// Copyright 2020 The Chromium Authors // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "base/threading/hang_watcher.h" #include #include #include "base/containers/flat_map.h" #include "base/debug/alias.h" #include "base/debug/crash_logging.h" #include "base/debug/dump_without_crashing.h" #include "base/debug/leak_annotations.h" #include "base/feature_list.h" #include "base/functional/bind.h" #include "base/functional/callback_helpers.h" #include "base/metrics/field_trial_params.h" #include "base/metrics/histogram_macros.h" #include "base/power_monitor/power_monitor.h" #include "base/ranges/algorithm.h" #include "base/strings/string_number_conversions.h" #include "base/synchronization/lock.h" #include "base/synchronization/waitable_event.h" #include "base/threading/platform_thread.h" #include "base/threading/thread_checker.h" #include "base/threading/thread_restrictions.h" #include "base/threading/threading_features.h" #include "base/time/default_tick_clock.h" #include "base/time/time.h" #include "base/trace_event/base_tracing.h" #include "build/build_config.h" #include "third_party/abseil-cpp/absl/base/attributes.h" namespace base { namespace { // Defines how much logging happens when the HangWatcher monitors the threads. // Logging levels are set per thread type through Finch. It's important that // the order of the enum members stay the same and that their numerical // values be in increasing order. The implementation of // ThreadTypeLoggingLevelGreaterOrEqual() depends on it. enum class LoggingLevel { kNone = 0, kUmaOnly = 1, kUmaAndCrash = 2 }; HangWatcher* g_instance = nullptr; ABSL_CONST_INIT thread_local internal::HangWatchState* hang_watch_state = nullptr; std::atomic g_use_hang_watcher{false}; std::atomic g_hang_watcher_process_type{ HangWatcher::ProcessType::kBrowserProcess}; std::atomic g_threadpool_log_level{LoggingLevel::kNone}; std::atomic g_io_thread_log_level{LoggingLevel::kNone}; std::atomic g_main_thread_log_level{LoggingLevel::kNone}; // Indicates whether HangWatcher::Run() should return after the next monitoring. std::atomic g_keep_monitoring{true}; // Emits the hung thread count histogram. |count| is the number of threads // of type |thread_type| that were hung or became hung during the last // monitoring window. This function should be invoked for each thread type // encountered on each call to Monitor(). void LogHungThreadCountHistogram(HangWatcher::ThreadType thread_type, int count) { // In the case of unique threads like the IO or UI/Main thread a count does // not make sense. const bool any_thread_hung = count >= 1; const HangWatcher::ProcessType process_type = g_hang_watcher_process_type.load(std::memory_order_relaxed); switch (process_type) { case HangWatcher::ProcessType::kUnknownProcess: break; case HangWatcher::ProcessType::kBrowserProcess: switch (thread_type) { case HangWatcher::ThreadType::kIOThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.BrowserProcess." "IOThread", any_thread_hung); break; case HangWatcher::ThreadType::kMainThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.BrowserProcess." "UIThread", any_thread_hung); break; case HangWatcher::ThreadType::kThreadPoolThread: // Not recorded for now. break; } break; case HangWatcher::ProcessType::kGPUProcess: // Not recorded for now. break; case HangWatcher::ProcessType::kRendererProcess: switch (thread_type) { case HangWatcher::ThreadType::kIOThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.RendererProcess." "IOThread", any_thread_hung); break; case HangWatcher::ThreadType::kMainThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.RendererProcess." "MainThread", any_thread_hung); break; case HangWatcher::ThreadType::kThreadPoolThread: // Not recorded for now. break; } break; case HangWatcher::ProcessType::kUtilityProcess: switch (thread_type) { case HangWatcher::ThreadType::kIOThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.UtilityProcess." "IOThread", any_thread_hung); break; case HangWatcher::ThreadType::kMainThread: