diff options
Diffstat (limited to '')
25 files changed, 11388 insertions, 0 deletions
diff --git a/mozglue/baseprofiler/public/BaseProfileJSONWriter.h b/mozglue/baseprofiler/public/BaseProfileJSONWriter.h new file mode 100644 index 0000000000..5dcf06f3f3 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfileJSONWriter.h @@ -0,0 +1,388 @@ +/* -*- 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 BASEPROFILEJSONWRITER_H +#define BASEPROFILEJSONWRITER_H + +#include "mozilla/HashFunctions.h" +#include "mozilla/HashTable.h" +#include "mozilla/JSONWriter.h" +#include "mozilla/TimeStamp.h" +#include "mozilla/UniquePtr.h" + +#include <functional> +#include <ostream> +#include <string_view> + +namespace mozilla { +namespace baseprofiler { + +class SpliceableJSONWriter; + +// On average, profile JSONs are large enough such that we want to avoid +// reallocating its buffer when expanding. Additionally, the contents of the +// profile are not accessed until the profile is entirely written. For these +// reasons we use a chunked writer that keeps an array of chunks, which is +// concatenated together after writing is finished. +class ChunkedJSONWriteFunc final : public JSONWriteFunc { + public: + friend class SpliceableJSONWriter; + + ChunkedJSONWriteFunc() : mChunkPtr{nullptr}, mChunkEnd{nullptr} { + AllocChunk(kChunkSize); + } + + bool IsEmpty() const { + MOZ_ASSERT_IF(!mChunkPtr, !mChunkEnd && mChunkList.length() == 0 && + mChunkLengths.length() == 0); + return !mChunkPtr; + } + + void Write(const Span<const char>& aStr) override { + MOZ_ASSERT(mChunkPtr >= mChunkList.back().get() && mChunkPtr <= mChunkEnd); + MOZ_ASSERT(mChunkEnd >= mChunkList.back().get() + mChunkLengths.back()); + MOZ_ASSERT(*mChunkPtr == '\0'); + + // Most strings to be written are small, but subprocess profiles (e.g., + // from the content process in e10s) may be huge. If the string is larger + // than a chunk, allocate its own chunk. + char* newPtr; + if (aStr.size() >= kChunkSize) { + AllocChunk(aStr.size() + 1); + newPtr = mChunkPtr + aStr.size(); + } else { + newPtr = mChunkPtr + aStr.size(); + if (newPtr >= mChunkEnd) { + AllocChunk(kChunkSize); + newPtr = mChunkPtr + aStr.size(); + } + } + + memcpy(mChunkPtr, aStr.data(), aStr.size()); + *newPtr = '\0'; + mChunkPtr = newPtr; + mChunkLengths.back() += aStr.size(); + } + void CopyDataIntoLazilyAllocatedBuffer( + const std::function<char*(size_t)>& aAllocator) const { + // Request a buffer for the full content plus a null terminator. + MOZ_ASSERT(mChunkLengths.length() == mChunkList.length()); + size_t totalLen = 1; + for (size_t i = 0; i < mChunkLengths.length(); i++) { + MOZ_ASSERT(strlen(mChunkList[i].get()) == mChunkLengths[i]); + totalLen += mChunkLengths[i]; + } + char* ptr = aAllocator(totalLen); + + if (!ptr) { + // Failed to allocate memory. + return; + } + + for (size_t i = 0; i < mChunkList.length(); i++) { + size_t len = mChunkLengths[i]; + memcpy(ptr, mChunkList[i].get(), len); + ptr += len; + } + *ptr = '\0'; + } + UniquePtr<char[]> CopyData() const { + UniquePtr<char[]> c; + CopyDataIntoLazilyAllocatedBuffer([&](size_t allocationSize) { + c = MakeUnique<char[]>(allocationSize); + return c.get(); + }); + return c; + } + void Take(ChunkedJSONWriteFunc&& aOther) { + for (size_t i = 0; i < aOther.mChunkList.length(); i++) { + MOZ_ALWAYS_TRUE(mChunkLengths.append(aOther.mChunkLengths[i])); + MOZ_ALWAYS_TRUE(mChunkList.append(std::move(aOther.mChunkList[i]))); + } + mChunkPtr = mChunkList.back().get() + mChunkLengths.back(); + mChunkEnd = mChunkPtr; + aOther.mChunkPtr = nullptr; + aOther.mChunkEnd = nullptr; + aOther.mChunkList.clear(); + aOther.mChunkLengths.clear(); + } + + private: + void AllocChunk(size_t aChunkSize) { + MOZ_ASSERT(mChunkLengths.length() == mChunkList.length()); + UniquePtr<char[]> newChunk = MakeUnique<char[]>(aChunkSize); + mChunkPtr = newChunk.get(); + mChunkEnd = mChunkPtr + aChunkSize; + *mChunkPtr = '\0'; + MOZ_ALWAYS_TRUE(mChunkLengths.append(0)); + MOZ_ALWAYS_TRUE(mChunkList.append(std::move(newChunk))); + } + + static const size_t kChunkSize = 4096 * 512; + + // Pointer for writing inside the current chunk. + // + // The current chunk is always at the back of mChunkList, i.e., + // mChunkList.back() <= mChunkPtr <= mChunkEnd. + char* mChunkPtr; + + // Pointer to the end of the current chunk. + // + // The current chunk is always at the back of mChunkList, i.e., + // mChunkEnd >= mChunkList.back() + mChunkLengths.back(). + char* mChunkEnd; + + // List of chunks and their lengths. + // + // For all i, the length of the string in mChunkList[i] is + // mChunkLengths[i]. + Vector<UniquePtr<char[]>> mChunkList; + Vector<size_t> mChunkLengths; +}; + +struct OStreamJSONWriteFunc final : public JSONWriteFunc { + explicit OStreamJSONWriteFunc(std::ostream& aStream) : mStream(aStream) {} + + void Write(const Span<const char>& aStr) override { + std::string_view sv(aStr.data(), aStr.size()); + mStream << sv; + } + + std::ostream& mStream; +}; + +class UniqueJSONStrings; + +class SpliceableJSONWriter : public JSONWriter { + public: + explicit SpliceableJSONWriter(UniquePtr<JSONWriteFunc> aWriter) + : JSONWriter(std::move(aWriter)) {} + + void StartBareList(CollectionStyle aStyle = MultiLineStyle) { + StartCollection(scEmptyString, scEmptyString, aStyle); + } + + void EndBareList() { EndCollection(scEmptyString); } + + // This function must be used to correctly stream timestamps in profiles. + // Null timestamps don't output anything. + void TimeProperty(const Span<const char>& aName, const TimeStamp& aTime) { + if (!aTime.IsNull()) { + DoubleProperty(aName, + (aTime - TimeStamp::ProcessCreation()).ToMilliseconds()); + } + } + + void NullElements(uint32_t aCount) { + for (uint32_t i = 0; i < aCount; i++) { + NullElement(); + } + } + + void Splice(const Span<const char>& aStr) { + Separator(); + WriteFunc()->Write(aStr); + mNeedComma[mDepth] = true; + } + + void Splice(const char* aStr, size_t aLen) { + Separator(); + WriteFunc()->Write(Span<const char>(aStr, aLen)); + mNeedComma[mDepth] = true; + } + + // Splice the given JSON directly in, without quoting. + void SplicedJSONProperty(const Span<const char>& aMaybePropertyName, + const Span<const char>& aJsonValue) { + Scalar(aMaybePropertyName, aJsonValue); + } + + void CopyAndSplice(const ChunkedJSONWriteFunc& aFunc) { + Separator(); + for (size_t i = 0; i < aFunc.mChunkList.length(); i++) { + WriteFunc()->Write( + Span<const char>(aFunc.mChunkList[i].get(), aFunc.mChunkLengths[i])); + } + mNeedComma[mDepth] = true; + } + + // Takes the chunks from aFunc and write them. If move is not possible + // (e.g., using OStreamJSONWriteFunc), aFunc's chunks are copied and its + // storage cleared. + virtual void TakeAndSplice(ChunkedJSONWriteFunc&& aFunc) { + Separator(); + for (size_t i = 0; i < aFunc.mChunkList.length(); i++) { + WriteFunc()->Write( + Span<const char>(aFunc.mChunkList[i].get(), aFunc.mChunkLengths[i])); + } + aFunc.mChunkPtr = nullptr; + aFunc.mChunkEnd = nullptr; + aFunc.mChunkList.clear(); + aFunc.mChunkLengths.clear(); + mNeedComma[mDepth] = true; + } + + // Set (or reset) the pointer to a UniqueJSONStrings. + void SetUniqueStrings(UniqueJSONStrings& aUniqueStrings) { + MOZ_RELEASE_ASSERT(!mUniqueStrings); + mUniqueStrings = &aUniqueStrings; + } + + // Set (or reset) the pointer to a UniqueJSONStrings. + void ResetUniqueStrings() { + MOZ_RELEASE_ASSERT(mUniqueStrings); + mUniqueStrings = nullptr; + } + + // Add `aStr` to the unique-strings list (if not already there), and write its + // index as a named object property. + inline void UniqueStringProperty(const Span<const char>& aName, + const Span<const char>& aStr); + + // Add `aStr` to the unique-strings list (if not already there), and write its + // index as an array element. + inline void UniqueStringElement(const Span<const char>& aStr); + + private: + UniqueJSONStrings* mUniqueStrings = nullptr; +}; + +class SpliceableChunkedJSONWriter final : public SpliceableJSONWriter { + public: + explicit SpliceableChunkedJSONWriter() + : SpliceableJSONWriter(MakeUnique<ChunkedJSONWriteFunc>()) {} + + // Access the ChunkedJSONWriteFunc as reference-to-const, usually to copy data + // out. + const ChunkedJSONWriteFunc& ChunkedWriteFunc() const { + MOZ_ASSERT(!mTaken); + // The WriteFunc was non-fallibly allocated as a ChunkedJSONWriteFunc in the + // only constructor above, so it's safe to cast to ChunkedJSONWriteFunc*. + return *static_cast<const ChunkedJSONWriteFunc*>(WriteFunc()); + } + + // Access the ChunkedJSONWriteFunc as rvalue-reference, usually to take its + // data out. This writer shouldn't be used anymore after this. + ChunkedJSONWriteFunc&& TakeChunkedWriteFunc() { +#ifdef DEBUG + MOZ_ASSERT(!mTaken); + mTaken = true; +#endif // + // The WriteFunc was non-fallibly allocated as a ChunkedJSONWriteFunc in the + // only constructor above, so it's safe to cast to ChunkedJSONWriteFunc*. + return std::move(*static_cast<ChunkedJSONWriteFunc*>(WriteFunc())); + } + + // Adopts the chunks from aFunc without copying. + void TakeAndSplice(ChunkedJSONWriteFunc&& aFunc) override { + MOZ_ASSERT(!mTaken); + Separator(); + // The WriteFunc was non-fallibly allocated as a ChunkedJSONWriteFunc in the + // only constructor above, so it's safe to cast to ChunkedJSONWriteFunc*. + static_cast<ChunkedJSONWriteFunc*>(WriteFunc())->Take(std::move(aFunc)); + mNeedComma[mDepth] = true; + } + +#ifdef DEBUG + private: + bool mTaken = false; +#endif // +}; + +class JSONSchemaWriter { + JSONWriter& mWriter; + uint32_t mIndex; + + public: + explicit JSONSchemaWriter(JSONWriter& aWriter) : mWriter(aWriter), mIndex(0) { + aWriter.StartObjectProperty("schema", + SpliceableJSONWriter::SingleLineStyle); + } + + void WriteField(const Span<const char>& aName) { + mWriter.IntProperty(aName, mIndex++); + } + + template <size_t Np1> + void WriteField(const char (&aName)[Np1]) { + WriteField(Span<const char>(aName, Np1 - 1)); + } + + ~JSONSchemaWriter() { mWriter.EndObject(); } +}; + +// This class helps create an indexed list of unique strings, and inserts the +// index as a JSON value. The collected list of unique strings can later be +// inserted as a JSON array. +// This can be useful for elements/properties with many repeated strings. +// +// With only JSONWriter w, +// `w.WriteElement("a"); w.WriteElement("b"); w.WriteElement("a");` +// when done inside a JSON array, will generate: +// `["a", "b", "c"]` +// +// With UniqueStrings u, +// `u.WriteElement(w, "a"); u.WriteElement(w, "b"); u.WriteElement(w, "a");` +// when done inside a JSON array, will generate: +// `[0, 1, 0]` +// and later, `u.SpliceStringTableElements(w)` (inside a JSON array), will +// output the corresponding indexed list of unique strings: +// `["a", "b"]` +class UniqueJSONStrings { + public: + // Start an empty list of unique strings. + MFBT_API explicit UniqueJSONStrings( + JSONWriter::CollectionStyle aStyle = JSONWriter::MultiLineStyle); + + // Start with a copy of the strings from another list. + MFBT_API explicit UniqueJSONStrings( + const UniqueJSONStrings& aOther, + JSONWriter::CollectionStyle aStyle = JSONWriter::MultiLineStyle); + + MFBT_API ~UniqueJSONStrings(); + + // Add `aStr` to the list (if not already there), and write its index as a + // named object property. + void WriteProperty(JSONWriter& aWriter, const Span<const char>& aName, + const Span<const char>& aStr) { + aWriter.IntProperty(aName, GetOrAddIndex(aStr)); + } + + // Add `aStr` to the list (if not already there), and write its index as an + // array element. + void WriteElement(JSONWriter& aWriter, const Span<const char>& aStr) { + aWriter.IntElement(GetOrAddIndex(aStr)); + } + + // Splice all collected unique strings into an array. This should only be done + // once, and then this UniqueStrings shouldn't be used anymore. + MFBT_API void SpliceStringTableElements(SpliceableJSONWriter& aWriter); + + private: + // If `aStr` is already listed, return its index. + // Otherwise add it to the list and return the new index. + MFBT_API uint32_t GetOrAddIndex(const Span<const char>& aStr); + + SpliceableChunkedJSONWriter mStringTableWriter; + HashMap<HashNumber, uint32_t> mStringHashToIndexMap; +}; + +void SpliceableJSONWriter::UniqueStringProperty(const Span<const char>& aName, + const Span<const char>& aStr) { + MOZ_RELEASE_ASSERT(mUniqueStrings); + mUniqueStrings->WriteProperty(*this, aName, aStr); +} + +// Add `aStr` to the list (if not already there), and write its index as an +// array element. +void SpliceableJSONWriter::UniqueStringElement(const Span<const char>& aStr) { + MOZ_RELEASE_ASSERT(mUniqueStrings); + mUniqueStrings->WriteElement(*this, aStr); +} + +} // namespace baseprofiler +} // namespace mozilla + +#endif // BASEPROFILEJSONWRITER_H diff --git a/mozglue/baseprofiler/public/BaseProfiler.h b/mozglue/baseprofiler/public/BaseProfiler.h new file mode 100644 index 0000000000..4bf1705041 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfiler.h @@ -0,0 +1,964 @@ +/* -*- 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/. */ + +// The Gecko Profiler is an always-on profiler that takes fast and low overhead +// samples of the program execution using only userspace functionality for +// portability. The goal of this module is to provide performance data in a +// generic cross-platform way without requiring custom tools or kernel support. +// +// Samples are collected to form a timeline with optional timeline event +// (markers) used for filtering. The samples include both native stacks and +// platform-independent "label stack" frames. + +#ifndef BaseProfiler_h +#define BaseProfiler_h + +// This file is safe to include unconditionally, and only defines +// empty macros if MOZ_GECKO_PROFILER is not set. + +// These headers are also safe to include unconditionally, with empty macros if +// MOZ_GECKO_PROFILER is not set. +#include "mozilla/BaseProfilerCounts.h" + +// BaseProfilerMarkers.h is #included in the middle of this header! +// #include "mozilla/BaseProfilerMarkers.h" + +#ifndef MOZ_GECKO_PROFILER + +# include "mozilla/BaseProfilerMarkers.h" +# include "mozilla/UniquePtr.h" + +// This file can be #included unconditionally. However, everything within this +// file must be guarded by a #ifdef MOZ_GECKO_PROFILER, *except* for the +// following macros and functions, which encapsulate the most common operations +// and thus avoid the need for many #ifdefs. + +# define AUTO_BASE_PROFILER_INIT + +# define BASE_PROFILER_REGISTER_THREAD(name) +# define BASE_PROFILER_UNREGISTER_THREAD() +# define AUTO_BASE_PROFILER_REGISTER_THREAD(name) + +# define AUTO_BASE_PROFILER_THREAD_SLEEP +# define AUTO_BASE_PROFILER_THREAD_WAKE + +# define AUTO_BASE_PROFILER_LABEL(label, categoryPair) +# define AUTO_BASE_PROFILER_LABEL_CATEGORY_PAIR(categoryPair) +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_CSTR(label, categoryPair, cStr) +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_STRING(label, categoryPair, str) +# define AUTO_BASE_PROFILER_LABEL_FAST(label, categoryPair, ctx) +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_FAST(label, dynamicString, \ + categoryPair, ctx, flags) + +# define AUTO_PROFILER_STATS(name) + +// Function stubs for when MOZ_GECKO_PROFILER is not defined. + +namespace mozilla { +// This won't be used, it's just there to allow the empty definition of +// `profiler_capture_backtrace`. +class ProfileChunkedBuffer {}; + +namespace baseprofiler { +// This won't be used, it's just there to allow the empty definition of +// `profiler_get_backtrace`. +struct ProfilerBacktrace {}; +using UniqueProfilerBacktrace = UniquePtr<ProfilerBacktrace>; + +// Get/Capture-backtrace functions can return nullptr or false, the result +// should be fed to another empty macro or stub anyway. + +static inline UniqueProfilerBacktrace profiler_get_backtrace() { + return nullptr; +} + +static inline bool profiler_capture_backtrace_into( + ProfileChunkedBuffer& aChunkedBuffer) { + return false; +} + +static inline UniquePtr<ProfileChunkedBuffer> profiler_capture_backtrace() { + return nullptr; +} +} // namespace baseprofiler +} // namespace mozilla + +#else // !MOZ_GECKO_PROFILER + +# include "BaseProfilingStack.h" + +# include "mozilla/Assertions.h" +# include "mozilla/Atomics.h" +# include "mozilla/Attributes.h" +# include "mozilla/Maybe.h" +# include "mozilla/PowerOfTwo.h" +# include "mozilla/Sprintf.h" +# include "mozilla/ThreadLocal.h" +# include "mozilla/TimeStamp.h" +# include "mozilla/UniquePtr.h" + +# include <functional> +# include <stdint.h> +# include <string> + +namespace mozilla { + +class MallocAllocPolicy; +class ProfileChunkedBuffer; +template <class T, size_t MinInlineCapacity, class AllocPolicy> +class Vector; + +namespace baseprofiler { + +class ProfilerBacktrace; +class SpliceableJSONWriter; + +// Macros used by the AUTO_PROFILER_* macros below. +# define BASE_PROFILER_RAII_PASTE(id, line) id##line +# define BASE_PROFILER_RAII_EXPAND(id, line) BASE_PROFILER_RAII_PASTE(id, line) +# define BASE_PROFILER_RAII BASE_PROFILER_RAII_EXPAND(raiiObject, __LINE__) + +//--------------------------------------------------------------------------- +// Profiler features +//--------------------------------------------------------------------------- + +// Higher-order macro containing all the feature info in one place. Define +// |MACRO| appropriately to extract the relevant parts. Note that the number +// values are used internally only and so can be changed without consequence. +// Any changes to this list should also be applied to the feature list in +// toolkit/components/extensions/schemas/geckoProfiler.json. +# define BASE_PROFILER_FOR_EACH_FEATURE(MACRO) \ + MACRO(0, "java", Java, "Profile Java code, Android only") \ + \ + MACRO(1, "js", JS, \ + "Get the JS engine to expose the JS stack to the profiler") \ + \ + /* The DevTools profiler doesn't want the native addresses. */ \ + MACRO(2, "leaf", Leaf, "Include the C++ leaf node if not stackwalking") \ + \ + MACRO(3, "mainthreadio", MainThreadIO, "Add main thread file I/O") \ + \ + MACRO(4, "fileio", FileIO, \ + "Add file I/O from all profiled threads, implies mainthreadio") \ + \ + MACRO(5, "fileioall", FileIOAll, \ + "Add file I/O from all threads, implies fileio") \ + \ + MACRO(6, "noiostacks", NoIOStacks, \ + "File I/O markers do not capture stacks, to reduce overhead") \ + \ + MACRO(7, "screenshots", Screenshots, \ + "Take a snapshot of the window on every composition") \ + \ + MACRO(8, "seqstyle", SequentialStyle, \ + "Disable parallel traversal in styling") \ + \ + MACRO(9, "stackwalk", StackWalk, \ + "Walk the C++ stack, not available on all platforms") \ + \ + MACRO(10, "tasktracer", TaskTracer, \ + "Start profiling with feature TaskTracer") \ + \ + MACRO(11, "threads", Threads, "Profile the registered secondary threads") \ + \ + MACRO(12, "jstracer", JSTracer, "Enable tracing of the JavaScript engine") \ + \ + MACRO(13, "jsallocations", JSAllocations, \ + "Have the JavaScript engine track allocations") \ + \ + MACRO(14, "nostacksampling", NoStackSampling, \ + "Disable all stack sampling: Cancels \"js\", \"leaf\", " \ + "\"stackwalk\" and labels") \ + \ + MACRO(15, "preferencereads", PreferenceReads, \ + "Track when preferences are read") \ + \ + MACRO(16, "nativeallocations", NativeAllocations, \ + "Collect the stacks from a smaller subset of all native " \ + "allocations, biasing towards collecting larger allocations") \ + \ + MACRO(17, "ipcmessages", IPCMessages, \ + "Have the IPC layer track cross-process messages") \ + \ + MACRO(18, "audiocallbacktracing", AudioCallbackTracing, \ + "Audio callback tracing") \ + \ + MACRO(19, "cpu", CPUUtilization, "CPU utilization") + +struct ProfilerFeature { +# define DECLARE(n_, str_, Name_, desc_) \ + static constexpr uint32_t Name_ = (1u << n_); \ + static constexpr bool Has##Name_(uint32_t aFeatures) { \ + return aFeatures & Name_; \ + } \ + static constexpr void Set##Name_(uint32_t& aFeatures) { \ + aFeatures |= Name_; \ + } \ + static constexpr void Clear##Name_(uint32_t& aFeatures) { \ + aFeatures &= ~Name_; \ + } + + // Define a bitfield constant, a getter, and two setters for each feature. + BASE_PROFILER_FOR_EACH_FEATURE(DECLARE) + +# undef DECLARE +}; + +namespace detail { + +// RacyFeatures is only defined in this header file so that its methods can +// be inlined into profiler_is_active(). Please do not use anything from the +// detail namespace outside the profiler. + +// Within the profiler's code, the preferred way to check profiler activeness +// and features is via ActivePS(). However, that requires locking gPSMutex. +// There are some hot operations where absolute precision isn't required, so we +// duplicate the activeness/feature state in a lock-free manner in this class. +class RacyFeatures { + public: + MFBT_API static void SetActive(uint32_t aFeatures); + + MFBT_API static void SetInactive(); + + MFBT_API static void SetPaused(); + + MFBT_API static void SetUnpaused(); + + MFBT_API static void SetSamplingPaused(); + + MFBT_API static void SetSamplingUnpaused(); + + MFBT_API static bool IsActive(); + + MFBT_API static bool IsActiveWithFeature(uint32_t aFeature); + + // True if profiler is active, and not fully paused. + // Note that periodic sampling *could* be paused! + MFBT_API static bool IsActiveAndUnpaused(); + + // True if profiler is active, and sampling is not paused (though generic + // `SetPaused()` or specific `SetSamplingPaused()`). + MFBT_API static bool IsActiveAndSamplingUnpaused(); + + private: + static constexpr uint32_t Active = 1u << 31; + static constexpr uint32_t Paused = 1u << 30; + static constexpr uint32_t SamplingPaused = 1u << 29; + +// Ensure Active/Paused don't overlap with any of the feature bits. +# define NO_OVERLAP(n_, str_, Name_, desc_) \ + static_assert(ProfilerFeature::Name_ != SamplingPaused, \ + "bad feature value"); + + BASE_PROFILER_FOR_EACH_FEATURE(NO_OVERLAP); + +# undef NO_OVERLAP + + // We combine the active bit with the feature bits so they can be read or + // written in a single atomic operation. + // TODO: Could this be MFBT_DATA for better inlining optimization? + static Atomic<uint32_t, MemoryOrdering::Relaxed> sActiveAndFeatures; +}; + +MFBT_API bool IsThreadBeingProfiled(); + +} // namespace detail + +//--------------------------------------------------------------------------- +// Start and stop the profiler +//--------------------------------------------------------------------------- + +static constexpr PowerOfTwo32 BASE_PROFILER_DEFAULT_ENTRIES = +# if !defined(GP_PLAT_arm_android) + MakePowerOfTwo32<1024 * 1024>(); // 1M entries = 8MB +# else + MakePowerOfTwo32<128 * 1024>(); // 128k entries = 1MB +# endif + +// Startup profiling usually need to capture more data, especially on slow +// systems. +static constexpr PowerOfTwo32 BASE_PROFILER_DEFAULT_STARTUP_ENTRIES = +# if !defined(GP_PLAT_arm_android) + MakePowerOfTwo32<4 * 1024 * 1024>(); // 4M entries = 32MB +# else + MakePowerOfTwo32<256 * 1024>(); // 256k entries = 2MB +# endif + +# define BASE_PROFILER_DEFAULT_DURATION 20 +# define BASE_PROFILER_DEFAULT_INTERVAL 1 + +// Initialize the profiler. If MOZ_PROFILER_STARTUP is set the profiler will +// also be started. This call must happen before any other profiler calls +// (except profiler_start(), which will call profiler_init() if it hasn't +// already run). +MFBT_API void profiler_init(void* stackTop); + +# define AUTO_BASE_PROFILER_INIT \ + ::mozilla::baseprofiler::AutoProfilerInit BASE_PROFILER_RAII + +// Clean up the profiler module, stopping it if required. This function may +// also save a shutdown profile if requested. No profiler calls should happen +// after this point and all profiling stack labels should have been popped. +MFBT_API void profiler_shutdown(); + +// Start the profiler -- initializing it first if necessary -- with the +// selected options. Stops and restarts the profiler if it is already active. +// After starting the profiler is "active". The samples will be recorded in a +// circular buffer. +// "aCapacity" is the maximum number of 8-byte entries in the profiler's +// circular buffer. +// "aInterval" the sampling interval, measured in millseconds. +// "aFeatures" is the feature set. Features unsupported by this +// platform/configuration are ignored. +// "aFilters" is the list of thread filters. Threads that do not match any +// of the filters are not profiled. A filter matches a thread if +// (a) the thread name contains the filter as a case-insensitive +// substring, or +// (b) the filter is of the form "pid:<n>" where n is the process +// id of the process that the thread is running in. +// "aDuration" is the duration of entries in the profiler's circular buffer. +MFBT_API void profiler_start(PowerOfTwo32 aCapacity, double aInterval, + uint32_t aFeatures, const char** aFilters, + uint32_t aFilterCount, + const Maybe<double>& aDuration = Nothing()); + +// Stop the profiler and discard the profile without saving it. A no-op if the +// profiler is inactive. After stopping the profiler is "inactive". +MFBT_API void profiler_stop(); + +// If the profiler is inactive, start it. If it's already active, restart it if +// the requested settings differ from the current settings. Both the check and +// the state change are performed while the profiler state is locked. +// The only difference to profiler_start is that the current buffer contents are +// not discarded if the profiler is already running with the requested settings. +MFBT_API void profiler_ensure_started( + PowerOfTwo32 aCapacity, double aInterval, uint32_t aFeatures, + const char** aFilters, uint32_t aFilterCount, + const Maybe<double>& aDuration = Nothing()); + +//--------------------------------------------------------------------------- +// Control the profiler +//--------------------------------------------------------------------------- + +// Register/unregister threads with the profiler. Both functions operate the +// same whether the profiler is active or inactive. +# define BASE_PROFILER_REGISTER_THREAD(name) \ + do { \ + char stackTop; \ + ::mozilla::baseprofiler::profiler_register_thread(name, &stackTop); \ + } while (0) +# define BASE_PROFILER_UNREGISTER_THREAD() \ + ::mozilla::baseprofiler::profiler_unregister_thread() +MFBT_API ProfilingStack* profiler_register_thread(const char* name, + void* guessStackTop); +MFBT_API void profiler_unregister_thread(); + +// Registers a DOM Window (the JS global `window`) with the profiler. Each +// Window _roughly_ corresponds to a single document loaded within a +// BrowsingContext. The unique IDs for both the Window and BrowsingContext are +// recorded to allow correlating different Windows loaded within the same tab or +// frame element. +// +// We register pages for each navigations but we do not register +// history.pushState or history.replaceState since they correspond to the same +// Inner Window ID. 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. +// +// "aBrowsingContextID" is the ID of the browsing context that document +// belongs to. That's used to determine the tab of +// that page. +// "aInnerWindowID" is the ID of the `window` global object of that +// document. +// "aUrl" is the URL of the page. +// "aEmbedderInnerWindowID" is the inner window id of embedder. It's used to +// determine sub documents of a page. +MFBT_API void profiler_register_page(uint64_t aBrowsingContextID, + uint64_t aInnerWindowID, + const std::string& aUrl, + uint64_t aEmbedderInnerWindowID); +// Unregister page with the profiler. +// +// Take a Inner Window ID and unregister the page entry that has the same ID. +MFBT_API void profiler_unregister_page(uint64_t aRegisteredInnerWindowID); + +// Remove all registered and unregistered pages in the profiler. +void profiler_clear_all_pages(); + +class BaseProfilerCount; +MFBT_API void profiler_add_sampled_counter(BaseProfilerCount* aCounter); +MFBT_API void profiler_remove_sampled_counter(BaseProfilerCount* aCounter); + +// Register and unregister a thread within a scope. +# define AUTO_BASE_PROFILER_REGISTER_THREAD(name) \ + ::mozilla::baseprofiler::AutoProfilerRegisterThread BASE_PROFILER_RAII(name) + +// Pause and resume the profiler. No-ops if the profiler is inactive. While +// paused the profile will not take any samples and will not record any data +// into its buffers. The profiler remains fully initialized in this state. +// This feature will keep JavaScript profiling enabled, thus allowing toggling +// the profiler without invalidating the JIT. +MFBT_API void profiler_pause(); +MFBT_API void profiler_resume(); + +// Only pause and resume the periodic sampling loop, including stack sampling, +// counters, and profiling overheads. +MFBT_API void profiler_pause_sampling(); +MFBT_API void profiler_resume_sampling(); + +// These functions tell the profiler that a thread went to sleep so that we can +// avoid sampling it while it's sleeping. Calling profiler_thread_sleep() +// twice without an intervening profiler_thread_wake() is an error. All three +// functions operate the same whether the profiler is active or inactive. +MFBT_API void profiler_thread_sleep(); +MFBT_API void profiler_thread_wake(); + +// Mark a thread as asleep/awake within a scope. +# define AUTO_BASE_PROFILER_THREAD_SLEEP \ + ::mozilla::baseprofiler::AutoProfilerThreadSleep BASE_PROFILER_RAII +# define AUTO_BASE_PROFILER_THREAD_WAKE \ + ::mozilla::baseprofiler::AutoProfilerThreadWake BASE_PROFILER_RAII + +//--------------------------------------------------------------------------- +// Get information from the profiler +//--------------------------------------------------------------------------- + +// Is the profiler active? Note: the return value of this function can become +// immediately out-of-date. E.g. the profile might be active but then +// profiler_stop() is called immediately afterward. One common and reasonable +// pattern of usage is the following: +// +// if (profiler_is_active()) { +// ExpensiveData expensiveData = CreateExpensiveData(); +// PROFILER_OPERATION(expensiveData); +// } +// +// where PROFILER_OPERATION is a no-op if the profiler is inactive. In this +// case the profiler_is_active() check is just an optimization -- it prevents +// us calling CreateExpensiveData() unnecessarily in most cases, but the +// expensive data will end up being created but not used if another thread +// stops the profiler between the CreateExpensiveData() and PROFILER_OPERATION +// calls. +inline bool profiler_is_active() { + return baseprofiler::detail::RacyFeatures::IsActive(); +} + +// Same as profiler_is_active(), but with the same extra checks that determine +// if the profiler would currently store markers. So this should be used before +// doing some potentially-expensive work that's used in a marker. E.g.: +// +// if (profiler_can_accept_markers()) { +// BASE_PROFILER_MARKER(name, OTHER, SomeMarkerType, expensivePayload); +// } +inline bool profiler_can_accept_markers() { + return baseprofiler::detail::RacyFeatures::IsActiveAndUnpaused(); +} + +// Is the profiler active, and is the current thread being profiled? +// (Same caveats and recommented usage as profiler_is_active().) +inline bool profiler_thread_is_being_profiled() { + return profiler_is_active() && baseprofiler::detail::IsThreadBeingProfiled(); +} + +// Is the profiler active and paused? Returns false if the profiler is inactive. +MFBT_API bool profiler_is_paused(); + +// Is the profiler active and sampling is paused? Returns false if the profiler +// is inactive. +MFBT_API bool profiler_is_sampling_paused(); + +// Is the current thread sleeping? +MFBT_API bool profiler_thread_is_sleeping(); + +// Get all the features supported by the profiler that are accepted by +// profiler_start(). The result is the same whether the profiler is active or +// not. +MFBT_API uint32_t profiler_get_available_features(); + +// Check if a profiler feature (specified via the ProfilerFeature type) is +// active. Returns false if the profiler is inactive. Note: the return value +// can become immediately out-of-date, much like the return value of +// profiler_is_active(). +MFBT_API bool profiler_feature_active(uint32_t aFeature); + +// Get the params used to start the profiler. Returns 0 and an empty vector +// (via outparams) if the profile is inactive. It's possible that the features +// returned may be slightly different to those requested due to required +// adjustments. +MFBT_API void profiler_get_start_params( + int* aEntrySize, Maybe<double>* aDuration, double* aInterval, + uint32_t* aFeatures, Vector<const char*, 0, MallocAllocPolicy>* aFilters); + +// The number of milliseconds since the process started. Operates the same +// whether the profiler is active or inactive. +MFBT_API double profiler_time(); + +// Get the current process's ID. +MFBT_API int profiler_current_process_id(); + +// Get the current thread's ID. +MFBT_API int profiler_current_thread_id(); + +// Statically initialized to 0, then set once from profiler_init(), which should +// be called from the main thread before any other use of the profiler. +extern MFBT_DATA int scProfilerMainThreadId; + +inline int profiler_main_thread_id() { return scProfilerMainThreadId; } + +inline bool profiler_is_main_thread() { + return profiler_current_thread_id() == profiler_main_thread_id(); +} + +// An object of this class is passed to profiler_suspend_and_sample_thread(). +// For each stack frame, one of the Collect methods will be called. +class ProfilerStackCollector { + public: + // Some collectors need to worry about possibly overwriting previous + // generations of data. If that's not an issue, this can return Nothing, + // which is the default behaviour. + virtual Maybe<uint64_t> SamplePositionInBuffer() { return Nothing(); } + virtual Maybe<uint64_t> BufferRangeStart() { return Nothing(); } + + // This method will be called once if the thread being suspended is the main + // thread. Default behaviour is to do nothing. + virtual void SetIsMainThread() {} + + // WARNING: The target thread is suspended when the Collect methods are + // called. Do not try to allocate or acquire any locks, or you could + // deadlock. The target thread will have resumed by the time this function + // returns. + + virtual void CollectNativeLeafAddr(void* aAddr) = 0; + + virtual void CollectProfilingStackFrame( + const ProfilingStackFrame& aFrame) = 0; +}; + +// This method suspends the thread identified by aThreadId, samples its +// profiling stack, JS stack, and (optionally) native stack, passing the +// collected frames into aCollector. aFeatures dictates which compiler features +// are used. |Leaf| is the only relevant one. +MFBT_API void profiler_suspend_and_sample_thread( + int aThreadId, uint32_t aFeatures, ProfilerStackCollector& aCollector, + bool aSampleNative = true); + +struct ProfilerBacktraceDestructor { + MFBT_API void operator()(ProfilerBacktrace*); +}; + +using UniqueProfilerBacktrace = + UniquePtr<ProfilerBacktrace, ProfilerBacktraceDestructor>; + +// Immediately capture the current thread's call stack, store it in the provided +// buffer (usually to avoid allocations if you can construct the buffer on the +// stack). Returns false if unsuccessful, or if the profiler is inactive. +MFBT_API bool profiler_capture_backtrace_into( + ProfileChunkedBuffer& aChunkedBuffer); + +// Immediately capture the current thread's call stack, and return it in a +// ProfileChunkedBuffer (usually for later use in MarkerStack::TakeBacktrace()). +// May be null if unsuccessful, or if the profiler is inactive. +MFBT_API UniquePtr<ProfileChunkedBuffer> profiler_capture_backtrace(); + +// Immediately capture the current thread's call stack, and return it in a +// ProfilerBacktrace (usually for later use in marker function that take a +// ProfilerBacktrace). May be null if unsuccessful, or if the profiler is +// inactive. +MFBT_API UniqueProfilerBacktrace profiler_get_backtrace(); + +struct ProfilerStats { + unsigned n = 0; + double sum = 0; + double min = std::numeric_limits<double>::max(); + double max = 0; + void Count(double v) { + ++n; + sum += v; + if (v < min) { + min = v; + } + if (v > max) { + max = v; + } + } +}; + +struct ProfilerBufferInfo { + // Index of the oldest entry. + uint64_t mRangeStart; + // Index of the newest entry. + uint64_t mRangeEnd; + // Buffer capacity in number of 8-byte entries. + uint32_t mEntryCount; + // 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; +}; + +// Get information about the current buffer status. +// Returns Nothing() if the profiler is inactive. +// +// This information may be useful to a user-interface displaying the current +// status of the profiler, allowing the user to get a sense for how fast the +// buffer is being written to, and how much data is visible. +MFBT_API Maybe<ProfilerBufferInfo> profiler_get_buffer_info(); + +// Uncomment the following line to display profiler runtime statistics at +// shutdown. +// # define PROFILER_RUNTIME_STATS + +# ifdef PROFILER_RUNTIME_STATS +// This class gathers durations and displays some basic stats when destroyed. +// It is intended to be used as a static variable (see `AUTO_PROFILER_STATS` +// below), to display stats at the end of the program. +class StaticBaseProfilerStats { + public: + explicit StaticBaseProfilerStats(const char* aName) : mName(aName) {} + + ~StaticBaseProfilerStats() { + // Using unsigned long long for computations and printfs. + using ULL = unsigned long long; + ULL n = static_cast<ULL>(mNumberDurations); + if (n != 0) { + ULL sumNs = static_cast<ULL>(mSumDurationsNs); + printf( + "[%d] Profiler stats `%s`: %llu ns / %llu = %llu ns, max %llu ns\n", + profiler_current_process_id(), mName, sumNs, n, sumNs / n, + static_cast<ULL>(mLongestDurationNs)); + } else { + printf("[%d] Profiler stats `%s`: (nothing)\n", + profiler_current_process_id(), mName); + } + } + + void AddDurationFrom(TimeStamp aStart) { + DurationNs duration = static_cast<DurationNs>( + (TimeStamp::NowUnfuzzed() - aStart).ToMicroseconds() * 1000 + 0.5); + mSumDurationsNs += duration; + ++mNumberDurations; + // Update mLongestDurationNs if this one is longer. + for (;;) { + DurationNs longest = mLongestDurationNs; + if (MOZ_LIKELY(longest >= duration)) { + // This duration is not the longest, nothing to do. + break; + } + if (MOZ_LIKELY(mLongestDurationNs.compareExchange(longest, duration))) { + // Successfully updated `mLongestDurationNs` with the new value. + break; + } + // Otherwise someone else just updated `mLongestDurationNs`, we need to + // try again by looping. + } + } + + private: + using DurationNs = uint64_t; + using Count = uint32_t; + + Atomic<DurationNs> mSumDurationsNs{0}; + Atomic<DurationNs> mLongestDurationNs{0}; + Atomic<Count> mNumberDurations{0}; + const char* mName; +}; + +// RAII object that measure its scoped lifetime duration and reports it to a +// `StaticBaseProfilerStats`. +class MOZ_RAII AutoProfilerStats { + public: + explicit AutoProfilerStats(StaticBaseProfilerStats& aStats) + : mStats(aStats), mStart(TimeStamp::NowUnfuzzed()) {} + + ~AutoProfilerStats() { mStats.AddDurationFrom(mStart); } + + private: + StaticBaseProfilerStats& mStats; + TimeStamp mStart; +}; + +// Macro that should be used to collect basic statistics from measurements of +// block durations, from where this macro is, until the end of its enclosing +// scope. The name is used in the static variable name and when displaying stats +// at the end of the program; Another location could use the same name but their +// stats will not be combined, so use different name if these locations should +// be distinguished. +# define AUTO_PROFILER_STATS(name) \ + static ::mozilla::baseprofiler::StaticBaseProfilerStats sStat##name( \ + #name); \ + ::mozilla::baseprofiler::AutoProfilerStats autoStat##name(sStat##name); + +# else // PROFILER_RUNTIME_STATS + +# define AUTO_PROFILER_STATS(name) + +# endif // PROFILER_RUNTIME_STATS else + +} // namespace baseprofiler +} // namespace mozilla + +// BaseProfilerMarkers.h requires some stuff from this header. +// TODO: Move common stuff to shared header, and move this #include to the top. +# include "mozilla/BaseProfilerMarkers.h" + +namespace mozilla { +namespace baseprofiler { + +//--------------------------------------------------------------------------- +// Put profiling data into the profiler (labels and markers) +//--------------------------------------------------------------------------- + +// Insert an RAII object in this scope to enter a label stack frame. Any +// samples collected in this scope will contain this label in their stack. +// The label argument must be a static C string. It is usually of the +// form "ClassName::FunctionName". (Ideally we'd use the compiler to provide +// that for us, but __func__ gives us the function name without the class +// name.) If the label applies to only part of a function, you can qualify it +// like this: "ClassName::FunctionName:PartName". +// +// Use AUTO_BASE_PROFILER_LABEL_DYNAMIC_* if you want to add additional / +// dynamic information to the label stack frame. +# define AUTO_BASE_PROFILER_LABEL(label, categoryPair) \ + ::mozilla::baseprofiler::AutoProfilerLabel BASE_PROFILER_RAII( \ + label, nullptr, \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair) + +// Similar to AUTO_BASE_PROFILER_LABEL, but with only one argument: the category +// pair. The label string is taken from the category pair. This is convenient +// for labels like +// AUTO_BASE_PROFILER_LABEL_CATEGORY_PAIR(GRAPHICS_LayerBuilding) which would +// otherwise just repeat the string. +# define AUTO_BASE_PROFILER_LABEL_CATEGORY_PAIR(categoryPair) \ + ::mozilla::baseprofiler::AutoProfilerLabel BASE_PROFILER_RAII( \ + "", nullptr, \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair, \ + uint32_t(::mozilla::baseprofiler::ProfilingStackFrame::Flags:: \ + LABEL_DETERMINED_BY_CATEGORY_PAIR)) + +// Similar to AUTO_BASE_PROFILER_LABEL, but with an additional string. The +// inserted RAII object stores the cStr pointer in a field; it does not copy the +// string. +// +// WARNING: This means that the string you pass to this macro needs to live at +// least until the end of the current scope. Be careful using this macro with +// ns[C]String; the other AUTO_BASE_PROFILER_LABEL_DYNAMIC_* macros below are +// preferred because they avoid this problem. +// +// If the profiler samples the current thread and walks the label stack while +// this RAII object is on the stack, it will copy the supplied string into the +// profile buffer. So there's one string copy operation, and it happens at +// sample time. +// +// Compare this to the plain AUTO_BASE_PROFILER_LABEL macro, which only accepts +// literal strings: When the label stack frames generated by +// AUTO_BASE_PROFILER_LABEL are sampled, no string copy needs to be made because +// the profile buffer can just store the raw pointers to the literal strings. +// Consequently, AUTO_BASE_PROFILER_LABEL frames take up considerably less space +// in the profile buffer than AUTO_BASE_PROFILER_LABEL_DYNAMIC_* frames. +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_CSTR(label, categoryPair, cStr) \ + ::mozilla::baseprofiler::AutoProfilerLabel BASE_PROFILER_RAII( \ + label, cStr, \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair) + +// Similar to AUTO_BASE_PROFILER_LABEL_DYNAMIC_CSTR, but takes an std::string. +// +// Note: The use of the Maybe<>s ensures the scopes for the dynamic string and +// the AutoProfilerLabel are appropriate, while also not incurring the runtime +// cost of the string assignment unless the profiler is active. Therefore, +// unlike AUTO_BASE_PROFILER_LABEL and AUTO_BASE_PROFILER_LABEL_DYNAMIC_CSTR, +// this macro doesn't push/pop a label when the profiler is inactive. +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_STRING(label, categoryPair, str) \ + Maybe<std::string> autoStr; \ + Maybe<::mozilla::baseprofiler::AutoProfilerLabel> raiiObjectString; \ + if (::mozilla::baseprofiler::profiler_is_active()) { \ + autoStr.emplace(str); \ + raiiObjectString.emplace( \ + label, autoStr->c_str(), \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair); \ + } + +// Similar to AUTO_BASE_PROFILER_LABEL, but accepting a JSContext* parameter, +// and a no-op if the profiler is disabled. Used to annotate functions for which +// overhead in the range of nanoseconds is noticeable. It avoids overhead from +// the TLS lookup because it can get the ProfilingStack from the JS context, and +// avoids almost all overhead in the case where the profiler is disabled. +# define AUTO_BASE_PROFILER_LABEL_FAST(label, categoryPair, ctx) \ + ::mozilla::baseprofiler::AutoProfilerLabel BASE_PROFILER_RAII( \ + ctx, label, nullptr, \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair) + +// Similar to AUTO_BASE_PROFILER_LABEL_FAST, but also takes an extra string and +// an additional set of flags. The flags parameter should carry values from the +// ProfilingStackFrame::Flags enum. +# define AUTO_BASE_PROFILER_LABEL_DYNAMIC_FAST(label, dynamicString, \ + categoryPair, ctx, flags) \ + ::mozilla::baseprofiler::AutoProfilerLabel BASE_PROFILER_RAII( \ + ctx, label, dynamicString, \ + ::mozilla::baseprofiler::ProfilingCategoryPair::categoryPair, flags) + +MFBT_API void profiler_add_js_marker(const char* aMarkerName, + const char* aMarkerText); + +// Returns true if any of the profiler mutexes are currently locked *on the +// current thread*. This may be used by re-entrant code that may call profiler +// functions while the same of a different profiler mutex is locked, which could +// deadlock. +bool profiler_is_locked_on_current_thread(); + +//--------------------------------------------------------------------------- +// Output profiles +//--------------------------------------------------------------------------- + +// Set a user-friendly process name, used in JSON stream. +MFBT_API void profiler_set_process_name(const std::string& aProcessName, + const std::string* aETLDplus1); + +// Get the profile encoded as a JSON string. A no-op (returning nullptr) if the +// profiler is inactive. +// If aIsShuttingDown is true, the current time is included as the process +// shutdown time in the JSON's "meta" object. +MFBT_API UniquePtr<char[]> profiler_get_profile(double aSinceTime = 0, + bool aIsShuttingDown = false, + bool aOnlyThreads = false); + +// Write the profile for this process (excluding subprocesses) into aWriter. +// Returns false if the profiler is inactive. +MFBT_API bool profiler_stream_json_for_this_process( + SpliceableJSONWriter& aWriter, double aSinceTime = 0, + bool aIsShuttingDown = false, bool aOnlyThreads = false); + +// Get the profile and write it into a file. A no-op if the profile is +// inactive. +MFBT_API void profiler_save_profile_to_file(const char* aFilename); + +//--------------------------------------------------------------------------- +// RAII classes +//--------------------------------------------------------------------------- + +class MOZ_RAII AutoProfilerInit { + public: + explicit AutoProfilerInit() { profiler_init(this); } + + ~AutoProfilerInit() { profiler_shutdown(); } + + private: +}; + +// Convenience class to register and unregister a thread with the profiler. +// Needs to be the first object on the stack of the thread. +class MOZ_RAII AutoProfilerRegisterThread final { + public: + explicit AutoProfilerRegisterThread(const char* aName) { + profiler_register_thread(aName, this); + } + + ~AutoProfilerRegisterThread() { profiler_unregister_thread(); } + + private: + AutoProfilerRegisterThread(const AutoProfilerRegisterThread&) = delete; + AutoProfilerRegisterThread& operator=(const AutoProfilerRegisterThread&) = + delete; +}; + +class MOZ_RAII AutoProfilerThreadSleep { + public: + explicit AutoProfilerThreadSleep() { profiler_thread_sleep(); } + + ~AutoProfilerThreadSleep() { profiler_thread_wake(); } + + private: +}; + +// Temporarily wake up the profiling of a thread while servicing events such as +// Asynchronous Procedure Calls (APCs). +class MOZ_RAII AutoProfilerThreadWake { + public: + explicit AutoProfilerThreadWake() + : mIssuedWake(profiler_thread_is_sleeping()) { + if (mIssuedWake) { + profiler_thread_wake(); + } + } + + ~AutoProfilerThreadWake() { + if (mIssuedWake) { + MOZ_ASSERT(!profiler_thread_is_sleeping()); + profiler_thread_sleep(); + } + } + + private: + bool mIssuedWake; +}; + +// This class creates a non-owning ProfilingStack reference. Objects of this +// class are stack-allocated, and so exist within a thread, and are thus bounded +// by the lifetime of the thread, which ensures that the references held can't +// be used after the ProfilingStack is destroyed. +class MOZ_RAII AutoProfilerLabel { + public: + // This is the AUTO_BASE_PROFILER_LABEL and AUTO_BASE_PROFILER_LABEL_DYNAMIC + // variant. + AutoProfilerLabel(const char* aLabel, const char* aDynamicString, + ProfilingCategoryPair aCategoryPair, uint32_t aFlags = 0) { + // Get the ProfilingStack from TLS. + Push(GetProfilingStack(), aLabel, aDynamicString, aCategoryPair, aFlags); + } + + void Push(ProfilingStack* aProfilingStack, const char* aLabel, + const char* aDynamicString, ProfilingCategoryPair aCategoryPair, + uint32_t aFlags = 0) { + // This function runs both on and off the main thread. + + mProfilingStack = aProfilingStack; + if (mProfilingStack) { + mProfilingStack->pushLabelFrame(aLabel, aDynamicString, this, + aCategoryPair, aFlags); + } + } + + ~AutoProfilerLabel() { + // This function runs both on and off the main thread. + + if (mProfilingStack) { + mProfilingStack->pop(); + } + } + + MFBT_API static ProfilingStack* GetProfilingStack(); + + private: + // We save a ProfilingStack pointer in the ctor so we don't have to redo the + // TLS lookup in the dtor. + ProfilingStack* mProfilingStack; + + public: + // See the comment on the definition in platform.cpp for details about this. + static MOZ_THREAD_LOCAL(ProfilingStack*) sProfilingStack; +}; + +// Get the MOZ_PROFILER_STARTUP* environment variables that should be +// supplied to a child process that is about to be launched, in order +// to make that child process start with the same profiler settings as +// in the current process. The given function is invoked once for +// each variable to be set. +MFBT_API void GetProfilerEnvVarsForChildProcess( + std::function<void(const char* key, const char* value)>&& aSetEnv); + +} // namespace baseprofiler +} // namespace mozilla + +#endif // !MOZ_GECKO_PROFILER + +#endif // BaseProfiler_h diff --git a/mozglue/baseprofiler/public/BaseProfilerCounts.h b/mozglue/baseprofiler/public/BaseProfilerCounts.h new file mode 100644 index 0000000000..fbcc713744 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerCounts.h @@ -0,0 +1,280 @@ +/* -*- 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 BaseProfilerCounts_h +#define BaseProfilerCounts_h + +#ifndef MOZ_GECKO_PROFILER + +# define BASE_PROFILER_DEFINE_COUNT_TOTAL(label, category, description) +# define BASE_PROFILER_DEFINE_COUNT(label, category, description) +# define BASE_PROFILER_DEFINE_STATIC_COUNT_TOTAL(label, category, description) +# define AUTO_BASE_PROFILER_COUNT_TOTAL(label, count) +# define AUTO_BASE_PROFILER_COUNT(label) +# define AUTO_BASE_PROFILER_STATIC_COUNT(label, count) +# define AUTO_BASE_PROFILER_FORCE_ALLOCATION(label) + +#else + +# include "mozilla/Atomics.h" + +namespace mozilla { +namespace baseprofiler { + +class BaseProfilerCount; +MFBT_API void profiler_add_sampled_counter(BaseProfilerCount* aCounter); +MFBT_API void profiler_remove_sampled_counter(BaseProfilerCount* aCounter); + +typedef Atomic<int64_t, MemoryOrdering::Relaxed> ProfilerAtomicSigned; +typedef Atomic<uint64_t, MemoryOrdering::Relaxed> ProfilerAtomicUnsigned; + +// Counter support +// There are two types of counters: +// 1) a simple counter which can be added to or subtracted from. This could +// track the number of objects of a type, the number of calls to something +// (reflow, JIT, etc). +// 2) a combined counter which has the above, plus a number-of-calls counter +// that is incremented by 1 for each call to modify the count. This provides +// an optional source for a 'heatmap' of access. This can be used (for +// example) to track the amount of memory allocated, and provide a heatmap of +// memory operations (allocs/frees). +// +// Counters are sampled by the profiler once per sample-period. At this time, +// all counters are global to the process. In the future, there might be more +// versions with per-thread or other discriminators. +// +// Typical usage: +// There