/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ /* vim: set ts=8 sts=2 et sw=2 tw=80: */ /* This Source Code Form is subject to the terms of the Mozilla Public * License, v. 2.0. If a copy of the MPL was not distributed with this * file, You can obtain one at http://mozilla.org/MPL/2.0/. */ #include "nsTimerImpl.h" #include "TimerThread.h" #include "GeckoProfiler.h" #include "nsThreadUtils.h" #include "pratom.h" #include "nsIObserverService.h" #include "mozilla/Services.h" #include "mozilla/ChaosMode.h" #include "mozilla/ArenaAllocator.h" #include "mozilla/ArrayUtils.h" #include "mozilla/BinarySearch.h" #include "mozilla/OperatorNewExtensions.h" #include "mozilla/StaticPrefs_timer.h" #include "mozilla/glean/GleanMetrics.h" #include using namespace mozilla; // Bug 1829983 reports an assertion failure that (so far) has only failed once // in over a month of the assert existing. This #define enables some additional // output that should get printed out if the assert fails again. #if defined(XP_WIN) && defined(DEBUG) # define HACK_OUTPUT_FOR_BUG_1829983 #endif // Uncomment the following line to enable runtime stats during development. // #define TIMERS_RUNTIME_STATS #ifdef TIMERS_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_TIMERS_STATS` // below), to display stats at the end of the program. class StaticTimersStats { public: explicit StaticTimersStats(const char* aName) : mName(aName) {} ~StaticTimersStats() { // Using unsigned long long for computations and printfs. using ULL = unsigned long long; ULL n = static_cast(mCount); if (n == 0) { printf("[%d] Timers stats `%s`: (nothing)\n", int(profiler_current_process_id().ToNumber()), mName); } else if (ULL sumNs = static_cast(mSumDurationsNs); sumNs == 0) { printf("[%d] Timers stats `%s`: %llu\n", int(profiler_current_process_id().ToNumber()), mName, n); } else { printf("[%d] Timers stats `%s`: %llu ns / %llu = %llu ns, max %llu ns\n", int(profiler_current_process_id().ToNumber()), mName, sumNs, n, sumNs / n, static_cast(mLongestDurationNs)); } } void AddDurationFrom(TimeStamp aStart) { // Duration between aStart and now, rounded to the nearest nanosecond. DurationNs duration = static_cast( (TimeStamp::Now() - aStart).ToMicroseconds() * 1000 + 0.5); mSumDurationsNs += duration; ++mCount; // 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. } } void AddCount() { MOZ_ASSERT(mSumDurationsNs == 0, "Don't mix counts and durations"); ++mCount; } private: using DurationNs = uint64_t; using Count = uint32_t; Atomic mSumDurationsNs{0}; Atomic mLongestDurationNs{0}; Atomic mCount{0}; const char* mName; }; // RAII object that measures its scoped lifetime duration and reports it to a // `StaticTimersStats`. class MOZ_RAII AutoTimersStats { public: explicit AutoTimersStats(StaticTimersStats& aStats) : mStats(aStats), mStart(TimeStamp::Now()) {} ~AutoTimersStats() { mStats.AddDurationFrom(mStart); } private: StaticTimersStats& 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_TIMERS_STATS(name) \ static ::StaticTimersStats sStat##name(#name); \ ::AutoTimersStats autoStat##name(sStat##name); // This macro only counts the number of times it's used, not durations. // Don't mix with AUTO_TIMERS_STATS! # define COUNT_TIMERS_STATS(name) \ static ::StaticTimersStats sStat##name(#name); \ sStat##name.AddCount(); #else // TIMERS_RUNTIME_STATS # define AUTO_TIMERS_STATS(name) # define COUNT_TIMERS_STATS(name) #endif // TIMERS_RUNTIME_STATS else NS_IMPL_ISUPPORTS_INHERITED(TimerThread, Runnable, nsIObserver) TimerThread::TimerThread() : Runnable("TimerThread"), mInitialized(false), mMonitor("TimerThread.mMonitor"), mShutdown(false), mWaiting(false), mNotified(false), mSleeping(false), mAllowedEarlyFiringMicroseconds(0) {} TimerThread::~TimerThread() { mThread = nullptr; NS_ASSERTION(mTimers.IsEmpty(), "Timers remain in TimerThread::~TimerThread"); #if TIMER_THREAD_STATISTICS { MonitorAutoLock lock(mMonitor); PrintStatistics(); } #endif } namespace { class TimerObserverRunnable : public Runnable { public: explicit TimerObserverRunnable(nsIObserver* aObserver) : mozilla::Runnable("TimerObserverRunnable"), mObserver(aObserver) {} NS_DECL_NSIRUNNABLE private: nsCOMPtr mObserver; }; NS_IMETHODIMP TimerObserverRunnable::Run() { nsCOMPtr observerService = mozilla::services::GetObserverService(); if (observerService) { observerService->AddObserver(mObserver, "sleep_notification", false); observerService->AddObserver(mObserver, "wake_notification", false); observerService->AddObserver(mObserver, "suspend_process_notification", false); observerService->AddObserver(mObserver, "resume_process_notification", false); } return NS_OK; } } // namespace namespace { // TimerEventAllocator is a thread-safe allocator used only for nsTimerEvents. // It's needed to avoid contention over the default allocator lock when // firing timer events (see bug 733277). The thread-safety is required because // nsTimerEvent objects are allocated on the timer thread, and freed on another // thread. Because TimerEventAllocator has its own lock, contention over that // lock is limited to the allocation and deallocation of nsTimerEvent objects. // // Because this is layered over ArenaAllocator, it never shrinks -- even // "freed" nsTimerEvents aren't truly freed, they're just put onto a free-list // for later recycling. So the amount of memory consumed will always be equal // to the high-water mark consumption. But nsTimerEvents are small and it's // unusual to have more than a few hundred of them, so this shouldn't be a // problem in practice. class TimerEventAllocator { private: struct FreeEntry { FreeEntry* mNext; }; ArenaAllocator<4096> mPool MOZ_GUARDED_BY(mMonitor); FreeEntry* mFirstFree