xsp.Value() - rEDZONE_SIZE; +# elif defined(GP_PLAT_arm_linux) || defined(GP_PLAT_arm_android) + uintptr_t rEDZONE_SIZE = 0; + uintptr_t start = startRegs.r13.Value() - rEDZONE_SIZE; +# elif defined(GP_PLAT_arm64_linux) || defined(GP_PLAT_arm64_android) || \ + defined(GP_PLAT_arm64_freebsd) + uintptr_t rEDZONE_SIZE = 0; + uintptr_t start = startRegs.sp.Value() - rEDZONE_SIZE; +# elif defined(GP_PLAT_x86_linux) || defined(GP_PLAT_x86_android) + uintptr_t rEDZONE_SIZE = 0; + uintptr_t start = startRegs.xsp.Value() - rEDZONE_SIZE; +# elif defined(GP_PLAT_mips64_linux) + uintptr_t rEDZONE_SIZE = 0; + uintptr_t start = startRegs.sp.Value() - rEDZONE_SIZE; +# else +# error "Unknown plat" +# endif + uintptr_t end = reinterpret_cast(aRegisteredThread.StackTop()); + uintptr_t ws = sizeof(void*); + start &= ~(ws - 1); + end &= ~(ws - 1); + uintptr_t nToCopy = 0; + if (start < end) { + nToCopy = end - start; + if (nToCopy > lul::N_STACK_BYTES) nToCopy = lul::N_STACK_BYTES; + } + MOZ_ASSERT(nToCopy <= lul::N_STACK_BYTES); + stackImg.mLen = nToCopy; + stackImg.mStartAvma = start; + if (nToCopy > 0) { + // If this is a vanilla memcpy(), ASAN makes the following complaint: + // + // ERROR: AddressSanitizer: stack-buffer-underflow ... + // ... + // HINT: this may be a false positive if your program uses some custom + // stack unwind mechanism or swapcontext + // + // This code is very much a custom stack unwind mechanism! So we use an + // alternative memcpy() implementation that is ignored by ASAN. +# if defined(MOZ_HAVE_ASAN_IGNORE) + ASAN_memcpy(&stackImg.mContents[0], (void*)start, nToCopy); +# else + memcpy(&stackImg.mContents[0], (void*)start, nToCopy); +# endif + (void)VALGRIND_MAKE_MEM_DEFINED(&stackImg.mContents[0], nToCopy); + } + } + + size_t framePointerFramesAcquired = 0; + lul::LUL* lul = CorePS::Lul(aLock); + lul->Unwind(reinterpret_cast(aNativeStack.mPCs), + reinterpret_cast(aNativeStack.mSPs), + &aNativeStack.mCount, &framePointerFramesAcquired, + MAX_NATIVE_FRAMES, &startRegs, &stackImg); + + // Update stats in the LUL stats object. Unfortunately this requires + // three global memory operations. + lul->mStats.mContext += 1; + lul->mStats.mCFI += aNativeStack.mCount - 1 - framePointerFramesAcquired; + lul->mStats.mFP += framePointerFramesAcquired; +} + +#endif + +#ifdef HAVE_NATIVE_UNWIND +static void DoNativeBacktrace(PSLockRef aLock, + const RegisteredThread& aRegisteredThread, + const Registers& aRegs, + NativeStack& aNativeStack) { + // This method determines which stackwalker is used for periodic and + // synchronous samples. (Backtrace samples are treated differently, see + // profiler_suspend_and_sample_thread() for details). The only part of the + // ordering that matters is that LUL must precede FRAME_POINTER, because on + // Linux they can both be present. +# if defined(USE_LUL_STACKWALK) + DoLULBacktrace(aLock, aRegisteredThread, aRegs, aNativeStack); +# elif defined(USE_EHABI_STACKWALK) + DoEHABIBacktrace(aLock, aRegisteredThread, aRegs, aNativeStack); +# elif defined(USE_FRAME_POINTER_STACK_WALK) + DoFramePointerBacktrace(aLock, aRegisteredThread, aRegs, aNativeStack); +# elif defined(USE_MOZ_STACK_WALK) + DoMozStackWalkBacktrace(aLock, aRegisteredThread, aRegs, aNativeStack); +# else +# error "Invalid configuration" +# endif +} +#endif + +// Writes some components shared by periodic and synchronous profiles to +// ActivePS's ProfileBuffer. (This should only be called from DoSyncSample() +// and DoPeriodicSample().) +// +// The grammar for entry sequences is in a comment above +// ProfileBuffer::StreamSamplesToJSON. +static inline void DoSharedSample( + PSLockRef aLock, bool aIsSynchronous, RegisteredThread& aRegisteredThread, + const Registers& aRegs, uint64_t aSamplePos, uint64_t aBufferRangeStart, + ProfileBuffer& aBuffer, + StackCaptureOptions aCaptureOptions = StackCaptureOptions::Full) { + // WARNING: this function runs within the profiler's "critical section". + + MOZ_ASSERT(!aBuffer.IsThreadSafe(), + "Mutexes cannot be used inside this critical section"); + + MOZ_RELEASE_ASSERT(ActivePS::Exists(aLock)); + + ProfileBufferCollector collector(aBuffer, aSamplePos, aBufferRangeStart); + NativeStack nativeStack; +#if defined(HAVE_NATIVE_UNWIND) + if (ActivePS::FeatureStackWalk(aLock) && + aCaptureOptions == StackCaptureOptions::Full) { + DoNativeBacktrace(aLock, aRegisteredThread, aRegs, nativeStack); + + MergeStacks(ActivePS::Features(aLock), aIsSynchronous, aRegisteredThread, + aRegs, nativeStack, collector); + } else +#endif + { + MergeStacks(ActivePS::Features(aLock), aIsSynchronous, aRegisteredThread, + aRegs, nativeStack, collector); + + // We can't walk the whole native stack, but we can record the top frame. + if (aCaptureOptions == StackCaptureOptions::Full) { + aBuffer.AddEntry(ProfileBufferEntry::NativeLeafAddr((void*)aRegs.mPC)); + } + } +} + +// Writes the components of a synchronous sample to the given ProfileBuffer. +static void DoSyncSample(PSLockRef aLock, RegisteredThread& aRegisteredThread, + const TimeStamp& aNow, const Registers& aRegs, + ProfileBuffer& aBuffer, + StackCaptureOptions aCaptureOptions) { + // WARNING: this function runs within the profiler's "critical section". + + MOZ_ASSERT(aCaptureOptions != StackCaptureOptions::NoStack, + "DoSyncSample should not be called when no capture is needed"); + + const uint64_t bufferRangeStart = aBuffer.BufferRangeStart(); + + const uint64_t samplePos = + aBuffer.AddThreadIdEntry(aRegisteredThread.Info()->ThreadId()); + + TimeDuration delta = aNow - CorePS::ProcessStartTime(); + aBuffer.AddEntry(ProfileBufferEntry::Time(delta.ToMilliseconds())); + + DoSharedSample(aLock, /* aIsSynchronous = */ true, aRegisteredThread, aRegs, + samplePos, bufferRangeStart, aBuffer, aCaptureOptions); +} + +// Writes the components of a periodic sample to ActivePS's ProfileBuffer. +// The ThreadId entry is already written in the main ProfileBuffer, its location +// is `aSamplePos`, we can write the rest to `aBuffer` (which may be different). +static void DoPeriodicSample(PSLockRef aLock, + RegisteredThread& aRegisteredThread, + ProfiledThreadData& aProfiledThreadData, + const Registers& aRegs, uint64_t aSamplePos, + uint64_t aBufferRangeStart, + ProfileBuffer& aBuffer) { + // WARNING: this function runs within the profiler's "critical section". + + DoSharedSample(aLock, /* aIsSynchronous = */ false, aRegisteredThread, aRegs, + aSamplePos, aBufferRangeStart, aBuffer); +} + +// END sampling/unwinding code +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN saving/streaming code + +const static uint64_t kJS_MAX_SAFE_UINTEGER = +9007199254740991ULL; + +static int64_t SafeJSInteger(uint64_t aValue) { + return aValue <= kJS_MAX_SAFE_UINTEGER ? int64_t(aValue) : -1; +} + +static void AddSharedLibraryInfoToStream(JSONWriter& aWriter, + const SharedLibrary& aLib) { + aWriter.StartObjectElement(); + aWriter.IntProperty("start", SafeJSInteger(aLib.GetStart())); + aWriter.IntProperty("end", SafeJSInteger(aLib.GetEnd())); + aWriter.IntProperty("offset", SafeJSInteger(aLib.GetOffset())); + aWriter.StringProperty("name", aLib.GetModuleName()); + aWriter.StringProperty("path", aLib.GetModulePath()); + aWriter.StringProperty("debugName", aLib.GetDebugName()); + aWriter.StringProperty("debugPath", aLib.GetDebugPath()); + aWriter.StringProperty("breakpadId", aLib.GetBreakpadId()); + aWriter.StringProperty("codeId", aLib.GetCodeId()); + aWriter.StringProperty("arch", aLib.GetArch()); + aWriter.EndObject(); +} + +void AppendSharedLibraries(JSONWriter& aWriter) { + SharedLibraryInfo info = SharedLibraryInfo::GetInfoForSelf(); + info.SortByAddress(); + for (size_t i = 0; i < info.GetSize(); i++) { + AddSharedLibraryInfoToStream(aWriter, info.GetEntry(i)); + } +} + +static void StreamCategories(SpliceableJSONWriter& aWriter) { + // Same order as ProfilingCategory. Format: + // [ + // { + // name: "Idle", + // color: "transparent", + // subcategories: ["Other"], + // }, + // { + // name: "Other", + // color: "grey", + // subcategories: [ + // "JSM loading", + // "Subprocess launching", + // "DLL loading" + // ] + // }, + // ... + // ] + +#define CATEGORY_JSON_BEGIN_CATEGORY(name, labelAsString, color) \ + aWriter.Start(); \ + aWriter.StringProperty("name", labelAsString); \ + aWriter.StringProperty("color", color); \ + aWriter.StartArrayProperty("subcategories"); +#define CATEGORY_JSON_SUBCATEGORY(supercategory, name, labelAsString) \ + aWriter.StringElement(labelAsString); +#define CATEGORY_JSON_END_CATEGORY \ + aWriter.EndArray(); \ + aWriter.EndObject(); + + MOZ_PROFILING_CATEGORY_LIST(CATEGORY_JSON_BEGIN_CATEGORY, + CATEGORY_JSON_SUBCATEGORY, + CATEGORY_JSON_END_CATEGORY) + +#undef CATEGORY_JSON_BEGIN_CATEGORY +#undef CATEGORY_JSON_SUBCATEGORY +#undef CATEGORY_JSON_END_CATEGORY +} + +static void StreamMarkerSchema(SpliceableJSONWriter& aWriter) { + // Get an array view with all registered marker-type-specific functions. + base_profiler_markers_detail::Streaming::LockedMarkerTypeFunctionsList + markerTypeFunctionsArray; + // List of streamed marker names, this is used to spot duplicates. + std::set names; + // Stream the display schema for each different one. (Duplications may come + // from the same code potentially living in different libraries.) + for (const auto& markerTypeFunctions : markerTypeFunctionsArray) { + auto name = markerTypeFunctions.mMarkerTypeNameFunction(); + // std::set.insert(T&&) returns a pair, its `second` is true if the element + // was actually inserted (i.e., it was not there yet.) + const bool didInsert = + names.insert(std::string(name.data(), name.size())).second; + if (didInsert) { + markerTypeFunctions.mMarkerSchemaFunction().Stream(aWriter, name); + } + } +} + +static int64_t MicrosecondsSince1970(); + +static void StreamMetaJSCustomObject(PSLockRef aLock, + SpliceableJSONWriter& aWriter, + bool aIsShuttingDown) { + MOZ_RELEASE_ASSERT(CorePS::Exists() && ActivePS::Exists(aLock)); + + aWriter.IntProperty("version", 27); + + // The "startTime" field holds the number of milliseconds since midnight + // January 1, 1970 GMT. This grotty code computes (Now - (Now - + // ProcessStartTime)) to convert CorePS::ProcessStartTime() into that form. + // Note: This is the only absolute time in the profile! All other timestamps + // are relative to this startTime. + TimeDuration delta = TimeStamp::Now() - CorePS::ProcessStartTime(); + aWriter.DoubleProperty( + "startTime", MicrosecondsSince1970() / 1000.0 - delta.ToMilliseconds()); + + aWriter.DoubleProperty("profilingStartTime", (ActivePS::ProfilingStartTime() - + CorePS::ProcessStartTime()) + .ToMilliseconds()); + + if (const TimeStamp contentEarliestTime = + ActivePS::Buffer(aLock) + .UnderlyingChunkedBuffer() + .GetEarliestChunkStartTimeStamp(); + !contentEarliestTime.IsNull()) { + aWriter.DoubleProperty( + "contentEarliestTime", + (contentEarliestTime - CorePS::ProcessStartTime()).ToMilliseconds()); + } else { + aWriter.NullProperty("contentEarliestTime"); + } + + const double profilingEndTime = profiler_time(); + aWriter.DoubleProperty("profilingEndTime", profilingEndTime); + + if (aIsShuttingDown) { + aWriter.DoubleProperty("shutdownTime", profilingEndTime); + } else { + aWriter.NullProperty("shutdownTime"); + } + + aWriter.StartArrayProperty("categories"); + StreamCategories(aWriter); + aWriter.EndArray(); + + aWriter.StartArrayProperty("markerSchema"); + StreamMarkerSchema(aWriter); + aWriter.EndArray(); + + if (!profiler_is_main_thread()) { + // Leave the rest of the properties out if we're not on the main thread. + // At the moment, the only case in which this function is called on a + // background thread is if we're in a content process and are going to + // send this profile to the parent process. In that case, the parent + // process profile's "meta" object already has the rest of the properties, + // and the parent process profile is dumped on that process's main thread. + return; + } + + aWriter.DoubleProperty("interval", ActivePS::Interval(aLock)); + aWriter.IntProperty("stackwalk", ActivePS::FeatureStackWalk(aLock)); + +#ifdef DEBUG + aWriter.IntProperty("debug", 1); +#else + aWriter.IntProperty("debug", 0); +#endif + + aWriter.IntProperty("gcpoison", 0); + + aWriter.IntProperty("asyncstack", 0); + + aWriter.IntProperty("processType", 0); +} + +static void StreamPages(PSLockRef aLock, SpliceableJSONWriter& aWriter) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + ActivePS::DiscardExpiredPages(aLock); + for (const auto& page : ActivePS::ProfiledPages(aLock)) { + page->StreamJSON(aWriter); + } +} + +static void locked_profiler_stream_json_for_this_process( + PSLockRef aLock, SpliceableJSONWriter& aWriter, double aSinceTime, + bool aIsShuttingDown, bool aOnlyThreads = false) { + LOG("locked_profiler_stream_json_for_this_process"); + + MOZ_RELEASE_ASSERT(CorePS::Exists() && ActivePS::Exists(aLock)); + + AUTO_PROFILER_STATS(base_locked_profiler_stream_json_for_this_process); + + const double collectionStartMs = profiler_time(); + + ProfileBuffer& buffer = ActivePS::Buffer(aLock); + + // If there is a set "Window length", discard older data. + Maybe durationS = ActivePS::Duration(aLock); + if (durationS.isSome()) { + const double durationStartMs = collectionStartMs - *durationS * 1000; + buffer.DiscardSamplesBeforeTime(durationStartMs); + } + + if (!aOnlyThreads) { + // Put shared library info + aWriter.StartArrayProperty("libs"); + AppendSharedLibraries(aWriter); + aWriter.EndArray(); + + // Put meta data + aWriter.StartObjectProperty("meta"); + { StreamMetaJSCustomObject(aLock, aWriter, aIsShuttingDown); } + aWriter.EndObject(); + + // Put page data + aWriter.StartArrayProperty("pages"); + { StreamPages(aLock, aWriter); } + aWriter.EndArray(); + + buffer.StreamProfilerOverheadToJSON(aWriter, CorePS::ProcessStartTime(), + aSinceTime); + buffer.StreamCountersToJSON(aWriter, CorePS::ProcessStartTime(), + aSinceTime); + + // Lists the samples for each thread profile + aWriter.StartArrayProperty("threads"); + } + + // if aOnlyThreads is true, the only output will be the threads array items. + { + ActivePS::DiscardExpiredDeadProfiledThreads(aLock); + Vector> threads = + ActivePS::ProfiledThreads(aLock); + for (auto& thread : threads) { + ProfiledThreadData* profiledThreadData = thread.second; + profiledThreadData->StreamJSON( + buffer, aWriter, CorePS::ProcessName(aLock), CorePS::ETLDplus1(aLock), + CorePS::ProcessStartTime(), aSinceTime); + } + } + + if (!aOnlyThreads) { + aWriter.EndArray(); + + aWriter.StartArrayProperty("pausedRanges"); + { buffer.StreamPausedRangesToJSON(aWriter, aSinceTime); } + aWriter.EndArray(); + } + + const double collectionEndMs = profiler_time(); + + // Record timestamps for the collection into the buffer, so that consumers + // know why we didn't collect any samples for its duration. + // We put these entries into the buffer after we've collected the profile, + // so they'll be visible for the *next* profile collection (if they haven't + // been overwritten due to buffer wraparound by then). + buffer.AddEntry(ProfileBufferEntry::CollectionStart(collectionStartMs)); + buffer.AddEntry(ProfileBufferEntry::CollectionEnd(collectionEndMs)); +} + +bool profiler_stream_json_for_this_process(SpliceableJSONWriter& aWriter, + double aSinceTime, + bool aIsShuttingDown, + bool aOnlyThreads) { + LOG("profiler_stream_json_for_this_process"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return false; + } + + locked_profiler_stream_json_for_this_process(lock, aWriter, aSinceTime, + aIsShuttingDown, aOnlyThreads); + return true; +} + +// END saving/streaming code +//////////////////////////////////////////////////////////////////////// + +static char FeatureCategory(uint32_t aFeature) { + if (aFeature & DefaultFeatures()) { + if (aFeature & AvailableFeatures()) { + return 'D'; + } + return 'd'; + } + + if (aFeature & StartupExtraDefaultFeatures()) { + if (aFeature & AvailableFeatures()) { + return 'S'; + } + return 's'; + } + + if (aFeature & AvailableFeatures()) { + return '-'; + } + return 'x'; +} + +static void PrintUsage() { + PrintToConsole( + "\n" + "Profiler environment variable usage:\n" + "\n" + " MOZ_BASE_PROFILER_HELP\n" + " If set to any value, prints this message.\n" + " (Only BaseProfiler features are known here; Use MOZ_PROFILER_HELP\n" + " for Gecko Profiler help, with more features).\n" + "\n" + " MOZ_BASE_PROFILER_{,DEBUG_,VERBOSE}LOGGING\n" + " Enables BaseProfiler logging to stdout. The levels of logging\n" + " available are MOZ_BASE_PROFILER_LOGGING' (least verbose),\n" + " '..._DEBUG_LOGGING', '..._VERBOSE_LOGGING' (most verbose)\n" + "\n" + " MOZ_PROFILER_STARTUP\n" + " If set to any value other than '' or '0'/'N'/'n', starts the\n" + " profiler immediately on start-up.\n" + " Useful if you want profile code that runs very early.\n" + "\n" + " MOZ_PROFILER_STARTUP_ENTRIES=<%u..%u>\n" + " If MOZ_PROFILER_STARTUP is set, specifies the number of entries\n" + " per process in the profiler's circular buffer when the profiler is\n" + " first started.\n" + " If unset, the platform default is used:\n" + " %u entries per process, or %u when MOZ_PROFILER_STARTUP is set.\n" + " (%u bytes per entry -> %u or %u total bytes per process)\n" + " Optional units in bytes: KB, KiB, MB, MiB, GB, GiB\n" + "\n" + " MOZ_PROFILER_STARTUP_DURATION=<1..>\n" + " If MOZ_PROFILER_STARTUP is set, specifies the maximum life time\n" + " of entries in the the profiler's circular buffer when the profiler\n" + " is first started, in seconds.\n" + " If unset, the life time of the entries will only be restricted by\n" + " MOZ_PROFILER_STARTUP_ENTRIES (or its default value), and no\n" + " additional time duration restriction will be applied.\n" + "\n" + " MOZ_PROFILER_STARTUP_INTERVAL=<1..1000>\n" + " If MOZ_PROFILER_STARTUP is set, specifies the sample interval,\n" + " measured in milliseconds, when the profiler is first started.\n" + " If unset, the platform default is used.\n" + "\n" + " MOZ_PROFILER_STARTUP_FEATURES_BITFIELD=\n" + " If MOZ_PROFILER_STARTUP is set, specifies the profiling\n" + " features, as the integer value of the features bitfield.\n" + " If unset, the value from MOZ_PROFILER_STARTUP_FEATURES is used.\n" + "\n" + " MOZ_PROFILER_STARTUP_FEATURES=\n" + " If MOZ_PROFILER_STARTUP is set, specifies the profiling\n" + " features, as a comma-separated list of strings.\n" + " Ignored if MOZ_PROFILER_STARTUP_FEATURES_BITFIELD is set.\n" + " If unset, the platform default is used.\n" + "\n" + " Features: (x=unavailable, D/d=default/unavailable,\n" + " S/s=MOZ_PROFILER_STARTUP extra " + "default/unavailable)\n", + unsigned(ActivePS::scMinimumBufferEntries), + unsigned(ActivePS::scMaximumBufferEntries), + unsigned(BASE_PROFILER_DEFAULT_ENTRIES.Value()), + unsigned(BASE_PROFILER_DEFAULT_STARTUP_ENTRIES.Value()), + unsigned(scBytesPerEntry), + unsigned(BASE_PROFILER_DEFAULT_ENTRIES.Value() * scBytesPerEntry), + unsigned(BASE_PROFILER_DEFAULT_STARTUP_ENTRIES.Value() * + scBytesPerEntry)); + +#define PRINT_FEATURE(n_, str_, Name_, desc_) \ + PrintToConsole(" %c %7u: \"%s\" (%s)\n", \ + FeatureCategory(ProfilerFeature::Name_), \ + ProfilerFeature::Name_, str_, desc_); + + BASE_PROFILER_FOR_EACH_FEATURE(PRINT_FEATURE) + +#undef PRINT_FEATURE + + PrintToConsole( + " - \"default\" (All above D+S defaults)\n" + "\n" + " MOZ_PROFILER_STARTUP_FILTERS=\n" + " If MOZ_PROFILER_STARTUP is set, specifies the thread filters, as " + "a\n" + " comma-separated list of strings. A given thread will be sampled if\n" + " any of the filters is a case-insensitive substring of the thread\n" + " name. If unset, a default is used.\n" + "\n" + " MOZ_PROFILER_SHUTDOWN\n" + " If set, the profiler saves a profile to the named file on shutdown.\n" + "\n" + " MOZ_PROFILER_SYMBOLICATE\n" + " If set, the profiler will pre-symbolicate profiles.\n" + " *Note* This will add a significant pause when gathering data, and\n" + " is intended mainly for local development.\n" + "\n" + " MOZ_PROFILER_LUL_TEST\n" + " If set to any value, runs LUL unit tests at startup.\n" + "\n" + " This platform %s native unwinding.\n" + "\n", +#if defined(HAVE_NATIVE_UNWIND) + "supports" +#else + "does not support" +#endif + ); +} + +//////////////////////////////////////////////////////////////////////// +// BEGIN Sampler + +#if defined(GP_OS_linux) || defined(GP_OS_android) +struct SigHandlerCoordinator; +#endif + +// Sampler performs setup and teardown of the state required to sample with the +// profiler. Sampler may exist when ActivePS is not present. +// +// SuspendAndSampleAndResumeThread must only be called from a single thread, +// and must not sample the thread it is being called from. A separate Sampler +// instance must be used for each thread which wants to capture samples. + +// WARNING WARNING WARNING WARNING WARNING WARNING WARNING WARNING +// +// With the exception of SamplerThread, all Sampler objects must be Disable-d +// before releasing the lock which was used to create them. This avoids races +// on linux with the SIGPROF signal handler. + +class Sampler { + public: + // Sets up the profiler such that it can begin sampling. + explicit Sampler(PSLockRef aLock); + + // Disable the sampler, restoring it to its previous state. This must be + // called once, and only once, before the Sampler is destroyed. + void Disable(PSLockRef aLock); + + // This method suspends and resumes the samplee thread. It calls the passed-in + // function-like object aProcessRegs (passing it a populated |const + // Registers&| arg) while the samplee thread is suspended. + // + // Func must be a function-like object of type `void()`. + template + void SuspendAndSampleAndResumeThread( + PSLockRef aLock, const RegisteredThread& aRegisteredThread, + const TimeStamp& aNow, const Func& aProcessRegs); + + private: +#if defined(GP_OS_linux) || defined(GP_OS_android) || defined(GP_OS_freebsd) + // Used to restore the SIGPROF handler when ours is removed. + struct sigaction mOldSigprofHandler; + + // This process' ID. Needed as an argument for tgkill in + // SuspendAndSampleAndResumeThread. + BaseProfilerProcessId mMyPid; + + // The sampler thread's ID. Used to assert that it is not sampling itself, + // which would lead to deadlock. + BaseProfilerThreadId mSamplerTid; + + public: + // This is the one-and-only variable used to communicate between the sampler + // thread and the samplee thread's signal handler. It's static because the + // samplee thread's signal handler is static. + static struct SigHandlerCoordinator* sSigHandlerCoordinator; +#endif +}; + +// END Sampler +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN SamplerThread + +// The sampler thread controls sampling and runs whenever the profiler is +// active. It periodically runs through all registered threads, finds those +// that should be sampled, then pauses and samples them. + +class SamplerThread { + public: + // Creates a sampler thread, but doesn't start it. + SamplerThread(PSLockRef aLock, uint32_t aActivityGeneration, + double aIntervalMilliseconds, uint32_t aFeatures); + ~SamplerThread(); + + // This runs on (is!) the sampler thread. + void Run(); + + // This runs on the main thread. + void Stop(PSLockRef aLock); + + private: + // This suspends the calling thread for the given number of microseconds. + // Best effort timing. + void SleepMicro(uint32_t aMicroseconds); + + // The sampler used to suspend and sample threads. + Sampler mSampler; + + // The activity generation, for detecting when the sampler thread must stop. + const uint32_t mActivityGeneration; + + // The interval between samples, measured in microseconds. + const int mIntervalMicroseconds; + + // The OS-specific handle for the sampler thread. +#if defined(GP_OS_windows) + HANDLE mThread; +#elif defined(GP_OS_darwin) || defined(GP_OS_linux) || \ + defined(GP_OS_android) || defined(GP_OS_freebsd) + pthread_t mThread; +#endif + +#if defined(GP_OS_windows) + bool mNoTimerResolutionChange = true; +#endif + + SamplerThread(const SamplerThread&) = delete; + void operator=(const SamplerThread&) = delete; +}; + +// This function is required because we need to create a SamplerThread within +// ActivePS's constructor, but SamplerThread is defined after ActivePS. It +// could probably be removed by moving some code around. +static SamplerThread* NewSamplerThread(PSLockRef aLock, uint32_t aGeneration, + double aInterval, uint32_t aFeatures) { + return new SamplerThread(aLock, aGeneration, aInterval, aFeatures); +} + +// This function is the sampler thread. This implementation is used for all +// targets. +void SamplerThread::Run() { + // TODO: If possible, name this thread later on, after NSPR becomes available. + // PR_SetCurrentThreadName("SamplerThread"); + + // Features won't change during this SamplerThread's lifetime, so we can read + // them once and store them locally. + const uint32_t features = []() -> uint32_t { + PSAutoLock lock; + if (!ActivePS::Exists(lock)) { + // If there is no active profiler, it doesn't matter what we return, + // because this thread will exit before any feature is used. + return 0; + } + return ActivePS::Features(lock); + }(); + + // Not *no*-stack-sampling means we do want stack sampling. + const bool stackSampling = !ProfilerFeature::HasNoStackSampling(features); + + // Use local BlocksRingBuffer&ProfileBuffer to capture the stack. + // (This is to avoid touching the CorePS::CoreBuffer lock while + // a thread is suspended, because that thread could be working with + // the CorePS::CoreBuffer as well.) + ProfileBufferChunkManagerSingle localChunkManager( + ProfileBufferChunkManager::scExpectedMaximumStackSize); + ProfileChunkedBuffer localBuffer( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex, localChunkManager); + ProfileBuffer localProfileBuffer(localBuffer); + + // Will be kept between collections, to know what each collection does. + auto previousState = localBuffer.GetState(); + + // This will be positive if we are running behind schedule (sampling less + // frequently than desired) and negative if we are ahead of schedule. + TimeDuration lastSleepOvershoot = 0; + TimeStamp sampleStart = TimeStamp::Now(); + + while (true) { + // This scope is for |lock|. It ends before we sleep below. + { + PSAutoLock lock; + TimeStamp lockAcquired = TimeStamp::Now(); + + if (!ActivePS::Exists(lock)) { + return; + } + + // At this point profiler_stop() might have been called, and + // profiler_start() might have been called on another thread. If this + // happens the generation won't match. + if (ActivePS::Generation(lock) != mActivityGeneration) { + return; + } + + ActivePS::ClearExpiredExitProfiles(lock); + + TimeStamp expiredMarkersCleaned = TimeStamp::Now(); + + if (int(gSkipSampling) <= 0 && !ActivePS::IsSamplingPaused(lock)) { + TimeDuration delta = sampleStart - CorePS::ProcessStartTime(); + ProfileBuffer& buffer = ActivePS::Buffer(lock); + + // handle per-process generic counters + const Vector& counters = CorePS::Counters(lock); + for (auto& counter : counters) { + // create Buffer entries for each counter + buffer.AddEntry(ProfileBufferEntry::CounterId(counter)); + buffer.AddEntry(ProfileBufferEntry::Time(delta.ToMilliseconds())); + // XXX support keyed maps of counts + // In the future, we'll support keyed counters - for example, counters + // with a key which is a thread ID. For "simple" counters we'll just + // use a key of 0. + int64_t count; + uint64_t number; + counter->Sample(count, number); + buffer.AddEntry(ProfileBufferEntry::CounterKey(0)); + buffer.AddEntry(ProfileBufferEntry::Count(count)); + if (number) { + buffer.AddEntry(ProfileBufferEntry::Number(number)); + } + } + TimeStamp countersSampled = TimeStamp::Now(); + + if (stackSampling) { + const Vector& liveThreads = + ActivePS::LiveProfiledThreads(lock); + + for (auto& thread : liveThreads) { + RegisteredThread* registeredThread = thread.mRegisteredThread; + ProfiledThreadData* profiledThreadData = + thread.mProfiledThreadData.get(); + RefPtr info = registeredThread->Info(); + + // If the thread is asleep and has been sampled before in the same + // sleep episode, find and copy the previous sample, as that's + // cheaper than taking a new sample. + if (registeredThread->RacyRegisteredThread() + .CanDuplicateLastSampleDueToSleep()) { + bool dup_ok = ActivePS::Buffer(lock).DuplicateLastSample( + info->ThreadId(), CorePS::ProcessStartTime(), + profiledThreadData->LastSample()); + if (dup_ok) { + continue; + } + } + + AUTO_PROFILER_STATS(base_SamplerThread_Run_DoPeriodicSample); + + TimeStamp now = TimeStamp::Now(); + + // Record the global profiler buffer's range start now, before + // adding the first entry for this thread's sample. + const uint64_t bufferRangeStart = buffer.BufferRangeStart(); + + // Add the thread ID now, so we know its position in the main + // buffer, which is used by some JS data. (DoPeriodicSample only + // knows about the temporary local buffer.) + const uint64_t samplePos = + buffer.AddThreadIdEntry(registeredThread->Info()->ThreadId()); + profiledThreadData->LastSample() = Some(samplePos); + + // Also add the time, so it's always there after the thread ID, as + // expected by the parser. (Other stack data is optional.) + TimeDuration delta = now - CorePS::ProcessStartTime(); + buffer.AddEntry(ProfileBufferEntry::Time(delta.ToMilliseconds())); + + mSampler.SuspendAndSampleAndResumeThread( + lock, *registeredThread, now, + [&](const Registers& aRegs, const TimeStamp& aNow) { + DoPeriodicSample(lock, *registeredThread, *profiledThreadData, + aRegs, samplePos, bufferRangeStart, + localProfileBuffer); + }); + + // If data is complete, copy it into the global buffer. + auto state = localBuffer.GetState(); + if (state.mClearedBlockCount != previousState.mClearedBlockCount) { + LOG("Stack sample too big for local storage, needed %u bytes", + unsigned(state.mRangeEnd - previousState.mRangeEnd)); + } else if (state.mRangeEnd - previousState.mRangeEnd >= + *profiler_get_core_buffer().BufferLength()) { + LOG("Stack sample too big for profiler storage, needed %u bytes", + unsigned(state.mRangeEnd - previousState.mRangeEnd)); + } else { + profiler_get_core_buffer().AppendContents(localBuffer); + } + + // Clean up for the next run. + localBuffer.Clear(); + previousState = localBuffer.GetState(); + } + } + +#if defined(USE_LUL_STACKWALK) + // The LUL unwind object accumulates frame statistics. Periodically we + // should poke it to give it a chance to print those statistics. This + // involves doing I/O (fprintf, __android_log_print, etc.) and so + // can't safely be done from the critical section inside + // SuspendAndSampleAndResumeThread, which is why it is done here. + lul::LUL* lul = CorePS::Lul(lock); + if (lul) { + lul->MaybeShowStats(); + } +#endif + TimeStamp threadsSampled = TimeStamp::Now(); + + { + AUTO_PROFILER_STATS(Sampler_FulfillChunkRequests); + ActivePS::FulfillChunkRequests(lock); + } + + buffer.CollectOverheadStats(delta, lockAcquired - sampleStart, + expiredMarkersCleaned - lockAcquired, + countersSampled - expiredMarkersCleaned, + threadsSampled - countersSampled); + } + } + // gPSMutex is not held after this point. + + // Calculate how long a sleep to request. After the sleep, measure how + // long we actually slept and take the difference into account when + // calculating the sleep interval for the next iteration. This is an + // attempt to keep "to schedule" in the presence of inaccuracy of the + // actual sleep intervals. + TimeStamp targetSleepEndTime = + sampleStart + TimeDuration::FromMicroseconds(mIntervalMicroseconds); + TimeStamp beforeSleep = TimeStamp::Now(); + TimeDuration targetSleepDuration = targetSleepEndTime - beforeSleep; + double sleepTime = std::max( + 0.0, (targetSleepDuration - lastSleepOvershoot).ToMicroseconds()); + SleepMicro(static_cast(sleepTime)); + sampleStart = TimeStamp::Now(); + lastSleepOvershoot = + sampleStart - (beforeSleep + TimeDuration::FromMicroseconds(sleepTime)); + } +} + +// Temporary closing namespaces from enclosing platform.cpp. +} // namespace baseprofiler +} // namespace mozilla + +// We #include these files directly because it means those files can use +// declarations from this file trivially. These provide target-specific +// implementations of all SamplerThread methods except Run(). +#if defined(GP_OS_windows) +# include "platform-win32.cpp" +#elif defined(GP_OS_darwin) +# include "platform-macos.cpp" +#elif defined(GP_OS_linux) || defined(GP_OS_android) || defined(GP_OS_freebsd) +# include "platform-linux-android.cpp" +#else +# error "bad platform" +#endif + +namespace mozilla { +namespace baseprofiler { + +UniquePlatformData AllocPlatformData(BaseProfilerThreadId aThreadId) { + return UniquePlatformData(new PlatformData(aThreadId)); +} + +void PlatformDataDestructor::operator()(PlatformData* aData) { delete aData; } + +// END SamplerThread +//////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////// +// BEGIN externally visible functions + +static uint32_t ParseFeature(const char* aFeature, bool aIsStartup) { + if (strcmp(aFeature, "default") == 0) { + return (aIsStartup ? (DefaultFeatures() | StartupExtraDefaultFeatures()) + : DefaultFeatures()) & + AvailableFeatures(); + } + +#define PARSE_FEATURE_BIT(n_, str_, Name_, desc_) \ + if (strcmp(aFeature, str_) == 0) { \ + return ProfilerFeature::Name_; \ + } + + BASE_PROFILER_FOR_EACH_FEATURE(PARSE_FEATURE_BIT) + +#undef PARSE_FEATURE_BIT + + PrintToConsole("\nUnrecognized feature \"%s\".