are two ways to use counters: With heap-created counter objects, +// or using macros. Note: the macros use statics, and will be slightly +// faster/smaller, and you need to care about creating them before using +// them. They're similar to the use-pattern for the other AUTO_PROFILER* +// macros, but they do need the PROFILER_DEFINE* to be use to instantiate +// the statics. +// +// PROFILER_DEFINE_COUNT(mything, "JIT", "Some JIT byte count") +// ... +// void foo() { ... AUTO_PROFILER_COUNT(mything, number_of_bytes_used); ... } +// +// or (to also get a heatmap) +// +// PROFILER_DEFINE_COUNT_TOTAL(mything, "JIT", "Some JIT byte count") +// ... +// void foo() { +// ... +// AUTO_PROFILER_COUNT_TOTAL(mything, number_of_bytes_generated); +// ... +// } +// +// To use without statics/macros: +// +// UniquePtr<ProfilerCounter> myCounter; +// ... +// myCounter = +// MakeUnique<ProfilerCounter>("mything", "JIT", "Some JIT byte count")); +// ... +// void foo() { ... myCounter->Add(number_of_bytes_generated0; ... } + +class BaseProfilerCount { + public: + BaseProfilerCount(const char* aLabel, ProfilerAtomicSigned* aCounter, + ProfilerAtomicUnsigned* aNumber, const char* aCategory, + const char* aDescription) + : mLabel(aLabel), + mCategory(aCategory), + mDescription(aDescription), + mCounter(aCounter), + mNumber(aNumber) { +# define COUNTER_CANARY 0xDEADBEEF +# ifdef DEBUG + mCanary = COUNTER_CANARY; + mPrevNumber = 0; +# endif + // Can't call profiler_* here since this may be non-xul-library + } +# ifdef DEBUG + ~BaseProfilerCount() { mCanary = 0; } +# endif + + void Sample(int64_t& aCounter, uint64_t& aNumber) { + MOZ_ASSERT(mCanary == COUNTER_CANARY); + + aCounter = *mCounter; + aNumber = mNumber ? *mNumber : 0; +# ifdef DEBUG + MOZ_ASSERT(aNumber >= mPrevNumber); + mPrevNumber = aNumber; +# endif + } + + // We don't define ++ and Add() here, since the static defines directly + // increment the atomic counters, and the subclasses implement ++ and + // Add() directly. + + // These typically are static strings (for example if you use the macros + // below) + const char* mLabel; + const char* mCategory; + const char* mDescription; + // We're ok with these being un-ordered in race conditions. These are + // pointers because we want to be able to use statics and increment them + // directly. Otherwise we could just have them inline, and not need the + // constructor args. + // These can be static globals (using the macros below), though they + // don't have to be - their lifetime must be longer than the use of them + // by the profiler (see profiler_add/remove_sampled_counter()). If you're + // using a lot of these, they probably should be allocated at runtime (see + // class ProfilerCountOnly below). + ProfilerAtomicSigned* mCounter; + ProfilerAtomicUnsigned* mNumber; // may be null + +# ifdef DEBUG + uint32_t mCanary; + uint64_t mPrevNumber; // value of number from the last Sample() +# endif +}; + +// Designed to be allocated dynamically, and simply incremented with obj++ +// or obj->Add(n) +class ProfilerCounter final : public BaseProfilerCount { + public: + ProfilerCounter(const char* aLabel, const char* aCategory, + const char* aDescription) + : BaseProfilerCount(aLabel, &mCounter, nullptr, aCategory, aDescription) { + // Assume we're in libxul + profiler_add_sampled_counter(this); + } + + virtual ~ProfilerCounter() { profiler_remove_sampled_counter(this); } + + BaseProfilerCount& operator++() { + Add(1); + return *this; + } + + void Add(int64_t aNumber) { mCounter += aNumber; } + + ProfilerAtomicSigned mCounter; +}; + +// Also keeps a heatmap (number of calls to ++/Add()) +class ProfilerCounterTotal final : public BaseProfilerCount { + public: + ProfilerCounterTotal(const char* aLabel, const char* aCategory, + const char* aDescription) + : BaseProfilerCount(aLabel, &mCounter, &mNumber, aCategory, + aDescription) { + // Assume we're in libxul + profiler_add_sampled_counter(this); + } + + virtual ~ProfilerCounterTotal() { profiler_remove_sampled_counter(this); } + + BaseProfilerCount& operator++() { + Add(1); + return *this; + } + + void Add(int64_t aNumber) { + mCounter += aNumber; + mNumber++; + } + + ProfilerAtomicSigned mCounter; + ProfilerAtomicUnsigned mNumber; +}; + +// Defines a counter that is sampled on each profiler tick, with a running +// count (signed), and number-of-instances. Note that because these are two +// independent Atomics, there is a possiblity that count will not include +// the last call, but number of uses will. I think this is not worth +// worrying about +# define BASE_PROFILER_DEFINE_COUNT_TOTAL(label, category, description) \ + ProfilerAtomicSigned profiler_count_##label(0); \ + ProfilerAtomicUnsigned profiler_number_##label(0); \ + const char profiler_category_##label[] = category; \ + const char profiler_description_##label[] = description; \ + UniquePtr<::mozilla::baseprofiler::BaseProfilerCount> AutoCount_##label; + +// This counts, but doesn't keep track of the number of calls to +// AUTO_PROFILER_COUNT() +# define BASE_PROFILER_DEFINE_COUNT(label, category, description) \ + ProfilerAtomicSigned profiler_count_##label(0); \ + const char profiler_category_##label[] = category; \ + const char profiler_description_##label[] = description; \ + UniquePtr<::mozilla::baseprofiler::BaseProfilerCount> AutoCount_##label; + +// This will create a static initializer if used, but avoids a possible +// allocation. +# define BASE_PROFILER_DEFINE_STATIC_COUNT_TOTAL(label, category, \ + description) \ + ProfilerAtomicSigned profiler_count_##label(0); \ + ProfilerAtomicUnsigned profiler_number_##label(0); \ + ::mozilla::baseprofiler::BaseProfilerCount AutoCount_##label( \ + #label, &profiler_count_##label, &profiler_number_##label, category, \ + description); + +// If we didn't care about static initializers, we could avoid the need for +// a ptr to the BaseProfilerCount object. + +// XXX It would be better to do this without the if() and without the +// theoretical race to set the UniquePtr (i.e. possible leak). +# define AUTO_BASE_PROFILER_COUNT_TOTAL(label, count) \ + do { \ + profiler_number_##label++; /* do this first*/ \ + profiler_count_##label += count; \ + if (!AutoCount_##label) { \ + /* Ignore that we could call this twice in theory, and that we leak \ + * them \ + */ \ + AutoCount_##label.reset(new BaseProfilerCount( \ + #label, &profiler_count_##label, &profiler_number_##label, \ + profiler_category_##label, profiler_description_##label)); \ + ::mozilla::baseprofiler::profiler_add_sampled_counter( \ + AutoCount_##label.get()); \ + } \ + } while (0) + +# define AUTO_BASE_PROFILER_COUNT(label, count) \ + do { \ + profiler_count_##label += count; /* do this first*/ \ + if (!AutoCount_##label) { \ + /* Ignore that we could call this twice in theory, and that we leak \ + * them \ + */ \ + AutoCount_##label.reset(new BaseProfilerCount( \ + #label, nullptr, &profiler_number_##label, \ + profiler_category_##label, profiler_description_##label)); \ + ::mozilla::baseprofiler::profiler_add_sampled_counter( \ + AutoCount_##label.get()); \ + } \ + } while (0) + +# define AUTO_BASE_PROFILER_STATIC_COUNT(label, count) \ + do { \ + profiler_number_##label++; /* do this first*/ \ + profiler_count_##label += count; \ + } while (0) + +// if we need to force the allocation +# define AUTO_BASE_PROFILER_FORCE_ALLOCATION(label) \ + do { \ + if (!AutoCount_##label) { \ + /* Ignore that we could call this twice in theory, and that we leak \ + * them \ + */ \ + AutoCount_##label.reset( \ + new ::mozilla::baseprofiler::BaseProfilerCount( \ + #label, &profiler_count_##label, &profiler_number_##label, \ + profiler_category_##label, profiler_description_##label)); \ + } \ + } while (0) + +} // namespace baseprofiler +} // namespace mozilla + +#endif // !MOZ_GECKO_PROFILER + +#endif // BaseProfilerCounts_h diff --git a/mozglue/baseprofiler/public/BaseProfilerDetail.h b/mozglue/baseprofiler/public/BaseProfilerDetail.h new file mode 100644 index 0000000000..9027f32bc7 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerDetail.h @@ -0,0 +1,189 @@ +/* -*- 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/. */ + +// Internal Base Profiler utilities. + +#ifndef BaseProfilerDetail_h +#define BaseProfilerDetail_h + +#include "mozilla/Atomics.h" +#include "mozilla/Maybe.h" +#include "mozilla/PlatformMutex.h" + +#ifndef MOZ_GECKO_PROFILER +# error Do not #include this header when MOZ_GECKO_PROFILER is not #defined. +#endif + +namespace mozilla { +namespace baseprofiler { + +// Implemented in platform.cpp +MFBT_API int profiler_current_thread_id(); + +namespace detail { + +// Thin shell around mozglue PlatformMutex, for Base Profiler internal use. +class BaseProfilerMutex : private ::mozilla::detail::MutexImpl { + public: + BaseProfilerMutex() : ::mozilla::detail::MutexImpl() {} + explicit BaseProfilerMutex(const char* aName) + : ::mozilla::detail::MutexImpl(), mName(aName) {} + + BaseProfilerMutex(const BaseProfilerMutex&) = delete; + BaseProfilerMutex& operator=(const BaseProfilerMutex&) = delete; + BaseProfilerMutex(BaseProfilerMutex&&) = delete; + BaseProfilerMutex& operator=(BaseProfilerMutex&&) = delete; + +#ifdef DEBUG + ~BaseProfilerMutex() { MOZ_ASSERT(mOwningThreadId == 0); } +#endif // DEBUG + + [[nodiscard]] bool IsLockedOnCurrentThread() const { + return mOwningThreadId == baseprofiler::profiler_current_thread_id(); + } + + void AssertCurrentThreadOwns() const { + MOZ_ASSERT(IsLockedOnCurrentThread()); + } + + void Lock() { + const int tid = baseprofiler::profiler_current_thread_id(); + MOZ_ASSERT(tid != 0); + MOZ_ASSERT(!IsLockedOnCurrentThread(), "Recursive locking"); + ::mozilla::detail::MutexImpl::lock(); + MOZ_ASSERT(mOwningThreadId == 0, "Not unlocked properly"); + mOwningThreadId = tid; + } + + [[nodiscard]] bool TryLock() { + const int tid = baseprofiler::profiler_current_thread_id(); + MOZ_ASSERT(tid != 0); + MOZ_ASSERT(!IsLockedOnCurrentThread(), "Recursive locking"); + if (!::mozilla::detail::MutexImpl::tryLock()) { + // Failed to lock, nothing more to do. + return false; + } + MOZ_ASSERT(mOwningThreadId == 0, "Not unlocked properly"); + mOwningThreadId = tid; + return true; + } + + void Unlock() { + MOZ_ASSERT(IsLockedOnCurrentThread(), "Unlocking when not locked here"); + // We're still holding the mutex here, so it's safe to just reset + // `mOwningThreadId`. + mOwningThreadId = 0; + ::mozilla::detail::MutexImpl::unlock(); + } + + const char* GetName() const { return mName; } + + private: + // Thread currently owning the lock, or 0. + // Atomic because it may be read at any time independent of the mutex. + // Relaxed because threads only need to know if they own it already, so: + // - If it's their id, only *they* wrote that value with a locked mutex. + // - If it's different from their thread id it doesn't matter what other + // number it is (0 or another id) and that it can change again at any time. + Atomic<int, MemoryOrdering::Relaxed> mOwningThreadId{0}; + + const char* mName = nullptr; +}; + +// RAII class to lock a mutex. +class MOZ_RAII BaseProfilerAutoLock { + public: + explicit BaseProfilerAutoLock(BaseProfilerMutex& aMutex) : mMutex(aMutex) { + mMutex.Lock(); + } + + BaseProfilerAutoLock(const BaseProfilerAutoLock&) = delete; + BaseProfilerAutoLock& operator=(const BaseProfilerAutoLock&) = delete; + BaseProfilerAutoLock(BaseProfilerAutoLock&&) = delete; + BaseProfilerAutoLock& operator=(BaseProfilerAutoLock&&) = delete; + + ~BaseProfilerAutoLock() { mMutex.Unlock(); } + + private: + BaseProfilerMutex& mMutex; +}; + +// Thin shell around mozglue PlatformMutex, for Base Profiler internal use. +// Actual mutex may be disabled at construction time. +class BaseProfilerMaybeMutex : private ::mozilla::detail::MutexImpl { + public: + explicit BaseProfilerMaybeMutex(bool aActivate) { + if (aActivate) { + mMaybeMutex.emplace(); + } + } + + BaseProfilerMaybeMutex(const BaseProfilerMaybeMutex&) = delete; + BaseProfilerMaybeMutex& operator=(const BaseProfilerMaybeMutex&) = delete; + BaseProfilerMaybeMutex(BaseProfilerMaybeMutex&&) = delete; + BaseProfilerMaybeMutex& operator=(BaseProfilerMaybeMutex&&) = delete; + + ~BaseProfilerMaybeMutex() = default; + + bool IsActivated() const { return mMaybeMutex.isSome(); } + + [[nodiscard]] bool IsActivatedAndLockedOnCurrentThread() const { + if (!IsActivated()) { + // Not activated, so we can never be locked. + return false; + } + return mMaybeMutex->IsLockedOnCurrentThread(); + } + + void AssertCurrentThreadOwns() const { +#ifdef DEBUG + if (IsActivated()) { + mMaybeMutex->AssertCurrentThreadOwns(); + } +#endif // DEBUG + } + + void Lock() { + if (IsActivated()) { + mMaybeMutex->Lock(); + } + } + + void Unlock() { + if (IsActivated()) { + mMaybeMutex->Unlock(); + } + } + + private: + Maybe<BaseProfilerMutex> mMaybeMutex; +}; + +// RAII class to lock a mutex. +class MOZ_RAII BaseProfilerMaybeAutoLock { + public: + explicit BaseProfilerMaybeAutoLock(BaseProfilerMaybeMutex& aMaybeMutex) + : mMaybeMutex(aMaybeMutex) { + mMaybeMutex.Lock(); + } + + BaseProfilerMaybeAutoLock(const BaseProfilerMaybeAutoLock&) = delete; + BaseProfilerMaybeAutoLock& operator=(const BaseProfilerMaybeAutoLock&) = + delete; + BaseProfilerMaybeAutoLock(BaseProfilerMaybeAutoLock&&) = delete; + BaseProfilerMaybeAutoLock& operator=(BaseProfilerMaybeAutoLock&&) = delete; + + ~BaseProfilerMaybeAutoLock() { mMaybeMutex.Unlock(); } + + private: + BaseProfilerMaybeMutex& mMaybeMutex; +}; + +} // namespace detail +} // namespace baseprofiler +} // namespace mozilla + +#endif // BaseProfilerDetail_h diff --git a/mozglue/baseprofiler/public/BaseProfilerMarkerTypes.h b/mozglue/baseprofiler/public/BaseProfilerMarkerTypes.h new file mode 100644 index 0000000000..1556b7a272 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerMarkerTypes.h @@ -0,0 +1,69 @@ +/* -*- 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 BaseProfilerMarkerTypes_h +#define BaseProfilerMarkerTypes_h + +// This header contains common marker type definitions. +// +// It #include's "mozilla/BaseProfilerMarkers.h", see that file for how to +// define other marker types, and how to add markers to the profiler buffers. +// +// If you don't need to use these common types, #include +// "mozilla/BaseProfilerMarkers.h" instead. +// +// Types in this files can be defined without relying on xpcom. +// Others are defined in "ProfilerMarkerTypes.h". + +// !!! /!\ WORK IN PROGRESS /!\ !!! +// This file contains draft marker definitions, but most are not used yet. +// Further work is needed to complete these definitions, and use them to convert +// existing PROFILER_ADD_MARKER calls. See meta bug 1661394. + +#include "mozilla/BaseProfilerMarkers.h" + +#ifdef MOZ_GECKO_PROFILER + +namespace mozilla::baseprofiler::markers { + +struct MediaSampleMarker { + static constexpr Span<const char> MarkerTypeName() { + return MakeStringSpan("MediaSample"); + } + static void StreamJSONMarkerData(SpliceableJSONWriter& aWriter, + int64_t aSampleStartTimeUs, + int64_t aSampleEndTimeUs) { + aWriter.IntProperty("sampleStartTimeUs", aSampleStartTimeUs); + aWriter.IntProperty("sampleEndTimeUs", aSampleEndTimeUs); + } + static MarkerSchema MarkerTypeDisplay() { + using MS = MarkerSchema; + MS schema{MS::Location::markerChart, MS::Location::markerTable}; + schema.AddKeyLabelFormat("sampleStartTimeUs", "Sample start time", + MS::Format::microseconds); + schema.AddKeyLabelFormat("sampleEndTimeUs", "Sample end time", + MS::Format::microseconds); + return schema; + } +}; + +struct ContentBuildMarker { + static constexpr Span<const char> MarkerTypeName() { + return MakeStringSpan("CONTENT_FULL_PAINT_TIME"); + } + static void StreamJSONMarkerData(SpliceableJSONWriter& aWriter) {} + static MarkerSchema MarkerTypeDisplay() { + using MS = MarkerSchema; + MS schema{MS::Location::markerChart, MS::Location::markerTable}; + return schema; + } +}; + +} // namespace mozilla::baseprofiler::markers + +#endif // MOZ_GECKO_PROFILER + +#endif // BaseProfilerMarkerTypes_h diff --git a/mozglue/baseprofiler/public/BaseProfilerMarkers.h b/mozglue/baseprofiler/public/BaseProfilerMarkers.h new file mode 100644 index 0000000000..c63b018f95 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerMarkers.h @@ -0,0 +1,242 @@ +/* -*- 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/. */ + +// Markers are useful to delimit something important happening such as the first +// paint. Unlike labels, which are only recorded in the profile buffer if a +// sample is collected while the label is on the label stack, markers will +// always be recorded in the profile buffer. +// +// This header contains basic definitions necessary to create marker types, and +// to add markers to the profiler buffers. +// +// If basic marker types are needed, #include +// "mozilla/BaseProfilerMarkerTypes.h" instead. +// +// But if you want to create your own marker type locally, you can #include this +// header only; look at mozilla/BaseProfilerMarkerTypes.h for examples of how to +// define types, and mozilla/BaseProfilerMarkerPrerequisites.h for some +// supporting types. +// +// To then record markers: +// - Use `baseprofiler::AddMarker(...)` from mozglue or other libraries that +// are outside of xul, especially if they may happen outside of xpcom's +// lifetime (typically startup, shutdown, or tests). +// - Otherwise #include "ProfilerMarkers.h" instead, and use +// `profiler_add_marker(...)`. +// See these functions for more details. + +#ifndef BaseProfilerMarkers_h +#define BaseProfilerMarkers_h + +#include "mozilla/BaseProfilerMarkersDetail.h" + +#ifndef MOZ_GECKO_PROFILER + +# define BASE_PROFILER_MARKER_UNTYPED(markerName, categoryName, ...) +# define BASE_PROFILER_MARKER(markerName, categoryName, options, MarkerType, \ + ...) +# define BASE_PROFILER_MARKER_TEXT(markerName, categoryName, options, text) +# define AUTO_BASE_PROFILER_MARKER_TEXT(markerName, categoryName, options, \ + text) + +#else // ndef MOZ_GECKO_PROFILER + +# include "mozilla/ProfileChunkedBuffer.h" +# include "mozilla/TimeStamp.h" +# include "mozilla/Unused.h" + +# include <functional> +# include <string> +# include <utility> + +namespace mozilla::baseprofiler { + +// Add a marker to a given buffer. `AddMarker()` and related macros should be +// used in most cases, see below for more information about them and the +// parameters; This function may be useful when markers need to be recorded in a +// local buffer outside of the main profiler buffer. +template <typename MarkerType, typename... PayloadArguments> +ProfileBufferBlockIndex AddMarkerToBuffer( + ProfileChunkedBuffer& aBuffer, const ProfilerString8View& aName, + const MarkerCategory& aCategory, MarkerOptions&& aOptions, + MarkerType aMarkerType, const PayloadArguments&... aPayloadArguments) { + Unused << aMarkerType; // Only the empty object type is useful. + return base_profiler_markers_detail::AddMarkerToBuffer<MarkerType>( + aBuffer, aName, aCategory, std::move(aOptions), + ::mozilla::baseprofiler::profiler_capture_backtrace_into, + aPayloadArguments...); +} + +// Add a marker (without payload) to a given buffer. +inline ProfileBufferBlockIndex AddMarkerToBuffer( + ProfileChunkedBuffer& aBuffer, const ProfilerString8View& aName, + const MarkerCategory& aCategory, MarkerOptions&& aOptions = {}) { + return AddMarkerToBuffer(aBuffer, aName, aCategory, std::move(aOptions), + markers::NoPayload{}); +} + +// Add a marker to the Base Profiler buffer. +// - aName: Main name of this marker. +// - aCategory: Category for this marker. +// - aOptions: Optional settings (such as timing, inner window id, +// backtrace...), see `MarkerOptions` for details. +// - aMarkerType: Empty object that specifies the type of marker. +// - aPayloadArguments: Arguments expected by this marker type's +// ` StreamJSONMarkerData` function. +template <typename MarkerType, typename... PayloadArguments> +ProfileBufferBlockIndex AddMarker( + const ProfilerString8View& aName, const MarkerCategory& aCategory, + MarkerOptions&& aOptions, MarkerType aMarkerType, + const PayloadArguments&... aPayloadArguments) { + if (!baseprofiler::profiler_can_accept_markers()) { + return {}; + } + return ::mozilla::baseprofiler::AddMarkerToBuffer( + base_profiler_markers_detail::CachedBaseCoreBuffer(), aName, aCategory, + std::move(aOptions), aMarkerType, aPayloadArguments...); +} + +// Add a marker (without payload) to the Base Profiler buffer. +inline ProfileBufferBlockIndex AddMarker(const ProfilerString8View& aName, + const MarkerCategory& aCategory, + MarkerOptions&& aOptions = {}) { + return AddMarker(aName, aCategory, std::move(aOptions), markers::NoPayload{}); +} + +} // namespace mozilla::baseprofiler + +// Same as `AddMarker()` (without payload). This macro is safe to use even if +// MOZ_GECKO_PROFILER is not #defined. +# define BASE_PROFILER_MARKER_UNTYPED(markerName, categoryName, ...) \ + do { \ + AUTO_PROFILER_STATS(BASE_PROFILER_MARKER_UNTYPED); \ + ::mozilla::baseprofiler::AddMarker( \ + markerName, ::mozilla::baseprofiler::category::categoryName, \ + ##__VA_ARGS__); \ + } while (false) + +// Same as `AddMarker()` (with payload). This macro is safe to use even if +// MOZ_GECKO_PROFILER is not #defined. +# define BASE_PROFILER_MARKER(markerName, categoryName, options, MarkerType, \ + ...) \ + do { \ + AUTO_PROFILER_STATS(BASE_PROFILER_MARKER_with_##MarkerType); \ + ::mozilla::baseprofiler::AddMarker( \ + markerName, ::mozilla::baseprofiler::category::categoryName, \ + options, ::mozilla::baseprofiler::markers::MarkerType{}, \ + ##__VA_ARGS__); \ + } while (false) + +namespace mozilla::baseprofiler::markers { +// Most common marker type. Others are in BaseProfilerMarkerTypes.h. +struct TextMarker { + static constexpr Span<const char> MarkerTypeName() { + return MakeStringSpan("Text"); + } + static void StreamJSONMarkerData(baseprofiler::SpliceableJSONWriter& aWriter, + const ProfilerString8View& aText) { + aWriter.StringProperty("name", aText); + } + static MarkerSchema MarkerTypeDisplay() { + using MS = MarkerSchema; + MS schema{MS::Location::markerChart, MS::Location::markerTable}; + schema.SetChartLabel("{marker.data.name}"); + schema.SetTableLabel("{marker.name} - {marker.data.name}"); + schema.AddKeyLabelFormat("name", "Details", MarkerSchema::Format::string); + return schema; + } +}; + +struct Tracing { + static constexpr Span<const char> MarkerTypeName() { + return MakeStringSpan("tracing"); + } + static void StreamJSONMarkerData(SpliceableJSONWriter& aWriter, + const ProfilerString8View& aCategory) { + if (aCategory.Length() != 0) { + aWriter.StringProperty("category", aCategory); + } + } + static MarkerSchema MarkerTypeDisplay() { + using MS = MarkerSchema; + MS schema{MS::Location::markerChart, MS::Location::markerTable, + MS::Location::timelineOverview}; + schema.AddKeyLabelFormat("category", "Type", MS::Format::string); + return schema; + } +}; +} // namespace mozilla::baseprofiler::markers + +// Add a text marker. This macro is safe to use even if MOZ_GECKO_PROFILER is +// not #defined. +# define BASE_PROFILER_MARKER_TEXT(markerName, categoryName, options, text) \ + do { \ + AUTO_PROFILER_STATS(BASE_PROFILER_MARKER_TEXT); \ + ::mozilla::baseprofiler::AddMarker( \ + markerName, ::mozilla::baseprofiler::category::categoryName, \ + options, ::mozilla::baseprofiler::markers::TextMarker{}, text); \ + } while (false) + +namespace mozilla::baseprofiler { + +// RAII object that adds a BASE_PROFILER_MARKER_TEXT when destroyed; the +// marker's timing will be the interval from construction (unless an instant or +// start time is already specified in the provided options) until destruction. +class MOZ_RAII AutoProfilerTextMarker { + public: + AutoProfilerTextMarker(const char* aMarkerName, + const MarkerCategory& aCategory, + MarkerOptions&& aOptions, const std::string& aText) + : mMarkerName(aMarkerName), + mCategory(aCategory), + mOptions(std::move(aOptions)), + mText(aText) { + MOZ_ASSERT(mOptions.Timing().EndTime().IsNull(), + "AutoProfilerTextMarker options shouldn't have an end time"); + if (mOptions.Timing().StartTime().IsNull()) { + mOptions.Set(MarkerTiming::InstantNow()); + } + } + + ~AutoProfilerTextMarker() { + mOptions.TimingRef().SetIntervalEnd(); + AUTO_PROFILER_STATS(AUTO_BASE_PROFILER_MARKER_TEXT); + AddMarker(ProfilerString8View::WrapNullTerminatedString(mMarkerName), + mCategory, std::move(mOptions), markers::TextMarker{}, mText); + } + + protected: + const char* mMarkerName; + MarkerCategory mCategory; + MarkerOptions mOptions; + std::string mText; +}; + +extern template MFBT_API ProfileBufferBlockIndex +AddMarker(const ProfilerString8View&, const MarkerCategory&, MarkerOptions&&, + markers::TextMarker, const std::string&); + +extern template MFBT_API ProfileBufferBlockIndex +AddMarkerToBuffer(ProfileChunkedBuffer&, const ProfilerString8View&, + const MarkerCategory&, MarkerOptions&&, markers::NoPayload); + +extern template MFBT_API ProfileBufferBlockIndex AddMarkerToBuffer( + ProfileChunkedBuffer&, const ProfilerString8View&, const MarkerCategory&, + MarkerOptions&&, markers::TextMarker, const std::string&); + +} // namespace mozilla::baseprofiler + +// Creates an AutoProfilerTextMarker RAII object. This macro is safe to use +// even if MOZ_GECKO_PROFILER is not #defined. +# define AUTO_BASE_PROFILER_MARKER_TEXT(markerName, categoryName, options, \ + text) \ + ::mozilla::baseprofiler::AutoProfilerTextMarker BASE_PROFILER_RAII( \ + markerName, ::mozilla::baseprofiler::category::categoryName, options, \ + text) + +#endif // nfed MOZ_GECKO_PROFILER else + +#endif // BaseProfilerMarkers_h diff --git a/mozglue/baseprofiler/public/BaseProfilerMarkersDetail.h b/mozglue/baseprofiler/public/BaseProfilerMarkersDetail.h new file mode 100644 index 0000000000..b5dbe27343 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerMarkersDetail.h @@ -0,0 +1,674 @@ +/* -*- 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 BaseProfilerMarkersDetail_h +#define BaseProfilerMarkersDetail_h + +#ifndef BaseProfilerMarkers_h +# error "This header should only be #included by BaseProfilerMarkers.h" +#endif + +#include "mozilla/BaseProfilerMarkersPrerequisites.h" + +#ifdef MOZ_GECKO_PROFILER + +// ~~ HERE BE DRAGONS ~~ +// +// Everything below is internal implementation detail, you shouldn't need to +// look at it unless working on the profiler code. + +# include "mozilla/BaseProfileJSONWriter.h" +# include "mozilla/ProfileBufferEntryKinds.h" + +# include <limits> +# include <tuple> + +namespace mozilla::baseprofiler { +// Implemented in platform.cpp +MFBT_API ProfileChunkedBuffer& profiler_get_core_buffer(); +} // namespace mozilla::baseprofiler + +namespace mozilla::base_profiler_markers_detail { + +// Get the core buffer from the profiler, and cache it in a +// non-templated-function static reference. +inline ProfileChunkedBuffer& CachedBaseCoreBuffer() { + static ProfileChunkedBuffer& coreBuffer = + baseprofiler::profiler_get_core_buffer(); + return coreBuffer; +} + +struct Streaming { + // A `MarkerDataDeserializer` is a free function that can read a serialized + // payload from an `EntryReader` and streams it as JSON object properties. + using MarkerDataDeserializer = void (*)(ProfileBufferEntryReader&, + baseprofiler::SpliceableJSONWriter&); + + // A `MarkerTypeNameFunction` is a free function that returns the name of the + // marker type. + using MarkerTypeNameFunction = Span<const char> (*)(); + + // A `MarkerSchemaFunction` is a free function that returns a + // `MarkerSchema`, which contains all the information needed to stream + // the display schema associated with a marker type. + using MarkerSchemaFunction = MarkerSchema (*)(); + + struct MarkerTypeFunctions { + MarkerDataDeserializer mMarkerDataDeserializer = nullptr; + MarkerTypeNameFunction mMarkerTypeNameFunction = nullptr; + MarkerSchemaFunction mMarkerSchemaFunction = nullptr; + }; + + // A `DeserializerTag` will be added before the payload, to help select the + // correct deserializer when reading back the payload. + using DeserializerTag = uint8_t; + + // Store a deserializer (and other marker-type-specific functions) and get its + // `DeserializerTag`. + // This is intended to be only used once per deserializer when a new marker + // type is used for the first time, so it should be called to initialize a + // `static const` tag that will be re-used by all markers of the corresponding + // payload type -- see use below. + MFBT_API static DeserializerTag TagForMarkerTypeFunctions( + MarkerDataDeserializer aDeserializer, + MarkerTypeNameFunction aMarkerTypeNameFunction, + MarkerSchemaFunction aMarkerSchemaFunction); + + // Get the `MarkerDataDeserializer` for a given `DeserializerTag`. + MFBT_API static MarkerDataDeserializer DeserializerForTag( + DeserializerTag aTag); + + // Retrieve all MarkerTypeFunctions's. + MFBT_API static Span<const MarkerTypeFunctions> MarkerTypeFunctionsArray(); +}; + +// This helper will examine a marker type's `StreamJSONMarkerData` function, see +// specialization below. +template <typename T> +struct StreamFunctionTypeHelper; + +// Helper specialization that takes the expected +// `StreamJSONMarkerData(baseprofiler::SpliceableJSONWriter&, ...)` function and +// provide information about the `...` parameters. +template <typename R, typename... As> +struct StreamFunctionTypeHelper<R(baseprofiler::SpliceableJSONWriter&, As...)> { + constexpr static size_t scArity = sizeof...(As); + using TupleType = + std::tuple<std::remove_cv_t<std::remove_reference_t<As>>...>; + + // Serialization function that takes the exact same parameter types + // (const-ref'd) as `StreamJSONMarkerData`. This has to be inside the helper + // because only here can we access the raw parameter pack `As...`. + // And because we're using the same argument types through + // references-to-const, permitted implicit conversions can happen. + static ProfileBufferBlockIndex Serialize( + ProfileChunkedBuffer& aBuffer, const ProfilerString8View& aName, + const MarkerCategory& aCategory, MarkerOptions&& aOptions, + Streaming::DeserializerTag aDeserializerTag, const As&... aAs) { + // Note that options are first after the entry kind, because they contain + // the thread id, which is handled first to filter markers by threads. + return aBuffer.PutObjects(ProfileBufferEntryKind::Marker, aOptions, aName, + aCategory, aDeserializerTag, aAs...); + } +}; + +// Helper for a marker type. +// A marker type is defined in a `struct` with some expected static member +// functions. See example in BaseProfilerMarkers.h. +template <typename MarkerType> +struct MarkerTypeSerialization { + // Definitions to access the expected + // `StreamJSONMarkerData(baseprofiler::SpliceableJSONWriter&, ...)` function + // and its parameters. + using StreamFunctionType = + StreamFunctionTypeHelper<decltype(MarkerType::StreamJSONMarkerData)>; + constexpr static size_t scStreamFunctionParameterCount = + StreamFunctionType::scArity; + using StreamFunctionUserParametersTuple = + typename StreamFunctionType::TupleType; + template <size_t i> + using StreamFunctionParameter = + std::tuple_element_t<i, StreamFunctionUserParametersTuple>; + + template <typename... Ts> + static ProfileBufferBlockIndex Serialize(ProfileChunkedBuffer& aBuffer, + const ProfilerString8View& aName, + const MarkerCategory& aCategory, + MarkerOptions&& aOptions, + const Ts&... aTs) { + static_assert(!std::is_same_v<MarkerType, + ::mozilla::baseprofiler::markers::NoPayload>, + "NoPayload should have been handled in the caller."); + // Register marker type functions, and get the tag for this deserializer. + // Note that the tag is stored in a function-static object, and this + // function is static in a templated struct, so there should only be one tag + // per MarkerType. + // Making the tag class-static may have been more efficient (to avoid a + // thread-safe init check at every call), but random global static + // initialization order would make it more complex to coordinate with + // `Streaming::TagForMarkerTypeFunctions()`, and also would add a (small) + // cost for everybody, even the majority of users not using the profiler. + static const Streaming::DeserializerTag tag = + Streaming::TagForMarkerTypeFunctions(Deserialize, + MarkerType::MarkerTypeName, + MarkerType::MarkerTypeDisplay); + return StreamFunctionType::Serialize(aBuffer, aName, aCategory, + std::move(aOptions), tag, aTs...); + } + + private: + // This templated function will recursively deserialize each argument expected + // by `MarkerType::StreamJSONMarkerData()` on the stack, and call it at the + // end. E.g., for `StreamJSONMarkerData(int, char)`: + // - DeserializeArguments<0>(aER, aWriter) reads an int and calls: + // - DeserializeArguments<1>(aER, aWriter, const int&) reads a char and calls: + // - MarkerType::StreamJSONMarkerData(aWriter, const int&, const char&). + // Prototyping on godbolt showed that clang and gcc can flatten these + // recursive calls into one function with successive reads followed by the one + // stream call; tested up to 40 arguments: https://godbolt.org/z/5KeeM4 + template <size_t i = 0, typename... Args> + static void DeserializeArguments(ProfileBufferEntryReader& aEntryReader, + baseprofiler::SpliceableJSONWriter& aWriter, + const Args&... aArgs) { + static_assert(sizeof...(Args) == i, + "We should have collected `i` arguments so far"); + if constexpr (i < scStreamFunctionParameterCount) { + // Deserialize the i-th argument on this stack. + auto argument = aEntryReader.ReadObject<StreamFunctionParameter<i>>(); + // Add our local argument to the next recursive call. + DeserializeArguments<i + 1>(aEntryReader, aWriter, aArgs..., argument); + } else { + // We've read all the arguments, finally call the `StreamJSONMarkerData` + // function, which should write the appropriate JSON elements for this + // marker type. Note that the MarkerType-specific "type" element is + // already written. + MarkerType::StreamJSONMarkerData(aWriter, aArgs...); + } + } + + public: + static void Deserialize(ProfileBufferEntryReader& aEntryReader, + baseprofiler::SpliceableJSONWriter& aWriter) { + aWriter.StringProperty("type", MarkerType::MarkerTypeName()); + DeserializeArguments(aEntryReader, aWriter); + } +}; + +template <> +struct MarkerTypeSerialization<::mozilla::baseprofiler::markers::NoPayload> { + // Nothing! NoPayload has special handling avoiding payload work. +}; + +template <typename MarkerType, typename... Ts> +static ProfileBufferBlockIndex AddMarkerWithOptionalStackToBuffer( + ProfileChunkedBuffer& aBuffer, const ProfilerString8View& aName, + const MarkerCategory& aCategory, MarkerOptions&& aOptions, + const Ts&... aTs) { + if constexpr (std::is_same_v<MarkerType, + ::mozilla::baseprofiler::markers::NoPayload>) { + static_assert(sizeof...(Ts) == 0, + "NoPayload does not accept any payload arguments."); + // Special case for NoPayload where there is a stack or inner window id: + // Because these options would be stored in the payload 'data' object, but + // there is no such object for NoPayload, we convert the marker to another + // type (without user fields in the 'data' object), so that the stack and/or + // inner window id are not lost. + // TODO: Remove this when bug 1646714 lands. + if (aOptions.Stack().GetChunkedBuffer() || + !aOptions.InnerWindowId().IsUnspecified()) { + struct NoPayloadUserData { + static constexpr Span<const char> MarkerTypeName() { + return MakeStringSpan("NoPayloadUserData"); + } + static void StreamJSONMarkerData( + baseprofiler::SpliceableJSONWriter& aWriter) { + // No user payload. + } + static mozilla::MarkerSchema MarkerTypeDisplay() { + using MS = mozilla::MarkerSchema; + MS schema{MS::Location::markerChart, MS::Location::markerTable}; + // No user data to display. + return schema; + } + }; + return MarkerTypeSerialization<NoPayloadUserData>::Serialize( + aBuffer, aName, aCategory, std::move(aOptions)); + } + + // Note that options are first after the entry kind, because they contain + // the thread id, which is handled first to filter markers by threads. + return aBuffer.PutObjects( + ProfileBufferEntryKind::Marker, std::move(aOptions), aName, aCategory, + base_profiler_markers_detail::Streaming::DeserializerTag(0)); + } else { + return MarkerTypeSerialization<MarkerType>::Serialize( + aBuffer, aName, aCategory, std::move(aOptions), aTs...); + } +} + +// Pointer to a function that can capture a backtrace into the provided +// `ProfileChunkedBuffer`, and returns true when successful. +using BacktraceCaptureFunction = bool (*)(ProfileChunkedBuffer&); + +// Add a marker with the given name, options, and arguments to the given buffer. +// Because this may be called from either Base or Gecko Profiler functions, the +// appropriate backtrace-capturing function must also be provided. +template <typename MarkerType, typename... Ts> +ProfileBufferBlockIndex AddMarkerToBuffer( + ProfileChunkedBuffer& aBuffer, const ProfilerString8View& aName, + const MarkerCategory& aCategory, MarkerOptions&& aOptions, + BacktraceCaptureFunction aBacktraceCaptureFunction, const Ts&... aTs) { + if (aOptions.ThreadId().IsUnspecified()) { + // If yet unspecified, set thread to this thread where the marker is added. + aOptions.Set(MarkerThreadId::CurrentThread()); + } + + if (aOptions.IsTimingUnspecified()) { + // If yet unspecified, set timing to this instant of adding the marker. + aOptions.Set(MarkerTiming::InstantNow()); + } + + if (aOptions.Stack().IsCaptureNeeded()) { + // A capture was requested, let's attempt to do it here&now. This avoids a + // lot of allocations that would be necessary if capturing a backtrace + // separately. + // TODO use a local on-stack byte buffer to remove last allocation. + // TODO reduce internal profiler stack levels, see bug 1659872. + ProfileBufferChunkManagerSingle chunkManager( + ProfileBufferChunkManager::scExpectedMaximumStackSize); + ProfileChunkedBuffer chunkedBuffer( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex, chunkManager); + aOptions.StackRef().UseRequestedBacktrace( + aBacktraceCaptureFunction(chunkedBuffer) ? &chunkedBuffer : nullptr); + // This call must be made from here, while chunkedBuffer is in scope. + return AddMarkerWithOptionalStackToBuffer<MarkerType>( + aBuffer, aName, aCategory, std::move(aOptions), aTs...); + } + + return AddMarkerWithOptionalStackToBuffer<MarkerType>( + aBuffer, aName, aCategory, std::move(aOptions), aTs...); +} + +template <typename StackCallback> +[[nodiscard]] bool DeserializeAfterKindAndStream( + ProfileBufferEntryReader& aEntryReader, + baseprofiler::SpliceableJSONWriter& aWriter, int aThreadIdOrZero, + StackCallback&& aStackCallback) { + // Each entry is made up of the following: + // ProfileBufferEntry::Kind::Marker, <- already read by caller + // options, <- next location in entries + // name, + // payload + const MarkerOptions options = aEntryReader.ReadObject<MarkerOptions>(); + if (aThreadIdOrZero != 0 && + options.ThreadId().ThreadId() != aThreadIdOrZero) { + // A specific thread is being read, we're not in it. + return false; + } + // Write the information to JSON with the following schema: + // [name, startTime, endTime, phase, category, data] + aWriter.StartArrayElement(); + { + aWriter.UniqueStringElement(aEntryReader.ReadObject<ProfilerString8View>()); + + const double startTime = options.Timing().GetStartTime(); + aWriter.DoubleElement(startTime); + + const double endTime = options.Timing().GetEndTime(); + aWriter.DoubleElement(endTime); + + aWriter.IntElement(static_cast<int64_t>(options.Timing().MarkerPhase())); + + MarkerCategory category = aEntryReader.ReadObject<MarkerCategory>(); + aWriter.IntElement(static_cast<int64_t>(category.GetCategory())); + + if (const auto tag = + aEntryReader.ReadObject<mozilla::base_profiler_markers_detail:: + Streaming::DeserializerTag>(); + tag != 0) { + aWriter.StartObjectElement(JSONWriter::SingleLineStyle); + { + // Stream "common props". + + // TODO: Move this to top-level tuple, when frontend supports it. + if (!options.InnerWindowId().IsUnspecified()) { + // Here, we are converting uint64_t to double. Both Browsing Context + // 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( + "innerWindowID", + static_cast<double>(options.InnerWindowId().Id())); + } + + // TODO: Move this to top-level tuple, when frontend supports it. + if (ProfileChunkedBuffer* chunkedBuffer = + options.Stack().GetChunkedBuffer(); + chunkedBuffer) { + aWriter.StartObjectProperty("stack"); + { std::forward<StackCallback>(aStackCallback)(*chunkedBuffer); } + aWriter.EndObject(); + } + + // Stream the payload, including the type. + mozilla::base_profiler_markers_detail::Streaming::MarkerDataDeserializer + deserializer = mozilla::base_profiler_markers_detail::Streaming:: + DeserializerForTag(tag); + MOZ_RELEASE_ASSERT(deserializer); + deserializer(aEntryReader, aWriter); + } + aWriter.EndObject(); + } + } + aWriter.EndArray(); + return true; +} + +} // namespace mozilla::base_profiler_markers_detail + +namespace mozilla { + +// ---------------------------------------------------------------------------- +// Serializer, Deserializer: ProfilerStringView<CHAR> + +// The serialization starts with a ULEB128 number that encodes both whether the +// ProfilerStringView is literal (Least Significant Bit = 0) or not (LSB = 1), +// plus the string length (excluding null terminator) in bytes, shifted left by +// 1 bit. Following that number: +// - If literal, the string pointer value. +// - If non-literal, the contents as bytes (excluding null terminator if any). +template <typename CHAR> +struct ProfileBufferEntryWriter::Serializer<ProfilerStringView<CHAR>> { + static Length Bytes(const ProfilerStringView<CHAR>& aString) { + MOZ_RELEASE_ASSERT( + aString.Length() < std::numeric_limits<Length>::max() / 2, + "Double the string length doesn't fit in Length type"); + const Length stringLength = static_cast<Length>(aString.Length()); + if (aString.IsLiteral()) { + // Literal -> Length shifted left and LSB=0, then pointer. + return ULEB128Size(stringLength << 1 | 0u) + + static_cast<ProfileChunkedBuffer::Length>(sizeof(const CHAR*)); + } + // Non-literal -> Length shifted left and LSB=1, then string size in bytes. + return ULEB128Size((stringLength << 1) | 1u) + stringLength * sizeof(CHAR); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const ProfilerStringView<CHAR>& aString) { + MOZ_RELEASE_ASSERT( + aString.Length() < std::numeric_limits<Length>::max() / 2, + "Double the string length doesn't fit in Length type"); + const Length stringLength = static_cast<Length>(aString.Length()); + if (aString.IsLiteral()) { + // Literal -> Length shifted left and LSB=0, then pointer. + aEW.WriteULEB128(stringLength << 1 | 0u); + aEW.WriteObject(WrapProfileBufferRawPointer(aString.Data())); + return; + } + // Non-literal -> Length shifted left and LSB=1, then string size in bytes. + aEW.WriteULEB128(stringLength << 1 | 1u); + aEW.WriteBytes(aString.Data(), stringLength * sizeof(CHAR)); + } +}; + +template <typename CHAR> +struct ProfileBufferEntryReader::Deserializer<ProfilerStringView<CHAR>> { + static void ReadInto(ProfileBufferEntryReader& aER, + ProfilerStringView<CHAR>& aString) { + const Length lengthAndIsLiteral = aER.ReadULEB128<Length>(); + const Length stringLength = lengthAndIsLiteral >> 1; + if ((lengthAndIsLiteral & 1u) == 0u) { + // LSB==0 -> Literal string, read the string pointer. + aString.mStringView = std::basic_string_view<CHAR>( + aER.ReadObject<const CHAR*>(), stringLength); + aString.mOwnership = ProfilerStringView<CHAR>::Ownership::Literal; + return; + } + // LSB==1 -> Not a literal string, allocate a buffer to store the string + // (plus terminal, for safety), and give it to the ProfilerStringView; Note + // that this is a secret use of ProfilerStringView, which is intended to + // only be used between deserialization and JSON streaming. + CHAR* buffer = new CHAR[stringLength + 1]; + aER.ReadBytes(buffer, stringLength * sizeof(CHAR)); + buffer[stringLength] = CHAR(0); + aString.mStringView = std::basic_string_view<CHAR>(buffer, stringLength); + aString.mOwnership = + ProfilerStringView<CHAR>::Ownership::OwnedThroughStringView; + } + + static ProfilerStringView<CHAR> Read(ProfileBufferEntryReader& aER) { + const Length lengthAndIsLiteral = aER.ReadULEB128<Length>(); + const Length stringLength = lengthAndIsLiteral >> 1; + if ((lengthAndIsLiteral & 1u) == 0u) { + // LSB==0 -> Literal string, read the string pointer. + return ProfilerStringView<CHAR>( + aER.ReadObject<const CHAR*>(), stringLength, + ProfilerStringView<CHAR>::Ownership::Literal); + } + // LSB==1 -> Not a literal string, allocate a buffer to store the string + // (plus terminal, for safety), and give it to the ProfilerStringView; Note + // that this is a secret use of ProfilerStringView, which is intended to + // only be used between deserialization and JSON streaming. + CHAR* buffer = new CHAR[stringLength + 1]; + aER.ReadBytes(buffer, stringLength * sizeof(CHAR)); + buffer[stringLength] = CHAR(0); + return ProfilerStringView<CHAR>( + buffer, stringLength, + ProfilerStringView<CHAR>::Ownership::OwnedThroughStringView); + } +}; + +// Serializer, Deserializer: MarkerCategory + +// The serialization contains both category numbers encoded as ULEB128. +template <> +struct ProfileBufferEntryWriter::Serializer<MarkerCategory> { + static Length Bytes(const MarkerCategory& aCategory) { + return ULEB128Size(static_cast<uint32_t>(aCategory.CategoryPair())); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const MarkerCategory& aCategory) { + aEW.WriteULEB128(static_cast<uint32_t>(aCategory.CategoryPair())); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<MarkerCategory> { + static void ReadInto(ProfileBufferEntryReader& aER, + MarkerCategory& aCategory) { + aCategory = Read(aER); + } + + static MarkerCategory Read(ProfileBufferEntryReader& aER) { + return MarkerCategory(static_cast<baseprofiler::ProfilingCategoryPair>( + aER.ReadULEB128<uint32_t>())); + } +}; + +// ---------------------------------------------------------------------------- +// Serializer, Deserializer: MarkerTiming + +// The serialization starts with the marker phase, followed by one or two +// timestamps as needed. +template <> +struct ProfileBufferEntryWriter::Serializer<MarkerTiming> { + static Length Bytes(const MarkerTiming& aTiming) { + MOZ_ASSERT(!aTiming.IsUnspecified()); + const auto phase = aTiming.MarkerPhase(); + switch (phase) { + case MarkerTiming::Phase::Instant: + return SumBytes(phase, aTiming.StartTime()); + case MarkerTiming::Phase::Interval: + return SumBytes(phase, aTiming.StartTime(), aTiming.EndTime()); + case MarkerTiming::Phase::IntervalStart: + return SumBytes(phase, aTiming.StartTime()); + case MarkerTiming::Phase::IntervalEnd: + return SumBytes(phase, aTiming.EndTime()); + default: + MOZ_RELEASE_ASSERT(phase == MarkerTiming::Phase::Instant || + phase == MarkerTiming::Phase::Interval || + phase == MarkerTiming::Phase::IntervalStart || + phase == MarkerTiming::Phase::IntervalEnd); + return 0; // Only to avoid build errors. + } + } + + static void Write(ProfileBufferEntryWriter& aEW, + const MarkerTiming& aTiming) { + MOZ_ASSERT(!aTiming.IsUnspecified()); + const auto phase = aTiming.MarkerPhase(); + switch (phase) { + case MarkerTiming::Phase::Instant: + aEW.WriteObjects(phase, aTiming.StartTime()); + return; + case MarkerTiming::Phase::Interval: + aEW.WriteObjects(phase, aTiming.StartTime(), aTiming.EndTime()); + return; + case MarkerTiming::Phase::IntervalStart: + aEW.WriteObjects(phase, aTiming.StartTime()); + return; + case MarkerTiming::Phase::IntervalEnd: + aEW.WriteObjects(phase, aTiming.EndTime()); + return; + default: + MOZ_RELEASE_ASSERT(phase == MarkerTiming::Phase::Instant || + phase == MarkerTiming::Phase::Interval || + phase == MarkerTiming::Phase::IntervalStart || + phase == MarkerTiming::Phase::IntervalEnd); + return; + } + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<MarkerTiming> { + static void ReadInto(ProfileBufferEntryReader& aER, MarkerTiming& aTiming) { + aTiming.mPhase = aER.ReadObject<MarkerTiming::Phase>(); + switch (aTiming.mPhase) { + case MarkerTiming::Phase::Instant: + aTiming.mStartTime = aER.ReadObject<TimeStamp>(); + aTiming.mEndTime = TimeStamp{}; + break; + case MarkerTiming::Phase::Interval: + aTiming.mStartTime = aER.ReadObject<TimeStamp>(); + aTiming.mEndTime = aER.ReadObject<TimeStamp>(); + break; + case MarkerTiming::Phase::IntervalStart: + aTiming.mStartTime = aER.ReadObject<TimeStamp>(); + aTiming.mEndTime = TimeStamp{}; + break; + case MarkerTiming::Phase::IntervalEnd: + aTiming.mStartTime = TimeStamp{}; + aTiming.mEndTime = aER.ReadObject<TimeStamp>(); + break; + default: + MOZ_RELEASE_ASSERT(aTiming.mPhase == MarkerTiming::Phase::Instant || + aTiming.mPhase == MarkerTiming::Phase::Interval || + aTiming.mPhase == + MarkerTiming::Phase::IntervalStart || + aTiming.mPhase == MarkerTiming::Phase::IntervalEnd); + break; + } + } + + static MarkerTiming Read(ProfileBufferEntryReader& aER) { + TimeStamp start; + TimeStamp end; + auto phase = aER.ReadObject<MarkerTiming::Phase>(); + switch (phase) { + case MarkerTiming::Phase::Instant: + start = aER.ReadObject<TimeStamp>(); + break; + case MarkerTiming::Phase::Interval: + start = aER.ReadObject<TimeStamp>(); + end = aER.ReadObject<TimeStamp>(); + break; + case MarkerTiming::Phase::IntervalStart: + start = aER.ReadObject<TimeStamp>(); + break; + case MarkerTiming::Phase::IntervalEnd: + end = aER.ReadObject<TimeStamp>(); + break; + default: + MOZ_RELEASE_ASSERT(phase == MarkerTiming::Phase::Instant || + phase == MarkerTiming::Phase::Interval || + phase == MarkerTiming::Phase::IntervalStart || + phase == MarkerTiming::Phase::IntervalEnd); + break; + } + return MarkerTiming(start, end, phase); + } +}; + +// ---------------------------------------------------------------------------- +// Serializer, Deserializer: MarkerStack + +// The serialization only contains the `ProfileChunkedBuffer` from the +// backtrace; if there is no backtrace or if it's empty, this will implicitly +// store a nullptr (see +// `ProfileBufferEntryWriter::Serializer<ProfilerChunkedBuffer*>`). +template <> +struct ProfileBufferEntryWriter::Serializer<MarkerStack> { + static Length Bytes(const MarkerStack& aStack) { + return SumBytes(aStack.GetChunkedBuffer()); + } + + static void Write(ProfileBufferEntryWriter& aEW, const MarkerStack& aStack) { + aEW.WriteObject(aStack.GetChunkedBuffer()); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<MarkerStack> { + static void ReadInto(ProfileBufferEntryReader& aER, MarkerStack& aStack) { + aStack = Read(aER); + } + + static MarkerStack Read(ProfileBufferEntryReader& aER) { + return MarkerStack(aER.ReadObject<UniquePtr<ProfileChunkedBuffer>>()); + } +}; + +// ---------------------------------------------------------------------------- +// Serializer, Deserializer: MarkerOptions + +// The serialization contains all members (either trivially-copyable, or they +// provide their specialization above). +template <> +struct ProfileBufferEntryWriter::Serializer<MarkerOptions> { + static Length Bytes(const MarkerOptions& aOptions) { + return SumBytes(aOptions.ThreadId(), aOptions.Timing(), aOptions.Stack(), + aOptions.InnerWindowId()); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const MarkerOptions& aOptions) { + aEW.WriteObjects(aOptions.ThreadId(), aOptions.Timing(), aOptions.Stack(), + aOptions.InnerWindowId()); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<MarkerOptions> { + static void ReadInto(ProfileBufferEntryReader& aER, MarkerOptions& aOptions) { + aER.ReadIntoObjects(aOptions.mThreadId, aOptions.mTiming, aOptions.mStack, + aOptions.mInnerWindowId); + } + + static MarkerOptions Read(ProfileBufferEntryReader& aER) { + MarkerOptions options; + ReadInto(aER, options); + return options; + } +}; + +} // namespace mozilla + +#endif // MOZ_GECKO_PROFILER + +#endif // BaseProfilerMarkersDetail_h diff --git a/mozglue/baseprofiler/public/BaseProfilerMarkersPrerequisites.h b/mozglue/baseprofiler/public/BaseProfilerMarkersPrerequisites.h new file mode 100644 index 0000000000..aa85b41896 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerMarkersPrerequisites.h @@ -0,0 +1,866 @@ +/* -*- 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 header contains basic definitions required to create marker types, and +// to add markers to the profiler buffers. +// +// In most cases, #include "mozilla/BaseProfilerMarkers.h" instead, or +// #include "mozilla/BaseProfilerMarkerTypes.h" for common marker types. + +#ifndef BaseProfilerMarkersPrerequisites_h +#define BaseProfilerMarkersPrerequisites_h + +#ifdef MOZ_GECKO_PROFILER + +# include "BaseProfilingCategory.h" +# include "mozilla/Maybe.h" +# include "mozilla/ProfileChunkedBuffer.h" +# include "mozilla/TimeStamp.h" +# include "mozilla/UniquePtr.h" +# include "mozilla/Variant.h" + +# include <initializer_list> +# include <string_view> +# include <string> +# include <type_traits> +# include <utility> +# include <vector> + +// TODO: Move common stuff to shared header instead. +# include "BaseProfiler.h" + +namespace mozilla { + +// Return a NotNull<const CHAR*> pointing at the literal empty string `""`. +template <typename CHAR> +constexpr const CHAR* LiteralEmptyStringPointer() { + static_assert(std::is_same_v<CHAR, char> || std::is_same_v<CHAR, char16_t>, + "Only char and char16_t are supported in Firefox"); + if constexpr (std::is_same_v<CHAR, char>) { + return ""; + } + if constexpr (std::is_same_v<CHAR, char16_t>) { + return u""; + } +} + +// Return a string_view<CHAR> pointing at the literal empty string. +template <typename CHAR> +constexpr std::basic_string_view<CHAR> LiteralEmptyStringView() { + static_assert(std::is_same_v<CHAR, char> || std::is_same_v<CHAR, char16_t>, + "Only char and char16_t are supported in Firefox"); + // Use `operator""sv()` from <string_view>. + using namespace std::literals::string_view_literals; + if constexpr (std::is_same_v<CHAR, char>) { + return ""sv; + } + if constexpr (std::is_same_v<CHAR, char16_t>) { + return u""sv; + } +} + +// General string view, optimized for short on-stack life before serialization, +// and between deserialization and JSON-streaming. +template <typename CHAR> +class MOZ_STACK_CLASS ProfilerStringView { + public: + // Default constructor points at "" (literal empty string). + constexpr ProfilerStringView() = default; + + // Don't allow copy. + ProfilerStringView(const ProfilerStringView&) = delete; + ProfilerStringView& operator=(const ProfilerStringView&) = delete; + + // Allow move. For consistency the moved-from string is always reset to "". + constexpr ProfilerStringView(ProfilerStringView&& aOther) + : mStringView(std::move(aOther.mStringView)), + mOwnership(aOther.mOwnership) { + if (mOwnership == Ownership::OwnedThroughStringView) { + // We now own the buffer, make the other point at the literal "". + aOther.mStringView = LiteralEmptyStringView<CHAR>(); + aOther.mOwnership = Ownership::Literal; + } + } + constexpr ProfilerStringView& operator=(ProfilerStringView&& aOther) { + mStringView = std::move(aOther.mStringView); + mOwnership = aOther.mOwnership; + if (mOwnership == Ownership::OwnedThroughStringView) { + // We now own the buffer, make the other point at the literal "". + aOther.mStringView = LiteralEmptyStringView<CHAR>(); + aOther.mOwnership = Ownership::Literal; + } + return *this; + } + + ~ProfilerStringView() { + if (MOZ_UNLIKELY(mOwnership == Ownership::OwnedThroughStringView)) { + // We own the buffer pointed at by mStringView, destroy it. + // This is only used between deserialization and streaming. + delete mStringView.data(); + } + } + + // Implicit construction from nullptr, points at "" (literal empty string). + constexpr MOZ_IMPLICIT ProfilerStringView(decltype(nullptr)) {} + + // Implicit constructor from a literal string. + template <size_t Np1> + constexpr MOZ_IMPLICIT ProfilerStringView(const CHAR (&aLiteralString)[Np1]) + : ProfilerStringView(aLiteralString, Np1 - 1, Ownership::Literal) {} + + // Constructor from a non-literal string. + constexpr ProfilerStringView(const CHAR* aString, size_t aLength) + : ProfilerStringView(aString, aLength, Ownership::Reference) {} + + // Implicit constructor from a string_view. + constexpr MOZ_IMPLICIT ProfilerStringView( + const std::basic_string_view<CHAR>& aStringView) + : ProfilerStringView(aStringView.data(), aStringView.length(), + Ownership::Reference) {} + + // Implicit constructor from an expiring string_view. We assume that the + // pointed-at string will outlive this ProfilerStringView. + constexpr MOZ_IMPLICIT ProfilerStringView( + std::basic_string_view<CHAR>&& aStringView) + : ProfilerStringView(aStringView.data(), aStringView.length(), + Ownership::Reference) {} + + // Implicit constructor from std::string. + constexpr MOZ_IMPLICIT ProfilerStringView( + const std::basic_string<CHAR>& aString) + : ProfilerStringView(aString.data(), aString.length(), + Ownership::Reference) {} + + // Construction from a raw pointer to a null-terminated string. + // This is a named class-static function to make it more obvious where work is + // being done (to determine the string length), and encourage users to instead + // provide a length, if already known. + // TODO: Find callers and convert them to constructor instead if possible. + static constexpr ProfilerStringView WrapNullTerminatedString( + const CHAR* aString) { + return ProfilerStringView( + aString, aString ? std::char_traits<CHAR>::length(aString) : 0, + Ownership::Reference); + } + + // Implicit constructor for an object with member functions `Data()` + // `Length()`, and `IsLiteral()`, common in xpcom strings. + template < + typename String, + typename DataReturnType = decltype(std::declval<const String>().Data()), + typename LengthReturnType = + decltype(std::declval<const String>().Length()), + typename IsLiteralReturnType = + decltype(std::declval<const String>().IsLiteral()), + typename = + std::enable_if_t<std::is_convertible_v<DataReturnType, const CHAR*> && + std::is_integral_v<LengthReturnType> && + std::is_same_v<IsLiteralReturnType, bool>>> + constexpr MOZ_IMPLICIT ProfilerStringView(const String& aString) + : ProfilerStringView( + static_cast<const CHAR*>(aString.Data()), aString.Length(), + aString.IsLiteral() ? Ownership::Literal : Ownership::Reference) {} + + [[nodiscard]] constexpr const std::basic_string_view<CHAR>& StringView() + const { + return mStringView; + } + + [[nodiscard]] constexpr const CHAR* Data() const { + return mStringView.data(); + } + + [[nodiscard]] constexpr size_t Length() const { return mStringView.length(); } + + [[nodiscard]] constexpr bool IsLiteral() const { + return mOwnership == Ownership::Literal; + } + [[nodiscard]] constexpr bool IsReference() const { + return mOwnership == Ownership::Reference; + } + // No `IsOwned...