MOZ_GUARDED_BY(mMonitor); mozilla::Monitor mMonitor; public: TimerEventAllocator() : mPool(), mFirstFree(nullptr), mMonitor("TimerEventAllocator") {} ~TimerEventAllocator() = default; void* Alloc(size_t aSize); void Free(void* aPtr); }; } // namespace // This is a nsICancelableRunnable because we can dispatch it to Workers and // those can be shut down at any time, and in these cases, Cancel() is called // instead of Run(). class nsTimerEvent final : public CancelableRunnable { public: NS_IMETHOD Run() override; nsresult Cancel() override { mTimer->Cancel(); return NS_OK; } #ifdef MOZ_COLLECTING_RUNNABLE_TELEMETRY NS_IMETHOD GetName(nsACString& aName) override; #endif explicit nsTimerEvent(already_AddRefed aTimer, ProfilerThreadId aTimerThreadId) : mozilla::CancelableRunnable("nsTimerEvent"), mTimer(aTimer), mGeneration(mTimer->GetGeneration()), mTimerThreadId(aTimerThreadId) { // Note: We override operator new for this class, and the override is // fallible! sAllocatorUsers++; if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug) || profiler_thread_is_being_profiled_for_markers(mTimerThreadId)) { mInitTime = TimeStamp::Now(); } } static void Init(); static void Shutdown(); static void DeleteAllocatorIfNeeded(); static void* operator new(size_t aSize) noexcept(true) { return sAllocator->Alloc(aSize); } void operator delete(void* aPtr) { sAllocator->Free(aPtr); sAllocatorUsers--; DeleteAllocatorIfNeeded(); } already_AddRefed ForgetTimer() { return mTimer.forget(); } private: nsTimerEvent(const nsTimerEvent&) = delete; nsTimerEvent& operator=(const nsTimerEvent&) = delete; nsTimerEvent& operator=(const nsTimerEvent&&) = delete; ~nsTimerEvent() { MOZ_ASSERT(!sCanDeleteAllocator || sAllocatorUsers > 0, "This will result in us attempting to deallocate the " "nsTimerEvent allocator twice"); } TimeStamp mInitTime; RefPtr mTimer; const int32_t mGeneration; ProfilerThreadId mTimerThreadId; static TimerEventAllocator* sAllocator; static Atomic sAllocatorUsers; static Atomic sCanDeleteAllocator; }; TimerEventAllocator* nsTimerEvent::sAllocator = nullptr; Atomic nsTimerEvent::sAllocatorUsers; Atomic nsTimerEvent::sCanDeleteAllocator; namespace { void* TimerEventAllocator::Alloc(size_t aSize) { MOZ_ASSERT(aSize == sizeof(nsTimerEvent)); mozilla::MonitorAutoLock lock(mMonitor); void* p; if (mFirstFree) { p = mFirstFree; mFirstFree = mFirstFree->mNext; } else { p = mPool.Allocate(aSize, fallible); } return p; } void TimerEventAllocator::Free(void* aPtr) { mozilla::MonitorAutoLock lock(mMonitor); FreeEntry* entry = reinterpret_cast(aPtr); entry->mNext = mFirstFree; mFirstFree = entry; } } // namespace struct TimerMarker { static constexpr Span MarkerTypeName() { return MakeStringSpan("Timer"); } static void StreamJSONMarkerData(baseprofiler::SpliceableJSONWriter& aWriter, uint32_t aDelay, uint8_t aType, MarkerThreadId aThreadId, bool aCanceled) { aWriter.IntProperty("delay", aDelay); if (!aThreadId.IsUnspecified()) { // Tech note: If `ToNumber()` returns a uint64_t, the conversion to // int64_t is "implementation-defined" before C++20. This is // acceptable here, because this is a one-way conversion to a unique // identifier that's used to visually separate data by thread on the // front-end. aWriter.IntProperty( "threadId", static_cast(aThreadId.ThreadId().ToNumber())); } if (aCanceled) { aWriter.BoolProperty("canceled", true); // Show a red 'X' as a prefix on the marker chart for canceled timers. aWriter.StringProperty("prefix", "❌"); } // The string property for the timer type is not written when the type is // one shot, as that's the type used almost all the time, and that would // consume space in the profiler buffer and then in the profile JSON, // getting in the way of capturing long power profiles. // Bug 1815677 might make this cheap to capture. if (aType != nsITimer::TYPE_ONE_SHOT) { if (aType == nsITimer::TYPE_REPEATING_SLACK) { aWriter.StringProperty("ttype", "repeating slack"); } else if (aType == nsITimer::TYPE_REPEATING_PRECISE) { aWriter.StringProperty("ttype", "repeating precise"); } else if (aType == nsITimer::TYPE_REPEATING_PRECISE_CAN_SKIP) { aWriter.StringProperty("ttype", "repeating precise can skip"); } else if (aType == nsITimer::TYPE_REPEATING_SLACK_LOW_PRIORITY) { aWriter.StringProperty("ttype", "repeating slack low priority"); } else if (aType == nsITimer::TYPE_ONE_SHOT_LOW_PRIORITY) { aWriter.StringProperty("ttype", "low priority"); } } } static MarkerSchema MarkerTypeDisplay() { using MS = MarkerSchema; MS schema{MS::Location::MarkerChart, MS::Location::MarkerTable}; schema.AddKeyLabelFormat("delay", "Delay", MS::Format::Milliseconds); schema.AddKeyLabelFormat("ttype", "Timer Type", MS::Format::String); schema.AddKeyLabelFormat("canceled", "Canceled", MS::Format::String); schema.SetChartLabel("{marker.data.prefix} {marker.data.delay}"); schema.SetTableLabel( "{marker.name} - {marker.data.prefix} {marker.data.delay}"); return schema; } }; struct AddRemoveTimerMarker { static constexpr Span MarkerTypeName() { return MakeStringSpan("AddRemoveTimer"); } static void StreamJSONMarkerData(baseprofiler::SpliceableJSONWriter& aWriter, const ProfilerString8View& aTimerName, uint32_t aDelay, MarkerThreadId aThreadId) { aWriter.StringProperty("name", aTimerName); aWriter.IntProperty("delay", aDelay); if (!aThreadId.IsUnspecified()) { // Tech note: If `ToNumber()` returns a uint64_t, the conversion to // int64_t is "implementation-defined" before C++20. This is // acceptable here, because this is a one-way conversion to a unique // identifier that's used to visually separate data by thread on the // front-end. aWriter.IntProperty( "threadId", static_cast(aThreadId.ThreadId().ToNumber())); } } static MarkerSchema