\n\n", aFeature); + // Since we may have an old feature we don't implement anymore, don't exit. + PrintUsage(); + return 0; +} + +uint32_t ParseFeaturesFromStringArray(const char** aFeatures, + uint32_t aFeatureCount, + bool aIsStartup /* = false */) { + uint32_t features = 0; + for (size_t i = 0; i < aFeatureCount; i++) { + features |= ParseFeature(aFeatures[i], aIsStartup); + } + return features; +} + +// Find the RegisteredThread for the current thread. This should only be called +// in places where TLSRegisteredThread can't be used. +static RegisteredThread* FindCurrentThreadRegisteredThread(PSLockRef aLock) { + BaseProfilerThreadId id = profiler_current_thread_id(); + const Vector>& registeredThreads = + CorePS::RegisteredThreads(aLock); + for (auto& registeredThread : registeredThreads) { + if (registeredThread->Info()->ThreadId() == id) { + return registeredThread.get(); + } + } + + return nullptr; +} + +static ProfilingStack* locked_register_thread(PSLockRef aLock, + const char* aName, + void* aStackTop) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + MOZ_ASSERT(!FindCurrentThreadRegisteredThread(aLock)); + + VTUNE_REGISTER_THREAD(aName); + + if (!TLSRegisteredThread::Init(aLock)) { + return nullptr; + } + + RefPtr info = new ThreadInfo(aName, profiler_current_thread_id(), + profiler_is_main_thread()); + UniquePtr registeredThread = + MakeUnique(info, aStackTop); + + TLSRegisteredThread::SetRegisteredThread(aLock, registeredThread.get()); + + if (ActivePS::Exists(aLock) && ActivePS::ShouldProfileThread(aLock, info)) { + registeredThread->RacyRegisteredThread().SetIsBeingProfiled(true); + ActivePS::AddLiveProfiledThread(aLock, registeredThread.get(), + MakeUnique(info)); + } + + ProfilingStack* profilingStack = + ®isteredThread->RacyRegisteredThread().ProfilingStack(); + + CorePS::AppendRegisteredThread(aLock, std::move(registeredThread)); + + return profilingStack; +} + +static void locked_profiler_start(PSLockRef aLock, PowerOfTwo32 aCapacity, + double aInterval, uint32_t aFeatures, + const char** aFilters, uint32_t aFilterCount, + const Maybe& aDuration); + +static Vector SplitAtCommas(const char* aString, + UniquePtr& aStorage) { + size_t len = strlen(aString); + aStorage = MakeUnique(len + 1); + PodCopy(aStorage.get(), aString, len + 1); + + // Iterate over all characters in aStorage and split at commas, by + // overwriting commas with the null char. + Vector array; + size_t currentElementStart = 0; + for (size_t i = 0; i <= len; i++) { + if (aStorage[i] == ',') { + aStorage[i] = '\0'; + } + if (aStorage[i] == '\0') { + // Only add non-empty elements, otherwise ParseFeatures would later + // complain about unrecognized features. + if (currentElementStart != i) { + MOZ_RELEASE_ASSERT(array.append(&aStorage[currentElementStart])); + } + currentElementStart = i + 1; + } + } + return array; +} + +static const char* get_size_suffix(const char* str) { + const char* ptr = str; + + while (isdigit(*ptr)) { + ptr++; + } + + return ptr; +} + +void profiler_init(void* aStackTop) { + LOG("profiler_init"); + + profiler_init_main_thread_id(); + + VTUNE_INIT(); + + MOZ_RELEASE_ASSERT(!CorePS::Exists()); + + if (getenv("MOZ_BASE_PROFILER_HELP")) { + PrintUsage(); + exit(0); + } + + SharedLibraryInfo::Initialize(); + + uint32_t features = DefaultFeatures() & AvailableFeatures(); + + UniquePtr filterStorage; + + Vector filters; + MOZ_RELEASE_ASSERT(filters.append(kMainThreadName)); + + PowerOfTwo32 capacity = BASE_PROFILER_DEFAULT_ENTRIES; + Maybe duration = Nothing(); + double interval = BASE_PROFILER_DEFAULT_INTERVAL; + + { + PSAutoLock lock; + + // We've passed the possible failure point. Instantiate CorePS, which + // indicates that the profiler has initialized successfully. + CorePS::Create(lock); + + Unused << locked_register_thread(lock, kMainThreadName, aStackTop); + + // Platform-specific initialization. + PlatformInit(lock); + + // (Linux-only) We could create CorePS::mLul and read unwind info into it + // at this point. That would match the lifetime implied by destruction of + // it in profiler_shutdown() just below. However, that gives a big delay on + // startup, even if no profiling is actually to be done. So, instead, it is + // created on demand at the first call to PlatformStart(). + + const char* startupEnv = getenv("MOZ_PROFILER_STARTUP"); + if (!startupEnv || startupEnv[0] == '\0' || + ((startupEnv[0] == '0' || startupEnv[0] == 'N' || + startupEnv[0] == 'n') && + startupEnv[1] == '\0')) { + return; + } + + // Hidden option to stop Base Profiler, mostly due to Talos intermittents, + // see https://bugzilla.mozilla.org/show_bug.cgi?id=1638851#c3 + // TODO: Investigate root cause and remove this in bugs 1648324 and 1648325. + if (getenv("MOZ_PROFILER_STARTUP_NO_BASE")) { + return; + } + + LOG("- MOZ_PROFILER_STARTUP is set"); + + // Startup default capacity may be different. + capacity = BASE_PROFILER_DEFAULT_STARTUP_ENTRIES; + + const char* startupCapacity = getenv("MOZ_PROFILER_STARTUP_ENTRIES"); + if (startupCapacity && startupCapacity[0] != '\0') { + errno = 0; + long capacityLong = strtol(startupCapacity, nullptr, 10); + std::string_view sizeSuffix = get_size_suffix(startupCapacity); + + if (sizeSuffix == "KB") { + capacityLong *= 1000 / scBytesPerEntry; + } else if (sizeSuffix == "KiB") { + capacityLong *= 1024 / scBytesPerEntry; + } else if (sizeSuffix == "MB") { + capacityLong *= (1000 * 1000) / scBytesPerEntry; + } else if (sizeSuffix == "MiB") { + capacityLong *= (1024 * 1024) / scBytesPerEntry; + } else if (sizeSuffix == "GB") { + capacityLong *= (1000 * 1000 * 1000) / scBytesPerEntry; + } else if (sizeSuffix == "GiB") { + capacityLong *= (1024 * 1024 * 1024) / scBytesPerEntry; + } else if (!sizeSuffix.empty()) { + PrintToConsole( + "- MOZ_PROFILER_STARTUP_ENTRIES unit must be one of the " + "following: KB, KiB, MB, MiB, GB, GiB"); + PrintUsage(); + exit(1); + } + + // `long` could be 32 or 64 bits, so we force a 64-bit comparison with + // the maximum 32-bit signed number (as more than that is clamped down to + // 2^31 anyway). + if (errno == 0 && capacityLong > 0 && + static_cast(capacityLong) <= + static_cast(INT32_MAX)) { + capacity = PowerOfTwo32(ActivePS::ClampToAllowedEntries( + static_cast(capacityLong))); + LOG("- MOZ_PROFILER_STARTUP_ENTRIES = %u", unsigned(capacity.Value())); + } else { + PrintToConsole("- MOZ_PROFILER_STARTUP_ENTRIES not a valid integer: %s", + startupCapacity); + PrintUsage(); + exit(1); + } + } + + const char* startupDuration = getenv("MOZ_PROFILER_STARTUP_DURATION"); + if (startupDuration && startupDuration[0] != '\0') { + // The duration is a floating point number. Use StringToDouble rather than + // strtod, so that "." is used as the decimal separator regardless of OS + // locale. + auto durationVal = StringToDouble(std::string(startupDuration)); + if (durationVal && *durationVal >= 0.0) { + if (*durationVal > 0.0) { + duration = Some(*durationVal); + } + LOG("- MOZ_PROFILER_STARTUP_DURATION = %f", *durationVal); + } else { + PrintToConsole("- MOZ_PROFILER_STARTUP_DURATION not a valid float: %s", + startupDuration); + PrintUsage(); + exit(1); + } + } + + const char* startupInterval = getenv("MOZ_PROFILER_STARTUP_INTERVAL"); + if (startupInterval && startupInterval[0] != '\0') { + // The interval is a floating point number. Use StringToDouble rather than + // strtod, so that "." is used as the decimal separator regardless of OS + // locale. + auto intervalValue = StringToDouble(MakeStringSpan(startupInterval)); + if (intervalValue && *intervalValue > 0.0 && *intervalValue <= 1000.0) { + interval = *intervalValue; + LOG("- MOZ_PROFILER_STARTUP_INTERVAL = %f", interval); + } else { + PrintToConsole("- MOZ_PROFILER_STARTUP_INTERVAL not a valid float: %s", + startupInterval); + PrintUsage(); + exit(1); + } + } + + features |= StartupExtraDefaultFeatures() & AvailableFeatures(); + + const char* startupFeaturesBitfield = + getenv("MOZ_PROFILER_STARTUP_FEATURES_BITFIELD"); + if (startupFeaturesBitfield && startupFeaturesBitfield[0] != '\0') { + errno = 0; + features = strtol(startupFeaturesBitfield, nullptr, 10); + if (errno == 0) { + LOG("- MOZ_PROFILER_STARTUP_FEATURES_BITFIELD = %d", features); + } else { + PrintToConsole( + "- MOZ_PROFILER_STARTUP_FEATURES_BITFIELD not a valid integer: %s", + startupFeaturesBitfield); + PrintUsage(); + exit(1); + } + } else { + const char* startupFeatures = getenv("MOZ_PROFILER_STARTUP_FEATURES"); + if (startupFeatures) { + // Interpret startupFeatures as a list of feature strings, separated by + // commas. + UniquePtr featureStringStorage; + Vector featureStringArray = + SplitAtCommas(startupFeatures, featureStringStorage); + features = ParseFeaturesFromStringArray(featureStringArray.begin(), + featureStringArray.length(), + /* aIsStartup */ true); + LOG("- MOZ_PROFILER_STARTUP_FEATURES = %d", features); + } + } + + const char* startupFilters = getenv("MOZ_PROFILER_STARTUP_FILTERS"); + if (startupFilters && startupFilters[0] != '\0') { + filters = SplitAtCommas(startupFilters, filterStorage); + LOG("- MOZ_PROFILER_STARTUP_FILTERS = %s", startupFilters); + + if (mozilla::profiler::detail::FiltersExcludePid(filters)) { + LOG(" -> This process is excluded and won't be profiled"); + return; + } + } + + locked_profiler_start(lock, capacity, interval, features, filters.begin(), + filters.length(), duration); + } + + // TODO: Install memory counter if it is possible from mozglue. + // #if defined(MOZ_REPLACE_MALLOC) && defined(MOZ_PROFILER_MEMORY) + // // start counting memory allocations (outside of lock because this may + // call + // // profiler_add_sampled_counter which would attempt to take the lock.) + // mozilla::profiler::install_memory_counter(true); + // #endif +} + +static void locked_profiler_save_profile_to_file(PSLockRef aLock, + const char* aFilename, + bool aIsShuttingDown); + +static SamplerThread* locked_profiler_stop(PSLockRef aLock); + +void profiler_shutdown() { + LOG("profiler_shutdown"); + + VTUNE_SHUTDOWN(); + + MOZ_RELEASE_ASSERT(profiler_is_main_thread()); + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + // If the profiler is active we must get a handle to the SamplerThread before + // ActivePS is destroyed, in order to delete it. + SamplerThread* samplerThread = nullptr; + { + PSAutoLock lock; + + // Save the profile on shutdown if requested. + if (ActivePS::Exists(lock)) { + const char* filename = getenv("MOZ_PROFILER_SHUTDOWN"); + if (filename && filename[0] != '\0') { + locked_profiler_save_profile_to_file(lock, filename, + /* aIsShuttingDown */ true); + } + + samplerThread = locked_profiler_stop(lock); + } + + CorePS::Destroy(lock); + + // We just destroyed CorePS and the ThreadInfos it contains, so we can + // clear this thread's TLSRegisteredThread. + TLSRegisteredThread::SetRegisteredThread(lock, nullptr); + } + + // We do these operations with gPSMutex unlocked. The comments in + // profiler_stop() explain why. + if (samplerThread) { + delete samplerThread; + } +} + +static bool WriteProfileToJSONWriter(SpliceableChunkedJSONWriter& aWriter, + double aSinceTime, bool aIsShuttingDown, + bool aOnlyThreads = false) { + LOG("WriteProfileToJSONWriter"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + if (!aOnlyThreads) { + aWriter.Start(); + { + if (!profiler_stream_json_for_this_process( + aWriter, aSinceTime, aIsShuttingDown, aOnlyThreads)) { + return false; + } + + // Don't include profiles from other processes because this is a + // synchronous function. + aWriter.StartArrayProperty("processes"); + aWriter.EndArray(); + } + aWriter.End(); + } else { + aWriter.StartBareList(); + if (!profiler_stream_json_for_this_process(aWriter, aSinceTime, + aIsShuttingDown, aOnlyThreads)) { + return false; + } + aWriter.EndBareList(); + } + return true; +} + +void profiler_set_process_name(const std::string& aProcessName, + const std::string* aETLDplus1) { + LOG("profiler_set_process_name(\"%s\", \"%s\")", aProcessName.c_str(), + aETLDplus1 ? aETLDplus1->c_str() : ""); + PSAutoLock lock; + CorePS::SetProcessName(lock, aProcessName); + if (aETLDplus1) { + CorePS::SetETLDplus1(lock, *aETLDplus1); + } +} + +UniquePtr profiler_get_profile(double aSinceTime, bool aIsShuttingDown, + bool aOnlyThreads) { + LOG("profiler_get_profile"); + + SpliceableChunkedJSONWriter b{FailureLatchInfallibleSource::Singleton()}; + if (!WriteProfileToJSONWriter(b, aSinceTime, aIsShuttingDown, aOnlyThreads)) { + return nullptr; + } + return b.ChunkedWriteFunc().CopyData(); +} + +void profiler_get_start_params(int* aCapacity, Maybe* aDuration, + double* aInterval, uint32_t* aFeatures, + Vector* aFilters) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + if (!aCapacity || !aDuration || !aInterval || !aFeatures || !aFilters) { + return; + } + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + *aCapacity = 0; + *aDuration = Nothing(); + *aInterval = 0; + *aFeatures = 0; + aFilters->clear(); + return; + } + + *aCapacity = ActivePS::Capacity(lock).Value(); + *aDuration = ActivePS::Duration(lock); + *aInterval = ActivePS::Interval(lock); + *aFeatures = ActivePS::Features(lock); + + const Vector& filters = ActivePS::Filters(lock); + MOZ_ALWAYS_TRUE(aFilters->resize(filters.length())); + for (uint32_t i = 0; i < filters.length(); ++i) { + (*aFilters)[i] = filters[i].c_str(); + } +} + +void GetProfilerEnvVarsForChildProcess( + std::function&& aSetEnv) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + aSetEnv("MOZ_PROFILER_STARTUP", ""); + return; + } + + aSetEnv("MOZ_PROFILER_STARTUP", "1"); + auto capacityString = + Smprintf("%u", unsigned(ActivePS::Capacity(lock).Value())); + aSetEnv("MOZ_PROFILER_STARTUP_ENTRIES", capacityString.get()); + + // Use AppendFloat instead of Smprintf with %f because the decimal + // separator used by %f is locale-dependent. But the string we produce needs + // to be parseable by strtod, which only accepts the period character as a + // decimal separator. AppendFloat always uses the period character. + std::string intervalString = std::to_string(ActivePS::Interval(lock)); + aSetEnv("MOZ_PROFILER_STARTUP_INTERVAL", intervalString.c_str()); + + auto featuresString = Smprintf("%d", ActivePS::Features(lock)); + aSetEnv("MOZ_PROFILER_STARTUP_FEATURES_BITFIELD", featuresString.get()); + + std::string filtersString; + const Vector& filters = ActivePS::Filters(lock); + for (uint32_t i = 0; i < filters.length(); ++i) { + filtersString += filters[i]; + if (i != filters.length() - 1) { + filtersString += ","; + } + } + aSetEnv("MOZ_PROFILER_STARTUP_FILTERS", filtersString.c_str()); +} + +void profiler_received_exit_profile(const std::string& aExitProfile) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + PSAutoLock lock; + if (!ActivePS::Exists(lock)) { + return; + } + ActivePS::AddExitProfile(lock, aExitProfile); +} + +Vector profiler_move_exit_profiles() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + PSAutoLock lock; + Vector profiles; + if (ActivePS::Exists(lock)) { + profiles = ActivePS::MoveExitProfiles(lock); + } + return profiles; +} + +static void locked_profiler_save_profile_to_file(PSLockRef aLock, + const char* aFilename, + bool aIsShuttingDown = false) { + LOG("locked_profiler_save_profile_to_file(%s)", aFilename); + + MOZ_RELEASE_ASSERT(CorePS::Exists() && ActivePS::Exists(aLock)); + + std::ofstream stream; + stream.open(aFilename); + if (stream.is_open()) { + OStreamJSONWriteFunc jw(stream); + SpliceableJSONWriter w(jw, FailureLatchInfallibleSource::Singleton()); + w.Start(); + { + locked_profiler_stream_json_for_this_process(aLock, w, /* sinceTime */ 0, + aIsShuttingDown); + + w.StartArrayProperty("processes"); + Vector exitProfiles = ActivePS::MoveExitProfiles(aLock); + for (auto& exitProfile : exitProfiles) { + if (!exitProfile.empty()) { + w.Splice(exitProfile); + } + } + w.EndArray(); + } + w.End(); + + stream.close(); + } +} + +void baseprofiler_save_profile_to_file(const char* aFilename) { + LOG("baseprofiler_save_profile_to_file(%s)", aFilename); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + locked_profiler_save_profile_to_file(lock, aFilename); +} + +uint32_t profiler_get_available_features() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + return AvailableFeatures(); +} + +Maybe profiler_get_buffer_info() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return Nothing(); + } + + return Some(ActivePS::Buffer(lock).GetProfilerBufferInfo()); +} + +// This basically duplicates AutoProfilerLabel's constructor. +static void* MozGlueBaseLabelEnter(const char* aLabel, + const char* aDynamicString, void* aSp) { + ProfilingStack* profilingStack = AutoProfilerLabel::sProfilingStack.get(); + if (profilingStack) { + profilingStack->pushLabelFrame(aLabel, aDynamicString, aSp, + ProfilingCategoryPair::OTHER); + } + return profilingStack; +} + +// This basically duplicates AutoProfilerLabel's destructor. +static void MozGlueBaseLabelExit(void* sProfilingStack) { + if (sProfilingStack) { + reinterpret_cast(sProfilingStack)->pop(); + } +} + +static void locked_profiler_start(PSLockRef aLock, PowerOfTwo32 aCapacity, + double aInterval, uint32_t aFeatures, + const char** aFilters, uint32_t aFilterCount, + const Maybe& aDuration) { + const TimeStamp profilingStartTime = TimeStamp::Now(); + + if (LOG_TEST) { + LOG("locked_profiler_start"); + LOG("- capacity = %d", int(aCapacity.Value())); + LOG("- duration = %.2f", aDuration ? *aDuration : -1); + LOG("- interval = %.2f", aInterval); + +#define LOG_FEATURE(n_, str_, Name_, desc_) \ + if (ProfilerFeature::Has##Name_(aFeatures)) { \ + LOG("- feature = %s", str_); \ + } + + BASE_PROFILER_FOR_EACH_FEATURE(LOG_FEATURE) + +#undef LOG_FEATURE + + for (uint32_t i = 0; i < aFilterCount; i++) { + LOG("- threads = %s", aFilters[i]); + } + } + + MOZ_RELEASE_ASSERT(CorePS::Exists() && !ActivePS::Exists(aLock)); + + mozilla::base_profiler_markers_detail::EnsureBufferForMainThreadAddMarker(); + +#if defined(GP_PLAT_amd64_windows) + InitializeWin64ProfilerHooks(); +#endif + + // Fall back to the default values if the passed-in values are unreasonable. + // We want to be able to store at least one full stack. + // TODO: Review magic numbers. + PowerOfTwo32 capacity = + (aCapacity.Value() >= + ProfileBufferChunkManager::scExpectedMaximumStackSize / scBytesPerEntry) + ? aCapacity + : BASE_PROFILER_DEFAULT_ENTRIES; + Maybe duration = aDuration; + + if (aDuration && *aDuration <= 0) { + duration = Nothing(); + } + double interval = aInterval > 0 ? aInterval : BASE_PROFILER_DEFAULT_INTERVAL; + + ActivePS::Create(aLock, profilingStartTime, capacity, interval, aFeatures, + aFilters, aFilterCount, duration); + + // Set up profiling for each registered thread, if appropriate. + const Vector>& registeredThreads = + CorePS::RegisteredThreads(aLock); + for (auto& registeredThread : registeredThreads) { + RefPtr info = registeredThread->Info(); + + if (ActivePS::ShouldProfileThread(aLock, info)) { + registeredThread->RacyRegisteredThread().SetIsBeingProfiled(true); + ActivePS::AddLiveProfiledThread(aLock, registeredThread.get(), + MakeUnique(info)); + registeredThread->RacyRegisteredThread().ReinitializeOnResume(); + } + } + + // Setup support for pushing/popping labels in mozglue. + RegisterProfilerLabelEnterExit(MozGlueBaseLabelEnter, MozGlueBaseLabelExit); + + // At the very end, set up RacyFeatures. + RacyFeatures::SetActive(ActivePS::Features(aLock)); +} + +void profiler_start(PowerOfTwo32 aCapacity, double aInterval, + uint32_t aFeatures, const char** aFilters, + uint32_t aFilterCount, const Maybe& aDuration) { + LOG("profiler_start"); + + SamplerThread* samplerThread = nullptr; + { + PSAutoLock lock; + + // Initialize if necessary. + if (!CorePS::Exists()) { + profiler_init(nullptr); + } + + // Reset the current state if the profiler is running. + if (ActivePS::Exists(lock)) { + samplerThread = locked_profiler_stop(lock); + } + + locked_profiler_start(lock, aCapacity, aInterval, aFeatures, aFilters, + aFilterCount, aDuration); + } + + // TODO: Install memory counter if it is possible from mozglue. + // #if defined(MOZ_REPLACE_MALLOC) && defined(MOZ_PROFILER_MEMORY) + // // start counting memory allocations (outside of lock because this may + // call + // // profiler_add_sampled_counter which would attempt to take the lock.) + // mozilla::profiler::install_memory_counter(true); + // #endif + + // We do these operations with gPSMutex unlocked. The comments in + // profiler_stop() explain why. + if (samplerThread) { + delete samplerThread; + } +} + +void profiler_ensure_started(PowerOfTwo32 aCapacity, double aInterval, + uint32_t aFeatures, const char** aFilters, + uint32_t aFilterCount, + const Maybe& aDuration) { + LOG("profiler_ensure_started"); + + // bool startedProfiler = false; (See TODO below) + SamplerThread* samplerThread = nullptr; + { + PSAutoLock lock; + + // Initialize if necessary. + if (!CorePS::Exists()) { + profiler_init(nullptr); + } + + if (ActivePS::Exists(lock)) { + // The profiler is active. + if (!ActivePS::Equals(lock, aCapacity, aDuration, aInterval, aFeatures, + aFilters, aFilterCount)) { + // Stop and restart with different settings. + samplerThread = locked_profiler_stop(lock); + locked_profiler_start(lock, aCapacity, aInterval, aFeatures, aFilters, + aFilterCount, aDuration); + // startedProfiler = true; (See TODO below) + } + } else { + // The profiler is stopped. + locked_profiler_start(lock, aCapacity, aInterval, aFeatures, aFilters, + aFilterCount, aDuration); + // startedProfiler = true; (See TODO below) + } + } + + // TODO: Install memory counter if it is possible from mozglue. + // #if defined(MOZ_REPLACE_MALLOC) && defined(MOZ_PROFILER_MEMORY) + // // start counting memory allocations (outside of lock because this may + // // call profiler_add_sampled_counter which would attempt to take the + // // lock.) + // mozilla::profiler::install_memory_counter(true); + // #endif + + // We do these operations with gPSMutex unlocked. The comments in + // profiler_stop() explain why. + if (samplerThread) { + delete samplerThread; + } +} + +[[nodiscard]] static SamplerThread* locked_profiler_stop(PSLockRef aLock) { + LOG("locked_profiler_stop"); + + MOZ_RELEASE_ASSERT(CorePS::Exists() && ActivePS::Exists(aLock)); + + // At the very start, clear RacyFeatures. + RacyFeatures::SetInactive(); + + // TODO: Uninstall memory counter if it is possible from mozglue. + // #if defined(MOZ_REPLACE_MALLOC) && defined(MOZ_PROFILER_MEMORY) + // mozilla::profiler::install_memory_counter(false); + // #endif + + // Remove support for pushing/popping labels in mozglue. + RegisterProfilerLabelEnterExit(nullptr, nullptr); + + // Stop sampling live threads. + const Vector& liveProfiledThreads = + ActivePS::LiveProfiledThreads(aLock); + for (auto& thread : liveProfiledThreads) { + RegisteredThread* registeredThread = thread.mRegisteredThread; + registeredThread->RacyRegisteredThread().SetIsBeingProfiled(false); + } + + // The Stop() call doesn't actually stop Run(); that happens in this + // function's caller when the sampler thread is destroyed. Stop() just gives + // the SamplerThread a chance to do some cleanup with gPSMutex locked. + SamplerThread* samplerThread = ActivePS::Destroy(aLock); + samplerThread->Stop(aLock); + + mozilla::base_profiler_markers_detail::ReleaseBufferForMainThreadAddMarker(); + + return samplerThread; +} + +void profiler_stop() { + LOG("profiler_stop"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + SamplerThread* samplerThread; + { + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + samplerThread = locked_profiler_stop(lock); + } + + // We delete with gPSMutex unlocked. Otherwise we would get a deadlock: we + // would be waiting here with gPSMutex locked for SamplerThread::Run() to + // return so the join operation within the destructor can complete, but Run() + // needs to lock gPSMutex to return. + // + // Because this call occurs with gPSMutex unlocked, it -- including the final + // iteration of Run()'s loop -- must be able detect deactivation and return + // in a way that's safe with respect to other gPSMutex-locking operations + // that may have occurred in the meantime. + delete samplerThread; +} + +bool profiler_is_paused() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return false; + } + + return ActivePS::IsPaused(lock); +} + +void profiler_pause() { + LOG("profiler_pause"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + { + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + RacyFeatures::SetPaused(); + ActivePS::SetIsPaused(lock, true); + ActivePS::Buffer(lock).AddEntry(ProfileBufferEntry::Pause(profiler_time())); + } +} + +void profiler_resume() { + LOG("profiler_resume"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + { + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + ActivePS::Buffer(lock).AddEntry( + ProfileBufferEntry::Resume(profiler_time())); + ActivePS::SetIsPaused(lock, false); + RacyFeatures::SetUnpaused(); + } +} + +bool profiler_is_sampling_paused() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return false; + } + + return ActivePS::IsSamplingPaused(lock); +} + +void profiler_pause_sampling() { + LOG("profiler_pause_sampling"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + { + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + RacyFeatures::SetSamplingPaused(); + ActivePS::SetIsSamplingPaused(lock, true); + ActivePS::Buffer(lock).AddEntry( + ProfileBufferEntry::PauseSampling(profiler_time())); + } +} + +void profiler_resume_sampling() { + LOG("profiler_resume_sampling"); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + { + PSAutoLock lock; + + if (!ActivePS::Exists(lock)) { + return; + } + + ActivePS::Buffer(lock).AddEntry( + ProfileBufferEntry::ResumeSampling(profiler_time())); + ActivePS::SetIsSamplingPaused(lock, false); + RacyFeatures::SetSamplingUnpaused(); + } +} + +bool profiler_feature_active(uint32_t aFeature) { + // This function runs both on and off the main thread. + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + // This function is hot enough that we use RacyFeatures, not ActivePS. + return RacyFeatures::IsActiveWithFeature(aFeature); +} + +bool profiler_active_without_feature(uint32_t aFeature) { + // This function runs both on and off the main thread. + + // This function is hot enough that we use RacyFeatures, not ActivePS. + return RacyFeatures::IsActiveWithoutFeature(aFeature); +} + +void profiler_add_sampled_counter(BaseProfilerCount* aCounter) { + DEBUG_LOG("profiler_add_sampled_counter(%s)", aCounter->mLabel); + PSAutoLock lock; + CorePS::AppendCounter(lock, aCounter); +} + +void profiler_remove_sampled_counter(BaseProfilerCount* aCounter) { + DEBUG_LOG("profiler_remove_sampled_counter(%s)", aCounter->mLabel); + PSAutoLock lock; + // Note: we don't enforce a final sample, though we could do so if the + // profiler was active + CorePS::RemoveCounter(lock, aCounter); +} + +ProfilingStack* profiler_register_thread(const char* aName, + void* aGuessStackTop) { + DEBUG_LOG("profiler_register_thread(%s)", aName); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (RegisteredThread* thread = FindCurrentThreadRegisteredThread(lock); + thread) { + LOG("profiler_register_thread(%s) - thread %" PRIu64 + " already registered as %s", + aName, uint64_t(profiler_current_thread_id().ToNumber()), + thread->Info()->Name()); + // TODO: Use new name. This is currently not possible because the + // RegisteredThread's ThreadInfo cannot be changed. + // In the meantime, we record a marker that could be used in the frontend. + std::string text("Thread "); + text += std::to_string(profiler_current_thread_id().ToNumber()); + text += " \""; + text += thread->Info()->Name(); + text += "\" attempted to re-register as \""; + text += aName; + text += "\""; + BASE_PROFILER_MARKER_TEXT("profiler_register_thread again", OTHER_Profiling, + MarkerThreadId::MainThread(), text); + + return &thread->RacyRegisteredThread().ProfilingStack(); + } + + void* stackTop = GetStackTop(aGuessStackTop); + return locked_register_thread(lock, aName, stackTop); +} + +void profiler_unregister_thread() { + if (!CorePS::Exists()) { + // This function can be called after the main thread has already shut down. + return; + } + + PSAutoLock lock; + + RegisteredThread* registeredThread = FindCurrentThreadRegisteredThread(lock); + MOZ_RELEASE_ASSERT(registeredThread == + TLSRegisteredThread::RegisteredThread(lock)); + if (registeredThread) { + RefPtr info = registeredThread->Info(); + + DEBUG_LOG("profiler_unregister_thread: %s", info->Name()); + + if (ActivePS::Exists(lock)) { + ActivePS::UnregisterThread(lock, registeredThread); + } + + // Clear the pointer to the RegisteredThread object that we're about to + // destroy. + TLSRegisteredThread::SetRegisteredThread(lock, nullptr); + + // Remove the thread from the list of registered threads. This deletes the + // registeredThread object. + CorePS::RemoveRegisteredThread(lock, registeredThread); + } else { + LOG("profiler_unregister_thread() - thread %" PRIu64 + " already unregistered", + uint64_t(profiler_current_thread_id().ToNumber())); + // We cannot record a marker on this thread because it was already + // unregistered. Send it to the main thread (unless this *is* already the + // main thread, which has been unregistered); this may be useful to catch + // mismatched register/unregister pairs in Firefox. + if (BaseProfilerThreadId tid = profiler_current_thread_id(); + tid != profiler_main_thread_id()) { + BASE_PROFILER_MARKER_TEXT( + "profiler_unregister_thread again", OTHER_Profiling, + MarkerThreadId::MainThread(), + std::to_string(profiler_current_thread_id().ToNumber())); + } + // There are two ways FindCurrentThreadRegisteredThread() might have failed. + // + // - TLSRegisteredThread::Init() failed in locked_register_thread(). + // + // - We've already called profiler_unregister_thread() for this thread. + // (Whether or not it should, this does happen in practice.) + // + // Either way, TLSRegisteredThread should be empty. + MOZ_RELEASE_ASSERT(!TLSRegisteredThread::RegisteredThread(lock)); + } +} + +void profiler_register_page(uint64_t aTabID, uint64_t aInnerWindowID, + const std::string& aUrl, + uint64_t aEmbedderInnerWindowID) { + DEBUG_LOG("profiler_register_page(%" PRIu64 ", %" PRIu64 ", %s, %" PRIu64 ")", + aTabID, aInnerWindowID, aUrl.c_str(), aEmbedderInnerWindowID); + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + // When a Browsing context is first loaded, the first url loaded in it will be + // about:blank. Because of that, this call keeps the first non-about:blank + // registration of window and discards the previous one. + RefPtr pageInfo = + new PageInformation(aTabID, aInnerWindowID, aUrl, aEmbedderInnerWindowID); + CorePS::AppendRegisteredPage(lock, std::move(pageInfo)); + + // After appending the given page to CorePS, look for the expired + // pages and remove them if there are any. + if (ActivePS::Exists(lock)) { + ActivePS::DiscardExpiredPages(lock); + } +} + +void profiler_unregister_page(uint64_t aRegisteredInnerWindowID) { + if (!CorePS::Exists()) { + // This function can be called after the main thread has already shut down. + return; + } + + PSAutoLock lock; + + // During unregistration, if the profiler is active, we have to keep the + // page information since there may be some markers associated with the given + // page. But if profiler is not active. we have no reason to keep the + // page information here because there can't be any marker associated with it. + if (ActivePS::Exists(lock)) { + ActivePS::UnregisterPage(lock, aRegisteredInnerWindowID); + } else { + CorePS::RemoveRegisteredPage(lock, aRegisteredInnerWindowID); + } +} + +void profiler_clear_all_pages() { + if (!CorePS::Exists()) { + // This function can be called after the main thread has already shut down. + return; + } + + { + PSAutoLock lock; + CorePS::ClearRegisteredPages(lock); + if (ActivePS::Exists(lock)) { + ActivePS::ClearUnregisteredPages(lock); + } + } +} + +void profiler_thread_sleep() { + // This function runs both on and off the main thread. + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + RacyRegisteredThread* racyRegisteredThread = + TLSRegisteredThread::RacyRegisteredThread(); + if (!racyRegisteredThread) { + return; + } + + racyRegisteredThread->SetSleeping(); +} + +void profiler_thread_wake() { + // This function runs both on and off the main thread. + + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + RacyRegisteredThread* racyRegisteredThread = + TLSRegisteredThread::RacyRegisteredThread(); + if (!racyRegisteredThread) { + return; + } + + racyRegisteredThread->SetAwake(); +} + +bool detail::IsThreadBeingProfiled() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + const RacyRegisteredThread* racyRegisteredThread = + TLSRegisteredThread::RacyRegisteredThread(); + return racyRegisteredThread && racyRegisteredThread->IsBeingProfiled(); +} + +bool profiler_thread_is_sleeping() { + MOZ_RELEASE_ASSERT(profiler_is_main_thread()); + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + RacyRegisteredThread* racyRegisteredThread = + TLSRegisteredThread::RacyRegisteredThread(); + if (!racyRegisteredThread) { + return false; + } + return racyRegisteredThread->IsSleeping(); +} + +double profiler_time() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + TimeDuration delta = TimeStamp::Now() - CorePS::ProcessStartTime(); + return delta.ToMilliseconds(); +} + +bool profiler_capture_backtrace_into(ProfileChunkedBuffer& aChunkedBuffer, + StackCaptureOptions aCaptureOptions) { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + + PSAutoLock lock; + + if (!ActivePS::Exists(lock) || + aCaptureOptions == StackCaptureOptions::NoStack) { + return false; + } + + RegisteredThread* registeredThread = + TLSRegisteredThread::RegisteredThread(lock); + if (!registeredThread) { + MOZ_ASSERT(registeredThread); + return false; + } + + ProfileBuffer profileBuffer(aChunkedBuffer); + + Registers regs; +#if defined(HAVE_NATIVE_UNWIND) + REGISTERS_SYNC_POPULATE(regs); +#else + regs.Clear(); +#endif + + DoSyncSample(lock, *registeredThread, TimeStamp::Now(), regs, profileBuffer, + aCaptureOptions); + + return true; +} + +UniquePtr profiler_capture_backtrace() { + MOZ_RELEASE_ASSERT(CorePS::Exists()); + AUTO_BASE_PROFILER_LABEL("baseprofiler::profiler_capture_backtrace", + PROFILER); + + // Quick is-active check before allocating a buffer. + // If NoMarkerStacks is set, we don't want to capture a backtrace. + if (!profiler_active_without_feature(ProfilerFeature::NoMarkerStacks)) { + return nullptr; + } + + auto buffer = MakeUnique( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex, + MakeUnique( + ProfileBufferChunkManager::scExpectedMaximumStackSize)); + + if (!profiler_capture_backtrace_into(*buffer, StackCaptureOptions::Full)) { + return nullptr; + } + + return buffer; +} + +UniqueProfilerBacktrace profiler_get_backtrace() { + UniquePtr buffer = profiler_capture_backtrace(); + + if (!buffer) { + return nullptr; + } + + return UniqueProfilerBacktrace( + new ProfilerBacktrace("SyncProfile", std::move(buffer))); +} + +void ProfilerBacktraceDestructor::operator()(ProfilerBacktrace* aBacktrace) { + delete aBacktrace; +} + +bool profiler_is_locked_on_current_thread() { + // This function is used to help users avoid calling `profiler_...