()` because it's a secret, only used internally! + + [[nodiscard]] operator Span<const CHAR>() const { + return Span<const CHAR>(Data(), Length()); + } + + private: + enum class Ownership { Literal, Reference, OwnedThroughStringView }; + + // Allow deserializer to store anything here. + friend ProfileBufferEntryReader::Deserializer<ProfilerStringView>; + + constexpr ProfilerStringView(const CHAR* aString, size_t aLength, + Ownership aOwnership) + : mStringView(aString ? std::basic_string_view<CHAR>(aString, aLength) + : LiteralEmptyStringView<CHAR>()), + mOwnership(aString ? aOwnership : Ownership::Literal) {} + + // String view to an outside string (literal or reference). + // We may actually own the pointed-at buffer, but it is only used internally + // between deserialization and JSON streaming. + std::basic_string_view<CHAR> mStringView = LiteralEmptyStringView<CHAR>(); + + Ownership mOwnership = Ownership::Literal; +}; + +using ProfilerString8View = ProfilerStringView<char>; +using ProfilerString16View = ProfilerStringView<char16_t>; + +// This compulsory marker parameter contains the required category information. +class MarkerCategory { + public: + // Constructor from category pair (includes both super- and sub-categories). + constexpr explicit MarkerCategory( + baseprofiler::ProfilingCategoryPair aCategoryPair) + : mCategoryPair(aCategoryPair) {} + + // Returns the stored category pair. + constexpr baseprofiler::ProfilingCategoryPair CategoryPair() const { + return mCategoryPair; + } + + // Returns the super-category from the stored category pair. + baseprofiler::ProfilingCategory GetCategory() const { + return GetProfilingCategoryPairInfo(mCategoryPair).mCategory; + } + + private: + baseprofiler::ProfilingCategoryPair mCategoryPair = + baseprofiler::ProfilingCategoryPair::OTHER; +}; + +namespace baseprofiler::category { + +// Each category pair name constructs a MarkerCategory. +// E.g.: mozilla::baseprofiler::category::OTHER_Profiling +// Profiler macros will take the category name alone without namespace. +// E.g.: `PROFILER_MARKER_UNTYPED("name", OTHER_Profiling)` +# define CATEGORY_ENUM_BEGIN_CATEGORY(name, labelAsString, color) +# define CATEGORY_ENUM_SUBCATEGORY(supercategory, name, labelAsString) \ + static constexpr MarkerCategory name{ProfilingCategoryPair::name}; +# define CATEGORY_ENUM_END_CATEGORY +MOZ_PROFILING_CATEGORY_LIST(CATEGORY_ENUM_BEGIN_CATEGORY, + CATEGORY_ENUM_SUBCATEGORY, + CATEGORY_ENUM_END_CATEGORY) +# undef CATEGORY_ENUM_BEGIN_CATEGORY +# undef CATEGORY_ENUM_SUBCATEGORY +# undef CATEGORY_ENUM_END_CATEGORY + +// Import `MarkerCategory` into this namespace. This will allow using this type +// dynamically in macros that prepend `::mozilla::baseprofiler::category::` to +// the given category, e.g.: +// `PROFILER_MARKER_UNTYPED("name", MarkerCategory(...))` +using MarkerCategory = ::mozilla::MarkerCategory; + +} // namespace baseprofiler::category + +// The classes below are all embedded in a `MarkerOptions` object. +class MarkerOptions; + +// This marker option captures a given thread id. +// If left unspecified (by default construction) during the add-marker call, the +// current thread id will be used then. +class MarkerThreadId { + public: + // Default constructor, keeps the thread id unspecified. + constexpr MarkerThreadId() = default; + + // Constructor from a given thread id. + constexpr explicit MarkerThreadId(int aThreadId) : mThreadId(aThreadId) {} + + // Use the current thread's id. + static MarkerThreadId CurrentThread() { + return MarkerThreadId(baseprofiler::profiler_current_thread_id()); + } + + // Use the main thread's id. This can be useful to record a marker from a + // possibly-unregistered thread, and display it in the main thread track. + static MarkerThreadId MainThread() { + return MarkerThreadId(baseprofiler::profiler_main_thread_id()); + } + + [[nodiscard]] constexpr int ThreadId() const { return mThreadId; } + + [[nodiscard]] constexpr bool IsUnspecified() const { return mThreadId == 0; } + + private: + int mThreadId = 0; +}; + +// This marker option contains marker timing information. +// This class encapsulates the logic for correctly storing a marker based on its +// Use the static methods to create the MarkerTiming. This is a transient object +// that is being used to enforce the constraints of the combinations of the +// data. +class MarkerTiming { + public: + // The following static methods are used to create the MarkerTiming based on + // the type that it is. + + static MarkerTiming InstantAt(const TimeStamp& aTime) { + MOZ_ASSERT(!aTime.IsNull(), "Time is null for an instant marker."); + return MarkerTiming{aTime, TimeStamp{}, MarkerTiming::Phase::Instant}; + } + + static MarkerTiming InstantNow() { + return InstantAt(TimeStamp::NowUnfuzzed()); + } + + static MarkerTiming Interval(const TimeStamp& aStartTime, + const TimeStamp& aEndTime) { + MOZ_ASSERT(!aStartTime.IsNull(), + "Start time is null for an interval marker."); + MOZ_ASSERT(!aEndTime.IsNull(), "End time is null for an interval marker."); + return MarkerTiming{aStartTime, aEndTime, MarkerTiming::Phase::Interval}; + } + + static MarkerTiming IntervalUntilNowFrom(const TimeStamp& aStartTime) { + return Interval(aStartTime, TimeStamp::NowUnfuzzed()); + } + + static MarkerTiming IntervalStart( + const TimeStamp& aTime = TimeStamp::NowUnfuzzed()) { + MOZ_ASSERT(!aTime.IsNull(), "Time is null for an interval start marker."); + return MarkerTiming{aTime, TimeStamp{}, MarkerTiming::Phase::IntervalStart}; + } + + static MarkerTiming IntervalEnd( + const TimeStamp& aTime = TimeStamp::NowUnfuzzed()) { + MOZ_ASSERT(!aTime.IsNull(), "Time is null for an interval end marker."); + return MarkerTiming{TimeStamp{}, aTime, MarkerTiming::Phase::IntervalEnd}; + } + + // Set the interval end in this timing. + // If there was already a start time, this makes it a full interval. + void SetIntervalEnd(const TimeStamp& aTime = TimeStamp::NowUnfuzzed()) { + MOZ_ASSERT(!aTime.IsNull(), "Time is null for an interval end marker."); + mEndTime = aTime; + mPhase = mStartTime.IsNull() ? Phase::IntervalEnd : Phase::Interval; + } + + [[nodiscard]] const TimeStamp& StartTime() const { return mStartTime; } + [[nodiscard]] const TimeStamp& EndTime() const { return mEndTime; } + + enum class Phase : uint8_t { + Instant = 0, + Interval = 1, + IntervalStart = 2, + IntervalEnd = 3, + }; + + [[nodiscard]] Phase MarkerPhase() const { + MOZ_ASSERT(!IsUnspecified()); + return mPhase; + } + + // The following getter methods are used to put the value into the buffer for + // storage. + [[nodiscard]] double GetStartTime() const { + MOZ_ASSERT(!IsUnspecified()); + // If mStartTime is null (e.g., for IntervalEnd), this will output 0.0 as + // expected. + return MarkerTiming::timeStampToDouble(mStartTime); + } + + [[nodiscard]] double GetEndTime() const { + MOZ_ASSERT(!IsUnspecified()); + // If mEndTime is null (e.g., for Instant or IntervalStart), this will + // output 0.0 as expected. + return MarkerTiming::timeStampToDouble(mEndTime); + } + + [[nodiscard]] uint8_t GetPhase() const { + MOZ_ASSERT(!IsUnspecified()); + return static_cast<uint8_t>(mPhase); + } + + private: + friend ProfileBufferEntryWriter::Serializer<MarkerTiming>; + friend ProfileBufferEntryReader::Deserializer<MarkerTiming>; + friend MarkerOptions; + + // Default timing leaves it internally "unspecified", serialization getters + // and add-marker functions will default to `InstantNow()`. + constexpr MarkerTiming() = default; + + // This should only be used by internal profiler code. + [[nodiscard]] bool IsUnspecified() const { + return mStartTime.IsNull() && mEndTime.IsNull(); + } + + // Full constructor, used by static factory functions. + constexpr MarkerTiming(const TimeStamp& aStartTime, const TimeStamp& aEndTime, + Phase aPhase) + : mStartTime(aStartTime), mEndTime(aEndTime), mPhase(aPhase) {} + + static double timeStampToDouble(const TimeStamp& time) { + if (time.IsNull()) { + // The Phase lets us know not to use this value. + return 0; + } + return (time - TimeStamp::ProcessCreation()).ToMilliseconds(); + } + + TimeStamp mStartTime; + TimeStamp mEndTime; + Phase mPhase = Phase::Instant; +}; + +// This marker option allows three cases: +// - By default, no stacks are captured. +// - The caller can request a stack capture, and the add-marker code will take +// care of it in the most efficient way. +// - The caller can still provide an existing backtrace, for cases where a +// marker reports something that happened elsewhere. +class MarkerStack { + public: + // Default constructor, no capture. + constexpr MarkerStack() = default; + + // Disallow copy. + MarkerStack(const MarkerStack&) = delete; + MarkerStack& operator=(const MarkerStack&) = delete; + + // Allow move. + MarkerStack(MarkerStack&& aOther) + : mIsCaptureRequested(aOther.mIsCaptureRequested), + mOptionalChunkedBufferStorage( + std::move(aOther.mOptionalChunkedBufferStorage)), + mChunkedBuffer(aOther.mChunkedBuffer) { + AssertInvariants(); + aOther.Clear(); + } + MarkerStack& operator=(MarkerStack&& aOther) { + mIsCaptureRequested = aOther.mIsCaptureRequested; + mOptionalChunkedBufferStorage = + std::move(aOther.mOptionalChunkedBufferStorage); + mChunkedBuffer = aOther.mChunkedBuffer; + AssertInvariants(); + aOther.Clear(); + return *this; + } + + // Take ownership of a backtrace. If null or empty, equivalent to NoStack(). + explicit MarkerStack(UniquePtr<ProfileChunkedBuffer>&& aExternalChunkedBuffer) + : mIsCaptureRequested(false), + mOptionalChunkedBufferStorage( + (!aExternalChunkedBuffer || aExternalChunkedBuffer->IsEmpty()) + ? nullptr + : std::move(aExternalChunkedBuffer)), + mChunkedBuffer(mOptionalChunkedBufferStorage.get()) { + AssertInvariants(); + } + + // Use an existing backtrace stored elsewhere, which the user must guarantee + // is alive during the add-marker call. If empty, equivalent to NoStack(). + explicit MarkerStack(ProfileChunkedBuffer& aExternalChunkedBuffer) + : mIsCaptureRequested(false), + mChunkedBuffer(aExternalChunkedBuffer.IsEmpty() + ? nullptr + : &aExternalChunkedBuffer) { + AssertInvariants(); + } + + // Don't capture a stack in this marker. + static MarkerStack NoStack() { return MarkerStack(false); } + + // Capture a stack when adding this marker. + static MarkerStack Capture() { + // Actual capture will be handled inside profiler_add_marker. + return MarkerStack(true); + } + + // Optionally capture a stack, useful for avoiding long-winded ternaries. + static MarkerStack MaybeCapture(bool aDoCapture) { + return MarkerStack(aDoCapture); + } + + // Use an existing backtrace stored elsewhere, which the user must guarantee + // is alive during the add-marker call. If empty, equivalent to NoStack(). + static MarkerStack UseBacktrace( + ProfileChunkedBuffer& aExternalChunkedBuffer) { + return MarkerStack(aExternalChunkedBuffer); + } + + // Take ownership of a backtrace previously captured with + // `profiler_capture_backtrace()`. If null, equivalent to NoStack(). + static MarkerStack TakeBacktrace( + UniquePtr<ProfileChunkedBuffer>&& aExternalChunkedBuffer) { + return MarkerStack(std::move(aExternalChunkedBuffer)); + } + + [[nodiscard]] bool IsCaptureNeeded() const { + // If the chunked buffer already contains something, consider the capture + // request already fulfilled. + return mIsCaptureRequested; + } + + ProfileChunkedBuffer* GetChunkedBuffer() const { return mChunkedBuffer; } + + // Use backtrace after a request. If null, equivalent to NoStack(). + void UseRequestedBacktrace(ProfileChunkedBuffer* aExternalChunkedBuffer) { + MOZ_RELEASE_ASSERT(IsCaptureNeeded()); + mIsCaptureRequested = false; + if (aExternalChunkedBuffer && !aExternalChunkedBuffer->IsEmpty()) { + // We only need to use the provided buffer if it is not empty. + mChunkedBuffer = aExternalChunkedBuffer; + } + AssertInvariants(); + } + + void Clear() { + mIsCaptureRequested = false; + mOptionalChunkedBufferStorage.reset(); + mChunkedBuffer = nullptr; + AssertInvariants(); + } + + private: + explicit MarkerStack(bool aIsCaptureRequested) + : mIsCaptureRequested(aIsCaptureRequested) { + AssertInvariants(); + } + + // This should be called after every constructor and non-const function. + void AssertInvariants() const { +# ifdef DEBUG + if (mIsCaptureRequested) { + MOZ_ASSERT(!mOptionalChunkedBufferStorage, + "We should not hold a buffer when capture is requested"); + MOZ_ASSERT(!mChunkedBuffer, + "We should not point at a buffer when capture is requested"); + } else { + if (mOptionalChunkedBufferStorage) { + MOZ_ASSERT(mChunkedBuffer == mOptionalChunkedBufferStorage.get(), + "Non-null mOptionalChunkedBufferStorage must be pointed-at " + "by mChunkedBuffer"); + } + if (mChunkedBuffer) { + MOZ_ASSERT(!mChunkedBuffer->IsEmpty(), + "Non-null mChunkedBuffer must not be empty"); + } + } +# endif // DEBUG + } + + // True if a capture is requested when marker is added to the profile buffer. + bool mIsCaptureRequested = false; + + // Optional storage for the backtrace, in case it was captured before the + // add-marker call. + UniquePtr<ProfileChunkedBuffer> mOptionalChunkedBufferStorage; + + // If not null, this points to the backtrace. It may point to a backtrace + // temporarily stored on the stack, or to mOptionalChunkedBufferStorage. + ProfileChunkedBuffer* mChunkedBuffer = nullptr; +}; + +// This marker option captures a given inner window id. +class MarkerInnerWindowId { + public: + // Default constructor, it leaves the id unspecified. + constexpr MarkerInnerWindowId() = default; + + // Constructor with a specified inner window id. + constexpr explicit MarkerInnerWindowId(uint64_t i) : mInnerWindowId(i) {} + + // Constructor with either specified inner window id or Nothing. + constexpr explicit MarkerInnerWindowId(const Maybe<uint64_t>& i) + : mInnerWindowId(i.valueOr(scNoId)) {} + + // Explicit option with unspecified id. + constexpr static MarkerInnerWindowId NoId() { return MarkerInnerWindowId{}; } + + [[nodiscard]] bool IsUnspecified() const { return mInnerWindowId == scNoId; } + + [[nodiscard]] constexpr uint64_t Id() const { return mInnerWindowId; } + + private: + static constexpr uint64_t scNoId = 0; + uint64_t mInnerWindowId = scNoId; +}; + +// This class combines each of the possible marker options above. +class MarkerOptions { + public: + // Constructor from individual options (including none). + // Implicit to allow `{}` and one option type as-is. + // Options that are not provided here are defaulted. In particular, timing + // defaults to `MarkerTiming::InstantNow()` when the marker is recorded. + template <typename... Options> + MOZ_IMPLICIT MarkerOptions(Options&&... aOptions) { + (Set(std::forward<Options>(aOptions)), ...); + } + + // Disallow copy. + MarkerOptions(const MarkerOptions&) = delete; + MarkerOptions& operator=(const MarkerOptions&) = delete; + + // Allow move. + MarkerOptions(MarkerOptions&&) = default; + MarkerOptions& operator=(MarkerOptions&&) = default; + + // The embedded `MarkerTiming` hasn't been specified yet. + [[nodiscard]] bool IsTimingUnspecified() const { + return mTiming.IsUnspecified(); + } + + // Each option may be added in a chain by e.g.: + // `options.Set(MarkerThreadId(123)).Set(MarkerTiming::IntervalEnd())`. + // When passed to an add-marker function, it must be an rvalue, either created + // on the spot, or `std::move`d from storage, e.g.: + // `PROFILER_MARKER_UNTYPED("...", std::move(options).Set(...))`; + // + // Options can be read by their name (without "Marker"), e.g.: `o.ThreadId()`. + // Add "Ref" for a non-const reference, e.g.: `o.ThreadIdRef() = ...;` +# define FUNCTIONS_ON_MEMBER(NAME) \ + MarkerOptions& Set(Marker##NAME&& a##NAME)& { \ + m##NAME = std::move(a##NAME); \ + return *this; \ + } \ + \ + MarkerOptions&& Set(Marker##NAME&& a##NAME)&& { \ + m##NAME = std::move(a##NAME); \ + return std::move(*this); \ + } \ + \ + const Marker##NAME& NAME() const { return m##NAME; } \ + \ + Marker##NAME& NAME##Ref() { return m##NAME; } + + FUNCTIONS_ON_MEMBER(ThreadId); + FUNCTIONS_ON_MEMBER(Timing); + FUNCTIONS_ON_MEMBER(Stack); + FUNCTIONS_ON_MEMBER(InnerWindowId); +# undef FUNCTIONS_ON_MEMBER + + private: + friend ProfileBufferEntryReader::Deserializer<MarkerOptions>; + + MarkerThreadId mThreadId; + MarkerTiming mTiming; + MarkerStack mStack; + MarkerInnerWindowId mInnerWindowId; +}; + +} // namespace mozilla + +namespace mozilla::baseprofiler::markers { + +// Default marker payload types, with no extra information, not even a marker +// type and payload. This is intended for label-only markers. +struct NoPayload final {}; + +} // namespace mozilla::baseprofiler::markers + +namespace mozilla { + +class JSONWriter; + +// This class collects all the information necessary to stream the JSON schema +// that informs the front-end how to display a type of markers. +// It will be created and populated in `MarkerTypeDisplay()` functions in each +// marker type definition, see Add/Set functions. +class MarkerSchema { + public: + enum class Location : unsigned { + markerChart, + markerTable, + // This adds markers to the main marker timeline in the header. + timelineOverview, + // In the timeline, this is a section that breaks out markers that are + // related to memory. When memory counters are enabled, this is its own + // track, otherwise it is displayed with the main thread. + timelineMemory, + // This adds markers to the IPC timeline area in the header. + timelineIPC, + // This adds markers to the FileIO timeline area in the header. + timelineFileIO, + // TODO - This is not supported yet. + stackChart + }; + + // Used as constructor parameter, to explicitly specify that the location (and + // other display options) are handled as a special case in the front-end. + // In this case, *no* schema will be output for this type. + struct SpecialFrontendLocation {}; + + enum class Format { + // ---------------------------------------------------- + // String types. + + // Show the URL, and handle PII sanitization + url, + // Show the file path, and handle PII sanitization. + filePath, + // Important, do not put URL or file path information here, as it will not + // be sanitized. Please be careful with including other types of PII here as + // well. + // e.g. "Label: Some String" + string, + + // ---------------------------------------------------- + // Numeric types + + // For time data that represents a duration of time. + // e.g. "Label: 5s, 5ms, 5μs" + duration, + // Data that happened at a specific time, relative to the start of the + // profile. e.g. "Label: 15.5s, 20.5ms, 30.5μs" + time, + // The following are alternatives to display a time only in a specific unit + // of time. + seconds, // "Label: 5s" + milliseconds, // "Label: 5ms" + microseconds, // "Label: 5μs" + nanoseconds, // "Label: 5ns" + // e.g. "Label: 5.55mb, 5 bytes, 312.5kb" + bytes, + // This should be a value between 0 and 1. + // "Label: 50%" + percentage, + // The integer should be used for generic representations of numbers. + // Do not use it for time information. + // "Label: 52, 5,323, 1,234,567" + integer, + // The decimal should be used for generic representations of numbers. + // Do not use it for time information. + // "Label: 52.23, 0.0054, 123,456.78" + decimal + }; + + enum class Searchable { notSearchable, searchable }; + + // Marker schema, with a non-empty list of locations where markers should be + // shown. + // Tech note: Even though `aLocations` are templated arguments, they are + // assigned to an `enum class` object, so they can only be of that enum type. + template <typename... Locations> + explicit MarkerSchema(Location aLocation, Locations... aLocations) + : mLocations{aLocation, aLocations...} {} + + // Marker schema for types that have special frontend handling. + // Nothing else should be set in this case. + // Implicit to allow quick return from MarkerTypeDisplay functions. + MOZ_IMPLICIT MarkerSchema(SpecialFrontendLocation) {} + + // Caller must specify location(s) or SpecialFrontendLocation above. + MarkerSchema() = delete; + + // Optional labels in the marker chart, the chart tooltip, and the marker + // table. If not provided, the marker "name" will be used. The given string + // can contain element keys in braces to include data elements streamed by + // `StreamJSONMarkerData()`. E.g.: "This is {text}" + +# define LABEL_SETTER(name) \ + MarkerSchema& Set##name(std::string a##name) { \ + m##name = std::move(a##name); \ + return *this; \ + } + + LABEL_SETTER(ChartLabel) + LABEL_SETTER(TooltipLabel) + LABEL_SETTER(TableLabel) + +# undef LABEL_SETTER + + MarkerSchema& SetAllLabels(std::string aText) { + // Here we set the same text in each label. + // TODO: Move to a single "label" field once the front-end allows it. + SetChartLabel(aText); + SetTooltipLabel(aText); + SetTableLabel(std::move(aText)); + return *this; + } + + // Each data element that is streamed by `StreamJSONMarkerData()` can be + // displayed as indicated by using one of the `Add...` function below. + // Each `Add...` will add a line in the full marker description. Parameters: + // - `aKey`: Element property name as streamed by `StreamJSONMarkerData()`. + // - `aLabel`: Optional prefix. Defaults to the key name. + // - `aFormat`: How to format the data element value, see `Format` above. + // - `aSearchable`: Optional, indicates if the value is used in searches, + // defaults to false. + + MarkerSchema& AddKeyFormat(std::string aKey, Format aFormat) { + mData.emplace_back(mozilla::VariantType<DynamicData>{}, + DynamicData{std::move(aKey), mozilla::Nothing{}, aFormat, + mozilla::Nothing{}}); + return *this; + } + + MarkerSchema& AddKeyLabelFormat(std::string aKey, std::string aLabel, + Format aFormat) { + mData.emplace_back( + mozilla::VariantType<DynamicData>{}, + DynamicData{std::move(aKey), mozilla::Some(std::move(aLabel)), aFormat, + mozilla::Nothing{}}); + return *this; + } + + MarkerSchema& AddKeyFormatSearchable(std::string aKey, Format aFormat, + Searchable aSearchable) { + mData.emplace_back(mozilla::VariantType<DynamicData>{}, + DynamicData{std::move(aKey), mozilla::Nothing{}, aFormat, + mozilla::Some(aSearchable)}); + return *this; + } + + MarkerSchema& AddKeyLabelFormatSearchable(std::string aKey, + std::string aLabel, Format aFormat, + Searchable aSearchable) { + mData.emplace_back( + mozilla::VariantType<DynamicData>{}, + DynamicData{std::move(aKey), mozilla::Some(std::move(aLabel)), aFormat, + mozilla::Some(aSearchable)}); + return *this; + } + + // The display may also include static rows. + + MarkerSchema& AddStaticLabelValue(std::string aLabel, std::string aValue) { + mData.emplace_back(mozilla::VariantType<StaticData>{}, + StaticData{std::move(aLabel), std::move(aValue)}); + return *this; + } + + // Internal streaming function. + MFBT_API void Stream(JSONWriter& aWriter, const Span<const char>& aName) &&; + + private: + MFBT_API static Span<const char> LocationToStringSpan(Location aLocation); + MFBT_API static Span<const char> FormatToStringSpan(Format aFormat); + + // List of marker display locations. Empty for SpecialFrontendLocation. + std::vector<Location> mLocations; + // Labels for different places. + std::string mChartLabel; + std::string mTooltipLabel; + std::string mTableLabel; + // Main display, made of zero or more rows of key+label+format or label+value. + private: + struct DynamicData { + std::string mKey; + mozilla::Maybe<std::string> mLabel; + Format mFormat; + mozilla::Maybe<Searchable> mSearchable; + }; + struct StaticData { + std::string mLabel; + std::string mValue; + }; + using DataRow = mozilla::Variant<DynamicData, StaticData>; + using DataRowVector = std::vector<DataRow>; + + DataRowVector mData; +}; + +} // namespace mozilla + +#endif // MOZ_GECKO_PROFILER + +#endif // BaseProfilerMarkersPrerequisites_h diff --git a/mozglue/baseprofiler/public/BaseProfilerSharedLibraries.h b/mozglue/baseprofiler/public/BaseProfilerSharedLibraries.h new file mode 100644 index 0000000000..0a104193c3 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilerSharedLibraries.h @@ -0,0 +1,146 @@ +/* -*- Mode: C++; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ +/* vim:set ts=2 sw=2 sts=2 et cindent: */ +/* 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 BASE_PROFILER_SHARED_LIBRARIES_H_ +#define BASE_PROFILER_SHARED_LIBRARIES_H_ + +#include "BaseProfiler.h" + +#ifndef MOZ_GECKO_PROFILER +# error Do not #include this header when MOZ_GECKO_PROFILER is not #defined. +#endif + +#include <algorithm> +#include <stdint.h> +#include <stdlib.h> +#include <string> +#include <vector> + +class SharedLibrary { + public: + SharedLibrary(uintptr_t aStart, uintptr_t aEnd, uintptr_t aOffset, + const std::string& aBreakpadId, const std::string& aModuleName, + const std::string& aModulePath, const std::string& aDebugName, + const std::string& aDebugPath, const std::string& aVersion, + const char* aArch) + : mStart(aStart), + mEnd(aEnd), + mOffset(aOffset), + mBreakpadId(aBreakpadId), + mModuleName(aModuleName), + mModulePath(aModulePath), + mDebugName(aDebugName), + mDebugPath(aDebugPath), + mVersion(aVersion), + mArch(aArch) {} + + SharedLibrary(const SharedLibrary& aEntry) + : mStart(aEntry.mStart), + mEnd(aEntry.mEnd), + mOffset(aEntry.mOffset), + mBreakpadId(aEntry.mBreakpadId), + mModuleName(aEntry.mModuleName), + mModulePath(aEntry.mModulePath), + mDebugName(aEntry.mDebugName), + mDebugPath(aEntry.mDebugPath), + mVersion(aEntry.mVersion), + mArch(aEntry.mArch) {} + + SharedLibrary& operator=(const SharedLibrary& aEntry) { + // Gracefully handle self assignment + if (this == &aEntry) return *this; + + mStart = aEntry.mStart; + mEnd = aEntry.mEnd; + mOffset = aEntry.mOffset; + mBreakpadId = aEntry.mBreakpadId; + mModuleName = aEntry.mModuleName; + mModulePath = aEntry.mModulePath; + mDebugName = aEntry.mDebugName; + mDebugPath = aEntry.mDebugPath; + mVersion = aEntry.mVersion; + mArch = aEntry.mArch; + return *this; + } + + bool operator==(const SharedLibrary& other) const { + return (mStart == other.mStart) && (mEnd == other.mEnd) && + (mOffset == other.mOffset) && (mModuleName == other.mModuleName) && + (mModulePath == other.mModulePath) && + (mDebugName == other.mDebugName) && + (mDebugPath == other.mDebugPath) && + (mBreakpadId == other.mBreakpadId) && (mVersion == other.mVersion) && + (mArch == other.mArch); + } + + uintptr_t GetStart() const { return mStart; } + uintptr_t GetEnd() const { return mEnd; } + uintptr_t GetOffset() const { return mOffset; } + const std::string& GetBreakpadId() const { return mBreakpadId; } + const std::string& GetModuleName() const { return mModuleName; } + const std::string& GetModulePath() const { return mModulePath; } + const std::string& GetDebugName() const { return mDebugName; } + const std::string& GetDebugPath() const { return mDebugPath; } + const std::string& GetVersion() const { return mVersion; } + const std::string& GetArch() const { return mArch; } + + private: + SharedLibrary() : mStart{0}, mEnd{0}, mOffset{0} {} + + uintptr_t mStart; + uintptr_t mEnd; + uintptr_t mOffset; + std::string mBreakpadId; + std::string mModuleName; + std::string mModulePath; + std::string mDebugName; + std::string mDebugPath; + std::string mVersion; + std::string mArch; +}; + +static bool CompareAddresses(const SharedLibrary& first, + const SharedLibrary& second) { + return first.GetStart() < second.GetStart(); +} + +class SharedLibraryInfo { + public: + static SharedLibraryInfo GetInfoForSelf(); + static void Initialize(); + + SharedLibraryInfo() {} + + void AddSharedLibrary(SharedLibrary entry) { mEntries.push_back(entry); } + + const SharedLibrary& GetEntry(size_t i) const { return mEntries[i]; } + + SharedLibrary& GetMutableEntry(size_t i) { return mEntries[i]; } + + // Removes items in the range [first, last) + // i.e. element at the "last" index is not removed + void RemoveEntries(size_t first, size_t last) { + mEntries.erase(mEntries.begin() + first, mEntries.begin() + last); + } + + bool Contains(const SharedLibrary& searchItem) const { + return (mEntries.end() != + std::find(mEntries.begin(), mEntries.end(), searchItem)); + } + + size_t GetSize() const { return mEntries.size(); } + + void SortByAddress() { + std::sort(mEntries.begin(), mEntries.end(), CompareAddresses); + } + + void Clear() { mEntries.clear(); } + + private: + std::vector<SharedLibrary> mEntries; +}; + +#endif // BASE_PROFILER_SHARED_LIBRARIES_H_ diff --git a/mozglue/baseprofiler/public/BaseProfilingCategory.h b/mozglue/baseprofiler/public/BaseProfilingCategory.h new file mode 100644 index 0000000000..6892ec40f4 --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilingCategory.h @@ -0,0 +1,72 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 4 -*- + * vim: set ts=8 sts=4 et sw=4 tw=99: + * 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 BaseProfilingCategory_h +#define BaseProfilingCategory_h + +#ifndef MOZ_GECKO_PROFILER +# error Do not #include this header when MOZ_GECKO_PROFILER is not #defined. +#endif + +#include "mozilla/Types.h" + +#include <cstdint> + +#include "ProfilingCategoryList.h" + +namespace mozilla { +namespace baseprofiler { + +// clang-format off + +// An enum that lists all possible category pairs in one list. +// This is the enum that is used in profiler stack labels. Having one list that +// includes subcategories from all categories in one list allows assigning the +// category pair to a stack label with just one number. +#define CATEGORY_ENUM_BEGIN_CATEGORY(name, labelAsString, color) +#define CATEGORY_ENUM_SUBCATEGORY(supercategory, name, labelAsString) name, +#define CATEGORY_ENUM_END_CATEGORY +enum class ProfilingCategoryPair : uint32_t { + MOZ_PROFILING_CATEGORY_LIST(CATEGORY_ENUM_BEGIN_CATEGORY, + CATEGORY_ENUM_SUBCATEGORY, + CATEGORY_ENUM_END_CATEGORY) + COUNT, + LAST = COUNT - 1, +}; +#undef CATEGORY_ENUM_BEGIN_CATEGORY +#undef CATEGORY_ENUM_SUBCATEGORY +#undef CATEGORY_ENUM_END_CATEGORY + +// An enum that lists just the categories without their subcategories. +#define SUPERCATEGORY_ENUM_BEGIN_CATEGORY(name, labelAsString, color) name, +#define SUPERCATEGORY_ENUM_SUBCATEGORY(supercategory, name, labelAsString) +#define SUPERCATEGORY_ENUM_END_CATEGORY +enum class ProfilingCategory : uint32_t { + MOZ_PROFILING_CATEGORY_LIST(SUPERCATEGORY_ENUM_BEGIN_CATEGORY, + SUPERCATEGORY_ENUM_SUBCATEGORY, + SUPERCATEGORY_ENUM_END_CATEGORY) + COUNT, + LAST = COUNT - 1, +}; +#undef SUPERCATEGORY_ENUM_BEGIN_CATEGORY +#undef SUPERCATEGORY_ENUM_SUBCATEGORY +#undef SUPERCATEGORY_ENUM_END_CATEGORY + +// clang-format on + +struct ProfilingCategoryPairInfo { + ProfilingCategory mCategory; + uint32_t mSubcategoryIndex; + const char* mLabel; +}; + +MFBT_API const ProfilingCategoryPairInfo& GetProfilingCategoryPairInfo( + ProfilingCategoryPair aCategoryPair); + +} // namespace baseprofiler +} // namespace mozilla + +#endif /* BaseProfilingCategory_h */ diff --git a/mozglue/baseprofiler/public/BaseProfilingStack.h b/mozglue/baseprofiler/public/BaseProfilingStack.h new file mode 100644 index 0000000000..214fc1ebbf --- /dev/null +++ b/mozglue/baseprofiler/public/BaseProfilingStack.h @@ -0,0 +1,520 @@ +/* -*- 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/. */ + +#ifndef BaseProfilingStack_h +#define BaseProfilingStack_h + +#include "BaseProfilingCategory.h" + +#include "mozilla/Atomics.h" + +#include "BaseProfiler.h" + +#ifndef MOZ_GECKO_PROFILER +# error Do not #include this header when MOZ_GECKO_PROFILER is not #defined. +#endif + +#include <algorithm> +#include <stdint.h> + +// This file defines the classes ProfilingStack and ProfilingStackFrame. +// The ProfilingStack manages an array of ProfilingStackFrames. +// It keeps track of the "label stack" and the JS interpreter stack. +// The two stack types are interleaved. +// +// Usage: +// +// ProfilingStack* profilingStack = ...; +// +// // For label frames: +// profilingStack->pushLabelFrame(...); +// // Execute some code. When finished, pop the frame: +// profilingStack->pop(); +// +// // For JS stack frames: +// profilingStack->pushJSFrame(...); +// // Execute some code. When finished, pop the frame: +// profilingStack->pop(); +// +// +// Concurrency considerations +// +// A thread's profiling stack (and the frames inside it) is only modified by +// that thread. However, the profiling stack can be *read* by a different +// thread, the sampler thread: Whenever the profiler wants to sample a given +// thread A, the following happens: +// (1) Thread A is suspended. +// (2) The sampler thread (thread S) reads the ProfilingStack of thread A, +// including all ProfilingStackFrames that are currently in that stack +// (profilingStack->frames[0..profilingStack->stackSize()]). +// (3) Thread A is resumed. +// +// Thread suspension is achieved using platform-specific APIs; refer to each +// platform's Sampler::SuspendAndSampleAndResumeThread implementation in +// platform-*.cpp for details. +// +// When the thread is suspended, the values in profilingStack->stackPointer and +// in the stack frame range +// profilingStack->frames[0..profilingStack->stackPointer] need to be in a +// consistent state, so that thread S does not read partially- constructed stack +// frames. More specifically, we have two requirements: +// (1) When adding a new frame at the top of the stack, its ProfilingStackFrame +// data needs to be put in place *before* the stackPointer is incremented, +// and the compiler + CPU need to know that this order matters. +// (2) When popping an frame from the stack and then preparing the +// ProfilingStackFrame data for the next frame that is about to be pushed, +// the decrement of the stackPointer in pop() needs to happen *before* the +// ProfilingStackFrame for the new frame is being popuplated, and the +// compiler + CPU need to know that this order matters. +// +// We can express the relevance of these orderings in multiple ways. +// Option A is to make stackPointer an atomic with SequentiallyConsistent +// memory ordering. This would ensure that no writes in thread A would be +// reordered across any writes to stackPointer, which satisfies requirements +// (1) and (2) at the same time. Option A is the simplest. +// Option B is to use ReleaseAcquire memory ordering both for writes to +// stackPointer *and* for writes to ProfilingStackFrame fields. Release-stores +// ensure that all writes that happened *before this write in program order* are +// not reordered to happen after this write. ReleaseAcquire ordering places no +// requirements on the ordering of writes that happen *after* this write in +// program order. +// Using release-stores for writes to stackPointer expresses requirement (1), +// and using release-stores for writes to the ProfilingStackFrame fields +// expresses requirement (2). +// +// Option B is more complicated than option A, but has much better performance +// on x86/64: In a microbenchmark run on a Macbook Pro from 2017, switching +// from option A to option B reduced the overhead of pushing+popping a +// ProfilingStackFrame by 10 nanoseconds. +// On x86/64, release-stores require no explicit hardware barriers or lock +// instructions. +// On ARM/64, option B may be slower than option A, because the compiler will +// generate hardware barriers for every single release-store instead of just +// for the writes to stackPointer. However, the actual performance impact of +// this has not yet been measured on ARM, so we're currently using option B +// everywhere. This is something that we may want to change in the future once +// we've done measurements. + +namespace mozilla { +namespace baseprofiler { + +// A call stack can be specified to the JS engine such that all JS entry/exits +// to functions push/pop a stack frame to/from the specified stack. +// +// For more detailed information, see vm/GeckoProfiler.h. +// +class ProfilingStackFrame { + // A ProfilingStackFrame represents either a label frame or a JS frame. + + // WARNING WARNING WARNING + // + // All the fields below are Atomic<...,ReleaseAcquire>. This is needed so + // that writes to these fields are release-writes, which ensures that + // earlier writes in this thread don't get reordered after the writes to + // these fields. In particular, the decrement of the stack pointer in + // ProfilingStack::pop() is a write that *must* happen before the values in + // this ProfilingStackFrame are changed. Otherwise, the sampler thread might + // see an inconsistent state where the stack pointer still points to a + // ProfilingStackFrame which has already been popped off the stack and whose + // fields have now been partially repopulated with new values. + // See the "Concurrency considerations" paragraph at the top of this file + // for more details. + + // Descriptive label for this stack frame. Must be a static string! Can be + // an empty string, but not a null pointer. + Atomic<const char*, ReleaseAcquire> label_; + + // An additional descriptive string of this frame which is combined with + // |label_| in profiler output. Need not be (and usually isn't) static. Can + // be null. + Atomic<const char*, ReleaseAcquire> dynamicString_; + + // Stack pointer for non-JS stack frames, the script pointer otherwise. + Atomic<void*, ReleaseAcquire> spOrScript; + + // ID of the JS Realm for JS stack frames. + // Must not be used on non-JS frames; it'll contain either the default 0, + // or a leftover value from a previous JS stack frame that was using this + // ProfilingStackFrame object. + mozilla::Atomic<uint64_t, mozilla::ReleaseAcquire> realmID_; + + // The bytecode offset for JS stack frames. + // Must not be used on non-JS frames; it'll contain either the default 0, + // or a leftover value from a previous JS stack frame that was using this + // ProfilingStackFrame object. + Atomic<int32_t, ReleaseAcquire> pcOffsetIfJS_; + + // Bits 0...8 hold the Flags. Bits 9...31 hold the category pair. + Atomic<uint32_t, ReleaseAcquire> flagsAndCategoryPair_; + + public: + ProfilingStackFrame() = default; + ProfilingStackFrame& operator=(const ProfilingStackFrame& other) { + label_ = other.label(); + dynamicString_ = other.dynamicString(); + void* spScript = other.spOrScript; + spOrScript = spScript; + int32_t offsetIfJS = other.pcOffsetIfJS_; + pcOffsetIfJS_ = offsetIfJS; + int64_t realmID = other.realmID_; + realmID_ = realmID; + uint32_t flagsAndCategory = other.flagsAndCategoryPair_; + flagsAndCategoryPair_ = flagsAndCategory; + return *this; + } + + // Reserve up to 16 bits for flags, and 16 for category pair. + enum class Flags : uint32_t { + // The first three flags describe the kind of the frame and are + // mutually exclusive. (We still give them individual bits for + // simplicity.) + + // A regular label frame. These usually come from AutoProfilerLabel. + IS_LABEL_FRAME = 1 << 0, + + // A special frame indicating the start of a run of JS profiling stack + // frames. IS_SP_MARKER_FRAME frames are ignored, except for the sp + // field. These frames are needed to get correct ordering between JS + // and LABEL frames because JS frames don't carry sp information. + // SP is short for "stack pointer". + IS_SP_MARKER_FRAME = 1 << 1, + + // A JS frame. + IS_JS_FRAME = 1 << 2, + + // An interpreter JS frame that has OSR-ed into baseline. IS_JS_FRAME + // frames can have this flag set and unset during their lifetime. + // JS_OSR frames are ignored. + JS_OSR = 1 << 3, + + // The next three are mutually exclusive. + // By default, for profiling stack frames that have both a label and a + // dynamic string, the two strings are combined into one string of the + // form "<label> <dynamicString>" during JSON serialization. The + // following flags can be used to change this preset. + STRING_TEMPLATE_METHOD = 1 << 4, // "<label>.<dynamicString>" + STRING_TEMPLATE_GETTER = 1 << 5, // "get <label>.<dynamicString>" + STRING_TEMPLATE_SETTER = 1 << 6, // "set <label>.<dynamicString>" + + // If set, causes this stack frame to be marked as "relevantForJS" in + // the profile JSON, which will make it show up in the "JS only" call + // tree view. + RELEVANT_FOR_JS = 1 << 7, + + // If set, causes the label on this ProfilingStackFrame to be ignored + // and to be replaced by the subcategory's label. + LABEL_DETERMINED_BY_CATEGORY_PAIR = 1 << 8, + + // Frame dynamic string does not contain user data. + NONSENSITIVE = 1 << 9, + + // A JS Baseline Interpreter frame. + IS_BLINTERP_FRAME = 1 << 10, + + FLAGS_BITCOUNT = 16, + FLAGS_MASK = (1 << FLAGS_BITCOUNT) - 1 + }; + + static_assert( + uint32_t(ProfilingCategoryPair::LAST) <= + (UINT32_MAX >> uint32_t(Flags::FLAGS_BITCOUNT)), + "Too many category pairs to fit into u32 with together with the " + "reserved bits for the flags"); + + bool isLabelFrame() const { + return uint32_t(flagsAndCategoryPair_) & uint32_t(Flags::IS_LABEL_FRAME); + } + + bool isSpMarkerFrame() const { + return uint32_t(flagsAndCategoryPair_) & + uint32_t(Flags::IS_SP_MARKER_FRAME); + } + + bool isJsFrame() const { + return uint32_t(flagsAndCategoryPair_) & uint32_t(Flags::IS_JS_FRAME); + } + + bool isOSRFrame() const { + return uint32_t(flagsAndCategoryPair_) & uint32_t(Flags::JS_OSR); + } + + void setIsOSRFrame(bool isOSR) { + if (isOSR) { + flagsAndCategoryPair_ = + uint32_t(flagsAndCategoryPair_) | uint32_t(Flags::JS_OSR); + } else { + flagsAndCategoryPair_ = + uint32_t(flagsAndCategoryPair_) & ~uint32_t(Flags::JS_OSR); + } + } + + const char* label() const { + uint32_t flagsAndCategoryPair = flagsAndCategoryPair_; + if (flagsAndCategoryPair & + uint32_t(Flags::LABEL_DETERMINED_BY_CATEGORY_PAIR)) { + auto categoryPair = ProfilingCategoryPair( + flagsAndCategoryPair >> uint32_t(Flags::FLAGS_BITCOUNT)); + return GetProfilingCategoryPairInfo(categoryPair).mLabel; + } + return label_; + } + + const char* dynamicString() const { return dynamicString_; } + + void initLabelFrame(const char* aLabel, const char* aDynamicString, void* sp, + ProfilingCategoryPair aCategoryPair, uint32_t aFlags) { + label_ = aLabel; + dynamicString_ = aDynamicString; + spOrScript = sp; + // pcOffsetIfJS_ is not set and must not be used on label frames. + flagsAndCategoryPair_ = + uint32_t(Flags::IS_LABEL_FRAME) | + (uint32_t(aCategoryPair) << uint32_t(Flags::FLAGS_BITCOUNT)) | aFlags; + MOZ_ASSERT(isLabelFrame()); + } + + void initSpMarkerFrame(void* sp) { + label_ = ""; + dynamicString_ = nullptr; + spOrScript = sp; + // pcOffsetIfJS_ is not set and must not be used on sp marker frames. + flagsAndCategoryPair_ = uint32_t(Flags::IS_SP_MARKER_FRAME) | + (uint32_t(ProfilingCategoryPair::OTHER) + << uint32_t(Flags::FLAGS_BITCOUNT)); + MOZ_ASSERT(isSpMarkerFrame()); + } + + void initJsFrame(const char* aLabel, const char* aDynamicString, + void* /* JSScript* */ aScript, int32_t aOffset, + uint64_t aRealmID) { + label_ = aLabel; + dynamicString_ = aDynamicString; + spOrScript = aScript; + pcOffsetIfJS_ = aOffset; + realmID_ = aRealmID; + flagsAndCategoryPair_ = + uint32_t(Flags::IS_JS_FRAME) | (uint32_t(ProfilingCategoryPair::JS) + << uint32_t(Flags::FLAGS_BITCOUNT)); + MOZ_ASSERT(isJsFrame()); + } + + uint32_t flags() const { + return uint32_t(flagsAndCategoryPair_) & uint32_t(Flags::FLAGS_MASK); + } + + ProfilingCategoryPair categoryPair() const { + return ProfilingCategoryPair(flagsAndCategoryPair_ >> + uint32_t(Flags::FLAGS_BITCOUNT)); + } + + uint64_t realmID() const { return realmID_; } + + void* stackAddress() const { + MOZ_ASSERT(!isJsFrame()); + return spOrScript; + } + + // Note that the pointer returned might be invalid. + void* rawScript() const { + MOZ_ASSERT(isJsFrame()); + return spOrScript; + } + void setRawScript(void* aScript) { + MOZ_ASSERT(isJsFrame()); + spOrScript = aScript; + } + + int32_t pcOffset() const { + MOZ_ASSERT(isJsFrame()); + return pcOffsetIfJS_; + } + + void setPCOffset(int32_t aOffset) { + MOZ_ASSERT(isJsFrame()); + pcOffsetIfJS_ = aOffset; + } + + // The offset of a pc into a script's code can actually be 0, so to + // signify a nullptr pc, use a -1 index. This is checked against in + // pc() and setPC() to set/get the right pc. + static const int32_t NullPCOffset = -1; +}; + +// Each thread has its own ProfilingStack. That thread modifies the +// ProfilingStack, pushing and popping elements as necessary. +// +// The ProfilingStack is also read periodically by the profiler's sampler +// thread. This happens only when the thread that owns the ProfilingStack is +// suspended. So there are no genuine parallel accesses. +// +// However, it is possible for pushing/popping to be interrupted by a periodic +// sample. Because of this, we need pushing/popping to be effectively atomic. +// +// - When pushing a new frame, we increment the stack pointer -- making the new +// frame visible to the sampler thread -- only after the new frame has been +// fully written. The stack pointer is Atomic<uint32_t,ReleaseAcquire>, so +// the increment is a release-store, which ensures that this store is not +// reordered before the writes of the frame. +// +// - When popping an old frame, the only operation is the decrementing of the +// stack pointer, which is obviously atomic. +// +class ProfilingStack final { + public: + ProfilingStack() = default; + + MFBT_API ~ProfilingStack(); + + void pushLabelFrame(const char* label, const char* dynamicString, void* sp, + ProfilingCategoryPair categoryPair, uint32_t flags = 0) { + // This thread is the only one that ever changes the value of + // stackPointer. + // Store the value of the atomic in a non-atomic local variable so that + // the compiler won't generate two separate loads from the atomic for + // the size check and the frames[] array indexing operation. + uint32_t stackPointerVal = stackPointer; + + if (MOZ_UNLIKELY(stackPointerVal >= capacity)) { + ensureCapacitySlow(); + } + frames[stackPointerVal].initLabelFrame(label, dynamicString, sp, + categoryPair, flags); + + // This must happen at the end! The compiler will not reorder this + // update because stackPointer is Atomic<..., ReleaseAcquire>, so any + // the writes above will not be reordered below the stackPointer store. + // Do the read and the write as two separate statements, in order to + // make it clear that we don't need an atomic increment, which would be + // more expensive on x86 than the separate operations done here. + // However, don't use stackPointerVal here; instead, allow the compiler + // to turn this store into a non-atomic increment instruction which + // takes up less code size. + stackPointer = stackPointer + 1; + } + + void pushSpMarkerFrame(void* sp) { + uint32_t oldStackPointer = stackPointer; + + if (MOZ_UNLIKELY(oldStackPointer >= capacity)) { + ensureCapacitySlow(); + } + frames[oldStackPointer].initSpMarkerFrame(sp); + + // This must happen at the end, see the comment in pushLabelFrame. + stackPointer = oldStackPointer + 1; + } + + void pushJsOffsetFrame(const char* label, const char* dynamicString, + void* script, int32_t offset, uint64_t aRealmID) { + // This thread is the only one that ever changes the value of + // stackPointer. Only load the atomic once. + uint32_t oldStackPointer = stackPointer; + + if (MOZ_UNLIKELY(oldStackPointer >= capacity)) { + ensureCapacitySlow(); + } + frames[oldStackPointer].initJsFrame(label, dynamicString, script, offset, + aRealmID); + + // This must happen at the end, see the comment in pushLabelFrame. + stackPointer = stackPointer + 1; + } + + void pop() { + MOZ_ASSERT(stackPointer > 0); + // Do the read and the write as two separate statements, in order to + // make it clear that we don't need an atomic decrement, which would be + // more expensive on x86 than the separate operations done here. + // This thread is the only one that ever changes the value of + // stackPointer. + uint32_t oldStackPointer = stackPointer; + stackPointer = oldStackPointer - 1; + } + + uint32_t stackSize() const { return stackPointer; } + uint32_t stackCapacity() const { return capacity; } + + private: + // Out of line path for expanding the buffer, since otherwise this would get + // inlined in every DOM WebIDL call. + MFBT_API MOZ_COLD void ensureCapacitySlow(); + + // No copying. + ProfilingStack(const ProfilingStack&) = delete; + void operator=(const ProfilingStack&) = delete; + + // No moving either. + ProfilingStack(ProfilingStack&&) = delete; + void operator=(ProfilingStack&&) = delete; + + uint32_t capacity = 0; + + public: + // The pointer to the stack frames, this is read from the profiler thread and + // written from the current thread. + // + // This is effectively a unique pointer. + Atomic<ProfilingStackFrame*, SequentiallyConsistent> frames{nullptr}; + + // This may exceed the capacity, so instead use the stackSize() method to + // determine the number of valid frames in stackFrames. When this is less + // than stackCapacity(), it refers to the first free stackframe past the top + // of the in-use stack (i.e. frames[stackPointer - 1] is the top stack + // frame). + // + // WARNING WARNING WARNING + // + // This is an atomic variable that uses ReleaseAcquire memory ordering. + // See the "Concurrency considerations" paragraph at the top of this file + // for more details. + Atomic<uint32_t, ReleaseAcquire> stackPointer{0}; +}; + +class AutoGeckoProfilerEntry; +class GeckoProfilerEntryMarker; +class GeckoProfilerBaselineOSRMarker; + +class GeckoProfilerThread { + friend class AutoGeckoProfilerEntry; + friend class GeckoProfilerEntryMarker; + friend class GeckoProfilerBaselineOSRMarker; + + ProfilingStack* profilingStack_; + + // Same as profilingStack_ if the profiler is currently active, otherwise + // null. + ProfilingStack* profilingStackIfEnabled_; + + public: + MFBT_API GeckoProfilerThread(); + + uint32_t stackPointer() { + MOZ_ASSERT(infraInstalled()); + return profilingStack_->stackPointer; + } + ProfilingStackFrame* stack() { return profilingStack_->frames; } + ProfilingStack* getProfilingStack() { return profilingStack_; } + ProfilingStack* getProfilingStackIfEnabled() { + return profilingStackIfEnabled_; + } + + /* + * True if the profiler infrastructure is setup. Should be true in builds + * that include profiler support except during early startup or late + * shutdown. Unrelated to the presence of the Gecko Profiler addon. + */ + bool infraInstalled() { return profilingStack_ != nullptr; } + + MFBT_API void setProfilingStack(ProfilingStack* profilingStack, bool enabled); + void enable(bool enable) { + profilingStackIfEnabled_ = enable ? profilingStack_ : nullptr; + } +}; + +} // namespace baseprofiler +} // namespace mozilla + +#endif /* BaseProfilingStack_h */ diff --git a/mozglue/baseprofiler/public/BlocksRingBuffer.h b/mozglue/baseprofiler/public/BlocksRingBuffer.h new file mode 100644 index 0000000000..6948ab8cf4 --- /dev/null +++ b/mozglue/baseprofiler/public/BlocksRingBuffer.h @@ -0,0 +1,1000 @@ +/* -*- 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 BlocksRingBuffer_h +#define BlocksRingBuffer_h + +#include "mozilla/BaseProfilerDetail.h" +#include "mozilla/ModuloBuffer.h" +#include "mozilla/ProfileBufferIndex.h" +#include "mozilla/ScopeExit.h" + +#include <functional> +#include <string> +#include <tuple> +#include <utility> + +namespace mozilla { + +// Thread-safe Ring buffer that can store blocks of different sizes during +// defined sessions. +// Each *block* contains an *entry* and the entry size: +// [ entry_size | entry ] [ entry_size | entry ] ... +// *In-session* is a period of time during which `BlocksRingBuffer` allows +// reading and writing. *Out-of-session*, the `BlocksRingBuffer` object is +// still valid, but contains no data, and gracefully denies accesses. +// +// To write an entry, the buffer reserves a block of sufficient size (to contain +// user data of predetermined size), writes the entry size, and lets the caller +// fill the entry contents using ModuloBuffer::Iterator APIs and a few entry- +// specific APIs. E.g.: +// ``` +// BlockRingsBuffer brb(PowerOfTwo<BlockRingsBuffer::Length>(1024)); +// brb.ReserveAndPut([]() { return sizeof(123); }, +// [&](ProfileBufferEntryWriter& aEW) { +// aEW.WriteObject(123); +// }); +// ``` +// Other `Put...