MarkerTypeDisplay() { using MS = MarkerSchema; MS schema{MS::Location::MarkerChart, MS::Location::MarkerTable}; schema.AddKeyLabelFormatSearchable("name", "Name", MS::Format::String, MS::Searchable::Searchable); schema.AddKeyLabelFormat("delay", "Delay", MS::Format::Milliseconds); schema.SetTableLabel( "{marker.name} - {marker.data.name} - {marker.data.delay}"); return schema; } }; void nsTimerEvent::Init() { sAllocator = new TimerEventAllocator(); } void nsTimerEvent::Shutdown() { sCanDeleteAllocator = true; DeleteAllocatorIfNeeded(); } void nsTimerEvent::DeleteAllocatorIfNeeded() { if (sCanDeleteAllocator && sAllocatorUsers == 0) { delete sAllocator; sAllocator = nullptr; } } #ifdef MOZ_COLLECTING_RUNNABLE_TELEMETRY NS_IMETHODIMP nsTimerEvent::GetName(nsACString& aName) { bool current; MOZ_RELEASE_ASSERT( NS_SUCCEEDED(mTimer->mEventTarget->IsOnCurrentThread(¤t)) && current); mTimer->GetName(aName); return NS_OK; } #endif NS_IMETHODIMP nsTimerEvent::Run() { if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug)) { TimeStamp now = TimeStamp::Now(); MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("[this=%p] time between PostTimerEvent() and Fire(): %fms\n", this, (now - mInitTime).ToMilliseconds())); } if (profiler_thread_is_being_profiled_for_markers(mTimerThreadId)) { MutexAutoLock lock(mTimer->mMutex); nsAutoCString name; mTimer->GetName(name, lock); // This adds a marker with the timer name as the marker name, to make it // obvious which timers are being used. This marker will be useful to // understand which timers might be added and firing excessively often. profiler_add_marker( name, geckoprofiler::category::TIMER, MarkerOptions(MOZ_LIKELY(mInitTime) ? MarkerTiming::Interval( mTimer->mTimeout - mTimer->mDelay, mInitTime) : MarkerTiming::IntervalUntilNowFrom( mTimer->mTimeout - mTimer->mDelay), MarkerThreadId(mTimerThreadId)), TimerMarker{}, mTimer->mDelay.ToMilliseconds(), mTimer->mType, MarkerThreadId::CurrentThread(), false); // This marker is meant to help understand the behavior of the timer thread. profiler_add_marker( "PostTimerEvent", geckoprofiler::category::OTHER, MarkerOptions(MOZ_LIKELY(mInitTime) ? MarkerTiming::IntervalUntilNowFrom(mInitTime) : MarkerTiming::InstantNow(), MarkerThreadId(mTimerThreadId)), AddRemoveTimerMarker{}, name, mTimer->mDelay.ToMilliseconds(), MarkerThreadId::CurrentThread()); } mTimer->Fire(mGeneration); return NS_OK; } nsresult TimerThread::Init() { mMonitor.AssertCurrentThreadOwns(); MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("TimerThread::Init [%d]\n", mInitialized)); if (!mInitialized) { nsTimerEvent::Init(); // We hold on to mThread to keep the thread alive. nsresult rv = NS_NewNamedThread("Timer", getter_AddRefs(mThread), this, {.stackSize = nsIThreadManager::DEFAULT_STACK_SIZE, .blockDispatch = true}); if (NS_FAILED(rv)) { mThread = nullptr; } else { RefPtr r = new TimerObserverRunnable(this); if (NS_IsMainThread()) { r->Run(); } else { NS_DispatchToMainThread(r); } } mInitialized = true; } if (!mThread) { return NS_ERROR_FAILURE; } return NS_OK; } nsresult TimerThread::Shutdown() { MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("TimerThread::Shutdown begin\n")); if (!mThread) { return NS_ERROR_NOT_INITIALIZED; } nsTArray> timers; { // lock scope MonitorAutoLock lock(mMonitor); mShutdown = true; // notify the cond var so that Run() can return if (mWaiting) { mNotified = true; mMonitor.Notify(); } // Need to copy content of mTimers array to a local array // because call to timers' Cancel() (and release its self) // must not be done under the lock. Destructor of a callback // might potentially call some code reentering the same lock // that leads to unexpected behavior or deadlock. // See bug 422472. timers.SetCapacity(mTimers.Length()); for (Entry& entry : mTimers) { if (entry.Value()) { timers.AppendElement(entry.Take()); } } mTimers.Clear(); } for (const RefPtr& timer : timers) { MOZ_ASSERT(timer); timer->Cancel(); } mThread->Shutdown(); // wait for the thread to die nsTimerEvent::Shutdown(); MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("TimerThread::Shutdown end\n")); return NS_OK; } namespace { struct MicrosecondsToInterval { PRIntervalTime operator[](size_t aMs) const { return PR_MicrosecondsToInterval(aMs); } }; struct IntervalComparator { int operator()(PRIntervalTime aInterval) const { return (0 < aInterval) ? -1 : 1; } }; } // namespace #ifdef DEBUG void TimerThread::VerifyTimerListConsistency() const { mMonitor.AssertCurrentThreadOwns(); // Find the first non-canceled timer (and check its cached timeout if we find // it). const size_t timerCount = mTimers.Length(); size_t lastNonCanceledTimerIndex = 0; while (lastNonCanceledTimerIndex < timerCount && !mTimers[lastNonCanceledTimerIndex].Value()) { ++lastNonCanceledTimerIndex; } MOZ_ASSERT(lastNonCanceledTimerIndex == timerCount || mTimers[lastNonCanceledTimerIndex].Value()); MOZ_ASSERT(lastNonCanceledTimerIndex == timerCount || mTimers[lastNonCanceledTimerIndex].Value()->mTimeout == mTimers[lastNonCanceledTimerIndex].Timeout()); // Verify that mTimers is sorted and the cached timeouts are consistent. for (size_t timerIndex = lastNonCanceledTimerIndex + 1; timerIndex < timerCount; ++timerIndex) { if (mTimers[timerIndex].Value()) { MOZ_ASSERT(mTimers[timerIndex].Timeout() == mTimers[timerIndex].Value()->mTimeout); MOZ_ASSERT(mTimers[timerIndex].Timeout() >= mTimers[lastNonCanceledTimerIndex].Timeout()); lastNonCanceledTimerIndex = timerIndex; } } } #endif size_t TimerThread::ComputeTimerInsertionIndex(const TimeStamp& timeout) const { mMonitor.AssertCurrentThreadOwns(); const size_t timerCount = mTimers.Length(); size_t firstGtIndex = 0; while (firstGtIndex < timerCount && (!mTimers[firstGtIndex].Value() || mTimers[firstGtIndex].Timeout() <= timeout)) { ++firstGtIndex; } return firstGtIndex; } TimeStamp TimerThread::ComputeWakeupTimeFromTimers() const { mMonitor.AssertCurrentThreadOwns(); // Timer list should be non-empty and first timer should always be // non-canceled at this point and we rely on that here. MOZ_ASSERT(!mTimers.IsEmpty()); MOZ_ASSERT(mTimers[0].Value()); // Overview: Find the last timer in the list that can be "bundled" together in // the same wake-up with mTimers[0] and use its timeout as our target wake-up // time. // bundleWakeup is when we should wake up in order to be able to fire all of // the timers in our selected bundle. It will always be the timeout of the // last timer in the bundle. TimeStamp bundleWakeup = mTimers[0].Timeout(); // cutoffTime is the latest that we can wake up for the timers currently // accepted into the bundle. These needs to be updated as we go through the // list because later timers may have more strict delay tolerances. const TimeDuration minTimerDelay = TimeDuration::FromMilliseconds( StaticPrefs::timer_minimum_firing_delay_tolerance_ms()); const TimeDuration maxTimerDelay = TimeDuration::FromMilliseconds( StaticPrefs::timer_maximum_firing_delay_tolerance_ms()); TimeStamp cutoffTime = bundleWakeup + ComputeAcceptableFiringDelay(mTimers[0].Delay(), minTimerDelay, maxTimerDelay); const size_t timerCount = mTimers.Length(); for (size_t entryIndex = 1; entryIndex < timerCount; ++entryIndex) { const Entry& curEntry = mTimers[entryIndex]; const nsTimerImpl* curTimer = curEntry.Value(); if (!curTimer) { // Canceled timer - skip it continue; } const TimeStamp curTimerDue = curEntry.Timeout(); if (curTimerDue > cutoffTime) { // Can't include this timer in the bundle - it fires too late. break; } // This timer can be included in the bundle. Update bundleWakeup and // cutoffTime. bundleWakeup = curTimerDue; cutoffTime = std::min( curTimerDue + ComputeAcceptableFiringDelay( curEntry.Delay(), minTimerDelay, maxTimerDelay), cutoffTime); MOZ_ASSERT(bundleWakeup <= cutoffTime); } #ifdef HACK_OUTPUT_FOR_BUG_1829983 const bool assertCondition = bundleWakeup - mTimers[0].Timeout() <= ComputeAcceptableFiringDelay(mTimers[0].Delay(), minTimerDelay, maxTimerDelay); if (!assertCondition) { printf_stderr("*** Special TimerThread debug output ***\n"); const int64_t tDMin = minTimerDelay.GetValue(); const int64_t tDMax = maxTimerDelay.GetValue(); printf_stderr("%16llx / %16llx\n", tDMin, tDMax); const size_t l = mTimers.Length(); for (size_t i = 0; i < l; ++i) { const Entry& e = mTimers[i]; const TimeStamp tS = e.Timeout(); const TimeStampValue tSV = tS.GetValue(); const TimeDuration d = e.Delay(); printf_stderr("[%5zu] %16llx / %16llx / %d / %d / %16llx\n", i, tSV.GTC(), tSV.QPC(), (int)tSV.IsNull(), (int)tSV.HasQPC(), d.GetValue()); } } #endif MOZ_ASSERT(bundleWakeup - mTimers[0].Timeout() <= ComputeAcceptableFiringDelay(mTimers[0].Delay(), minTimerDelay, maxTimerDelay)); return bundleWakeup; } TimeDuration TimerThread::ComputeAcceptableFiringDelay( TimeDuration timerDuration, TimeDuration minDelay, TimeDuration maxDelay) const { // Use the timer's duration divided by this value as a base for how much // firing delay a timer can accept. 8 was chosen specifically because it is a // power of two which means that this division turns nicely into a shift. constexpr int64_t timerDurationDivider = 8; static_assert(IsPowerOfTwo(static_cast(timerDurationDivider))); const TimeDuration tmp = timerDuration / timerDurationDivider; return std::min(std::max(minDelay, tmp), maxDelay); } NS_IMETHODIMP TimerThread::Run() { MonitorAutoLock lock(mMonitor); mProfilerThreadId = profiler_current_thread_id(); // We need to know how many microseconds give a positive PRIntervalTime. This // is platform-dependent and we calculate it at runtime, finding a value |v| // such that |PR_MicrosecondsToInterval(v) > 0| and then binary-searching in // the range [0, v) to find the ms-to-interval scale. uint32_t usForPosInterval = 1; while (PR_MicrosecondsToInterval(usForPosInterval) == 0) { usForPosInterval <<= 1; } size_t usIntervalResolution; BinarySearchIf(MicrosecondsToInterval(), 0, usForPosInterval, IntervalComparator(), &usIntervalResolution); MOZ_ASSERT(PR_MicrosecondsToInterval(usIntervalResolution - 1) == 0); MOZ_ASSERT(PR_MicrosecondsToInterval(usIntervalResolution) == 1); // Half of the amount of microseconds needed to get positive PRIntervalTime. // We use this to decide how to round our wait times later mAllowedEarlyFiringMicroseconds = usIntervalResolution / 2; bool forceRunNextTimer = false; // Queue for tracking of how many timers are fired on each wake-up. We need to // buffer these locally and only send off to glean occasionally to avoid // performance hit. static constexpr size_t kMaxQueuedTimerFired = 128; size_t queuedTimerFiredCount = 0; AutoTArray queuedTimersFiredPerWakeup; queuedTimersFiredPerWakeup.SetLengthAndRetainStorage(kMaxQueuedTimerFired); uint64_t timersFiredThisWakeup = 0; while (!mShutdown) { // Have to use PRIntervalTime here, since PR_WaitCondVar takes it TimeDuration waitFor; bool forceRunThisTimer = forceRunNextTimer; forceRunNextTimer = false; #ifdef DEBUG VerifyTimerListConsistency(); #endif if (mSleeping) { // Sleep for 0.1 seconds while not firing timers. uint32_t milliseconds = 100; if (ChaosMode::isActive(ChaosFeature::TimerScheduling)) { milliseconds = ChaosMode::randomUint32LessThan(200); } waitFor = TimeDuration::FromMilliseconds(milliseconds); } else { waitFor = TimeDuration::Forever(); TimeStamp now = TimeStamp::Now(); #if TIMER_THREAD_STATISTICS if (!mNotified && !mIntendedWakeupTime.IsNull() && now < mIntendedWakeupTime) { ++mEarlyWakeups; const double earlinessms = (mIntendedWakeupTime - now).ToMilliseconds(); mTotalEarlyWakeupTime += earlinessms; } #endif RemoveLeadingCanceledTimersInternal(); if (!mTimers.IsEmpty()) { if (now >= mTimers[0].Value()->mTimeout || forceRunThisTimer) { next: // NB: AddRef before the Release under RemoveTimerInternal to avoid // mRefCnt passing through zero, in case all other refs than the one // from mTimers have gone away (the last non-mTimers[i]-ref's Release // must be racing with us, blocked in gThread->RemoveTimer waiting // for TimerThread::mMonitor, under nsTimerImpl::Release. RefPtr timerRef(mTimers[0].Take()); RemoveFirstTimerInternal(); MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("Timer thread woke up %fms from when it was supposed to\n", fabs((now - timerRef->mTimeout).ToMilliseconds()))); // We are going to let the call to PostTimerEvent here handle the // release of the timer so that we don't end up releasing the timer // on the TimerThread instead of on the thread it targets. { ++timersFiredThisWakeup; LogTimerEvent::Run run(timerRef.get()); PostTimerEvent(timerRef.forget()); } if (mShutdown) { break; } // Update now, as PostTimerEvent plus the locking may have taken a // tick or two, and we may goto next below. now = TimeStamp::Now(); } } RemoveLeadingCanceledTimersInternal(); if (!mTimers.IsEmpty()) { TimeStamp timeout = mTimers[0].Value()->mTimeout; // Don't wait at all (even for PR_INTERVAL_NO_WAIT) if the next timer // is due now or overdue. // // Note that we can only sleep for integer values of a certain // resolution. We use mAllowedEarlyFiringMicroseconds, calculated // before, to do the optimal rounding (i.e., of how to decide what // interval is so small we should not wait at all). double microseconds = (timeout - now).ToMicroseconds(); // The mean value of sFractions must be 1 to ensure that the average of // a long sequence of timeouts converges to the actual sum of their // times. static constexpr double sChaosFractions[] = {0.0, 0.25, 0.5, 0.75, 1.0, 1.75, 2.75}; if (ChaosMode::isActive(ChaosFeature::TimerScheduling)) { microseconds *= sChaosFractions[ChaosMode::randomUint32LessThan( ArrayLength(sChaosFractions))]; forceRunNextTimer = true; } if (microseconds < mAllowedEarlyFiringMicroseconds) { forceRunNextTimer = false; goto next; // round down; execute event now } // TECHNICAL NOTE: Determining waitFor (by subtracting |now| from our // desired wake-up time) at this point is not ideal. For one thing, the // |now| that we have at this point is somewhat old. Secondly, there is // quite a bit of code between here and where we actually use waitFor to // request sleep. If I am thinking about this correctly, both of these // will contribute to us requesting more sleep than is actually needed // to wake up at our desired time. We could avoid this problem by only // determining our desired wake-up time here and then calculating the // wait time when we're actually about to sleep. const TimeStamp wakeupTime = ComputeWakeupTimeFromTimers(); waitFor = wakeupTime - now; // If this were to fail that would mean that we had more timers that we // should have fired. MOZ_ASSERT(!waitFor.IsZero()); if (ChaosMode::isActive(ChaosFeature::TimerScheduling)) { // If chaos mode is active then mess with the amount of time that we // request to sleep (without changing what we record as our expected // wake-up time). This will simulate unintended early/late wake-ups. const double waitInMs = waitFor.ToMilliseconds(); const double chaosWaitInMs = waitInMs * sChaosFractions[ChaosMode::randomUint32LessThan( ArrayLength(sChaosFractions))]; waitFor = TimeDuration::FromMilliseconds(chaosWaitInMs); } mIntendedWakeupTime = wakeupTime; } else { mIntendedWakeupTime = TimeStamp{}; } if (MOZ_LOG_TEST(GetTimerLog(), LogLevel::Debug)) { if (waitFor == TimeDuration::Forever()) MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("waiting forever\n")); else MOZ_LOG(GetTimerLog(), LogLevel::Debug, ("waiting for %f\n", waitFor.ToMilliseconds())); } } { // About to sleep - let's make note of how many timers we processed and // see if we should send out a new batch of telemetry. queuedTimersFiredPerWakeup[queuedTimerFiredCount] = timersFiredThisWakeup; ++queuedTimerFiredCount; if (queuedTimerFiredCount == kMaxQueuedTimerFired) { glean::timer_thread::timers_fired_per_wakeup.AccumulateSamples( queuedTimersFiredPerWakeup); queuedTimerFiredCount = 0; } } #if TIMER_THREAD_STATISTICS { size_t bucketIndex = 0; while (bucketIndex < sTimersFiredPerWakeupBucketCount - 1 && timersFiredThisWakeup > sTimersFiredPerWakeupThresholds[bucketIndex]) { ++bucketIndex; } MOZ_ASSERT(bucketIndex < sTimersFiredPerWakeupBucketCount); ++mTimersFiredPerWakeup[bucketIndex]; ++mTotalWakeupCount; if (mNotified) { ++mTimersFiredPerNotifiedWakeup[bucketIndex]; ++mTotalNotifiedWakeupCount; } else { ++mTimersFiredPerUnnotifiedWakeup[bucketIndex]; ++mTotalUnnotifiedWakeupCount; } } #endif timersFiredThisWakeup = 0; mWaiting = true; mNotified = false; { AUTO_PROFILER_TRACING_MARKER("TimerThread", "Wait", OTHER); mMonitor.Wait(waitFor); } if (mNotified) { forceRunNextTimer = false; } mWaiting = false; } // About to shut down - let's send out the final batch of timers fired counts. if (queuedTimerFiredCount != 0) { queuedTimersFiredPerWakeup.SetLengthAndRetainStorage(queuedTimerFiredCount); glean::timer_thread::timers_fired_per_wakeup.AccumulateSamples( queuedTimersFiredPerWakeup); } return NS_OK; } nsresult TimerThread::AddTimer(nsTimerImpl* aTimer, const MutexAutoLock& aProofOfLock) { MonitorAutoLock lock(mMonitor); AUTO_TIMERS_STATS(TimerThread_AddTimer); if (!aTimer->mEventTarget) { return NS_ERROR_NOT_INITIALIZED; } nsresult