` functions + // when the profiler may already have a lock in place, which would prevent a + // 2nd recursive lock (resulting in a crash or a never-ending wait). + // So we must return `true` for any of: + // - The main profiler mutex, used by most functions, and/or + // - The buffer mutex, used directly in some functions without locking the + // main mutex, e.g., marker-related functions. + return PSAutoLock::IsLockedOnCurrentThread() || + profiler_get_core_buffer().IsThreadSafeAndLockedOnCurrentThread(); +} + +// This is a simplified version of profiler_add_marker that can be easily passed +// into the JS engine. +void profiler_add_js_marker(const char* aMarkerName, const char* aMarkerText) { + BASE_PROFILER_MARKER_TEXT( + ProfilerString8View::WrapNullTerminatedString(aMarkerName), JS, {}, + ProfilerString8View::WrapNullTerminatedString(aMarkerText)); +} + +// NOTE: aCollector's methods will be called while the target thread is paused. +// Doing things in those methods like allocating -- which may try to claim +// locks -- is a surefire way to deadlock. +void profiler_suspend_and_sample_thread(BaseProfilerThreadId aThreadId, + uint32_t aFeatures, + ProfilerStackCollector& aCollector, + bool aSampleNative /* = true */) { + const bool isSynchronous = [&aThreadId]() { + const BaseProfilerThreadId currentThreadId = profiler_current_thread_id(); + if (!aThreadId.IsSpecified()) { + aThreadId = currentThreadId; + return true; + } + return aThreadId == currentThreadId; + }(); + + // Lock the profiler mutex + PSAutoLock lock; + + const Vector>& registeredThreads = + CorePS::RegisteredThreads(lock); + for (auto& thread : registeredThreads) { + RefPtr info = thread->Info(); + RegisteredThread& registeredThread = *thread.get(); + + if (info->ThreadId() == aThreadId) { + if (info->IsMainThread()) { + aCollector.SetIsMainThread(); + } + + // Allocate the space for the native stack + NativeStack nativeStack; + + auto collectStack = [&](const Registers& aRegs, const TimeStamp& aNow) { + // The target thread is now suspended. Collect a native + // backtrace, and call the callback. +#if defined(HAVE_FASTINIT_NATIVE_UNWIND) + if (aSampleNative) { + // We can only use FramePointerStackWalk or MozStackWalk from + // suspend_and_sample_thread as other stackwalking methods may not be + // initialized. +# if defined(USE_FRAME_POINTER_STACK_WALK) + DoFramePointerBacktrace(lock, registeredThread, aRegs, nativeStack); +# elif defined(USE_MOZ_STACK_WALK) + DoMozStackWalkBacktrace(lock, registeredThread, aRegs, nativeStack); +# else +# error "Invalid configuration" +# endif + + MergeStacks(aFeatures, isSynchronous, registeredThread, aRegs, + nativeStack, aCollector); + } else +#endif + { + MergeStacks(aFeatures, isSynchronous, registeredThread, aRegs, + nativeStack, aCollector); + + aCollector.CollectNativeLeafAddr((void*)aRegs.mPC); + } + }; + + if (isSynchronous) { + // Sampling the current thread, do NOT suspend it! + Registers regs; +#if defined(HAVE_NATIVE_UNWIND) + REGISTERS_SYNC_POPULATE(regs); +#else + regs.Clear(); +#endif + collectStack(regs, TimeStamp::Now()); + } else { + // Suspend, sample, and then resume the target thread. + Sampler sampler(lock); + TimeStamp now = TimeStamp::Now(); + sampler.SuspendAndSampleAndResumeThread(lock, registeredThread, now, + collectStack); + + // NOTE: Make sure to disable the sampler before it is destroyed, in + // case the profiler is running at the same time. + sampler.Disable(lock); + } + break; + } + } +} + +// END externally visible functions +//////////////////////////////////////////////////////////////////////// + +} // namespace baseprofiler +} // namespace mozilla diff --git a/mozglue/baseprofiler/core/platform.h b/mozglue/baseprofiler/core/platform.h new file mode 100644 index 0000000000..f77fc12726 --- /dev/null +++ b/mozglue/baseprofiler/core/platform.h @@ -0,0 +1,149 @@ +// Copyright (c) 2006-2011 The Chromium Authors. All rights reserved. +// +// Redistribution and use in source and binary forms, with or without +// modification, are permitted provided that the following conditions +// are met: +// * Redistributions of source code must retain the above copyright +// notice, this list of conditions and the following disclaimer. +// * 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. +// * Neither the name of Google, Inc. nor the names of its contributors +// may be used to endorse or promote products derived from this +// software without specific prior written permission. +// +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +// "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 THE +// COPYRIGHT OWNER 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. + +#ifndef TOOLS_PLATFORM_H_ +#define TOOLS_PLATFORM_H_ + +#include "PlatformMacros.h" + +#include "BaseProfiler.h" + +#include "mozilla/Atomics.h" +#include "mozilla/Logging.h" +#include "mozilla/UniquePtr.h" +#include "mozilla/Vector.h" + +#include +#include +#include + +namespace mozilla { +namespace baseprofiler { +bool LogTest(int aLevelToTest); +void PrintToConsole(const char* aFmt, ...) MOZ_FORMAT_PRINTF(1, 2); +} // namespace baseprofiler +} // namespace mozilla + +// These are for MOZ_BASE_PROFILER_LOGGING and above. It's the default logging +// level for the profiler, and should be used sparingly. +#define LOG_TEST ::mozilla::baseprofiler::LogTest(3) +#define LOG(arg, ...) \ + do { \ + if (LOG_TEST) { \ + ::mozilla::baseprofiler::PrintToConsole( \ + "[I %d/%d] " arg "\n", \ + int(::mozilla::baseprofiler::profiler_current_process_id() \ + .ToNumber()), \ + int(::mozilla::baseprofiler::profiler_current_thread_id() \ + .ToNumber()), \ + ##__VA_ARGS__); \ + } \ + } while (0) + +// These are for MOZ_BASE_PROFILER_DEBUG_LOGGING. It should be used for logging +// that is somewhat more verbose than LOG. +#define DEBUG_LOG_TEST ::mozilla::baseprofiler::LogTest(4) +#define DEBUG_LOG(arg, ...) \ + do { \ + if (DEBUG_LOG_TEST) { \ + ::mozilla::baseprofiler::PrintToConsole( \ + "[D %d/%d] " arg "\n", \ + int(::mozilla::baseprofiler::profiler_current_process_id() \ + .ToNumber()), \ + int(::mozilla::baseprofiler::profiler_current_thread_id() \ + .ToNumber()), \ + ##__VA_ARGS__); \ + } \ + } while (0) + +// These are for MOZ_BASE_PROFILER_VERBOSE_LOGGING. It should be used for +// logging that is somewhat more verbose than DEBUG_LOG. +#define VERBOSE_LOG_TEST ::mozilla::baseprofiler::LogTest(5) +#define VERBOSE_LOG(arg, ...) \ + do { \ + if (VERBOSE_LOG_TEST) { \ + ::mozilla::baseprofiler::PrintToConsole( \ + "[V %d/%d] " arg "\n", \ + int(::mozilla::baseprofiler::profiler_current_process_id() \ + .ToNumber()), \ + int(::mozilla::baseprofiler::profiler_current_thread_id() \ + .ToNumber()), \ + ##__VA_ARGS__); \ + } \ + } while (0) + +namespace mozilla { + +class JSONWriter; + +namespace baseprofiler { + +// If positive, skip stack-sampling in the sampler thread loop. +// Users should increment it atomically when samplings should be avoided, and +// later decrement it back. Multiple uses can overlap. +// There could be a sampling in progress when this is first incremented, so if +// it is critical to prevent any sampling, lock the profiler mutex instead. +// Relaxed ordering, because it's used to request that the profiler pause +// future sampling; this is not time critical, nor dependent on anything else. +extern mozilla::Atomic gSkipSampling; + +typedef uint8_t* Address; + +class PlatformData; + +// We can't new/delete the type safely without defining it +// (-Wdelete-incomplete). Use these to hide the details from clients. +struct PlatformDataDestructor { + void operator()(PlatformData*); +}; + +typedef UniquePtr UniquePlatformData; +UniquePlatformData AllocPlatformData(BaseProfilerThreadId aThreadId); + +// Convert the array of strings to a bitfield. +uint32_t ParseFeaturesFromStringArray(const char** aFeatures, + uint32_t aFeatureCount, + bool aIsStartup = false); + +// Flags to conveniently track various JS instrumentations. +enum class JSInstrumentationFlags { + StackSampling = 0x1, + Allocations = 0x2, +}; + +// Record an exit profile from a child process. +void profiler_received_exit_profile(const std::string& aExitProfile); + +// Extract all received exit profiles that have not yet expired (i.e., they +// still intersect with this process' buffer range). +Vector profiler_move_exit_profiles(); + +} // namespace baseprofiler +} // namespace mozilla + +#endif /* ndef TOOLS_PLATFORM_H_ */ diff --git a/mozglue/baseprofiler/core/shared-libraries-linux.cc b/mozglue/baseprofiler/core/shared-libraries-linux.cc new file mode 100644 index 0000000000..a12d60cf73 --- /dev/null +++ b/mozglue/baseprofiler/core/shared-libraries-linux.cc @@ -0,0 +1,854 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim: set ts=8 sts=2 et sw=2 tw=80: */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "BaseProfilerSharedLibraries.h" + +#define PATH_MAX_TOSTRING(x) #x +#define PATH_MAX_STRING(x) PATH_MAX_TOSTRING(x) +#include +#include +#include +#include +#include +#include "platform.h" +#include "mozilla/Sprintf.h" + +#include +#include +#include +#include +#if defined(GP_OS_linux) || defined(GP_OS_android) +# include +#endif +#include +#include +#include +#include + +#if defined(GP_OS_linux) || defined(GP_OS_android) || defined(GP_OS_freebsd) +# include // dl_phdr_info, ElfW() +#else +# error "Unexpected configuration" +#endif + +#if defined(GP_OS_android) +extern "C" MOZ_EXPORT __attribute__((weak)) int dl_iterate_phdr( + int (*callback)(struct dl_phdr_info* info, size_t size, void* data), + void* data); +#endif + +#if defined(GP_OS_freebsd) && !defined(ElfW) +# define ElfW(type) Elf_##type +#endif + +// ---------------------------------------------------------------------------- +// Starting imports from toolkit/crashreporter/google-breakpad/, as needed by +// this file when moved to mozglue. + +// Imported from +// toolkit/crashreporter/google-breakpad/src/common/memory_range.h. +// A lightweight wrapper with a pointer and a length to encapsulate a contiguous +// range of memory. It provides helper methods for checked access of a subrange +// of the memory. Its implemementation does not allocate memory or call into +// libc functions, and is thus safer to use in a crashed environment. +class MemoryRange { + public: + MemoryRange() : data_(NULL), length_(0) {} + + MemoryRange(const void* data, size_t length) { Set(data, length); } + + // Returns true if this memory range contains no data. + bool IsEmpty() const { + // Set() guarantees that |length_| is zero if |data_| is NULL. + return length_ == 0; + } + + // Resets to an empty range. + void Reset() { + data_ = NULL; + length_ = 0; + } + + // Sets this memory range to point to |data| and its length to |length|. + void Set(const void* data, size_t length) { + data_ = reinterpret_cast(data); + // Always set |length_| to zero if |data_| is NULL. + length_ = data ? length : 0; + } + + // Returns true if this range covers a subrange of |sub_length| bytes + // at |sub_offset| bytes of this memory range, or false otherwise. + bool Covers(size_t sub_offset, size_t sub_length) const { + // The following checks verify that: + // 1. sub_offset is within [ 0 .. length_ - 1 ] + // 2. sub_offset + sub_length is within + // [ sub_offset .. length_ ] + return sub_offset < length_ && sub_offset + sub_length >= sub_offset && + sub_offset + sub_length <= length_; + } + + // Returns a raw data pointer to a subrange of |sub_length| bytes at + // |sub_offset| bytes of this memory range, or NULL if the subrange + // is out of bounds. + const void* GetData(size_t sub_offset, size_t sub_length) const { + return Covers(sub_offset, sub_length) ? (data_ + sub_offset) : NULL; + } + + // Same as the two-argument version of GetData() but uses sizeof(DataType) + // as the subrange length and returns an |DataType| pointer for convenience. + template + const DataType* GetData(size_t sub_offset) const { + return reinterpret_cast( + GetData(sub_offset, sizeof(DataType))); + } + + // Returns a raw pointer to the |element_index|-th element of an array + // of elements of length |element_size| starting at |sub_offset| bytes + // of this memory range, or NULL if the element is out of bounds. + const void* GetArrayElement(size_t element_offset, size_t element_size, + unsigned element_index) const { + size_t sub_offset = element_offset + element_index * element_size; + return GetData(sub_offset, element_size); + } + + // Same as the three-argument version of GetArrayElement() but deduces + // the element size using sizeof(ElementType) and returns an |ElementType| + // pointer for convenience. + template + const ElementType* GetArrayElement(size_t element_offset, + unsigned element_index) const { + return reinterpret_cast( + GetArrayElement(element_offset, sizeof(ElementType), element_index)); + } + + // Returns a subrange of |sub_length| bytes at |sub_offset| bytes of + // this memory range, or an empty range if the subrange is out of bounds. + MemoryRange Subrange(size_t sub_offset, size_t sub_length) const { + return Covers(sub_offset, sub_length) + ? MemoryRange(data_ + sub_offset, sub_length) + : MemoryRange(); + } + + // Returns a pointer to the beginning of this memory range. + const uint8_t* data() const { return data_; } + + // Returns the length, in bytes, of this memory range. + size_t length() const { return length_; } + + private: + // Pointer to the beginning of this memory range. + const uint8_t* data_; + + // Length, in bytes, of this memory range. + size_t length_; +}; + +// Imported from +// toolkit/crashreporter/google-breakpad/src/common/linux/memory_mapped_file.h +// and inlined .cc. +// A utility class for mapping a file into memory for read-only access of the +// file content. Its implementation avoids calling into libc functions by +// directly making system calls for open, close, mmap, and munmap. +class MemoryMappedFile { + public: + MemoryMappedFile() {} + + // Constructor that calls Map() to map a file at |path| into memory. + // If Map() fails, the object behaves as if it is default constructed. + MemoryMappedFile(const char* path, size_t offset) { Map(path, offset); } + + MemoryMappedFile(const MemoryMappedFile&) = delete; + MemoryMappedFile& operator=(const MemoryMappedFile&) = delete; + + ~MemoryMappedFile() {} + + // Maps a file at |path| into memory, which can then be accessed via + // content() as a MemoryRange object or via data(), and returns true on + // success. Mapping an empty file will succeed but with data() and size() + // returning NULL and 0, respectively. An existing mapping is unmapped + // before a new mapping is created. + bool Map(const char* path, size_t offset) { + Unmap(); + + int fd = open(path, O_RDONLY, 0); + if (fd == -1) { + return false; + } + +#if defined(__x86_64__) || defined(__aarch64__) || \ + (defined(__mips__) && _MIPS_SIM == _ABI64) || \ + !(defined(GP_OS_linux) || defined(GP_OS_android)) + + struct stat st; + if (fstat(fd, &st) == -1 || st.st_size < 0) { +#else + struct stat64 st; + if (fstat64(fd, &st) == -1 || st.st_size < 0) { +#endif + close(fd); + return false; + } + + // Strangely file size can be negative, but we check above that it is not. + size_t file_len = static_cast(st.st_size); + // If the file does not extend beyond the offset, simply use an empty + // MemoryRange and return true. Don't bother to call mmap() + // even though mmap() can handle an empty file on some platforms. + if (offset >= file_len) { + close(fd); + return true; + } + + void* data = mmap(NULL, file_len, PROT_READ, MAP_PRIVATE, fd, offset); + close(fd); + if (data == MAP_FAILED) { + return false; + } + + content_.Set(data, file_len - offset); + return true; + } + + // Unmaps the memory for the mapped file. It's a no-op if no file is + // mapped. + void Unmap() { + if (content_.data()) { + munmap(const_cast(content_.data()), content_.length()); + content_.Set(NULL, 0); + } + } + + // Returns a MemoryRange object that covers the memory for the mapped + // file. The MemoryRange object is empty if no file is mapped. + const MemoryRange& content() const { return content_; } + + // Returns a pointer to the beginning of the memory for the mapped file. + // or NULL if no file is mapped or the mapped file is empty. + const void* data() const { return content_.data(); } + + // Returns the size in bytes of the mapped file, or zero if no file + // is mapped. + size_t size() const { return content_.length(); } + + private: + // Mapped file content as a MemoryRange object. + MemoryRange content_; +}; + +// Imported from +// toolkit/crashreporter/google-breakpad/src/common/linux/file_id.h and inlined +// .cc. +// GNU binutils' ld defaults to 'sha1', which is 160 bits == 20 bytes, +// so this is enough to fit that, which most binaries will use. +// This is just a sensible default for vectors so most callers can get away with +// stack allocation. +static const size_t kDefaultBuildIdSize = 20; + +// Used in a few places for backwards-compatibility. +typedef struct { + uint32_t data1; + uint16_t data2; + uint16_t data3; + uint8_t data4[8]; +} MDGUID; /* GUID */ + +const size_t kMDGUIDSize = sizeof(MDGUID); + +class FileID { + public: + explicit FileID(const char* path) : path_(path) {} + ~FileID() {} + + // Load the identifier for the elf file path specified in the constructor into + // |identifier|. + // + // The current implementation will look for a .note.gnu.build-id + // section and use that as the file id, otherwise it falls back to + // XORing the first 4096 bytes of the .text section to generate an identifier. + bool ElfFileIdentifier(std::vector& identifier) { + MemoryMappedFile mapped_file(path_.c_str(), 0); + if (!mapped_file.data()) // Should probably check if size >= ElfW(Ehdr)? + return false; + + return ElfFileIdentifierFromMappedFile(mapped_file.data(), identifier); + } + + // Traits classes so consumers can write templatized code to deal + // with specific ELF bits. + struct ElfClass32 { + typedef Elf32_Addr Addr; + typedef Elf32_Ehdr Ehdr; + typedef Elf32_Nhdr Nhdr; + typedef Elf32_Phdr Phdr; + typedef Elf32_Shdr Shdr; + typedef Elf32_Half Half; + typedef Elf32_Off Off; + typedef Elf32_Sym Sym; + typedef Elf32_Word Word; + + static const int kClass = ELFCLASS32; + static const uint16_t kMachine = EM_386; + static const size_t kAddrSize = sizeof(Elf32_Addr); + static constexpr const char* kMachineName = "x86"; + }; + + struct ElfClass64 { + typedef Elf64_Addr Addr; + typedef Elf64_Ehdr Ehdr; + typedef Elf64_Nhdr Nhdr; + typedef Elf64_Phdr Phdr; + typedef Elf64_Shdr Shdr; + typedef Elf64_Half Half; + typedef Elf64_Off Off; + typedef Elf64_Sym Sym; + typedef Elf64_Word Word; + + static const int kClass = ELFCLASS64; + static const uint16_t kMachine = EM_X86_64; + static const size_t kAddrSize = sizeof(Elf64_Addr); + static constexpr const char* kMachineName = "x86_64"; + }; + + // Internal helper method, exposed for convenience for callers + // that already have more info. + template + static const typename ElfClass::Shdr* FindElfSectionByName( + const char* name, typename ElfClass::Word section_type, + const typename ElfClass::Shdr* sections, const char* section_names, + const char* names_end, int nsection) { + if (!name || !sections || nsection == 0) { + return NULL; + } + + int name_len = strlen(name); + if (name_len == 0) return NULL; + + for (int i = 0; i < nsection; ++i) { + const char* section_name = section_names + sections[i].sh_name; + if (sections[i].sh_type == section_type && + names_end - section_name >= name_len + 1 && + strcmp(name, section_name) == 0) { + return sections + i; + } + } + return NULL; + } + + struct ElfSegment { + const void* start; + size_t size; + }; + + // Convert an offset from an Elf header into a pointer to the mapped + // address in the current process. Takes an extra template parameter + // to specify the return type to avoid having to dynamic_cast the + // result. + template + static const T* GetOffset(const typename ElfClass::Ehdr* elf_header, + typename ElfClass::Off offset) { + return reinterpret_cast(reinterpret_cast(elf_header) + + offset); + } + +// ELF note name and desc are 32-bits word padded. +#define NOTE_PADDING(a) ((a + 3) & ~3) + + static bool ElfClassBuildIDNoteIdentifier(const void* section, size_t length, + std::vector& identifier) { + static_assert(sizeof(ElfClass32::Nhdr) == sizeof(ElfClass64::Nhdr), + "Elf32_Nhdr and Elf64_Nhdr should be the same"); + typedef typename ElfClass32::Nhdr Nhdr; + + const void* section_end = reinterpret_cast(section) + length; + const Nhdr* note_header = reinterpret_cast(section); + while (reinterpret_cast(note_header) < section_end) { + if (note_header->n_type == NT_GNU_BUILD_ID) break; + note_header = reinterpret_cast( + reinterpret_cast(note_header) + sizeof(Nhdr) + + NOTE_PADDING(note_header->n_namesz) + + NOTE_PADDING(note_header->n_descsz)); + } + if (reinterpret_cast(note_header) >= section_end || + note_header->n_descsz == 0) { + return false; + } + + const uint8_t* build_id = reinterpret_cast(note_header) + + sizeof(Nhdr) + + NOTE_PADDING(note_header->n_namesz); + identifier.insert(identifier.end(), build_id, + build_id + note_header->n_descsz); + + return true; + } + + template + static bool FindElfClassSection(const char* elf_base, + const char* section_name, + typename ElfClass::Word section_type, + const void** section_start, + size_t* section_size) { + typedef typename ElfClass::Ehdr Ehdr; + typedef typename ElfClass::Shdr Shdr; + + if (!elf_base || !section_start || !section_size) { + return false; + } + + if (strncmp(elf_base, ELFMAG, SELFMAG) != 0) { + return false; + } + + const Ehdr* elf_header = reinterpret_cast(elf_base); + if (elf_header->e_ident[EI_CLASS] != ElfClass::kClass) { + return false; + } + + const Shdr* sections = + GetOffset(elf_header, elf_header->e_shoff); + const Shdr* section_names = sections + elf_header->e_shstrndx; + const char* names = + GetOffset(elf_header, section_names->sh_offset); + const char* names_end = names + section_names->sh_size; + + const Shdr* section = + FindElfSectionByName(section_name, section_type, sections, + names, names_end, elf_header->e_shnum); + + if (section != NULL && section->sh_size > 0) { + *section_start = elf_base + section->sh_offset; + *section_size = section->sh_size; + } + + return true; + } + + template + static bool FindElfClassSegment(const char* elf_base, + typename ElfClass::Word segment_type, + std::vector* segments) { + typedef typename ElfClass::Ehdr Ehdr; + typedef typename ElfClass::Phdr Phdr; + + if (!elf_base || !segments) { + return false; + } + + if (strncmp(elf_base, ELFMAG, SELFMAG) != 0) { + return false; + } + + const Ehdr* elf_header = reinterpret_cast(elf_base); + if (elf_header->e_ident[EI_CLASS] != ElfClass::kClass) { + return false; + } + + const Phdr* phdrs = + GetOffset(elf_header, elf_header->e_phoff); + + for (int i = 0; i < elf_header->e_phnum; ++i) { + if (phdrs[i].p_type == segment_type) { + ElfSegment seg = {}; + seg.start = elf_base + phdrs[i].p_offset; + seg.size = phdrs[i].p_filesz; + segments->push_back(seg); + } + } + + return true; + } + + static bool IsValidElf(const void* elf_base) { + return strncmp(reinterpret_cast(elf_base), ELFMAG, SELFMAG) == + 0; + } + + static int ElfClass(const void* elf_base) { + const ElfW(Ehdr)* elf_header = + reinterpret_cast(elf_base); + + return elf_header->e_ident[EI_CLASS]; + } + + static bool FindElfSection(const void* elf_mapped_base, + const char* section_name, uint32_t section_type, + const void** section_start, size_t* section_size) { + if (!elf_mapped_base || !section_start || !section_size) { + return false; + } + + *section_start = NULL; + *section_size = 0; + + if (!IsValidElf(elf_mapped_base)) return false; + + int cls = ElfClass(elf_mapped_base); + const char* elf_base = static_cast(elf_mapped_base); + + if (cls == ELFCLASS32) { + return FindElfClassSection(elf_base, section_name, + section_type, section_start, + section_size) && + *section_start != NULL; + } else if (cls == ELFCLASS64) { + return FindElfClassSection(elf_base, section_name, + section_type, section_start, + section_size) && + *section_start != NULL; + } + + return false; + } + + static bool FindElfSegments(const void* elf_mapped_base, + uint32_t segment_type, + std::vector* segments) { + if (!elf_mapped_base || !segments) { + return false; + } + + if (!IsValidElf(elf_mapped_base)) return false; + + int cls = ElfClass(elf_mapped_base); + const char* elf_base = static_cast(elf_mapped_base); + + if (cls == ELFCLASS32) { + return FindElfClassSegment(elf_base, segment_type, segments); + } else if (cls == ELFCLASS64) { + return FindElfClassSegment(elf_base, segment_type, segments); + } + + return false; + } + + // Attempt to locate a .note.gnu.build-id section in an ELF binary + // and copy it into |identifier|. + static bool FindElfBuildIDNote(const void* elf_mapped_base, + std::vector& identifier) { + // lld normally creates 2 PT_NOTEs, gold normally creates 1. + std::vector segs; + if (FindElfSegments(elf_mapped_base, PT_NOTE, &segs)) { + for (ElfSegment& seg : segs) { + if (ElfClassBuildIDNoteIdentifier(seg.start, seg.size, identifier)) { + return true; + } + } + } + + void* note_section; + size_t note_size; + if (FindElfSection(elf_mapped_base, ".note.gnu.build-id", SHT_NOTE, + (const void**)¬e_section, ¬e_size)) { + return ElfClassBuildIDNoteIdentifier(note_section, note_size, identifier); + } + + return false; + } + + // Attempt to locate the .text section of an ELF binary and generate + // a simple hash by XORing the first page worth of bytes into |identifier|. + static bool HashElfTextSection(const void* elf_mapped_base, + std::vector& identifier) { + identifier.resize(kMDGUIDSize); + + void* text_section; + size_t text_size; + if (!FindElfSection(elf_mapped_base, ".text", SHT_PROGBITS, + (const void**)&text_section, &text_size) || + text_size == 0) { + return false; + } + + // Only provide |kMDGUIDSize| bytes to keep identifiers produced by this + // function backwards-compatible. + memset(&identifier[0], 0, kMDGUIDSize); + const uint8_t* ptr = reinterpret_cast(text_section); + const uint8_t* ptr_end = + ptr + std::min(text_size, static_cast(4096)); + while (ptr < ptr_end) { + for (unsigned i = 0; i < kMDGUIDSize; i++) identifier[i] ^= ptr[i]; + ptr += kMDGUIDSize; + } + return true; + } + + // Load the identifier for the elf file mapped into memory at |base| into + // |identifier|. Return false if the identifier could not be created for this + // file. + static bool ElfFileIdentifierFromMappedFile( + const void* base, std::vector& identifier) { + // Look for a build id note first. + if (FindElfBuildIDNote(base, identifier)) return true; + + // Fall back on hashing the first page of the text section. + return HashElfTextSection(base, identifier); + } + + // These three functions are not ever called in an unsafe context, so it's OK + // to allocate memory and use libc. + static std::string bytes_to_hex_string(const uint8_t* bytes, size_t count, + bool lowercase = false) { + std::string result; + for (unsigned int idx = 0; idx < count; ++idx) { + char buf[3]; + SprintfLiteral(buf, lowercase ? "%02x" : "%02X", bytes[idx]); + result.append(buf); + } + return result; + } + + // Convert the |identifier| data to a string. The string will + // be formatted as a UUID in all uppercase without dashes. + // (e.g., 22F065BBFC9C49F780FE26A7CEBD7BCE). + static std::string ConvertIdentifierToUUIDString( + const std::vector& identifier) { + uint8_t identifier_swapped[kMDGUIDSize] = {0}; + + // Endian-ness swap to match dump processor expectation. + memcpy(identifier_swapped, &identifier[0], + std::min(kMDGUIDSize, identifier.size())); + uint32_t* data1 = reinterpret_cast(identifier_swapped); + *data1 = htonl(*data1); + uint16_t* data2 = reinterpret_cast(identifier_swapped + 4); + *data2 = htons(*data2); + uint16_t* data3 = reinterpret_cast(identifier_swapped + 6); + *data3 = htons(*data3); + + return bytes_to_hex_string(identifier_swapped, kMDGUIDSize); + } + + // Convert the entire |identifier| data to a lowercase hex string. + static std::string ConvertIdentifierToString( + const std::vector& identifier) { + return bytes_to_hex_string(&identifier[0], identifier.size(), + /* lowercase */ true); + } + + private: + // Storage for the path specified + std::string path_; +}; + +// End of imports from toolkit/crashreporter/google-breakpad/. +// ---------------------------------------------------------------------------- + +struct LoadedLibraryInfo { + LoadedLibraryInfo(const char* aName, unsigned long aBaseAddress, + unsigned long aFirstMappingStart, + unsigned long aLastMappingEnd) + : mName(aName), + mBaseAddress(aBaseAddress), + mFirstMappingStart(aFirstMappingStart), + mLastMappingEnd(aLastMappingEnd) {} + + std::string mName; + unsigned long mBaseAddress; + unsigned long mFirstMappingStart; + unsigned long mLastMappingEnd; +}; + +static std::string IDtoUUIDString(const std::vector& aIdentifier) { + std::string uuid = FileID::ConvertIdentifierToUUIDString(aIdentifier); + // This is '0', not '\0', since it represents the breakpad id age. + uuid += '0'; + return uuid; +} + +// Return raw Build ID in hex. +static std::string IDtoString(const std::vector& aIdentifier) { + std::string uuid = FileID::ConvertIdentifierToString(aIdentifier); + return uuid; +} + +// Get the breakpad Id for the binary file pointed by bin_name +static std::string getBreakpadId(const char* bin_name) { + std::vector identifier; + identifier.reserve(kDefaultBuildIdSize); + + FileID file_id(bin_name); + if (file_id.ElfFileIdentifier(identifier)) { + return IDtoUUIDString(identifier); + } + + return {}; +} + +// Get the code Id for the binary file pointed by bin_name +static std::string getCodeId(const char* bin_name) { + std::vector identifier; + identifier.reserve(kDefaultBuildIdSize); + + FileID file_id(bin_name); + if (file_id.ElfFileIdentifier(identifier)) { + return IDtoString(identifier); + } + + return {}; +} + +static SharedLibrary SharedLibraryAtPath(const char* path, + unsigned long libStart, + unsigned long libEnd, + unsigned long offset = 0) { + std::string pathStr = path; + + size_t pos = pathStr.rfind('\\'); + std::string nameStr = + (pos != std::string::npos) ? pathStr.substr(pos + 1) : pathStr; + + return SharedLibrary(libStart, libEnd, offset, getBreakpadId(path), + getCodeId(path), nameStr, pathStr, nameStr, pathStr, + std::string{}, ""); +} + +static int dl_iterate_callback(struct dl_phdr_info* dl_info, size_t size, + void* data) { + auto libInfoList = reinterpret_cast*>(data); + + if (dl_info->dlpi_phnum <= 0) return 0; + + unsigned long baseAddress = dl_info->dlpi_addr; + unsigned long firstMappingStart = -1; + unsigned long lastMappingEnd = 0; + + for (size_t i = 0; i < dl_info->dlpi_phnum; i++) { + if (dl_info->dlpi_phdr[i].p_type != PT_LOAD) { + continue; + } + unsigned long start = dl_info->dlpi_addr + dl_info->dlpi_phdr[i].p_vaddr; + unsigned long end = start + dl_info->dlpi_phdr[i].p_memsz; + if (start < firstMappingStart) { + firstMappingStart = start; + } + if (end > lastMappingEnd) { + lastMappingEnd = end; + } + } + + libInfoList->push_back(LoadedLibraryInfo(dl_info->dlpi_name, baseAddress, + firstMappingStart, lastMappingEnd)); + + return 0; +} + +SharedLibraryInfo SharedLibraryInfo::GetInfoForSelf() { + SharedLibraryInfo info; + +#if defined(GP_OS_linux) + // We need to find the name of the executable (exeName, exeNameLen) and the + // address of its executable section (exeExeAddr) in the running image. + char exeName[PATH_MAX]; + memset(exeName, 0, sizeof(exeName)); + + ssize_t exeNameLen = readlink("/proc/self/exe", exeName, sizeof(exeName) - 1); + if (exeNameLen == -1) { + // readlink failed for whatever reason. Note this, but keep going. + exeName[0] = '\0'; + exeNameLen = 0; + // LOG("SharedLibraryInfo::GetInfoForSelf(): readlink