` functions may be used as shortcuts for simple entries. +// The objects given to the caller's callbacks should only be used inside the +// callbacks and not stored elsewhere, because they keep their own references to +// the BlocksRingBuffer and therefore should not live longer. +// Different type of objects may be serialized into an entry, see `Serializer` +// for more information. +// +// When reading data, the buffer iterates over blocks (it knows how to read the +// entry size, and therefore move to the next block), and lets the caller read +// the entry inside of each block. E.g.: +// ``` +// brb.Read([](BlocksRingBuffer::Reader aR) {} +// for (ProfileBufferEntryReader aER : aR) { +// /* Use ProfileBufferEntryReader functions to read serialized objects. */ +// int n = aER.ReadObject<int>(); +// } +// }); +// ``` +// Different type of objects may be deserialized from an entry, see +// `Deserializer` for more information. +// +// The caller may retrieve the `ProfileBufferBlockIndex` corresponding to an +// entry (`ProfileBufferBlockIndex` is an opaque type preventing the user from +// modifying it). That index may later be used to get back to that particular +// entry if it still exists. +class BlocksRingBuffer { + public: + // Using ModuloBuffer as underlying circular byte buffer. + using Buffer = ModuloBuffer<uint32_t, ProfileBufferIndex>; + using Byte = Buffer::Byte; + + // Length type for total buffer (as PowerOfTwo<Length>) and each entry. + using Length = uint32_t; + + enum class ThreadSafety { WithoutMutex, WithMutex }; + + // Default constructor starts out-of-session (nothing to read or write). + explicit BlocksRingBuffer(ThreadSafety aThreadSafety) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex) {} + + // Create a buffer of the given length. + explicit BlocksRingBuffer(ThreadSafety aThreadSafety, + PowerOfTwo<Length> aLength) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex), + mMaybeUnderlyingBuffer(Some(UnderlyingBuffer(aLength))) {} + + // Take ownership of an existing buffer. + BlocksRingBuffer(ThreadSafety aThreadSafety, + UniquePtr<Buffer::Byte[]> aExistingBuffer, + PowerOfTwo<Length> aLength) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex), + mMaybeUnderlyingBuffer( + Some(UnderlyingBuffer(std::move(aExistingBuffer), aLength))) {} + + // Use an externally-owned buffer. + BlocksRingBuffer(ThreadSafety aThreadSafety, Buffer::Byte* aExternalBuffer, + PowerOfTwo<Length> aLength) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex), + mMaybeUnderlyingBuffer( + Some(UnderlyingBuffer(aExternalBuffer, aLength))) {} + + // Destructor doesn't need to do anything special. (Clearing entries would + // only update indices and stats, which won't be accessible after the object + // is destroyed anyway.) + ~BlocksRingBuffer() = default; + + // Remove underlying buffer, if any. + void Reset() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + ResetUnderlyingBuffer(); + } + + // Create a buffer of the given length. + void Set(PowerOfTwo<Length> aLength) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + ResetUnderlyingBuffer(); + mMaybeUnderlyingBuffer.emplace(aLength); + } + + // Take ownership of an existing buffer. + void Set(UniquePtr<Buffer::Byte[]> aExistingBuffer, + PowerOfTwo<Length> aLength) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + ResetUnderlyingBuffer(); + mMaybeUnderlyingBuffer.emplace(std::move(aExistingBuffer), aLength); + } + + // Use an externally-owned buffer. + void Set(Buffer::Byte* aExternalBuffer, PowerOfTwo<Length> aLength) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + ResetUnderlyingBuffer(); + mMaybeUnderlyingBuffer.emplace(aExternalBuffer, aLength); + } + + // This cannot change during the lifetime of this buffer, so there's no need + // to lock. + bool IsThreadSafe() const { return mMutex.IsActivated(); } + + [[nodiscard]] bool IsInSession() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return !!mMaybeUnderlyingBuffer; + } + + // Lock the buffer mutex and run the provided callback. + // This can be useful when the caller needs to explicitly lock down this + // buffer, but not do anything else with it. + template <typename Callback> + auto LockAndRun(Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return std::forward<Callback>(aCallback)(); + } + + // Buffer length in bytes. + Maybe<PowerOfTwo<Length>> BufferLength() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return mMaybeUnderlyingBuffer.map([](const UnderlyingBuffer& aBuffer) { + return aBuffer.mBuffer.BufferLength(); + }); + ; + } + + // Size of external resources. + size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { + if (!mMaybeUnderlyingBuffer) { + return 0; + } + return mMaybeUnderlyingBuffer->mBuffer.SizeOfExcludingThis(aMallocSizeOf); + } + + size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); + } + + // Snapshot of the buffer state. + struct State { + // Index to the first block. + ProfileBufferBlockIndex mRangeStart; + + // Index past the last block. Equals mRangeStart if empty. + ProfileBufferBlockIndex mRangeEnd; + + // Number of blocks that have been pushed into this buffer. + uint64_t mPushedBlockCount = 0; + + // Number of blocks that have been removed from this buffer. + // Note: Live entries = pushed - cleared. + uint64_t mClearedBlockCount = 0; + }; + + // Get a snapshot of the current state. + // When out-of-session, mFirstReadIndex==mNextWriteIndex, and + // mPushedBlockCount==mClearedBlockCount==0. + // Note that these may change right after this thread-safe call, so they + // should only be used for statistical purposes. + State GetState() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return { + mFirstReadIndex, mNextWriteIndex, + mMaybeUnderlyingBuffer ? mMaybeUnderlyingBuffer->mPushedBlockCount : 0, + mMaybeUnderlyingBuffer ? mMaybeUnderlyingBuffer->mClearedBlockCount + : 0}; + } + + class Reader; + + // Class that can iterate through blocks and provide + // `ProfileBufferEntryReader`s. + // Created through `Reader`, lives within a lock guard lifetime. + class BlockIterator { + public: +#ifdef DEBUG + ~BlockIterator() { + // No BlockIterator should live outside of a mutexed call. + mRing->mMutex.AssertCurrentThreadOwns(); + } +#endif // DEBUG + + // Comparison with other iterator, mostly used in range-for loops. + bool operator==(const BlockIterator aRhs) const { + MOZ_ASSERT(mRing == aRhs.mRing); + return mBlockIndex == aRhs.mBlockIndex; + } + bool operator!=(const BlockIterator aRhs) const { + MOZ_ASSERT(mRing == aRhs.mRing); + return mBlockIndex != aRhs.mBlockIndex; + } + + // Advance to next BlockIterator. + BlockIterator& operator++() { + mBlockIndex = NextBlockIndex(); + return *this; + } + + // Dereferencing creates a `ProfileBufferEntryReader` for the entry inside + // this block. + ProfileBufferEntryReader operator*() const { + return mRing->ReaderInBlockAt(mBlockIndex); + } + + // True if this iterator is just past the last entry. + bool IsAtEnd() const { + MOZ_ASSERT(mBlockIndex <= BufferRangeEnd()); + return mBlockIndex == BufferRangeEnd(); + } + + // Can be used as reference to come back to this entry with `ReadAt()`. + ProfileBufferBlockIndex CurrentBlockIndex() const { return mBlockIndex; } + + // Index past the end of this block, which is the start of the next block. + ProfileBufferBlockIndex NextBlockIndex() const { + MOZ_ASSERT(!IsAtEnd()); + const Length entrySize = + mRing->ReaderInBlockAt(mBlockIndex).RemainingBytes(); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mBlockIndex.ConvertToProfileBufferIndex() + ULEB128Size(entrySize) + + entrySize); + } + + // Index of the first block in the whole buffer. + ProfileBufferBlockIndex BufferRangeStart() const { + return mRing->mFirstReadIndex; + } + + // Index past the last block in the whole buffer. + ProfileBufferBlockIndex BufferRangeEnd() const { + return mRing->mNextWriteIndex; + } + + private: + // Only a Reader can instantiate a BlockIterator. + friend class Reader; + + BlockIterator(const BlocksRingBuffer& aRing, + ProfileBufferBlockIndex aBlockIndex) + : mRing(WrapNotNull(&aRing)), mBlockIndex(aBlockIndex) { + // No BlockIterator should live outside of a mutexed call. + mRing->mMutex.AssertCurrentThreadOwns(); + } + + // Using a non-null pointer instead of a reference, to allow copying. + // This BlockIterator should only live inside one of the thread-safe + // BlocksRingBuffer functions, for this reference to stay valid. + NotNull<const BlocksRingBuffer*> mRing; + ProfileBufferBlockIndex mBlockIndex; + }; + + // Class that can create `BlockIterator`s (e.g., for range-for), or just + // iterate through entries; lives within a lock guard lifetime. + class MOZ_RAII Reader { + public: + Reader(const Reader&) = delete; + Reader& operator=(const Reader&) = delete; + Reader(Reader&&) = delete; + Reader& operator=(Reader&&) = delete; + +#ifdef DEBUG + ~Reader() { + // No Reader should live outside of a mutexed call. + mRing.mMutex.AssertCurrentThreadOwns(); + } +#endif // DEBUG + + // Index of the first block in the whole buffer. + ProfileBufferBlockIndex BufferRangeStart() const { + return mRing.mFirstReadIndex; + } + + // Index past the last block in the whole buffer. + ProfileBufferBlockIndex BufferRangeEnd() const { + return mRing.mNextWriteIndex; + } + + // Iterators to the first and past-the-last blocks. + // Compatible with range-for (see `ForEach` below as example). + BlockIterator begin() const { + return BlockIterator(mRing, BufferRangeStart()); + } + // Note that a `BlockIterator` at the `end()` should not be dereferenced, as + // there is no actual block there! + BlockIterator end() const { return BlockIterator(mRing, BufferRangeEnd()); } + + // Get a `BlockIterator` at the given `ProfileBufferBlockIndex`, clamped to + // the stored range. Note that a `BlockIterator` at the `end()` should not + // be dereferenced, as there is no actual block there! + BlockIterator At(ProfileBufferBlockIndex aBlockIndex) const { + if (aBlockIndex < BufferRangeStart()) { + // Anything before the range (including null ProfileBufferBlockIndex) is + // clamped at the beginning. + return begin(); + } + // Otherwise we at least expect the index to be valid (pointing exactly at + // a live block, or just past the end.) + mRing.AssertBlockIndexIsValidOrEnd(aBlockIndex); + return BlockIterator(mRing, aBlockIndex); + } + + // Run `aCallback(ProfileBufferEntryReader&)` on each entry from first to + // last. Callback should not store `ProfileBufferEntryReader`, as it may + // become invalid after this thread-safe call. + template <typename Callback> + void ForEach(Callback&& aCallback) const { + for (ProfileBufferEntryReader reader : *this) { + aCallback(reader); + } + } + + private: + friend class BlocksRingBuffer; + + explicit Reader(const BlocksRingBuffer& aRing) : mRing(aRing) { + // No Reader should live outside of a mutexed call. + mRing.mMutex.AssertCurrentThreadOwns(); + } + + // This Reader should only live inside one of the thread-safe + // BlocksRingBuffer functions, for this reference to stay valid. + const BlocksRingBuffer& mRing; + }; + + // Call `aCallback(BlocksRingBuffer::Reader*)` (nullptr when out-of-session), + // and return whatever `aCallback` returns. Callback should not store + // `Reader`, because it may become invalid after this call. + template <typename Callback> + auto Read(Callback&& aCallback) const { + { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_LIKELY(mMaybeUnderlyingBuffer)) { + Reader reader(*this); + return std::forward<Callback>(aCallback)(&reader); + } + } + return std::forward<Callback>(aCallback)(nullptr); + } + + // Call `aCallback(ProfileBufferEntryReader&)` on each item. + // Callback should not store `ProfileBufferEntryReader`, because it may become + // invalid after this call. + template <typename Callback> + void ReadEach(Callback&& aCallback) const { + Read([&](Reader* aReader) { + if (MOZ_LIKELY(aReader)) { + aReader->ForEach(aCallback); + } + }); + } + + // Call `aCallback(Maybe<ProfileBufferEntryReader>&&)` on the entry at + // the given ProfileBufferBlockIndex; The `Maybe` will be `Nothing` if + // out-of-session, or if that entry doesn't exist anymore, or if we've reached + // just past the last entry. Return whatever `aCallback` returns. Callback + // should not store `ProfileBufferEntryReader`, because it may become invalid + // after this call. + template <typename Callback> + auto ReadAt(ProfileBufferBlockIndex aBlockIndex, Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + MOZ_ASSERT(aBlockIndex <= mNextWriteIndex); + Maybe<ProfileBufferEntryReader> maybeEntryReader; + if (MOZ_LIKELY(mMaybeUnderlyingBuffer) && aBlockIndex >= mFirstReadIndex && + aBlockIndex < mNextWriteIndex) { + AssertBlockIndexIsValid(aBlockIndex); + maybeEntryReader.emplace(ReaderInBlockAt(aBlockIndex)); + } + return std::forward<Callback>(aCallback)(std::move(maybeEntryReader)); + } + + // Main function to write entries. + // Reserve `aCallbackBytes()` bytes, call `aCallback()` with a pointer to an + // on-stack temporary ProfileBufferEntryWriter (nullptr when out-of-session), + // and return whatever `aCallback` returns. Callback should not store + // `ProfileBufferEntryWriter`, because it may become invalid after this + // thread-safe call. Note: `aCallbackBytes` is a callback instead of a simple + // value, to delay this potentially-expensive computation until after we're + // checked that we're in-session; use `Put(Length, Callback)` below if you + // know the size already. + template <typename CallbackBytes, typename Callback> + auto ReserveAndPut(CallbackBytes aCallbackBytes, Callback&& aCallback) { + Maybe<ProfileBufferEntryWriter> maybeEntryWriter; + + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + + if (MOZ_LIKELY(mMaybeUnderlyingBuffer)) { + const Length entryBytes = std::forward<CallbackBytes>(aCallbackBytes)(); + MOZ_RELEASE_ASSERT(entryBytes > 0); + const Length bufferBytes = + mMaybeUnderlyingBuffer->mBuffer.BufferLength().Value(); + MOZ_RELEASE_ASSERT(entryBytes <= bufferBytes - ULEB128Size(entryBytes), + "Entry would wrap and overwrite itself"); + // Compute block size from the requested entry size. + const Length blockBytes = ULEB128Size(entryBytes) + entryBytes; + // We will put this new block at the end of the current buffer. + const ProfileBufferIndex blockIndex = + mNextWriteIndex.ConvertToProfileBufferIndex(); + // Compute the end of this new block. + const ProfileBufferIndex blockEnd = blockIndex + blockBytes; + while (blockEnd > + mFirstReadIndex.ConvertToProfileBufferIndex() + bufferBytes) { + // About to trample on an old block. + ProfileBufferEntryReader reader = ReaderInBlockAt(mFirstReadIndex); + mMaybeUnderlyingBuffer->mClearedBlockCount += 1; + // Move the buffer reading start past this cleared block. + mFirstReadIndex = ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mFirstReadIndex.ConvertToProfileBufferIndex() + + ULEB128Size(reader.RemainingBytes()) + reader.RemainingBytes()); + } + // Store the new end of buffer. + mNextWriteIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(blockEnd); + mMaybeUnderlyingBuffer->mPushedBlockCount += 1; + // Finally, let aCallback write into the entry. + mMaybeUnderlyingBuffer->mBuffer.EntryWriterFromTo(maybeEntryWriter, + blockIndex, blockEnd); + MOZ_ASSERT(maybeEntryWriter.isSome(), + "Non-empty entry should always create an EntryWriter"); + maybeEntryWriter->WriteULEB128(entryBytes); + MOZ_ASSERT(maybeEntryWriter->RemainingBytes() == entryBytes); + } + +#ifdef DEBUG + auto checkAllWritten = MakeScopeExit([&]() { + MOZ_ASSERT(!maybeEntryWriter || maybeEntryWriter->RemainingBytes() == 0); + }); +#endif // DEBUG + return std::forward<Callback>(aCallback)(maybeEntryWriter); + } + + // Add a new entry of known size, call `aCallback` with a pointer to a + // temporary ProfileBufferEntryWriter (can be null when out-of-session), and + // return whatever `aCallback` returns. Callback should not store the + // `ProfileBufferEntryWriter`, as it may become invalid after this thread-safe + // call. + template <typename Callback> + auto Put(Length aBytes, Callback&& aCallback) { + return ReserveAndPut([aBytes]() { return aBytes; }, + std::forward<Callback>(aCallback)); + } + + // Add a new entry copied from the given buffer, return block index. + ProfileBufferBlockIndex PutFrom(const void* aSrc, Length aBytes) { + return ReserveAndPut([aBytes]() { return aBytes; }, + [&](Maybe<ProfileBufferEntryWriter>& aEntryWriter) { + if (MOZ_UNLIKELY(aEntryWriter.isNothing())) { + // Out-of-session, return "empty" index. + return ProfileBufferBlockIndex{}; + } + aEntryWriter->WriteBytes(aSrc, aBytes); + return aEntryWriter->CurrentBlockIndex(); + }); + } + + // Add a new single entry with *all* given object (using a Serializer for + // each), return block index. + template <typename... Ts> + ProfileBufferBlockIndex PutObjects(const Ts&... aTs) { + static_assert(sizeof...(Ts) > 0, + "PutObjects must be given at least one object."); + return ReserveAndPut( + [&]() { return ProfileBufferEntryWriter::SumBytes(aTs...); }, + [&](Maybe<ProfileBufferEntryWriter>& aEntryWriter) { + if (MOZ_UNLIKELY(aEntryWriter.isNothing())) { + // Out-of-session, return "empty" index. + return ProfileBufferBlockIndex{}; + } + aEntryWriter->WriteObjects(aTs...); + return aEntryWriter->CurrentBlockIndex(); + }); + } + + // Add a new entry copied from the given object, return block index. + template <typename T> + ProfileBufferBlockIndex PutObject(const T& aOb) { + return PutObjects(aOb); + } + + // Append the contents of another BlocksRingBuffer to this one. + ProfileBufferBlockIndex AppendContents(const BlocksRingBuffer& aSrc) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + + if (MOZ_UNLIKELY(!mMaybeUnderlyingBuffer)) { + // We are out-of-session, could not append contents. + return ProfileBufferBlockIndex{}; + } + + baseprofiler::detail::BaseProfilerMaybeAutoLock srcLock(aSrc.mMutex); + + if (MOZ_UNLIKELY(!aSrc.mMaybeUnderlyingBuffer)) { + // The other BRB is out-of-session, nothing to copy, we're done. + return ProfileBufferBlockIndex{}; + } + + const ProfileBufferIndex srcStartIndex = + aSrc.mFirstReadIndex.ConvertToProfileBufferIndex(); + const ProfileBufferIndex srcEndIndex = + aSrc.mNextWriteIndex.ConvertToProfileBufferIndex(); + const Length bytesToCopy = static_cast<Length>(srcEndIndex - srcStartIndex); + + if (MOZ_UNLIKELY(bytesToCopy == 0)) { + // The other BRB is empty, nothing to copy, we're done. + return ProfileBufferBlockIndex{}; + } + + const Length bufferBytes = + mMaybeUnderlyingBuffer->mBuffer.BufferLength().Value(); + + MOZ_RELEASE_ASSERT(bytesToCopy <= bufferBytes, + "Entry would wrap and overwrite itself"); + + // We will put all copied blocks at the end of the current buffer. + const ProfileBufferIndex dstStartIndex = + mNextWriteIndex.ConvertToProfileBufferIndex(); + // Compute where the copy will end... + const ProfileBufferIndex dstEndIndex = dstStartIndex + bytesToCopy; + + while (dstEndIndex > + mFirstReadIndex.ConvertToProfileBufferIndex() + bufferBytes) { + // About to trample on an old block. + ProfileBufferEntryReader reader = ReaderInBlockAt(mFirstReadIndex); + mMaybeUnderlyingBuffer->mClearedBlockCount += 1; + // Move the buffer reading start past this cleared block. + mFirstReadIndex = ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mFirstReadIndex.ConvertToProfileBufferIndex() + + ULEB128Size(reader.RemainingBytes()) + reader.RemainingBytes()); + } + + // Store the new end of buffer. + mNextWriteIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(dstEndIndex); + // Update our pushed count with the number of live blocks we are copying. + mMaybeUnderlyingBuffer->mPushedBlockCount += + aSrc.mMaybeUnderlyingBuffer->mPushedBlockCount - + aSrc.mMaybeUnderlyingBuffer->mClearedBlockCount; + + auto reader = aSrc.mMaybeUnderlyingBuffer->mBuffer.EntryReaderFromTo( + srcStartIndex, srcEndIndex, nullptr, nullptr); + auto writer = mMaybeUnderlyingBuffer->mBuffer.EntryWriterFromTo( + dstStartIndex, dstEndIndex); + writer.WriteFromReader(reader, bytesToCopy); + + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex(dstStartIndex); + } + + // Clear all entries: Move read index to the end so that these entries cannot + // be read anymore. + void Clear() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + ClearAllEntries(); + } + + // Clear all entries strictly before aBlockIndex, and move read index to the + // end so that these entries cannot be read anymore. + void ClearBefore(ProfileBufferBlockIndex aBlockIndex) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (!mMaybeUnderlyingBuffer) { + return; + } + // Don't accept a not-yet-written index. One-past-the-end is ok. + MOZ_ASSERT(aBlockIndex <= mNextWriteIndex); + if (aBlockIndex <= mFirstReadIndex) { + // Already cleared. + return; + } + if (aBlockIndex == mNextWriteIndex) { + // Right past the end, just clear everything. + ClearAllEntries(); + return; + } + // Otherwise we need to clear a subset of entries. + AssertBlockIndexIsValid(aBlockIndex); + // Just count skipped entries. + Reader reader(*this); + BlockIterator it = reader.begin(); + for (; it.CurrentBlockIndex() < aBlockIndex; ++it) { + MOZ_ASSERT(it.CurrentBlockIndex() < reader.end().CurrentBlockIndex()); + mMaybeUnderlyingBuffer->mClearedBlockCount += 1; + } + MOZ_ASSERT(it.CurrentBlockIndex() == aBlockIndex); + // Move read index to given index, so there's effectively no more entries + // before. + mFirstReadIndex = aBlockIndex; + } + +#ifdef DEBUG + void Dump() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (!mMaybeUnderlyingBuffer) { + printf("empty BlocksRingBuffer\n"); + return; + } + using ULL = unsigned long long; + printf("start=%llu (%llu) end=%llu (%llu) - ", + ULL(mFirstReadIndex.ConvertToProfileBufferIndex()), + ULL(mFirstReadIndex.ConvertToProfileBufferIndex() & + (mMaybeUnderlyingBuffer->mBuffer.BufferLength().Value() - 1)), + ULL(mNextWriteIndex.ConvertToProfileBufferIndex()), + ULL(mNextWriteIndex.ConvertToProfileBufferIndex() & + (mMaybeUnderlyingBuffer->mBuffer.BufferLength().Value() - 1))); + mMaybeUnderlyingBuffer->mBuffer.Dump(); + } +#endif // DEBUG + + private: + // In DEBUG mode, assert that `aBlockIndex` is a valid index for a live block. + // (Not just in range, but points exactly at the start of a block.) + // Slow, so avoid it for internal checks; this is more to check what callers + // provide us. + void AssertBlockIndexIsValid(ProfileBufferBlockIndex aBlockIndex) const { +#ifdef DEBUG + mMutex.AssertCurrentThreadOwns(); + MOZ_ASSERT(aBlockIndex >= mFirstReadIndex); + MOZ_ASSERT(aBlockIndex < mNextWriteIndex); + // Quick check (default), or slow check (change '1' to '0') below: +# if 1 + // Quick check that this looks like a valid block start. + // Read the entry size at the start of the block. + const Length entryBytes = ReaderInBlockAt(aBlockIndex).RemainingBytes(); + MOZ_ASSERT(entryBytes > 0, "Empty entries are not allowed"); + MOZ_ASSERT( + entryBytes < mMaybeUnderlyingBuffer->mBuffer.BufferLength().Value() - + ULEB128Size(entryBytes), + "Entry would wrap and overwrite itself"); + // The end of the block should be inside the live buffer range. + MOZ_ASSERT(aBlockIndex.ConvertToProfileBufferIndex() + + ULEB128Size(entryBytes) + entryBytes <= + mNextWriteIndex.ConvertToProfileBufferIndex()); +# else + // Slow check that the index is really the start of the block. + // This kills performances, as it reads from the first index until + // aBlockIndex. Only use to debug issues locally. + Reader reader(*this); + BlockIterator it = reader.begin(); + for (; it.CurrentBlockIndex() < aBlockIndex; ++it) { + MOZ_ASSERT(it.CurrentBlockIndex() < reader.end().CurrentBlockIndex()); + } + MOZ_ASSERT(it.CurrentBlockIndex() == aBlockIndex); +# endif +#endif // DEBUG + } + + // In DEBUG mode, assert that `aBlockIndex` is a valid index for a live block, + // or is just past-the-end. (Not just in range, but points exactly at the + // start of a block.) Slow, so avoid it for internal checks; this is more to + // check what callers provide us. + void AssertBlockIndexIsValidOrEnd(ProfileBufferBlockIndex aBlockIndex) const { +#ifdef DEBUG + mMutex.AssertCurrentThreadOwns(); + if (aBlockIndex == mNextWriteIndex) { + return; + } + AssertBlockIndexIsValid(aBlockIndex); +#endif // DEBUG + } + + // Create a reader for the block starting at aBlockIndex. + ProfileBufferEntryReader ReaderInBlockAt( + ProfileBufferBlockIndex aBlockIndex) const { + mMutex.AssertCurrentThreadOwns(); + MOZ_ASSERT(mMaybeUnderlyingBuffer.isSome()); + MOZ_ASSERT(aBlockIndex >= mFirstReadIndex); + MOZ_ASSERT(aBlockIndex < mNextWriteIndex); + // Create a reader from the given index until the end of the buffer. + ProfileBufferEntryReader reader = + mMaybeUnderlyingBuffer->mBuffer.EntryReaderFromTo( + aBlockIndex.ConvertToProfileBufferIndex(), + mNextWriteIndex.ConvertToProfileBufferIndex(), nullptr, nullptr); + // Read the block size at the beginning. + const Length entryBytes = reader.ReadULEB128<Length>(); + // Make sure we don't overshoot the buffer. + MOZ_RELEASE_ASSERT(entryBytes <= reader.RemainingBytes()); + ProfileBufferIndex nextBlockIndex = + aBlockIndex.ConvertToProfileBufferIndex() + ULEB128Size(entryBytes) + + entryBytes; + // And reduce the reader to the entry area. Only provide a next-block-index + // if it's not at the end of the buffer (i.e., there's an actual block + // there). + reader = mMaybeUnderlyingBuffer->mBuffer.EntryReaderFromTo( + aBlockIndex.ConvertToProfileBufferIndex() + ULEB128Size(entryBytes), + nextBlockIndex, aBlockIndex, + (nextBlockIndex < mNextWriteIndex.ConvertToProfileBufferIndex()) + ? ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + nextBlockIndex) + : ProfileBufferBlockIndex{}); + return reader; + } + + ProfileBufferEntryReader FullBufferReader() const { + mMutex.AssertCurrentThreadOwns(); + if (!mMaybeUnderlyingBuffer) { + return {}; + } + return mMaybeUnderlyingBuffer->mBuffer.EntryReaderFromTo( + mFirstReadIndex.ConvertToProfileBufferIndex(), + mNextWriteIndex.ConvertToProfileBufferIndex(), nullptr, nullptr); + } + + // Clear all entries: Move read index to the end so that these entries cannot + // be read anymore. + void ClearAllEntries() { + mMutex.AssertCurrentThreadOwns(); + if (!mMaybeUnderlyingBuffer) { + return; + } + // Mark all entries pushed so far as cleared. + mMaybeUnderlyingBuffer->mClearedBlockCount = + mMaybeUnderlyingBuffer->mPushedBlockCount; + // Move read index to write index, so there's effectively no more entries + // that can be read. (Not setting both to 0, in case user is keeping + // `ProfileBufferBlockIndex`'es to old entries.) + mFirstReadIndex = mNextWriteIndex; + } + + // If there is an underlying buffer, clear all entries, and discard the + // buffer. This BlocksRingBuffer will now gracefully reject all API calls, and + // is in a state where a new underlying buffer may be set. + void ResetUnderlyingBuffer() { + mMutex.AssertCurrentThreadOwns(); + if (!mMaybeUnderlyingBuffer) { + return; + } + ClearAllEntries(); + mMaybeUnderlyingBuffer.reset(); + } + + // Used to de/serialize a BlocksRingBuffer (e.g., containing a backtrace). + friend ProfileBufferEntryWriter::Serializer<BlocksRingBuffer>; + friend ProfileBufferEntryReader::Deserializer<BlocksRingBuffer>; + friend ProfileBufferEntryWriter::Serializer<UniquePtr<BlocksRingBuffer>>; + friend ProfileBufferEntryReader::Deserializer<UniquePtr<BlocksRingBuffer>>; + + // Mutex guarding the following members. + mutable baseprofiler::detail::BaseProfilerMaybeMutex mMutex; + + struct UnderlyingBuffer { + // Create a buffer of the given length. + explicit UnderlyingBuffer(PowerOfTwo<Length> aLength) : mBuffer(aLength) { + MOZ_ASSERT(aLength.Value() > ULEB128MaxSize<Length>(), + "Buffer should be able to contain more than a block size"); + } + + // Take ownership of an existing buffer. + UnderlyingBuffer(UniquePtr<Buffer::Byte[]> aExistingBuffer, + PowerOfTwo<Length> aLength) + : mBuffer(std::move(aExistingBuffer), aLength) { + MOZ_ASSERT(aLength.Value() > ULEB128MaxSize<Length>(), + "Buffer should be able to contain more than a block size"); + } + + // Use an externally-owned buffer. + UnderlyingBuffer(Buffer::Byte* aExternalBuffer, PowerOfTwo<Length> aLength) + : mBuffer(aExternalBuffer, aLength) { + MOZ_ASSERT(aLength.Value() > ULEB128MaxSize<Length>(), + "Buffer should be able to contain more than a block size"); + } + + // Only allow move-construction. + UnderlyingBuffer(UnderlyingBuffer&&) = default; + + // Copies and move-assignment are explictly disallowed. + UnderlyingBuffer(const UnderlyingBuffer&) = delete; + UnderlyingBuffer& operator=(const UnderlyingBuffer&) = delete; + UnderlyingBuffer& operator=(UnderlyingBuffer&&) = delete; + + // Underlying circular byte buffer. + Buffer mBuffer; + + // Statistics. + uint64_t mPushedBlockCount = 0; + uint64_t mClearedBlockCount = 0; + }; + + // Underlying buffer, with stats. + // Only valid during in-session period. + Maybe<UnderlyingBuffer> mMaybeUnderlyingBuffer; + + // Index to the first block to be read (or cleared). Initialized to 1 because + // 0 is reserved for the "empty" ProfileBufferBlockIndex value. Kept between + // sessions, so that stored indices from one session will be gracefully denied + // in future sessions. + ProfileBufferBlockIndex mFirstReadIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + ProfileBufferIndex(1)); + // Index where the next new block should be allocated. Initialized to 1 + // because 0 is reserved for the "empty" ProfileBufferBlockIndex value. Kept + // between sessions, so that stored indices from one session will be + // gracefully denied in future sessions. + ProfileBufferBlockIndex mNextWriteIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + ProfileBufferIndex(1)); +}; + +// ---------------------------------------------------------------------------- +// BlocksRingBuffer serialization + +// A BlocksRingBuffer can hide another one! +// This will be used to store marker backtraces; They can be read back into a +// UniquePtr<BlocksRingBuffer>. +// Format: len (ULEB128) | start | end | buffer (len bytes) | pushed | cleared +// len==0 marks an out-of-session buffer, or empty buffer. +template <> +struct ProfileBufferEntryWriter::Serializer<BlocksRingBuffer> { + static Length Bytes(const BlocksRingBuffer& aBuffer) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(aBuffer.mMutex); + if (aBuffer.mMaybeUnderlyingBuffer.isNothing()) { + // Out-of-session, we only need 1 byte to store a length of 0. + return ULEB128Size<Length>(0); + } + const auto start = aBuffer.mFirstReadIndex.ConvertToProfileBufferIndex(); + const auto end = aBuffer.mNextWriteIndex.ConvertToProfileBufferIndex(); + const auto len = end - start; + if (len == 0) { + // In-session but empty, also store a length of 0. + return ULEB128Size<Length>(0); + } + return ULEB128Size(len) + sizeof(start) + sizeof(end) + len + + sizeof(aBuffer.mMaybeUnderlyingBuffer->mPushedBlockCount) + + sizeof(aBuffer.mMaybeUnderlyingBuffer->mClearedBlockCount); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const BlocksRingBuffer& aBuffer) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(aBuffer.mMutex); + if (aBuffer.mMaybeUnderlyingBuffer.isNothing()) { + // Out-of-session, only store a length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + const auto start = aBuffer.mFirstReadIndex.ConvertToProfileBufferIndex(); + const auto end = aBuffer.mNextWriteIndex.ConvertToProfileBufferIndex(); + MOZ_ASSERT(end - start <= std::numeric_limits<Length>::max()); + const auto len = static_cast<Length>(end - start); + if (len == 0) { + // In-session but empty, only store a length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + // In-session. + // Store buffer length, start and end indices. + aEW.WriteULEB128<Length>(len); + aEW.WriteObject(start); + aEW.WriteObject(end); + // Write all the bytes. + auto reader = aBuffer.FullBufferReader(); + aEW.WriteFromReader(reader, reader.RemainingBytes()); + // And write stats. + aEW.WriteObject(aBuffer.mMaybeUnderlyingBuffer->mPushedBlockCount); + aEW.WriteObject(aBuffer.mMaybeUnderlyingBuffer->mClearedBlockCount); + } +}; + +// A serialized BlocksRingBuffer can be read into an empty buffer (either +// out-of-session, or in-session with enough room). +template <> +struct ProfileBufferEntryReader::Deserializer<BlocksRingBuffer> { + static void ReadInto(ProfileBufferEntryReader& aER, + BlocksRingBuffer& aBuffer) { + // Expect an empty buffer, as we're going to overwrite it. + MOZ_ASSERT(aBuffer.GetState().mRangeStart == aBuffer.GetState().mRangeEnd); + // Read the stored buffer length. + const auto len = aER.ReadULEB128<Length>(); + if (len == 0) { + // 0-length means an "uninteresting" buffer, just return now. + return; + } + // We have a non-empty buffer to read. + if (aBuffer.BufferLength().isSome()) { + // Output buffer is in-session (i.e., it already has a memory buffer + // attached). Make sure the caller allocated enough space. + MOZ_RELEASE_ASSERT(aBuffer.BufferLength()->Value() >= len); + } else { + // Output buffer is out-of-session, attach a new memory buffer. + aBuffer.Set(PowerOfTwo<Length>(len)); + MOZ_ASSERT(aBuffer.BufferLength()->Value() >= len); + } + // Read start and end indices. + const auto start = aER.ReadObject<ProfileBufferIndex>(); + aBuffer.mFirstReadIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(start); + const auto end = aER.ReadObject<ProfileBufferIndex>(); + aBuffer.mNextWriteIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(end); + MOZ_ASSERT(end - start == len); + // Copy bytes into the buffer. + auto writer = + aBuffer.mMaybeUnderlyingBuffer->mBuffer.EntryWriterFromTo(start, end); + writer.WriteFromReader(aER, end - start); + MOZ_ASSERT(writer.RemainingBytes() == 0); + // Finally copy stats. + aBuffer.mMaybeUnderlyingBuffer->mPushedBlockCount = aER.ReadObject<decltype( + aBuffer.mMaybeUnderlyingBuffer->mPushedBlockCount)>(); + aBuffer.mMaybeUnderlyingBuffer->mClearedBlockCount = + aER.ReadObject<decltype( + aBuffer.mMaybeUnderlyingBuffer->mClearedBlockCount)>(); + } + + // We cannot output a BlocksRingBuffer object (not copyable), use `ReadInto()` + // or `aER.ReadObject<UniquePtr<BlocksRinbBuffer>>()` instead. + static BlocksRingBuffer Read(ProfileBufferEntryReader& aER) = delete; +}; + +// A BlocksRingBuffer is usually refererenced through a UniquePtr, for +// convenience we support (de)serializing that UniquePtr directly. +// This is compatible with the non-UniquePtr serialization above, with a null +// pointer being treated like an out-of-session or empty buffer; and any of +// these would be deserialized into a null pointer. +template <> +struct ProfileBufferEntryWriter::Serializer<UniquePtr<BlocksRingBuffer>> { + static Length Bytes(const UniquePtr<BlocksRingBuffer>& aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + return ULEB128Size<Length>(0); + } + // Otherwise write the pointed-at BlocksRingBuffer (which could be + // out-of-session or empty.) + return SumBytes(*aBufferUPtr); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const UniquePtr<BlocksRingBuffer>& aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + // Otherwise write the pointed-at BlocksRingBuffer (which could be + // out-of-session or empty.) + aEW.WriteObject(*aBufferUPtr); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<UniquePtr<BlocksRingBuffer>> { + static void ReadInto(ProfileBufferEntryReader& aER, + UniquePtr<BlocksRingBuffer>& aBuffer) { + aBuffer = Read(aER); + } + + static UniquePtr<BlocksRingBuffer> Read(ProfileBufferEntryReader& aER) { + UniquePtr<BlocksRingBuffer> bufferUPtr; + // Keep a copy of the reader before reading the length, so we can restart + // from here below. + ProfileBufferEntryReader readerBeforeLen = aER; + // Read the stored buffer length. + const auto len = aER.ReadULEB128<Length>(); + if (len == 0) { + // 0-length means an "uninteresting" buffer, just return nullptr. + return bufferUPtr; + } + // We have a non-empty buffer. + // allocate an empty BlocksRingBuffer without mutex. + bufferUPtr = MakeUnique<BlocksRingBuffer>( + BlocksRingBuffer::ThreadSafety::WithoutMutex); + // Rewind the reader before the length and deserialize the contents, using + // the non-UniquePtr Deserializer. + aER = readerBeforeLen; + aER.ReadIntoObject(*bufferUPtr); + return bufferUPtr; + } +}; + +} // namespace mozilla + +#endif // BlocksRingBuffer_h diff --git a/mozglue/baseprofiler/public/ModuloBuffer.h b/mozglue/baseprofiler/public/ModuloBuffer.h new file mode 100644 index 0000000000..80e765279e --- /dev/null +++ b/mozglue/baseprofiler/public/ModuloBuffer.h @@ -0,0 +1,618 @@ +/* -*- 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 ModuloBuffer_h +#define ModuloBuffer_h + +#include "mozilla/leb128iterator.h" +#include "mozilla/Maybe.h" +#include "mozilla/MemoryReporting.h" +#include "mozilla/NotNull.h" +#include "mozilla/PowerOfTwo.h" +#include "mozilla/ProfileBufferEntrySerialization.h" +#include "mozilla/UniquePtr.h" + +#include <functional> +#include <iterator> +#include <limits> +#include <type_traits> + +namespace mozilla { + +// The ModuloBuffer class is a circular buffer that holds raw byte values, with +// data-read/write helpers. +// +// OffsetT: Type of the internal offset into the buffer of bytes, it should be +// large enough to access all bytes of the buffer. It will also be used as +// Length (in bytes) of the buffer and of any subset. Default uint32_t +// IndexT: Type of the external index, it should be large enough that overflows +// should not happen during the lifetime of the ModuloBuffer. +// +// The basic usage is to create an iterator-like object with `ReaderAt(Index)` +// or `WriterAt(Index)`, and use it to read/write data blobs. Iterators +// automatically manage the wrap-around (through "Modulo", which is effectively +// an AND-masking with the PowerOfTwo buffer size.) +// +// There is zero safety: No thread safety, no checks that iterators may be +// overwriting data that's still to be read, etc. It's up to the caller to add +// adequate checks. +// The intended use is as an underlying buffer for a safer container. +template <typename OffsetT = uint32_t, typename IndexT = uint64_t> +class ModuloBuffer { + public: + using Byte = uint8_t; + static_assert(sizeof(Byte) == 1, "ModuloBuffer::Byte must be 1 byte"); + using Offset = OffsetT; + static_assert(!std::numeric_limits<Offset>::is_signed, + "ModuloBuffer::Offset must be an unsigned integral type"); + using Length = Offset; + using Index = IndexT; + static_assert(!std::numeric_limits<Index>::is_signed, + "ModuloBuffer::Index must be an unsigned integral type"); + static_assert(sizeof(Index) >= sizeof(Offset), + "ModuloBuffer::Index size must >= Offset"); + + // Create a buffer of the given length. + explicit ModuloBuffer(PowerOfTwo<Length> aLength) + : mMask(aLength.Mask()), + mBuffer(WrapNotNull(new Byte[aLength.Value()])), + mBufferDeleter([](Byte* aBuffer) { delete[] aBuffer; }) {} + + // Take ownership of an existing buffer. Existing contents is ignored. + // Done by extracting the raw pointer from UniquePtr<Byte[]>, and adding + // an equivalent `delete[]` in `mBufferDeleter`. + ModuloBuffer(UniquePtr<Byte[]> aExistingBuffer, PowerOfTwo<Length> aLength) + : mMask(aLength.Mask()), + mBuffer(WrapNotNull(aExistingBuffer.release())), + mBufferDeleter([](Byte* aBuffer) { delete[] aBuffer; }) {} + + // Use an externally-owned buffer. Existing contents is ignored. + ModuloBuffer(Byte* aExternalBuffer, PowerOfTwo<Length> aLength) + : mMask(aLength.Mask()), mBuffer(WrapNotNull(aExternalBuffer)) {} + + // Disallow copying, as we may uniquely own the resource. + ModuloBuffer(const ModuloBuffer& aOther) = delete; + ModuloBuffer& operator=(const ModuloBuffer& aOther) = delete; + + // Allow move-construction. Stealing ownership if the original had it. + // This effectively prevents copy construction, and all assignments; needed so + // that a ModuloBuffer may be initialized from a separate construction. + // The moved-from ModuloBuffer still points at the resource but doesn't own + // it, so it won't try to free it; but accesses are not guaranteed, so it + // should not be used anymore. + ModuloBuffer(ModuloBuffer&& aOther) + : mMask(std::move(aOther.mMask)), + mBuffer(std::move(aOther.mBuffer)), + mBufferDeleter(std::move(aOther.mBufferDeleter)) { + // The above move leaves `aOther.mBufferDeleter` in a valid state but with + // an unspecified value, so it could theoretically still contain the + // original function, which would be bad because we don't want aOther to + // delete the resource that `this` now owns. + if (aOther.mBufferDeleter) { + // `aOther` still had a non-empty deleter, reset it. + aOther.mBufferDeleter = nullptr; + } + } + + // Disallow assignment, as we have some `const` members. + ModuloBuffer& operator=(ModuloBuffer&& aOther) = delete; + + // Destructor, deletes the resource if we uniquely own it. + ~ModuloBuffer() { + if (mBufferDeleter) { + mBufferDeleter(mBuffer); + } + } + + PowerOfTwo<Length> BufferLength() const { + return PowerOfTwo<Length>(mMask.MaskValue() + 1); + } + + // Size of external resources. + // Note: `mBufferDeleter`'s potential external data (for its captures) is not + // included, as it's hidden in the `std::function` implementation. + size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { + if (!mBufferDeleter) { + // If we don't have a buffer deleter, assume we don't own the data, so + // it's probably on the stack, or should be reported by its owner. + return 0; + } + return aMallocSizeOf(mBuffer); + } + + size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); + } + + ProfileBufferEntryReader EntryReaderFromTo( + Index aStart, Index aEnd, ProfileBufferBlockIndex aBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) const { + using EntrySpan = Span<const ProfileBufferEntryReader::Byte>; + if (aStart == aEnd) { + return ProfileBufferEntryReader{}; + } + // Don't allow over-wrapping. + MOZ_ASSERT(aEnd - aStart <= mMask.MaskValue() + 1); + // Start offset in 0 .. (buffer size - 1) + Offset start = static_cast<Offset>(aStart) & mMask; + // End offset in 1 .. (buffer size) + Offset end = (static_cast<Offset>(aEnd - 1) & mMask) + 1; + if (start < end) { + // Segment doesn't cross buffer threshold, one span is enough. + return ProfileBufferEntryReader{EntrySpan(&mBuffer[start], end - start), + aBlockIndex, aNextBlockIndex}; + } + // Segment crosses buffer threshold, we need one span until the end and one + // span restarting at the beginning of the buffer. + return ProfileBufferEntryReader{ + EntrySpan(&mBuffer[start], mMask.MaskValue() + 1 - start), + EntrySpan(&mBuffer[0], end), aBlockIndex, aNextBlockIndex}; + } + + // Return an entry writer for the given range. + ProfileBufferEntryWriter EntryWriterFromTo(Index aStart, Index aEnd) const { + using EntrySpan = Span<ProfileBufferEntryReader::Byte>; + if (aStart == aEnd) { + return ProfileBufferEntryWriter{}; + } + MOZ_ASSERT(aEnd - aStart <= mMask.MaskValue() + 1); + // Start offset in 0 .. (buffer size - 1) + Offset start = static_cast<Offset>(aStart) & mMask; + // End offset in 1 .. (buffer size) + Offset end = (static_cast<Offset>(aEnd - 1) & mMask) + 1; + if (start < end) { + // Segment doesn't cross buffer threshold, one span is enough. + return ProfileBufferEntryWriter{ + EntrySpan(&mBuffer[start], end - start), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aStart), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aEnd)}; + } + // Segment crosses buffer threshold, we need one span until the end and one + // span restarting at the beginning of the buffer. + return ProfileBufferEntryWriter{ + EntrySpan(&mBuffer[start], mMask.MaskValue() + 1 - start), + EntrySpan(&mBuffer[0], end), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aStart), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aEnd)}; + } + + // Emplace an entry writer into `aMaybeEntryWriter` for the given range. + void EntryWriterFromTo(Maybe<ProfileBufferEntryWriter>& aMaybeEntryWriter, + Index aStart, Index aEnd) const { + MOZ_ASSERT(aMaybeEntryWriter.isNothing(), + "Reference entry writer should be Nothing."); + using EntrySpan = Span<ProfileBufferEntryReader::Byte>; + if (aStart == aEnd) { + return; + } + MOZ_ASSERT(aEnd - aStart <= mMask.MaskValue() + 1); + // Start offset in 0 .. (buffer size - 1) + Offset start = static_cast<Offset>(aStart) & mMask; + // End offset in 1 .. (buffer size) + Offset end = (static_cast<Offset>(aEnd - 1) & mMask) + 1; + if (start < end) { + // Segment doesn't cross buffer threshold, one span is enough. + aMaybeEntryWriter.emplace( + EntrySpan(&mBuffer[start], end - start), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aStart), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aEnd)); + } else { + // Segment crosses buffer threshold, we need one span until the end and + // one span restarting at the beginning of the buffer. + aMaybeEntryWriter.emplace( + EntrySpan(&mBuffer[start], mMask.MaskValue() + 1 - start), + EntrySpan(&mBuffer[0], end), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aStart), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex(aEnd)); + } + } + + // All ModuloBuffer operations should be done through this iterator, which has + // an effectively infinite range. The underlying wrapping-around is hidden. + // Use `ReaderAt(Index)` or `WriterAt(Index)` to create it. + // + // `const Iterator<...>` means the iterator itself cannot change, i.e., it + // cannot move, and only its const methods are available. Note that these + // const methods may still be used to modify the buffer contents (e.g.: + // `operator*()`, `Poke()`). + // + // `Iterator</*IsBufferConst=*/true>` means the buffer contents cannot be + // modified, i.e., write operations are forbidden, but the iterator may still + // move if non-const itself. + template <bool IsBufferConst> + class Iterator { + // Alias to const- or mutable-`ModuloBuffer` depending on `IsBufferConst`. + using ConstOrMutableBuffer = + std::conditional_t<IsBufferConst, const ModuloBuffer, ModuloBuffer>; + + // Implementation note about the strange enable-if's below: + // `template <bool NotIBC = !IsBufferConst> enable_if_t<NotIBC>` + // which intuitively could be simplified to: + // `enable_if_t<!IsBufferConst>` + // The former extra-templated syntax is in fact necessary to delay + // instantiation of these functions until they are actually needed. + // + // If we were just doing `enable_if_t<!IsBufferConst>`, this would only + // depend on the *class* (`ModuloBuffer<...>::Iterator`), which gets + // instantiated when a `ModuloBuffer` is created with some template + // arguments; at that point, all non-templated methods get instantiated, so + // there's no "SFINAE" happening, and `enable_if_t<...