UMA_HISTOGRAM_BOOLEAN( "HangWatcher.IsThreadHung.UtilityProcess." "MainThread", any_thread_hung); break; case HangWatcher::ThreadType::kThreadPoolThread: // Not recorded for now. break; } break; } } // Returns true if |thread_type| was configured through Finch to have a logging // level that is equal to or exceeds |logging_level|. bool ThreadTypeLoggingLevelGreaterOrEqual(HangWatcher::ThreadType thread_type, LoggingLevel logging_level) { switch (thread_type) { case HangWatcher::ThreadType::kIOThread: return g_io_thread_log_level.load(std::memory_order_relaxed) >= logging_level; case HangWatcher::ThreadType::kMainThread: return g_main_thread_log_level.load(std::memory_order_relaxed) >= logging_level; case HangWatcher::ThreadType::kThreadPoolThread: return g_threadpool_log_level.load(std::memory_order_relaxed) >= logging_level; } } } // namespace // Determines if the HangWatcher is activated. When false the HangWatcher // thread never started. BASE_FEATURE(kEnableHangWatcher, "EnableHangWatcher", FEATURE_ENABLED_BY_DEFAULT); // Browser process. constexpr base::FeatureParam kIOThreadLogLevel{ &kEnableHangWatcher, "io_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kUIThreadLogLevel{ &kEnableHangWatcher, "ui_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kThreadPoolLogLevel{ &kEnableHangWatcher, "threadpool_log_level", static_cast(LoggingLevel::kUmaOnly)}; // GPU process. constexpr base::FeatureParam kGPUProcessIOThreadLogLevel{ &kEnableHangWatcher, "gpu_process_io_thread_log_level", static_cast(LoggingLevel::kNone)}; constexpr base::FeatureParam kGPUProcessMainThreadLogLevel{ &kEnableHangWatcher, "gpu_process_main_thread_log_level", static_cast(LoggingLevel::kNone)}; constexpr base::FeatureParam kGPUProcessThreadPoolLogLevel{ &kEnableHangWatcher, "gpu_process_threadpool_log_level", static_cast(LoggingLevel::kNone)}; // Renderer process. constexpr base::FeatureParam kRendererProcessIOThreadLogLevel{ &kEnableHangWatcher, "renderer_process_io_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kRendererProcessMainThreadLogLevel{ &kEnableHangWatcher, "renderer_process_main_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kRendererProcessThreadPoolLogLevel{ &kEnableHangWatcher, "renderer_process_threadpool_log_level", static_cast(LoggingLevel::kUmaOnly)}; // Utility process. constexpr base::FeatureParam kUtilityProcessIOThreadLogLevel{ &kEnableHangWatcher, "utility_process_io_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kUtilityProcessMainThreadLogLevel{ &kEnableHangWatcher, "utility_process_main_thread_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr base::FeatureParam kUtilityProcessThreadPoolLogLevel{ &kEnableHangWatcher, "utility_process_threadpool_log_level", static_cast(LoggingLevel::kUmaOnly)}; constexpr const char* kThreadName = "HangWatcher"; // The time that the HangWatcher thread will sleep for between calls to // Monitor(). Increasing or decreasing this does not modify the type of hangs // that can be detected. It instead increases the probability that a call to // Monitor() will happen at the right time to catch a hang. This has to be // balanced with power/cpu use concerns as busy looping would catch amost all // hangs but present unacceptable overhead. NOTE: If this period is ever changed // then all metrics that depend on it like // HangWatcher.IsThreadHung need to be updated. constexpr auto kMonitoringPeriod = base::Seconds(10); WatchHangsInScope::WatchHangsInScope(TimeDelta timeout) { internal::HangWatchState* current_hang_watch_state = HangWatcher::IsEnabled() ? internal::HangWatchState::GetHangWatchStateForCurrentThread() : nullptr; DCHECK(timeout >= base::TimeDelta()) << "Negative timeouts are invalid."; // Thread is not monitored, noop. if (!current_hang_watch_state) { took_effect_ = false; return; } #if DCHECK_IS_ON() previous_watch_hangs_in_scope_ = current_hang_watch_state->GetCurrentWatchHangsInScope(); current_hang_watch_state->SetCurrentWatchHangsInScope(this); #endif auto [old_flags, old_deadline] = current_hang_watch_state->GetFlagsAndDeadline(); // TODO(crbug.com/1034046): Check