failed"); + } else { + // Assert no buffer overflow. + MOZ_RELEASE_ASSERT(exeNameLen >= 0 && + exeNameLen < static_cast(sizeof(exeName))); + } + + unsigned long exeExeAddr = 0; +#endif + +#if defined(GP_OS_android) + // If dl_iterate_phdr doesn't exist, we give up immediately. + if (!dl_iterate_phdr) { + // On ARM Android, dl_iterate_phdr is provided by the custom linker. + // So if libxul was loaded by the system linker (e.g. as part of + // xpcshell when running tests), it won't be available and we should + // not call it. + return info; + } +#endif + +#if defined(GP_OS_linux) || defined(GP_OS_android) + // Read info from /proc/self/maps. We ignore most of it. + pid_t pid = mozilla::baseprofiler::profiler_current_process_id().ToNumber(); + char path[PATH_MAX]; + SprintfLiteral(path, "/proc/%d/maps", pid); + std::ifstream maps(path); + std::string line; + while (std::getline(maps, line)) { + int ret; + unsigned long start; + unsigned long end; + char perm[6 + 1] = ""; + unsigned long offset; + char modulePath[PATH_MAX + 1] = ""; + ret = sscanf(line.c_str(), + "%lx-%lx %6s %lx %*s %*x %" PATH_MAX_STRING(PATH_MAX) "s\n", + &start, &end, perm, &offset, modulePath); + if (!strchr(perm, 'x')) { + // Ignore non executable entries + continue; + } + if (ret != 5 && ret != 4) { + // LOG("SharedLibraryInfo::GetInfoForSelf(): " + // "reading /proc/self/maps failed"); + continue; + } + +# if defined(GP_OS_linux) + // Try to establish the main executable's load address. + if (exeNameLen > 0 && strcmp(modulePath, exeName) == 0) { + exeExeAddr = start; + } +# elif defined(GP_OS_android) + // Use /proc/pid/maps to get the dalvik-jit section since it has no + // associated phdrs. + if (0 == strcmp(modulePath, "/dev/ashmem/dalvik-jit-code-cache")) { + info.AddSharedLibrary( + SharedLibraryAtPath(modulePath, start, end, offset)); + if (info.GetSize() > 10000) { + // LOG("SharedLibraryInfo::GetInfoForSelf(): " + // "implausibly large number of mappings acquired"); + break; + } + } +# endif + } +#endif + + std::vector libInfoList; + + // We collect the bulk of the library info using dl_iterate_phdr. + dl_iterate_phdr(dl_iterate_callback, &libInfoList); + + for (const auto& libInfo : libInfoList) { + info.AddSharedLibrary( + SharedLibraryAtPath(libInfo.mName.c_str(), libInfo.mFirstMappingStart, + libInfo.mLastMappingEnd, + libInfo.mFirstMappingStart - libInfo.mBaseAddress)); + } + +#if defined(GP_OS_linux) + // Make another pass over the information we just harvested from + // dl_iterate_phdr. If we see a nameless object mapped at what we earlier + // established to be the main executable's load address, attach the + // executable's name to that entry. + for (size_t i = 0; i < info.GetSize(); i++) { + SharedLibrary& lib = info.GetMutableEntry(i); + if (lib.GetStart() <= exeExeAddr && exeExeAddr <= lib.GetEnd() && + lib.GetDebugPath().empty()) { + lib = SharedLibraryAtPath(exeName, lib.GetStart(), lib.GetEnd(), + lib.GetOffset()); + + // We only expect to see one such entry. + break; + } + } +#endif + + return info; +} + +void SharedLibraryInfo::Initialize() { /* do nothing */ +} diff --git a/mozglue/baseprofiler/core/shared-libraries-macos.cc b/mozglue/baseprofiler/core/shared-libraries-macos.cc new file mode 100644 index 0000000000..f995581001 --- /dev/null +++ b/mozglue/baseprofiler/core/shared-libraries-macos.cc @@ -0,0 +1,206 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include "BaseProfilerSharedLibraries.h" + +#include "platform.h" + +#include "mozilla/Unused.h" +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +// Architecture specific abstraction. +#if defined(GP_ARCH_x86) +typedef mach_header platform_mach_header; +typedef segment_command mach_segment_command_type; +# define MACHO_MAGIC_NUMBER MH_MAGIC +# define CMD_SEGMENT LC_SEGMENT +# define seg_size uint32_t +#else +typedef mach_header_64 platform_mach_header; +typedef segment_command_64 mach_segment_command_type; +# define MACHO_MAGIC_NUMBER MH_MAGIC_64 +# define CMD_SEGMENT LC_SEGMENT_64 +# define seg_size uint64_t +#endif + +struct NativeSharedLibrary { + const platform_mach_header* header; + std::string path; +}; +static std::vector* sSharedLibrariesList = nullptr; + +class MOZ_RAII SharedLibrariesLock { + public: + SharedLibrariesLock() { sSharedLibrariesMutex.Lock(); } + + ~SharedLibrariesLock() { sSharedLibrariesMutex.Unlock(); } + + SharedLibrariesLock(const SharedLibrariesLock&) = delete; + void operator=(const SharedLibrariesLock&) = delete; + + private: + static mozilla::baseprofiler::detail::BaseProfilerMutex sSharedLibrariesMutex; +}; + +mozilla::baseprofiler::detail::BaseProfilerMutex + SharedLibrariesLock::sSharedLibrariesMutex; + +static void SharedLibraryAddImage(const struct mach_header* mh, + intptr_t vmaddr_slide) { + // NOTE: Presumably for backwards-compatibility reasons, this function accepts + // a mach_header even on 64-bit where it ought to be a mach_header_64. We cast + // it to the right type here. + auto header = reinterpret_cast(mh); + + Dl_info info; + if (!dladdr(header, &info)) { + return; + } + + SharedLibrariesLock lock; + if (!sSharedLibrariesList) { + return; + } + + NativeSharedLibrary lib = {header, info.dli_fname}; + sSharedLibrariesList->push_back(lib); +} + +static void SharedLibraryRemoveImage(const struct mach_header* mh, + intptr_t vmaddr_slide) { + // NOTE: Presumably for backwards-compatibility reasons, this function accepts + // a mach_header even on 64-bit where it ought to be a mach_header_64. We cast + // it to the right type here. + auto header = reinterpret_cast(mh); + + SharedLibrariesLock lock; + if (!sSharedLibrariesList) { + return; + } + + uint32_t count = sSharedLibrariesList->size(); + for (uint32_t i = 0; i < count; ++i) { + if ((*sSharedLibrariesList)[i].header == header) { + sSharedLibrariesList->erase(sSharedLibrariesList->begin() + i); + return; + } + } +} + +void SharedLibraryInfo::Initialize() { + // NOTE: We intentionally leak this memory here. We're allocating dynamically + // in order to avoid static initializers. + sSharedLibrariesList = new std::vector(); + + _dyld_register_func_for_add_image(SharedLibraryAddImage); + _dyld_register_func_for_remove_image(SharedLibraryRemoveImage); +} + +static void addSharedLibrary(const platform_mach_header* header, + const char* path, SharedLibraryInfo& info) { + const struct load_command* cmd = + reinterpret_cast(header + 1); + + seg_size size = 0; + unsigned long long start = reinterpret_cast(header); + // Find the cmd segment in the macho image. It will contain the offset we care + // about. + const uint8_t* uuid_bytes = nullptr; + for (unsigned int i = 0; + cmd && (i < header->ncmds) && (uuid_bytes == nullptr || size == 0); + ++i) { + if (cmd->cmd == CMD_SEGMENT) { + const mach_segment_command_type* seg = + reinterpret_cast(cmd); + + if (!strcmp(seg->segname, "__TEXT")) { + size = seg->vmsize; + } + } else if (cmd->cmd == LC_UUID) { + const uuid_command* ucmd = reinterpret_cast(cmd); + uuid_bytes = ucmd->uuid; + } + + cmd = reinterpret_cast( + reinterpret_cast(cmd) + cmd->cmdsize); + } + + std::string uuid; + std::string breakpadId; + if (uuid_bytes != nullptr) { + static constexpr char digits[16] = {'0', '1', '2', '3', '4', '5', '6', '7', + '8', '9', 'A', 'B', 'C', 'D', 'E', 'F'}; + for (int i = 0; i < 16; ++i) { + uint8_t byte = uuid_bytes[i]; + uuid += digits[byte >> 4]; + uuid += digits[byte & 0xFu]; + } + + // Breakpad id is the same as the uuid but with the additional trailing 0 + // for the breakpad id age. + breakpadId = uuid; + // breakpad id age. + breakpadId += '0'; + } + + std::string pathStr = path; + + size_t pos = pathStr.rfind('\\'); + std::string nameStr = + (pos != std::string::npos) ? pathStr.substr(pos + 1) : pathStr; + + const NXArchInfo* archInfo = + NXGetArchInfoFromCpuType(header->cputype, header->cpusubtype); + + info.AddSharedLibrary(SharedLibrary( + start, start + size, 0, breakpadId, uuid, nameStr, pathStr, nameStr, + pathStr, std::string{}, archInfo ? archInfo->name : "")); +} + +// Translate the statically stored sSharedLibrariesList information into a +// SharedLibraryInfo object. +SharedLibraryInfo SharedLibraryInfo::GetInfoForSelf() { + SharedLibrariesLock lock; + SharedLibraryInfo sharedLibraryInfo; + + for (auto& info : *sSharedLibrariesList) { + addSharedLibrary(info.header, info.path.c_str(), sharedLibraryInfo); + } + + // Add the entry for dyld itself. + // We only support macOS 10.12+, which corresponds to dyld version 15+. + // dyld version 15 added the dyldPath property. + task_dyld_info_data_t task_dyld_info; + mach_msg_type_number_t count = TASK_DYLD_INFO_COUNT; + if (task_info(mach_task_self(), TASK_DYLD_INFO, (task_info_t)&task_dyld_info, + &count) != KERN_SUCCESS) { + return sharedLibraryInfo; + } + + struct dyld_all_image_infos* aii = + (struct dyld_all_image_infos*)task_dyld_info.all_image_info_addr; + if (aii->version >= 15) { + const platform_mach_header* header = + reinterpret_cast( + aii->dyldImageLoadAddress); + addSharedLibrary(header, aii->dyldPath, sharedLibraryInfo); + } + + return sharedLibraryInfo; +} diff --git a/mozglue/baseprofiler/core/shared-libraries-win32.cc b/mozglue/baseprofiler/core/shared-libraries-win32.cc new file mode 100644 index 0000000000..eb6ca3572a --- /dev/null +++ b/mozglue/baseprofiler/core/shared-libraries-win32.cc @@ -0,0 +1,222 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* This Source Code Form is subject to the terms of the Mozilla Public + * License, v. 2.0. If a copy of the MPL was not distributed with this + * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ + +#include + +#include "BaseProfilerSharedLibraries.h" + +#include "mozilla/glue/WindowsUnicode.h" +#include "mozilla/NativeNt.h" +#include "mozilla/WindowsEnumProcessModules.h" +#include "mozilla/WindowsVersion.h" + +#include +#include + +static constexpr char uppercaseDigits[16] = {'0', '1', '2', '3', '4', '5', + '6', '7', '8', '9', 'A', 'B', + 'C', 'D', 'E', 'F'}; +static constexpr char lowercaseDigits[16] = {'0', '1', '2', '3', '4', '5', + '6', '7', '8', '9', 'a', 'b', + 'c', 'd', 'e', 'f'}; + +static void AppendHex(const unsigned char* aBegin, const unsigned char* aEnd, + std::string& aOut) { + for (const unsigned char* p = aBegin; p < aEnd; ++p) { + unsigned char c = *p; + aOut += uppercaseDigits[c >> 4]; + aOut += uppercaseDigits[c & 0xFu]; + } +} + +static constexpr bool WITH_PADDING = true; +static constexpr bool WITHOUT_PADDING = false; +static constexpr bool LOWERCASE = true; +static constexpr bool UPPERCASE = false; +template +static void AppendHex(T aValue, std::string& aOut, bool aWithPadding, + bool aLowercase = UPPERCASE) { + for (int i = sizeof(T) * 2 - 1; i >= 0; --i) { + unsigned nibble = (aValue >> (i * 4)) & 0xFu; + // If no-padding requested, skip starting zeroes -- unless we're on the very + // last nibble (so we don't output a blank). + if (!aWithPadding && i != 0) { + if (nibble == 0) { + // Requested no padding, skip zeroes. + continue; + } + // Requested no padding, got first non-zero, pretend we now want padding + // so we don't skip zeroes anymore. + aWithPadding = true; + } + aOut += aLowercase ? lowercaseDigits[nibble] : uppercaseDigits[nibble]; + } +} + +static bool IsModuleUnsafeToLoad(const std::string& aModuleName) { + auto LowerCaseEqualsLiteral = [](char aModuleChar, char aDetouredChar) { + return std::tolower(aModuleChar) == aDetouredChar; + }; + +#if defined(_M_AMD64) || defined(_M_IX86) + // Hackaround for Bug 1607574. Nvidia's shim driver nvd3d9wrap[x].dll detours + // LoadLibraryExW and it causes AV when the following conditions are met. + // 1. LoadLibraryExW was called for "detoured.dll" + // 2. nvinit[x].dll was unloaded + // 3. OS version is older than 6.2 +# if defined(_M_AMD64) + LPCWSTR kNvidiaShimDriver = L"nvd3d9wrapx.dll"; + LPCWSTR kNvidiaInitDriver = L"nvinitx.dll"; +# elif defined(_M_IX86) + LPCWSTR kNvidiaShimDriver = L"nvd3d9wrap.dll"; + LPCWSTR kNvidiaInitDriver = L"nvinit.dll"; +# endif + constexpr std::string_view detoured_dll = "detoured.dll"; + if (std::equal(aModuleName.cbegin(), aModuleName.cend(), + detoured_dll.cbegin(), detoured_dll.cend(), + LowerCaseEqualsLiteral) && + !mozilla::IsWin8OrLater() && ::GetModuleHandleW(kNvidiaShimDriver) && + !::GetModuleHandleW(kNvidiaInitDriver)) { + return true; + } +#endif // defined(_M_AMD64) || defined(_M_IX86) + + // Hackaround for Bug 1723868. There is no safe way to prevent the module + // Microsoft's VP9 Video Decoder from being unloaded because mfplat.dll may + // have posted more than one task to unload the module in the work queue + // without calling LoadLibrary. + constexpr std::string_view vp9_decoder_dll = "msvp9dec_store.dll"; + if (std::equal(aModuleName.cbegin(), aModuleName.cend(), + vp9_decoder_dll.cbegin(), vp9_decoder_dll.cend(), + LowerCaseEqualsLiteral)) { + return true; + } + + return false; +} + +void SharedLibraryInfo::AddSharedLibraryFromModuleInfo( + const wchar_t* aModulePath, mozilla::Maybe aModule) { + mozilla::UniquePtr utf8ModulePath( + mozilla::glue::WideToUTF8(aModulePath)); + if (!utf8ModulePath) { + return; + } + + std::string modulePathStr(utf8ModulePath.get()); + size_t pos = modulePathStr.find_last_of("\\/"); + std::string moduleNameStr = (pos != std::string::npos) + ? modulePathStr.substr(pos + 1) + : modulePathStr; + + // If the module is unsafe to call LoadLibraryEx for, we skip. + if (IsModuleUnsafeToLoad(moduleNameStr)) { + return; + } + + // Load the module again - to make sure that its handle will remain valid as + // we attempt to read the PDB information from it - or for the first time if + // we only have a path. We want to load the DLL without running the newly + // loaded module's DllMain function, but not as a data file because we want + // to be able to do RVA computations easily. Hence, we use the flag + // LOAD_LIBRARY_AS_IMAGE_RESOURCE which ensures that the sections (not PE + // headers) will be relocated by the loader. Otherwise GetPdbInfo() and/or + // GetVersionInfo() can cause a crash. If the original handle |aModule| is + // valid, LoadLibraryEx just increments its refcount. + nsModuleHandle handleLock( + ::LoadLibraryExW(aModulePath, NULL, LOAD_LIBRARY_AS_IMAGE_RESOURCE)); + if (!handleLock) { + return; + } + + mozilla::nt::PEHeaders headers(handleLock.get()); + if (!headers) { + return; + } + + mozilla::Maybe> bounds = headers.GetBounds(); + if (!bounds) { + return; + } + + // Put the original |aModule| into SharedLibrary, but we get debug info + // from |handleLock| as |aModule| might be inaccessible. + const uintptr_t modStart = + aModule.isSome() ? reinterpret_cast(*aModule) + : reinterpret_cast(handleLock.get()); + const uintptr_t modEnd = modStart + bounds->length(); + + std::string breakpadId; + std::string pdbPathStr; + std::string pdbNameStr; + if (const auto* debugInfo = headers.GetPdbInfo()) { + MOZ_ASSERT(breakpadId.empty()); + const GUID& pdbSig = debugInfo->pdbSignature; + AppendHex(pdbSig.Data1, breakpadId, WITH_PADDING); + AppendHex(pdbSig.Data2, breakpadId, WITH_PADDING); + AppendHex(pdbSig.Data3, breakpadId, WITH_PADDING); + AppendHex(reinterpret_cast(&pdbSig.Data4), + reinterpret_cast(&pdbSig.Data4) + + sizeof(pdbSig.Data4), + breakpadId); + AppendHex(debugInfo->pdbAge, breakpadId, WITHOUT_PADDING); + + // The PDB file name could be different from module filename, + // so report both + // e.g. The PDB for C:\Windows\SysWOW64\ntdll.dll is wntdll.pdb + pdbPathStr = debugInfo->pdbFileName; + size_t pos = pdbPathStr.find_last_of("\\/"); + pdbNameStr = + (pos != std::string::npos) ? pdbPathStr.substr(pos + 1) : pdbPathStr; + } + + std::string codeId; + DWORD timestamp; + DWORD imageSize; + if (headers.GetTimeStamp(timestamp) && headers.GetImageSize(imageSize)) { + AppendHex(timestamp, codeId, WITH_PADDING); + AppendHex(imageSize, codeId, WITHOUT_PADDING, LOWERCASE); + } + + std::string versionStr; + uint64_t version; + if (headers.GetVersionInfo(version)) { + versionStr += std::to_string((version >> 48) & 0xFFFF); + versionStr += '.'; + versionStr += std::to_string((version >> 32) & 0xFFFF); + versionStr += '.'; + versionStr += std::to_string((version >> 16) & 0xFFFF); + versionStr += '.'; + versionStr += std::to_string(version & 0xFFFF); + } + + SharedLibrary shlib(modStart, modEnd, + 0, // DLLs are always mapped at offset 0 on Windows + breakpadId, codeId, moduleNameStr, modulePathStr, + pdbNameStr, pdbPathStr, versionStr, ""); + AddSharedLibrary(shlib); +} + +SharedLibraryInfo SharedLibraryInfo::GetInfoForSelf() { + SharedLibraryInfo sharedLibraryInfo; + + auto addSharedLibraryFromModuleInfo = + [&sharedLibraryInfo](const wchar_t* aModulePath, HMODULE aModule) { + sharedLibraryInfo.AddSharedLibraryFromModuleInfo( + aModulePath, mozilla::Some(aModule)); + }; + + mozilla::EnumerateProcessModules(addSharedLibraryFromModuleInfo); + return sharedLibraryInfo; +} + +SharedLibraryInfo SharedLibraryInfo::GetInfoFromPath(const wchar_t* aPath) { + SharedLibraryInfo sharedLibraryInfo; + sharedLibraryInfo.AddSharedLibraryFromModuleInfo(aPath, mozilla::Nothing()); + return sharedLibraryInfo; +} + +void SharedLibraryInfo::Initialize() { /* do nothing */ +} diff --git a/mozglue/baseprofiler/core/vtune/ittnotify.h b/mozglue/baseprofiler/core/vtune/ittnotify.h new file mode 100644 index 0000000000..04adf9eb5e --- /dev/null +++ b/mozglue/baseprofiler/core/vtune/ittnotify.h @@ -0,0 +1,4127 @@ +// clang-format off + +/* + This file is provided under a dual BSD/GPLv2 license. When using or + redistributing this file, you may do so under either license. + + GPL LICENSE SUMMARY + + Copyright (c) 2005-2014 Intel Corporation. All rights reserved. + + This program is free software; you can redistribute it and/or modify + it under the terms of version 2 of the GNU General Public License as + published by the Free Software Foundation. + + This program is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program; if not, write to the Free Software + Foundation, Inc., 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. + The full GNU General Public License is included in this distribution + in the file called LICENSE.GPL. + + Contact Information: + http://software.intel.com/en-us/articles/intel-vtune-amplifier-xe/ + + BSD LICENSE + + Copyright (c) 2005-2014 Intel Corporation. All rights reserved. + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * 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. + * Neither the name of Intel Corporation nor the names of its + contributors may be used to endorse or promote products derived + from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "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 THE COPYRIGHT + OWNER 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. + */ +#ifndef _ITTNOTIFY_H_ +# define _ITTNOTIFY_H_ + +/** +@file +@brief Public User API functions and types +@mainpage + +The ITT API is used to annotate a user's program with additional information +that can be used by correctness and performance tools. The user inserts +calls in their program. Those calls generate information that is collected +at runtime, and used by Intel(R) Threading Tools. + +@section API Concepts +The following general concepts are used throughout the API. + +@subsection Unicode Support +Many API functions take character string arguments. On Windows, there +are two versions of each such function. The function name is suffixed +by W if Unicode support is enabled, and by A otherwise. Any API function +that takes a character string argument adheres to this convention. + +@subsection Conditional Compilation +Many users prefer having an option to modify ITT API code when linking it +inside their runtimes. ITT API header file provides a mechanism to replace +ITT API function names inside your code with empty strings. To do this, +define the macros INTEL_NO_ITTNOTIFY_API during compilation and remove the +static library from the linker script. + +@subsection Domains +[see domains] +Domains provide a way to separate notification for different modules or +libraries in a program. Domains are specified by dotted character strings, +e.g. TBB.Internal.Control. + +A mechanism (to be specified) is provided to enable and disable +domains. By default, all domains are enabled. +@subsection Named Entities and Instances +Named entities (frames, regions, tasks, and markers) communicate +information about the program to the analysis tools. A named entity often +refers to a section of program code, or to some set of logical concepts +that the programmer wants to group together. + +Named entities relate to the programmer's static view of the program. When +the program actually executes, many instances of a given named entity +may be created. + +The API annotations denote instances of named entities. The actual +named entities are displayed using the analysis tools. In other words, +the named entities come into existence when instances are created. + +Instances of named entities may have instance identifiers (IDs). Some +API calls use instance identifiers to create relationships between +different instances of named entities. Other API calls associate data +with instances of named entities. + +Some named entities must always have instance IDs. In particular, regions +and frames always have IDs. Task and markers need IDs only if the ID is +needed in another API call (such as adding a relation or metadata). + +The lifetime of instance IDs is distinct from the lifetime of +instances. This allows various relationships to be specified separate +from the actual execution of instances. This flexibility comes at the +expense of extra API calls. + +The same ID may not be reused for different instances, unless a previous +[ref] __itt_id_destroy call for that ID has been issued. +*/ + +/** @cond exclude_from_documentation */ +#ifndef ITT_OS_WIN +# define ITT_OS_WIN 1 +#endif /* ITT_OS_WIN */ + +#ifndef ITT_OS_LINUX +# define ITT_OS_LINUX 2 +#endif /* ITT_OS_LINUX */ + +#ifndef ITT_OS_MAC +# define ITT_OS_MAC 3 +#endif /* ITT_OS_MAC */ + +#ifndef ITT_OS_FREEBSD +# define ITT_OS_FREEBSD 4 +#endif /* ITT_OS_FREEBSD */ + +#ifndef ITT_OS +# if defined WIN32 || defined _WIN32 +# define ITT_OS ITT_OS_WIN +# elif defined( __APPLE__ ) && defined( __MACH__ ) +# define ITT_OS ITT_OS_MAC +# elif defined( __FreeBSD__ ) +# define ITT_OS ITT_OS_FREEBSD +# else +# define ITT_OS ITT_OS_LINUX +# endif +#endif /* ITT_OS */ + +#ifndef ITT_PLATFORM_WIN +# define ITT_PLATFORM_WIN 1 +#endif /* ITT_PLATFORM_WIN */ + +#ifndef ITT_PLATFORM_POSIX +# define ITT_PLATFORM_POSIX 2 +#endif /* ITT_PLATFORM_POSIX */ + +#ifndef ITT_PLATFORM_MAC +# define ITT_PLATFORM_MAC 3 +#endif /* ITT_PLATFORM_MAC */ + +#ifndef ITT_PLATFORM_FREEBSD +# define ITT_PLATFORM_FREEBSD 4 +#endif /* ITT_PLATFORM_FREEBSD */ + +#ifndef ITT_PLATFORM +# if ITT_OS==ITT_OS_WIN +# define ITT_PLATFORM ITT_PLATFORM_WIN +# elif ITT_OS==ITT_OS_MAC +# define ITT_PLATFORM ITT_PLATFORM_MAC +# elif ITT_OS==ITT_OS_FREEBSD +# define ITT_PLATFORM ITT_PLATFORM_FREEBSD +# else +# define ITT_PLATFORM ITT_PLATFORM_POSIX +# endif +#endif /* ITT_PLATFORM */ + +#if defined(_UNICODE) && !defined(UNICODE) +#define UNICODE +#endif + +#include +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#include +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#include +#if defined(UNICODE) || defined(_UNICODE) +#include +#endif /* UNICODE || _UNICODE */ +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +#ifndef ITTAPI_CDECL +# if ITT_PLATFORM==ITT_PLATFORM_WIN +# define ITTAPI_CDECL __cdecl +# else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +# if defined _M_IX86 || defined __i386__ +# define ITTAPI_CDECL __attribute__ ((cdecl)) +# else /* _M_IX86 || __i386__ */ +# define ITTAPI_CDECL /* actual only on x86 platform */ +# endif /* _M_IX86 || __i386__ */ +# endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* ITTAPI_CDECL */ + +#ifndef STDCALL +# if ITT_PLATFORM==ITT_PLATFORM_WIN +# define STDCALL __stdcall +# else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +# if defined _M_IX86 || defined __i386__ +# define STDCALL __attribute__ ((stdcall)) +# else /* _M_IX86 || __i386__ */ +# define STDCALL /* supported only on x86 platform */ +# endif /* _M_IX86 || __i386__ */ +# endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* STDCALL */ + +#define ITTAPI ITTAPI_CDECL +#define LIBITTAPI ITTAPI_CDECL + +/* TODO: Temporary for compatibility! */ +#define ITTAPI_CALL ITTAPI_CDECL +#define LIBITTAPI_CALL ITTAPI_CDECL + +#if ITT_PLATFORM==ITT_PLATFORM_WIN +/* use __forceinline (VC++ specific) */ +#define ITT_INLINE __forceinline +#define ITT_INLINE_ATTRIBUTE /* nothing */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +/* + * Generally, functions are not inlined unless optimization is specified. + * For functions declared inline, this attribute inlines the function even + * if no optimization level was specified. + */ +#ifdef __STRICT_ANSI__ +#define ITT_INLINE static +#define ITT_INLINE_ATTRIBUTE __attribute__((unused)) +#else /* __STRICT_ANSI__ */ +#define ITT_INLINE static inline +#define ITT_INLINE_ATTRIBUTE __attribute__((always_inline, unused)) +#endif /* __STRICT_ANSI__ */ +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +/** @endcond */ + +#ifdef INTEL_ITTNOTIFY_ENABLE_LEGACY +# if ITT_PLATFORM==ITT_PLATFORM_WIN +# pragma message("WARNING!!! Deprecated API is used. Please undefine INTEL_ITTNOTIFY_ENABLE_LEGACY macro") +# else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +# warning "Deprecated API is used. Please undefine INTEL_ITTNOTIFY_ENABLE_LEGACY macro" +# endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +# include "vtune/legacy/ittnotify.h" +#endif /* INTEL_ITTNOTIFY_ENABLE_LEGACY */ + +/** @cond exclude_from_documentation */ +/* Helper macro for joining tokens */ +#define ITT_JOIN_AUX(p,n) p##n +#define ITT_JOIN(p,n) ITT_JOIN_AUX(p,n) + +#ifdef ITT_MAJOR +#undef ITT_MAJOR +#endif +#ifdef ITT_MINOR +#undef ITT_MINOR +#endif +#define ITT_MAJOR 3 +#define ITT_MINOR 0 + +/* Standard versioning of a token with major and minor version numbers */ +#define ITT_VERSIONIZE(x) \ + ITT_JOIN(x, \ + ITT_JOIN(_, \ + ITT_JOIN(ITT_MAJOR, \ + ITT_JOIN(_, ITT_MINOR)))) + +#ifndef INTEL_ITTNOTIFY_PREFIX +# define INTEL_ITTNOTIFY_PREFIX __itt_ +#endif /* INTEL_ITTNOTIFY_PREFIX */ +#ifndef INTEL_ITTNOTIFY_POSTFIX +# define INTEL_ITTNOTIFY_POSTFIX _ptr_ +#endif /* INTEL_ITTNOTIFY_POSTFIX */ + +#define ITTNOTIFY_NAME_AUX(n) ITT_JOIN(INTEL_ITTNOTIFY_PREFIX,n) +#define ITTNOTIFY_NAME(n) ITT_VERSIONIZE(ITTNOTIFY_NAME_AUX(ITT_JOIN(n,INTEL_ITTNOTIFY_POSTFIX))) + +#define ITTNOTIFY_VOID(n) (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n) +#define ITTNOTIFY_DATA(n) (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n) + +#define ITTNOTIFY_VOID_D0(n,d) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d) +#define ITTNOTIFY_VOID_D1(n,d,x) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x) +#define ITTNOTIFY_VOID_D2(n,d,x,y) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x,y) +#define ITTNOTIFY_VOID_D3(n,d,x,y,z) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x,y,z) +#define ITTNOTIFY_VOID_D4(n,d,x,y,z,a) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x,y,z,a) +#define ITTNOTIFY_VOID_D5(n,d,x,y,z,a,b) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x,y,z,a,b) +#define ITTNOTIFY_VOID_D6(n,d,x,y,z,a,b,c) (!(d)->flags) ? (void)0 : (!ITTNOTIFY_NAME(n)) ? (void)0 : ITTNOTIFY_NAME(n)(d,x,y,z,a,b,c) +#define ITTNOTIFY_DATA_D0(n,d) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d) +#define ITTNOTIFY_DATA_D1(n,d,x) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x) +#define ITTNOTIFY_DATA_D2(n,d,x,y) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x,y) +#define ITTNOTIFY_DATA_D3(n,d,x,y,z) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x,y,z) +#define ITTNOTIFY_DATA_D4(n,d,x,y,z,a) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x,y,z,a) +#define ITTNOTIFY_DATA_D5(n,d,x,y,z,a,b) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x,y,z,a,b) +#define ITTNOTIFY_DATA_D6(n,d,x,y,z,a,b,c) (!(d)->flags) ? 0 : (!ITTNOTIFY_NAME(n)) ? 0 : ITTNOTIFY_NAME(n)(d,x,y,z,a,b,c) + +#ifdef ITT_STUB +#undef ITT_STUB +#endif +#ifdef ITT_STUBV +#undef ITT_STUBV +#endif +#define ITT_STUBV(api,type,name,args) \ + typedef type (api* ITT_JOIN(ITTNOTIFY_NAME(name),_t)) args; \ + extern ITT_JOIN(ITTNOTIFY_NAME(name),_t) ITTNOTIFY_NAME(name); +#define ITT_STUB ITT_STUBV +/** @endcond */ + +#ifdef __cplusplus +extern "C" { +#endif /* __cplusplus */ + +/** @cond exclude_from_gpa_documentation */ +/** + * @defgroup public Public API + * @{ + * @} + */ + +/** + * @defgroup control Collection Control + * @ingroup public + * General behavior: application continues to run, but no profiling information is being collected + * + * Pausing occurs not only for the current thread but for all process as well as spawned processes + * - Intel(R) Parallel Inspector and Intel(R) Inspector XE: + * - Does not analyze or report errors that involve memory access. + * - Other errors are reported as usual. Pausing data collection in + * Intel(R) Parallel Inspector and Intel(R) Inspector XE + * only pauses tracing and analyzing memory access. + * It does not pause tracing or analyzing threading APIs. + * . + * - Intel(R) Parallel Amplifier and Intel(R) VTune(TM) Amplifier XE: + * - Does continue to record when new threads are started. + * . + * - Other effects: + * - Possible reduction of runtime overhead. + * . + * @{ + */ +/** @brief Pause collection */ +void ITTAPI __itt_pause(void); +/** @brief Resume collection */ +void ITTAPI __itt_resume(void); +/** @brief Detach collection */ +void ITTAPI __itt_detach(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, pause, (void)) +ITT_STUBV(ITTAPI, void, resume, (void)) +ITT_STUBV(ITTAPI, void, detach, (void)) +#define __itt_pause ITTNOTIFY_VOID(pause) +#define __itt_pause_ptr ITTNOTIFY_NAME(pause) +#define __itt_resume ITTNOTIFY_VOID(resume) +#define __itt_resume_ptr ITTNOTIFY_NAME(resume) +#define __itt_detach ITTNOTIFY_VOID(detach) +#define __itt_detach_ptr ITTNOTIFY_NAME(detach) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_pause() +#define __itt_pause_ptr 0 +#define __itt_resume() +#define __itt_resume_ptr 0 +#define __itt_detach() +#define __itt_detach_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_pause_ptr 0 +#define __itt_resume_ptr 0 +#define __itt_detach_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} control group */ +/** @endcond */ + +/** + * @defgroup threads Threads + * @ingroup public + * Give names to threads + * @{ + */ +/** + * @brief Sets thread name of calling thread + * @param[in] name - name of thread + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_thread_set_nameA(const char *name); +void ITTAPI __itt_thread_set_nameW(const wchar_t *name); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_thread_set_name __itt_thread_set_nameW +# define __itt_thread_set_name_ptr __itt_thread_set_nameW_ptr +#else /* UNICODE */ +# define __itt_thread_set_name __itt_thread_set_nameA +# define __itt_thread_set_name_ptr __itt_thread_set_nameA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_thread_set_name(const char *name); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, thread_set_nameA, (const char *name)) +ITT_STUBV(ITTAPI, void, thread_set_nameW, (const wchar_t *name)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUBV(ITTAPI, void, thread_set_name, (const char *name)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_thread_set_nameA ITTNOTIFY_VOID(thread_set_nameA) +#define __itt_thread_set_nameA_ptr ITTNOTIFY_NAME(thread_set_nameA) +#define __itt_thread_set_nameW ITTNOTIFY_VOID(thread_set_nameW) +#define __itt_thread_set_nameW_ptr ITTNOTIFY_NAME(thread_set_nameW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_thread_set_name ITTNOTIFY_VOID(thread_set_name) +#define __itt_thread_set_name_ptr ITTNOTIFY_NAME(thread_set_name) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_thread_set_nameA(name) +#define __itt_thread_set_nameA_ptr 0 +#define __itt_thread_set_nameW(name) +#define __itt_thread_set_nameW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_thread_set_name(name) +#define __itt_thread_set_name_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_thread_set_nameA_ptr 0 +#define __itt_thread_set_nameW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_thread_set_name_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @cond exclude_from_gpa_documentation */ + +/** + * @brief Mark current thread as ignored from this point on, for the duration of its existence. + */ +void ITTAPI __itt_thread_ignore(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, thread_ignore, (void)) +#define __itt_thread_ignore ITTNOTIFY_VOID(thread_ignore) +#define __itt_thread_ignore_ptr ITTNOTIFY_NAME(thread_ignore) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_thread_ignore() +#define __itt_thread_ignore_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_thread_ignore_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} threads group */ + +/** + * @defgroup suppress Error suppression + * @ingroup public + * General behavior: application continues to run, but errors are suppressed + * + * @{ + */ + +/*****************************************************************//** + * @name group of functions used for error suppression in correctness tools + *********************************************************************/ +/** @{ */ +/** + * @hideinitializer + * @brief possible value for