>` is actually doing + // `typename enable_if<...>::type` on the spot, but there is no `type` if + // `IsBufferConst` is true, so it just fails right away. E.g.: + // error: no type named 'type' in 'std::enable_if<false, void>'; + // 'enable_if' cannot be used to disable this declaration + // note: in instantiation of template type alias 'enable_if_t' + // > std::enable_if_t<!IsBufferConst> WriteObject(const T& aObject) { + // in instantiation of template class + // 'mozilla::ModuloBuffer<...>::Iterator<true>' + // > auto it = mb.ReaderAt(1); + // + // By adding another template level `template <bool NotIsBufferConst = + // !IsBufferConst>`, the instantiation is delayed until the function is + // actually invoked somewhere, e.g. `it.Poke(...);`. + // So at that invocation point, the compiler looks for a "Poke" name in it, + // and considers potential template instantiations that could work. The + // `enable_if_t` is *now* attempted, with `NotIsBufferConst` taking its + // value from `!IsBufferConst`: + // - If `IsBufferConst` is false, `NotIsBufferConst` is true, + // `enable_if<NotIsBufferConst>` does define a `type` (`void` by default), + // so `enable_if_t` happily becomes `void`, the function exists and may be + // called. + // - Otherwise if `IsBufferConst` is true, `NotIsBufferConst` is false, + // `enable_if<NotIsBufferConst>` does *not* define a `type`, therefore + // `enable_if_t` produces an error because there is no `type`. Now "SFINAE" + // happens: This "Substitution Failure Is Not An Error" (by itself)... But + // then, there are no other functions named "Poke" as requested in the + // `it.Poke(...);` call, so we are now getting an error (can't find + // function), as expected because `it` had `IsBufferConst`==true. (But at + // least the compiler waited until this invocation attempt before outputting + // an error.) + // + // C++ is fun! + + public: + // These definitions are expected by std functions, to recognize this as an + // iterator. See https://en.cppreference.com/w/cpp/iterator/iterator_traits + using difference_type = Index; + using value_type = Byte; + using pointer = std::conditional_t<IsBufferConst, const Byte*, Byte*>; + using reference = std::conditional_t<IsBufferConst, const Byte&, Byte&>; + using iterator_category = std::random_access_iterator_tag; + + // Can always copy/assign from the same kind of iterator. + Iterator(const Iterator& aRhs) = default; + Iterator& operator=(const Iterator& aRhs) = default; + + // Can implicitly copy an Iterator-to-mutable (reader+writer) to + // Iterator-to-const (reader-only), but not the reverse. + template <bool IsRhsBufferConst, + typename = std::enable_if_t<(!IsRhsBufferConst) && IsBufferConst>> + MOZ_IMPLICIT Iterator(const Iterator<IsRhsBufferConst>& aRhs) + : mModuloBuffer(aRhs.mModuloBuffer), mIndex(aRhs.mIndex) {} + + // Can implicitly assign from an Iterator-to-mutable (reader+writer) to + // Iterator-to-const (reader-only), but not the reverse. + template <bool IsRhsBufferConst, + typename = std::enable_if_t<(!IsRhsBufferConst) && IsBufferConst>> + Iterator& operator=(const Iterator<IsRhsBufferConst>& aRhs) { + mModuloBuffer = aRhs.mModuloBuffer; + mIndex = aRhs.mIndex; + return *this; + } + + // Current location of the iterator in the `Index` range. + // Note that due to wrapping, multiple indices may effectively point at the + // same byte in the buffer. + Index CurrentIndex() const { return mIndex; } + + // Location comparison in the `Index` range. I.e., two `Iterator`s may look + // unequal, but refer to the same buffer location. + // Must be on the same buffer. + bool operator==(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex == aRhs.mIndex; + } + bool operator!=(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex != aRhs.mIndex; + } + bool operator<(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex < aRhs.mIndex; + } + bool operator<=(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex <= aRhs.mIndex; + } + bool operator>(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex > aRhs.mIndex; + } + bool operator>=(const Iterator& aRhs) const { + MOZ_ASSERT(mModuloBuffer == aRhs.mModuloBuffer); + return mIndex >= aRhs.mIndex; + } + + // Movement in the `Index` range. + Iterator& operator++() { + ++mIndex; + return *this; + } + Iterator operator++(int) { + Iterator here(*mModuloBuffer, mIndex); + ++mIndex; + return here; + } + Iterator& operator--() { + --mIndex; + return *this; + } + Iterator operator--(int) { + Iterator here(*mModuloBuffer, mIndex); + --mIndex; + return here; + } + Iterator& operator+=(Length aLength) { + mIndex += aLength; + return *this; + } + Iterator operator+(Length aLength) const { + return Iterator(*mModuloBuffer, mIndex + aLength); + } + friend Iterator operator+(Length aLength, const Iterator& aIt) { + return aIt + aLength; + } + Iterator& operator-=(Length aLength) { + mIndex -= aLength; + return *this; + } + Iterator operator-(Length aLength) const { + return Iterator(*mModuloBuffer, mIndex - aLength); + } + + // Distance from `aRef` to here in the `Index` range. + // May be negative (as 2's complement) if `aRef > *this`. + Index operator-(const Iterator& aRef) const { + MOZ_ASSERT(mModuloBuffer == aRef.mModuloBuffer); + return mIndex - aRef.mIndex; + } + + // Dereference a single byte (read-only if `IsBufferConst` is true). + reference operator*() const { + return mModuloBuffer->mBuffer[OffsetInBuffer()]; + } + + // Random-access dereference. + reference operator[](Length aLength) const { return *(*this + aLength); } + + // Write data (if `IsBufferConst` is false) but don't move iterator. + template <bool NotIsBufferConst = !IsBufferConst> + std::enable_if_t<NotIsBufferConst> Poke(const void* aSrc, + Length aLength) const { + // Don't allow data larger than the buffer. + MOZ_ASSERT(aLength <= mModuloBuffer->BufferLength().Value()); + // Offset inside the buffer (corresponding to our Index). + Offset offset = OffsetInBuffer(); + // Compute remaining bytes between this offset and the end of the buffer. + Length remaining = mModuloBuffer->BufferLength().Value() - offset; + if (MOZ_LIKELY(remaining >= aLength)) { + // Enough space to write everything before the end. + memcpy(&mModuloBuffer->mBuffer[offset], aSrc, aLength); + } else { + // Not enough space. Write as much as possible before the end. + memcpy(&mModuloBuffer->mBuffer[offset], aSrc, remaining); + // And then continue from the beginning of the buffer. + memcpy(&mModuloBuffer->mBuffer[0], + static_cast<const Byte*>(aSrc) + remaining, + (aLength - remaining)); + } + } + + // Write object data (if `IsBufferConst` is false) but don't move iterator. + // Note that this copies bytes from the object, with the intent to read them + // back later. Restricted to trivially-copyable types, which support this + // without Undefined Behavior! + template <typename T, bool NotIsBufferConst = !IsBufferConst> + std::enable_if_t<NotIsBufferConst> PokeObject(const T& aObject) const { + static_assert(std::is_trivially_copyable<T>::value, + "PokeObject<T> - T must be trivially copyable"); + return Poke(&aObject, sizeof(T)); + } + + // Write data (if `IsBufferConst` is false) and move iterator ahead. + template <bool NotIsBufferConst = !IsBufferConst> + std::enable_if_t<NotIsBufferConst> Write(const void* aSrc, Length aLength) { + Poke(aSrc, aLength); + mIndex += aLength; + } + + // Write object data (if `IsBufferConst` is false) and move iterator ahead. + // Note that this copies bytes from the object, with the intent to read them + // back later. Restricted to trivially-copyable types, which support this + // without Undefined Behavior! + template <typename T, bool NotIsBufferConst = !IsBufferConst> + std::enable_if_t<NotIsBufferConst> WriteObject(const T& aObject) { + static_assert(std::is_trivially_copyable<T>::value, + "WriteObject<T> - T must be trivially copyable"); + return Write(&aObject, sizeof(T)); + } + + // Number of bytes needed to represent `aValue` in unsigned LEB128. + template <typename T> + static unsigned ULEB128Size(T aValue) { + return ::mozilla::ULEB128Size(aValue); + } + + // Write number as unsigned LEB128 (if `IsBufferConst` is false) and move + // iterator ahead. + template <typename T, bool NotIsBufferConst = !IsBufferConst> + std::enable_if_t<NotIsBufferConst> WriteULEB128(T aValue) { + ::mozilla::WriteULEB128(aValue, *this); + } + + // Read data but don't move iterator. + void Peek(void* aDst, Length aLength) const { + // Don't allow data larger than the buffer. + MOZ_ASSERT(aLength <= mModuloBuffer->BufferLength().Value()); + // Offset inside the buffer (corresponding to our Index). + Offset offset = OffsetInBuffer(); + // Compute remaining bytes between this offset and the end of the buffer. + Length remaining = mModuloBuffer->BufferLength().Value() - offset; + if (MOZ_LIKELY(remaining >= aLength)) { + // Can read everything we need before the end of the buffer. + memcpy(aDst, &mModuloBuffer->mBuffer[offset], aLength); + } else { + // Read as much as possible before the end of the buffer. + memcpy(aDst, &mModuloBuffer->mBuffer[offset], remaining); + // And then continue from the beginning of the buffer. + memcpy(static_cast<Byte*>(aDst) + remaining, &mModuloBuffer->mBuffer[0], + (aLength - remaining)); + } + } + + // Read data into an object but don't move iterator. + // Note that this overwrites `aObject` with bytes from the buffer. + // Restricted to trivially-copyable types, which support this without + // Undefined Behavior! + template <typename T> + void PeekIntoObject(T& aObject) const { + static_assert(std::is_trivially_copyable<T>::value, + "PeekIntoObject<T> - T must be trivially copyable"); + Peek(&aObject, sizeof(T)); + } + + // Read data as an object but don't move iterator. + // Note that this creates an default `T` first, and then overwrites it with + // bytes from the buffer. Restricted to trivially-copyable types, which + // support this without Undefined Behavior! + template <typename T> + T PeekObject() const { + static_assert(std::is_trivially_copyable<T>::value, + "PeekObject<T> - T must be trivially copyable"); + T object; + PeekIntoObject(object); + return object; + } + + // Read data and move iterator ahead. + void Read(void* aDst, Length aLength) { + Peek(aDst, aLength); + mIndex += aLength; + } + + // Read data into a mutable iterator and move both iterators ahead. + void ReadInto(Iterator</* IsBufferConst */ false>& aDst, Length aLength) { + // Don't allow data larger than the buffer. + MOZ_ASSERT(aLength <= mModuloBuffer->BufferLength().Value()); + MOZ_ASSERT(aLength <= aDst.mModuloBuffer->BufferLength().Value()); + // Offset inside the buffer (corresponding to our Index). + Offset offset = OffsetInBuffer(); + // Compute remaining bytes between this offset and the end of the buffer. + Length remaining = mModuloBuffer->BufferLength().Value() - offset; + if (MOZ_LIKELY(remaining >= aLength)) { + // Can read everything we need before the end of the buffer. + aDst.Write(&mModuloBuffer->mBuffer[offset], aLength); + } else { + // Read as much as possible before the end of the buffer. + aDst.Write(&mModuloBuffer->mBuffer[offset], remaining); + // And then continue from the beginning of the buffer. + aDst.Write(&mModuloBuffer->mBuffer[0], (aLength - remaining)); + } + mIndex += aLength; + } + + // Read data into an object and move iterator ahead. + // Note that this overwrites `aObject` with bytes from the buffer. + // Restricted to trivially-copyable types, which support this without + // Undefined Behavior! + template <typename T> + void ReadIntoObject(T& aObject) { + static_assert(std::is_trivially_copyable<T>::value, + "ReadIntoObject<T> - T must be trivially copyable"); + Read(&aObject, sizeof(T)); + } + + // Read data as an object and move iterator ahead. + // Note that this creates an default `T` first, and then overwrites it with + // bytes from the buffer. Restricted to trivially-copyable types, which + // support this without Undefined Behavior! + template <typename T> + T ReadObject() { + static_assert(std::is_trivially_copyable<T>::value, + "ReadObject<T> - T must be trivially copyable"); + T object; + ReadIntoObject(object); + return object; + } + + // Read an unsigned LEB128 number and move iterator ahead. + template <typename T> + T ReadULEB128() { + return ::mozilla::ReadULEB128<T>(*this); + } + + private: + // Only a ModuloBuffer can instantiate its iterator. + friend class ModuloBuffer; + + Iterator(ConstOrMutableBuffer& aBuffer, Index aIndex) + : mModuloBuffer(WrapNotNull(&aBuffer)), mIndex(aIndex) {} + + // Convert the Iterator's mIndex into an offset inside the byte buffer. + Offset OffsetInBuffer() const { + return static_cast<Offset>(mIndex) & mModuloBuffer->mMask; + } + + // ModuloBuffer that this Iterator operates on. + // Using a non-null pointer instead of a reference, to allow re-assignment + // of an Iterator variable. + NotNull<ConstOrMutableBuffer*> mModuloBuffer; + + // Position of this iterator in the wider `Index` range. (Will be wrapped + // around as needed when actually accessing bytes from the buffer.) + Index mIndex; + }; + + // Shortcut to iterator to const (read-only) data. + using Reader = Iterator<true>; + // Shortcut to iterator to non-const (read/write) data. + using Writer = Iterator<false>; + + // Create an iterator to const data at the given index. + Reader ReaderAt(Index aIndex) const { return Reader(*this, aIndex); } + + // Create an iterator to non-const data at the given index. + Writer WriterAt(Index aIndex) { return Writer(*this, aIndex); } + +#ifdef DEBUG + void Dump() const { + Length len = BufferLength().Value(); + if (len > 128) { + len = 128; + } + for (Length i = 0; i < len; ++i) { + printf("%02x ", mBuffer[i]); + } + printf("\n"); + } +#endif // DEBUG + + private: + // Mask used to convert an index to an offset in `mBuffer` + const PowerOfTwoMask<Offset> mMask; + + // Buffer data. `const NotNull<...>` shows that `mBuffer is `const`, and + // `Byte* const` shows that the pointer cannot be changed to point at + // something else, but the pointed-at `Byte`s are writable. + const NotNull<Byte* const> mBuffer; + + // Function used to release the buffer resource (if needed). + std::function<void(Byte*)> mBufferDeleter; +}; + +} // namespace mozilla + +#endif // ModuloBuffer_h diff --git a/mozglue/baseprofiler/public/PowerOfTwo.h b/mozglue/baseprofiler/public/PowerOfTwo.h new file mode 100644 index 0000000000..7d396c15e6 --- /dev/null +++ b/mozglue/baseprofiler/public/PowerOfTwo.h @@ -0,0 +1,322 @@ +/* -*- 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/. */ + +// PowerOfTwo is a value type that always hold a power of 2. +// It has the same size as their underlying unsigned type, but offer the +// guarantee of being a power of 2, which permits some optimizations when +// involved in modulo operations (using masking instead of actual modulo). +// +// PowerOfTwoMask contains a mask corresponding to a power of 2. +// E.g., 2^8 is 256 or 0x100, the corresponding mask is 2^8-1 or 255 or 0xFF. +// It should be used instead of PowerOfTwo in situations where most operations +// would be modulo, this saves having to recompute the mask from the stored +// power of 2. +// +// One common use would be for ring-buffer containers with a power-of-2 size, +// where an index is usually converted to an in-buffer offset by `i % size`. +// Instead, the container could store a PowerOfTwo or PowerOfTwoMask, and do +// `i % p2` or `i & p2m`, which is more efficient than for arbitrary sizes. +// +// Shortcuts for common 32- and 64-bit values: PowerOfTwo32, etc. +// +// To create constexpr constants, use MakePowerOfTwo<Type, Value>(), etc. + +#ifndef PowerOfTwo_h +#define PowerOfTwo_h + +#include "mozilla/MathAlgorithms.h" + +#include <limits> + +namespace mozilla { + +// Compute the smallest power of 2 greater than or equal to aInput, except if +// that would overflow in which case the highest possible power of 2 if chosen. +// 0->1, 1->1, 2->2, 3->4, ... 2^31->2^31, 2^31+1->2^31 (for uint32_t), etc. +template <typename T> +T FriendlyRoundUpPow2(T aInput) { + // This is the same code as `RoundUpPow2()`, except we handle any type (that + // CeilingLog2 supports) and allow the greater-than-max-power case. + constexpr T max = T(1) << (sizeof(T) * CHAR_BIT - 1); + if (aInput >= max) { + return max; + } + return T(1) << CeilingLog2(aInput); +} + +namespace detail { +// Same function name `CountLeadingZeroes` with uint32_t and uint64_t overloads. +inline uint_fast8_t CountLeadingZeroes(uint32_t aValue) { + MOZ_ASSERT(aValue != 0); + return detail::CountLeadingZeroes32(aValue); +} +inline uint_fast8_t CountLeadingZeroes(uint64_t aValue) { + MOZ_ASSERT(aValue != 0); + return detail::CountLeadingZeroes64(aValue); +} +// Refuse anything else. +template <typename T> +inline uint_fast8_t CountLeadingZeroes(T aValue) = delete; +} // namespace detail + +// Compute the smallest 2^N-1 mask where aInput can fit. +// I.e., `x & mask == x`, but `x & (mask >> 1) != x`. +// Or looking at binary, we want a mask with as many leading zeroes as the +// input, by right-shifting a full mask: (8-bit examples) +// input: 00000000 00000001 00000010 00010110 01111111 10000000 +// N leading 0s: ^^^^^^^^ 8 ^^^^^^^ 7 ^^^^^^ 6 ^^^ 3 ^ 1 0 +// full mask: 11111111 11111111 11111111 11111111 11111111 11111111 +// full mask >> N: 00000000 00000001 00000011 00011111 01111111 11111111 +template <typename T> +T RoundUpPow2Mask(T aInput) { + // Special case, as CountLeadingZeroes(0) is undefined. (And even if that was + // defined, shifting by the full type size is also undefined!) + if (aInput == 0) { + return 0; + } + return T(-1) >> detail::CountLeadingZeroes(aInput); +} + +template <typename T> +class PowerOfTwoMask; + +template <typename T, T Mask> +constexpr PowerOfTwoMask<T> MakePowerOfTwoMask(); + +template <typename T> +class PowerOfTwo; + +template <typename T, T Value> +constexpr PowerOfTwo<T> MakePowerOfTwo(); + +// PowerOfTwoMask will always contain a mask for a power of 2, which is useful +// for power-of-2 modulo operations (e.g., to keep an index inside a power-of-2 +// container). +// Use this instead of PowerOfTwo if masking is the primary use of the value. +// +// Note that this class can store a "full" mask where all bits are set, so it +// works for mask corresponding to the power of 2 that would overflow `T` +// (e.g., 2^32 for uint32_t gives a mask of 2^32-1, which fits in a uint32_t). +// For this reason there is no API that computes the power of 2 corresponding to +// the mask; But this can be done explicitly with `MaskValue() + 1`, which may +// be useful for computing things like distance-to-the-end by doing +// `MaskValue() + 1 - offset`, which works fine with unsigned number types. +template <typename T> +class PowerOfTwoMask { + static_assert(!std::numeric_limits<T>::is_signed, + "PowerOfTwoMask must use an unsigned type"); + + public: + // Construct a power of 2 mask where the given value can fit. + // Cannot be constexpr because of `RoundUpPow2Mask()`. + explicit PowerOfTwoMask(T aInput) : mMask(RoundUpPow2Mask(aInput)) {} + + // Compute the mask corresponding to a PowerOfTwo. + // This saves having to compute the nearest 2^N-1. + // Not a conversion constructor, as that could be ambiguous whether we'd want + // the mask corresponding to the power of 2 (2^N -> 2^N-1), or the mask that + // can *contain* the PowerOfTwo value (2^N -> 2^(N+1)-1). + // Note: Not offering reverse PowerOfTwoMark-to-PowerOfTwo conversion, because + // that could result in an unexpected 0 result for the largest possible mask. + template <typename U> + static constexpr PowerOfTwoMask<U> MaskForPowerOfTwo( + const PowerOfTwo<U>& aP2) { + return PowerOfTwoMask(aP2); + } + + // Allow smaller unsigned types as input. + // Bigger or signed types must be explicitly converted by the caller. + template <typename U> + explicit constexpr PowerOfTwoMask(U aInput) + : mMask(RoundUpPow2Mask(static_cast<T>(aInput))) { + static_assert(!std::numeric_limits<T>::is_signed, + "PowerOfTwoMask does not accept signed types"); + static_assert(sizeof(U) <= sizeof(T), + "PowerOfTwoMask does not accept bigger types"); + } + + constexpr T MaskValue() const { return mMask; } + + // `x & aPowerOfTwoMask` just works. + template <typename U> + friend U operator&(U aNumber, PowerOfTwoMask aP2M) { + return static_cast<U>(aNumber & aP2M.MaskValue()); + } + + // `aPowerOfTwoMask & x` just works. + template <typename U> + friend constexpr U operator&(PowerOfTwoMask aP2M, U aNumber) { + return static_cast<U>(aP2M.MaskValue() & aNumber); + } + + // `x % aPowerOfTwoMask(2^N-1)` is equivalent to `x % 2^N` but is more + // optimal by doing `x & (2^N-1)`. + // Useful for templated code doing modulo with a template argument type. + template <typename U> + friend constexpr U operator%(U aNumerator, PowerOfTwoMask aDenominator) { + return aNumerator & aDenominator.MaskValue(); + } + + constexpr bool operator==(const PowerOfTwoMask& aRhs) const { + return mMask == aRhs.mMask; + } + constexpr bool operator!=(const PowerOfTwoMask& aRhs) const { + return mMask != aRhs.mMask; + } + + private: + // Trust `PowerOfTwo` to call the private Trusted constructor below. + friend class PowerOfTwo<T>; + + // Trust `MakePowerOfTwoMask()` to call the private Trusted constructor below. + template <typename U, U Mask> + friend constexpr PowerOfTwoMask<U> MakePowerOfTwoMask(); + + struct Trusted { + T mMask; + }; + // Construct the mask corresponding to a PowerOfTwo. + // This saves having to compute the nearest 2^N-1. + // Note: Not a public PowerOfTwo->PowerOfTwoMask conversion constructor, as + // that could be ambiguous whether we'd want the mask corresponding to the + // power of 2 (2^N -> 2^N-1), or the mask that can *contain* the PowerOfTwo + // value (2^N -> 2^(N+1)-1). + explicit constexpr PowerOfTwoMask(const Trusted& aP2) : mMask(aP2.mMask) {} + + T mMask = 0; +}; + +// Make a PowerOfTwoMask constant, statically-checked. +template <typename T, T Mask> +constexpr PowerOfTwoMask<T> MakePowerOfTwoMask() { + static_assert(Mask == T(-1) || IsPowerOfTwo(Mask + 1), + "MakePowerOfTwoMask<T, Mask>: Mask must be 2^N-1"); + using Trusted = typename PowerOfTwoMask<T>::Trusted; + return PowerOfTwoMask<T>(Trusted{Mask}); +} + +// PowerOfTwo will always contain a power of 2. +template <typename T> +class PowerOfTwo { + static_assert(!std::numeric_limits<T>::is_signed, + "PowerOfTwo must use an unsigned type"); + + public: + // Construct a power of 2 that can fit the given value, or the highest power + // of 2 possible. + // Caller should explicitly check/assert `Value() <= aInput` if they want to. + // Cannot be constexpr because of `FriendlyRoundUpPow2()`. + explicit PowerOfTwo(T aInput) : mValue(FriendlyRoundUpPow2(aInput)) {} + + // Allow smaller unsigned types as input. + // Bigger or signed types must be explicitly converted by the caller. + template <typename U> + explicit PowerOfTwo(U aInput) + : mValue(FriendlyRoundUpPow2(static_cast<T>(aInput))) { + static_assert(!std::numeric_limits<T>::is_signed, + "PowerOfTwo does not accept signed types"); + static_assert(sizeof(U) <= sizeof(T), + "PowerOfTwo does not accept bigger types"); + } + + constexpr T Value() const { return mValue; } + + // Binary mask corresponding to the power of 2, useful for modulo. + // E.g., `x & powerOfTwo(y).Mask()` == `x % powerOfTwo(y)`. + // Consider PowerOfTwoMask class instead of PowerOfTwo if masking is the + // primary use case. + constexpr T MaskValue() const { return mValue - 1; } + + // PowerOfTwoMask corresponding to this power of 2, useful for modulo. + constexpr PowerOfTwoMask<T> Mask() const { + using Trusted = typename PowerOfTwoMask<T>::Trusted; + return PowerOfTwoMask<T>(Trusted{MaskValue()}); + } + + // `x % aPowerOfTwo` works optimally. + // Useful for templated code doing modulo with a template argument type. + // Use PowerOfTwoMask class instead if masking is the primary use case. + template <typename U> + friend constexpr U operator%(U aNumerator, PowerOfTwo aDenominator) { + return aNumerator & aDenominator.MaskValue(); + } + + constexpr bool operator==(const PowerOfTwo& aRhs) const { + return mValue == aRhs.mValue; + } + constexpr bool operator!=(const PowerOfTwo& aRhs) const { + return mValue != aRhs.mValue; + } + constexpr bool operator<(const PowerOfTwo& aRhs) const { + return mValue < aRhs.mValue; + } + constexpr bool operator<=(const PowerOfTwo& aRhs) const { + return mValue <= aRhs.mValue; + } + constexpr bool operator>(const PowerOfTwo& aRhs) const { + return mValue > aRhs.mValue; + } + constexpr bool operator>=(const PowerOfTwo& aRhs) const { + return mValue >= aRhs.mValue; + } + + private: + // Trust `MakePowerOfTwo()` to call the private Trusted constructor below. + template <typename U, U Value> + friend constexpr PowerOfTwo<U> MakePowerOfTwo(); + + struct Trusted { + T mValue; + }; + // Construct a PowerOfTwo with the given trusted value. + // This saves having to compute the nearest 2^N. + // Note: Not offering PowerOfTwoMark-to-PowerOfTwo conversion, because that + // could result in an unexpected 0 result for the largest possible mask. + explicit constexpr PowerOfTwo(const Trusted& aP2) : mValue(aP2.mValue) {} + + // The smallest power of 2 is 2^0 == 1. + T mValue = 1; +}; + +// Make a PowerOfTwo constant, statically-checked. +template <typename T, T Value> +constexpr PowerOfTwo<T> MakePowerOfTwo() { + static_assert(IsPowerOfTwo(Value), + "MakePowerOfTwo<T, Value>: Value must be 2^N"); + using Trusted = typename PowerOfTwo<T>::Trusted; + return PowerOfTwo<T>(Trusted{Value}); +} + +// Shortcuts for the most common types and functions. + +using PowerOfTwoMask32 = PowerOfTwoMask<uint32_t>; +using PowerOfTwo32 = PowerOfTwo<uint32_t>; +using PowerOfTwoMask64 = PowerOfTwoMask<uint64_t>; +using PowerOfTwo64 = PowerOfTwo<uint64_t>; + +template <uint32_t Mask> +constexpr PowerOfTwoMask32 MakePowerOfTwoMask32() { + return MakePowerOfTwoMask<uint32_t, Mask>(); +} + +template <uint32_t Value> +constexpr PowerOfTwo32 MakePowerOfTwo32() { + return MakePowerOfTwo<uint32_t, Value>(); +} + +template <uint64_t Mask> +constexpr PowerOfTwoMask64 MakePowerOfTwoMask64() { + return MakePowerOfTwoMask<uint64_t, Mask>(); +} + +template <uint64_t Value> +constexpr PowerOfTwo64 MakePowerOfTwo64() { + return MakePowerOfTwo<uint64_t, Value>(); +} + +} // namespace mozilla + +#endif // PowerOfTwo_h diff --git a/mozglue/baseprofiler/public/ProfileBufferChunk.h b/mozglue/baseprofiler/public/ProfileBufferChunk.h new file mode 100644 index 0000000000..24a516bcaf --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferChunk.h @@ -0,0 +1,543 @@ +/* -*- 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 ProfileBufferChunk_h +#define ProfileBufferChunk_h + +#include "mozilla/MemoryReporting.h" +#include "mozilla/ProfileBufferIndex.h" +#include "mozilla/Span.h" +#include "mozilla/TimeStamp.h" +#include "mozilla/UniquePtr.h" + +#if defined(MOZ_MEMORY) +# include "mozmemory.h" +#endif + +#include <algorithm> +#include <limits> +#include <type_traits> + +#ifdef DEBUG +# include <cstdio> +#endif + +namespace mozilla { + +// Represents a single chunk of memory, with a link to the next chunk (or null). +// +// A chunk is made of an internal header (which contains a public part) followed +// by user-accessible bytes. +// +// +---------------+---------+----------------------------------------------+ +// | public Header | private | memory containing user blocks | +// +---------------+---------+----------------------------------------------+ +// <---------------BufferBytes()------------------> +// <------------------------------ChunkBytes()------------------------------> +// +// The chunk can reserve "blocks", but doesn't know the internal contents of +// each block, it only knows where the first one starts, and where the last one +// ends (which is where the next one will begin, if not already out of range). +// It is up to the user to add structure to each block so that they can be +// distinguished when later read. +// +// +---------------+---------+----------------------------------------------+ +// | public Header | private | [1st block]...[last full block] | +// +---------------+---------+----------------------------------------------+ +// ChunkHeader().mOffsetFirstBlock ^ ^ +// ChunkHeader().mOffsetPastLastBlock --' +// +// It is possible to attempt to reserve more than the remaining space, in which +// case only what is available is returned. The caller is responsible for using +// another chunk, reserving a block "tail" in it, and using both parts to +// constitute a full block. (This initial tail may be empty in some chunks.) +// +// +---------------+---------+----------------------------------------------+ +// | public Header | private | tail][1st block]...[last full block][head... | +// +---------------+---------+----------------------------------------------+ +// ChunkHeader().mOffsetFirstBlock ^ ^ +// ChunkHeader().mOffsetPastLastBlock --' +// +// Each Chunk has an internal state (checked in DEBUG builds) that directs how +// to use it during creation, initialization, use, end of life, recycling, and +// destruction. See `State` below for details. +// In particular: +// - `ReserveInitialBlockAsTail()` must be called before the first `Reserve()` +// after construction or recycling, even with a size of 0 (no actual tail), +// - `MarkDone()` and `MarkRecycled()` must be called as appropriate. +class ProfileBufferChunk { + public: + using Byte = uint8_t; + using Length = uint32_t; + + using SpanOfBytes = Span<Byte>; + + // Hint about the size of the metadata (public and private headers). + // `Create()` below takes the minimum *buffer* size, so the minimum total + // Chunk size is at least `SizeofChunkMetadata() + aMinBufferBytes`. + [[nodiscard]] static constexpr Length SizeofChunkMetadata() { + return static_cast<Length>(sizeof(InternalHeader)); + } + + // Allocate space for a chunk with a given minimum size, and construct it. + // The actual size may be higher, to match the actual space taken in the + // memory pool. + [[nodiscard]] static UniquePtr<ProfileBufferChunk> Create( + Length aMinBufferBytes) { + // We need at least one byte, to cover the always-present `mBuffer` byte. + aMinBufferBytes = std::max(aMinBufferBytes, Length(1)); + // Trivial struct with the same alignment as `ProfileBufferChunk`, and size + // equal to that alignment, because typically the sizeof of an object is + // a multiple of its alignment. + struct alignas(alignof(InternalHeader)) ChunkStruct { + Byte c[alignof(InternalHeader)]; + }; + static_assert(std::is_trivial_v<ChunkStruct>, + "ChunkStruct must be trivial to avoid any construction"); + // Allocate an array of that struct, enough to contain the expected + // `ProfileBufferChunk` (with its header+buffer). + size_t count = (sizeof(InternalHeader) + aMinBufferBytes + + (alignof(InternalHeader) - 1)) / + alignof(InternalHeader); +#if defined(MOZ_MEMORY) + // Potentially expand the array to use more of the effective allocation. + count = (malloc_good_size(count * sizeof(ChunkStruct)) + + (sizeof(ChunkStruct) - 1)) / + sizeof(ChunkStruct); +#endif + auto chunkStorage = MakeUnique<ChunkStruct[]>(count); + MOZ_ASSERT(reinterpret_cast<uintptr_t>(chunkStorage.get()) % + alignof(InternalHeader) == + 0); + // After the allocation, compute the actual chunk size (including header). + const size_t chunkBytes = count * sizeof(ChunkStruct); + MOZ_ASSERT(chunkBytes >= sizeof(ProfileBufferChunk), + "Not enough space to construct a ProfileBufferChunk"); + MOZ_ASSERT(chunkBytes <= + static_cast<size_t>(std::numeric_limits<Length>::max())); + // Compute the size of the user-accessible buffer inside the chunk. + const Length bufferBytes = + static_cast<Length>(chunkBytes - sizeof(InternalHeader)); + MOZ_ASSERT(bufferBytes >= aMinBufferBytes, + "Not enough space for minimum buffer size"); + // Construct the header at the beginning of the allocated array, with the + // known buffer size. + new (chunkStorage.get()) ProfileBufferChunk(bufferBytes); + // We now have a proper `ProfileBufferChunk` object, create the appropriate + // UniquePtr for it. + UniquePtr<ProfileBufferChunk> chunk{ + reinterpret_cast<ProfileBufferChunk*>(chunkStorage.release())}; + MOZ_ASSERT( + size_t(reinterpret_cast<const char*>( + &chunk.get()->BufferSpan()[bufferBytes - 1]) - + reinterpret_cast<const char*>(chunk.get())) == chunkBytes - 1, + "Buffer span spills out of chunk allocation"); + return chunk; + } + +#ifdef DEBUG + ~ProfileBufferChunk() { + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::InUse); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Full); + MOZ_ASSERT(mInternalHeader.mState == InternalHeader::State::Created || + mInternalHeader.mState == InternalHeader::State::Done || + mInternalHeader.mState == InternalHeader::State::Recycled); + } +#endif + + // Must be called with the first block tail (may be empty), which will be + // skipped if the reader starts with this ProfileBufferChunk. + [[nodiscard]] SpanOfBytes ReserveInitialBlockAsTail(Length aTailSize) { +#ifdef DEBUG + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::InUse); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Full); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Done); + MOZ_ASSERT(mInternalHeader.mState == InternalHeader::State::Created || + mInternalHeader.mState == InternalHeader::State::Recycled); + mInternalHeader.mState = InternalHeader::State::InUse; +#endif + mInternalHeader.mHeader.mOffsetFirstBlock = aTailSize; + mInternalHeader.mHeader.mOffsetPastLastBlock = aTailSize; + return SpanOfBytes(&mBuffer, aTailSize); + } + + struct ReserveReturn { + SpanOfBytes mSpan; + ProfileBufferBlockIndex mBlockRangeIndex; + }; + + // Reserve a block of up to `aBlockSize` bytes, and return a Span to it, and + // its starting index. The actual size may be smaller, if the block cannot fit + // in the remaining space. + [[nodiscard]] ReserveReturn ReserveBlock(Length aBlockSize) { + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Created); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Full); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Done); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Recycled); + MOZ_ASSERT(mInternalHeader.mState == InternalHeader::State::InUse); + MOZ_ASSERT(RangeStart() != 0, + "Expected valid range start before first Reserve()"); + const Length blockOffset = mInternalHeader.mHeader.mOffsetPastLastBlock; + Length reservedSize = aBlockSize; + if (MOZ_UNLIKELY(aBlockSize >= RemainingBytes())) { + reservedSize = RemainingBytes(); +#ifdef DEBUG + mInternalHeader.mState = InternalHeader::State::Full; +#endif + } + mInternalHeader.mHeader.mOffsetPastLastBlock += reservedSize; + mInternalHeader.mHeader.mBlockCount += 1; + return {SpanOfBytes(&mBuffer + blockOffset, reservedSize), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mInternalHeader.mHeader.mRangeStart + blockOffset)}; + } + + // When a chunk will not be used to store more blocks (because it is full, or + // because the profiler will not add more data), it should be marked "done". + // Access to its content is still allowed. + void MarkDone() { +#ifdef DEBUG + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Created); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Done); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Recycled); + MOZ_ASSERT(mInternalHeader.mState == InternalHeader::State::InUse || + mInternalHeader.mState == InternalHeader::State::Full); + mInternalHeader.mState = InternalHeader::State::Done; +#endif + mInternalHeader.mHeader.mDoneTimeStamp = TimeStamp::NowUnfuzzed(); + } + + // A "Done" chunk may be recycled, to avoid allocating a new one. + void MarkRecycled() { +#ifdef DEBUG + // We also allow Created and already-Recycled chunks to be recycled, this + // way it's easier to recycle chunks when their state is not easily + // trackable. + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::InUse); + MOZ_ASSERT(mInternalHeader.mState != InternalHeader::State::Full); + MOZ_ASSERT(mInternalHeader.mState == InternalHeader::State::Created || + mInternalHeader.mState == InternalHeader::State::Done || + mInternalHeader.mState == InternalHeader::State::Recycled); + mInternalHeader.mState = InternalHeader::State::Recycled; +#endif + // Reset all header fields, in case this recycled chunk gets read. + mInternalHeader.mHeader.Reset(); + } + + // Public header, meant to uniquely identify a chunk, it may be shared with + // other processes to coordinate global memory handling. + struct Header { + explicit Header(Length aBufferBytes) : mBufferBytes(aBufferBytes) {} + + // Reset all members to their as-new values (apart from the buffer size, + // which cannot change), ready for re-use. + void Reset() { + mOffsetFirstBlock = 0; + mOffsetPastLastBlock = 0; + mDoneTimeStamp = TimeStamp{}; + mBlockCount = 0; + mRangeStart = 0; + mProcessId = 0; + } + + // Note: Part of the ordering of members below is to avoid unnecessary + // padding. + + // Members managed by the ProfileBufferChunk. + + // Offset of the first block (past the initial tail block, which may be 0). + Length mOffsetFirstBlock = 0; + // Offset past the last byte of the last reserved block + // It may be past mBufferBytes when last block continues in the next + // ProfileBufferChunk. It may be before mBufferBytes if ProfileBufferChunk + // is marked "Done" before the end is reached. + Length mOffsetPastLastBlock = 0; + // Timestamp when buffer is "Done" (which happens when the last block is + // written). This will be used to find and discard the oldest + // ProfileBufferChunk. + TimeStamp mDoneTimeStamp; + // Number of bytes in the buffer, set once at construction time. + const Length mBufferBytes; + // Number of reserved blocks (including final one even if partial, but + // excluding initial tail). + Length mBlockCount = 0; + + // Meta-data set by the user. + + // Index of the first byte of this ProfileBufferChunk, relative to all + // Chunks for this process. Index 0 is reserved as nullptr-like index, + // mRangeStart should be set to a non-0 value before the first `Reserve()`. + ProfileBufferIndex mRangeStart = 0; + // Process writing to this ProfileBufferChunk. + int mProcessId = 0; + + // A bit of spare space (necessary here because of the alignment due to + // other members), may be later repurposed for extra data. + const int mPADDING = 0; + }; + + [[nodiscard]] const Header& ChunkHeader() const { + return mInternalHeader.mHeader; + } + + [[nodiscard]] Length BufferBytes() const { + return ChunkHeader().mBufferBytes; + } + + // Total size of the chunk (buffer + header). + [[nodiscard]] Length ChunkBytes() const { + return static_cast<Length>(sizeof(InternalHeader)) + BufferBytes(); + } + + // Size of external resources, in this case all the following chunks. + [[nodiscard]] size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { + const ProfileBufferChunk* const next = GetNext(); + return next ? next->SizeOfIncludingThis(aMallocSizeOf) : 0; + } + + // Size of this chunk and all following ones. + [[nodiscard]] size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + // Just in case `aMallocSizeOf` falls back on just `sizeof`, make sure we + // account for at least the actual Chunk requested allocation size. + return std::max<size_t>(aMallocSizeOf(this), ChunkBytes()) + + SizeOfExcludingThis(aMallocSizeOf); + } + + [[nodiscard]] Length RemainingBytes() const { + return BufferBytes() - OffsetPastLastBlock(); + } + + [[nodiscard]] Length OffsetFirstBlock() const { + return ChunkHeader().mOffsetFirstBlock; + } + + [[nodiscard]] Length OffsetPastLastBlock() const { + return ChunkHeader().mOffsetPastLastBlock; + } + + [[nodiscard]] Length BlockCount() const { return ChunkHeader().mBlockCount; } + + [[nodiscard]] int ProcessId() const { return ChunkHeader().mProcessId; } + + void SetProcessId(int aProcessId) { + mInternalHeader.mHeader.mProcessId = aProcessId; + } + + // Global range index at the start of this Chunk. + [[nodiscard]] ProfileBufferIndex RangeStart() const { + return ChunkHeader().mRangeStart; + } + + void SetRangeStart(ProfileBufferIndex aRangeStart) { + mInternalHeader.mHeader.mRangeStart = aRangeStart; + } + + // Get a read-only Span to the buffer. It is up to the caller to decypher the + // contents, based on known offsets and the internal block structure. + [[nodiscard]] Span<const Byte> BufferSpan() const { + return Span<const Byte>(&mBuffer, BufferBytes()); + } + + [[nodiscard]] Byte ByteAt(Length aOffset) const { + MOZ_ASSERT(aOffset < OffsetPastLastBlock()); + return *(&mBuffer + aOffset); + } + + [[nodiscard]] ProfileBufferChunk* GetNext() { + return mInternalHeader.mNext.get(); + } + [[nodiscard]] const ProfileBufferChunk* GetNext() const { + return mInternalHeader.mNext.get(); + } + + [[nodiscard]] UniquePtr<ProfileBufferChunk> ReleaseNext() { + return std::move(mInternalHeader.mNext); + } + + void InsertNext(UniquePtr<ProfileBufferChunk>&& aChunk) { + if (!aChunk) { + return; + } + aChunk->SetLast(ReleaseNext()); + mInternalHeader.mNext = std::move(aChunk); + } + + // Find the last chunk in this chain (it may be `this`). + [[nodiscard]] ProfileBufferChunk* Last() { + ProfileBufferChunk* chunk = this; + for (;;) { + ProfileBufferChunk* next = chunk->GetNext(); + if (!next) { + return chunk; + } + chunk = next; + } + } + [[nodiscard]] const ProfileBufferChunk* Last() const { + const ProfileBufferChunk* chunk = this; + for (;;) { + const ProfileBufferChunk* next = chunk->GetNext(); + if (!next) { + return chunk; + } + chunk = next; + } + } + + void SetLast(UniquePtr<ProfileBufferChunk>&& aChunk) { + if (!aChunk) { + return; + } + Last()->mInternalHeader.mNext = std::move(aChunk); + } + + // Join two possibly-null chunk lists. + [[nodiscard]] static UniquePtr<ProfileBufferChunk> Join( + UniquePtr<ProfileBufferChunk>&& aFirst, + UniquePtr<ProfileBufferChunk>&& aLast) { + if (aFirst) { + aFirst->SetLast(std::move(aLast)); + return std::move(aFirst); + } + return std::move(aLast); + } + +#ifdef DEBUG + void Dump(std::FILE* aFile = stdout) const { + fprintf(aFile, + "Chunk[%p] chunkSize=%u bufferSize=%u state=%s rangeStart=%u " + "firstBlockOffset=%u offsetPastLastBlock=%u blockCount=%u", + this, unsigned(ChunkBytes()), unsigned(BufferBytes()), + mInternalHeader.StateString(), unsigned(RangeStart()), + unsigned(OffsetFirstBlock()), unsigned(OffsetPastLastBlock()), + unsigned(BlockCount())); + const auto len = OffsetPastLastBlock(); + constexpr unsigned columns = 16; + unsigned char ascii[columns + 1]; + ascii[columns] = '\0'; + for (Length i = 0; i < len; ++i) { + if (i % columns == 0) { + fprintf(aFile, "\n %4u=0x%03x:", unsigned(i), unsigned(i)); + for (unsigned a = 0; a < columns; ++a) { + ascii[a] = ' '; + } + } + unsigned char sep = ' '; + if (i == OffsetFirstBlock()) { + if (i == OffsetPastLastBlock()) { + sep = '#'; + } else { + sep = '['; + } + } else if (i == OffsetPastLastBlock()) { + sep = ']'; + } + unsigned char c = *(&mBuffer + i); + fprintf(aFile, "%c%02x", sep, c); + + if (i == len - 1) { + if (i + 1 == OffsetPastLastBlock()) { + // Special case when last block ends right at the end. + fprintf(aFile, "]"); + } else { + fprintf(aFile, " "); + } + } else if (i % columns == columns - 1) { + fprintf(aFile, " "); + } + + ascii[i % columns] = (c >= ' ' && c <= '~') ? c : '.'; + + if (i % columns == columns - 1) { + fprintf(aFile, " %s", ascii); + } + } + + if (len % columns < columns - 1) { + for (Length i = len % columns; i < columns; ++i) { + fprintf(aFile, " "); + } + fprintf(aFile, " %s", ascii); + } + + fprintf(aFile, "\n"); + } +#endif // DEBUG + + private: + // ProfileBufferChunk constructor. Use static `Create()` to allocate and + // construct a ProfileBufferChunk. + explicit ProfileBufferChunk(Length aBufferBytes) + : mInternalHeader(aBufferBytes) {} + + // This internal header starts with the public `Header`, and adds some data + // only necessary for local handling. + // This encapsulation is also necessary to perform placement-new in + // `Create()`. + struct InternalHeader { + explicit InternalHeader(Length aBufferBytes) : mHeader(aBufferBytes) {} + + Header mHeader; + UniquePtr<ProfileBufferChunk> mNext; + +#ifdef DEBUG + enum class State { + Created, // Self-set. Just constructed, waiting for initial block tail. + InUse, // Ready to accept blocks. + Full, // Self-set. Blocks reach the end (or further). + Done, // Blocks won't be added anymore. + Recycled // Still full of data, but expecting an initial block tail. + }; + + State mState = State::Created; + // Transition table: (X=unexpected) + // Method \ State Created InUse Full Done Recycled + // ReserveInitialBlockAsTail InUse X X X InUse + // Reserve X InUse/Full X X X + // MarkDone X Done Done X X + // MarkRecycled X X X Recycled X + // destructor ok X X ok ok + + const char* StateString() const { + switch (mState) { + case State::Created: + return "Created"; + case State::InUse: + return "InUse"; + case State::Full: + return "Full"; + case State::Done: + return "Done"; + case State::Recycled: + return "Recycled"; + default: + return "?"; + } + } +#else // DEBUG + const char* StateString() const { return "(non-DEBUG)"; } +#endif + }; + + InternalHeader mInternalHeader; + + // KEEP THIS LAST! + // First byte of the buffer. Note that ProfileBufferChunk::Create allocates a + // bigger block, such that `mBuffer` is the first of `mBufferBytes` available + // bytes. + // The initialization is not strictly needed, because bytes should only be + // read after they have been written and `mOffsetPastLastBlock` has been + // updated. However: + // - Reviewbot complains that it's not initialized. + // - It's cheap to initialize one byte. + // - In the worst case (reading does happen), zero is not a valid entry size + // and should get caught in entry readers. + Byte mBuffer = '\0'; +}; + +} // namespace mozilla + +#endif // ProfileBufferChunk_h diff --git a/mozglue/baseprofiler/public/ProfileBufferChunkManager.h b/mozglue/baseprofiler/public/ProfileBufferChunkManager.h new file mode 100644 index 0000000000..e7f12bf21f --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferChunkManager.h @@ -0,0 +1,134 @@ +/* -*- 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 ProfileBufferChunkManager_h +#define ProfileBufferChunkManager_h + +#include "mozilla/ProfileBufferChunk.h" +#include "mozilla/ScopeExit.h" + +#include <functional> + +namespace mozilla { + +// Manages the ProfileBufferChunks for this process. +// The main user of this class is the buffer that needs chunks to store its +// data. +// The main ProfileBufferChunks responsibilities are: +// - It can create new chunks, they are called "unreleased". +// - Later these chunks are returned here, and become "released". +// - The manager is free to destroy or recycle the oldest released chunks +// (usually to reclaim memory), and will inform the user through a provided +// callback. +// - The user may access still-alive released chunks. +class ProfileBufferChunkManager { + public: + virtual ~ProfileBufferChunkManager() +#ifdef DEBUG + { + MOZ_ASSERT(!mUser, "Still registered when being destroyed"); + } +#else + = default; +#endif + + // Expected maximum size needed to store one stack sample. + // Most ChunkManager sub-classes will require chunk sizes, this can serve as + // a minimum recommendation to hold most backtraces. + constexpr static ProfileBufferChunk::Length scExpectedMaximumStackSize = + 128 * 1024; + + // Estimated maximum buffer size. + [[nodiscard]] virtual size_t MaxTotalSize() const = 0; + + // Create or recycle a chunk right now. May return null in case of allocation + // failure. + // Note that the chunk-destroyed callback may be invoked during this call; + // user should be careful with reentrancy issues. + [[nodiscard]] virtual UniquePtr<ProfileBufferChunk> GetChunk() = 0; + + // `aChunkReceiver` may be called with a new or recycled chunk, or nullptr. + // (See `FulfillChunkRequests()` regarding when the callback may happen.) + virtual void RequestChunk( + std::function<void(UniquePtr<ProfileBufferChunk>)>&& aChunkReceiver) = 0; + + // This method may be invoked at any time on any thread (and not necessarily + // by the main user of this class), to do the work necessary to respond to a + // previous `RequestChunk()`. + // It is optional: If it is never called, or called too late, the user is + // responsible for directly calling `GetChunk()` when a chunk is really + // needed (or it should at least fail gracefully). + // The idea is to fulfill chunk request on a separate thread, and most + // importantly outside of profiler calls, to avoid doing expensive memory + // allocations during these calls. + virtual void FulfillChunkRequests() = 0; + + // One chunk is released by the user, the ProfileBufferChunkManager should + // keep it as long as possible (depending on local or global memory/time + // limits). Note that the chunk-destroyed callback may be invoked during this + // call; user should be careful with reentrancy issues. + virtual void ReleaseChunk(UniquePtr<ProfileBufferChunk> aChunk) = 0; + + // `aChunkDestroyedCallback` will be called whenever the contents of a + // previously-released chunk is about to be destroyed or recycled. + // Note that it may be called during other functions above, or at other times + // from the same or other threads; user should be careful with reentrancy + // issues. + virtual void SetChunkDestroyedCallback( + std::function<void(const ProfileBufferChunk&)>&& + aChunkDestroyedCallback) = 0; + + // Give away all released chunks that have not yet been destroyed. + [[nodiscard]] virtual UniquePtr<ProfileBufferChunk> + GetExtantReleasedChunks() = 0; + + // Let a callback see all released chunks that have not yet been destroyed, if + // any. Return whatever the callback returns. + template <typename Callback> + [[nodiscard]] auto PeekExtantReleasedChunks(Callback&& aCallback) { + const ProfileBufferChunk* chunks = PeekExtantReleasedChunksAndLock(); + auto unlock = + MakeScopeExit([&]() { UnlockAfterPeekExtantReleasedChunks(); }); + return std::forward<Callback>(aCallback)(chunks); + } + + // Chunks that were still unreleased will never be released. + virtual void ForgetUnreleasedChunks() = 0; + + [[nodiscard]] virtual size_t SizeOfExcludingThis( + MallocSizeOf aMallocSizeOf) const = 0; + [[nodiscard]] virtual size_t SizeOfIncludingThis( + MallocSizeOf aMallocSizeOf) const = 0; + + protected: + // Derived classes to implement `PeekExtantReleasedChunks` through these: + virtual const ProfileBufferChunk* PeekExtantReleasedChunksAndLock() = 0; + virtual void UnlockAfterPeekExtantReleasedChunks() = 0; + +#ifdef DEBUG + public: + // DEBUG checks ensuring that this manager and its users avoid UAFs. + // Derived classes should assert that mUser is not null in their functions. + + void RegisteredWith(const void* aUser) { + MOZ_ASSERT(!mUser); + MOZ_ASSERT(aUser); + mUser = aUser; + } + + void DeregisteredFrom(const void* aUser) { + MOZ_ASSERT(mUser == aUser); + mUser = nullptr; + } + + protected: + const void* mUser = nullptr; +#endif // DEBUG +}; + +} // namespace mozilla + +#endif // ProfileBufferChunkManager_h diff --git a/mozglue/baseprofiler/public/ProfileBufferChunkManagerSingle.h b/mozglue/baseprofiler/public/ProfileBufferChunkManagerSingle.h new file mode 100644 index 0000000000..c91b38cbdb --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferChunkManagerSingle.h @@ -0,0 +1,172 @@ +/* -*- 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 ProfileBufferChunkManagerSingle_h +#define ProfileBufferChunkManagerSingle_h + +#include "mozilla/ProfileBufferChunkManager.h" + +#ifdef DEBUG +# include "mozilla/Atomics.h" +#endif // DEBUG + +namespace mozilla { + +// Manages only one Chunk. +// The first call to `Get`/`RequestChunk()` will retrieve the one chunk, and all +// subsequent calls will return nullptr. That chunk may still be released, but +// it will never be destroyed or recycled. +// Unlike others, this manager may be `Reset()`, to allow another round of +// small-data gathering. +// The main use is with short-lived ProfileChunkedBuffers that collect little +// data that can fit in one chunk, e.g., capturing one stack. +// It is not thread-safe. +class ProfileBufferChunkManagerSingle final : public ProfileBufferChunkManager { + public: + using Length = ProfileBufferChunk::Length; + + // Use a preallocated chunk. (Accepting null to gracefully handle OOM.) + explicit