rv = Init(); if (NS_FAILED(rv)) { return rv; } // Awaken the timer thread if: // - This timer needs to fire *before* the Timer Thread is scheduled to wake // up. // AND/OR // - The delay is 0, which is usually meant to be run as soon as possible. // Note: Even if the thread is scheduled to wake up now/soon, on some // systems there could be a significant delay compared to notifying, which // is almost immediate; and some users of 0-delay depend on it being this // fast! const TimeDuration minTimerDelay = TimeDuration::FromMilliseconds( StaticPrefs::timer_minimum_firing_delay_tolerance_ms()); const TimeDuration maxTimerDelay = TimeDuration::FromMilliseconds( StaticPrefs::timer_maximum_firing_delay_tolerance_ms()); const TimeDuration firingDelay = ComputeAcceptableFiringDelay( aTimer->mDelay, minTimerDelay, maxTimerDelay); const bool firingBeforeNextWakeup = mIntendedWakeupTime.IsNull() || (aTimer->mTimeout + firingDelay < mIntendedWakeupTime); const bool wakeUpTimerThread = mWaiting && (firingBeforeNextWakeup || aTimer->mDelay.IsZero()); #if TIMER_THREAD_STATISTICS if (mTotalTimersAdded == 0) { mFirstTimerAdded = TimeStamp::Now(); } ++mTotalTimersAdded; #endif // Add the timer to our list. if (!AddTimerInternal(*aTimer)) { return NS_ERROR_OUT_OF_MEMORY; } if (wakeUpTimerThread) { mNotified = true; mMonitor.Notify(); } if (profiler_thread_is_being_profiled_for_markers(mProfilerThreadId)) { nsAutoCString name; aTimer->GetName(name, aProofOfLock); nsLiteralCString prefix("Anonymous_"); profiler_add_marker( "AddTimer", geckoprofiler::category::OTHER, MarkerOptions(MarkerThreadId(mProfilerThreadId), MarkerStack::MaybeCapture( name.Equals("nonfunction:JS") || StringHead(name, prefix.Length()) == prefix)), AddRemoveTimerMarker{}, name, aTimer->mDelay.ToMilliseconds(), MarkerThreadId::CurrentThread()); } return NS_OK; } nsresult TimerThread::RemoveTimer(nsTimerImpl* aTimer, const MutexAutoLock& aProofOfLock) { MonitorAutoLock lock(mMonitor); AUTO_TIMERS_STATS(TimerThread_RemoveTimer); // Remove the timer from our array. Tell callers that aTimer was not found // by returning NS_ERROR_NOT_AVAILABLE. if (!RemoveTimerInternal(*aTimer)) { return NS_ERROR_NOT_AVAILABLE; } #if TIMER_THREAD_STATISTICS ++mTotalTimersRemoved; #endif // Note: The timer thread is *not* awoken. // The removed-timer entry is just left null, and will be reused (by a new or // re-set timer) or discarded (when the timer thread logic handles non-null // timers around it). // If this was the front timer, and in the unlikely case that its entry is not // soon reused by a re-set timer, the timer thread will wake up at the // previously-scheduled time, but will quickly notice that there is no actual // pending timer, and will restart its wait until the following real timeout. if (profiler_thread_is_being_profiled_for_markers(mProfilerThreadId)) { nsAutoCString name; aTimer->GetName(name, aProofOfLock); nsLiteralCString prefix("Anonymous_"); // This marker is meant to help understand the behavior of the timer thread. profiler_add_marker( "RemoveTimer", geckoprofiler::category::OTHER, MarkerOptions(MarkerThreadId(mProfilerThreadId), MarkerStack::MaybeCapture( name.Equals("nonfunction:JS") || StringHead(name, prefix.Length()) == prefix)), AddRemoveTimerMarker{}, name, aTimer->mDelay.ToMilliseconds(), MarkerThreadId::CurrentThread()); // This adds a marker with the timer name as the marker name, to make it // obvious which timers are being used. This marker will be useful to // understand which timers might be added and removed excessively often. profiler_add_marker(name, geckoprofiler::category::TIMER, MarkerOptions(MarkerTiming::IntervalUntilNowFrom( aTimer->mTimeout - aTimer->mDelay), MarkerThreadId(mProfilerThreadId)), TimerMarker{}, aTimer->mDelay.ToMilliseconds(), aTimer->mType, MarkerThreadId::CurrentThread(), true); } return NS_OK; } TimeStamp TimerThread::FindNextFireTimeForCurrentThread(TimeStamp aDefault, uint32_t aSearchBound) { MonitorAutoLock lock(mMonitor); AUTO_TIMERS_STATS(TimerThread_FindNextFireTimeForCurrentThread); for (const Entry& entry : mTimers) { const nsTimerImpl* timer = entry.Value(); if (timer) { if (entry.Timeout() > aDefault) { return aDefault; } // Don't yield to timers created with the *_LOW_PRIORITY type. if (!timer->IsLowPriority()) { bool isOnCurrentThread = false; nsresult rv = timer->mEventTarget->IsOnCurrentThread(&isOnCurrentThread); if (NS_SUCCEEDED(rv) && isOnCurrentThread) { return entry.Timeout(); } } if (aSearchBound == 0) { // Return the currently highest timeout when we reach the bound. // This won't give accurate information if we stop before finding // any timer for the current thread, but at least won't report too // long idle period. return timer->mTimeout; } --aSearchBound; } } // No timers for this thread, return the default. return aDefault; } // This function must be called from within a lock // Also: we hold the mutex for the nsTimerImpl. bool TimerThread::AddTimerInternal(nsTimerImpl& aTimer) { mMonitor.AssertCurrentThreadOwns(); aTimer.mMutex.AssertCurrentThreadOwns(); AUTO_TIMERS_STATS(TimerThread_AddTimerInternal); if (mShutdown) { return false; } LogTimerEvent::LogDispatch(&aTimer); const TimeStamp& timeout = aTimer.mTimeout; const size_t insertionIndex = ComputeTimerInsertionIndex(timeout); if (insertionIndex != 0 && !mTimers[insertionIndex - 1].Value()) { // Very common scenario in practice: The timer just before the insertion // point is canceled, overwrite it. AUTO_TIMERS_STATS(TimerThread_AddTimerInternal_overwrite_before); mTimers[insertionIndex - 1] = Entry{aTimer}; return true; } const size_t length = mTimers.Length(); if (insertionIndex == length) { // We're at the end (including it's the