suppression mask + */ +#define __itt_suppress_all_errors 0x7fffffff + +/** + * @hideinitializer + * @brief possible value for suppression mask (suppresses errors from threading analysis) + */ +#define __itt_suppress_threading_errors 0x000000ff + +/** + * @hideinitializer + * @brief possible value for suppression mask (suppresses errors from memory analysis) + */ +#define __itt_suppress_memory_errors 0x0000ff00 + +/** + * @brief Start suppressing errors identified in mask on this thread + */ +void ITTAPI __itt_suppress_push(unsigned int mask); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, suppress_push, (unsigned int mask)) +#define __itt_suppress_push ITTNOTIFY_VOID(suppress_push) +#define __itt_suppress_push_ptr ITTNOTIFY_NAME(suppress_push) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_suppress_push(mask) +#define __itt_suppress_push_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_suppress_push_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Undo the effects of the matching call to __itt_suppress_push + */ +void ITTAPI __itt_suppress_pop(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, suppress_pop, (void)) +#define __itt_suppress_pop ITTNOTIFY_VOID(suppress_pop) +#define __itt_suppress_pop_ptr ITTNOTIFY_NAME(suppress_pop) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_suppress_pop() +#define __itt_suppress_pop_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_suppress_pop_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @enum __itt_model_disable + * @brief Enumerator for the disable methods + */ +typedef enum __itt_suppress_mode { + __itt_unsuppress_range, + __itt_suppress_range +} __itt_suppress_mode_t; + +/** + * @brief Mark a range of memory for error suppression or unsuppression for error types included in mask + */ +void ITTAPI __itt_suppress_mark_range(__itt_suppress_mode_t mode, unsigned int mask, void * address, size_t size); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, suppress_mark_range, (__itt_suppress_mode_t mode, unsigned int mask, void * address, size_t size)) +#define __itt_suppress_mark_range ITTNOTIFY_VOID(suppress_mark_range) +#define __itt_suppress_mark_range_ptr ITTNOTIFY_NAME(suppress_mark_range) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_suppress_mark_range(mask) +#define __itt_suppress_mark_range_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_suppress_mark_range_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Undo the effect of a matching call to __itt_suppress_mark_range. If not matching + * call is found, nothing is changed. + */ +void ITTAPI __itt_suppress_clear_range(__itt_suppress_mode_t mode, unsigned int mask, void * address, size_t size); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, suppress_clear_range, (__itt_suppress_mode_t mode, unsigned int mask, void * address, size_t size)) +#define __itt_suppress_clear_range ITTNOTIFY_VOID(suppress_clear_range) +#define __itt_suppress_clear_range_ptr ITTNOTIFY_NAME(suppress_clear_range) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_suppress_clear_range(mask) +#define __itt_suppress_clear_range_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_suppress_clear_range_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} */ +/** @} suppress group */ + +/** + * @defgroup sync Synchronization + * @ingroup public + * Indicate user-written synchronization code + * @{ + */ +/** + * @hideinitializer + * @brief possible value of attribute argument for sync object type + */ +#define __itt_attr_barrier 1 + +/** + * @hideinitializer + * @brief possible value of attribute argument for sync object type + */ +#define __itt_attr_mutex 2 + +/** +@brief Name a synchronization object +@param[in] addr Handle for the synchronization object. You should +use a real address to uniquely identify the synchronization object. +@param[in] objtype null-terminated object type string. If NULL is +passed, the name will be "User Synchronization". +@param[in] objname null-terminated object name string. If NULL, +no name will be assigned to the object. +@param[in] attribute one of [#__itt_attr_barrier, #__itt_attr_mutex] + */ + +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_sync_createA(void *addr, const char *objtype, const char *objname, int attribute); +void ITTAPI __itt_sync_createW(void *addr, const wchar_t *objtype, const wchar_t *objname, int attribute); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_sync_create __itt_sync_createW +# define __itt_sync_create_ptr __itt_sync_createW_ptr +#else /* UNICODE */ +# define __itt_sync_create __itt_sync_createA +# define __itt_sync_create_ptr __itt_sync_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_sync_create (void *addr, const char *objtype, const char *objname, int attribute); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, sync_createA, (void *addr, const char *objtype, const char *objname, int attribute)) +ITT_STUBV(ITTAPI, void, sync_createW, (void *addr, const wchar_t *objtype, const wchar_t *objname, int attribute)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUBV(ITTAPI, void, sync_create, (void *addr, const char* objtype, const char* objname, int attribute)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_createA ITTNOTIFY_VOID(sync_createA) +#define __itt_sync_createA_ptr ITTNOTIFY_NAME(sync_createA) +#define __itt_sync_createW ITTNOTIFY_VOID(sync_createW) +#define __itt_sync_createW_ptr ITTNOTIFY_NAME(sync_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_create ITTNOTIFY_VOID(sync_create) +#define __itt_sync_create_ptr ITTNOTIFY_NAME(sync_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_createA(addr, objtype, objname, attribute) +#define __itt_sync_createA_ptr 0 +#define __itt_sync_createW(addr, objtype, objname, attribute) +#define __itt_sync_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_create(addr, objtype, objname, attribute) +#define __itt_sync_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_createA_ptr 0 +#define __itt_sync_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** +@brief Rename a synchronization object + +You can use the rename call to assign or reassign a name to a given +synchronization object. +@param[in] addr handle for the synchronization object. +@param[in] name null-terminated object name string. +*/ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_sync_renameA(void *addr, const char *name); +void ITTAPI __itt_sync_renameW(void *addr, const wchar_t *name); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_sync_rename __itt_sync_renameW +# define __itt_sync_rename_ptr __itt_sync_renameW_ptr +#else /* UNICODE */ +# define __itt_sync_rename __itt_sync_renameA +# define __itt_sync_rename_ptr __itt_sync_renameA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_sync_rename(void *addr, const char *name); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, sync_renameA, (void *addr, const char *name)) +ITT_STUBV(ITTAPI, void, sync_renameW, (void *addr, const wchar_t *name)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUBV(ITTAPI, void, sync_rename, (void *addr, const char *name)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_renameA ITTNOTIFY_VOID(sync_renameA) +#define __itt_sync_renameA_ptr ITTNOTIFY_NAME(sync_renameA) +#define __itt_sync_renameW ITTNOTIFY_VOID(sync_renameW) +#define __itt_sync_renameW_ptr ITTNOTIFY_NAME(sync_renameW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_rename ITTNOTIFY_VOID(sync_rename) +#define __itt_sync_rename_ptr ITTNOTIFY_NAME(sync_rename) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_renameA(addr, name) +#define __itt_sync_renameA_ptr 0 +#define __itt_sync_renameW(addr, name) +#define __itt_sync_renameW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_rename(addr, name) +#define __itt_sync_rename_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_sync_renameA_ptr 0 +#define __itt_sync_renameW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_sync_rename_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + @brief Destroy a synchronization object. + @param addr Handle for the synchronization object. + */ +void ITTAPI __itt_sync_destroy(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, sync_destroy, (void *addr)) +#define __itt_sync_destroy ITTNOTIFY_VOID(sync_destroy) +#define __itt_sync_destroy_ptr ITTNOTIFY_NAME(sync_destroy) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_sync_destroy(addr) +#define __itt_sync_destroy_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_sync_destroy_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/*****************************************************************//** + * @name group of functions is used for performance measurement tools + *********************************************************************/ +/** @{ */ +/** + * @brief Enter spin loop on user-defined sync object + */ +void ITTAPI __itt_sync_prepare(void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, sync_prepare, (void *addr)) +#define __itt_sync_prepare ITTNOTIFY_VOID(sync_prepare) +#define __itt_sync_prepare_ptr ITTNOTIFY_NAME(sync_prepare) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_sync_prepare(addr) +#define __itt_sync_prepare_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_sync_prepare_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Quit spin loop without acquiring spin object + */ +void ITTAPI __itt_sync_cancel(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, sync_cancel, (void *addr)) +#define __itt_sync_cancel ITTNOTIFY_VOID(sync_cancel) +#define __itt_sync_cancel_ptr ITTNOTIFY_NAME(sync_cancel) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_sync_cancel(addr) +#define __itt_sync_cancel_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_sync_cancel_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Successful spin loop completion (sync object acquired) + */ +void ITTAPI __itt_sync_acquired(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, sync_acquired, (void *addr)) +#define __itt_sync_acquired ITTNOTIFY_VOID(sync_acquired) +#define __itt_sync_acquired_ptr ITTNOTIFY_NAME(sync_acquired) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_sync_acquired(addr) +#define __itt_sync_acquired_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_sync_acquired_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Start sync object releasing code. Is called before the lock release call. + */ +void ITTAPI __itt_sync_releasing(void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, sync_releasing, (void *addr)) +#define __itt_sync_releasing ITTNOTIFY_VOID(sync_releasing) +#define __itt_sync_releasing_ptr ITTNOTIFY_NAME(sync_releasing) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_sync_releasing(addr) +#define __itt_sync_releasing_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_sync_releasing_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} */ + +/** @} sync group */ + +/**************************************************************//** + * @name group of functions is used for correctness checking tools + ******************************************************************/ +/** @{ */ +/** + * @ingroup legacy + * @deprecated Legacy API + * @brief Fast synchronization which does no require spinning. + * - This special function is to be used by TBB and OpenMP libraries only when they know + * there is no spin but they need to suppress TC warnings about shared variable modifications. + * - It only has corresponding pointers in static library and does not have corresponding function + * in dynamic library. + * @see void __itt_sync_prepare(void* addr); + */ +void ITTAPI __itt_fsync_prepare(void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, fsync_prepare, (void *addr)) +#define __itt_fsync_prepare ITTNOTIFY_VOID(fsync_prepare) +#define __itt_fsync_prepare_ptr ITTNOTIFY_NAME(fsync_prepare) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_fsync_prepare(addr) +#define __itt_fsync_prepare_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_fsync_prepare_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup legacy + * @deprecated Legacy API + * @brief Fast synchronization which does no require spinning. + * - This special function is to be used by TBB and OpenMP libraries only when they know + * there is no spin but they need to suppress TC warnings about shared variable modifications. + * - It only has corresponding pointers in static library and does not have corresponding function + * in dynamic library. + * @see void __itt_sync_cancel(void *addr); + */ +void ITTAPI __itt_fsync_cancel(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, fsync_cancel, (void *addr)) +#define __itt_fsync_cancel ITTNOTIFY_VOID(fsync_cancel) +#define __itt_fsync_cancel_ptr ITTNOTIFY_NAME(fsync_cancel) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_fsync_cancel(addr) +#define __itt_fsync_cancel_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_fsync_cancel_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup legacy + * @deprecated Legacy API + * @brief Fast synchronization which does no require spinning. + * - This special function is to be used by TBB and OpenMP libraries only when they know + * there is no spin but they need to suppress TC warnings about shared variable modifications. + * - It only has corresponding pointers in static library and does not have corresponding function + * in dynamic library. + * @see void __itt_sync_acquired(void *addr); + */ +void ITTAPI __itt_fsync_acquired(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, fsync_acquired, (void *addr)) +#define __itt_fsync_acquired ITTNOTIFY_VOID(fsync_acquired) +#define __itt_fsync_acquired_ptr ITTNOTIFY_NAME(fsync_acquired) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_fsync_acquired(addr) +#define __itt_fsync_acquired_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_fsync_acquired_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup legacy + * @deprecated Legacy API + * @brief Fast synchronization which does no require spinning. + * - This special function is to be used by TBB and OpenMP libraries only when they know + * there is no spin but they need to suppress TC warnings about shared variable modifications. + * - It only has corresponding pointers in static library and does not have corresponding function + * in dynamic library. + * @see void __itt_sync_releasing(void* addr); + */ +void ITTAPI __itt_fsync_releasing(void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, fsync_releasing, (void *addr)) +#define __itt_fsync_releasing ITTNOTIFY_VOID(fsync_releasing) +#define __itt_fsync_releasing_ptr ITTNOTIFY_NAME(fsync_releasing) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_fsync_releasing(addr) +#define __itt_fsync_releasing_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_fsync_releasing_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} */ + +/** + * @defgroup model Modeling by Intel(R) Parallel Advisor + * @ingroup public + * This is the subset of itt used for modeling by Intel(R) Parallel Advisor. + * This API is called ONLY using annotate.h, by "Annotation" macros + * the user places in their sources during the parallelism modeling steps. + * + * site_begin/end and task_begin/end take the address of handle variables, + * which are writeable by the API. Handles must be 0 initialized prior + * to the first call to begin, or may cause a run-time failure. + * The handles are initialized in a multi-thread safe way by the API if + * the handle is 0. The commonly expected idiom is one static handle to + * identify a site or task. If a site or task of the same name has already + * been started during this collection, the same handle MAY be returned, + * but is not required to be - it is unspecified if data merging is done + * based on name. These routines also take an instance variable. Like + * the lexical instance, these must be 0 initialized. Unlike the lexical + * instance, this is used to track a single dynamic instance. + * + * API used by the Intel(R) Parallel Advisor to describe potential concurrency + * and related activities. User-added source annotations expand to calls + * to these procedures to enable modeling of a hypothetical concurrent + * execution serially. + * @{ + */ +#if !defined(_ADVISOR_ANNOTATE_H_) || defined(ANNOTATE_EXPAND_NULL) + +typedef void* __itt_model_site; /*!< @brief handle for lexical site */ +typedef void* __itt_model_site_instance; /*!< @brief handle for dynamic instance */ +typedef void* __itt_model_task; /*!< @brief handle for lexical site */ +typedef void* __itt_model_task_instance; /*!< @brief handle for dynamic instance */ + +/** + * @enum __itt_model_disable + * @brief Enumerator for the disable methods + */ +typedef enum { + __itt_model_disable_observation, + __itt_model_disable_collection +} __itt_model_disable; + +#endif /* !_ADVISOR_ANNOTATE_H_ || ANNOTATE_EXPAND_NULL */ + +/** + * @brief ANNOTATE_SITE_BEGIN/ANNOTATE_SITE_END support. + * + * site_begin/end model a potential concurrency site. + * site instances may be recursively nested with themselves. + * site_end exits the most recently started but unended site for the current + * thread. The handle passed to end may be used to validate structure. + * Instances of a site encountered on different threads concurrently + * are considered completely distinct. If the site name for two different + * lexical sites match, it is unspecified whether they are treated as the + * same or different for data presentation. + */ +void ITTAPI __itt_model_site_begin(__itt_model_site *site, __itt_model_site_instance *instance, const char *name); +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_model_site_beginW(const wchar_t *name); +#endif +void ITTAPI __itt_model_site_beginA(const char *name); +void ITTAPI __itt_model_site_beginAL(const char *name, size_t siteNameLen); +void ITTAPI __itt_model_site_end (__itt_model_site *site, __itt_model_site_instance *instance); +void ITTAPI __itt_model_site_end_2(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_site_begin, (__itt_model_site *site, __itt_model_site_instance *instance, const char *name)) +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, model_site_beginW, (const wchar_t *name)) +#endif +ITT_STUBV(ITTAPI, void, model_site_beginA, (const char *name)) +ITT_STUBV(ITTAPI, void, model_site_beginAL, (const char *name, size_t siteNameLen)) +ITT_STUBV(ITTAPI, void, model_site_end, (__itt_model_site *site, __itt_model_site_instance *instance)) +ITT_STUBV(ITTAPI, void, model_site_end_2, (void)) +#define __itt_model_site_begin ITTNOTIFY_VOID(model_site_begin) +#define __itt_model_site_begin_ptr ITTNOTIFY_NAME(model_site_begin) +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_site_beginW ITTNOTIFY_VOID(model_site_beginW) +#define __itt_model_site_beginW_ptr ITTNOTIFY_NAME(model_site_beginW) +#endif +#define __itt_model_site_beginA ITTNOTIFY_VOID(model_site_beginA) +#define __itt_model_site_beginA_ptr ITTNOTIFY_NAME(model_site_beginA) +#define __itt_model_site_beginAL ITTNOTIFY_VOID(model_site_beginAL) +#define __itt_model_site_beginAL_ptr ITTNOTIFY_NAME(model_site_beginAL) +#define __itt_model_site_end ITTNOTIFY_VOID(model_site_end) +#define __itt_model_site_end_ptr ITTNOTIFY_NAME(model_site_end) +#define __itt_model_site_end_2 ITTNOTIFY_VOID(model_site_end_2) +#define __itt_model_site_end_2_ptr ITTNOTIFY_NAME(model_site_end_2) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_site_begin(site, instance, name) +#define __itt_model_site_begin_ptr 0 +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_site_beginW(name) +#define __itt_model_site_beginW_ptr 0 +#endif +#define __itt_model_site_beginA(name) +#define __itt_model_site_beginA_ptr 0 +#define __itt_model_site_beginAL(name, siteNameLen) +#define __itt_model_site_beginAL_ptr 0 +#define __itt_model_site_end(site, instance) +#define __itt_model_site_end_ptr 0 +#define __itt_model_site_end_2() +#define __itt_model_site_end_2_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_site_begin_ptr 0 +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_site_beginW_ptr 0 +#endif +#define __itt_model_site_beginA_ptr 0 +#define __itt_model_site_beginAL_ptr 0 +#define __itt_model_site_end_ptr 0 +#define __itt_model_site_end_2_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_TASK_BEGIN/ANNOTATE_TASK_END support + * + * task_begin/end model a potential task, which is contained within the most + * closely enclosing dynamic site. task_end exits the most recently started + * but unended task. The handle passed to end may be used to validate + * structure. It is unspecified if bad dynamic nesting is detected. If it + * is, it should be encoded in the resulting data collection. The collector + * should not fail due to construct nesting issues, nor attempt to directly + * indicate the problem. + */ +void ITTAPI __itt_model_task_begin(__itt_model_task *task, __itt_model_task_instance *instance, const char *name); +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_model_task_beginW(const wchar_t *name); +void ITTAPI __itt_model_iteration_taskW(const wchar_t *name); +#endif +void ITTAPI __itt_model_task_beginA(const char *name); +void ITTAPI __itt_model_task_beginAL(const char *name, size_t taskNameLen); +void ITTAPI __itt_model_iteration_taskA(const char *name); +void ITTAPI __itt_model_iteration_taskAL(const char *name, size_t taskNameLen); +void ITTAPI __itt_model_task_end (__itt_model_task *task, __itt_model_task_instance *instance); +void ITTAPI __itt_model_task_end_2(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_task_begin, (__itt_model_task *task, __itt_model_task_instance *instance, const char *name)) +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, model_task_beginW, (const wchar_t *name)) +ITT_STUBV(ITTAPI, void, model_iteration_taskW, (const wchar_t *name)) +#endif +ITT_STUBV(ITTAPI, void, model_task_beginA, (const char *name)) +ITT_STUBV(ITTAPI, void, model_task_beginAL, (const char *name, size_t taskNameLen)) +ITT_STUBV(ITTAPI, void, model_iteration_taskA, (const char *name)) +ITT_STUBV(ITTAPI, void, model_iteration_taskAL, (const char *name, size_t taskNameLen)) +ITT_STUBV(ITTAPI, void, model_task_end, (__itt_model_task *task, __itt_model_task_instance *instance)) +ITT_STUBV(ITTAPI, void, model_task_end_2, (void)) +#define __itt_model_task_begin ITTNOTIFY_VOID(model_task_begin) +#define __itt_model_task_begin_ptr ITTNOTIFY_NAME(model_task_begin) +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_task_beginW ITTNOTIFY_VOID(model_task_beginW) +#define __itt_model_task_beginW_ptr ITTNOTIFY_NAME(model_task_beginW) +#define __itt_model_iteration_taskW ITTNOTIFY_VOID(model_iteration_taskW) +#define __itt_model_iteration_taskW_ptr ITTNOTIFY_NAME(model_iteration_taskW) +#endif +#define __itt_model_task_beginA ITTNOTIFY_VOID(model_task_beginA) +#define __itt_model_task_beginA_ptr ITTNOTIFY_NAME(model_task_beginA) +#define __itt_model_task_beginAL ITTNOTIFY_VOID(model_task_beginAL) +#define __itt_model_task_beginAL_ptr ITTNOTIFY_NAME(model_task_beginAL) +#define __itt_model_iteration_taskA ITTNOTIFY_VOID(model_iteration_taskA) +#define __itt_model_iteration_taskA_ptr ITTNOTIFY_NAME(model_iteration_taskA) +#define __itt_model_iteration_taskAL ITTNOTIFY_VOID(model_iteration_taskAL) +#define __itt_model_iteration_taskAL_ptr ITTNOTIFY_NAME(model_iteration_taskAL) +#define __itt_model_task_end ITTNOTIFY_VOID(model_task_end) +#define __itt_model_task_end_ptr ITTNOTIFY_NAME(model_task_end) +#define __itt_model_task_end_2 ITTNOTIFY_VOID(model_task_end_2) +#define __itt_model_task_end_2_ptr ITTNOTIFY_NAME(model_task_end_2) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_task_begin(task, instance, name) +#define __itt_model_task_begin_ptr 0 +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_task_beginW(name) +#define __itt_model_task_beginW_ptr 0 +#endif +#define __itt_model_task_beginA(name) +#define __itt_model_task_beginA_ptr 0 +#define __itt_model_task_beginAL(name, siteNameLen) +#define __itt_model_task_beginAL_ptr 0 +#define __itt_model_iteration_taskA(name) +#define __itt_model_iteration_taskA_ptr 0 +#define __itt_model_iteration_taskAL(name, siteNameLen) +#define __itt_model_iteration_taskAL_ptr 0 +#define __itt_model_task_end(task, instance) +#define __itt_model_task_end_ptr 0 +#define __itt_model_task_end_2() +#define __itt_model_task_end_2_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_task_begin_ptr 0 +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_model_task_beginW_ptr 0 +#endif +#define __itt_model_task_beginA_ptr 0 +#define __itt_model_task_beginAL_ptr 0 +#define __itt_model_iteration_taskA_ptr 0 +#define __itt_model_iteration_taskAL_ptr 0 +#define __itt_model_task_end_ptr 0 +#define __itt_model_task_end_2_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_LOCK_ACQUIRE/ANNOTATE_LOCK_RELEASE support + * + * lock_acquire/release model a potential lock for both lockset and + * performance modeling. Each unique address is modeled as a separate + * lock, with invalid addresses being valid lock IDs. Specifically: + * no storage is accessed by the API at the specified address - it is only + * used for lock identification. Lock acquires may be self-nested and are + * unlocked by a corresponding number of releases. + * (These closely correspond to __itt_sync_acquired/__itt_sync_releasing, + * but may not have identical semantics.) + */ +void ITTAPI __itt_model_lock_acquire(void *lock); +void ITTAPI __itt_model_lock_acquire_2(void *lock); +void ITTAPI __itt_model_lock_release(void *lock); +void ITTAPI __itt_model_lock_release_2(void *lock); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_lock_acquire, (void *lock)) +ITT_STUBV(ITTAPI, void, model_lock_acquire_2, (void *lock)) +ITT_STUBV(ITTAPI, void, model_lock_release, (void *lock)) +ITT_STUBV(ITTAPI, void, model_lock_release_2, (void *lock)) +#define __itt_model_lock_acquire ITTNOTIFY_VOID(model_lock_acquire) +#define __itt_model_lock_acquire_ptr ITTNOTIFY_NAME(model_lock_acquire) +#define __itt_model_lock_acquire_2 ITTNOTIFY_VOID(model_lock_acquire_2) +#define __itt_model_lock_acquire_2_ptr ITTNOTIFY_NAME(model_lock_acquire_2) +#define __itt_model_lock_release ITTNOTIFY_VOID(model_lock_release) +#define __itt_model_lock_release_ptr ITTNOTIFY_NAME(model_lock_release) +#define __itt_model_lock_release_2 ITTNOTIFY_VOID(model_lock_release_2) +#define __itt_model_lock_release_2_ptr ITTNOTIFY_NAME(model_lock_release_2) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_lock_acquire(lock) +#define __itt_model_lock_acquire_ptr 0 +#define __itt_model_lock_acquire_2(lock) +#define __itt_model_lock_acquire_2_ptr 0 +#define __itt_model_lock_release(lock) +#define __itt_model_lock_release_ptr 0 +#define __itt_model_lock_release_2(lock) +#define __itt_model_lock_release_2_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_lock_acquire_ptr 0 +#define __itt_model_lock_acquire_2_ptr 0 +#define __itt_model_lock_release_ptr 0 +#define __itt_model_lock_release_2_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_RECORD_ALLOCATION/ANNOTATE_RECORD_DEALLOCATION support + * + * record_allocation/deallocation describe user-defined memory allocator + * behavior, which may be required for correctness modeling to understand + * when storage is not expected to be actually reused across threads. + */ +void ITTAPI __itt_model_record_allocation (void *addr, size_t size); +void ITTAPI __itt_model_record_deallocation(void *addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_record_allocation, (void *addr, size_t size)) +ITT_STUBV(ITTAPI, void, model_record_deallocation, (void *addr)) +#define __itt_model_record_allocation ITTNOTIFY_VOID(model_record_allocation) +#define __itt_model_record_allocation_ptr ITTNOTIFY_NAME(model_record_allocation) +#define __itt_model_record_deallocation ITTNOTIFY_VOID(model_record_deallocation) +#define __itt_model_record_deallocation_ptr ITTNOTIFY_NAME(model_record_deallocation) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_record_allocation(addr, size) +#define __itt_model_record_allocation_ptr 0 +#define __itt_model_record_deallocation(addr) +#define __itt_model_record_deallocation_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_record_allocation_ptr 0 +#define __itt_model_record_deallocation_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_INDUCTION_USES support + * + * Note particular storage is inductive through the end of the current site + */ +void ITTAPI __itt_model_induction_uses(void* addr, size_t size); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_induction_uses, (void *addr, size_t size)) +#define __itt_model_induction_uses ITTNOTIFY_VOID(model_induction_uses) +#define __itt_model_induction_uses_ptr ITTNOTIFY_NAME(model_induction_uses) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_induction_uses(addr, size) +#define __itt_model_induction_uses_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_induction_uses_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_REDUCTION_USES support + * + * Note particular storage is used for reduction through the end + * of the current site + */ +void ITTAPI __itt_model_reduction_uses(void* addr, size_t size); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_reduction_uses, (void *addr, size_t size)) +#define __itt_model_reduction_uses ITTNOTIFY_VOID(model_reduction_uses) +#define __itt_model_reduction_uses_ptr ITTNOTIFY_NAME(model_reduction_uses) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_reduction_uses(addr, size) +#define __itt_model_reduction_uses_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_reduction_uses_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_OBSERVE_USES support + * + * Have correctness modeling record observations about uses of storage + * through the end of the current site + */ +void ITTAPI __itt_model_observe_uses(void* addr, size_t size); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_observe_uses, (void *addr, size_t size)) +#define __itt_model_observe_uses ITTNOTIFY_VOID(model_observe_uses) +#define __itt_model_observe_uses_ptr ITTNOTIFY_NAME(model_observe_uses) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_observe_uses(addr, size) +#define __itt_model_observe_uses_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_observe_uses_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_CLEAR_USES support + * + * Clear the special handling of a piece of storage related to induction, + * reduction or observe_uses + */ +void ITTAPI __itt_model_clear_uses(void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_clear_uses, (void *addr)) +#define __itt_model_clear_uses ITTNOTIFY_VOID(model_clear_uses) +#define __itt_model_clear_uses_ptr ITTNOTIFY_NAME(model_clear_uses) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_clear_uses(addr) +#define __itt_model_clear_uses_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_clear_uses_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief ANNOTATE_DISABLE_*_PUSH/ANNOTATE_DISABLE_*_POP support + * + * disable_push/disable_pop push and pop disabling based on a parameter. + * Disabling observations stops processing of memory references during + * correctness modeling, and all annotations that occur in the disabled + * region. This allows description of code that is expected to be handled + * specially during conversion to parallelism or that is not recognized + * by tools (e.g. some kinds of synchronization operations.) + * This mechanism causes all annotations in the disabled region, other + * than disable_push and disable_pop, to be ignored. (For example, this + * might validly be used to disable an entire parallel site and the contained + * tasks and locking in it for data collection purposes.) + * The disable for collection is a more expensive operation, but reduces + * collector overhead significantly. This applies to BOTH correctness data + * collection and performance data collection. For example, a site + * containing a task might only enable data collection for the first 10 + * iterations. Both performance and correctness data should reflect this, + * and the program should run as close to full speed as possible when + * collection is disabled. + */ +void ITTAPI __itt_model_disable_push(__itt_model_disable x); +void ITTAPI __itt_model_disable_pop(void); +void ITTAPI __itt_model_aggregate_task(size_t x); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, model_disable_push, (__itt_model_disable x)) +ITT_STUBV(ITTAPI, void, model_disable_pop, (void)) +ITT_STUBV(ITTAPI, void, model_aggregate_task, (size_t x)) +#define __itt_model_disable_push ITTNOTIFY_VOID(model_disable_push) +#define __itt_model_disable_push_ptr ITTNOTIFY_NAME(model_disable_push) +#define __itt_model_disable_pop ITTNOTIFY_VOID(model_disable_pop) +#define __itt_model_disable_pop_ptr ITTNOTIFY_NAME(model_disable_pop) +#define __itt_model_aggregate_task ITTNOTIFY_VOID(model_aggregate_task) +#define __itt_model_aggregate_task_ptr ITTNOTIFY_NAME(model_aggregate_task) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_model_disable_push(x) +#define __itt_model_disable_push_ptr 0 +#define __itt_model_disable_pop() +#define __itt_model_disable_pop_ptr 0 +#define __itt_model_aggregate_task(x) +#define __itt_model_aggregate_task_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_model_disable_push_ptr 0 +#define __itt_model_disable_pop_ptr 0 +#define __itt_model_aggregate_task_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} model group */ + +/** + * @defgroup heap Heap + * @ingroup public + * Heap