ProfileBufferChunkManagerSingle(UniquePtr<ProfileBufferChunk> aChunk) + : mInitialChunk(std::move(aChunk)), + mBufferBytes(mInitialChunk ? mInitialChunk->BufferBytes() : 0) { + MOZ_ASSERT(!mInitialChunk || !mInitialChunk->GetNext(), + "Expected at most one chunk"); + } + + // ChunkMinBufferBytes: Minimum number of user-available bytes in the Chunk. + // Note that Chunks use a bit more memory for their header. + explicit ProfileBufferChunkManagerSingle(Length aChunkMinBufferBytes) + : mInitialChunk(ProfileBufferChunk::Create(aChunkMinBufferBytes)), + mBufferBytes(mInitialChunk ? mInitialChunk->BufferBytes() : 0) {} + +#ifdef DEBUG + ~ProfileBufferChunkManagerSingle() { MOZ_ASSERT(mVirtuallyLocked == false); } +#endif // DEBUG + + // Reset this manager, using the provided chunk (probably coming from the + // ProfileChunkedBuffer that just used it); if null, fallback on current or + // released chunk. + void Reset(UniquePtr<ProfileBufferChunk> aPossibleChunk) { + if (aPossibleChunk) { + mInitialChunk = std::move(aPossibleChunk); + mReleasedChunk = nullptr; + } else if (!mInitialChunk) { + MOZ_ASSERT(!!mReleasedChunk, "Can't reset properly!"); + mInitialChunk = std::move(mReleasedChunk); + } + + if (mInitialChunk) { + mInitialChunk->MarkRecycled(); + mBufferBytes = mInitialChunk->BufferBytes(); + } else { + mBufferBytes = 0; + } + } + + [[nodiscard]] size_t MaxTotalSize() const final { return mBufferBytes; } + + // One of `GetChunk` and `RequestChunk` will only work the very first time (if + // there's even a chunk). + [[nodiscard]] UniquePtr<ProfileBufferChunk> GetChunk() final { + MOZ_ASSERT(mUser, "Not registered yet"); + return std::move(mInitialChunk); + } + + void RequestChunk(std::function<void(UniquePtr<ProfileBufferChunk>)>&& + aChunkReceiver) final { + MOZ_ASSERT(mUser, "Not registered yet"); + // Simple retrieval. + std::move(aChunkReceiver)(GetChunk()); + } + + void FulfillChunkRequests() final { + // Nothing to do here. + } + + void ReleaseChunk(UniquePtr<ProfileBufferChunk> aChunk) final { + MOZ_ASSERT(mUser, "Not registered yet"); + if (!aChunk) { + return; + } + MOZ_ASSERT(!mReleasedChunk, "Unexpected 2nd released chunk"); + MOZ_ASSERT(!aChunk->GetNext(), "Only expected one released chunk"); + mReleasedChunk = std::move(aChunk); + } + + void SetChunkDestroyedCallback( + std::function<void(const ProfileBufferChunk&)>&& aChunkDestroyedCallback) + final { + MOZ_ASSERT(mUser, "Not registered yet"); + // The chunk-destroyed callback will never actually be called, but we keep + // the callback here in case the caller expects it to live as long as this + // manager. + mChunkDestroyedCallback = std::move(aChunkDestroyedCallback); + } + + [[nodiscard]] UniquePtr<ProfileBufferChunk> GetExtantReleasedChunks() final { + MOZ_ASSERT(mUser, "Not registered yet"); + return std::move(mReleasedChunk); + } + + void ForgetUnreleasedChunks() final { + MOZ_ASSERT(mUser, "Not registered yet"); + } + + [[nodiscard]] size_t SizeOfExcludingThis( + MallocSizeOf aMallocSizeOf) const final { + MOZ_ASSERT(mUser, "Not registered yet"); + size_t size = 0; + if (mInitialChunk) { + size += mInitialChunk->SizeOfIncludingThis(aMallocSizeOf); + } + if (mReleasedChunk) { + size += mReleasedChunk->SizeOfIncludingThis(aMallocSizeOf); + } + // Note: Missing size of std::function external resources (if any). + return size; + } + + [[nodiscard]] size_t SizeOfIncludingThis( + MallocSizeOf aMallocSizeOf) const final { + MOZ_ASSERT(mUser, "Not registered yet"); + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf); + } + + protected: + // This manager is not thread-safe, so there's not actual locking needed. + const ProfileBufferChunk* PeekExtantReleasedChunksAndLock() final { + MOZ_ASSERT(mVirtuallyLocked.compareExchange(false, true)); + MOZ_ASSERT(mUser, "Not registered yet"); + return mReleasedChunk.get(); + } + void UnlockAfterPeekExtantReleasedChunks() final { + MOZ_ASSERT(mVirtuallyLocked.compareExchange(true, false)); + } + + private: + // Initial chunk created with this manager, given away at first Get/Request. + UniquePtr<ProfileBufferChunk> mInitialChunk; + + // Storage for the released chunk (which should probably not happen, as it + // means the chunk is full). + UniquePtr<ProfileBufferChunk> mReleasedChunk; + + // Size of the one chunk we're managing. Stored here, because the chunk may + // be moved out and inaccessible from here. + Length mBufferBytes; + + // The chunk-destroyed callback will never actually be called, but we keep it + // here in case the caller expects it to live as long as this manager. + std::function<void(const ProfileBufferChunk&)> mChunkDestroyedCallback; + +#ifdef DEBUG + mutable Atomic<bool> mVirtuallyLocked{false}; +#endif // DEBUG +}; + +} // namespace mozilla + +#endif // ProfileBufferChunkManagerSingle_h diff --git a/mozglue/baseprofiler/public/ProfileBufferChunkManagerWithLocalLimit.h b/mozglue/baseprofiler/public/ProfileBufferChunkManagerWithLocalLimit.h new file mode 100644 index 0000000000..5b1af6d66c --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferChunkManagerWithLocalLimit.h @@ -0,0 +1,428 @@ +/* -*- 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 ProfileBufferChunkManagerWithLocalLimit_h +#define ProfileBufferChunkManagerWithLocalLimit_h + +#include "BaseProfiler.h" +#include "mozilla/BaseProfilerDetail.h" +#include "mozilla/ProfileBufferChunkManager.h" +#include "mozilla/ProfileBufferControlledChunkManager.h" + +#include <utility> + +namespace mozilla { + +// Manages the Chunks for this process in a thread-safe manner, with a maximum +// size per process. +// +// "Unreleased" chunks are not owned here, only "released" chunks can be +// destroyed or recycled when reaching the memory limit, so it is theoretically +// possible to break that limit, if: +// - The user of this class doesn't release their chunks, AND/OR +// - The limit is too small (e.g., smaller than 2 or 3 chunks, which should be +// the usual number of unreleased chunks in flight). +// In this case, it just means that we will use more memory than allowed, +// potentially risking OOMs. Hopefully this shouldn't happen in real code, +// assuming that the user is doing the right thing and releasing chunks ASAP, +// and that the memory limit is reasonably large. +class ProfileBufferChunkManagerWithLocalLimit final + : public ProfileBufferChunkManager, + public ProfileBufferControlledChunkManager { + public: + using Length = ProfileBufferChunk::Length; + + // MaxTotalBytes: Maximum number of bytes allocated in all local Chunks. + // ChunkMinBufferBytes: Minimum number of user-available bytes in each Chunk. + // Note that Chunks use a bit more memory for their header. + explicit ProfileBufferChunkManagerWithLocalLimit(size_t aMaxTotalBytes, + Length aChunkMinBufferBytes) + : mMaxTotalBytes(aMaxTotalBytes), + mChunkMinBufferBytes(aChunkMinBufferBytes) {} + + ~ProfileBufferChunkManagerWithLocalLimit() { + if (mUpdateCallback) { + // Signal the end of this callback. + std::move(mUpdateCallback)(Update(nullptr)); + } + } + + [[nodiscard]] size_t MaxTotalSize() const final { + // `mMaxTotalBytes` is `const` so there is no need to lock the mutex. + return mMaxTotalBytes; + } + + [[nodiscard]] UniquePtr<ProfileBufferChunk> GetChunk() final { + AUTO_PROFILER_STATS(Local_GetChunk); + + ChunkAndUpdate chunkAndUpdate = [&]() { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + return GetChunk(lock); + }(); + + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback && !chunkAndUpdate.second.IsNotUpdate()) { + mUpdateCallback(std::move(chunkAndUpdate.second)); + } + + return std::move(chunkAndUpdate.first); + } + + void RequestChunk(std::function<void(UniquePtr<ProfileBufferChunk>)>&& + aChunkReceiver) final { + AUTO_PROFILER_STATS(Local_RequestChunk); + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + if (mChunkReceiver) { + // We already have a chunk receiver, meaning a request is pending. + return; + } + // Store the chunk receiver. This indicates that a request is pending, and + // it will be handled in the next `FulfillChunkRequests()` call. + mChunkReceiver = std::move(aChunkReceiver); + } + + void FulfillChunkRequests() final { + AUTO_PROFILER_STATS(Local_FulfillChunkRequests); + std::function<void(UniquePtr<ProfileBufferChunk>)> chunkReceiver; + ChunkAndUpdate chunkAndUpdate = [&]() -> ChunkAndUpdate { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + if (!mChunkReceiver) { + // No receiver means no pending request, we're done. + return {}; + } + // Otherwise there is a request, extract the receiver to call below. + std::swap(chunkReceiver, mChunkReceiver); + MOZ_ASSERT(!mChunkReceiver, "mChunkReceiver should have been emptied"); + // And allocate the requested chunk. This may fail, it's fine, we're + // letting the receiver know about it. + AUTO_PROFILER_STATS(Local_FulfillChunkRequests_GetChunk); + return GetChunk(lock); + }(); + + if (chunkReceiver) { + { + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback && !chunkAndUpdate.second.IsNotUpdate()) { + mUpdateCallback(std::move(chunkAndUpdate.second)); + } + } + + // Invoke callback outside of lock, so that it can use other chunk manager + // functions if needed. + // Note that this means there could be a race, where another request + // happens now and even gets fulfilled before this one is! It should be + // rare, and shouldn't be a problem anyway, the user will still get their + // requested chunks, new/recycled chunks look the same so their order + // doesn't matter. + std::move(chunkReceiver)(std::move(chunkAndUpdate.first)); + } + } + + void ReleaseChunk(UniquePtr<ProfileBufferChunk> aChunk) final { + if (!aChunk) { + return; + } + + MOZ_RELEASE_ASSERT(!aChunk->GetNext(), "ReleaseChunk only accepts 1 chunk"); + MOZ_RELEASE_ASSERT(!aChunk->ChunkHeader().mDoneTimeStamp.IsNull(), + "Released chunk should have a 'Done' timestamp"); + + Update update = [&]() { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + MOZ_ASSERT(mUser, "Not registered yet"); + // Keep a pointer to the first newly-released chunk, so we can use it to + // prepare an update (after `aChunk` is moved-from). + const ProfileBufferChunk* const newlyReleasedChunk = aChunk.get(); + // Transfer the chunk size from the unreleased bucket to the released one. + mUnreleasedBufferBytes -= aChunk->BufferBytes(); + mReleasedBufferBytes += aChunk->BufferBytes(); + if (!mReleasedChunks) { + // No other released chunks at the moment, we're starting the list. + MOZ_ASSERT(mReleasedBufferBytes == aChunk->BufferBytes()); + mReleasedChunks = std::move(aChunk); + } else { + // Insert aChunk in mReleasedChunks to keep done-timestamp order. + const TimeStamp& releasedChunkDoneTimeStamp = + aChunk->ChunkHeader().mDoneTimeStamp; + if (releasedChunkDoneTimeStamp < + mReleasedChunks->ChunkHeader().mDoneTimeStamp) { + // aChunk is the oldest -> Insert at the beginning. + aChunk->SetLast(std::move(mReleasedChunks)); + mReleasedChunks = std::move(aChunk); + } else { + // Go through the already-released chunk list, and insert aChunk + // before the first younger released chunk, or at the end. + ProfileBufferChunk* chunk = mReleasedChunks.get(); + for (;;) { + ProfileBufferChunk* const nextChunk = chunk->GetNext(); + if (!nextChunk || releasedChunkDoneTimeStamp < + nextChunk->ChunkHeader().mDoneTimeStamp) { + // Either we're at the last released chunk, or the next released + // chunk is younger -> Insert right after this released chunk. + chunk->InsertNext(std::move(aChunk)); + break; + } + chunk = nextChunk; + } + } + } + + return Update(mUnreleasedBufferBytes, mReleasedBufferBytes, + mReleasedChunks.get(), newlyReleasedChunk); + }(); + + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback && !update.IsNotUpdate()) { + mUpdateCallback(std::move(update)); + } + } + + void SetChunkDestroyedCallback( + std::function<void(const ProfileBufferChunk&)>&& aChunkDestroyedCallback) + final { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + MOZ_ASSERT(mUser, "Not registered yet"); + mChunkDestroyedCallback = std::move(aChunkDestroyedCallback); + } + + [[nodiscard]] UniquePtr<ProfileBufferChunk> GetExtantReleasedChunks() final { + UniquePtr<ProfileBufferChunk> chunks; + size_t unreleasedBufferBytes = [&]() { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + MOZ_ASSERT(mUser, "Not registered yet"); + mReleasedBufferBytes = 0; + chunks = std::move(mReleasedChunks); + return mUnreleasedBufferBytes; + }(); + + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback) { + mUpdateCallback(Update(unreleasedBufferBytes, 0, nullptr, nullptr)); + } + + return chunks; + } + + void ForgetUnreleasedChunks() final { + Update update = [&]() { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + MOZ_ASSERT(mUser, "Not registered yet"); + mUnreleasedBufferBytes = 0; + return Update(0, mReleasedBufferBytes, mReleasedChunks.get(), nullptr); + }(); + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback) { + mUpdateCallback(std::move(update)); + } + } + + [[nodiscard]] size_t SizeOfExcludingThis( + MallocSizeOf aMallocSizeOf) const final { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + return SizeOfExcludingThis(aMallocSizeOf, lock); + } + + [[nodiscard]] size_t SizeOfIncludingThis( + MallocSizeOf aMallocSizeOf) const final { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + MOZ_ASSERT(mUser, "Not registered yet"); + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf, lock); + } + + void SetUpdateCallback(UpdateCallback&& aUpdateCallback) final { + { + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + if (mUpdateCallback) { + // Signal the end of the previous callback. + std::move(mUpdateCallback)(Update(nullptr)); + mUpdateCallback = nullptr; + } + } + + if (aUpdateCallback) { + Update initialUpdate = [&]() { + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + return Update(mUnreleasedBufferBytes, mReleasedBufferBytes, + mReleasedChunks.get(), nullptr); + }(); + + baseprofiler::detail::BaseProfilerAutoLock lock(mUpdateCallbackMutex); + MOZ_ASSERT(!mUpdateCallback, "Only one update callback allowed"); + mUpdateCallback = std::move(aUpdateCallback); + mUpdateCallback(std::move(initialUpdate)); + } + } + + void DestroyChunksAtOrBefore(TimeStamp aDoneTimeStamp) final { + MOZ_ASSERT(!aDoneTimeStamp.IsNull()); + baseprofiler::detail::BaseProfilerAutoLock lock(mMutex); + for (;;) { + if (!mReleasedChunks) { + // We don't own any released chunks (anymore), we're done. + break; + } + if (mReleasedChunks->ChunkHeader().mDoneTimeStamp > aDoneTimeStamp) { + // The current chunk is strictly after the given timestamp, we're done. + break; + } + // We've found a chunk at or before the timestamp, discard it. + DiscardOldestReleasedChunk(lock); + } + } + + protected: + const ProfileBufferChunk* PeekExtantReleasedChunksAndLock() final { + mMutex.Lock(); + MOZ_ASSERT(mUser, "Not registered yet"); + return mReleasedChunks.get(); + } + void UnlockAfterPeekExtantReleasedChunks() final { mMutex.Unlock(); } + + private: + void MaybeRecycleChunk( + UniquePtr<ProfileBufferChunk>&& chunk, + const baseprofiler::detail::BaseProfilerAutoLock& aLock) { + // Try to recycle big-enough chunks. (All chunks should have the same size, + // but it's a cheap test and may allow future adjustments based on actual + // data rate.) + if (chunk->BufferBytes() >= mChunkMinBufferBytes) { + // We keep up to two recycled chunks at any time. + if (!mRecycledChunks) { + mRecycledChunks = std::move(chunk); + } else if (!mRecycledChunks->GetNext()) { + mRecycledChunks->InsertNext(std::move(chunk)); + } + } + } + + UniquePtr<ProfileBufferChunk> TakeRecycledChunk( + const baseprofiler::detail::BaseProfilerAutoLock& aLock) { + UniquePtr<ProfileBufferChunk> recycled; + if (mRecycledChunks) { + recycled = std::exchange(mRecycledChunks, mRecycledChunks->ReleaseNext()); + recycled->MarkRecycled(); + } + return recycled; + } + + void DiscardOldestReleasedChunk( + const baseprofiler::detail::BaseProfilerAutoLock& aLock) { + MOZ_ASSERT(!!mReleasedChunks); + UniquePtr<ProfileBufferChunk> oldest = + std::exchange(mReleasedChunks, mReleasedChunks->ReleaseNext()); + mReleasedBufferBytes -= oldest->BufferBytes(); + if (mChunkDestroyedCallback) { + // Inform the user that we're going to destroy this chunk. + mChunkDestroyedCallback(*oldest); + } + MaybeRecycleChunk(std::move(oldest), aLock); + } + + using ChunkAndUpdate = std::pair<UniquePtr<ProfileBufferChunk>, Update>; + [[nodiscard]] ChunkAndUpdate GetChunk( + const baseprofiler::detail::BaseProfilerAutoLock& aLock) { + MOZ_ASSERT(mUser, "Not registered yet"); + // After this function, the total memory consumption will be the sum of: + // - Bytes from released (i.e., full) chunks, + // - Bytes from unreleased (still in use) chunks, + // - Bytes from the chunk we want to create/recycle. (Note that we don't + // count the extra bytes of chunk header, and of extra allocation ability, + // for the new chunk, as it's assumed to be negligible compared to the + // total memory limit.) + // If this total is higher than the local limit, we'll want to destroy + // the oldest released chunks until we're under the limit; if any, we may + // recycle one of them to avoid a deallocation followed by an allocation. + while (mReleasedBufferBytes + mUnreleasedBufferBytes + + mChunkMinBufferBytes >= + mMaxTotalBytes && + !!mReleasedChunks) { + // We have reached the local limit, discard the oldest released chunk. + DiscardOldestReleasedChunk(aLock); + } + + // Extract the recycled chunk, if any. + ChunkAndUpdate chunkAndUpdate{TakeRecycledChunk(aLock), Update()}; + UniquePtr<ProfileBufferChunk>& chunk = chunkAndUpdate.first; + + if (!chunk) { + // No recycled chunk -> Create a chunk now. (This could still fail.) + chunk = ProfileBufferChunk::Create(mChunkMinBufferBytes); + } + + if (chunk) { + // We do have a chunk (recycled or new), record its size as "unreleased". + mUnreleasedBufferBytes += chunk->BufferBytes(); + + chunkAndUpdate.second = + Update(mUnreleasedBufferBytes, mReleasedBufferBytes, + mReleasedChunks.get(), nullptr); + } + + return chunkAndUpdate; + } + + [[nodiscard]] size_t SizeOfExcludingThis( + MallocSizeOf aMallocSizeOf, + const baseprofiler::detail::BaseProfilerAutoLock&) const { + MOZ_ASSERT(mUser, "Not registered yet"); + size_t size = 0; + if (mReleasedChunks) { + size += mReleasedChunks->SizeOfIncludingThis(aMallocSizeOf); + } + if (mRecycledChunks) { + size += mRecycledChunks->SizeOfIncludingThis(aMallocSizeOf); + } + // Note: Missing size of std::function external resources (if any). + return size; + } + + // Maxumum number of bytes that should be used by all unreleased and released + // chunks. Note that only released chunks can be destroyed here, so it is the + // responsibility of the user to properly release their chunks when possible. + const size_t mMaxTotalBytes; + + // Minimum number of bytes that new chunks should be able to store. + // Used when calling `ProfileBufferChunk::Create()`. + const Length mChunkMinBufferBytes; + + // Mutex guarding the following members. + mutable baseprofiler::detail::BaseProfilerMutex mMutex; + + // Number of bytes currently held in chunks that have been given away (through + // `GetChunk` or `RequestChunk`) and not released yet. + size_t mUnreleasedBufferBytes = 0; + + // Number of bytes currently held in chunks that have been released and stored + // in `mReleasedChunks` below. + size_t mReleasedBufferBytes = 0; + + // List of all released chunks. The oldest one should be at the start of the + // list, and may be destroyed or recycled when the memory limit is reached. + UniquePtr<ProfileBufferChunk> mReleasedChunks; + + // This may hold chunks that were released then slated for destruction, they + // will be reused next time an allocation would have been needed. + UniquePtr<ProfileBufferChunk> mRecycledChunks; + + // Optional callback used to notify the user when a chunk is about to be + // destroyed or recycled. (The data content is always destroyed, but the chunk + // container may be reused.) + std::function<void(const ProfileBufferChunk&)> mChunkDestroyedCallback; + + // Callback set from `RequestChunk()`, until it is serviced in + // `FulfillChunkRequests()`. There can only be one request in flight. + std::function<void(UniquePtr<ProfileBufferChunk>)> mChunkReceiver; + + // Separate mutex guarding mUpdateCallback, so that it may be invoked outside + // of the main buffer `mMutex`. + mutable baseprofiler::detail::BaseProfilerMutex mUpdateCallbackMutex; + + UpdateCallback mUpdateCallback; +}; + +} // namespace mozilla + +#endif // ProfileBufferChunkManagerWithLocalLimit_h diff --git a/mozglue/baseprofiler/public/ProfileBufferControlledChunkManager.h b/mozglue/baseprofiler/public/ProfileBufferControlledChunkManager.h new file mode 100644 index 0000000000..45b39b163c --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferControlledChunkManager.h @@ -0,0 +1,203 @@ +/* -*- 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 ProfileBufferControlledChunkManager_h +#define ProfileBufferControlledChunkManager_h + +#include "mozilla/ProfileBufferChunk.h" + +#include <functional> +#include <vector> + +namespace mozilla { + +// A "Controlled" chunk manager will provide updates about chunks that it +// creates, releases, and destroys; and it can destroy released chunks as +// requested. +class ProfileBufferControlledChunkManager { + public: + using Length = ProfileBufferChunk::Length; + + virtual ~ProfileBufferControlledChunkManager() = default; + + // Minimum amount of chunk metadata to be transferred between processes. + struct ChunkMetadata { + // Timestamp when chunk was marked "done", which is used to: + // - determine its age, so the oldest one will be destroyed first, + // - uniquely identify this chunk in this process. (The parent process is + // responsible for associating this timestamp to its process id.) + TimeStamp mDoneTimeStamp; + // Size of this chunk's buffer. + Length mBufferBytes; + + ChunkMetadata(TimeStamp aDoneTimeStamp, Length aBufferBytes) + : mDoneTimeStamp(aDoneTimeStamp), mBufferBytes(aBufferBytes) {} + }; + + // Class collecting all information necessary to describe updates that + // happened in a chunk manager. + // An update can be folded into a previous update. + class Update { + public: + // Construct a "not-an-Update" object, which should only be used after a + // real update is folded into it. + Update() = default; + + // Construct a "final" Update, which marks the end of all updates from a + // chunk manager. + explicit Update(decltype(nullptr)) : mUnreleasedBytes(FINAL) {} + + // Construct an Update from the given data and released chunks. + // The chunk pointers may be null, and it doesn't matter if + // `aNewlyReleasedChunks` is already linked to `aExistingReleasedChunks` or + // not. + Update(size_t aUnreleasedBytes, size_t aReleasedBytes, + const ProfileBufferChunk* aExistingReleasedChunks, + const ProfileBufferChunk* aNewlyReleasedChunks) + : mUnreleasedBytes(aUnreleasedBytes), + mReleasedBytes(aReleasedBytes), + mOldestDoneTimeStamp( + aExistingReleasedChunks + ? aExistingReleasedChunks->ChunkHeader().mDoneTimeStamp + : TimeStamp{}) { + MOZ_RELEASE_ASSERT( + !IsNotUpdate(), + "Empty update should only be constructed with default constructor"); + MOZ_RELEASE_ASSERT( + !IsFinal(), + "Final update should only be constructed with nullptr constructor"); + for (const ProfileBufferChunk* chunk = aNewlyReleasedChunks; chunk; + chunk = chunk->GetNext()) { + mNewlyReleasedChunks.emplace_back(ChunkMetadata{ + chunk->ChunkHeader().mDoneTimeStamp, chunk->BufferBytes()}); + } + } + + // Construct an Update from raw data. + // This may be used to re-construct an Update that was previously + // serialized. + Update(size_t aUnreleasedBytes, size_t aReleasedBytes, + TimeStamp aOldestDoneTimeStamp, + std::vector<ChunkMetadata>&& aNewlyReleasedChunks) + : mUnreleasedBytes(aUnreleasedBytes), + mReleasedBytes(aReleasedBytes), + mOldestDoneTimeStamp(aOldestDoneTimeStamp), + mNewlyReleasedChunks(std::move(aNewlyReleasedChunks)) {} + + // Clear the Update completely and return it to a "not-an-Update" state. + void Clear() { + mUnreleasedBytes = NO_UPDATE; + mReleasedBytes = 0; + mOldestDoneTimeStamp = TimeStamp{}; + mNewlyReleasedChunks.clear(); + } + + bool IsNotUpdate() const { return mUnreleasedBytes == NO_UPDATE; } + + bool IsFinal() const { return mUnreleasedBytes == FINAL; } + + size_t UnreleasedBytes() const { + MOZ_RELEASE_ASSERT(!IsNotUpdate(), + "Cannot access UnreleasedBytes from empty update"); + MOZ_RELEASE_ASSERT(!IsFinal(), + "Cannot access UnreleasedBytes from final update"); + return mUnreleasedBytes; + } + + size_t ReleasedBytes() const { + MOZ_RELEASE_ASSERT(!IsNotUpdate(), + "Cannot access ReleasedBytes from empty update"); + MOZ_RELEASE_ASSERT(!IsFinal(), + "Cannot access ReleasedBytes from final update"); + return mReleasedBytes; + } + + TimeStamp OldestDoneTimeStamp() const { + MOZ_RELEASE_ASSERT(!IsNotUpdate(), + "Cannot access OldestDoneTimeStamp from empty update"); + MOZ_RELEASE_ASSERT(!IsFinal(), + "Cannot access OldestDoneTimeStamp from final update"); + return mOldestDoneTimeStamp; + } + + const std::vector<ChunkMetadata>& NewlyReleasedChunksRef() const { + MOZ_RELEASE_ASSERT( + !IsNotUpdate(), + "Cannot access NewlyReleasedChunksRef from empty update"); + MOZ_RELEASE_ASSERT( + !IsFinal(), "Cannot access NewlyReleasedChunksRef from final update"); + return mNewlyReleasedChunks; + } + + // Fold a later update into this one. + void Fold(Update&& aNewUpdate) { + MOZ_ASSERT( + !IsFinal() || aNewUpdate.IsFinal(), + "There shouldn't be another non-final update after the final update"); + + if (IsNotUpdate() || aNewUpdate.IsFinal()) { + // We were empty, or the new update is the final update, we just switch + // to that new update. + *this = std::move(aNewUpdate); + return; + } + + mUnreleasedBytes = aNewUpdate.mUnreleasedBytes; + mReleasedBytes = aNewUpdate.mReleasedBytes; + if (!aNewUpdate.mOldestDoneTimeStamp.IsNull()) { + MOZ_ASSERT(mOldestDoneTimeStamp.IsNull() || + mOldestDoneTimeStamp <= aNewUpdate.mOldestDoneTimeStamp); + mOldestDoneTimeStamp = aNewUpdate.mOldestDoneTimeStamp; + auto it = mNewlyReleasedChunks.begin(); + while (it != mNewlyReleasedChunks.end() && + it->mDoneTimeStamp < mOldestDoneTimeStamp) { + it = mNewlyReleasedChunks.erase(it); + } + } + if (!aNewUpdate.mNewlyReleasedChunks.empty()) { + mNewlyReleasedChunks.reserve(mNewlyReleasedChunks.size() + + aNewUpdate.mNewlyReleasedChunks.size()); + mNewlyReleasedChunks.insert(mNewlyReleasedChunks.end(), + aNewUpdate.mNewlyReleasedChunks.begin(), + aNewUpdate.mNewlyReleasedChunks.end()); + } + } + + private: + static const size_t NO_UPDATE = size_t(-1); + static const size_t FINAL = size_t(-2); + + size_t mUnreleasedBytes = NO_UPDATE; + size_t mReleasedBytes = 0; + TimeStamp mOldestDoneTimeStamp; + std::vector<ChunkMetadata> mNewlyReleasedChunks; + }; + + using UpdateCallback = std::function<void(Update&&)>; + + // This *may* be set (or reset) by an object that needs to know about all + // chunk updates that happen in this manager. The main use will be to + // coordinate the global memory usage of Firefox. + // If a non-empty callback is given, it will be immediately invoked with the + // current state. + // When the callback is about to be destroyed (by overwriting it here, or in + // the class destructor), it will be invoked one last time with an empty + // update. + // Note that the callback (even the first current-state callback) will be + // invoked from inside a locked scope, so it should *not* call other functions + // of the chunk manager. A side benefit of this locking is that it guarantees + // that no two invocations can overlap. + virtual void SetUpdateCallback(UpdateCallback&& aUpdateCallback) = 0; + + // This is a request to destroy all chunks before the given timestamp. + // This timestamp should be one that was given in a previous UpdateCallback + // call. Obviously, only released chunks can be destroyed. + virtual void DestroyChunksAtOrBefore(TimeStamp aDoneTimeStamp) = 0; +}; + +} // namespace mozilla + +#endif // ProfileBufferControlledChunkManager_h diff --git a/mozglue/baseprofiler/public/ProfileBufferEntryKinds.h b/mozglue/baseprofiler/public/ProfileBufferEntryKinds.h new file mode 100644 index 0000000000..c8280a92d7 --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferEntryKinds.h @@ -0,0 +1,94 @@ +/* -*- 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 ProfileBufferEntryKinds_h +#define ProfileBufferEntryKinds_h + +#include <cstdint> + +namespace mozilla { + +// This is equal to sizeof(double), which is the largest non-char variant in +// |u|. +static constexpr size_t ProfileBufferEntryNumChars = 8; + +// NOTE! If you add entries, you need to verify if they need to be added to the +// switch statement in DuplicateLastSample! +// This will evaluate the MACRO with (KIND, TYPE, SIZE) +#define FOR_EACH_PROFILE_BUFFER_ENTRY_KIND(MACRO) \ + MACRO(CategoryPair, int, sizeof(int)) \ + MACRO(CollectionStart, double, sizeof(double)) \ + MACRO(CollectionEnd, double, sizeof(double)) \ + MACRO(Label, const char*, sizeof(const char*)) \ + MACRO(FrameFlags, uint64_t, sizeof(uint64_t)) \ + MACRO(DynamicStringFragment, char*, ProfileBufferEntryNumChars) \ + MACRO(JitReturnAddr, void*, sizeof(void*)) \ + MACRO(InnerWindowID, uint64_t, sizeof(uint64_t)) \ + MACRO(LineNumber, int, sizeof(int)) \ + MACRO(ColumnNumber, int, sizeof(int)) \ + MACRO(NativeLeafAddr, void*, sizeof(void*)) \ + MACRO(Pause, double, sizeof(double)) \ + MACRO(Resume, double, sizeof(double)) \ + MACRO(PauseSampling, double, sizeof(double)) \ + MACRO(ResumeSampling, double, sizeof(double)) \ + MACRO(Responsiveness, double, sizeof(double)) \ + MACRO(ThreadId, int, sizeof(int)) \ + MACRO(Time, double, sizeof(double)) \ + MACRO(TimeBeforeCompactStack, double, sizeof(double)) \ + MACRO(CounterId, void*, sizeof(void*)) \ + MACRO(CounterKey, uint64_t, sizeof(uint64_t)) \ + MACRO(Number, uint64_t, sizeof(uint64_t)) \ + MACRO(Count, int64_t, sizeof(int64_t)) \ + MACRO(ProfilerOverheadTime, double, sizeof(double)) \ + MACRO(ProfilerOverheadDuration, double, sizeof(double)) + +// The `Kind` is a single byte identifying the type of data that is actually +// stored in a `ProfileBufferEntry`, as per the list in +// `FOR_EACH_PROFILE_BUFFER_ENTRY_KIND`. +// +// This byte is also used to identify entries in ProfileChunkedBuffer blocks, +// for both "legacy" entries that do contain a `ProfileBufferEntry`, and for +// new types of entries that may carry more data of different types. +// TODO: Eventually each type of "legacy" entry should be replaced with newer, +// more efficient kinds of entries (e.g., stack frames could be stored in one +// bigger entry, instead of multiple `ProfileBufferEntry`s); then we could +// discard `ProfileBufferEntry` and move this enum to a more appropriate spot. +using ProfileBufferEntryKindUnderlyingType = uint8_t; + +enum class ProfileBufferEntryKind : ProfileBufferEntryKindUnderlyingType { + INVALID = 0, +#define KIND(KIND, TYPE, SIZE) KIND, + FOR_EACH_PROFILE_BUFFER_ENTRY_KIND(KIND) +#undef KIND + + // Any value under `LEGACY_LIMIT` represents a `ProfileBufferEntry`. + LEGACY_LIMIT, + + // Any value starting here does *not* represent a `ProfileBufferEntry` and + // requires separate decoding and handling. + + // Markers and their data. + Marker = LEGACY_LIMIT, + + // Entry with "running times", such as CPU usage measurements. + // Optional between TimeBeforeCompactStack and CompactStack. + RunningTimes, + + // Optional between TimeBeforeCompactStack and CompactStack. + UnresponsiveDurationMs, + + // Collection of legacy stack entries, must follow a ThreadId and + // TimeBeforeCompactStack (which are not included in the CompactStack; + // TimeBeforeCompactStack is equivalent to Time, but indicates that a + // CompactStack follows shortly afterwards). + CompactStack, + + MODERN_LIMIT +}; + +} // namespace mozilla + +#endif // ProfileBufferEntryKinds_h diff --git a/mozglue/baseprofiler/public/ProfileBufferEntrySerialization.h b/mozglue/baseprofiler/public/ProfileBufferEntrySerialization.h new file mode 100644 index 0000000000..267b99f10d --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferEntrySerialization.h @@ -0,0 +1,1166 @@ +/* -*- 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 ProfileBufferEntrySerialization_h +#define ProfileBufferEntrySerialization_h + +#include "mozilla/Assertions.h" +#include "mozilla/leb128iterator.h" +#include "mozilla/Likely.h" +#include "mozilla/Maybe.h" +#include "mozilla/ProfileBufferIndex.h" +#include "mozilla/Span.h" +#include "mozilla/Tuple.h" +#include "mozilla/UniquePtrExtensions.h" +#include "mozilla/Unused.h" +#include "mozilla/Variant.h" + +#include <string> +#include <tuple> + +namespace mozilla { + +class ProfileBufferEntryWriter; + +// Iterator-like class used to read from an entry. +// An entry may be split in two memory segments (e.g., the ends of a ring +// buffer, or two chunks of a chunked buffer); it doesn't deal with this +// underlying buffer, but only with one or two spans pointing at the space +// where the entry lives. +class ProfileBufferEntryReader { + public: + using Byte = uint8_t; + using Length = uint32_t; + + using SpanOfConstBytes = Span<const Byte>; + + // Class to be specialized for types to be read from a profile buffer entry. + // See common specializations at the bottom of this header. + // The following static functions must be provided: + // static void ReadInto(EntryReader aER&, T& aT) + // { + // /* Call `aER.ReadX(...)` function to deserialize into aT, be sure to + // read exactly `Bytes(aT)`! */ + // } + // static T Read(EntryReader& aER) { + // /* Call `aER.ReadX(...)` function to deserialize and return a `T`, be + // sure to read exactly `Bytes(returned value)`! */ + // } + template <typename T> + struct Deserializer; + + ProfileBufferEntryReader() = default; + + // Reader over one Span. + ProfileBufferEntryReader(SpanOfConstBytes aSpan, + ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) + : mCurrentSpan(aSpan), + mNextSpanOrEmpty(aSpan.Last(0)), + mCurrentBlockIndex(aCurrentBlockIndex), + mNextBlockIndex(aNextBlockIndex) { + // 2nd internal Span points at the end of the 1st internal Span, to enforce + // invariants. + CheckInvariants(); + } + + // Reader over two Spans, the second one must not be empty. + ProfileBufferEntryReader(SpanOfConstBytes aSpanHead, + SpanOfConstBytes aSpanTail, + ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) + : mCurrentSpan(aSpanHead), + mNextSpanOrEmpty(aSpanTail), + mCurrentBlockIndex(aCurrentBlockIndex), + mNextBlockIndex(aNextBlockIndex) { + MOZ_RELEASE_ASSERT(!mNextSpanOrEmpty.IsEmpty()); + if (MOZ_UNLIKELY(mCurrentSpan.IsEmpty())) { + // First span is already empty, skip it. + mCurrentSpan = mNextSpanOrEmpty; + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + CheckInvariants(); + } + + // Allow copying, which is needed when used as an iterator in some std + // functions (e.g., string assignment), and to occasionally backtrack. + // Be aware that the main profile buffer APIs give a reference to an entry + // reader, and expect that reader to advance to the end of the entry, so don't + // just advance copies! + ProfileBufferEntryReader(const ProfileBufferEntryReader&) = default; + ProfileBufferEntryReader& operator=(const ProfileBufferEntryReader&) = + default; + + // Don't =default moving, as it doesn't bring any benefit in this class. + + [[nodiscard]] Length RemainingBytes() const { + return mCurrentSpan.LengthBytes() + mNextSpanOrEmpty.LengthBytes(); + } + + void SetRemainingBytes(Length aBytes) { + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + if (aBytes <= mCurrentSpan.LengthBytes()) { + mCurrentSpan = mCurrentSpan.First(aBytes); + mNextSpanOrEmpty = mCurrentSpan.Last(0); + } else { + mNextSpanOrEmpty = + mNextSpanOrEmpty.First(aBytes - mCurrentSpan.LengthBytes()); + } + } + + [[nodiscard]] ProfileBufferBlockIndex CurrentBlockIndex() const { + return mCurrentBlockIndex; + } + + [[nodiscard]] ProfileBufferBlockIndex NextBlockIndex() const { + return mNextBlockIndex; + } + + // Create a reader of size zero, pointing at aOffset past the current position + // of this Reader, so it can be used as end iterator. + [[nodiscard]] ProfileBufferEntryReader EmptyIteratorAtOffset( + Length aOffset) const { + MOZ_RELEASE_ASSERT(aOffset <= RemainingBytes()); + if (MOZ_LIKELY(aOffset < mCurrentSpan.LengthBytes())) { + // aOffset is before the end of mCurrentSpan. + return ProfileBufferEntryReader(mCurrentSpan.Subspan(aOffset, 0), + mCurrentBlockIndex, mNextBlockIndex); + } + // aOffset is right at the end of mCurrentSpan, or inside mNextSpanOrEmpty. + return ProfileBufferEntryReader( + mNextSpanOrEmpty.Subspan(aOffset - mCurrentSpan.LengthBytes(), 0), + mCurrentBlockIndex, mNextBlockIndex); + } + + // Be like a limited input iterator, with only `*`, prefix-`++`, `==`, `!=`. + // These definitions are expected by std functions, to recognize this as an + // iterator. See https://en.cppreference.com/w/cpp/iterator/iterator_traits + using difference_type = std::make_signed_t<Length>; + using value_type = Byte; + using pointer = const Byte*; + using reference = const Byte&; + using iterator_category = std::input_iterator_tag; + + [[nodiscard]] const Byte& operator*() { + // Assume the caller will read from the returned reference (and not just + // take the address). + MOZ_RELEASE_ASSERT(mCurrentSpan.LengthBytes() >= 1); + return *(mCurrentSpan.Elements()); + } + + ProfileBufferEntryReader& operator++() { + MOZ_RELEASE_ASSERT(mCurrentSpan.LengthBytes() >= 1); + if (MOZ_LIKELY(mCurrentSpan.LengthBytes() > 1)) { + // More than 1 byte left in mCurrentSpan, just eat it. + mCurrentSpan = mCurrentSpan.From(1); + } else { + // mCurrentSpan will be empty, move mNextSpanOrEmpty to mCurrentSpan. + mCurrentSpan = mNextSpanOrEmpty; + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + CheckInvariants(); + return *this; + } + + ProfileBufferEntryReader& operator+=(Length aBytes) { + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + if (MOZ_LIKELY(aBytes <= mCurrentSpan.LengthBytes())) { + // All bytes are in mCurrentSpan. + // Update mCurrentSpan past the read bytes. + mCurrentSpan = mCurrentSpan.From(aBytes); + if (mCurrentSpan.IsEmpty() && !mNextSpanOrEmpty.IsEmpty()) { + // Don't leave mCurrentSpan empty, move non-empty mNextSpanOrEmpty into + // mCurrentSpan. + mCurrentSpan = mNextSpanOrEmpty; + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + } else { + // mCurrentSpan does not hold enough bytes. + // This should only happen at most once: Only for double spans, and when + // data crosses the gap. + const Length tail = + aBytes - static_cast<Length>(mCurrentSpan.LengthBytes()); + // Move mNextSpanOrEmpty to mCurrentSpan, past the data. So the next call + // will go back to the true case above. + mCurrentSpan = mNextSpanOrEmpty.From(tail); + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + CheckInvariants(); + return *this; + } + + [[nodiscard]] bool operator==(const ProfileBufferEntryReader& aOther) const { + return mCurrentSpan.Elements() == aOther.mCurrentSpan.Elements(); + } + [[nodiscard]] bool operator!=(const ProfileBufferEntryReader& aOther) const { + return mCurrentSpan.Elements() != aOther.mCurrentSpan.Elements(); + } + + // Read an unsigned LEB128 number and move iterator ahead. + template <typename T> + [[nodiscard]] T ReadULEB128() { + return ::mozilla::ReadULEB128<T>(*this); + } + + // Read a sequence of bytes, like memcpy. + void ReadBytes(void* aDest, Length aBytes) { + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + if (MOZ_LIKELY(aBytes <= mCurrentSpan.LengthBytes())) { + // All bytes are in mCurrentSpan. + memcpy(aDest, mCurrentSpan.Elements(), aBytes); + // Update mCurrentSpan past the read bytes. + mCurrentSpan = mCurrentSpan.From(aBytes); + if (mCurrentSpan.IsEmpty() && !mNextSpanOrEmpty.IsEmpty()) { + // Don't leave mCurrentSpan empty, move non-empty mNextSpanOrEmpty into + // mCurrentSpan. + mCurrentSpan = mNextSpanOrEmpty; + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + } else { + // mCurrentSpan does not hold enough bytes. + // This should only happen at most once: Only for double spans, and when + // data crosses the gap. + // Split data between the end of mCurrentSpan and the beginning of + // mNextSpanOrEmpty. + memcpy(aDest, mCurrentSpan.Elements(), mCurrentSpan.LengthBytes()); + const Length tail = + aBytes - static_cast<Length>(mCurrentSpan.LengthBytes()); + memcpy(reinterpret_cast<Byte*>(aDest) + mCurrentSpan.LengthBytes(), + mNextSpanOrEmpty.Elements(), tail); + // Move mNextSpanOrEmpty to mCurrentSpan, past the data. So the next call + // will go back to the true case above. + mCurrentSpan = mNextSpanOrEmpty.From(tail); + mNextSpanOrEmpty = mNextSpanOrEmpty.Last(0); + } + CheckInvariants(); + } + + template <typename T> + void ReadIntoObject(T& aObject) { + Deserializer<T>::ReadInto(*this, aObject); + } + + // Read into one or more objects, sequentially. + // `EntryReader::ReadIntoObjects()` with nothing is implicitly allowed, this + // could be useful for generic programming. + template <typename... Ts> + void ReadIntoObjects(Ts&... aTs) { + (ReadIntoObject(aTs), ...); + } + + // Read data as an object and move iterator ahead. + template <typename T> + [[nodiscard]] T ReadObject() { + T ob = Deserializer<T>::Read(*this); + return ob; + } + + private: + friend class ProfileBufferEntryWriter; + + // Invariants: + // - mCurrentSpan cannot be empty unless mNextSpanOrEmpty is also empty. So + // mCurrentSpan always points at the next byte to read or the end. + // - If mNextSpanOrEmpty is empty, it points at the end of mCurrentSpan. So + // when reaching the end of mCurrentSpan, we can blindly move + // mNextSpanOrEmpty to mCurrentSpan and keep the invariants. + SpanOfConstBytes mCurrentSpan; + SpanOfConstBytes mNextSpanOrEmpty; + ProfileBufferBlockIndex mCurrentBlockIndex; + ProfileBufferBlockIndex mNextBlockIndex; + + void CheckInvariants() const { + MOZ_ASSERT(!mCurrentSpan.IsEmpty() || mNextSpanOrEmpty.IsEmpty()); + MOZ_ASSERT(!mNextSpanOrEmpty.IsEmpty() || + (mNextSpanOrEmpty == mCurrentSpan.Last(0))); + } +}; + +// Iterator-like class used to write into an entry. +// An entry may be split in two memory segments (e.g., the ends of a ring +// buffer, or two chunks of a chunked buffer); it doesn't deal with this +// underlying buffer, but only with one or two spans pointing at the space +// reserved for the entry. +class ProfileBufferEntryWriter { + public: + using Byte = uint8_t; + using Length = uint32_t; + + using SpanOfBytes = Span<Byte>; + + // Class to be specialized for types to be written in an entry. + // See common specializations at the bottom of this header. + // The following static functions must be provided: + // static Length Bytes(const T& aT) { + // /* Return number of bytes that will be written. */ + // } + // static void Write(ProfileBufferEntryWriter& aEW, + // const T& aT) { + // /* Call `aEW.WriteX(...)` functions to serialize aT, be sure to write + // exactly `Bytes(aT)` bytes! */ + // } + template <typename T> + struct Serializer; + + ProfileBufferEntryWriter() = default; + + ProfileBufferEntryWriter(SpanOfBytes aSpan, + ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) + : mCurrentSpan(aSpan), + mCurrentBlockIndex(aCurrentBlockIndex), + mNextBlockIndex(aNextBlockIndex) {} + + ProfileBufferEntryWriter(SpanOfBytes aSpanHead, SpanOfBytes aSpanTail, + ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) + : mCurrentSpan(aSpanHead), + mNextSpanOrEmpty(aSpanTail), + mCurrentBlockIndex(aCurrentBlockIndex), + mNextBlockIndex(aNextBlockIndex) { + // Either: + // - mCurrentSpan is not empty, OR + // - mNextSpanOrEmpty is empty if mNextSpanOrEmpty is empty as well. + MOZ_RELEASE_ASSERT(!mCurrentSpan.IsEmpty() || mNextSpanOrEmpty.IsEmpty()); + } + + // Disable copying and moving, so we can't have multiple writing heads. + ProfileBufferEntryWriter(const ProfileBufferEntryWriter&) = delete; + ProfileBufferEntryWriter& operator=(const ProfileBufferEntryWriter&) = delete; + ProfileBufferEntryWriter(ProfileBufferEntryWriter&&) = delete; + ProfileBufferEntryWriter& operator=(ProfileBufferEntryWriter&&) = delete; + + void Set() { + mCurrentSpan = SpanOfBytes{}; + mNextSpanOrEmpty = SpanOfBytes{}; + mCurrentBlockIndex = nullptr; + mNextBlockIndex = nullptr; + } + + void Set(SpanOfBytes aSpan, ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) { + mCurrentSpan = aSpan; + mNextSpanOrEmpty = SpanOfBytes{}; + mCurrentBlockIndex = aCurrentBlockIndex; + mNextBlockIndex = aNextBlockIndex; + } + + void Set(SpanOfBytes aSpan0, SpanOfBytes aSpan1, + ProfileBufferBlockIndex aCurrentBlockIndex, + ProfileBufferBlockIndex aNextBlockIndex) { + mCurrentSpan = aSpan0; + mNextSpanOrEmpty = aSpan1; + mCurrentBlockIndex = aCurrentBlockIndex; + mNextBlockIndex = aNextBlockIndex; + // Either: + // - mCurrentSpan is not empty, OR + // - mNextSpanOrEmpty is empty if mNextSpanOrEmpty is empty as well. + MOZ_RELEASE_ASSERT(!mCurrentSpan.IsEmpty() || mNextSpanOrEmpty.IsEmpty()); + } + + [[nodiscard]] Length RemainingBytes() const { + return mCurrentSpan.LengthBytes() + mNextSpanOrEmpty.LengthBytes(); + } + + [[nodiscard]] ProfileBufferBlockIndex CurrentBlockIndex() const { + return mCurrentBlockIndex; + } + + [[nodiscard]] ProfileBufferBlockIndex NextBlockIndex() const { + return mNextBlockIndex; + } + + // Be like a limited output iterator, with only `*` and prefix-`++`. + // These definitions are expected by std functions, to recognize this as an + // iterator. See https://en.cppreference.com/w/cpp/iterator/iterator_traits + using value_type = Byte; + using pointer = Byte*; + using reference = Byte&; + using iterator_category = std::output_iterator_tag; + + [[nodiscard]] Byte& operator*() { + MOZ_RELEASE_ASSERT(RemainingBytes() >= 1); + return *( + (MOZ_LIKELY(!mCurrentSpan.IsEmpty()) ? mCurrentSpan : mNextSpanOrEmpty) + .Elements()); + } + + ProfileBufferEntryWriter& operator++() { + if (MOZ_LIKELY(mCurrentSpan.LengthBytes() >= 1)) { + // There is at least 1 byte in mCurrentSpan, eat it. + mCurrentSpan = mCurrentSpan.From(1); + } else { + // mCurrentSpan is empty, move mNextSpanOrEmpty (past the first byte) to + // mCurrentSpan. + MOZ_RELEASE_ASSERT(mNextSpanOrEmpty.LengthBytes() >= 1); + mCurrentSpan = mNextSpanOrEmpty.From(1); + mNextSpanOrEmpty = mNextSpanOrEmpty.First(0); + } + return *this; + } + + ProfileBufferEntryWriter& operator+=(Length aBytes) { + // Note: This is a rare operation. The code below is a copy of `WriteBytes` + // but without the `memcpy`s. + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + if (MOZ_LIKELY(aBytes <= mCurrentSpan.LengthBytes())) { + // Data fits in mCurrentSpan. + // Update mCurrentSpan. It may become empty, so in case of a double span, + // the next call will go to the false case below. + mCurrentSpan = mCurrentSpan.From(aBytes); + } else { + // Data does not fully fit in mCurrentSpan. + // This should only happen at most once: Only for double spans, and when + // data crosses the gap or starts there. + const Length tail = + aBytes - static_cast<Length>(mCurrentSpan.LengthBytes()); + // Move mNextSpanOrEmpty to mCurrentSpan, past the data. So the next call + // will go back to the true case above. + mCurrentSpan = mNextSpanOrEmpty.From(tail); + mNextSpanOrEmpty = mNextSpanOrEmpty.First(0); + } + return *this; + } + + // Number of bytes needed to represent `aValue` in unsigned LEB128. + template <typename T> + [[nodiscard]] static unsigned ULEB128Size(T aValue) { + return ::mozilla::ULEB128Size(aValue); + } + + // Write number as unsigned LEB128 and move iterator ahead. + template <typename T> + void WriteULEB128(T aValue) { + ::mozilla::WriteULEB128(aValue, *this); + } + + // Number of bytes needed to serialize objects. + template <typename... Ts> + [[nodiscard]] static Length SumBytes(const Ts&... aTs) { + return (0 + ... + Serializer<Ts>::Bytes(aTs)); + } + + // Write a sequence of bytes, like memcpy. + void WriteBytes(const void* aSrc, Length aBytes) { + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + if (MOZ_LIKELY(aBytes <= mCurrentSpan.LengthBytes())) { + // Data fits in mCurrentSpan. + memcpy(mCurrentSpan.Elements(), aSrc, aBytes); + // Update mCurrentSpan. It may become empty, so in case of a double span, + // the next call will go to the false case below. + mCurrentSpan = mCurrentSpan.From(aBytes); + } else { + // Data does not fully fit in mCurrentSpan. + // This should only happen at most once: Only for double spans, and when + // data crosses the gap or starts there. + // Split data between the end of mCurrentSpan and the beginning of + // mNextSpanOrEmpty. (mCurrentSpan could be empty, it's ok to do a memcpy + // because Span::Elements() is never null.) + memcpy(mCurrentSpan.Elements(), aSrc, mCurrentSpan.LengthBytes()); + const Length tail = + aBytes - static_cast<Length>(mCurrentSpan.LengthBytes()); + memcpy(mNextSpanOrEmpty.Elements(), + reinterpret_cast<const Byte*>(aSrc) + mCurrentSpan.LengthBytes(), + tail); + // Move mNextSpanOrEmpty to mCurrentSpan, past the data. So the next call + // will go back to the true case above. + mCurrentSpan = mNextSpanOrEmpty.From(tail); + mNextSpanOrEmpty = mNextSpanOrEmpty.First(0); + } + } + + void WriteFromReader(ProfileBufferEntryReader& aReader, Length aBytes) { + MOZ_RELEASE_ASSERT(aBytes <= RemainingBytes()); + MOZ_RELEASE_ASSERT(aBytes <= aReader.RemainingBytes()); + Length read0 = std::min( + aBytes, static_cast<Length>(aReader.mCurrentSpan.LengthBytes())); + if (read0 != 0) { + WriteBytes(aReader.mCurrentSpan.Elements(), read0); + } + Length read1 = aBytes - read0; + if (read1 != 0) { + WriteBytes(aReader.mNextSpanOrEmpty.Elements(), read1); + } + aReader += aBytes; + } + + // Write a single object by using the appropriate Serializer. + template <typename T> + void WriteObject(const T& aObject) { + Serializer<T>::Write(*this, aObject); + } + + // Write one or more objects, sequentially. + // Allow `EntryWrite::WriteObjects()` with nothing, this could be useful + // for generic programming. + template <typename... Ts> + void WriteObjects(const Ts&... aTs) { + (WriteObject(aTs), ...); + } + + private: + // The two spans covering the memory still to be written. + SpanOfBytes mCurrentSpan; + SpanOfBytes mNextSpanOrEmpty; + ProfileBufferBlockIndex mCurrentBlockIndex; + ProfileBufferBlockIndex mNextBlockIndex; +}; + +// ============================================================================ +// Serializer and Deserializer ready-to-use specializations. + +// ---------------------------------------------------------------------------- +// Trivially-copyable types (default) + +// The default implementation works for all trivially-copyable types (e.g., +// PODs). +// +// Usage: `aEW.WriteObject(123);`. +// +// Raw pointers, though trivially-copyable, are explictly forbidden when writing +// (to avoid unexpected leaks/UAFs), instead use one of +// `WrapProfileBufferLiteralCStringPointer`, `WrapProfileBufferUnownedCString`, +// or `WrapProfileBufferRawPointer` as needed. +template <typename T> +struct ProfileBufferEntryWriter::Serializer { + static_assert(std::is_trivially_copyable<T>::value, + "Serializer only works with trivially-copyable types by " + "default, use/add specialization for other types."); + + static constexpr Length Bytes(const T&) { return sizeof(T); } + + static void Write(ProfileBufferEntryWriter& aEW, const T& aT) { + static_assert(!std::is_pointer<T>::value, + "Serializer won't write raw pointers by default, use " + "WrapProfileBufferRawPointer or other."); + aEW.WriteBytes(&aT, sizeof(T)); + } +}; + +// Usage: `aER.ReadObject<int>();` or `int x; aER.ReadIntoObject(x);`. +template <typename T> +struct ProfileBufferEntryReader::Deserializer { + static_assert(std::is_trivially_copyable<T>::value, + "Deserializer only works with trivially-copyable types by " + "default, use/add specialization for other types."); + + static void ReadInto(ProfileBufferEntryReader& aER, T& aT) { + aER.ReadBytes(&aT, sizeof(T)); + } + + static T Read(ProfileBufferEntryReader& aER) { + // Note that this creates a default `T` first, and then overwrites it with + // bytes from the buffer. Trivially-copyable types support this without UB. + T ob; + ReadInto(aER, ob); + return ob; + } +}; + +// ---------------------------------------------------------------------------- +// Strip const/volatile/reference from types. + +// Automatically strip `const`. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<const T> + : public ProfileBufferEntryWriter::Serializer<T> {}; + +template <typename T> +struct ProfileBufferEntryReader::Deserializer<const T> + : public ProfileBufferEntryReader::Deserializer<T> {}; + +// Automatically strip `volatile`. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<volatile T> + : public ProfileBufferEntryWriter::Serializer<T> {}; + +template <typename T> +struct ProfileBufferEntryReader::Deserializer<volatile T> + : public ProfileBufferEntryReader::Deserializer<T> {}; + +// Automatically strip `lvalue-reference`. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<T&> + : public ProfileBufferEntryWriter::Serializer<T> {}; + +template <typename T> +struct ProfileBufferEntryReader::Deserializer<T&> + : public ProfileBufferEntryReader::Deserializer<T> {}; + +// Automatically strip `rvalue-reference`. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<T&&> + : public ProfileBufferEntryWriter::Serializer<T> {}; + +template <typename T> +struct ProfileBufferEntryReader::Deserializer<T&&> + : public ProfileBufferEntryReader::Deserializer<T> {}; + +// ---------------------------------------------------------------------------- +// ProfileBufferBlockIndex + +// ProfileBufferBlockIndex, serialized as the underlying value. +template <> +struct ProfileBufferEntryWriter::Serializer<ProfileBufferBlockIndex> { + static constexpr Length Bytes(const ProfileBufferBlockIndex& aBlockIndex) { + return sizeof(ProfileBufferBlockIndex); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const ProfileBufferBlockIndex& aBlockIndex) { + aEW.WriteBytes(&aBlockIndex, sizeof(aBlockIndex)); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<ProfileBufferBlockIndex> { + static void ReadInto(ProfileBufferEntryReader& aER, + ProfileBufferBlockIndex& aBlockIndex) { + aER.ReadBytes(&aBlockIndex, sizeof(aBlockIndex)); + } + + static ProfileBufferBlockIndex Read(ProfileBufferEntryReader& aER) { + ProfileBufferBlockIndex blockIndex; + ReadInto(aER, blockIndex); + return blockIndex; + } +}; + +// ---------------------------------------------------------------------------- +// Literal C string pointer + +// Wrapper around a pointer to a literal C string. +template <size_t NonTerminalCharacters> +struct ProfileBufferLiteralCStringPointer { + const char* mCString; +}; + +// Wrap a pointer to a literal C string. +template <size_t CharactersIncludingTerminal> +ProfileBufferLiteralCStringPointer<CharactersIncludingTerminal - 1> +WrapProfileBufferLiteralCStringPointer( + const char (&aCString)[CharactersIncludingTerminal]) { + return {aCString}; +} + +// Literal C strings, serialized as the raw pointer because it is unique and +// valid for the whole program lifetime. +// +// Usage: `aEW.WriteObject(WrapProfileBufferLiteralCStringPointer("hi"));`. +// +// No deserializer is provided for this type, instead it must be deserialized as +// a raw pointer: `aER.ReadObject<const char*>();` +template <size_t CharactersIncludingTerminal> +struct ProfileBufferEntryReader::Deserializer< + ProfileBufferLiteralCStringPointer<CharactersIncludingTerminal>> { + static constexpr Length Bytes( + const ProfileBufferLiteralCStringPointer<CharactersIncludingTerminal>&) { + // We're only storing a pointer, its size is independent from the pointer + // value. + return sizeof(const char*); + } + + static void Write( + ProfileBufferEntryWriter& aEW, + const ProfileBufferLiteralCStringPointer<CharactersIncludingTerminal>& + aWrapper) { + // Write the pointer *value*, not the string contents. + aEW.WriteBytes(aWrapper.mCString, sizeof(aWrapper.mCString)); + } +}; + +// ---------------------------------------------------------------------------- +// C string contents + +// Wrapper around a pointer to a C string whose contents will be serialized. +struct ProfileBufferUnownedCString { + const char* mCString; +}; + +// Wrap a pointer to a C string whose contents will be serialized. +inline ProfileBufferUnownedCString WrapProfileBufferUnownedCString( + const char* aCString) { + return {aCString}; +} + +// The contents of a (probably) unowned C string are serialized as the number of +// characters (encoded as ULEB128) and all the characters in the string. The +// terminal '\0' is omitted. +// +// Usage: `aEW.WriteObject(WrapProfileBufferUnownedCString(str.c_str()))`. +// +// No deserializer is provided for this pointer type, instead it must be +// deserialized as one of the other string types that manages its contents, +// e.g.: `aER.ReadObject<std::string>();` +template <> +struct ProfileBufferEntryWriter::Serializer<ProfileBufferUnownedCString> { + static Length Bytes(const ProfileBufferUnownedCString& aS) { + const auto len = strlen(aS.mCString); + return ULEB128Size(len) + len; + } + + static void Write(ProfileBufferEntryWriter& aEW, + const ProfileBufferUnownedCString& aS) { + const auto len = strlen(aS.mCString); + aEW.WriteULEB128(len); + aEW.WriteBytes(aS.mCString, len); + } +}; + +// ---------------------------------------------------------------------------- +// Raw pointers + +// Wrapper around a pointer to be serialized as the raw pointer value. +template <typename T> +struct ProfileBufferRawPointer { + T* mRawPointer; +}; + +// Wrap a pointer to be serialized as the raw pointer value. +template <typename T> +ProfileBufferRawPointer<T> WrapProfileBufferRawPointer(T* aRawPointer) { + return {aRawPointer}; +} + +// Raw pointers are serialized as the raw pointer value. +// +// Usage: `aEW.WriteObject(WrapProfileBufferRawPointer(ptr));` +// +// The wrapper is compulsory when writing pointers (to avoid unexpected +// leaks/UAFs), but reading can be done straight into a raw pointer object, +// e.g.: `aER.ReadObject<Foo*>;`. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<ProfileBufferRawPointer<T>> { + template <typename U> + static constexpr Length Bytes(const U&) { + return sizeof(T*); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const ProfileBufferRawPointer<T>& aWrapper) { + aEW.WriteBytes(&aWrapper.mRawPointer, sizeof(aWrapper.mRawPointer)); + } +}; + +// Usage: `aER.ReadObject<Foo*>;` or `Foo* p; aER.ReadIntoObject(p);`, no +// wrapper necessary. +template <typename T> +struct ProfileBufferEntryReader::Deserializer<ProfileBufferRawPointer<T>> { + static void ReadInto(ProfileBufferEntryReader& aER, + ProfileBufferRawPointer<T>& aPtr) { + aER.ReadBytes(&aPtr.mRawPointer, sizeof(aPtr)); + } + + static ProfileBufferRawPointer<T> Read(ProfileBufferEntryReader& aER) { + ProfileBufferRawPointer<T> rawPointer; + ReadInto(aER, rawPointer); + return rawPointer; + } +}; + +// ---------------------------------------------------------------------------- +// std::string contents + +// std::string contents are serialized as the number of characters (encoded as +// ULEB128) and all the characters in the string. The terminal '\0' is omitted. +// +// Usage: `std::string s = ...; aEW.WriteObject(s);` +template <typename CHAR> +struct ProfileBufferEntryWriter::Serializer<std::basic_string<CHAR>> { + static Length Bytes(const std::basic_string<CHAR>& aS) { + const Length len = static_cast<Length>(aS.length()); + return ULEB128Size(len) + len; + } + + static void Write(ProfileBufferEntryWriter& aEW, + const std::basic_string<CHAR>& aS) { + const Length len = static_cast<Length>(aS.length()); + aEW.WriteULEB128(len); + aEW.WriteBytes(aS.c_str(), len * sizeof(CHAR)); + } +}; + +// Usage: `std::string s = aEW.ReadObject<std::string>(s);` or +// `std::string s; aER.ReadIntoObject(s);` +template <typename CHAR> +struct ProfileBufferEntryReader::Deserializer<std::basic_string<CHAR>> { + static void ReadCharsInto(ProfileBufferEntryReader& aER, + std::basic_string<CHAR>& aS, size_t aLength) { + // Assign to `aS` by using iterators. + // (`aER+0` so we get the same iterator type as `aER+len`.) + aS.assign(aER, aER.EmptyIteratorAtOffset(aLength)); + aER += aLength; + } + + static void ReadInto(ProfileBufferEntryReader& aER, + std::basic_string<CHAR>& aS) { + ReadCharsInto( + aER, aS, + aER.ReadULEB128<typename std::basic_string<CHAR>::size_type>()); + } + + static std::basic_string<CHAR> ReadChars(ProfileBufferEntryReader& aER, + size_t aLength) { + // Construct a string by using iterators. + // (`aER+0` so we get the same iterator type as `aER+len`.) + std::basic_string<CHAR> s(aER, aER.EmptyIteratorAtOffset(aLength)); + aER += aLength; + return s; + } + + static std::basic_string<CHAR> Read(ProfileBufferEntryReader& aER) { + return ReadChars( + aER, aER.ReadULEB128<typename std::basic_string<CHAR>::size_type>()); + } +}; + +// ---------------------------------------------------------------------------- +// mozilla::UniqueFreePtr<CHAR> + +// UniqueFreePtr<CHAR>, which points at a string allocated with `malloc` +// (typically generated by `strdup()`), is serialized as the number of +// *bytes* (encoded as ULEB128) and all the characters in the string. The +// null terminator is omitted. +// `CHAR` can be any type that has a specialization for +// `std::char_traits<CHAR>::length(const CHAR*)`. +// +// Note: A nullptr pointer will be serialized like an empty string, so when +// deserializing it will result in an allocated buffer only containing a +// single null terminator. +template <typename CHAR> +struct ProfileBufferEntryWriter::Serializer<UniqueFreePtr<CHAR>> { + static Length Bytes(const UniqueFreePtr<CHAR>& aS) { + if (!aS) { + // Null pointer, store it as if it was an empty string (so: 0 bytes). + return ULEB128Size(0u); + } + // Note that we store the size in *bytes*, not in number of characters. + const auto bytes = std::char_traits<CHAR>::length(aS.get()) * sizeof(CHAR); + return ULEB128Size(bytes) + bytes; + } + + static void Write(ProfileBufferEntryWriter& aEW, + const UniqueFreePtr<CHAR>& aS) { + if (!aS) { + // Null pointer, store it as if it was an empty string (so we write a + // length of 0 bytes). + aEW.WriteULEB128(0u); + return; + } + // Note that we store the size in *bytes*, not in number of characters. + const auto bytes = std::char_traits<CHAR>::length(aS.get()) * sizeof(CHAR); + aEW.WriteULEB128(bytes); + aEW.WriteBytes(aS.get(), bytes); + } +}; + +template <typename CHAR> +struct ProfileBufferEntryReader::Deserializer<UniqueFreePtr<CHAR>> { + static void ReadInto(ProfileBufferEntryReader& aER, UniqueFreePtr<CHAR>& aS) { + aS = Read(aER); + } + + static UniqueFreePtr<CHAR> Read(ProfileBufferEntryReader& aER) { + // Read the number of *bytes* that follow. + const auto bytes = aER.ReadULEB128<size_t>(); + // We need a buffer of the non-const character type. + using NC_CHAR = std::remove_const_t<CHAR>; + // We allocate the required number of bytes, plus one extra character for + // the null terminator. + NC_CHAR* buffer = static_cast<NC_CHAR*>(malloc(bytes + sizeof(NC_CHAR))); + // Copy the characters into the buffer. + aER.ReadBytes(buffer, bytes); + // And append a null terminator. + buffer[bytes / sizeof(NC_CHAR)] = NC_CHAR(0); + return UniqueFreePtr<CHAR>(buffer); + } +}; + +// ---------------------------------------------------------------------------- +// std::tuple + +// std::tuple is serialized as a sequence of each recursively-serialized item. +// +// This is equivalent to manually serializing each item, so reading/writing +// tuples is equivalent to reading/writing their elements in order, e.g.: +// ``` +// std::tuple<int, std::string> is = ...; +// aEW.WriteObject(is); // Write the tuple, equivalent to: +// aEW.WriteObject(/* int */ std::get<0>(is), /* string */ std::get<1>(is)); +// ... +// // Reading back can be done directly into a tuple: +// auto is = aER.ReadObject<std::tuple<int, std::string>>(); +// // Or each item could be read separately: +// auto i = aER.ReadObject<int>(); auto s = aER.ReadObject<std::string>(); +// ``` +template <typename... Ts> +struct ProfileBufferEntryWriter::Serializer<std::tuple<Ts...>> { + private: + template <size_t... Is> + static Length TupleBytes(const std::tuple<Ts...>& aTuple, + std::index_sequence<Is...>) { + return (0 + ... + SumBytes(std::get<Is>(aTuple))); + } + + template <size_t... Is> + static void TupleWrite(ProfileBufferEntryWriter& aEW, + const std::tuple<Ts...>& aTuple, + std::index_sequence<Is...>) { + (aEW.WriteObject(std::get<Is>(aTuple)), ...); + } + + public: + static Length Bytes(const std::tuple<Ts...>& aTuple) { + // Generate a 0..N-1 index pack, we'll add the sizes of each item. + return TupleBytes(aTuple, std::index_sequence_for<Ts...>()); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const std::tuple<Ts...>& aTuple) { + // Generate a 0..N-1 index pack, we'll write each item. + TupleWrite(aEW, aTuple, std::index_sequence_for<Ts...>()); + } +}; + +template <typename... Ts> +struct ProfileBufferEntryReader::Deserializer<std::tuple<Ts...>> { + static void ReadInto(ProfileBufferEntryReader& aER, + std::tuple<Ts...>& aTuple) { + aER.ReadBytes(&aTuple, Bytes(aTuple)); + } + + static std::tuple<Ts...> Read(ProfileBufferEntryReader& aER) { + // Note that this creates default `Ts` first, and then overwrites them. + std::tuple<Ts...> ob; + ReadInto(aER, ob); + return ob; + } +}; + +// ---------------------------------------------------------------------------- +// mozilla::Tuple + +// Tuple is serialized as a sequence of each recursively-serialized +// item. +// +// This is equivalent to manually serializing each item, so reading/writing +// tuples is equivalent to reading/writing their elements in order, e.g.: +// ``` +// Tuple<int, std::string> is = ...; +// aEW.WriteObject(is); // Write the Tuple, equivalent to: +// aEW.WriteObject(/* int */ std::get<0>(is), /* string */ std::get<1>(is)); +// ... +// // Reading back can be done directly into a Tuple: +// auto is = aER.ReadObject<Tuple<int, std::string>>(); +// // Or each item could be read separately: +// auto i = aER.ReadObject<int>(); auto s = aER.ReadObject<std::string>(); +// ``` +template <typename... Ts> +struct ProfileBufferEntryWriter::Serializer<Tuple<Ts...>> { + private: + template <size_t... Is> + static Length TupleBytes(const Tuple<Ts...>& aTuple, + std::index_sequence<Is...>) { + return (0 + ... + SumBytes(Get<Is>(aTuple))); + } + + template <size_t... Is> + static void TupleWrite(ProfileBufferEntryWriter& aEW, + const Tuple<Ts...>& aTuple, + std::index_sequence<Is...>) { + (aEW.WriteObject(Get<Is>(aTuple)), ...); + } + + public: + static Length Bytes(const Tuple<Ts...>& aTuple) { + // Generate a 0..N-1 index pack, we'll add the sizes of each item. + return TupleBytes(aTuple, std::index_sequence_for<Ts...>()); + } + + static void Write(ProfileBufferEntryWriter& aEW, const Tuple<Ts...>& aTuple) { + // Generate a 0..N-1 index pack, we'll write each item. + TupleWrite(aEW, aTuple, std::index_sequence_for<Ts...>()); + } +}; + +template <typename... Ts> +struct ProfileBufferEntryReader::Deserializer<Tuple<Ts...>> { + static void ReadInto(ProfileBufferEntryReader& aER, Tuple<Ts...>& aTuple) { + aER.ReadBytes(&aTuple, Bytes(aTuple)); + } + + static Tuple<Ts...> Read(ProfileBufferEntryReader& aER) { + // Note that this creates default `Ts` first, and then overwrites them. + Tuple<Ts...> ob; + ReadInto(aER, ob); + return ob; + } +}; + +// ---------------------------------------------------------------------------- +// mozilla::Span + +// Span. All elements are serialized in sequence. +// The caller is assumed to know the number of elements (they may manually +// write&read it before the span if needed). +// Similar to tuples, reading/writing spans is equivalent to reading/writing +// their elements in order. +template <class T, size_t N> +struct ProfileBufferEntryWriter::Serializer<Span<T, N>> { + static Length Bytes(const Span<T, N>& aSpan) { + Length bytes = 0; + for (const T& element : aSpan) { + bytes += SumBytes(element); + } + return bytes; + } + + static void Write(ProfileBufferEntryWriter& aEW, const Span<T, N>& aSpan) { + for (const T& element : aSpan) { + aEW.WriteObject(element); + } + } +}; + +template <class T, size_t N> +struct ProfileBufferEntryReader::Deserializer<Span<T, N>> { + // Read elements back into span pointing at a pre-allocated buffer. + static void ReadInto(ProfileBufferEntryReader& aER, Span<T, N>& aSpan) { + for (T& element : aSpan) { + aER.ReadIntoObject(element); + } + } + + // A Span does not own its data, this would probably leak so we forbid this. + static Span<T, N> Read(ProfileBufferEntryReader& aER) = delete; +}; + +// ---------------------------------------------------------------------------- +// mozilla::Maybe + +// Maybe<T> is serialized as one byte containing either 'm' (Nothing), +// or 'M' followed by the recursively-serialized `T` object. +template <typename T> +struct ProfileBufferEntryWriter::Serializer<Maybe<T>> { + static Length Bytes(const Maybe<T>& aMaybe) { + // 1 byte to store nothing/something flag, then object size if present. + return aMaybe.isNothing() ? 1 : (1 + SumBytes(aMaybe.ref())); + } + + static void Write(ProfileBufferEntryWriter& aEW, const Maybe<T>& aMaybe) { + // 'm'/'M' is just an arbitrary 1-byte value to distinguish states. + if (aMaybe.isNothing()) { + aEW.WriteObject<char>('m'); + } else { + aEW.WriteObject<char>('M'); + // Use the Serializer for the contained type. + aEW.WriteObject(aMaybe.ref()); + } + } +}; + +template <typename T> +struct ProfileBufferEntryReader::Deserializer<Maybe<T>> { + static void ReadInto(ProfileBufferEntryReader& aER, Maybe<T>& aMaybe) { + char c = aER.ReadObject<char>(); + if (c == 'm') { + aMaybe.reset(); + } else { + MOZ_ASSERT(c == 'M'); + // If aMaybe is empty, create a default `T` first, to be overwritten. + // Otherwise we'll just overwrite whatever was already there. + if (aMaybe.isNothing()) { + aMaybe.emplace(); + } + // Use the Deserializer for the contained type. + aER.ReadIntoObject(aMaybe.ref()); + } + } + + static Maybe<T> Read(ProfileBufferEntryReader& aER) { + Maybe<T> maybe; + char c = aER.ReadObject<char>(); + MOZ_ASSERT(c == 'M' || c == 'm'); + if (c == 'M') { + // Note that this creates a default `T` inside the Maybe first, and then + // overwrites it. + maybe = Some(T{}); + // Use the Deserializer for the contained type. + aER.ReadIntoObject(maybe.ref()); + } + return maybe; + } +}; + +// ---------------------------------------------------------------------------- +// mozilla::Variant + +// Variant is serialized as the tag (0-based index of the stored type, encoded +// as ULEB128), and the recursively-serialized object. +template <typename... Ts> +struct ProfileBufferEntryWriter::Serializer<Variant<Ts...>> { + public: + static Length Bytes(const Variant<Ts...>& aVariantTs) { + return aVariantTs.match([](auto aIndex, const auto& aAlternative) { + return ULEB128Size(aIndex) + SumBytes(aAlternative); + }); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const Variant<Ts...>& aVariantTs) { + aVariantTs.match([&aEW](auto aIndex, const auto& aAlternative) { + aEW.WriteULEB128(aIndex); + aEW.WriteObject(aAlternative); + }); + } +}; + +template <typename... Ts> +struct ProfileBufferEntryReader::Deserializer<Variant<Ts...>> { + private: + // Called from the fold expression in `VariantReadInto()`, only the selected + // variant will deserialize the object. + template <size_t I> + static void VariantIReadInto(ProfileBufferEntryReader& aER, + Variant<Ts...>& aVariantTs, unsigned aTag) { + if (I == aTag) { + // Ensure the variant contains the target type. Note that this may create + // a default object. + if (!aVariantTs.template is<I>()) { + aVariantTs = Variant<Ts...>(VariantIndex<I>{}); + } + aER.ReadIntoObject(aVariantTs.template as<I>()); + } + } + + template <size_t... Is> + static void VariantReadInto(ProfileBufferEntryReader& aER, + Variant<Ts...>& aVariantTs, + std::index_sequence<Is...>) { + unsigned tag = aER.ReadULEB128<unsigned>(); + (VariantIReadInto<Is>(aER, aVariantTs, tag), ...); + } + + public: + static void ReadInto(ProfileBufferEntryReader& aER, + Variant<Ts...>& aVariantTs) { + // Generate a 0..N-1 index pack, the selected variant will deserialize + // itself. + VariantReadInto(aER, aVariantTs, std::index_sequence_for<Ts...>()); + } + + static Variant<Ts...> Read(ProfileBufferEntryReader& aER) { + // Note that this creates a default `Variant` of the first type, and then + // overwrites it. Consider using `ReadInto` for more control if needed. + Variant<Ts...> variant(VariantIndex<0>{}); + ReadInto(aER, variant); + return variant; + } +}; + +} // namespace mozilla + +#endif // ProfileBufferEntrySerialization_h diff --git a/mozglue/baseprofiler/public/ProfileBufferIndex.h b/mozglue/baseprofiler/public/ProfileBufferIndex.h new file mode 100644 index 0000000000..5cda6bd89e --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileBufferIndex.h @@ -0,0 +1,97 @@ +/* -*- 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 ProfileBufferIndex_h +#define ProfileBufferIndex_h + +#include "mozilla/Attributes.h" + +#include <cstddef> +#include <cstdint> + +namespace mozilla { + +// Generic index into a Profiler buffer, mostly for internal usage. +// Intended to appear infinite (it should effectively never wrap). +// 0 (zero) is reserved as nullptr-like value; it may indicate failure result, +// or it may point at the earliest available block. +using ProfileBufferIndex = uint64_t; + +// Externally-opaque class encapsulating a block index, i.e. a +// ProfileBufferIndex that is guaranteed to point at the start of a Profile +// buffer block (until it is destroyed, but then that index cannot be reused and +// functions should gracefully handle expired blocks). +// Users may get these from Profile buffer functions, to later access previous +// blocks; they should avoid converting and operating on their value. +class ProfileBufferBlockIndex { + public: + // Default constructor with internal 0 value, for which Profile buffers must + // guarantee that it is before any valid entries; All public APIs should + // fail gracefully, doing and/or returning Nothing. + ProfileBufferBlockIndex() : mBlockIndex(0) {} + + // Implicit conversion from literal `nullptr` to internal 0 value, to allow + // convenient init/reset/comparison with 0 index. + MOZ_IMPLICIT ProfileBufferBlockIndex(std::nullptr_t) : mBlockIndex(0) {} + + // Explicit conversion to bool, works in `if` and other tests. + // Only returns false for default `ProfileBufferBlockIndex{}` value. + explicit operator bool() const { return mBlockIndex != 0; } + + // Comparison operators. Default `ProfileBufferBlockIndex{}` value is always + // the lowest. + [[nodiscard]] bool operator==(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex == aRhs.mBlockIndex; + } + [[nodiscard]] bool operator!=(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex != aRhs.mBlockIndex; + } + [[nodiscard]] bool operator<(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex < aRhs.mBlockIndex; + } + [[nodiscard]] bool operator<=(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex <= aRhs.mBlockIndex; + } + [[nodiscard]] bool operator>(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex > aRhs.mBlockIndex; + } + [[nodiscard]] bool operator>=(const ProfileBufferBlockIndex& aRhs) const { + return mBlockIndex >= aRhs.mBlockIndex; + } + + // Explicit conversion to ProfileBufferIndex, mostly used by internal Profile + // buffer code. + [[nodiscard]] ProfileBufferIndex ConvertToProfileBufferIndex() const { + return mBlockIndex; + } + + // Explicit creation from ProfileBufferIndex, mostly used by internal + // Profile buffer code. + [[nodiscard]] static ProfileBufferBlockIndex CreateFromProfileBufferIndex( + ProfileBufferIndex aIndex) { + return ProfileBufferBlockIndex(aIndex); + } + + private: + // Private to prevent easy construction from any value. Use + // `CreateFromProfileBufferIndex()` instead. + // The main reason for this indirection is to make it harder to create these + // objects, because only the profiler code should need to do it. Ideally, this + // class should be used wherever a block index should be stored, but there is + // so much code that uses `uint64_t` that it would be a big task to change + // them all. So for now we allow conversions to/from numbers, but it's as ugly + // as possible to make sure it doesn't get too common; and if one day we want + // to tackle a global change, it should be easy to find all these locations + // thanks to the explicit conversion functions. + explicit ProfileBufferBlockIndex(ProfileBufferIndex aBlockIndex) + : mBlockIndex(aBlockIndex) {} + + ProfileBufferIndex mBlockIndex; +}; + +} // namespace mozilla + +#endif // ProfileBufferIndex_h diff --git a/mozglue/baseprofiler/public/ProfileChunkedBuffer.h b/mozglue/baseprofiler/public/ProfileChunkedBuffer.h new file mode 100644 index 0000000000..d4d55eafcb --- /dev/null +++ b/mozglue/baseprofiler/public/ProfileChunkedBuffer.h @@ -0,0 +1,1872 @@ +/* -*- 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 ProfileChunkedBuffer_h +#define ProfileChunkedBuffer_h + +#include "mozilla/BaseProfilerDetail.h" +#include "mozilla/NotNull.h" +#include "mozilla/ProfileBufferChunkManager.h" +#include "mozilla/ProfileBufferChunkManagerSingle.h" +#include "mozilla/ProfileBufferEntrySerialization.h" +#include "mozilla/RefCounted.h" +#include "mozilla/RefPtr.h" +#include "mozilla/ScopeExit.h" +#include "mozilla/Unused.h" + +#include <cstdio> +#include <utility> + +namespace mozilla { + +namespace detail { + +// Internal accessor pointing at a position inside a chunk. +// It can handle two groups of chunks (typically the extant chunks stored in +// the store manager, and the current chunk). +// The main operations are: +// - ReadEntrySize() to read an entry size, 0 means failure. +// - operator+=(Length) to skip a number of bytes. +// - EntryReader() creates an entry reader at the current position for a given +// size (it may fail with an empty reader), and skips the entry. +// Note that there is no "past-the-end" position -- as soon as InChunkPointer +// reaches the end, it becomes effectively null. +class InChunkPointer { + public: + using Byte = ProfileBufferChunk::Byte; + using Length = ProfileBufferChunk::Length; + + // Nullptr-like InChunkPointer, may be used as end iterator. + InChunkPointer() + : mChunk(nullptr), mNextChunkGroup(nullptr), mOffsetInChunk(0) {} + + // InChunkPointer over one or two chunk groups, pointing at the given + // block index (if still in range). + // This constructor should only be used with *trusted* block index values! + InChunkPointer(const ProfileBufferChunk* aChunk, + const ProfileBufferChunk* aNextChunkGroup, + ProfileBufferBlockIndex aBlockIndex) + : mChunk(aChunk), mNextChunkGroup(aNextChunkGroup) { + if (mChunk) { + mOffsetInChunk = mChunk->OffsetFirstBlock(); + Adjust(); + } else if (mNextChunkGroup) { + mChunk = mNextChunkGroup; + mNextChunkGroup = nullptr; + mOffsetInChunk = mChunk->OffsetFirstBlock(); + Adjust(); + } else { + mOffsetInChunk = 0; + } + + // Try to advance to given position. + if (!AdvanceToGlobalRangePosition(aBlockIndex)) { + // Block does not exist anymore (or block doesn't look valid), reset the + // in-chunk pointer. + mChunk = nullptr; + mNextChunkGroup = nullptr; + } + } + + // InChunkPointer over one or two chunk groups, will start at the first + // block (if any). This may be slow, so avoid using it too much. + InChunkPointer(const ProfileBufferChunk* aChunk, + const ProfileBufferChunk* aNextChunkGroup, + ProfileBufferIndex aIndex = ProfileBufferIndex(0)) + : mChunk(aChunk), mNextChunkGroup(aNextChunkGroup) { + if (mChunk) { + mOffsetInChunk = mChunk->OffsetFirstBlock(); + Adjust(); + } else if (mNextChunkGroup) { + mChunk = mNextChunkGroup; + mNextChunkGroup = nullptr; + mOffsetInChunk = mChunk->OffsetFirstBlock(); + Adjust(); + } else { + mOffsetInChunk = 0; + } + + // Try to advance to given position. + if (!AdvanceToGlobalRangePosition(aIndex)) { + // Block does not exist anymore, reset the in-chunk pointer. + mChunk = nullptr; + mNextChunkGroup = nullptr; + } + } + + // Compute the current position in the global range. + // 0 if null (including if we're reached the end). + [[nodiscard]] ProfileBufferIndex GlobalRangePosition() const { + if (IsNull()) { + return 0; + } + return mChunk->RangeStart() + mOffsetInChunk; + } + + // Move InChunkPointer forward to the block at the given global block + // position, which is assumed to be valid exactly -- but it may be obsolete. + // 0 stays where it is (if valid already). + // MOZ_ASSERTs if the index is invalid. + [[nodiscard]] bool AdvanceToGlobalRangePosition( + ProfileBufferBlockIndex aBlockIndex) { + if (IsNull()) { + // Pointer is null already. (Not asserting because it's acceptable.) + return false; + } + if (!aBlockIndex) { + // Special null position, just stay where we are. + return ShouldPointAtValidBlock(); + } + if (aBlockIndex.ConvertToProfileBufferIndex() < GlobalRangePosition()) { + // Past the requested position, stay where we are (assuming the current + // position was valid). + return ShouldPointAtValidBlock(); + } + for (;;) { + if (aBlockIndex.ConvertToProfileBufferIndex() < + mChunk->RangeStart() + mChunk->OffsetPastLastBlock()) { + // Target position is in this chunk's written space, move to it. + mOffsetInChunk = + aBlockIndex.ConvertToProfileBufferIndex() - mChunk->RangeStart(); + return ShouldPointAtValidBlock(); + } + // Position is after this chunk, try next chunk. + GoToNextChunk(); + if (IsNull()) { + return false; + } + // Skip whatever block tail there is, we don't allow pointing in the + // middle of a block. + mOffsetInChunk = mChunk->OffsetFirstBlock(); + if (aBlockIndex.ConvertToProfileBufferIndex() < GlobalRangePosition()) { + // Past the requested position, meaning that the given position was in- + // between blocks -> Failure. + MOZ_ASSERT(false, "AdvanceToGlobalRangePosition - In-between blocks"); + return false; + } + } + } + + // Move InChunkPointer forward to the block at or after the given global + // range position. + // 0 stays where it is (if valid already). + [[nodiscard]] bool AdvanceToGlobalRangePosition( + ProfileBufferIndex aPosition) { + if (aPosition == 0) { + // Special position '0', just stay where we are. + // Success if this position is already valid. + return !IsNull(); + } + for (;;) { + ProfileBufferIndex currentPosition = GlobalRangePosition(); + if (currentPosition == 0) { + // Pointer is null. + return false; + } + if (aPosition <= currentPosition) { + // At or past the requested position, stay where we are. + return true; + } + if (aPosition < mChunk->RangeStart() + mChunk->OffsetPastLastBlock()) { + // Target position is in this chunk's written space, move to it. + for (;;) { + // Skip the current block. + mOffsetInChunk += ReadEntrySize(); + if (mOffsetInChunk >= mChunk->OffsetPastLastBlock()) { + // Reached the end of the chunk, this can happen for the last + // block, let's just continue to the next chunk. + break; + } + if (aPosition <= mChunk->RangeStart() + mOffsetInChunk) { + // We're at or after the position, return at this block position. + return true; + } + } + } + // Position is after this chunk, try next chunk. + GoToNextChunk(); + if (IsNull()) { + return false; + } + // Skip whatever block tail there is, we don't allow pointing in the + // middle of a block. + mOffsetInChunk = mChunk->OffsetFirstBlock(); + } + } + + [[nodiscard]] Byte ReadByte() { + MOZ_ASSERT(!IsNull()); + MOZ_ASSERT(mOffsetInChunk < mChunk->OffsetPastLastBlock()); + Byte byte = mChunk->ByteAt(mOffsetInChunk); + if (MOZ_UNLIKELY(++mOffsetInChunk == mChunk->OffsetPastLastBlock())) { + Adjust(); + } + return byte; + } + + // Read and skip a ULEB128-encoded size. + // 0 means failure (0-byte entries are not allowed.) + // Note that this doesn't guarantee that there are actually that many bytes + // available to read! (EntryReader() below may gracefully fail.) + [[nodiscard]] Length ReadEntrySize() { + ULEB128Reader<Length> reader; + if (IsNull()) { + return 0; + } + for (;;) { + const bool isComplete = reader.FeedByteIsComplete(ReadByte()); + if (MOZ_UNLIKELY(IsNull())) { + // End of chunks, so there's no actual entry after this anyway. + return 0; + } + if (MOZ_LIKELY(isComplete)) { + if (MOZ_UNLIKELY(reader.Value() > mChunk->BufferBytes())) { + // Don't allow entries larger than a chunk. + return 0; + } + return reader.Value(); + } + } + } + + InChunkPointer& operator+=(Length aLength) { + MOZ_ASSERT(!IsNull()); + mOffsetInChunk += aLength; + Adjust(); + return *this; + } + + [[nodiscard]] ProfileBufferEntryReader EntryReader(Length aLength) { + if (IsNull() || aLength == 0) { + return ProfileBufferEntryReader(); + } + + MOZ_ASSERT(mOffsetInChunk < mChunk->OffsetPastLastBlock()); + + // We should be pointing at the entry, past the entry size. + const ProfileBufferIndex entryIndex = GlobalRangePosition(); + // Verify that there's enough space before for the size (starting at index + // 1 at least). + MOZ_ASSERT(entryIndex >= 1u + ULEB128Size(aLength)); + + const Length remaining = mChunk->OffsetPastLastBlock() - mOffsetInChunk; + Span<const Byte> mem0 = mChunk->BufferSpan(); + mem0 = mem0.From(mOffsetInChunk); + if (aLength <= remaining) { + // Move to the end of this block, which could make this null if we have + // reached the end of all buffers. + *this += aLength; + return ProfileBufferEntryReader( + mem0.To(aLength), + // Block starts before the entry size. + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + entryIndex - ULEB128Size(aLength)), + // Block ends right after the entry (could be null for last entry). + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + GlobalRangePosition())); + } + + // We need to go to the next chunk for the 2nd part of this block. + GoToNextChunk(); + if (IsNull()) { + return ProfileBufferEntryReader(); + } + + Span<const Byte> mem1 = mChunk->BufferSpan(); + const Length tail = aLength - remaining; + MOZ_ASSERT(tail <= mChunk->BufferBytes()); + MOZ_ASSERT(tail == mChunk->OffsetFirstBlock()); + // We are in the correct chunk, move the offset to the end of the block. + mOffsetInChunk = tail; + // And adjust as needed, which could make this null if we have reached the + // end of all buffers. + Adjust(); + return ProfileBufferEntryReader( + mem0, mem1.To(tail), + // Block starts before the entry size. + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + entryIndex - ULEB128Size(aLength)), + // Block ends right after the entry (could be null for last entry). + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + GlobalRangePosition())); + } + + [[nodiscard]] bool IsNull() const { return !mChunk; } + + [[nodiscard]] bool operator==(const InChunkPointer& aOther) const { + if (IsNull() || aOther.IsNull()) { + return IsNull() && aOther.IsNull(); + } + return mChunk == aOther.mChunk && mOffsetInChunk == aOther.mOffsetInChunk; + } + + [[nodiscard]] bool operator!=(const InChunkPointer& aOther) const { + return !(*this == aOther); + } + + [[nodiscard]] Byte operator*() const { + MOZ_ASSERT(!IsNull()); + MOZ_ASSERT(mOffsetInChunk < mChunk->OffsetPastLastBlock()); + return mChunk->ByteAt(mOffsetInChunk); + } + + InChunkPointer& operator++() { + MOZ_ASSERT(!IsNull()); + MOZ_ASSERT(mOffsetInChunk < mChunk->OffsetPastLastBlock()); + if (MOZ_UNLIKELY(++mOffsetInChunk == mChunk->OffsetPastLastBlock())) { + mOffsetInChunk = 0; + GoToNextChunk(); + Adjust(); + } + return *this; + } + + private: + void GoToNextChunk() { + MOZ_ASSERT(!IsNull()); + const ProfileBufferIndex expectedNextRangeStart = + mChunk->RangeStart() + mChunk->BufferBytes(); + + mChunk = mChunk->GetNext(); + if (!mChunk) { + // Reached the end of the current chunk group, try the next one (which + // may be null too, especially on the 2nd try). + mChunk = mNextChunkGroup; + mNextChunkGroup = nullptr; + } + + if (mChunk && mChunk->RangeStart() == 0) { + // Reached a chunk without a valid (non-null) range start, assume there + // are only unused chunks from here on. + mChunk = nullptr; + } + + MOZ_ASSERT(!mChunk || mChunk->RangeStart() == expectedNextRangeStart, + "We don't handle discontinuous buffers (yet)"); + // Non-DEBUG fallback: Stop reading past discontinuities. + // (They should be rare, only happening on temporary OOMs.) + // TODO: Handle discontinuities (by skipping over incomplete blocks). + if (mChunk && mChunk->RangeStart() != expectedNextRangeStart) { + mChunk = nullptr; + } + } + + // We want `InChunkPointer` to always point at a valid byte (or be null). + // After some operations, `mOffsetInChunk` may point past the end of the + // current `mChunk`, in which case we need to adjust our position to be inside + // the appropriate chunk. E.g., if we're 10 bytes after the end of the current + // chunk, we should end up at offset 10 in the next chunk. + // Note that we may "fall off" the last chunk and make this `InChunkPointer` + // effectively null. + void Adjust() { + while (mChunk && mOffsetInChunk >= mChunk->OffsetPastLastBlock()) { + // TODO: Try to adjust offset between chunks relative to mRangeStart + // differences. But we don't handle discontinuities yet. + if (mOffsetInChunk < mChunk->BufferBytes()) { + mOffsetInChunk -= mChunk->BufferBytes(); + } else { + mOffsetInChunk -= mChunk->OffsetPastLastBlock(); + } + GoToNextChunk(); + } + } + + // Check if the current position is likely to point at a valid block. + // (Size should be reasonable, and block should fully fit inside buffer.) + // MOZ_ASSERTs on failure, to catch incorrect uses of block indices (which + // should only point at valid blocks if still in range). Non-asserting build + // fallback should still be handled. + [[nodiscard]] bool ShouldPointAtValidBlock() const { + if (IsNull()) { + // Pointer is null, no blocks here. + MOZ_ASSERT(false, "ShouldPointAtValidBlock - null pointer"); + return false; + } + // Use a copy, so we don't modify `*this`. + InChunkPointer pointer = *this; + // Try to read the entry size. + Length entrySize = pointer.ReadEntrySize(); + if (entrySize == 0) { + // Entry size of zero means we read 0 or a way-too-big value. + MOZ_ASSERT(false, "ShouldPointAtValidBlock - invalid size"); + return false; + } + // See if the last byte of the entry is still inside the buffer. + pointer += entrySize - 1; + MOZ_ASSERT(!IsNull(), "ShouldPointAtValidBlock - past end of buffer"); + return !IsNull(); + } + + const ProfileBufferChunk* mChunk; + const ProfileBufferChunk* mNextChunkGroup; + Length mOffsetInChunk; +}; + +} // namespace detail + +// Thread-safe buffer that can store blocks of different sizes during defined +// sessions, using Chunks (from a ChunkManager) as storage. +// +// Each *block* contains an *entry* and the entry size: +// [ entry_size | entry ] [ entry_size | entry ] ... +// +// *In-session* is a period of time during which `ProfileChunkedBuffer` allows +// reading and writing. +// *Out-of-session*, the `ProfileChunkedBuffer` object is still valid, but +// contains no data, and gracefully denies accesses. +// +// To write an entry, the buffer reserves a block of sufficient size (to contain +// user data of predetermined size), writes the entry size, and lets the caller +// fill the entry contents using a ProfileBufferEntryWriter. E.g.: +// ``` +// ProfileChunkedBuffer cb(...); +// cb.ReserveAndPut([]() { return sizeof(123); }, +// [&](Maybe<ProfileBufferEntryWriter>& aEW) { +// if (aEW) { aEW->WriteObject(123); } +// }); +// ``` +// Other `Put...` functions may be used as shortcuts for simple entries. +// The objects given to the caller's callbacks should only be used inside the +// callbacks and not stored elsewhere, because they keep their own references to +// chunk memory and therefore should not live longer. +// Different type of objects may be serialized into an entry, see +// `ProfileBufferEntryWriter::Serializer` for more information. +// +// When reading data, the buffer iterates over blocks (it knows how to read the +// entry size, and therefore move to the next block), and lets the caller read +// the entry inside of each block. E.g.: +// ``` +// cb.ReadEach([](ProfileBufferEntryReader& aER) { +// /* Use ProfileBufferEntryReader functions to read serialized objects. */ +// int n = aER.ReadObject<int>(); +// }); +// ``` +// Different type of objects may be deserialized from an entry, see +// `ProfileBufferEntryReader::Deserializer` for more information. +// +// Writers may retrieve the block index corresponding to an entry +// (`ProfileBufferBlockIndex` is an opaque type preventing the user from easily +// modifying it). That index may later be used with `ReadAt` to get back to the +// entry in that particular block -- if it still exists. +class ProfileChunkedBuffer { + public: + using Byte = ProfileBufferChunk::Byte; + using Length = ProfileBufferChunk::Length; + + enum class ThreadSafety { WithoutMutex, WithMutex }; + + // Default constructor starts out-of-session (nothing to read or write). + explicit ProfileChunkedBuffer(ThreadSafety aThreadSafety) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex) {} + + // Start in-session with external chunk manager. + ProfileChunkedBuffer(ThreadSafety aThreadSafety, + ProfileBufferChunkManager& aChunkManager) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex) { + SetChunkManager(aChunkManager); + } + + // Start in-session with owned chunk manager. + ProfileChunkedBuffer(ThreadSafety aThreadSafety, + UniquePtr<ProfileBufferChunkManager>&& aChunkManager) + : mMutex(aThreadSafety != ThreadSafety::WithoutMutex) { + SetChunkManager(std::move(aChunkManager)); + } + + ~ProfileChunkedBuffer() { + // Do proper clean-up by resetting the chunk manager. + ResetChunkManager(); + } + + // This cannot change during the lifetime of this buffer, so there's no need + // to lock. + [[nodiscard]] bool IsThreadSafe() const { return mMutex.IsActivated(); } + + [[nodiscard]] bool IsInSession() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return !!mChunkManager; + } + + // Stop using the current chunk manager. + // If we own the current chunk manager, it will be destroyed. + // This will always clear currently-held chunks, if any. + void ResetChunkManager() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + Unused << ResetChunkManager(lock); + } + + // Set the current chunk manager. + // The caller is responsible for keeping the chunk manager alive as along as + // it's used here (until the next (Re)SetChunkManager, or + // ~ProfileChunkedBuffer). + void SetChunkManager(ProfileBufferChunkManager& aChunkManager) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + Unused << ResetChunkManager(lock); + SetChunkManager(aChunkManager, lock); + } + + // Set the current chunk manager, and keep ownership of it. + void SetChunkManager(UniquePtr<ProfileBufferChunkManager>&& aChunkManager) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + Unused << ResetChunkManager(lock); + mOwnedChunkManager = std::move(aChunkManager); + if (mOwnedChunkManager) { + SetChunkManager(*mOwnedChunkManager, lock); + } + } + + // Stop using the current chunk manager, and return it if owned here. + [[nodiscard]] UniquePtr<ProfileBufferChunkManager> ExtractChunkManager() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return ResetChunkManager(lock); + } + + // Clear the contents of this buffer, ready to receive new chunks. + // Note that memory is not freed: No chunks are destroyed, they are all + // receycled. + // Also the range doesn't reset, instead it continues at some point after the + // previous range. This may be useful if the caller may be keeping indexes + // into old chunks that have now been cleared, using these indexes will fail + // gracefully (instead of potentially pointing into new data). + void Clear() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return; + } + + mRangeStart = mRangeEnd = mNextChunkRangeStart; + mPushedBlockCount = 0; + mClearedBlockCount = 0; + mFailedPutBytes = 0; + + // Recycle all released chunks as "next" chunks. This will reduce the number + // of future allocations. Also, when using ProfileBufferChunkManagerSingle, + // this retrieves the one chunk if it was released. + UniquePtr<ProfileBufferChunk> releasedChunks = + mChunkManager->GetExtantReleasedChunks(); + if (releasedChunks) { + // Released chunks should be in the "Done" state, they need to be marked + // "recycled" before they can be reused. + for (ProfileBufferChunk* chunk = releasedChunks.get(); chunk; + chunk = chunk->GetNext()) { + chunk->MarkRecycled(); + } + mNextChunks = ProfileBufferChunk::Join(std::move(mNextChunks), + std::move(releasedChunks)); + } + + if (mCurrentChunk) { + // We already have a current chunk (empty or in-use), mark it "done" and + // then "recycled", ready to be reused. + mCurrentChunk->MarkDone(); + mCurrentChunk->MarkRecycled(); + } else { + if (!mNextChunks) { + // No current chunk, and no next chunks to recycle, nothing more to do. + // The next "Put" operation will try to allocate a chunk as needed. + return; + } + + // No current chunk, take a next chunk. + mCurrentChunk = std::exchange(mNextChunks, mNextChunks->ReleaseNext()); + } + + // Here, there was already a current chunk, or one has just been taken. + // Make sure it's ready to receive new entries. + InitializeCurrentChunk(lock); + } + + // Buffer maximum length in bytes. + Maybe<size_t> BufferLength() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (!mChunkManager) { + return Nothing{}; + } + return Some(mChunkManager->MaxTotalSize()); + } + + [[nodiscard]] size_t SizeOfExcludingThis(MallocSizeOf aMallocSizeOf) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return SizeOfExcludingThis(aMallocSizeOf, lock); + } + + [[nodiscard]] size_t SizeOfIncludingThis(MallocSizeOf aMallocSizeOf) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return aMallocSizeOf(this) + SizeOfExcludingThis(aMallocSizeOf, lock); + } + + // Snapshot of the buffer state. + struct State { + // Index to/before the first block. + ProfileBufferIndex mRangeStart = 1; + + // Index past the last block. Equals mRangeStart if empty. + ProfileBufferIndex mRangeEnd = 1; + + // Number of blocks that have been pushed into this buffer. + uint64_t mPushedBlockCount = 0; + + // Number of blocks that have been removed from this buffer. + // Note: Live entries = pushed - cleared. + uint64_t mClearedBlockCount = 0; + + // Number of bytes that could not be put into this buffer. + uint64_t mFailedPutBytes = 0; + }; + + // Get a snapshot of the current state. + // When out-of-session, mFirstReadIndex==mNextWriteIndex, and + // mPushedBlockCount==mClearedBlockCount==0. + // Note that these may change right after this thread-safe call, so they + // should only be used for statistical purposes. + [[nodiscard]] State GetState() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return {mRangeStart, mRangeEnd, mPushedBlockCount, mClearedBlockCount, + mFailedPutBytes}; + } + + [[nodiscard]] bool IsEmpty() const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return mRangeStart == mRangeEnd; + } + + // True if this buffer is already locked on this thread. + // This should be used if some functions may call an already-locked buffer, + // e.g.: Put -> memory hook -> profiler_add_native_allocation_marker -> Put. + [[nodiscard]] bool IsThreadSafeAndLockedOnCurrentThread() const { + return mMutex.IsActivatedAndLockedOnCurrentThread(); + } + + // Lock the buffer mutex and run the provided callback. + // This can be useful when the caller needs to explicitly lock down this + // buffer, but not do anything else with it. + template <typename Callback> + auto LockAndRun(Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return std::forward<Callback>(aCallback)(); + } + + // Reserve a block that can hold an entry of the given `aCallbackEntryBytes()` + // size, write the entry size (ULEB128-encoded), and invoke and return + // `aCallback(Maybe<ProfileBufferEntryWriter>&)`. + // Note: `aCallbackEntryBytes` is a callback instead of a simple value, to + // delay this potentially-expensive computation until after we're checked that + // we're in-session; use `Put(Length, Callback)` below if you know the size + // already. + template <typename CallbackEntryBytes, typename Callback> + auto ReserveAndPut(CallbackEntryBytes&& aCallbackEntryBytes, + Callback&& aCallback) + -> decltype(std::forward<Callback>(aCallback)( + std::declval<Maybe<ProfileBufferEntryWriter>&>())) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + + // This can only be read in the 2nd lambda below after it has been written + // by the first lambda. + Length entryBytes; + + return ReserveAndPutRaw( + [&]() { + entryBytes = std::forward<CallbackEntryBytes>(aCallbackEntryBytes)(); + MOZ_ASSERT(entryBytes != 0, "Empty entries are not allowed"); + return ULEB128Size(entryBytes) + entryBytes; + }, + [&](Maybe<ProfileBufferEntryWriter>& aMaybeEntryWriter) { + if (aMaybeEntryWriter.isSome()) { + aMaybeEntryWriter->WriteULEB128(entryBytes); + MOZ_ASSERT(aMaybeEntryWriter->RemainingBytes() == entryBytes); + } + return std::forward<Callback>(aCallback)(aMaybeEntryWriter); + }, + lock); + } + + template <typename Callback> + auto Put(Length aEntryBytes, Callback&& aCallback) { + return ReserveAndPut([aEntryBytes]() { return aEntryBytes; }, + std::forward<Callback>(aCallback)); + } + + // Add a new entry copied from the given buffer, return block index. + ProfileBufferBlockIndex PutFrom(const void* aSrc, Length aBytes) { + return ReserveAndPut( + [aBytes]() { return aBytes; }, + [aSrc, aBytes](Maybe<ProfileBufferEntryWriter>& aMaybeEntryWriter) { + if (aMaybeEntryWriter.isNothing()) { + return ProfileBufferBlockIndex{}; + } + aMaybeEntryWriter->WriteBytes(aSrc, aBytes); + return aMaybeEntryWriter->CurrentBlockIndex(); + }); + } + + // Add a new single entry with *all* given object (using a Serializer for + // each), return block index. + template <typename... Ts> + ProfileBufferBlockIndex PutObjects(const Ts&... aTs) { + static_assert(sizeof...(Ts) > 0, + "PutObjects must be given at least one object."); + return ReserveAndPut( + [&]() { return ProfileBufferEntryWriter::SumBytes(aTs...); }, + [&](Maybe<ProfileBufferEntryWriter>& aMaybeEntryWriter) { + if (aMaybeEntryWriter.isNothing()) { + return ProfileBufferBlockIndex{}; + } + aMaybeEntryWriter->WriteObjects(aTs...); + return aMaybeEntryWriter->CurrentBlockIndex(); + }); + } + + // Add a new entry copied from the given object, return block index. + template <typename T> + ProfileBufferBlockIndex PutObject(const T& aOb) { + return PutObjects(aOb); + } + + // Get *all* chunks related to this buffer, including extant chunks in its + // ChunkManager, and yet-unused new/recycled chunks. + // We don't expect this buffer to be used again, though it's still possible + // and will allocate the first buffer when needed. + [[nodiscard]] UniquePtr<ProfileBufferChunk> GetAllChunks() { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return nullptr; + } + UniquePtr<ProfileBufferChunk> chunks = + mChunkManager->GetExtantReleasedChunks(); + Unused << HandleRequestedChunk_IsPending(lock); + if (MOZ_LIKELY(!!mCurrentChunk)) { + mCurrentChunk->MarkDone(); + chunks = + ProfileBufferChunk::Join(std::move(chunks), std::move(mCurrentChunk)); + } + chunks = + ProfileBufferChunk::Join(std::move(chunks), std::move(mNextChunks)); + mChunkManager->ForgetUnreleasedChunks(); + mRangeStart = mRangeEnd = mNextChunkRangeStart; + return chunks; + } + + class Reader; + + // Class that can iterate through blocks and provide + // `ProfileBufferEntryReader`s. + // Created through `Reader`, lives within a lock guard lifetime. + class BlockIterator { + public: +#ifdef DEBUG + ~BlockIterator() { + // No BlockIterator should live outside of a mutexed call. + mBuffer->mMutex.AssertCurrentThreadOwns(); + } +#endif // DEBUG + + // Comparison with other iterator, mostly used in range-for loops. + [[nodiscard]] bool operator==(const BlockIterator& aRhs) const { + MOZ_ASSERT(mBuffer == aRhs.mBuffer); + return mCurrentBlockIndex == aRhs.mCurrentBlockIndex; + } + [[nodiscard]] bool operator!