very first insertion), add new timer // at the end. AUTO_TIMERS_STATS(TimerThread_AddTimerInternal_append); return mTimers.AppendElement(Entry{aTimer}, mozilla::fallible); } if (!mTimers[insertionIndex].Value()) { // The timer at the insertion point is canceled, overwrite it. AUTO_TIMERS_STATS(TimerThread_AddTimerInternal_overwrite); mTimers[insertionIndex] = Entry{aTimer}; return true; } // The new timer has to be inserted. AUTO_TIMERS_STATS(TimerThread_AddTimerInternal_insert); // The capacity should be checked first, because if it needs to be increased // and the memory allocation fails, only the new timer should be lost. if (length == mTimers.Capacity() && mTimers[length - 1].Value()) { // We have reached capacity, and the last entry is not canceled, so we // really want to increase the capacity in case the extra slot is required. // To force-expand the array, append a canceled-timer entry with a timestamp // far in the future. // This empty Entry may be used below to receive the moved-from previous // entry. If not, it may be used in a later call if we need to append a new // timer at the end. AUTO_TIMERS_STATS(TimerThread_AddTimerInternal_insert_expand); if (!mTimers.AppendElement( Entry{mTimers[length - 1].Timeout() + TimeDuration::FromSeconds(365.0 * 24.0 * 60.0 * 60.0)}, mozilla::fallible)) { return false; } } // Extract the timer at the insertion point, and put the new timer in its // place. Entry extractedEntry = std::exchange(mTimers[insertionIndex], Entry{aTimer}); // Following entries can be pushed until we hit a canceled timer or the end. for (size_t i = insertionIndex + 1; i < length; ++i) { Entry& entryRef = mTimers[i]; if (!entryRef.Value()) { // Canceled entry, overwrite it with the extracted entry from before. COUNT_TIMERS_STATS(TimerThread_AddTimerInternal_insert_overwrite); entryRef = std::move(extractedEntry); return true; } // Write extracted entry from before, and extract current entry. COUNT_TIMERS_STATS(TimerThread_AddTimerInternal_insert_shifts); std::swap(entryRef, extractedEntry); } // We've reached the end of the list, with still one extracted entry to // re-insert. We've checked the capacity above, this cannot fail. COUNT_TIMERS_STATS(TimerThread_AddTimerInternal_insert_append); mTimers.AppendElement(std::move(extractedEntry)); return true; } // This function must be called from within a lock // Also: we hold the mutex for the nsTimerImpl. bool TimerThread::RemoveTimerInternal(nsTimerImpl& aTimer) { mMonitor.AssertCurrentThreadOwns(); aTimer.mMutex.AssertCurrentThreadOwns(); AUTO_TIMERS_STATS(TimerThread_RemoveTimerInternal); if (!aTimer.IsInTimerThread()) { COUNT_TIMERS_STATS(TimerThread_RemoveTimerInternal_not_in_list); return false; } AUTO_TIMERS_STATS(TimerThread_RemoveTimerInternal_in_list); for (auto& entry : mTimers) { if (entry.Value() == &aTimer) { entry.Forget(); return true; } } MOZ_ASSERT(!aTimer.IsInTimerThread(), "Not found in the list but it should be!?"); return false; } void TimerThread::RemoveLeadingCanceledTimersInternal() { mMonitor.AssertCurrentThreadOwns(); AUTO_TIMERS_STATS(TimerThread_RemoveLeadingCanceledTimersInternal); size_t toRemove = 0; while (toRemove < mTimers.Length() && !mTimers[toRemove].Value()) { ++toRemove; } mTimers.RemoveElementsAt(0, toRemove); } void TimerThread::RemoveFirstTimerInternal() { mMonitor.AssertCurrentThreadOwns(); AUTO_TIMERS_STATS(TimerThread_RemoveFirstTimerInternal); MOZ_ASSERT(!mTimers.IsEmpty()); mTimers.RemoveElementAt(0); } void TimerThread::PostTimerEvent(already_AddRefed aTimerRef) { mMonitor.AssertCurrentThreadOwns(); AUTO_TIMERS_STATS(TimerThread_PostTimerEvent); RefPtr timer(aTimerRef); #if TIMER_THREAD_STATISTICS const double actualFiringDelay = std::max((TimeStamp::Now() - timer->mTimeout).ToMilliseconds(), 0.0); if (mNotified) { ++mTotalTimersFiredNotified; mTotalActualTimerFiringDelayNotified += actualFiringDelay; } else { ++mTotalTimersFiredUnnotified; mTotalActualTimerFiringDelayUnnotified += actualFiringDelay; } #endif if (!timer->mEventTarget) { NS_ERROR("Attempt to post timer event to NULL event target"); return; } // XXX we may want to reuse this nsTimerEvent in the case of repeating timers. // Since we already addref'd 'timer', we don't need to addref here. // We will release either in ~nsTimerEvent(), or pass the reference back to // the caller. We need to copy the generation number from this timer into the // event, so we can avoid firing a timer that was re-initialized after being // canceled. nsCOMPtr target = timer->mEventTarget; void* p = nsTimerEvent::operator new(sizeof(nsTimerEvent)); if (!p) { return; } RefPtr event = ::new (KnownNotNull, p) nsTimerEvent(timer.forget(), mProfilerThreadId); nsresult rv; { // We release mMonitor around the Dispatch because if the Dispatch interacts // with the timer API we'll deadlock. MonitorAutoUnlock unlock(mMonitor); rv = target->Dispatch(event, NS_DISPATCH_NORMAL); if (NS_FAILED(rv)) { timer = event->ForgetTimer(); // We do this to avoid possible deadlock by taking the two locks in a // different order than is used in RemoveTimer(). RemoveTimer() has // aTimer->mMutex first. We use timer.get() to keep static analysis // happy // NOTE: I'm not sure that any of the below is actually necessary. It // seems to me that the timer that we're trying to fire will have already // been removed prior to this. MutexAutoLock lock1(timer.get()->mMutex); MonitorAutoLock lock2(mMonitor); RemoveTimerInternal(*timer); } } } void TimerThread::DoBeforeSleep() { // Mainthread MonitorAutoLock lock(mMonitor); mSleeping = true; } // Note: wake may be notified without preceding sleep notification void TimerThread::DoAfterSleep() { // Mainthread MonitorAutoLock lock(mMonitor); mSleeping = false; // Wake up the timer thread to re-process the array to ensure the sleep delay // is correct, and fire any expired timers (perhaps quite a few) mNotified = true; PROFILER_MARKER_UNTYPED("AfterSleep", OTHER, MarkerThreadId(mProfilerThreadId)); mMonitor.Notify(); } NS_IMETHODIMP TimerThread::Observe(nsISupports* /* aSubject */, const char* aTopic, const char16_t* /* aData */) { if (StaticPrefs::timer_ignore_sleep_wake_notifications()) { return NS_OK; } if (strcmp(aTopic, "sleep_notification") == 0 || strcmp(aTopic, "suspend_process_notification") == 0) { DoBeforeSleep(); } else if (strcmp(aTopic, "wake_notification") == 0 || strcmp(aTopic, "resume_process_notification") == 0) { DoAfterSleep(); } return NS_OK; } uint32_t TimerThread::AllowedEarlyFiringMicroseconds() { MonitorAutoLock lock(mMonitor); return mAllowedEarlyFiringMicroseconds; } #if TIMER_THREAD_STATISTICS void TimerThread::PrintStatistics() const { mMonitor.AssertCurrentThreadOwns(); const TimeStamp freshNow = TimeStamp::Now(); const double timeElapsed = mFirstTimerAdded.IsNull() ? 0.0 : (freshNow - mFirstTimerAdded).ToSeconds(); printf_stderr("TimerThread Stats (Total time %8.2fs)\n", timeElapsed); printf_stderr("Added: %6llu Removed: %6llu Fired: %6llu\n", mTotalTimersAdded, mTotalTimersRemoved, mTotalTimersFiredNotified + mTotalTimersFiredUnnotified); auto PrintTimersFiredBucket = [](const AutoTArray& buckets, const size_t wakeupCount, const size_t timersFiredCount, const double totalTimerDelay, const char* label) { printf_stderr("%s : [", label); for (size_t bucketVal : buckets) { printf_stderr(" %5llu", bucketVal); } printf_stderr( " ] Wake-ups/timer %6llu / %6llu (%7.4f) Avg Timer Delay %7.4f\n", wakeupCount, timersFiredCount, static_cast(wakeupCount) / timersFiredCount, totalTimerDelay / timersFiredCount); }; printf_stderr("Wake-ups:\n"); PrintTimersFiredBucket( mTimersFiredPerWakeup, mTotalWakeupCount, mTotalTimersFiredNotified + mTotalTimersFiredUnnotified, mTotalActualTimerFiringDelayNotified + mTotalActualTimerFiringDelayUnnotified, "Total "); PrintTimersFiredBucket(mTimersFiredPerNotifiedWakeup, mTotalNotifiedWakeupCount, mTotalTimersFiredNotified, mTotalActualTimerFiringDelayNotified, "Notified "); PrintTimersFiredBucket(mTimersFiredPerUnnotifiedWakeup, mTotalUnnotifiedWakeupCount, mTotalTimersFiredUnnotified, mTotalActualTimerFiringDelayUnnotified, "Unnotified "); printf_stderr("Early Wake-ups: %6llu Avg: %7.4fms\n", mEarlyWakeups, mTotalEarlyWakeupTime / mEarlyWakeups); } #endif /* This nsReadOnlyTimer class is used for the values returned by the * TimerThread::GetTimers method. * It is not possible to return a strong reference to the nsTimerImpl * instance (that could extend the lifetime of the timer and cause it to fire * a callback pointing to already freed memory) or a weak reference * (nsSupportsWeakReference doesn't support freeing the referee on a thread * that isn't the thread that owns the weak reference), so instead the timer * name, delay and type are copied to a new object. */ class nsReadOnlyTimer final : public nsITimer { public: explicit nsReadOnlyTimer(const nsACString& aName, uint32_t aDelay, uint32_t aType) : mName(aName), mDelay(aDelay), mType(aType) {} NS_DECL_ISUPPORTS NS_IMETHOD Init(nsIObserver* aObserver, uint32_t aDelayInMs, uint32_t aType) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD InitWithCallback(nsITimerCallback* aCallback, uint32_t aDelayInMs, uint32_t aType) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD InitHighResolutionWithCallback(nsITimerCallback* aCallback, const mozilla::TimeDuration& aDelay, uint32_t aType) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD Cancel(void) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD InitWithNamedFuncCallback(nsTimerCallbackFunc aCallback, void* aClosure, uint32_t aDelay, uint32_t aType, const char* aName) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD InitHighResolutionWithNamedFuncCallback( nsTimerCallbackFunc aCallback, void* aClosure, const mozilla::TimeDuration& aDelay, uint32_t aType, const char* aName) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetName(nsACString& aName) override { aName = mName; return NS_OK; } NS_IMETHOD GetDelay(uint32_t* aDelay) override { *aDelay = mDelay; return NS_OK; } NS_IMETHOD SetDelay(uint32_t aDelay) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetType(uint32_t* aType) override { *aType = mType; return NS_OK; } NS_IMETHOD SetType(uint32_t aType) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetClosure(void** aClosure) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetCallback(nsITimerCallback** aCallback) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetTarget(nsIEventTarget** aTarget) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD SetTarget(nsIEventTarget* aTarget) override { return NS_ERROR_NOT_IMPLEMENTED; } NS_IMETHOD GetAllowedEarlyFiringMicroseconds( uint32_t* aAllowedEarlyFiringMicroseconds) override { return NS_ERROR_NOT_IMPLEMENTED; } size_t SizeOfIncludingThis(mozilla::MallocSizeOf aMallocSizeOf) override { return sizeof(*this); } private: nsCString mName; uint32_t mDelay; uint32_t mType; ~nsReadOnlyTimer() = default; }; NS_IMPL_ISUPPORTS(nsReadOnlyTimer, nsITimer) nsresult TimerThread::GetTimers(nsTArray>& aRetVal) { nsTArray> timers; { MonitorAutoLock lock(mMonitor); for (const auto& entry : mTimers) { nsTimerImpl* timer = entry.Value(); if (!timer) { continue; } timers.AppendElement(timer); } } for (nsTimerImpl* timer : timers) { nsAutoCString name; timer->GetName(name); uint32_t delay; timer->GetDelay(&delay); uint32_t type; timer->GetType(&type); aRetVal.AppendElement(new nsReadOnlyTimer(name, delay, type)); } return NS_OK; }