group + * @{ + */ + +typedef void* __itt_heap_function; + +/** + * @brief Create an identification for heap function + * @return non-zero identifier or NULL + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_heap_function ITTAPI __itt_heap_function_createA(const char* name, const char* domain); +__itt_heap_function ITTAPI __itt_heap_function_createW(const wchar_t* name, const wchar_t* domain); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_heap_function_create __itt_heap_function_createW +# define __itt_heap_function_create_ptr __itt_heap_function_createW_ptr +#else +# define __itt_heap_function_create __itt_heap_function_createA +# define __itt_heap_function_create_ptr __itt_heap_function_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_heap_function ITTAPI __itt_heap_function_create(const char* name, const char* domain); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_heap_function, heap_function_createA, (const char* name, const char* domain)) +ITT_STUB(ITTAPI, __itt_heap_function, heap_function_createW, (const wchar_t* name, const wchar_t* domain)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_heap_function, heap_function_create, (const char* name, const char* domain)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_heap_function_createA ITTNOTIFY_DATA(heap_function_createA) +#define __itt_heap_function_createA_ptr ITTNOTIFY_NAME(heap_function_createA) +#define __itt_heap_function_createW ITTNOTIFY_DATA(heap_function_createW) +#define __itt_heap_function_createW_ptr ITTNOTIFY_NAME(heap_function_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_heap_function_create ITTNOTIFY_DATA(heap_function_create) +#define __itt_heap_function_create_ptr ITTNOTIFY_NAME(heap_function_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_heap_function_createA(name, domain) (__itt_heap_function)0 +#define __itt_heap_function_createA_ptr 0 +#define __itt_heap_function_createW(name, domain) (__itt_heap_function)0 +#define __itt_heap_function_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_heap_function_create(name, domain) (__itt_heap_function)0 +#define __itt_heap_function_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_heap_function_createA_ptr 0 +#define __itt_heap_function_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_heap_function_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an allocation begin occurrence. + */ +void ITTAPI __itt_heap_allocate_begin(__itt_heap_function h, size_t size, int initialized); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_allocate_begin, (__itt_heap_function h, size_t size, int initialized)) +#define __itt_heap_allocate_begin ITTNOTIFY_VOID(heap_allocate_begin) +#define __itt_heap_allocate_begin_ptr ITTNOTIFY_NAME(heap_allocate_begin) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_allocate_begin(h, size, initialized) +#define __itt_heap_allocate_begin_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_allocate_begin_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an allocation end occurrence. + */ +void ITTAPI __itt_heap_allocate_end(__itt_heap_function h, void** addr, size_t size, int initialized); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_allocate_end, (__itt_heap_function h, void** addr, size_t size, int initialized)) +#define __itt_heap_allocate_end ITTNOTIFY_VOID(heap_allocate_end) +#define __itt_heap_allocate_end_ptr ITTNOTIFY_NAME(heap_allocate_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_allocate_end(h, addr, size, initialized) +#define __itt_heap_allocate_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_allocate_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an free begin occurrence. + */ +void ITTAPI __itt_heap_free_begin(__itt_heap_function h, void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_free_begin, (__itt_heap_function h, void* addr)) +#define __itt_heap_free_begin ITTNOTIFY_VOID(heap_free_begin) +#define __itt_heap_free_begin_ptr ITTNOTIFY_NAME(heap_free_begin) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_free_begin(h, addr) +#define __itt_heap_free_begin_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_free_begin_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an free end occurrence. + */ +void ITTAPI __itt_heap_free_end(__itt_heap_function h, void* addr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_free_end, (__itt_heap_function h, void* addr)) +#define __itt_heap_free_end ITTNOTIFY_VOID(heap_free_end) +#define __itt_heap_free_end_ptr ITTNOTIFY_NAME(heap_free_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_free_end(h, addr) +#define __itt_heap_free_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_free_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an reallocation begin occurrence. + */ +void ITTAPI __itt_heap_reallocate_begin(__itt_heap_function h, void* addr, size_t new_size, int initialized); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_reallocate_begin, (__itt_heap_function h, void* addr, size_t new_size, int initialized)) +#define __itt_heap_reallocate_begin ITTNOTIFY_VOID(heap_reallocate_begin) +#define __itt_heap_reallocate_begin_ptr ITTNOTIFY_NAME(heap_reallocate_begin) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_reallocate_begin(h, addr, new_size, initialized) +#define __itt_heap_reallocate_begin_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_reallocate_begin_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an reallocation end occurrence. + */ +void ITTAPI __itt_heap_reallocate_end(__itt_heap_function h, void* addr, void** new_addr, size_t new_size, int initialized); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_reallocate_end, (__itt_heap_function h, void* addr, void** new_addr, size_t new_size, int initialized)) +#define __itt_heap_reallocate_end ITTNOTIFY_VOID(heap_reallocate_end) +#define __itt_heap_reallocate_end_ptr ITTNOTIFY_NAME(heap_reallocate_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_reallocate_end(h, addr, new_addr, new_size, initialized) +#define __itt_heap_reallocate_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_reallocate_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @brief internal access begin */ +void ITTAPI __itt_heap_internal_access_begin(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_internal_access_begin, (void)) +#define __itt_heap_internal_access_begin ITTNOTIFY_VOID(heap_internal_access_begin) +#define __itt_heap_internal_access_begin_ptr ITTNOTIFY_NAME(heap_internal_access_begin) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_internal_access_begin() +#define __itt_heap_internal_access_begin_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_internal_access_begin_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @brief internal access end */ +void ITTAPI __itt_heap_internal_access_end(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_internal_access_end, (void)) +#define __itt_heap_internal_access_end ITTNOTIFY_VOID(heap_internal_access_end) +#define __itt_heap_internal_access_end_ptr ITTNOTIFY_NAME(heap_internal_access_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_internal_access_end() +#define __itt_heap_internal_access_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_internal_access_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @brief record memory growth begin */ +void ITTAPI __itt_heap_record_memory_growth_begin(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_record_memory_growth_begin, (void)) +#define __itt_heap_record_memory_growth_begin ITTNOTIFY_VOID(heap_record_memory_growth_begin) +#define __itt_heap_record_memory_growth_begin_ptr ITTNOTIFY_NAME(heap_record_memory_growth_begin) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_record_memory_growth_begin() +#define __itt_heap_record_memory_growth_begin_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_record_memory_growth_begin_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @brief record memory growth end */ +void ITTAPI __itt_heap_record_memory_growth_end(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_record_memory_growth_end, (void)) +#define __itt_heap_record_memory_growth_end ITTNOTIFY_VOID(heap_record_memory_growth_end) +#define __itt_heap_record_memory_growth_end_ptr ITTNOTIFY_NAME(heap_record_memory_growth_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_record_memory_growth_end() +#define __itt_heap_record_memory_growth_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_record_memory_growth_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Specify the type of heap detection/reporting to modify. + */ +/** + * @hideinitializer + * @brief Report on memory leaks. + */ +#define __itt_heap_leaks 0x00000001 + +/** + * @hideinitializer + * @brief Report on memory growth. + */ +#define __itt_heap_growth 0x00000002 + + +/** @brief heap reset detection */ +void ITTAPI __itt_heap_reset_detection(unsigned int reset_mask); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_reset_detection, (unsigned int reset_mask)) +#define __itt_heap_reset_detection ITTNOTIFY_VOID(heap_reset_detection) +#define __itt_heap_reset_detection_ptr ITTNOTIFY_NAME(heap_reset_detection) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_reset_detection() +#define __itt_heap_reset_detection_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_reset_detection_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @brief report */ +void ITTAPI __itt_heap_record(unsigned int record_mask); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, heap_record, (unsigned int record_mask)) +#define __itt_heap_record ITTNOTIFY_VOID(heap_record) +#define __itt_heap_record_ptr ITTNOTIFY_NAME(heap_record) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_heap_record() +#define __itt_heap_record_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_heap_record_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @} heap group */ +/** @endcond */ +/* ========================================================================== */ + +/** + * @defgroup domains Domains + * @ingroup public + * Domains group + * @{ + */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_domain +{ + volatile int flags; /*!< Zero if disabled, non-zero if enabled. The meaning of different non-zero values is reserved to the runtime */ + const char* nameA; /*!< Copy of original name in ASCII. */ +#if defined(UNICODE) || defined(_UNICODE) + const wchar_t* nameW; /*!< Copy of original name in UNICODE. */ +#else /* UNICODE || _UNICODE */ + void* nameW; +#endif /* UNICODE || _UNICODE */ + int extra1; /*!< Reserved to the runtime */ + void* extra2; /*!< Reserved to the runtime */ + struct ___itt_domain* next; +} __itt_domain; + +#pragma pack(pop) +/** @endcond */ + +/** + * @ingroup domains + * @brief Create a domain. + * Create domain using some domain name: the URI naming style is recommended. + * Because the set of domains is expected to be static over the application's + * execution time, there is no mechanism to destroy a domain. + * Any domain can be accessed by any thread in the process, regardless of + * which thread created the domain. This call is thread-safe. + * @param[in] name name of domain + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_domain* ITTAPI __itt_domain_createA(const char *name); +__itt_domain* ITTAPI __itt_domain_createW(const wchar_t *name); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_domain_create __itt_domain_createW +# define __itt_domain_create_ptr __itt_domain_createW_ptr +#else /* UNICODE */ +# define __itt_domain_create __itt_domain_createA +# define __itt_domain_create_ptr __itt_domain_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_domain* ITTAPI __itt_domain_create(const char *name); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_domain*, domain_createA, (const char *name)) +ITT_STUB(ITTAPI, __itt_domain*, domain_createW, (const wchar_t *name)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_domain*, domain_create, (const char *name)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_domain_createA ITTNOTIFY_DATA(domain_createA) +#define __itt_domain_createA_ptr ITTNOTIFY_NAME(domain_createA) +#define __itt_domain_createW ITTNOTIFY_DATA(domain_createW) +#define __itt_domain_createW_ptr ITTNOTIFY_NAME(domain_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_domain_create ITTNOTIFY_DATA(domain_create) +#define __itt_domain_create_ptr ITTNOTIFY_NAME(domain_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_domain_createA(name) (__itt_domain*)0 +#define __itt_domain_createA_ptr 0 +#define __itt_domain_createW(name) (__itt_domain*)0 +#define __itt_domain_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_domain_create(name) (__itt_domain*)0 +#define __itt_domain_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_domain_createA_ptr 0 +#define __itt_domain_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_domain_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} domains group */ + +/** + * @defgroup ids IDs + * @ingroup public + * IDs group + * @{ + */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_id +{ + unsigned long long d1, d2, d3; +} __itt_id; + +#pragma pack(pop) +/** @endcond */ + +const __itt_id __itt_null = { 0, 0, 0 }; + +/** + * @ingroup ids + * @brief A convenience function is provided to create an ID without domain control. + * @brief This is a convenience function to initialize an __itt_id structure. This function + * does not affect the collector runtime in any way. After you make the ID with this + * function, you still must create it with the __itt_id_create function before using the ID + * to identify a named entity. + * @param[in] addr The address of object; high QWORD of the ID value. + * @param[in] extra The extra data to unique identify object; low QWORD of the ID value. + */ + +ITT_INLINE __itt_id ITTAPI __itt_id_make(void* addr, unsigned long long extra) ITT_INLINE_ATTRIBUTE; +ITT_INLINE __itt_id ITTAPI __itt_id_make(void* addr, unsigned long long extra) +{ + __itt_id id = __itt_null; + id.d1 = (unsigned long long)((uintptr_t)addr); + id.d2 = (unsigned long long)extra; + id.d3 = (unsigned long long)0; /* Reserved. Must be zero */ + return id; +} + +/** + * @ingroup ids + * @brief Create an instance of identifier. + * This establishes the beginning of the lifetime of an instance of + * the given ID in the trace. Once this lifetime starts, the ID + * can be used to tag named entity instances in calls such as + * __itt_task_begin, and to specify relationships among + * identified named entity instances, using the \ref relations APIs. + * Instance IDs are not domain specific! + * @param[in] domain The domain controlling the execution of this call. + * @param[in] id The ID to create. + */ +void ITTAPI __itt_id_create(const __itt_domain *domain, __itt_id id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, id_create, (const __itt_domain *domain, __itt_id id)) +#define __itt_id_create(d,x) ITTNOTIFY_VOID_D1(id_create,d,x) +#define __itt_id_create_ptr ITTNOTIFY_NAME(id_create) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_id_create(domain,id) +#define __itt_id_create_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_id_create_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup ids + * @brief Destroy an instance of identifier. + * This ends the lifetime of the current instance of the given ID value in the trace. + * Any relationships that are established after this lifetime ends are invalid. + * This call must be performed before the given ID value can be reused for a different + * named entity instance. + * @param[in] domain The domain controlling the execution of this call. + * @param[in] id The ID to destroy. + */ +void ITTAPI __itt_id_destroy(const __itt_domain *domain, __itt_id id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, id_destroy, (const __itt_domain *domain, __itt_id id)) +#define __itt_id_destroy(d,x) ITTNOTIFY_VOID_D1(id_destroy,d,x) +#define __itt_id_destroy_ptr ITTNOTIFY_NAME(id_destroy) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_id_destroy(domain,id) +#define __itt_id_destroy_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_id_destroy_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} ids group */ + +/** + * @defgroup handless String Handles + * @ingroup public + * String Handles group + * @{ + */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_string_handle +{ + const char* strA; /*!< Copy of original string in ASCII. */ +#if defined(UNICODE) || defined(_UNICODE) + const wchar_t* strW; /*!< Copy of original string in UNICODE. */ +#else /* UNICODE || _UNICODE */ + void* strW; +#endif /* UNICODE || _UNICODE */ + int extra1; /*!< Reserved. Must be zero */ + void* extra2; /*!< Reserved. Must be zero */ + struct ___itt_string_handle* next; +} __itt_string_handle; + +#pragma pack(pop) +/** @endcond */ + +/** + * @ingroup handles + * @brief Create a string handle. + * Create and return handle value that can be associated with a string. + * Consecutive calls to __itt_string_handle_create with the same name + * return the same value. Because the set of string handles is expected to remain + * static during the application's execution time, there is no mechanism to destroy a string handle. + * Any string handle can be accessed by any thread in the process, regardless of which thread created + * the string handle. This call is thread-safe. + * @param[in] name The input string + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_string_handle* ITTAPI __itt_string_handle_createA(const char *name); +__itt_string_handle* ITTAPI __itt_string_handle_createW(const wchar_t *name); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_string_handle_create __itt_string_handle_createW +# define __itt_string_handle_create_ptr __itt_string_handle_createW_ptr +#else /* UNICODE */ +# define __itt_string_handle_create __itt_string_handle_createA +# define __itt_string_handle_create_ptr __itt_string_handle_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_string_handle* ITTAPI __itt_string_handle_create(const char *name); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_string_handle*, string_handle_createA, (const char *name)) +ITT_STUB(ITTAPI, __itt_string_handle*, string_handle_createW, (const wchar_t *name)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_string_handle*, string_handle_create, (const char *name)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_string_handle_createA ITTNOTIFY_DATA(string_handle_createA) +#define __itt_string_handle_createA_ptr ITTNOTIFY_NAME(string_handle_createA) +#define __itt_string_handle_createW ITTNOTIFY_DATA(string_handle_createW) +#define __itt_string_handle_createW_ptr ITTNOTIFY_NAME(string_handle_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_string_handle_create ITTNOTIFY_DATA(string_handle_create) +#define __itt_string_handle_create_ptr ITTNOTIFY_NAME(string_handle_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_string_handle_createA(name) (__itt_string_handle*)0 +#define __itt_string_handle_createA_ptr 0 +#define __itt_string_handle_createW(name) (__itt_string_handle*)0 +#define __itt_string_handle_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_string_handle_create(name) (__itt_string_handle*)0 +#define __itt_string_handle_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_string_handle_createA_ptr 0 +#define __itt_string_handle_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_string_handle_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} handles group */ + +/** @cond exclude_from_documentation */ +typedef unsigned long long __itt_timestamp; +/** @endcond */ + +#define __itt_timestamp_none ((__itt_timestamp)-1LL) + +/** @cond exclude_from_gpa_documentation */ + +/** + * @ingroup timestamps + * @brief Return timestamp corresponding to the current moment. + * This returns the timestamp in the format that is the most relevant for the current + * host or platform (RDTSC, QPC, and others). You can use the "<" operator to + * compare __itt_timestamp values. + */ +__itt_timestamp ITTAPI __itt_get_timestamp(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, __itt_timestamp, get_timestamp, (void)) +#define __itt_get_timestamp ITTNOTIFY_DATA(get_timestamp) +#define __itt_get_timestamp_ptr ITTNOTIFY_NAME(get_timestamp) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_get_timestamp() +#define __itt_get_timestamp_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_get_timestamp_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} timestamps */ +/** @endcond */ + +/** @cond exclude_from_gpa_documentation */ + +/** + * @defgroup regions Regions + * @ingroup public + * Regions group + * @{ + */ +/** + * @ingroup regions + * @brief Begin of region instance. + * Successive calls to __itt_region_begin with the same ID are ignored + * until a call to __itt_region_end with the same ID + * @param[in] domain The domain for this region instance + * @param[in] id The instance ID for this region instance. Must not be __itt_null + * @param[in] parentid The instance ID for the parent of this region instance, or __itt_null + * @param[in] name The name of this region + */ +void ITTAPI __itt_region_begin(const __itt_domain *domain, __itt_id id, __itt_id parentid, __itt_string_handle *name); + +/** + * @ingroup regions + * @brief End of region instance. + * The first call to __itt_region_end with a given ID ends the + * region. Successive calls with the same ID are ignored, as are + * calls that do not have a matching __itt_region_begin call. + * @param[in] domain The domain for this region instance + * @param[in] id The instance ID for this region instance + */ +void ITTAPI __itt_region_end(const __itt_domain *domain, __itt_id id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, region_begin, (const __itt_domain *domain, __itt_id id, __itt_id parentid, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, region_end, (const __itt_domain *domain, __itt_id id)) +#define __itt_region_begin(d,x,y,z) ITTNOTIFY_VOID_D3(region_begin,d,x,y,z) +#define __itt_region_begin_ptr ITTNOTIFY_NAME(region_begin) +#define __itt_region_end(d,x) ITTNOTIFY_VOID_D1(region_end,d,x) +#define __itt_region_end_ptr ITTNOTIFY_NAME(region_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_region_begin(d,x,y,z) +#define __itt_region_begin_ptr 0 +#define __itt_region_end(d,x) +#define __itt_region_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_region_begin_ptr 0 +#define __itt_region_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} regions group */ + +/** + * @defgroup frames Frames + * @ingroup public + * Frames are similar to regions, but are intended to be easier to use and to implement. + * In particular: + * - Frames always represent periods of elapsed time + * - By default, frames have no nesting relationships + * @{ + */ + +/** + * @ingroup frames + * @brief Begin a frame instance. + * Successive calls to __itt_frame_begin with the + * same ID are ignored until a call to __itt_frame_end with the same ID. + * @param[in] domain The domain for this frame instance + * @param[in] id The instance ID for this frame instance or NULL + */ +void ITTAPI __itt_frame_begin_v3(const __itt_domain *domain, __itt_id *id); + +/** + * @ingroup frames + * @brief End a frame instance. + * The first call to __itt_frame_end with a given ID + * ends the frame. Successive calls with the same ID are ignored, as are + * calls that do not have a matching __itt_frame_begin call. + * @param[in] domain The domain for this frame instance + * @param[in] id The instance ID for this frame instance or NULL for current + */ +void ITTAPI __itt_frame_end_v3(const __itt_domain *domain, __itt_id *id); + +/** + * @ingroup frames + * @brief Submits a frame instance. + * Successive calls to __itt_frame_begin or __itt_frame_submit with the + * same ID are ignored until a call to __itt_frame_end or __itt_frame_submit + * with the same ID. + * Passing special __itt_timestamp_none value as "end" argument means + * take the current timestamp as the end timestamp. + * @param[in] domain The domain for this frame instance + * @param[in] id The instance ID for this frame instance or NULL + * @param[in] begin Timestamp of the beginning of the frame + * @param[in] end Timestamp of the end of the frame + */ +void ITTAPI __itt_frame_submit_v3(const __itt_domain *domain, __itt_id *id, + __itt_timestamp begin, __itt_timestamp end); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, frame_begin_v3, (const __itt_domain *domain, __itt_id *id)) +ITT_STUBV(ITTAPI, void, frame_end_v3, (const __itt_domain *domain, __itt_id *id)) +ITT_STUBV(ITTAPI, void, frame_submit_v3, (const __itt_domain *domain, __itt_id *id, __itt_timestamp begin, __itt_timestamp end)) +#define __itt_frame_begin_v3(d,x) ITTNOTIFY_VOID_D1(frame_begin_v3,d,x) +#define __itt_frame_begin_v3_ptr ITTNOTIFY_NAME(frame_begin_v3) +#define __itt_frame_end_v3(d,x) ITTNOTIFY_VOID_D1(frame_end_v3,d,x) +#define __itt_frame_end_v3_ptr ITTNOTIFY_NAME(frame_end_v3) +#define __itt_frame_submit_v3(d,x,b,e) ITTNOTIFY_VOID_D3(frame_submit_v3,d,x,b,e) +#define __itt_frame_submit_v3_ptr ITTNOTIFY_NAME(frame_submit_v3) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_frame_begin_v3(domain,id) +#define __itt_frame_begin_v3_ptr 0 +#define __itt_frame_end_v3(domain,id) +#define __itt_frame_end_v3_ptr 0 +#define __itt_frame_submit_v3(domain,id,begin,end) +#define __itt_frame_submit_v3_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_frame_begin_v3_ptr 0 +#define __itt_frame_end_v3_ptr 0 +#define __itt_frame_submit_v3_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} frames group */ +/** @endcond */ + +/** + * @defgroup taskgroup Task Group + * @ingroup public + * Task Group + * @{ + */ +/** + * @ingroup task_groups + * @brief Denotes a task_group instance. + * Successive calls to __itt_task_group with the same ID are ignored. + * @param[in] domain The domain for this task_group instance + * @param[in] id The instance ID for this task_group instance. Must not be __itt_null. + * @param[in] parentid The instance ID for the parent of this task_group instance, or __itt_null. + * @param[in] name The name of this task_group + */ +void ITTAPI __itt_task_group(const __itt_domain *domain, __itt_id id, __itt_id parentid, __itt_string_handle *name); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, task_group, (const __itt_domain *domain, __itt_id id, __itt_id parentid, __itt_string_handle *name)) +#define __itt_task_group(d,x,y,z) ITTNOTIFY_VOID_D3(task_group,d,x,y,z) +#define __itt_task_group_ptr ITTNOTIFY_NAME(task_group) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_task_group(d,x,y,z) +#define __itt_task_group_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_task_group_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} taskgroup group */ + +/** + * @defgroup tasks Tasks + * @ingroup public + * A task instance represents a piece of work performed by a particular + * thread for a period of time. A call to __itt_task_begin creates a + * task instance. This becomes the current instance for that task on that + * thread. A following call to __itt_task_end on the same thread ends the + * instance. There may be multiple simultaneous instances of tasks with the + * same name on different threads. If an ID is specified, the task instance + * receives that ID. Nested tasks are allowed. + * + * Note: The task is defined by the bracketing of __itt_task_begin and + * __itt_task_end on the same thread. If some scheduling mechanism causes + * task switching (the thread executes a different user task) or task + * switching (the user task switches to a different thread) then this breaks + * the notion of current instance. Additional API calls are required to + * deal with that possibility. + * @{ + */ + +/** + * @ingroup tasks + * @brief Begin a task instance. + * @param[in] domain The domain for this task + * @param[in] taskid The instance ID for this task instance, or __itt_null + * @param[in] parentid The parent instance to which this task instance belongs, or __itt_null + * @param[in] name The name of this task + */ +void ITTAPI __itt_task_begin(const __itt_domain *domain, __itt_id taskid, __itt_id parentid, __itt_string_handle *name); + +/** + * @ingroup tasks + * @brief Begin a task instance. + * @param[in] domain The domain for this task + * @param[in] taskid The identifier for this task instance (may be 0) + * @param[in] parentid The parent of this task (may be 0) + * @param[in] fn The pointer to the function you are tracing + */ +void ITTAPI __itt_task_begin_fn(const __itt_domain *domain, __itt_id taskid, __itt_id parentid, void* fn); + +/** + * @ingroup tasks + * @brief End the current task instance. + * @param[in] domain The domain for this task + */ +void ITTAPI __itt_task_end(const __itt_domain *domain); + +/** + * @ingroup tasks + * @brief Begin an overlapped task instance. + * @param[in] domain The domain for this task. + * @param[in] taskid The identifier for this task instance, *cannot* be __itt_null. + * @param[in] parentid The parent of this task, or __itt_null. + * @param[in] name The name of this task. + */ +void ITTAPI __itt_task_begin_overlapped(const __itt_domain* domain, __itt_id taskid, __itt_id parentid, __itt_string_handle* name); + +/** + * @ingroup tasks + * @brief End an overlapped task instance. + * @param[in] domain The domain for this task + * @param[in] taskid Explicit ID of finished task + */ +void ITTAPI __itt_task_end_overlapped(const __itt_domain *domain, __itt_id taskid); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, task_begin, (const __itt_domain *domain, __itt_id id, __itt_id parentid, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, task_begin_fn, (const __itt_domain *domain, __itt_id id, __itt_id parentid, void* fn)) +ITT_STUBV(ITTAPI, void, task_end, (const __itt_domain *domain)) +ITT_STUBV(ITTAPI, void, task_begin_overlapped, (const __itt_domain *domain, __itt_id taskid, __itt_id parentid, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, task_end_overlapped, (const __itt_domain *domain, __itt_id taskid)) +#define __itt_task_begin(d,x,y,z) ITTNOTIFY_VOID_D3(task_begin,d,x,y,z) +#define __itt_task_begin_ptr ITTNOTIFY_NAME(task_begin) +#define __itt_task_begin_fn(d,x,y,z) ITTNOTIFY_VOID_D3(task_begin_fn,d,x,y,z) +#define __itt_task_begin_fn_ptr ITTNOTIFY_NAME(task_begin_fn) +#define __itt_task_end(d) ITTNOTIFY_VOID_D0(task_end,d) +#define __itt_task_end_ptr ITTNOTIFY_NAME(task_end) +#define __itt_task_begin_overlapped(d,x,y,z) ITTNOTIFY_VOID_D3(task_begin_overlapped,d,x,y,z) +#define __itt_task_begin_overlapped_ptr ITTNOTIFY_NAME(task_begin_overlapped) +#define __itt_task_end_overlapped(d,x) ITTNOTIFY_VOID_D1(task_end_overlapped,d,x) +#define __itt_task_end_overlapped_ptr ITTNOTIFY_NAME(task_end_overlapped) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_task_begin(domain,id,parentid,name) +#define __itt_task_begin_ptr 0 +#define __itt_task_begin_fn(domain,id,parentid,fn) +#define __itt_task_begin_fn_ptr 0 +#define __itt_task_end(domain) +#define __itt_task_end_ptr 0 +#define __itt_task_begin_overlapped(domain,taskid,parentid,name) +#define __itt_task_begin_overlapped_ptr 0 +#define __itt_task_end_overlapped(domain,taskid) +#define __itt_task_end_overlapped_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_task_begin_ptr 0 +#define __itt_task_begin_fn_ptr 0 +#define __itt_task_end_ptr 0 +#define __itt_task_begin_overlapped_ptr 0 +#define __itt_task_end_overlapped_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} tasks group */ + + +/** + * @defgroup markers Markers + * Markers represent a single discreet event in time. Markers have a scope, + * described by an enumerated type __itt_scope. Markers are created by + * the API call __itt_marker. A marker instance can be given an ID for use in + * adding metadata. + * @{ + */ + +/** + * @brief Describes the scope of an event object in the trace. + */ +typedef enum +{ + __itt_scope_unknown = 0, + __itt_scope_global, + __itt_scope_track_group, + __itt_scope_track, + __itt_scope_task, + __itt_scope_marker +} __itt_scope; + +/** @cond exclude_from_documentation */ +#define __itt_marker_scope_unknown __itt_scope_unknown +#define __itt_marker_scope_global __itt_scope_global +#define __itt_marker_scope_process __itt_scope_track_group +#define __itt_marker_scope_thread __itt_scope_track +#define __itt_marker_scope_task __itt_scope_task +/** @endcond */ + +/** + * @ingroup markers + * @brief Create a marker instance + * @param[in] domain The domain for this marker + * @param[in] id The instance ID for this marker or __itt_null + * @param[in] name The name for this marker + * @param[in] scope The scope for this marker + */ +void ITTAPI __itt_marker(const __itt_domain *domain, __itt_id id, __itt_string_handle *name, __itt_scope scope); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, marker, (const __itt_domain *domain, __itt_id id, __itt_string_handle *name, __itt_scope scope)) +#define __itt_marker(d,x,y,z) ITTNOTIFY_VOID_D3(marker,d,x,y,z) +#define __itt_marker_ptr ITTNOTIFY_NAME(marker) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_marker(domain,id,name,scope) +#define __itt_marker_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_marker_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} markers group */ + +/** + * @defgroup metadata Metadata + * The metadata API is used to attach extra information to named + * entities. Metadata can be attached to an identified named entity by ID, + * or to the current