=(const BlockIterator& aRhs) const { + MOZ_ASSERT(mBuffer == aRhs.mBuffer); + return mCurrentBlockIndex != aRhs.mCurrentBlockIndex; + } + + // Advance to next BlockIterator. + BlockIterator& operator++() { + mBuffer->mMutex.AssertCurrentThreadOwns(); + mCurrentBlockIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mNextBlockPointer.GlobalRangePosition()); + mCurrentEntry = + mNextBlockPointer.EntryReader(mNextBlockPointer.ReadEntrySize()); + return *this; + } + + // Dereferencing creates a `ProfileBufferEntryReader` object for the entry + // inside this block. + // (Note: It would be possible to return a `const + // ProfileBufferEntryReader&`, but not useful in practice, because in most + // case the user will want to read, which is non-const.) + [[nodiscard]] ProfileBufferEntryReader operator*() const { + return mCurrentEntry; + } + + // True if this iterator is just past the last entry. + [[nodiscard]] bool IsAtEnd() const { + return mCurrentEntry.RemainingBytes() == 0; + } + + // Can be used as reference to come back to this entry with `GetEntryAt()`. + [[nodiscard]] ProfileBufferBlockIndex CurrentBlockIndex() const { + return mCurrentBlockIndex; + } + + // Index past the end of this block, which is the start of the next block. + [[nodiscard]] ProfileBufferBlockIndex NextBlockIndex() const { + MOZ_ASSERT(!IsAtEnd()); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mNextBlockPointer.GlobalRangePosition()); + } + + // Index of the first block in the whole buffer. + [[nodiscard]] ProfileBufferBlockIndex BufferRangeStart() const { + mBuffer->mMutex.AssertCurrentThreadOwns(); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mBuffer->mRangeStart); + } + + // Index past the last block in the whole buffer. + [[nodiscard]] ProfileBufferBlockIndex BufferRangeEnd() const { + mBuffer->mMutex.AssertCurrentThreadOwns(); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mBuffer->mRangeEnd); + } + + private: + // Only a Reader can instantiate a BlockIterator. + friend class Reader; + + BlockIterator(const ProfileChunkedBuffer& aBuffer, + const ProfileBufferChunk* aChunks0, + const ProfileBufferChunk* aChunks1, + ProfileBufferBlockIndex aBlockIndex) + : mNextBlockPointer(aChunks0, aChunks1, aBlockIndex), + mCurrentBlockIndex( + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mNextBlockPointer.GlobalRangePosition())), + mCurrentEntry( + mNextBlockPointer.EntryReader(mNextBlockPointer.ReadEntrySize())), + mBuffer(WrapNotNull(&aBuffer)) { + // No BlockIterator should live outside of a mutexed call. + mBuffer->mMutex.AssertCurrentThreadOwns(); + } + + detail::InChunkPointer mNextBlockPointer; + + ProfileBufferBlockIndex mCurrentBlockIndex; + + ProfileBufferEntryReader mCurrentEntry; + + // Using a non-null pointer instead of a reference, to allow copying. + // This BlockIterator should only live inside one of the thread-safe + // ProfileChunkedBuffer functions, for this reference to stay valid. + NotNull<const ProfileChunkedBuffer*> mBuffer; + }; + + // Class that can create `BlockIterator`s (e.g., for range-for), or just + // iterate through entries; lives within a lock guard lifetime. + class MOZ_RAII Reader { + public: + Reader(const Reader&) = delete; + Reader& operator=(const Reader&) = delete; + Reader(Reader&&) = delete; + Reader& operator=(Reader&&) = delete; + +#ifdef DEBUG + ~Reader() { + // No Reader should live outside of a mutexed call. + mBuffer.mMutex.AssertCurrentThreadOwns(); + } +#endif // DEBUG + + // Index of the first block in the whole buffer. + [[nodiscard]] ProfileBufferBlockIndex BufferRangeStart() const { + mBuffer.mMutex.AssertCurrentThreadOwns(); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mBuffer.mRangeStart); + } + + // Index past the last block in the whole buffer. + [[nodiscard]] ProfileBufferBlockIndex BufferRangeEnd() const { + mBuffer.mMutex.AssertCurrentThreadOwns(); + return ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + mBuffer.mRangeEnd); + } + + // Iterators to the first and past-the-last blocks. + // Compatible with range-for (see `ForEach` below as example). + [[nodiscard]] BlockIterator begin() const { + return BlockIterator(mBuffer, mChunks0, mChunks1, nullptr); + } + // Note that a `BlockIterator` at the `end()` should not be dereferenced, as + // there is no actual block there! + [[nodiscard]] BlockIterator end() const { + return BlockIterator(mBuffer, nullptr, nullptr, nullptr); + } + + // Get a `BlockIterator` at the given `ProfileBufferBlockIndex`, clamped to + // the stored range. Note that a `BlockIterator` at the `end()` should not + // be dereferenced, as there is no actual block there! + [[nodiscard]] BlockIterator At(ProfileBufferBlockIndex aBlockIndex) const { + if (aBlockIndex < BufferRangeStart()) { + // Anything before the range (including null ProfileBufferBlockIndex) is + // clamped at the beginning. + return begin(); + } + // Otherwise we at least expect the index to be valid (pointing exactly at + // a live block, or just past the end.) + return BlockIterator(mBuffer, mChunks0, mChunks1, aBlockIndex); + } + + // Run `aCallback(ProfileBufferEntryReader&)` on each entry from first to + // last. Callback should not store `ProfileBufferEntryReader`, as it may + // become invalid after this thread-safe call. + template <typename Callback> + void ForEach(Callback&& aCallback) const { + for (ProfileBufferEntryReader reader : *this) { + aCallback(reader); + } + } + + // If this reader only points at one chunk with some data, this data will be + // exposed as a single entry. + [[nodiscard]] ProfileBufferEntryReader SingleChunkDataAsEntry() { + const ProfileBufferChunk* onlyNonEmptyChunk = nullptr; + for (const ProfileBufferChunk* chunkList : {mChunks0, mChunks1}) { + for (const ProfileBufferChunk* chunk = chunkList; chunk; + chunk = chunk->GetNext()) { + if (chunk->OffsetFirstBlock() != chunk->OffsetPastLastBlock()) { + if (onlyNonEmptyChunk) { + // More than one non-empty chunk. + return ProfileBufferEntryReader(); + } + onlyNonEmptyChunk = chunk; + } + } + } + if (!onlyNonEmptyChunk) { + // No non-empty chunks. + return ProfileBufferEntryReader(); + } + // Here, we have found one chunk that had some data. + return ProfileBufferEntryReader( + onlyNonEmptyChunk->BufferSpan().FromTo( + onlyNonEmptyChunk->OffsetFirstBlock(), + onlyNonEmptyChunk->OffsetPastLastBlock()), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + onlyNonEmptyChunk->RangeStart()), + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + onlyNonEmptyChunk->RangeStart() + + (onlyNonEmptyChunk->OffsetPastLastBlock() - + onlyNonEmptyChunk->OffsetFirstBlock()))); + } + + private: + friend class ProfileChunkedBuffer; + + explicit Reader(const ProfileChunkedBuffer& aBuffer, + const ProfileBufferChunk* aChunks0, + const ProfileBufferChunk* aChunks1) + : mBuffer(aBuffer), mChunks0(aChunks0), mChunks1(aChunks1) { + // No Reader should live outside of a mutexed call. + mBuffer.mMutex.AssertCurrentThreadOwns(); + } + + // This Reader should only live inside one of the thread-safe + // ProfileChunkedBuffer functions, for this reference to stay valid. + const ProfileChunkedBuffer& mBuffer; + const ProfileBufferChunk* mChunks0; + const ProfileBufferChunk* mChunks1; + }; + + // In in-session, call `aCallback(ProfileChunkedBuffer::Reader&)` and return + // true. Callback should not store `Reader`, because it may become invalid + // after this call. + // If out-of-session, return false (callback is not invoked). + template <typename Callback> + [[nodiscard]] auto Read(Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return std::forward<Callback>(aCallback)(static_cast<Reader*>(nullptr)); + } + return mChunkManager->PeekExtantReleasedChunks( + [&](const ProfileBufferChunk* aOldestChunk) { + Reader reader(*this, aOldestChunk, mCurrentChunk.get()); + return std::forward<Callback>(aCallback)(&reader); + }); + } + + // Invoke `aCallback(ProfileBufferEntryReader& [, ProfileBufferBlockIndex])` + // on each entry, it must read or at least skip everything. Either/both chunk + // pointers may be null. + template <typename Callback> + static void ReadEach(const ProfileBufferChunk* aChunks0, + const ProfileBufferChunk* aChunks1, + Callback&& aCallback) { + static_assert(std::is_invocable_v<Callback, ProfileBufferEntryReader&> || + std::is_invocable_v<Callback, ProfileBufferEntryReader&, + ProfileBufferBlockIndex>, + "ReadEach callback must take ProfileBufferEntryReader& and " + "optionally a ProfileBufferBlockIndex"); + detail::InChunkPointer p{aChunks0, aChunks1}; + while (!p.IsNull()) { + // The position right before an entry size *is* a block index. + const ProfileBufferBlockIndex blockIndex = + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + p.GlobalRangePosition()); + Length entrySize = p.ReadEntrySize(); + if (entrySize == 0) { + return; + } + ProfileBufferEntryReader entryReader = p.EntryReader(entrySize); + if (entryReader.RemainingBytes() == 0) { + return; + } + MOZ_ASSERT(entryReader.RemainingBytes() == entrySize); + if constexpr (std::is_invocable_v<Callback, ProfileBufferEntryReader&, + ProfileBufferBlockIndex>) { + aCallback(entryReader, blockIndex); + } else { + Unused << blockIndex; + aCallback(entryReader); + } + MOZ_ASSERT(entryReader.RemainingBytes() == 0); + } + } + + // Invoke `aCallback(ProfileBufferEntryReader& [, ProfileBufferBlockIndex])` + // on each entry, it must read or at least skip everything. + template <typename Callback> + void ReadEach(Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return; + } + mChunkManager->PeekExtantReleasedChunks( + [&](const ProfileBufferChunk* aOldestChunk) { + ReadEach(aOldestChunk, mCurrentChunk.get(), + std::forward<Callback>(aCallback)); + }); + } + + // Call `aCallback(Maybe<ProfileBufferEntryReader>&&)` on the entry at + // the given ProfileBufferBlockIndex; The `Maybe` will be `Nothing` if + // out-of-session, or if that entry doesn't exist anymore, or if we've reached + // just past the last entry. Return whatever `aCallback` returns. Callback + // should not store `ProfileBufferEntryReader`, because it may become invalid + // after this call. + // Either/both chunk pointers may be null. + template <typename Callback> + [[nodiscard]] static auto ReadAt(ProfileBufferBlockIndex aMinimumBlockIndex, + const ProfileBufferChunk* aChunks0, + const ProfileBufferChunk* aChunks1, + Callback&& aCallback) { + static_assert( + std::is_invocable_v<Callback, Maybe<ProfileBufferEntryReader>&&>, + "ReadAt callback must take a Maybe<ProfileBufferEntryReader>&&"); + Maybe<ProfileBufferEntryReader> maybeEntryReader; + if (detail::InChunkPointer p{aChunks0, aChunks1}; !p.IsNull()) { + // If the pointer position is before the given position, try to advance. + if (p.GlobalRangePosition() >= + aMinimumBlockIndex.ConvertToProfileBufferIndex() || + p.AdvanceToGlobalRangePosition( + aMinimumBlockIndex.ConvertToProfileBufferIndex())) { + MOZ_ASSERT(p.GlobalRangePosition() >= + aMinimumBlockIndex.ConvertToProfileBufferIndex()); + // Here we're pointing at the start of a block, try to read the entry + // size. (Entries cannot be empty, so 0 means failure.) + if (Length entrySize = p.ReadEntrySize(); entrySize != 0) { + maybeEntryReader.emplace(p.EntryReader(entrySize)); + if (maybeEntryReader->RemainingBytes() == 0) { + // An empty entry reader means there was no complete block at the + // given index. + maybeEntryReader.reset(); + } else { + MOZ_ASSERT(maybeEntryReader->RemainingBytes() == entrySize); + } + } + } + } +#ifdef DEBUG + auto assertAllRead = MakeScopeExit([&]() { + MOZ_ASSERT(!maybeEntryReader || maybeEntryReader->RemainingBytes() == 0); + }); +#endif // DEBUG + return std::forward<Callback>(aCallback)(std::move(maybeEntryReader)); + } + + // Call `aCallback(Maybe<ProfileBufferEntryReader>&&)` on the entry at + // the given ProfileBufferBlockIndex; The `Maybe` will be `Nothing` if + // out-of-session, or if that entry doesn't exist anymore, or if we've reached + // just past the last entry. Return whatever `aCallback` returns. Callback + // should not store `ProfileBufferEntryReader`, because it may become invalid + // after this call. + template <typename Callback> + [[nodiscard]] auto ReadAt(ProfileBufferBlockIndex aBlockIndex, + Callback&& aCallback) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return std::forward<Callback>(aCallback)(Nothing{}); + } + return mChunkManager->PeekExtantReleasedChunks( + [&](const ProfileBufferChunk* aOldestChunk) { + return ReadAt(aBlockIndex, aOldestChunk, mCurrentChunk.get(), + std::forward<Callback>(aCallback)); + }); + } + + // Append the contents of another ProfileChunkedBuffer to this one. + ProfileBufferBlockIndex AppendContents(const ProfileChunkedBuffer& aSrc) { + ProfileBufferBlockIndex firstBlockIndex; + // If we start failing, we'll stop writing. + bool failed = false; + aSrc.ReadEach([&](ProfileBufferEntryReader& aER) { + if (failed) { + return; + } + failed = + !Put(aER.RemainingBytes(), [&](Maybe<ProfileBufferEntryWriter>& aEW) { + if (aEW.isNothing()) { + return false; + } + if (!firstBlockIndex) { + firstBlockIndex = aEW->CurrentBlockIndex(); + } + aEW->WriteFromReader(aER, aER.RemainingBytes()); + return true; + }); + }); + return failed ? nullptr : firstBlockIndex; + } + +#ifdef DEBUG + void Dump(std::FILE* aFile = stdout) const { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + fprintf(aFile, + "ProfileChunkedBuffer[%p] State: range %u-%u pushed=%u cleared=%u " + "(live=%u) failed-puts=%u bytes", + this, unsigned(mRangeStart), unsigned(mRangeEnd), + unsigned(mPushedBlockCount), unsigned(mClearedBlockCount), + unsigned(mPushedBlockCount) - unsigned(mClearedBlockCount), + unsigned(mFailedPutBytes)); + if (MOZ_UNLIKELY(!mChunkManager)) { + fprintf(aFile, " - Out-of-session\n"); + return; + } + fprintf(aFile, " - chunks:\n"); + bool hasChunks = false; + mChunkManager->PeekExtantReleasedChunks( + [&](const ProfileBufferChunk* aOldestChunk) { + for (const ProfileBufferChunk* chunk = aOldestChunk; chunk; + chunk = chunk->GetNext()) { + fprintf(aFile, "R "); + chunk->Dump(aFile); + hasChunks = true; + } + }); + if (mCurrentChunk) { + fprintf(aFile, "C "); + mCurrentChunk->Dump(aFile); + hasChunks = true; + } + for (const ProfileBufferChunk* chunk = mNextChunks.get(); chunk; + chunk = chunk->GetNext()) { + fprintf(aFile, "N "); + chunk->Dump(aFile); + hasChunks = true; + } + switch (mRequestedChunkHolder->GetState()) { + case RequestedChunkRefCountedHolder::State::Unused: + fprintf(aFile, " - No request pending.\n"); + break; + case RequestedChunkRefCountedHolder::State::Requested: + fprintf(aFile, " - Request pending.\n"); + break; + case RequestedChunkRefCountedHolder::State::Fulfilled: + fprintf(aFile, " - Request fulfilled.\n"); + break; + } + if (!hasChunks) { + fprintf(aFile, " No chunks.\n"); + } + } +#endif // DEBUG + + private: + // Used to de/serialize a ProfileChunkedBuffer (e.g., containing a backtrace). + friend ProfileBufferEntryWriter::Serializer<ProfileChunkedBuffer>; + friend ProfileBufferEntryReader::Deserializer<ProfileChunkedBuffer>; + friend ProfileBufferEntryWriter::Serializer<UniquePtr<ProfileChunkedBuffer>>; + friend ProfileBufferEntryReader::Deserializer< + UniquePtr<ProfileChunkedBuffer>>; + + [[nodiscard]] UniquePtr<ProfileBufferChunkManager> ResetChunkManager( + const baseprofiler::detail::BaseProfilerMaybeAutoLock&) { + UniquePtr<ProfileBufferChunkManager> chunkManager; + if (mChunkManager) { + mRequestedChunkHolder = nullptr; + mChunkManager->ForgetUnreleasedChunks(); +#ifdef DEBUG + mChunkManager->DeregisteredFrom(this); +#endif + mChunkManager = nullptr; + chunkManager = std::move(mOwnedChunkManager); + if (mCurrentChunk) { + mCurrentChunk->MarkDone(); + mCurrentChunk = nullptr; + } + mNextChunks = nullptr; + mNextChunkRangeStart = mRangeEnd; + mRangeStart = mRangeEnd; + mPushedBlockCount = 0; + mClearedBlockCount = 0; + mFailedPutBytes = 0; + } + return chunkManager; + } + + void SetChunkManager( + ProfileBufferChunkManager& aChunkManager, + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + MOZ_ASSERT(!mChunkManager); + mChunkManager = &aChunkManager; +#ifdef DEBUG + mChunkManager->RegisteredWith(this); +#endif + + mChunkManager->SetChunkDestroyedCallback( + [this](const ProfileBufferChunk& aChunk) { + for (;;) { + ProfileBufferIndex rangeStart = mRangeStart; + if (MOZ_LIKELY(rangeStart <= aChunk.RangeStart())) { + if (MOZ_LIKELY(mRangeStart.compareExchange( + rangeStart, + aChunk.RangeStart() + aChunk.BufferBytes()))) { + break; + } + } + } + mClearedBlockCount += aChunk.BlockCount(); + }); + + // We start with one chunk right away, and request a following one now + // so it should be available before the current chunk is full. + SetAndInitializeCurrentChunk(mChunkManager->GetChunk(), aLock); + mRequestedChunkHolder = MakeRefPtr<RequestedChunkRefCountedHolder>(); + RequestChunk(aLock); + } + + [[nodiscard]] size_t SizeOfExcludingThis( + MallocSizeOf aMallocSizeOf, + const baseprofiler::detail::BaseProfilerMaybeAutoLock&) const { + if (MOZ_UNLIKELY(!mChunkManager)) { + // Out-of-session. + return 0; + } + size_t size = mChunkManager->SizeOfIncludingThis(aMallocSizeOf); + if (mCurrentChunk) { + size += mCurrentChunk->SizeOfIncludingThis(aMallocSizeOf); + } + if (mNextChunks) { + size += mNextChunks->SizeOfIncludingThis(aMallocSizeOf); + } + return size; + } + + void InitializeCurrentChunk( + const baseprofiler::detail::BaseProfilerMaybeAutoLock&) { + MOZ_ASSERT(!!mCurrentChunk); + mCurrentChunk->SetRangeStart(mNextChunkRangeStart); + mNextChunkRangeStart += mCurrentChunk->BufferBytes(); + Unused << mCurrentChunk->ReserveInitialBlockAsTail(0); + } + + void SetAndInitializeCurrentChunk( + UniquePtr<ProfileBufferChunk>&& aChunk, + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + mCurrentChunk = std::move(aChunk); + if (mCurrentChunk) { + InitializeCurrentChunk(aLock); + } + } + + void RequestChunk( + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + if (HandleRequestedChunk_IsPending(aLock)) { + // There is already a pending request, don't start a new one. + return; + } + + // Ensure the `RequestedChunkHolder` knows we're starting a request. + mRequestedChunkHolder->StartRequest(); + + // Request a chunk, the callback carries a `RefPtr` of the + // `RequestedChunkHolder`, so it's guaranteed to live until it's invoked, + // even if this `ProfileChunkedBuffer` changes its `ChunkManager` or is + // destroyed. + mChunkManager->RequestChunk( + [requestedChunkHolder = RefPtr<RequestedChunkRefCountedHolder>( + mRequestedChunkHolder)](UniquePtr<ProfileBufferChunk> aChunk) { + requestedChunkHolder->AddRequestedChunk(std::move(aChunk)); + }); + } + + [[nodiscard]] bool HandleRequestedChunk_IsPending( + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + MOZ_ASSERT(!!mChunkManager); + MOZ_ASSERT(!!mRequestedChunkHolder); + + if (mRequestedChunkHolder->GetState() == + RequestedChunkRefCountedHolder::State::Unused) { + return false; + } + + // A request is either in-flight or fulfilled. + Maybe<UniquePtr<ProfileBufferChunk>> maybeChunk = + mRequestedChunkHolder->GetChunkIfFulfilled(); + if (maybeChunk.isNothing()) { + // Request is still pending. + return true; + } + + // Since we extracted the provided chunk, the holder should now be unused. + MOZ_ASSERT(mRequestedChunkHolder->GetState() == + RequestedChunkRefCountedHolder::State::Unused); + + // Request has been fulfilled. + UniquePtr<ProfileBufferChunk>& chunk = *maybeChunk; + if (chunk) { + // Try to use as current chunk if needed. + if (!mCurrentChunk) { + SetAndInitializeCurrentChunk(std::move(chunk), aLock); + // We've just received a chunk and made it current, request a next chunk + // for later. + MOZ_ASSERT(!mNextChunks); + RequestChunk(aLock); + return true; + } + + if (!mNextChunks) { + mNextChunks = std::move(chunk); + } else { + mNextChunks->InsertNext(std::move(chunk)); + } + } + + return false; + } + + // Get a pointer to the next chunk available + [[nodiscard]] ProfileBufferChunk* GetOrCreateCurrentChunk( + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + ProfileBufferChunk* current = mCurrentChunk.get(); + if (MOZ_UNLIKELY(!current)) { + // No current chunk ready. + MOZ_ASSERT(!mNextChunks, + "There shouldn't be next chunks when there is no current one"); + // See if a request has recently been fulfilled, ignore pending status. + Unused << HandleRequestedChunk_IsPending(aLock); + current = mCurrentChunk.get(); + if (MOZ_UNLIKELY(!current)) { + // There was no pending chunk, try to get one right now. + // This may still fail, but we can't do anything else about it, the + // caller must handle the nullptr case. + // Attempt a request for later. + SetAndInitializeCurrentChunk(mChunkManager->GetChunk(), aLock); + current = mCurrentChunk.get(); + } + } + return current; + } + + // Get a pointer to the next chunk available + [[nodiscard]] ProfileBufferChunk* GetOrCreateNextChunk( + const baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock) { + MOZ_ASSERT(!!mCurrentChunk, + "Why ask for a next chunk when there isn't even a current one?"); + ProfileBufferChunk* next = mNextChunks.get(); + if (MOZ_UNLIKELY(!next)) { + // No next chunk ready, see if a request has recently been fulfilled, + // ignore pending status. + Unused << HandleRequestedChunk_IsPending(aLock); + next = mNextChunks.get(); + if (MOZ_UNLIKELY(!next)) { + // There was no pending chunk, try to get one right now. + mNextChunks = mChunkManager->GetChunk(); + next = mNextChunks.get(); + // This may still fail, but we can't do anything else about it, the + // caller must handle the nullptr case. + if (MOZ_UNLIKELY(!next)) { + // Attempt a request for later. + RequestChunk(aLock); + } + } + } + return next; + } + + // Reserve a block of `aCallbackBlockBytes()` size, and invoke and return + // `aCallback(Maybe<ProfileBufferEntryWriter>&)`. Note that this is the "raw" + // version that doesn't write the entry size at the beginning of the block. + // Note: `aCallbackBlockBytes` is a callback instead of a simple value, to + // delay this potentially-expensive computation until after we're checked that + // we're in-session; use `Put(Length, Callback)` below if you know the size + // already. + template <typename CallbackBlockBytes, typename Callback> + auto ReserveAndPutRaw(CallbackBlockBytes&& aCallbackBlockBytes, + Callback&& aCallback, + baseprofiler::detail::BaseProfilerMaybeAutoLock& aLock, + uint64_t aBlockCount = 1) { + // The entry writer that will point into one or two chunks to write + // into, empty by default (failure). + Maybe<ProfileBufferEntryWriter> maybeEntryWriter; + + // The current chunk will be filled if we need to write more than its + // remaining space. + bool currentChunkFilled = false; + + // If the current chunk gets filled, we may or may not initialize the next + // chunk! + bool nextChunkInitialized = false; + + if (MOZ_LIKELY(mChunkManager)) { + // In-session. + + const Length blockBytes = + std::forward<CallbackBlockBytes>(aCallbackBlockBytes)(); + + if (ProfileBufferChunk* current = GetOrCreateCurrentChunk(aLock); + MOZ_LIKELY(current)) { + if (blockBytes <= current->RemainingBytes()) { + // Block fits in current chunk with only one span. + currentChunkFilled = blockBytes == current->RemainingBytes(); + const auto [mem0, blockIndex] = current->ReserveBlock(blockBytes); + MOZ_ASSERT(mem0.LengthBytes() == blockBytes); + maybeEntryWriter.emplace( + mem0, blockIndex, + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + blockIndex.ConvertToProfileBufferIndex() + blockBytes)); + MOZ_ASSERT(maybeEntryWriter->RemainingBytes() == blockBytes); + mRangeEnd += blockBytes; + mPushedBlockCount += aBlockCount; + } else { + // Block doesn't fit fully in current chunk, it needs to overflow into + // the next one. + // Whether or not we can write this entry, the current chunk is now + // considered full, so it will be released. (Otherwise we could refuse + // this entry, but later accept a smaller entry into this chunk, which + // would be somewhat inconsistent.) + currentChunkFilled = true; + // Make sure the next chunk is available (from a previous request), + // otherwise create one on the spot. + if (ProfileBufferChunk* next = GetOrCreateNextChunk(aLock); + MOZ_LIKELY(next)) { + // Here, we know we have a current and a next chunk. + // Reserve head of block at the end of the current chunk. + const auto [mem0, blockIndex] = + current->ReserveBlock(current->RemainingBytes()); + MOZ_ASSERT(mem0.LengthBytes() < blockBytes); + MOZ_ASSERT(current->RemainingBytes() == 0); + // Set the next chunk range, and reserve the needed space for the + // tail of the block. + next->SetRangeStart(mNextChunkRangeStart); + mNextChunkRangeStart += next->BufferBytes(); + const auto mem1 = next->ReserveInitialBlockAsTail( + blockBytes - mem0.LengthBytes()); + MOZ_ASSERT(next->RemainingBytes() != 0); + nextChunkInitialized = true; + // Block is split in two spans. + maybeEntryWriter.emplace( + mem0, mem1, blockIndex, + ProfileBufferBlockIndex::CreateFromProfileBufferIndex( + blockIndex.ConvertToProfileBufferIndex() + blockBytes)); + MOZ_ASSERT(maybeEntryWriter->RemainingBytes() == blockBytes); + mRangeEnd += blockBytes; + mPushedBlockCount += aBlockCount; + } else { + // Cannot get a new chunk. Record put failure. + mFailedPutBytes += blockBytes; + } + } + } else { + // Cannot get a current chunk. Record put failure. + mFailedPutBytes += blockBytes; + } + } // end of `if (MOZ_LIKELY(mChunkManager))` + + // Here, we either have a `Nothing` (failure), or a non-empty entry writer + // pointing at the start of the block. + + // After we invoke the callback and return, we may need to handle the + // current chunk being filled. + auto handleFilledChunk = MakeScopeExit([&]() { + // If the entry writer was not already empty, the callback *must* have + // filled the full entry. + MOZ_ASSERT(!maybeEntryWriter || maybeEntryWriter->RemainingBytes() == 0); + + if (currentChunkFilled) { + // Extract current (now filled) chunk. + UniquePtr<ProfileBufferChunk> filled = std::move(mCurrentChunk); + + if (mNextChunks) { + // Cycle to the next chunk. + mCurrentChunk = + std::exchange(mNextChunks, mNextChunks->ReleaseNext()); + + // Make sure it is initialized (it is now the current chunk). + if (!nextChunkInitialized) { + InitializeCurrentChunk(aLock); + } + } + + // And finally mark filled chunk done and release it. + filled->MarkDone(); + mChunkManager->ReleaseChunk(std::move(filled)); + + // Request another chunk if needed. + // In most cases, here we should have one current chunk and no next + // chunk, so we want to do a request so there hopefully will be a next + // chunk available when the current one gets filled. + // But we also for a request if we don't even have a current chunk (if + // it's too late, it's ok because the next `ReserveAndPutRaw` wil just + // allocate one on the spot.) + // And if we already have a next chunk, there's no need for more now. + if (!mCurrentChunk || !mNextChunks) { + RequestChunk(aLock); + } + } + }); + + return std::forward<Callback>(aCallback)(maybeEntryWriter); + } + + // Reserve a block of `aBlockBytes` size, and invoke and return + // `aCallback(Maybe<ProfileBufferEntryWriter>&)`. Note that this is the "raw" + // version that doesn't write the entry size at the beginning of the block. + template <typename Callback> + auto ReserveAndPutRaw(Length aBlockBytes, Callback&& aCallback, + uint64_t aBlockCount) { + baseprofiler::detail::BaseProfilerMaybeAutoLock lock(mMutex); + return ReserveAndPutRaw([aBlockBytes]() { return aBlockBytes; }, + std::forward<Callback>(aCallback), lock, + aBlockCount); + } + + // Mutex guarding the following members. + mutable baseprofiler::detail::BaseProfilerMaybeMutex mMutex; + + // Pointer to the current Chunk Manager (or null when out-of-session.) + // It may be owned locally (see below) or externally. + ProfileBufferChunkManager* mChunkManager = nullptr; + + // Only non-null when we own the current Chunk Manager. + UniquePtr<ProfileBufferChunkManager> mOwnedChunkManager; + + UniquePtr<ProfileBufferChunk> mCurrentChunk; + + UniquePtr<ProfileBufferChunk> mNextChunks; + + // Class used to transfer requested chunks from a `ChunkManager` to a + // `ProfileChunkedBuffer`. + // It needs to be ref-counted because the request may be fulfilled + // asynchronously, and either side may be destroyed during the request. + // It cannot use the `ProfileChunkedBuffer` mutex, because that buffer and its + // mutex could be destroyed during the request. + class RequestedChunkRefCountedHolder + : public external::AtomicRefCounted<RequestedChunkRefCountedHolder> { + public: + MOZ_DECLARE_REFCOUNTED_TYPENAME(RequestedChunkRefCountedHolder) + + enum class State { Unused, Requested, Fulfilled }; + + // Get the current state. Note that it may change after the function + // returns, so it should be used carefully, e.g., `ProfileChunkedBuffer` can + // see if a request is pending or fulfilled, to avoid starting another + // request. + [[nodiscard]] State GetState() const { + baseprofiler::detail::BaseProfilerAutoLock lock(mRequestMutex); + return mState; + } + + // Must be called by `ProfileChunkedBuffer` when it requests a chunk. + // There cannot be more than one request in-flight. + void StartRequest() { + baseprofiler::detail::BaseProfilerAutoLock lock(mRequestMutex); + MOZ_ASSERT(mState == State::Unused, "Already requested or fulfilled"); + mState = State::Requested; + } + + // Must be called by the `ChunkManager` with a chunk. + // If the `ChunkManager` cannot provide a chunk (because of memory limits, + // or it gets destroyed), it must call this anyway with a nullptr. + void AddRequestedChunk(UniquePtr<ProfileBufferChunk>&& aChunk) { + baseprofiler::detail::BaseProfilerAutoLock lock(mRequestMutex); + MOZ_ASSERT(mState == State::Requested); + mState = State::Fulfilled; + mRequestedChunk = std::move(aChunk); + } + + // The `ProfileChunkedBuffer` can try to extract the provided chunk after a + // request: + // - Nothing -> Request is not fulfilled yet. + // - Some(nullptr) -> The `ChunkManager` was not able to provide a chunk. + // - Some(chunk) -> Requested chunk. + [[nodiscard]] Maybe<UniquePtr<ProfileBufferChunk>> GetChunkIfFulfilled() { + Maybe<UniquePtr<ProfileBufferChunk>> maybeChunk; + baseprofiler::detail::BaseProfilerAutoLock lock(mRequestMutex); + MOZ_ASSERT(mState == State::Requested || mState == State::Fulfilled); + if (mState == State::Fulfilled) { + mState = State::Unused; + maybeChunk.emplace(std::move(mRequestedChunk)); + } + return maybeChunk; + } + + private: + // Mutex guarding the following members. + mutable baseprofiler::detail::BaseProfilerMutex mRequestMutex; + State mState = State::Unused; + UniquePtr<ProfileBufferChunk> mRequestedChunk; + }; + + // Requested-chunk holder, kept alive when in-session, but may also live + // longer if a request is in-flight. + RefPtr<RequestedChunkRefCountedHolder> mRequestedChunkHolder; + + // Range start of the next chunk to become current. Starting at 1 because + // 0 is a reserved index similar to nullptr. + ProfileBufferIndex mNextChunkRangeStart = 1; + + // Index to the first block. + // Atomic because it may be increased when a Chunk is destroyed, and the + // callback may be invoked from anywhere, including from inside one of our + // locked section, so we cannot protect it with a mutex. + Atomic<ProfileBufferIndex, MemoryOrdering::ReleaseAcquire> mRangeStart{1}; + + // Index past the last block. Equals mRangeStart if empty. + ProfileBufferIndex mRangeEnd = 1; + + // Number of blocks that have been pushed into this buffer. + uint64_t mPushedBlockCount = 0; + + // Number of blocks that have been removed from this buffer. + // Note: Live entries = pushed - cleared. + // Atomic because it may be updated when a Chunk is destroyed, and the + // callback may be invoked from anywhere, including from inside one of our + // locked section, so we cannot protect it with a mutex. + Atomic<uint64_t, MemoryOrdering::ReleaseAcquire> mClearedBlockCount{0}; + + // Number of bytes that could not be put into this buffer. + uint64_t mFailedPutBytes = 0; +}; + +// ---------------------------------------------------------------------------- +// ProfileChunkedBuffer serialization + +// A ProfileChunkedBuffer can hide another one! +// This will be used to store marker backtraces; They can be read back into a +// UniquePtr<ProfileChunkedBuffer>. +// Format: len (ULEB128) | start | end | buffer (len bytes) | pushed | cleared +// len==0 marks an out-of-session buffer, or empty buffer. +template <> +struct ProfileBufferEntryWriter::Serializer<ProfileChunkedBuffer> { + static Length Bytes(const ProfileChunkedBuffer& aBuffer) { + return aBuffer.Read([&](ProfileChunkedBuffer::Reader* aReader) { + if (!aReader) { + // Out-of-session, we only need 1 byte to store a length of 0. + return ULEB128Size<Length>(0); + } + ProfileBufferEntryReader reader = aReader->SingleChunkDataAsEntry(); + const ProfileBufferIndex start = + reader.CurrentBlockIndex().ConvertToProfileBufferIndex(); + const ProfileBufferIndex end = + reader.NextBlockIndex().ConvertToProfileBufferIndex(); + MOZ_ASSERT(end - start <= std::numeric_limits<Length>::max()); + const Length len = static_cast<Length>(end - start); + if (len == 0) { + // In-session but empty, also store a length of 0. + return ULEB128Size<Length>(0); + } + // In-session. + return static_cast<Length>(ULEB128Size(len) + sizeof(start) + len + + sizeof(aBuffer.mPushedBlockCount) + + sizeof(aBuffer.mClearedBlockCount)); + }); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const ProfileChunkedBuffer& aBuffer) { + aBuffer.Read([&](ProfileChunkedBuffer::Reader* aReader) { + if (!aReader) { + // Out-of-session, only store a length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + ProfileBufferEntryReader reader = aReader->SingleChunkDataAsEntry(); + const ProfileBufferIndex start = + reader.CurrentBlockIndex().ConvertToProfileBufferIndex(); + const ProfileBufferIndex end = + reader.NextBlockIndex().ConvertToProfileBufferIndex(); + MOZ_ASSERT(end - start <= std::numeric_limits<Length>::max()); + const Length len = static_cast<Length>(end - start); + MOZ_ASSERT(len <= aEW.RemainingBytes()); + if (len == 0) { + // In-session but empty, only store a length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + // In-session. + // Store buffer length, and start index. + aEW.WriteULEB128(len); + aEW.WriteObject(start); + // Write all the bytes. + aEW.WriteFromReader(reader, reader.RemainingBytes()); + // And write stats. + aEW.WriteObject(static_cast<uint64_t>(aBuffer.mPushedBlockCount)); + aEW.WriteObject(static_cast<uint64_t>(aBuffer.mClearedBlockCount)); + // Note: Failed pushes are not important to serialize. + }); + } +}; + +// A serialized ProfileChunkedBuffer can be read into an empty buffer (either +// out-of-session, or in-session with enough room). +template <> +struct ProfileBufferEntryReader::Deserializer<ProfileChunkedBuffer> { + static void ReadInto(ProfileBufferEntryReader& aER, + ProfileChunkedBuffer& aBuffer) { + // Expect an empty buffer, as we're going to overwrite it. + MOZ_ASSERT(aBuffer.GetState().mRangeStart == aBuffer.GetState().mRangeEnd); + // Read the stored buffer length. + const auto len = aER.ReadULEB128<ProfileChunkedBuffer::Length>(); + if (len == 0) { + // 0-length means an "uninteresting" buffer, just return now. + return; + } + // We have a non-empty buffer to read. + + // Read start and end indices. + const auto start = aER.ReadObject<ProfileBufferIndex>(); + aBuffer.mRangeStart = start; + // For now, set the end to be the start (the buffer is still empty). It will + // be updated in `ReserveAndPutRaw()` below. + aBuffer.mRangeEnd = start; + + if (aBuffer.IsInSession()) { + // Output buffer is in-session (i.e., it already has a memory buffer + // attached). Make sure the caller allocated enough space. + MOZ_RELEASE_ASSERT(aBuffer.BufferLength().value() >= len); + } else { + // Output buffer is out-of-session, set a new chunk manager that will + // provide a single chunk of just the right size. + aBuffer.SetChunkManager(MakeUnique<ProfileBufferChunkManagerSingle>(len)); + MOZ_ASSERT(aBuffer.BufferLength().value() >= len); + } + + // Copy bytes into the buffer. + aBuffer.ReserveAndPutRaw( + len, + [&](Maybe<ProfileBufferEntryWriter>& aEW) { + MOZ_RELEASE_ASSERT(aEW.isSome()); + aEW->WriteFromReader(aER, len); + }, + 0); + // Finally copy stats. + aBuffer.mPushedBlockCount = aER.ReadObject<uint64_t>(); + aBuffer.mClearedBlockCount = aER.ReadObject<uint64_t>(); + // Failed puts are not important to keep. + aBuffer.mFailedPutBytes = 0; + } + + // We cannot output a ProfileChunkedBuffer object (not copyable), use + // `ReadInto()` or `aER.ReadObject<UniquePtr<BlocksRinbBuffer>>()` instead. + static ProfileChunkedBuffer Read(ProfileBufferEntryReader& aER) = delete; +}; + +// A ProfileChunkedBuffer is usually refererenced through a UniquePtr, for +// convenience we support (de)serializing that UniquePtr directly. +// This is compatible with the non-UniquePtr serialization above, with a null +// pointer being treated like an out-of-session or empty buffer; and any of +// these would be deserialized into a null pointer. +template <> +struct ProfileBufferEntryWriter::Serializer<UniquePtr<ProfileChunkedBuffer>> { + static Length Bytes(const UniquePtr<ProfileChunkedBuffer>& aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + return ULEB128Size<Length>(0); + } + // Otherwise write the pointed-at ProfileChunkedBuffer (which could be + // out-of-session or empty.) + return SumBytes(*aBufferUPtr); + } + + static void Write(ProfileBufferEntryWriter& aEW, + const UniquePtr<ProfileChunkedBuffer>& aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + // Otherwise write the pointed-at ProfileChunkedBuffer (which could be + // out-of-session or empty.) + aEW.WriteObject(*aBufferUPtr); + } +}; + +// Serialization of a raw pointer to ProfileChunkedBuffer. +// Use Deserializer<UniquePtr<ProfileChunkedBuffer>> to read it back. +template <> +struct ProfileBufferEntryWriter::Serializer<ProfileChunkedBuffer*> { + static Length Bytes(ProfileChunkedBuffer* aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + return ULEB128Size<Length>(0); + } + // Otherwise write the pointed-at ProfileChunkedBuffer (which could be + // out-of-session or empty.) + return SumBytes(*aBufferUPtr); + } + + static void Write(ProfileBufferEntryWriter& aEW, + ProfileChunkedBuffer* aBufferUPtr) { + if (!aBufferUPtr) { + // Null pointer, treat it like an empty buffer, i.e., write length of 0. + aEW.WriteULEB128<Length>(0); + return; + } + // Otherwise write the pointed-at ProfileChunkedBuffer (which could be + // out-of-session or empty.) + aEW.WriteObject(*aBufferUPtr); + } +}; + +template <> +struct ProfileBufferEntryReader::Deserializer<UniquePtr<ProfileChunkedBuffer>> { + static void ReadInto(ProfileBufferEntryReader& aER, + UniquePtr<ProfileChunkedBuffer>& aBuffer) { + aBuffer = Read(aER); + } + + static UniquePtr<ProfileChunkedBuffer> Read(ProfileBufferEntryReader& aER) { + UniquePtr<ProfileChunkedBuffer> bufferUPtr; + // Keep a copy of the reader before reading the length, so we can restart + // from here below. + ProfileBufferEntryReader readerBeforeLen = aER; + // Read the stored buffer length. + const auto len = aER.ReadULEB128<ProfileChunkedBuffer::Length>(); + if (len == 0) { + // 0-length means an "uninteresting" buffer, just return nullptr. + return bufferUPtr; + } + // We have a non-empty buffer. + // allocate an empty ProfileChunkedBuffer without mutex. + bufferUPtr = MakeUnique<ProfileChunkedBuffer>( + ProfileChunkedBuffer::ThreadSafety::WithoutMutex); + // Rewind the reader before the length and deserialize the contents, using + // the non-UniquePtr Deserializer. + aER = readerBeforeLen; + aER.ReadIntoObject(*bufferUPtr); + return bufferUPtr; + } +}; + +} // namespace mozilla + +#endif // ProfileChunkedBuffer_h diff --git a/mozglue/baseprofiler/public/ProfilingCategoryList.h b/mozglue/baseprofiler/public/ProfilingCategoryList.h new file mode 100644 index 0000000000..437f24aaa1 --- /dev/null +++ b/mozglue/baseprofiler/public/ProfilingCategoryList.h @@ -0,0 +1,122 @@ +/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- + * vim: set ts=8 sts=2 et sw=2 tw=99: + * 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 baseprofiler_ProfilingCategoryList_h +#define baseprofiler_ProfilingCategoryList_h + +// Profiler sub-categories are applied to each sampled stack to describe the +// type of workload that the CPU is busy with. Only one sub-category can be +// assigned so be mindful that these are non-overlapping. The active category is +// set by pushing a label to the profiling stack, or by the unwinder in cases +// such as JITs. A profile sample in arbitrary C++/Rust will typically be +// categorized based on the top of the label stack. +// +// The list of available color names for categories is: +// transparent +// blue +// green +// grey +// lightblue +// magenta +// orange +// purple +// yellow + +// clang-format off + +#define MOZ_PROFILING_CATEGORY_LIST(BEGIN_CATEGORY, SUBCATEGORY, END_CATEGORY) \ + BEGIN_CATEGORY(IDLE, "Idle", "transparent") \ + SUBCATEGORY(IDLE, IDLE, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(OTHER, "Other", "grey") \ + SUBCATEGORY(OTHER, OTHER, "Other") \ + SUBCATEGORY(OTHER, OTHER_PreferenceRead, "Preference Read") \ + SUBCATEGORY(OTHER, OTHER_Profiling, "Profiling") \ + END_CATEGORY \ + BEGIN_CATEGORY(TEST, "Test", "darkgray") \ + SUBCATEGORY(TEST, TEST, "Test") \ + END_CATEGORY \ + BEGIN_CATEGORY(LAYOUT, "Layout", "purple") \ + SUBCATEGORY(LAYOUT, LAYOUT, "Other") \ + SUBCATEGORY(LAYOUT, LAYOUT_FrameConstruction, "Frame construction") \ + SUBCATEGORY(LAYOUT, LAYOUT_Reflow, "Reflow") \ + SUBCATEGORY(LAYOUT, LAYOUT_CSSParsing, "CSS parsing") \ + SUBCATEGORY(LAYOUT, LAYOUT_SelectorQuery, "Selector query") \ + SUBCATEGORY(LAYOUT, LAYOUT_StyleComputation, "Style computation") \ + END_CATEGORY \ + BEGIN_CATEGORY(JS, "JavaScript", "yellow") \ + SUBCATEGORY(JS, JS, "Other") \ + SUBCATEGORY(JS, JS_Parsing, "Parsing") \ + SUBCATEGORY(JS, JS_BaselineCompilation, "JIT Compile (baseline)") \ + SUBCATEGORY(JS, JS_IonCompilation, "JIT Compile (ion)") \ + SUBCATEGORY(JS, JS_Interpreter, "Interpreter") \ + SUBCATEGORY(JS, JS_BaselineInterpret, "JIT (baseline-interpreter)") \ + SUBCATEGORY(JS, JS_Baseline, "JIT (baseline)") \ + SUBCATEGORY(JS, JS_IonMonkey, "JIT (ion)") \ + END_CATEGORY \ + BEGIN_CATEGORY(GCCC, "GC / CC", "orange") \ + SUBCATEGORY(GCCC, GCCC, "Other") \ + SUBCATEGORY(GCCC, GCCC_MinorGC, "Minor GC") \ + SUBCATEGORY(GCCC, GCCC_MajorGC, "Major GC (Other)") \ + SUBCATEGORY(GCCC, GCCC_MajorGC_Mark, "Major GC (Mark)") \ + SUBCATEGORY(GCCC, GCCC_MajorGC_Sweep, "Major GC (Sweep)") \ + SUBCATEGORY(GCCC, GCCC_MajorGC_Compact, "Major GC (Compact)") \ + SUBCATEGORY(GCCC, GCCC_UnmarkGray, "Unmark Gray") \ + SUBCATEGORY(GCCC, GCCC_Barrier, "Barrier") \ + SUBCATEGORY(GCCC, GCCC_FreeSnowWhite, "CC (Free Snow White)") \ + SUBCATEGORY(GCCC, GCCC_BuildGraph, "CC (Build Graph)") \ + SUBCATEGORY(GCCC, GCCC_ScanRoots, "CC (Scan Roots)") \ + SUBCATEGORY(GCCC, GCCC_CollectWhite, "CC (Collect White)") \ + SUBCATEGORY(GCCC, GCCC_Finalize, "CC (Finalize)") \ + END_CATEGORY \ + BEGIN_CATEGORY(NETWORK, "Network", "lightblue") \ + SUBCATEGORY(NETWORK, NETWORK, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(GRAPHICS, "Graphics", "green") \ + SUBCATEGORY(GRAPHICS, GRAPHICS, "Other") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_DisplayListBuilding, "DisplayList building") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_DisplayListMerging, "DisplayList merging") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_LayerBuilding, "Layer building") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_TileAllocation, "Tile allocation") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_WRDisplayList, "WebRender display list") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_Rasterization, "Rasterization") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_FlushingAsyncPaints, "Flushing async paints") \ + SUBCATEGORY(GRAPHICS, GRAPHICS_ImageDecoding, "Image decoding") \ + END_CATEGORY \ + BEGIN_CATEGORY(DOM, "DOM", "blue") \ + SUBCATEGORY(DOM, DOM, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_ANDROID, "Android", "yellow") \ + SUBCATEGORY(JAVA_ANDROID, JAVA_ANDROID, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_ANDROIDX, "AndroidX", "orange") \ + SUBCATEGORY(JAVA_ANDROIDX, JAVA_ANDROIDX, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_LANGUAGE, "Java", "blue") \ + SUBCATEGORY(JAVA_LANGUAGE, JAVA_LANGUAGE, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_MOZILLA, "Mozilla", "green") \ + SUBCATEGORY(JAVA_MOZILLA, JAVA_MOZILLA, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_KOTLIN, "Kotlin", "purple") \ + SUBCATEGORY(JAVA_KOTLIN, JAVA_KOTLIN, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(JAVA_BLOCKED, "Blocked", "lightblue") \ + SUBCATEGORY(JAVA_BLOCKED, JAVA_BLOCKED, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(IPC, "IPC", "lightgreen") \ + SUBCATEGORY(IPC, IPC, "Other") \ + END_CATEGORY \ + BEGIN_CATEGORY(MEDIA, "Media", "orange") \ + SUBCATEGORY(MEDIA, MEDIA, "Other") \ + SUBCATEGORY(MEDIA, MEDIA_CUBEB, "Cubeb") \ + SUBCATEGORY(MEDIA, MEDIA_PLAYBACK, "Playback") \ + SUBCATEGORY(MEDIA, MEDIA_RT, "Real-time rendering") \ + END_CATEGORY + +// clang-format on + +#endif // baseprofiler_ProfilingCategoryList_h diff --git a/mozglue/baseprofiler/public/leb128iterator.h b/mozglue/baseprofiler/public/leb128iterator.h new file mode 100644 index 0000000000..636baf916f --- /dev/null +++ b/mozglue/baseprofiler/public/leb128iterator.h @@ -0,0 +1,207 @@ +/* -*- 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/. */ + +// LEB128 utilities that can read/write unsigned LEB128 numbers from/to +// iterators. +// +// LEB128 = Little Endian Base 128, where small numbers take few bytes, but +// large numbers are still allowed, which is ideal when serializing numbers that +// are likely to be small. +// Each byte contains 7 bits from the number, starting at the "little end", the +// top bit is 0 for the last byte, 1 otherwise. +// Numbers 0-127 only take 1 byte. 128-16383 take 2 bytes. Etc. +// +// Iterators only need to provide: +// - `*it` to return a reference to the next byte to be read from or written to. +// - `++it` to advance the iterator after a byte is written. +// +// The caller must always provide sufficient space to write any number, by: +// - pre-allocating a large enough buffer, or +// - allocating more space when `++it` reaches the end and/or `*it` is invoked +// after the end, or +// - moving the underlying pointer to an appropriate location (e.g., wrapping +// around a circular buffer). +// The caller must also provide enough bytes to read a full value (i.e., at +// least one byte should have its top bit unset), and a type large enough to +// hold the stored value. +// +// Note: There are insufficient checks for validity! These functions are +// intended to be used together, i.e., the user should only `ReadULEB128()` from +// a sufficiently-large buffer that the same user filled with `WriteULEB128()`. +// Using with externally-sourced data (e.g., DWARF) is *not* recommended. +// +// https://en.wikipedia.org/wiki/LEB128 + +#ifndef leb128iterator_h +#define leb128iterator_h + +#include "mozilla/Assertions.h" +#include "mozilla/Likely.h" + +#include <climits> +#include <cstdint> +#include <limits> +#include <type_traits> + +namespace mozilla { + +// Number of bytes needed to represent `aValue`. +template <typename T> +constexpr uint_fast8_t ULEB128Size(T aValue) { + static_assert(!std::numeric_limits<T>::is_signed, + "ULEB128Size only takes unsigned types"); + // We need one output byte per 7 bits of non-zero value. So we just remove + // 7 least significant bits at a time until the value becomes zero. + // Note the special case of 0, which still needs 1 output byte; this is done + // by starting the first loop before we check for 0. + uint_fast8_t size = 0; + for (;;) { + size += 1; + aValue >>= 7; + // Expecting small values, so it should be more likely that `aValue == 0`. + if (MOZ_LIKELY(aValue == 0)) { + return size; + } + } +} + +// Maximum number of bytes needed to represent any value of type `T`. +template <typename T> +constexpr uint_fast8_t ULEB128MaxSize() { + return ULEB128Size<T>(std::numeric_limits<T>::max()); +} + +// Write `aValue` in LEB128 to `aIterator`. +// The iterator will be moved past the last byte. +template <typename T, typename It> +void WriteULEB128(T aValue, It& aIterator) { + static_assert(!std::numeric_limits<T>::is_signed, + "WriteULEB128 only takes unsigned types"); + using IteratorValue = std::remove_reference_t<decltype(*aIterator)>; + static_assert(sizeof(IteratorValue) == 1, + "WriteULEB128 expects an iterator to single bytes"); + // 0. Don't test for 0 yet, as we want to output one byte for it. + for (;;) { + // 1. Extract the 7 least significant bits. + const uint_fast8_t byte = aValue & 0x7Fu; + // 2. Remove them from `aValue`. + aValue >>= 7; + // 3. Write the 7 bits, and set the 8th bit if `aValue` is not 0 yet + // (meaning there will be more bytes after this one.) + // Expecting small values, so it should be more likely that `aValue == 0`. + // Note: No absolute need to force-cast to IteratorValue, because we have + // only changed the bottom 8 bits above. However the compiler could warn + // about a narrowing conversion from potentially-multibyte uint_fast8_t down + // to whatever single-byte type `*iterator* expects, so we make it explicit. + *aIterator = static_cast<IteratorValue>( + MOZ_LIKELY(aValue == 0) ? byte : (byte | 0x80u)); + // 4. Always advance the iterator to the next byte. + ++aIterator; + // 5. We're done if `aValue` is 0. + // Expecting small values, so it should be more likely that `aValue == 0`. + if (MOZ_LIKELY(aValue == 0)) { + return; + } + } +} + +// Read an LEB128 value from `aIterator`. +// The iterator will be moved past the last byte. +template <typename T, typename It> +T ReadULEB128(It& aIterator) { + static_assert(!std::numeric_limits<T>::is_signed, + "ReadULEB128 must return an unsigned type"); + using IteratorValue = std::remove_reference_t<decltype(*aIterator)>; + static_assert(sizeof(IteratorValue) == 1, + "ReadULEB128 expects an iterator to single bytes"); + // Incoming bits will be added to `result`... + T result = 0; + // ... starting with the least significant bits. + uint_fast8_t shift = 0; + for (;;) { + // 1. Read one byte from the iterator. + // `static_cast` just in case IteratorValue is not implicitly convertible to + // uint_fast8_t. It wouldn't matter if the sign was extended, we're only + // dealing with the bottom 8 bits below. + const uint_fast8_t byte = static_cast<uint_fast8_t>(*aIterator); + // 2. Always advance the iterator. + ++aIterator; + // 3. Extract the 7 bits of value, and shift them in place into `result`. + result |= static_cast<T>(byte & 0x7fu) << shift; + // 4. If the 8th bit is *not* set, this was the last byte. + // Expecting small values, so it should be more likely that the bit is off. + if (MOZ_LIKELY((byte & 0x80u) == 0)) { + return result; + } + // There are more bytes to read. + // 5. Next byte will contain more significant bits above the past 7. + shift += 7; + // Safety check that we're not going to shift by >= than the type size, + // which is Undefined Behavior in C++. + MOZ_ASSERT(shift < CHAR_BIT * sizeof(T)); + } +} + +// constexpr ULEB128 reader class. +// Mostly useful when dealing with non-trivial byte feeds. +template <typename T> +class ULEB128Reader { + static_assert(!std::numeric_limits<T>::is_signed, + "ULEB128Reader must handle an unsigned type"); + + public: + constexpr ULEB128Reader() = default; + + // Don't allow copy/assignment, it doesn't make sense for a stateful parser. + constexpr ULEB128Reader(const ULEB128Reader&) = delete; + constexpr ULEB128Reader& operator=(const ULEB128Reader&) = delete; + + // Feed a byte into the parser. + // Returns true if this was the last byte. + [[nodiscard]] constexpr bool FeedByteIsComplete(unsigned aByte) { + MOZ_ASSERT(!IsComplete()); + // Extract the 7 bits of value, and shift them in place into the value. + mValue |= static_cast<T>(aByte & 0x7fu) << mShift; + // If the 8th bit is *not* set, this was the last byte. + // Expecting small values, so it should be more likely that the bit is off. + if (MOZ_LIKELY((aByte & 0x80u) == 0)) { + mShift = mCompleteShift; + return true; + } + // There are more bytes to read. + // Next byte will contain more significant bits above the past 7. + mShift += 7; + // Safety check that we're not going to shift by >= than the type size, + // which is Undefined Behavior in C++. + MOZ_ASSERT(mShift < CHAR_BIT * sizeof(T)); + return false; + } + + constexpr void Reset() { + mValue = 0; + mShift = 0; + } + + [[nodiscard]] constexpr bool IsComplete() const { + return mShift == mCompleteShift; + } + + [[nodiscard]] constexpr T Value() const { + MOZ_ASSERT(IsComplete()); + return mValue; + } + + private: + // Special value of `mShift` indicating that parsing is complete. + constexpr static unsigned mCompleteShift = 0x10000u; + + T mValue = 0; + unsigned mShift = 0; +}; + +} // namespace mozilla + +#endif // leb128iterator_h |