entity (which is always a task). + * + * Conceptually metadata has a type (what kind of metadata), a key (the + * name of the metadata), and a value (the actual data). The encoding of + * the value depends on the type of the metadata. + * + * The type of metadata is specified by an enumerated type __itt_metdata_type. + * @{ + */ + +/** + * @ingroup parameters + * @brief describes the type of metadata + */ +typedef enum { + __itt_metadata_unknown = 0, + __itt_metadata_u64, /**< Unsigned 64-bit integer */ + __itt_metadata_s64, /**< Signed 64-bit integer */ + __itt_metadata_u32, /**< Unsigned 32-bit integer */ + __itt_metadata_s32, /**< Signed 32-bit integer */ + __itt_metadata_u16, /**< Unsigned 16-bit integer */ + __itt_metadata_s16, /**< Signed 16-bit integer */ + __itt_metadata_float, /**< Signed 32-bit floating-point */ + __itt_metadata_double /**< SIgned 64-bit floating-point */ +} __itt_metadata_type; + +/** + * @ingroup parameters + * @brief Add metadata to an instance of a named entity. + * @param[in] domain The domain controlling the call + * @param[in] id The identifier of the instance to which the metadata is to be added, or __itt_null to add to the current task + * @param[in] key The name of the metadata + * @param[in] type The type of the metadata + * @param[in] count The number of elements of the given type. If count == 0, no metadata will be added. + * @param[in] data The metadata itself +*/ +void ITTAPI __itt_metadata_add(const __itt_domain *domain, __itt_id id, __itt_string_handle *key, __itt_metadata_type type, size_t count, void *data); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, metadata_add, (const __itt_domain *domain, __itt_id id, __itt_string_handle *key, __itt_metadata_type type, size_t count, void *data)) +#define __itt_metadata_add(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(metadata_add,d,x,y,z,a,b) +#define __itt_metadata_add_ptr ITTNOTIFY_NAME(metadata_add) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_metadata_add(d,x,y,z,a,b) +#define __itt_metadata_add_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_metadata_add_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup parameters + * @brief Add string metadata to an instance of a named entity. + * @param[in] domain The domain controlling the call + * @param[in] id The identifier of the instance to which the metadata is to be added, or __itt_null to add to the current task + * @param[in] key The name of the metadata + * @param[in] data The metadata itself + * @param[in] length The number of characters in the string, or -1 if the length is unknown but the string is null-terminated +*/ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_metadata_str_addA(const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const char *data, size_t length); +void ITTAPI __itt_metadata_str_addW(const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const wchar_t *data, size_t length); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_metadata_str_add __itt_metadata_str_addW +# define __itt_metadata_str_add_ptr __itt_metadata_str_addW_ptr +#else /* UNICODE */ +# define __itt_metadata_str_add __itt_metadata_str_addA +# define __itt_metadata_str_add_ptr __itt_metadata_str_addA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_metadata_str_add(const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const char *data, size_t length); +#endif + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, metadata_str_addA, (const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const char *data, size_t length)) +ITT_STUBV(ITTAPI, void, metadata_str_addW, (const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const wchar_t *data, size_t length)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUBV(ITTAPI, void, metadata_str_add, (const __itt_domain *domain, __itt_id id, __itt_string_handle *key, const char *data, size_t length)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_addA(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_addA,d,x,y,z,a) +#define __itt_metadata_str_addA_ptr ITTNOTIFY_NAME(metadata_str_addA) +#define __itt_metadata_str_addW(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_addW,d,x,y,z,a) +#define __itt_metadata_str_addW_ptr ITTNOTIFY_NAME(metadata_str_addW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_add,d,x,y,z,a) +#define __itt_metadata_str_add_ptr ITTNOTIFY_NAME(metadata_str_add) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_addA(d,x,y,z,a) +#define __itt_metadata_str_addA_ptr 0 +#define __itt_metadata_str_addW(d,x,y,z,a) +#define __itt_metadata_str_addW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add(d,x,y,z,a) +#define __itt_metadata_str_add_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_addA_ptr 0 +#define __itt_metadata_str_addW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup parameters + * @brief Add metadata to an instance of a named entity. + * @param[in] domain The domain controlling the call + * @param[in] scope The scope of the instance to which the metadata is to be added + + * @param[in] id The identifier of the instance to which the metadata is to be added, or __itt_null to add to the current task + + * @param[in] key The name of the metadata + * @param[in] type The type of the metadata + * @param[in] count The number of elements of the given type. If count == 0, no metadata will be added. + * @param[in] data The metadata itself +*/ +void ITTAPI __itt_metadata_add_with_scope(const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, __itt_metadata_type type, size_t count, void *data); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, metadata_add_with_scope, (const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, __itt_metadata_type type, size_t count, void *data)) +#define __itt_metadata_add_with_scope(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(metadata_add_with_scope,d,x,y,z,a,b) +#define __itt_metadata_add_with_scope_ptr ITTNOTIFY_NAME(metadata_add_with_scope) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_metadata_add_with_scope(d,x,y,z,a,b) +#define __itt_metadata_add_with_scope_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_metadata_add_with_scope_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup parameters + * @brief Add string metadata to an instance of a named entity. + * @param[in] domain The domain controlling the call + * @param[in] scope The scope of the instance to which the metadata is to be added + + * @param[in] id The identifier of the instance to which the metadata is to be added, or __itt_null to add to the current task + + * @param[in] key The name of the metadata + * @param[in] data The metadata itself + * @param[in] length The number of characters in the string, or -1 if the length is unknown but the string is null-terminated +*/ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_metadata_str_add_with_scopeA(const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const char *data, size_t length); +void ITTAPI __itt_metadata_str_add_with_scopeW(const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const wchar_t *data, size_t length); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_metadata_str_add_with_scope __itt_metadata_str_add_with_scopeW +# define __itt_metadata_str_add_with_scope_ptr __itt_metadata_str_add_with_scopeW_ptr +#else /* UNICODE */ +# define __itt_metadata_str_add_with_scope __itt_metadata_str_add_with_scopeA +# define __itt_metadata_str_add_with_scope_ptr __itt_metadata_str_add_with_scopeA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_metadata_str_add_with_scope(const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const char *data, size_t length); +#endif + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUBV(ITTAPI, void, metadata_str_add_with_scopeA, (const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const char *data, size_t length)) +ITT_STUBV(ITTAPI, void, metadata_str_add_with_scopeW, (const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const wchar_t *data, size_t length)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUBV(ITTAPI, void, metadata_str_add_with_scope, (const __itt_domain *domain, __itt_scope scope, __itt_string_handle *key, const char *data, size_t length)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_add_with_scopeA(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_add_with_scopeA,d,x,y,z,a) +#define __itt_metadata_str_add_with_scopeA_ptr ITTNOTIFY_NAME(metadata_str_add_with_scopeA) +#define __itt_metadata_str_add_with_scopeW(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_add_with_scopeW,d,x,y,z,a) +#define __itt_metadata_str_add_with_scopeW_ptr ITTNOTIFY_NAME(metadata_str_add_with_scopeW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add_with_scope(d,x,y,z,a) ITTNOTIFY_VOID_D4(metadata_str_add_with_scope,d,x,y,z,a) +#define __itt_metadata_str_add_with_scope_ptr ITTNOTIFY_NAME(metadata_str_add_with_scope) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_add_with_scopeA(d,x,y,z,a) +#define __itt_metadata_str_add_with_scopeA_ptr 0 +#define __itt_metadata_str_add_with_scopeW(d,x,y,z,a) +#define __itt_metadata_str_add_with_scopeW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add_with_scope(d,x,y,z,a) +#define __itt_metadata_str_add_with_scope_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_metadata_str_add_with_scopeA_ptr 0 +#define __itt_metadata_str_add_with_scopeW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_metadata_str_add_with_scope_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @} metadata group */ + +/** + * @defgroup relations Relations + * Instances of named entities can be explicitly associated with other + * instances using instance IDs and the relationship API calls. + * + * @{ + */ + +/** + * @ingroup relations + * @brief The kind of relation between two instances is specified by the enumerated type __itt_relation. + * Relations between instances can be added with an API call. The relation + * API uses instance IDs. Relations can be added before or after the actual + * instances are created and persist independently of the instances. This + * is the motivation for having different lifetimes for instance IDs and + * the actual instances. + */ +typedef enum +{ + __itt_relation_is_unknown = 0, + __itt_relation_is_dependent_on, /**< "A is dependent on B" means that A cannot start until B completes */ + __itt_relation_is_sibling_of, /**< "A is sibling of B" means that A and B were created as a group */ + __itt_relation_is_parent_of, /**< "A is parent of B" means that A created B */ + __itt_relation_is_continuation_of, /**< "A is continuation of B" means that A assumes the dependencies of B */ + __itt_relation_is_child_of, /**< "A is child of B" means that A was created by B (inverse of is_parent_of) */ + __itt_relation_is_continued_by, /**< "A is continued by B" means that B assumes the dependencies of A (inverse of is_continuation_of) */ + __itt_relation_is_predecessor_to /**< "A is predecessor to B" means that B cannot start until A completes (inverse of is_dependent_on) */ +} __itt_relation; + +/** + * @ingroup relations + * @brief Add a relation to the current task instance. + * The current task instance is the head of the relation. + * @param[in] domain The domain controlling this call + * @param[in] relation The kind of relation + * @param[in] tail The ID for the tail of the relation + */ +void ITTAPI __itt_relation_add_to_current(const __itt_domain *domain, __itt_relation relation, __itt_id tail); + +/** + * @ingroup relations + * @brief Add a relation between two instance identifiers. + * @param[in] domain The domain controlling this call + * @param[in] head The ID for the head of the relation + * @param[in] relation The kind of relation + * @param[in] tail The ID for the tail of the relation + */ +void ITTAPI __itt_relation_add(const __itt_domain *domain, __itt_id head, __itt_relation relation, __itt_id tail); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, relation_add_to_current, (const __itt_domain *domain, __itt_relation relation, __itt_id tail)) +ITT_STUBV(ITTAPI, void, relation_add, (const __itt_domain *domain, __itt_id head, __itt_relation relation, __itt_id tail)) +#define __itt_relation_add_to_current(d,x,y) ITTNOTIFY_VOID_D2(relation_add_to_current,d,x,y) +#define __itt_relation_add_to_current_ptr ITTNOTIFY_NAME(relation_add_to_current) +#define __itt_relation_add(d,x,y,z) ITTNOTIFY_VOID_D3(relation_add,d,x,y,z) +#define __itt_relation_add_ptr ITTNOTIFY_NAME(relation_add) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_relation_add_to_current(d,x,y) +#define __itt_relation_add_to_current_ptr 0 +#define __itt_relation_add(d,x,y,z) +#define __itt_relation_add_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_relation_add_to_current_ptr 0 +#define __itt_relation_add_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} relations group */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_clock_info +{ + unsigned long long clock_freq; /*!< Clock domain frequency */ + unsigned long long clock_base; /*!< Clock domain base timestamp */ +} __itt_clock_info; + +#pragma pack(pop) +/** @endcond */ + +/** @cond exclude_from_documentation */ +typedef void (ITTAPI *__itt_get_clock_info_fn)(__itt_clock_info* clock_info, void* data); +/** @endcond */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_clock_domain +{ + __itt_clock_info info; /*!< Most recent clock domain info */ + __itt_get_clock_info_fn fn; /*!< Callback function pointer */ + void* fn_data; /*!< Input argument for the callback function */ + int extra1; /*!< Reserved. Must be zero */ + void* extra2; /*!< Reserved. Must be zero */ + struct ___itt_clock_domain* next; +} __itt_clock_domain; + +#pragma pack(pop) +/** @endcond */ + +/** + * @ingroup clockdomains + * @brief Create a clock domain. + * Certain applications require the capability to trace their application using + * a clock domain different than the CPU, for instance the instrumentation of events + * that occur on a GPU. + * Because the set of domains is expected to be static over the application's execution time, + * there is no mechanism to destroy a domain. + * Any domain can be accessed by any thread in the process, regardless of which thread created + * the domain. This call is thread-safe. + * @param[in] fn A pointer to a callback function which retrieves alternative CPU timestamps + * @param[in] fn_data Argument for a callback function; may be NULL + */ +__itt_clock_domain* ITTAPI __itt_clock_domain_create(__itt_get_clock_info_fn fn, void* fn_data); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, __itt_clock_domain*, clock_domain_create, (__itt_get_clock_info_fn fn, void* fn_data)) +#define __itt_clock_domain_create ITTNOTIFY_DATA(clock_domain_create) +#define __itt_clock_domain_create_ptr ITTNOTIFY_NAME(clock_domain_create) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_clock_domain_create(fn,fn_data) (__itt_clock_domain*)0 +#define __itt_clock_domain_create_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_clock_domain_create_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup clockdomains + * @brief Recalculate clock domains frequences and clock base timestamps. + */ +void ITTAPI __itt_clock_domain_reset(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, clock_domain_reset, (void)) +#define __itt_clock_domain_reset ITTNOTIFY_VOID(clock_domain_reset) +#define __itt_clock_domain_reset_ptr ITTNOTIFY_NAME(clock_domain_reset) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_clock_domain_reset() +#define __itt_clock_domain_reset_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_clock_domain_reset_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup clockdomain + * @brief Create an instance of identifier. This establishes the beginning of the lifetime of + * an instance of the given ID in the trace. Once this lifetime starts, the ID can be used to + * tag named entity instances in calls such as __itt_task_begin, and to specify relationships among + * identified named entity instances, using the \ref relations APIs. + * @param[in] domain The domain controlling the execution of this call. + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] id The ID to create. + */ +void ITTAPI __itt_id_create_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id); + +/** + * @ingroup clockdomain + * @brief Destroy an instance of identifier. This ends the lifetime of the current instance of the + * given ID value in the trace. Any relationships that are established after this lifetime ends are + * invalid. This call must be performed before the given ID value can be reused for a different + * named entity instance. + * @param[in] domain The domain controlling the execution of this call. + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] id The ID to destroy. + */ +void ITTAPI __itt_id_destroy_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, id_create_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id)) +ITT_STUBV(ITTAPI, void, id_destroy_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id)) +#define __itt_id_create_ex(d,x,y,z) ITTNOTIFY_VOID_D3(id_create_ex,d,x,y,z) +#define __itt_id_create_ex_ptr ITTNOTIFY_NAME(id_create_ex) +#define __itt_id_destroy_ex(d,x,y,z) ITTNOTIFY_VOID_D3(id_destroy_ex,d,x,y,z) +#define __itt_id_destroy_ex_ptr ITTNOTIFY_NAME(id_destroy_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_id_create_ex(domain,clock_domain,timestamp,id) +#define __itt_id_create_ex_ptr 0 +#define __itt_id_destroy_ex(domain,clock_domain,timestamp,id) +#define __itt_id_destroy_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_id_create_ex_ptr 0 +#define __itt_id_destroy_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup clockdomain + * @brief Begin a task instance. + * @param[in] domain The domain for this task + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] taskid The instance ID for this task instance, or __itt_null + * @param[in] parentid The parent instance to which this task instance belongs, or __itt_null + * @param[in] name The name of this task + */ +void ITTAPI __itt_task_begin_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid, __itt_id parentid, __itt_string_handle* name); + +/** + * @ingroup clockdomain + * @brief Begin a task instance. + * @param[in] domain The domain for this task + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] taskid The identifier for this task instance, or __itt_null + * @param[in] parentid The parent of this task, or __itt_null + * @param[in] fn The pointer to the function you are tracing + */ +void ITTAPI __itt_task_begin_fn_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid, __itt_id parentid, void* fn); + +/** + * @ingroup clockdomain + * @brief End the current task instance. + * @param[in] domain The domain for this task + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + */ +void ITTAPI __itt_task_end_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, task_begin_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id, __itt_id parentid, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, task_begin_fn_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id, __itt_id parentid, void* fn)) +ITT_STUBV(ITTAPI, void, task_end_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp)) +#define __itt_task_begin_ex(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(task_begin_ex,d,x,y,z,a,b) +#define __itt_task_begin_ex_ptr ITTNOTIFY_NAME(task_begin_ex) +#define __itt_task_begin_fn_ex(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(task_begin_fn_ex,d,x,y,z,a,b) +#define __itt_task_begin_fn_ex_ptr ITTNOTIFY_NAME(task_begin_fn_ex) +#define __itt_task_end_ex(d,x,y) ITTNOTIFY_VOID_D2(task_end_ex,d,x,y) +#define __itt_task_end_ex_ptr ITTNOTIFY_NAME(task_end_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_task_begin_ex(domain,clock_domain,timestamp,id,parentid,name) +#define __itt_task_begin_ex_ptr 0 +#define __itt_task_begin_fn_ex(domain,clock_domain,timestamp,id,parentid,fn) +#define __itt_task_begin_fn_ex_ptr 0 +#define __itt_task_end_ex(domain,clock_domain,timestamp) +#define __itt_task_end_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_task_begin_ex_ptr 0 +#define __itt_task_begin_fn_ex_ptr 0 +#define __itt_task_end_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @defgroup counters Counters + * @ingroup public + * Counters are user-defined objects with a monotonically increasing + * value. Counter values are 64-bit unsigned integers. + * Counters have names that can be displayed in + * the tools. + * @{ + */ + +/** + * @brief opaque structure for counter identification + */ +/** @cond exclude_from_documentation */ + +typedef struct ___itt_counter* __itt_counter; + +/** + * @brief Create an unsigned 64 bits integer counter with given name/domain + * + * After __itt_counter_create() is called, __itt_counter_inc(id), __itt_counter_inc_delta(id, delta), + * __itt_counter_set_value(id, value_ptr) or __itt_counter_set_value_ex(id, clock_domain, timestamp, value_ptr) + * can be used to change the value of the counter, where value_ptr is a pointer to an unsigned 64 bits integer + * + * The call is equal to __itt_counter_create_typed(name, domain, __itt_metadata_u64) + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_counter ITTAPI __itt_counter_createA(const char *name, const char *domain); +__itt_counter ITTAPI __itt_counter_createW(const wchar_t *name, const wchar_t *domain); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_counter_create __itt_counter_createW +# define __itt_counter_create_ptr __itt_counter_createW_ptr +#else /* UNICODE */ +# define __itt_counter_create __itt_counter_createA +# define __itt_counter_create_ptr __itt_counter_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_counter ITTAPI __itt_counter_create(const char *name, const char *domain); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_counter, counter_createA, (const char *name, const char *domain)) +ITT_STUB(ITTAPI, __itt_counter, counter_createW, (const wchar_t *name, const wchar_t *domain)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_counter, counter_create, (const char *name, const char *domain)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_createA ITTNOTIFY_DATA(counter_createA) +#define __itt_counter_createA_ptr ITTNOTIFY_NAME(counter_createA) +#define __itt_counter_createW ITTNOTIFY_DATA(counter_createW) +#define __itt_counter_createW_ptr ITTNOTIFY_NAME(counter_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create ITTNOTIFY_DATA(counter_create) +#define __itt_counter_create_ptr ITTNOTIFY_NAME(counter_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_createA(name, domain) +#define __itt_counter_createA_ptr 0 +#define __itt_counter_createW(name, domain) +#define __itt_counter_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create(name, domain) +#define __itt_counter_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_createA_ptr 0 +#define __itt_counter_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Increment the unsigned 64 bits integer counter value + * + * Calling this function to non-unsigned 64 bits integer counters has no effect + */ +void ITTAPI __itt_counter_inc(__itt_counter id); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_inc, (__itt_counter id)) +#define __itt_counter_inc ITTNOTIFY_VOID(counter_inc) +#define __itt_counter_inc_ptr ITTNOTIFY_NAME(counter_inc) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_inc(id) +#define __itt_counter_inc_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_inc_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** + * @brief Increment the unsigned 64 bits integer counter value with x + * + * Calling this function to non-unsigned 64 bits integer counters has no effect + */ +void ITTAPI __itt_counter_inc_delta(__itt_counter id, unsigned long long value); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_inc_delta, (__itt_counter id, unsigned long long value)) +#define __itt_counter_inc_delta ITTNOTIFY_VOID(counter_inc_delta) +#define __itt_counter_inc_delta_ptr ITTNOTIFY_NAME(counter_inc_delta) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_inc_delta(id, value) +#define __itt_counter_inc_delta_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_inc_delta_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Decrement the unsigned 64 bits integer counter value + * + * Calling this function to non-unsigned 64 bits integer counters has no effect + */ +void ITTAPI __itt_counter_dec(__itt_counter id); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_dec, (__itt_counter id)) +#define __itt_counter_dec ITTNOTIFY_VOID(counter_dec) +#define __itt_counter_dec_ptr ITTNOTIFY_NAME(counter_dec) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_dec(id) +#define __itt_counter_dec_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_dec_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** + * @brief Decrement the unsigned 64 bits integer counter value with x + * + * Calling this function to non-unsigned 64 bits integer counters has no effect + */ +void ITTAPI __itt_counter_dec_delta(__itt_counter id, unsigned long long value); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_dec_delta, (__itt_counter id, unsigned long long value)) +#define __itt_counter_dec_delta ITTNOTIFY_VOID(counter_dec_delta) +#define __itt_counter_dec_delta_ptr ITTNOTIFY_NAME(counter_dec_delta) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_dec_delta(id, value) +#define __itt_counter_dec_delta_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_dec_delta_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup counters + * @brief Increment a counter by one. + * The first call with a given name creates a counter by that name and sets its + * value to zero. Successive calls increment the counter value. + * @param[in] domain The domain controlling the call. Counter names are not domain specific. + * The domain argument is used only to enable or disable the API calls. + * @param[in] name The name of the counter + */ +void ITTAPI __itt_counter_inc_v3(const __itt_domain *domain, __itt_string_handle *name); + +/** + * @ingroup counters + * @brief Increment a counter by the value specified in delta. + * @param[in] domain The domain controlling the call. Counter names are not domain specific. + * The domain argument is used only to enable or disable the API calls. + * @param[in] name The name of the counter + * @param[in] delta The amount by which to increment the counter + */ +void ITTAPI __itt_counter_inc_delta_v3(const __itt_domain *domain, __itt_string_handle *name, unsigned long long delta); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_inc_v3, (const __itt_domain *domain, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, counter_inc_delta_v3, (const __itt_domain *domain, __itt_string_handle *name, unsigned long long delta)) +#define __itt_counter_inc_v3(d,x) ITTNOTIFY_VOID_D1(counter_inc_v3,d,x) +#define __itt_counter_inc_v3_ptr ITTNOTIFY_NAME(counter_inc_v3) +#define __itt_counter_inc_delta_v3(d,x,y) ITTNOTIFY_VOID_D2(counter_inc_delta_v3,d,x,y) +#define __itt_counter_inc_delta_v3_ptr ITTNOTIFY_NAME(counter_inc_delta_v3) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_inc_v3(domain,name) +#define __itt_counter_inc_v3_ptr 0 +#define __itt_counter_inc_delta_v3(domain,name,delta) +#define __itt_counter_inc_delta_v3_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_inc_v3_ptr 0 +#define __itt_counter_inc_delta_v3_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + + +/** + * @ingroup counters + * @brief Decrement a counter by one. + * The first call with a given name creates a counter by that name and sets its + * value to zero. Successive calls decrement the counter value. + * @param[in] domain The domain controlling the call. Counter names are not domain specific. + * The domain argument is used only to enable or disable the API calls. + * @param[in] name The name of the counter + */ +void ITTAPI __itt_counter_dec_v3(const __itt_domain *domain, __itt_string_handle *name); + +/** + * @ingroup counters + * @brief Decrement a counter by the value specified in delta. + * @param[in] domain The domain controlling the call. Counter names are not domain specific. + * The domain argument is used only to enable or disable the API calls. + * @param[in] name The name of the counter + * @param[in] delta The amount by which to decrement the counter + */ +void ITTAPI __itt_counter_dec_delta_v3(const __itt_domain *domain, __itt_string_handle *name, unsigned long long delta); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_dec_v3, (const __itt_domain *domain, __itt_string_handle *name)) +ITT_STUBV(ITTAPI, void, counter_dec_delta_v3, (const __itt_domain *domain, __itt_string_handle *name, unsigned long long delta)) +#define __itt_counter_dec_v3(d,x) ITTNOTIFY_VOID_D1(counter_dec_v3,d,x) +#define __itt_counter_dec_v3_ptr ITTNOTIFY_NAME(counter_dec_v3) +#define __itt_counter_dec_delta_v3(d,x,y) ITTNOTIFY_VOID_D2(counter_dec_delta_v3,d,x,y) +#define __itt_counter_dec_delta_v3_ptr ITTNOTIFY_NAME(counter_dec_delta_v3) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_dec_v3(domain,name) +#define __itt_counter_dec_v3_ptr 0 +#define __itt_counter_dec_delta_v3(domain,name,delta) +#define __itt_counter_dec_delta_v3_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_dec_v3_ptr 0 +#define __itt_counter_dec_delta_v3_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @} counters group */ + + +/** + * @brief Set the counter value + */ +void ITTAPI __itt_counter_set_value(__itt_counter id, void *value_ptr); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_set_value, (__itt_counter id, void *value_ptr)) +#define __itt_counter_set_value ITTNOTIFY_VOID(counter_set_value) +#define __itt_counter_set_value_ptr ITTNOTIFY_NAME(counter_set_value) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_set_value(id, value_ptr) +#define __itt_counter_set_value_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_set_value_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Set the counter value + */ +void ITTAPI __itt_counter_set_value_ex(__itt_counter id, __itt_clock_domain *clock_domain, unsigned long long timestamp, void *value_ptr); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_set_value_ex, (__itt_counter id, __itt_clock_domain *clock_domain, unsigned long long timestamp, void *value_ptr)) +#define __itt_counter_set_value_ex ITTNOTIFY_VOID(counter_set_value_ex) +#define __itt_counter_set_value_ex_ptr ITTNOTIFY_NAME(counter_set_value_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_set_value_ex(id, clock_domain, timestamp, value_ptr) +#define __itt_counter_set_value_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_set_value_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Create a typed counter with given name/domain + * + * After __itt_counter_create_typed() is called, __itt_counter_inc(id), __itt_counter_inc_delta(id, delta), + * __itt_counter_set_value(id, value_ptr) or __itt_counter_set_value_ex(id, clock_domain, timestamp, value_ptr) + * can be used to change the value of the counter + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_counter ITTAPI __itt_counter_create_typedA(const char *name, const char *domain, __itt_metadata_type type); +__itt_counter ITTAPI __itt_counter_create_typedW(const wchar_t *name, const wchar_t *domain, __itt_metadata_type type); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_counter_create_typed __itt_counter_create_typedW +# define __itt_counter_create_typed_ptr __itt_counter_create_typedW_ptr +#else /* UNICODE */ +# define __itt_counter_create_typed __itt_counter_create_typedA +# define __itt_counter_create_typed_ptr __itt_counter_create_typedA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_counter ITTAPI __itt_counter_create_typed(const char *name, const char *domain, __itt_metadata_type type); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_counter, counter_create_typedA, (const char *name, const char *domain, __itt_metadata_type type)) +ITT_STUB(ITTAPI, __itt_counter, counter_create_typedW, (const wchar_t *name, const wchar_t *domain, __itt_metadata_type type)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_counter, counter_create_typed, (const char *name, const char *domain, __itt_metadata_type type)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_create_typedA ITTNOTIFY_DATA(counter_create_typedA) +#define __itt_counter_create_typedA_ptr ITTNOTIFY_NAME(counter_create_typedA) +#define __itt_counter_create_typedW ITTNOTIFY_DATA(counter_create_typedW) +#define __itt_counter_create_typedW_ptr ITTNOTIFY_NAME(counter_create_typedW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create_typed ITTNOTIFY_DATA(counter_create_typed) +#define __itt_counter_create_typed_ptr ITTNOTIFY_NAME(counter_create_typed) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_create_typedA(name, domain, type) +#define __itt_counter_create_typedA_ptr 0 +#define __itt_counter_create_typedW(name, domain, type) +#define __itt_counter_create_typedW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create_typed(name, domain, type) +#define __itt_counter_create_typed_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_counter_create_typedA_ptr 0 +#define __itt_counter_create_typedW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_counter_create_typed_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Destroy the counter identified by the pointer previously returned by __itt_counter_create() or + * __itt_counter_create_typed() + */ +void ITTAPI __itt_counter_destroy(__itt_counter id); + +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, counter_destroy, (__itt_counter id)) +#define __itt_counter_destroy ITTNOTIFY_VOID(counter_destroy) +#define __itt_counter_destroy_ptr ITTNOTIFY_NAME(counter_destroy) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_counter_destroy(id) +#define __itt_counter_destroy_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_counter_destroy_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} counters group */ + +/** + * @ingroup markers + * @brief Create a marker instance. + * @param[in] domain The domain for this marker + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] id The instance ID for this marker, or __itt_null + * @param[in] name The name for this marker + * @param[in] scope The scope for this marker + */ +void ITTAPI __itt_marker_ex(const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id, __itt_string_handle *name, __itt_scope scope); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, marker_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id id, __itt_string_handle *name, __itt_scope scope)) +#define __itt_marker_ex(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(marker_ex,d,x,y,z,a,b) +#define __itt_marker_ex_ptr ITTNOTIFY_NAME(marker_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_marker_ex(domain,clock_domain,timestamp,id,name,scope) +#define __itt_marker_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_marker_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @ingroup clockdomain + * @brief Add a relation to the current task instance. + * The current task instance is the head of the relation. + * @param[in] domain The domain controlling this call + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] relation The kind of relation + * @param[in] tail The ID for the tail of the relation + */ +void ITTAPI __itt_relation_add_to_current_ex(const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_relation relation, __itt_id tail); + +/** + * @ingroup clockdomain + * @brief Add a relation between two instance identifiers. + * @param[in] domain The domain controlling this call + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] head The ID for the head of the relation + * @param[in] relation The kind of relation + * @param[in] tail The ID for the tail of the relation + */ +void ITTAPI __itt_relation_add_ex(const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id head, __itt_relation relation, __itt_id tail); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, relation_add_to_current_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_relation relation, __itt_id tail)) +ITT_STUBV(ITTAPI, void, relation_add_ex, (const __itt_domain *domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id head, __itt_relation relation, __itt_id tail)) +#define __itt_relation_add_to_current_ex(d,x,y,z,a) ITTNOTIFY_VOID_D4(relation_add_to_current_ex,d,x,y,z,a) +#define __itt_relation_add_to_current_ex_ptr ITTNOTIFY_NAME(relation_add_to_current_ex) +#define __itt_relation_add_ex(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(relation_add_ex,d,x,y,z,a,b) +#define __itt_relation_add_ex_ptr ITTNOTIFY_NAME(relation_add_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_relation_add_to_current_ex(domain,clock_domain,timestame,relation,tail) +#define __itt_relation_add_to_current_ex_ptr 0 +#define __itt_relation_add_ex(domain,clock_domain,timestamp,head,relation,tail) +#define __itt_relation_add_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_relation_add_to_current_ex_ptr 0 +#define __itt_relation_add_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @cond exclude_from_documentation */ +typedef enum ___itt_track_group_type +{ + __itt_track_group_type_normal = 0 +} __itt_track_group_type; +/** @endcond */ + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_track_group +{ + __itt_string_handle* name; /*!< Name of the track group */ + struct ___itt_track* track; /*!< List of child tracks */ + __itt_track_group_type tgtype; /*!< Type of the track group */ + int extra1; /*!< Reserved. Must be zero */ + void* extra2; /*!< Reserved. Must be zero */ + struct ___itt_track_group* next; +} __itt_track_group; + +#pragma pack(pop) +/** @endcond */ + +/** + * @brief Placeholder for custom track types. Currently, "normal" custom track + * is the only available track type. + */ +typedef enum ___itt_track_type +{ + __itt_track_type_normal = 0 +#ifdef INTEL_ITTNOTIFY_API_PRIVATE + , __itt_track_type_queue +#endif /* INTEL_ITTNOTIFY_API_PRIVATE */ +} __itt_track_type; + +/** @cond exclude_from_documentation */ +#pragma pack(push, 8) + +typedef struct ___itt_track +{ + __itt_string_handle* name; /*!< Name of the track group */ + __itt_track_group* group; /*!< Parent group to a track */ + __itt_track_type ttype; /*!< Type of the track */ + int extra1; /*!< Reserved. Must be zero */ + void* extra2; /*!< Reserved. Must be zero */ + struct ___itt_track* next; +} __itt_track; + +#pragma pack(pop) +/** @endcond */ + +/** + * @brief Create logical track group. + */ +__itt_track_group* ITTAPI __itt_track_group_create(__itt_string_handle* name, __itt_track_group_type track_group_type); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, __itt_track_group*, track_group_create, (__itt_string_handle* name, __itt_track_group_type track_group_type)) +#define __itt_track_group_create ITTNOTIFY_DATA(track_group_create) +#define __itt_track_group_create_ptr ITTNOTIFY_NAME(track_group_create) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_track_group_create(name) (__itt_track_group*)0 +#define __itt_track_group_create_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_track_group_create_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Create logical track. + */ +__itt_track* ITTAPI __itt_track_create(__itt_track_group* track_group, __itt_string_handle* name, __itt_track_type track_type); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, __itt_track*, track_create, (__itt_track_group* track_group,__itt_string_handle* name, __itt_track_type track_type)) +#define __itt_track_create ITTNOTIFY_DATA(track_create) +#define __itt_track_create_ptr ITTNOTIFY_NAME(track_create) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_track_create(track_group,name,track_type) (__itt_track*)0 +#define __itt_track_create_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_track_create_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Set the logical track. + */ +void ITTAPI __itt_set_track(__itt_track* track); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, set_track, (__itt_track *track)) +#define __itt_set_track ITTNOTIFY_VOID(set_track) +#define __itt_set_track_ptr ITTNOTIFY_NAME(set_track) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_set_track(track) +#define __itt_set_track_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_set_track_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/* ========================================================================== */ +/** @cond exclude_from_gpa_documentation */ +/** + * @defgroup events Events + * @ingroup public + * Events group + * @{ + */ +/** @brief user event type */ +typedef int __itt_event; + +/** + * @brief Create an event notification + * @note name or namelen being null/name and namelen not matching, user event feature not enabled + * @return non-zero event identifier upon success and __itt_err otherwise + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_event LIBITTAPI __itt_event_createA(const char *name, int namelen); +__itt_event LIBITTAPI __itt_event_createW(const wchar_t *name, int namelen); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_event_create __itt_event_createW +# define __itt_event_create_ptr __itt_event_createW_ptr +#else +# define __itt_event_create __itt_event_createA +# define __itt_event_create_ptr __itt_event_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_event LIBITTAPI __itt_event_create(const char *name, int namelen); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(LIBITTAPI, __itt_event, event_createA, (const char *name, int namelen)) +ITT_STUB(LIBITTAPI, __itt_event, event_createW, (const wchar_t *name, int namelen)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(LIBITTAPI, __itt_event, event_create, (const char *name, int namelen)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_event_createA ITTNOTIFY_DATA(event_createA) +#define __itt_event_createA_ptr ITTNOTIFY_NAME(event_createA) +#define __itt_event_createW ITTNOTIFY_DATA(event_createW) +#define __itt_event_createW_ptr ITTNOTIFY_NAME(event_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_event_create ITTNOTIFY_DATA(event_create) +#define __itt_event_create_ptr ITTNOTIFY_NAME(event_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_event_createA(name, namelen) (__itt_event)0 +#define __itt_event_createA_ptr 0 +#define __itt_event_createW(name, namelen) (__itt_event)0 +#define __itt_event_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_event_create(name, namelen) (__itt_event)0 +#define __itt_event_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_event_createA_ptr 0 +#define __itt_event_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_event_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an event occurrence. + * @return __itt_err upon failure (invalid event id/user event feature not enabled) + */ +int LIBITTAPI __itt_event_start(__itt_event event); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(LIBITTAPI, int, event_start, (__itt_event event)) +#define __itt_event_start ITTNOTIFY_DATA(event_start) +#define __itt_event_start_ptr ITTNOTIFY_NAME(event_start) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_event_start(event) (int)0 +#define __itt_event_start_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_event_start_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Record an event end occurrence. + * @note It is optional if events do not have durations. + * @return __itt_err upon failure (invalid event id/user event feature not enabled) + */ +int LIBITTAPI __itt_event_end(__itt_event event); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(LIBITTAPI, int, event_end, (__itt_event event)) +#define __itt_event_end ITTNOTIFY_DATA(event_end) +#define __itt_event_end_ptr ITTNOTIFY_NAME(event_end) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_event_end(event) (int)0 +#define __itt_event_end_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_event_end_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} events group */ + + +/** + * @defgroup arrays Arrays Visualizer + * @ingroup public + * Visualize arrays + * @{ + */ + +/** + * @enum __itt_av_data_type + * @brief Defines types of arrays data (for C/C++ intrinsic types) + */ +typedef enum +{ + __itt_e_first = 0, + __itt_e_char = 0, /* 1-byte integer */ + __itt_e_uchar, /* 1-byte unsigned integer */ + __itt_e_int16, /* 2-byte integer */ + __itt_e_uint16, /* 2-byte unsigned integer */ + __itt_e_int32, /* 4-byte integer */ + __itt_e_uint32, /* 4-byte unsigned integer */ + __itt_e_int64, /* 8-byte integer */ + __itt_e_uint64, /* 8-byte unsigned integer */ + __itt_e_float, /* 4-byte floating */ + __itt_e_double, /* 8-byte floating */ + __itt_e_last = __itt_e_double +} __itt_av_data_type; + +/** + * @brief Save an array data to a file. + * Output format is defined by the file extension. The csv and bmp formats are supported (bmp - for 2-dimensional array only). + * @param[in] data - pointer to the array data + * @param[in] rank - the rank of the array + * @param[in] dimensions - pointer to an array of integers, which specifies the array dimensions. + * The size of dimensions must be equal to the rank + * @param[in] type - the type of the array, specified as one of the __itt_av_data_type values (for intrinsic types) + * @param[in] filePath - the file path; the output format is defined by the file extension + * @param[in] columnOrder - defines how the array is stored in the linear memory. + * It should be 1 for column-major order (e.g. in FORTRAN) or 0 - for row-major order (e.g. in C). + */ + +#if ITT_PLATFORM==ITT_PLATFORM_WIN +int ITTAPI __itt_av_saveA(void *data, int rank, const int *dimensions, int type, const char *filePath, int columnOrder); +int ITTAPI __itt_av_saveW(void *data, int rank, const int *dimensions, int type, const wchar_t *filePath, int columnOrder); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_av_save __itt_av_saveW +# define __itt_av_save_ptr __itt_av_saveW_ptr +#else /* UNICODE */ +# define __itt_av_save __itt_av_saveA +# define __itt_av_save_ptr __itt_av_saveA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +int ITTAPI __itt_av_save(void *data, int rank, const int *dimensions, int type, const char *filePath, int columnOrder); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, int, av_saveA, (void *data, int rank, const int *dimensions, int type, const char *filePath, int columnOrder)) +ITT_STUB(ITTAPI, int, av_saveW, (void *data, int rank, const int *dimensions, int type, const wchar_t *filePath, int columnOrder)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, int, av_save, (void *data, int rank, const int *dimensions, int type, const char *filePath, int columnOrder)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_av_saveA ITTNOTIFY_DATA(av_saveA) +#define __itt_av_saveA_ptr ITTNOTIFY_NAME(av_saveA) +#define __itt_av_saveW ITTNOTIFY_DATA(av_saveW) +#define __itt_av_saveW_ptr ITTNOTIFY_NAME(av_saveW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_av_save ITTNOTIFY_DATA(av_save) +#define __itt_av_save_ptr ITTNOTIFY_NAME(av_save) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_av_saveA(name) +#define __itt_av_saveA_ptr 0 +#define __itt_av_saveW(name) +#define __itt_av_saveW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_av_save(name) +#define __itt_av_save_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_av_saveA_ptr 0 +#define __itt_av_saveW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_av_save_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +void ITTAPI __itt_enable_attach(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, enable_attach, (void)) +#define __itt_enable_attach ITTNOTIFY_VOID(enable_attach) +#define __itt_enable_attach_ptr ITTNOTIFY_NAME(enable_attach) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_enable_attach() +#define __itt_enable_attach_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_enable_attach_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @cond exclude_from_gpa_documentation */ + +/** @} arrays group */ + +/** @endcond */ + +/** + * @brief Module load info + * This API is used to report necessary information in case of module relocation + * @param[in] start_addr - relocated module start address + * @param[in] end_addr - relocated module end address + * @param[in] path - file system path to the module + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +void ITTAPI __itt_module_loadA(void *start_addr, void *end_addr, const char *path); +void ITTAPI __itt_module_loadW(void *start_addr, void *end_addr, const wchar_t *path); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_module_load __itt_module_loadW +# define __itt_module_load_ptr __itt_module_loadW_ptr +#else /* UNICODE */ +# define __itt_module_load __itt_module_loadA +# define __itt_module_load_ptr __itt_module_loadA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +void ITTAPI __itt_module_load(void *start_addr, void *end_addr, const char *path); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, void, module_loadA, (void *start_addr, void *end_addr, const char *path)) +ITT_STUB(ITTAPI, void, module_loadW, (void *start_addr, void *end_addr, const wchar_t *path)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, void, module_load, (void *start_addr, void *end_addr, const char *path)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_module_loadA ITTNOTIFY_VOID(module_loadA) +#define __itt_module_loadA_ptr ITTNOTIFY_NAME(module_loadA) +#define __itt_module_loadW ITTNOTIFY_VOID(module_loadW) +#define __itt_module_loadW_ptr ITTNOTIFY_NAME(module_loadW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_module_load ITTNOTIFY_VOID(module_load) +#define __itt_module_load_ptr ITTNOTIFY_NAME(module_load) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_module_loadA(start_addr, end_addr, path) +#define __itt_module_loadA_ptr 0 +#define __itt_module_loadW(start_addr, end_addr, path) +#define __itt_module_loadW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_module_load(start_addr, end_addr, path) +#define __itt_module_load_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_module_loadA_ptr 0 +#define __itt_module_loadW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_module_load_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + + + +#ifdef __cplusplus +} +#endif /* __cplusplus */ + +#endif /* _ITTNOTIFY_H_ */ + +#ifdef INTEL_ITTNOTIFY_API_PRIVATE + +#ifndef _ITTNOTIFY_PRIVATE_ +#define _ITTNOTIFY_PRIVATE_ + +#ifdef __cplusplus +extern "C" { +#endif /* __cplusplus */ + +/** + * @ingroup clockdomain + * @brief Begin an overlapped task instance. + * @param[in] domain The domain for this task + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] taskid The identifier for this task instance, *cannot* be __itt_null. + * @param[in] parentid The parent of this task, or __itt_null. + * @param[in] name The name of this task. + */ +void ITTAPI __itt_task_begin_overlapped_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid, __itt_id parentid, __itt_string_handle* name); + +/** + * @ingroup clockdomain + * @brief End an overlapped task instance. + * @param[in] domain The domain for this task + * @param[in] clock_domain The clock domain controlling the execution of this call. + * @param[in] timestamp The user defined timestamp. + * @param[in] taskid Explicit ID of finished task + */ +void ITTAPI __itt_task_end_overlapped_ex(const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, task_begin_overlapped_ex, (const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid, __itt_id parentid, __itt_string_handle* name)) +ITT_STUBV(ITTAPI, void, task_end_overlapped_ex, (const __itt_domain* domain, __itt_clock_domain* clock_domain, unsigned long long timestamp, __itt_id taskid)) +#define __itt_task_begin_overlapped_ex(d,x,y,z,a,b) ITTNOTIFY_VOID_D5(task_begin_overlapped_ex,d,x,y,z,a,b) +#define __itt_task_begin_overlapped_ex_ptr ITTNOTIFY_NAME(task_begin_overlapped_ex) +#define __itt_task_end_overlapped_ex(d,x,y,z) ITTNOTIFY_VOID_D3(task_end_overlapped_ex,d,x,y,z) +#define __itt_task_end_overlapped_ex_ptr ITTNOTIFY_NAME(task_end_overlapped_ex) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_task_begin_overlapped_ex(domain,clock_domain,timestamp,taskid,parentid,name) +#define __itt_task_begin_overlapped_ex_ptr 0 +#define __itt_task_end_overlapped_ex(domain,clock_domain,timestamp,taskid) +#define __itt_task_end_overlapped_ex_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_task_begin_overlapped_ex_ptr 0 +#define __itt_task_end_overlapped_ptr 0 +#define __itt_task_end_overlapped_ex_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @defgroup makrs_internal Marks + * @ingroup internal + * Marks group + * @warning Internal API: + * - It is not shipped to outside of Intel + * - It is delivered to internal Intel teams using e-mail or SVN access only + * @{ + */ +/** @brief user mark type */ +typedef int __itt_mark_type; + +/** + * @brief Creates a user mark type with the specified name using char or Unicode string. + * @param[in] name - name of mark to create + * @return Returns a handle to the mark type + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +__itt_mark_type ITTAPI __itt_mark_createA(const char *name); +__itt_mark_type ITTAPI __itt_mark_createW(const wchar_t *name); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_mark_create __itt_mark_createW +# define __itt_mark_create_ptr __itt_mark_createW_ptr +#else /* UNICODE */ +# define __itt_mark_create __itt_mark_createA +# define __itt_mark_create_ptr __itt_mark_createA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +__itt_mark_type ITTAPI __itt_mark_create(const char *name); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, __itt_mark_type, mark_createA, (const char *name)) +ITT_STUB(ITTAPI, __itt_mark_type, mark_createW, (const wchar_t *name)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, __itt_mark_type, mark_create, (const char *name)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_createA ITTNOTIFY_DATA(mark_createA) +#define __itt_mark_createA_ptr ITTNOTIFY_NAME(mark_createA) +#define __itt_mark_createW ITTNOTIFY_DATA(mark_createW) +#define __itt_mark_createW_ptr ITTNOTIFY_NAME(mark_createW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_create ITTNOTIFY_DATA(mark_create) +#define __itt_mark_create_ptr ITTNOTIFY_NAME(mark_create) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_createA(name) (__itt_mark_type)0 +#define __itt_mark_createA_ptr 0 +#define __itt_mark_createW(name) (__itt_mark_type)0 +#define __itt_mark_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_create(name) (__itt_mark_type)0 +#define __itt_mark_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_createA_ptr 0 +#define __itt_mark_createW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_create_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Creates a "discrete" user mark type of the specified type and an optional parameter using char or Unicode string. + * + * - The mark of "discrete" type is placed to collection results in case of success. It appears in overtime view(s) as a special tick sign. + * - The call is "synchronous" - function returns after mark is actually added to results. + * - This function is useful, for example, to mark different phases of application + * (beginning of the next mark automatically meand end of current region). + * - Can be used together with "continuous" marks (see below) at the same collection session + * @param[in] mt - mark, created by __itt_mark_create(const char* name) function + * @param[in] parameter - string parameter of mark + * @return Returns zero value in case of success, non-zero value otherwise. + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +int ITTAPI __itt_markA(__itt_mark_type mt, const char *parameter); +int ITTAPI __itt_markW(__itt_mark_type mt, const wchar_t *parameter); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_mark __itt_markW +# define __itt_mark_ptr __itt_markW_ptr +#else /* UNICODE */ +# define __itt_mark __itt_markA +# define __itt_mark_ptr __itt_markA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +int ITTAPI __itt_mark(__itt_mark_type mt, const char *parameter); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, int, markA, (__itt_mark_type mt, const char *parameter)) +ITT_STUB(ITTAPI, int, markW, (__itt_mark_type mt, const wchar_t *parameter)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, int, mark, (__itt_mark_type mt, const char *parameter)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_markA ITTNOTIFY_DATA(markA) +#define __itt_markA_ptr ITTNOTIFY_NAME(markA) +#define __itt_markW ITTNOTIFY_DATA(markW) +#define __itt_markW_ptr ITTNOTIFY_NAME(markW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark ITTNOTIFY_DATA(mark) +#define __itt_mark_ptr ITTNOTIFY_NAME(mark) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_markA(mt, parameter) (int)0 +#define __itt_markA_ptr 0 +#define __itt_markW(mt, parameter) (int)0 +#define __itt_markW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark(mt, parameter) (int)0 +#define __itt_mark_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_markA_ptr 0 +#define __itt_markW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Use this if necessary to create a "discrete" user event type (mark) for process + * rather then for one thread + * @see int __itt_mark(__itt_mark_type mt, const char* parameter); + */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +int ITTAPI __itt_mark_globalA(__itt_mark_type mt, const char *parameter); +int ITTAPI __itt_mark_globalW(__itt_mark_type mt, const wchar_t *parameter); +#if defined(UNICODE) || defined(_UNICODE) +# define __itt_mark_global __itt_mark_globalW +# define __itt_mark_global_ptr __itt_mark_globalW_ptr +#else /* UNICODE */ +# define __itt_mark_global __itt_mark_globalA +# define __itt_mark_global_ptr __itt_mark_globalA_ptr +#endif /* UNICODE */ +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +int ITTAPI __itt_mark_global(__itt_mark_type mt, const char *parameter); +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#if ITT_PLATFORM==ITT_PLATFORM_WIN +ITT_STUB(ITTAPI, int, mark_globalA, (__itt_mark_type mt, const char *parameter)) +ITT_STUB(ITTAPI, int, mark_globalW, (__itt_mark_type mt, const wchar_t *parameter)) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +ITT_STUB(ITTAPI, int, mark_global, (__itt_mark_type mt, const char *parameter)) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_globalA ITTNOTIFY_DATA(mark_globalA) +#define __itt_mark_globalA_ptr ITTNOTIFY_NAME(mark_globalA) +#define __itt_mark_globalW ITTNOTIFY_DATA(mark_globalW) +#define __itt_mark_globalW_ptr ITTNOTIFY_NAME(mark_globalW) +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_global ITTNOTIFY_DATA(mark_global) +#define __itt_mark_global_ptr ITTNOTIFY_NAME(mark_global) +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#else /* INTEL_NO_ITTNOTIFY_API */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_globalA(mt, parameter) (int)0 +#define __itt_mark_globalA_ptr 0 +#define __itt_mark_globalW(mt, parameter) (int)0 +#define __itt_mark_globalW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_global(mt, parameter) (int)0 +#define __itt_mark_global_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#if ITT_PLATFORM==ITT_PLATFORM_WIN +#define __itt_mark_globalA_ptr 0 +#define __itt_mark_globalW_ptr 0 +#else /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#define __itt_mark_global_ptr 0 +#endif /* ITT_PLATFORM==ITT_PLATFORM_WIN */ +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Creates an "end" point for "continuous" mark with specified name. + * + * - Returns zero value in case of success, non-zero value otherwise. + * Also returns non-zero value when preceding "begin" point for the + * mark with the same name failed to be created or not created. + * - The mark of "continuous" type is placed to collection results in + * case of success. It appears in overtime view(s) as a special tick + * sign (different from "discrete" mark) together with line from + * corresponding "begin" mark to "end" mark. + * @note Continuous marks can overlap and be nested inside each other. + * Discrete mark can be nested inside marked region + * @param[in] mt - mark, created by __itt_mark_create(const char* name) function + * @return Returns zero value in case of success, non-zero value otherwise. + */ +int ITTAPI __itt_mark_off(__itt_mark_type mt); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, int, mark_off, (__itt_mark_type mt)) +#define __itt_mark_off ITTNOTIFY_DATA(mark_off) +#define __itt_mark_off_ptr ITTNOTIFY_NAME(mark_off) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_mark_off(mt) (int)0 +#define __itt_mark_off_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_mark_off_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Use this if necessary to create an "end" point for mark of process + * @see int __itt_mark_off(__itt_mark_type mt); + */ +int ITTAPI __itt_mark_global_off(__itt_mark_type mt); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, int, mark_global_off, (__itt_mark_type mt)) +#define __itt_mark_global_off ITTNOTIFY_DATA(mark_global_off) +#define __itt_mark_global_off_ptr ITTNOTIFY_NAME(mark_global_off) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_mark_global_off(mt) (int)0 +#define __itt_mark_global_off_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_mark_global_off_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ +/** @} marks group */ + +/** + * @defgroup counters_internal Counters + * @ingroup internal + * Counters group + * @{ + */ + + +/** + * @defgroup stitch Stack Stitching + * @ingroup internal + * Stack Stitching group + * @{ + */ +/** + * @brief opaque structure for counter identification + */ +typedef struct ___itt_caller *__itt_caller; + +/** + * @brief Create the stitch point e.g. a point in call stack where other stacks should be stitched to. + * The function returns a unique identifier which is used to match the cut points with corresponding stitch points. + */ +__itt_caller ITTAPI __itt_stack_caller_create(void); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUB(ITTAPI, __itt_caller, stack_caller_create, (void)) +#define __itt_stack_caller_create ITTNOTIFY_DATA(stack_caller_create) +#define __itt_stack_caller_create_ptr ITTNOTIFY_NAME(stack_caller_create) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_stack_caller_create() (__itt_caller)0 +#define __itt_stack_caller_create_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_stack_caller_create_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Destroy the inforamtion about stitch point identified by the pointer previously returned by __itt_stack_caller_create() + */ +void ITTAPI __itt_stack_caller_destroy(__itt_caller id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, stack_caller_destroy, (__itt_caller id)) +#define __itt_stack_caller_destroy ITTNOTIFY_VOID(stack_caller_destroy) +#define __itt_stack_caller_destroy_ptr ITTNOTIFY_NAME(stack_caller_destroy) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_stack_caller_destroy(id) +#define __itt_stack_caller_destroy_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_stack_caller_destroy_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief Sets the cut point. Stack from each event which occurs after this call will be cut + * at the same stack level the function was called and stitched to the corresponding stitch point. + */ +void ITTAPI __itt_stack_callee_enter(__itt_caller id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, stack_callee_enter, (__itt_caller id)) +#define __itt_stack_callee_enter ITTNOTIFY_VOID(stack_callee_enter) +#define __itt_stack_callee_enter_ptr ITTNOTIFY_NAME(stack_callee_enter) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_stack_callee_enter(id) +#define __itt_stack_callee_enter_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_stack_callee_enter_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** + * @brief This function eliminates the cut point which was set by latest __itt_stack_callee_enter(). + */ +void ITTAPI __itt_stack_callee_leave(__itt_caller id); + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +ITT_STUBV(ITTAPI, void, stack_callee_leave, (__itt_caller id)) +#define __itt_stack_callee_leave ITTNOTIFY_VOID(stack_callee_leave) +#define __itt_stack_callee_leave_ptr ITTNOTIFY_NAME(stack_callee_leave) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_stack_callee_leave(id) +#define __itt_stack_callee_leave_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_stack_callee_leave_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +/** @} stitch group */ + +/* ***************************************************************************************************************************** */ + +#include + +/** @cond exclude_from_documentation */ +typedef enum __itt_error_code +{ + __itt_error_success = 0, /*!< no error */ + __itt_error_no_module = 1, /*!< module can't be loaded */ + /* %1$s -- library name; win: %2$d -- system error code; unx: %2$s -- system error message. */ + __itt_error_no_symbol = 2, /*!< symbol not found */ + /* %1$s -- library name, %2$s -- symbol name. */ + __itt_error_unknown_group = 3, /*!< unknown group specified */ + /* %1$s -- env var name, %2$s -- group name. */ + __itt_error_cant_read_env = 4, /*!< GetEnvironmentVariable() failed */ + /* %1$s -- env var name, %2$d -- system error. */ + __itt_error_env_too_long = 5, /*!< variable value too long */ + /* %1$s -- env var name, %2$d -- actual length of the var, %3$d -- max allowed length. */ + __itt_error_system = 6 /*!< pthread_mutexattr_init or pthread_mutex_init failed */ + /* %1$s -- function name, %2$d -- errno. */ +} __itt_error_code; + +typedef void (__itt_error_handler_t)(__itt_error_code code, va_list); +__itt_error_handler_t* __itt_set_error_handler(__itt_error_handler_t*); + +const char* ITTAPI __itt_api_version(void); +/** @endcond */ + +/** @cond exclude_from_documentation */ +#ifndef INTEL_NO_MACRO_BODY +#ifndef INTEL_NO_ITTNOTIFY_API +#define __itt_error_handler ITT_JOIN(INTEL_ITTNOTIFY_PREFIX, error_handler) +void __itt_error_handler(__itt_error_code code, va_list args); +extern const int ITTNOTIFY_NAME(err); +#define __itt_err ITTNOTIFY_NAME(err) +ITT_STUB(ITTAPI, const char*, api_version, (void)) +#define __itt_api_version ITTNOTIFY_DATA(api_version) +#define __itt_api_version_ptr ITTNOTIFY_NAME(api_version) +#else /* INTEL_NO_ITTNOTIFY_API */ +#define __itt_api_version() (const char*)0 +#define __itt_api_version_ptr 0 +#endif /* INTEL_NO_ITTNOTIFY_API */ +#else /* INTEL_NO_MACRO_BODY */ +#define __itt_api_version_ptr 0 +#endif /* INTEL_NO_MACRO_BODY */ +/** @endcond */ + +#ifdef __cplusplus +} +#endif /* __cplusplus */ + +#endif /* _ITTNOTIFY_PRIVATE_ */ + +#endif /* INTEL_ITTNOTIFY_API_PRIVATE */ + +// clang-format on -- cgit v1.2.3