whether we are over deadline already for the // previous WatchHangsInScope here by issuing only one TimeTicks::Now() // and resuing the value. previous_deadline_ = old_deadline; TimeTicks deadline = TimeTicks::Now() + timeout; current_hang_watch_state->SetDeadline(deadline); current_hang_watch_state->IncrementNestingLevel(); const bool hangs_ignored_for_current_scope = internal::HangWatchDeadline::IsFlagSet( internal::HangWatchDeadline::Flag::kIgnoreCurrentWatchHangsInScope, old_flags); // If the current WatchHangsInScope is ignored, temporarily reactivate hang // watching for newly created WatchHangsInScopes. On exiting hang watching // is suspended again to return to the original state. if (hangs_ignored_for_current_scope) { current_hang_watch_state->UnsetIgnoreCurrentWatchHangsInScope(); set_hangs_ignored_on_exit_ = true; } } WatchHangsInScope::~WatchHangsInScope() { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); // If hang watching was not enabled at construction time there is nothing to // validate or undo. if (!took_effect_) { return; } // If the thread was unregistered since construction there is also nothing to // do. auto* const state = internal::HangWatchState::GetHangWatchStateForCurrentThread(); if (!state) { return; } // If a hang is currently being captured we should block here so execution // stops and we avoid recording unrelated stack frames in the crash. if (state->IsFlagSet(internal::HangWatchDeadline::Flag::kShouldBlockOnHang)) { base::HangWatcher::GetInstance()->BlockIfCaptureInProgress(); } #if DCHECK_IS_ON() // Verify that no Scope was destructed out of order. DCHECK_EQ(this, state->GetCurrentWatchHangsInScope()); state->SetCurrentWatchHangsInScope(previous_watch_hangs_in_scope_); #endif if (state->nesting_level() == 1) { // If a call to InvalidateActiveExpectations() suspended hang watching // during the lifetime of this or any nested WatchHangsInScope it can now // safely be reactivated by clearing the ignore bit since this is the // outer-most scope. state->UnsetIgnoreCurrentWatchHangsInScope(); } else if (set_hangs_ignored_on_exit_) { // Return to ignoring hangs since this was the previous state before hang // watching was temporarily enabled for this WatchHangsInScope only in the // constructor. state->SetIgnoreCurrentWatchHangsInScope(); } // Reset the deadline to the value it had before entering this // WatchHangsInScope. state->SetDeadline(previous_deadline_); // TODO(crbug.com/1034046): Log when a WatchHangsInScope exits after its // deadline and that went undetected by the HangWatcher. state->DecrementNestingLevel(); } // static void HangWatcher::InitializeOnMainThread(ProcessType process_type) { DCHECK(!g_use_hang_watcher); DCHECK(g_io_thread_log_level == LoggingLevel::kNone); DCHECK(g_main_thread_log_level == LoggingLevel::kNone); DCHECK(g_threadpool_log_level == LoggingLevel::kNone); bool enable_hang_watcher = base::FeatureList::IsEnabled(kEnableHangWatcher); // Do not start HangWatcher in the GPU process until the issue related to // invalid magic signature in the GPU WatchDog is fixed // (https://crbug.com/1297760). if (process_type == ProcessType::kGPUProcess) enable_hang_watcher = false; g_use_hang_watcher.store(enable_hang_watcher, std::memory_order_relaxed); // Keep the process type. g_hang_watcher_process_type.store(process_type, std::memory_order_relaxed); // If hang watching is disabled as a whole there is no need to read the // params. if (!enable_hang_watcher) return; // Retrieve thread-specific config for hang watching. switch (process_type) { case HangWatcher::ProcessType::kUnknownProcess: break; case HangWatcher::ProcessType::kBrowserProcess: g_threadpool_log_level.store( static_cast(kThreadPoolLogLevel.Get()), std::memory_order_relaxed); g_io_thread_log_level.store( static_cast(kIOThreadLogLevel.Get()), std::memory_order_relaxed); g_main_thread_log_level.store( static_cast(kUIThreadLogLevel.Get()), std::memory_order_relaxed); break; case HangWatcher::ProcessType::kGPUProcess: g_threadpool_log_level.store( static_cast(kGPUProcessThreadPoolLogLevel.Get()), std::memory_order_relaxed); g_io_thread_log_level.store( static_cast(kGPUProcessIOThreadLogLevel.Get()), std::memory_order_relaxed); g_main_thread_log_level.store( static_cast(kGPUProcessMainThreadLogLevel.Get()), std::memory_order_relaxed); break; case HangWatcher::ProcessType::kRendererProcess: g_threadpool_log_level.store( static_cast(kRendererProcessThreadPoolLogLevel.Get()), std::memory_order_relaxed); g_io_thread_log_level.store( static_cast(kRendererProcessIOThreadLogLevel.Get()), std::memory_order_relaxed); g_main_thread_log_level.store( static_cast(kRendererProcessMainThreadLogLevel.Get()), std::memory_order_relaxed); break; case HangWatcher::ProcessType::kUtilityProcess: g_threadpool_log_level.store( static_cast(kUtilityProcessThreadPoolLogLevel.Get()), std::memory_order_relaxed); g_io_thread_log_level.store( static_cast(kUtilityProcessIOThreadLogLevel.Get()), std::memory_order_relaxed); g_main_thread_log_level.store( static_cast(kUtilityProcessMainThreadLogLevel.Get()), std::memory_order_relaxed); break; } } void HangWatcher::UnitializeOnMainThreadForTesting() { g_use_hang_watcher.store(false, std::memory_order_relaxed); g_threadpool_log_level.store(LoggingLevel::kNone, std::memory_order_relaxed); g_io_thread_log_level.store(LoggingLevel::kNone, std::memory_order_relaxed); g_main_thread_log_level.store(LoggingLevel::kNone, std::memory_order_relaxed); } // static bool HangWatcher::IsEnabled() { return g_use_hang_watcher.load(std::memory_order_relaxed); } // static bool HangWatcher::IsThreadPoolHangWatchingEnabled() { return g_threadpool_log_level.load(std::memory_order_relaxed) != LoggingLevel::kNone; } // static bool HangWatcher::IsIOThreadHangWatchingEnabled() { return g_io_thread_log_level.load(std::memory_order_relaxed) != LoggingLevel::kNone; } // static bool HangWatcher::IsCrashReportingEnabled() { if (g_main_thread_log_level.load(std::memory_order_relaxed) == LoggingLevel::kUmaAndCrash) { return true; } if (g_io_thread_log_level.load(std::memory_order_relaxed) == LoggingLevel::kUmaAndCrash) { return true; } if (g_threadpool_log_level.load(std::memory_order_relaxed) == LoggingLevel::kUmaAndCrash) { return true; } return false; } // static void HangWatcher::InvalidateActiveExpectations() { auto* const state = internal::HangWatchState::GetHangWatchStateForCurrentThread(); if (!state) { // If the current thread is not under watch there is nothing to invalidate. return; } state->SetIgnoreCurrentWatchHangsInScope(); } HangWatcher::HangWatcher() : monitor_period_(kMonitoringPeriod), should_monitor_(WaitableEvent::ResetPolicy::AUTOMATIC), thread_(this, kThreadName), tick_clock_(base::DefaultTickClock::GetInstance()), memory_pressure_listener_( FROM_HERE, base::BindRepeating(&HangWatcher::OnMemoryPressure, base::Unretained(this))) { // |thread_checker_| should not be bound to the constructing thread. DETACH_FROM_THREAD(hang_watcher_thread_checker_); should_monitor_.declare_only_used_while_idle(); DCHECK(!g_instance); g_instance = this; } // static void HangWatcher::CreateHangWatcherInstance() { DCHECK(!g_instance); g_instance = new base::HangWatcher(); // The hang watcher is leaked to make sure it survives all watched threads. ANNOTATE_LEAKING_OBJECT_PTR(g_instance); } #if !BUILDFLAG(IS_NACL) debug::ScopedCrashKeyString HangWatcher::GetTimeSinceLastCriticalMemoryPressureCrashKey() { DCHECK_CALLED_ON_VALID_THREAD(hang_watcher_thread_checker_); // The crash key size is large enough to hold the biggest possible return // value from base::TimeDelta::InSeconds(). constexpr debug::CrashKeySize kCrashKeyContentSize = debug::CrashKeySize::Size32; DCHECK_GE(static_cast(kCrashKeyContentSize), base::NumberToString(std::numeric_limits::max()).size()); static debug::CrashKeyString* crash_key = AllocateCrashKeyString( "seconds-since-last-memory-pressure", kCrashKeyContentSize); const base::TimeTicks last_critical_memory_pressure_time = last_critical_memory_pressure_.load(std::memory_order_relaxed); if (last_critical_memory_pressure_time.is_null()) { constexpr char kNoMemoryPressureMsg[] = "No critical memory pressure"; static_assert( std::size(kNoMemoryPressureMsg) <= static_cast(kCrashKeyContentSize), "The crash key is too small to hold \"No critical memory pressure\"."); return debug::ScopedCrashKeyString(crash_key, kNoMemoryPressureMsg); } else { base::TimeDelta time_since_last_critical_memory_pressure = base::TimeTicks::Now() - last_critical_memory_pressure_time; return debug::ScopedCrashKeyString( crash_key, base::NumberToString( time_since_last_critical_memory_pressure.InSeconds())); } } #endif std::string HangWatcher::GetTimeSinceLastSystemPowerResumeCrashKeyValue() const { DCHECK_CALLED_ON_VALID_THREAD(hang_watcher_thread_checker_); const TimeTicks last_system_power_resume_time = PowerMonitor::GetLastSystemResumeTime(); if (last_system_power_resume_time.is_null()) return "Never suspended"; if (last_system_power_resume_time == TimeTicks::Max()) return "Power suspended"; const TimeDelta time_since_last_system_resume = TimeTicks::Now() - last_system_power_resume_time; return NumberToString(time_since_last_system_resume.InSeconds()); } void HangWatcher::OnMemoryPressure( base::MemoryPressureListener::MemoryPressureLevel memory_pressure_level) { if (memory_pressure_level == base::MemoryPressureListener::MEMORY_PRESSURE_LEVEL_CRITICAL) { last_critical_memory_pressure_.store(base::TimeTicks::Now(), std::memory_order_relaxed); } } HangWatcher::~HangWatcher() { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); DCHECK_EQ(g_instance, this); DCHECK(watch_states_.empty()); g_instance = nullptr; Stop(); } void HangWatcher::Start() { thread_.Start(); } void HangWatcher::Stop() { g_keep_monitoring.store(false, std::memory_order_relaxed); should_monitor_.Signal(); thread_.Join(); // In production HangWatcher is always leaked but during testing it's possibly // stopped and restarted using a new instance. This makes sure the next call // to Start() will actually monitor in that case. g_keep_monitoring.store(true, std::memory_order_relaxed); } bool HangWatcher::IsWatchListEmpty() { AutoLock auto_lock(watch_state_lock_); return watch_states_.empty(); } void HangWatcher::Wait() { while (true) { // Amount by which the actual time spent sleeping can deviate from // the target time and still be considered timely. constexpr base::TimeDelta kWaitDriftTolerance = base::Milliseconds(100); const base::TimeTicks time_before_wait = tick_clock_->NowTicks(); // Sleep until next scheduled monitoring or until signaled. const bool was_signaled = should_monitor_.TimedWait(monitor_period_); if (after_wait_callback_) after_wait_callback_.Run(time_before_wait); const base::TimeTicks time_after_wait = tick_clock_->NowTicks(); const base::TimeDelta wait_time = time_after_wait - time_before_wait; const bool wait_was_normal = wait_time <= (monitor_period_ + kWaitDriftTolerance); UMA_HISTOGRAM_TIMES("HangWatcher.SleepDrift.BrowserProcess", wait_time - monitor_period_); if (!wait_was_normal) { // If the time spent waiting was too high it might indicate the machine is // very slow or that that it went to sleep. In any case we can't trust the // WatchHangsInScopes that are currently live. Update the ignore // threshold to make sure they don't trigger a hang on subsequent monitors // then keep waiting. base::AutoLock auto_lock(watch_state_lock_); // Find the latest deadline among the live watch states. They might change // atomically while iterating but that's fine because if they do that // means the new WatchHangsInScope was constructed very soon after the // abnormal sleep happened and might be affected by the root cause still. // Ignoring it is cautious and harmless. base::TimeTicks latest_deadline; for (const auto& state : watch_states_) { base::TimeTicks deadline = state->GetDeadline(); if (deadline > latest_deadline) { latest_deadline = deadline; } } deadline_ignore_threshold_ = latest_deadline; } // Stop waiting. if (wait_was_normal || was_signaled) return; } } void HangWatcher::Run() { // Monitor() should only run on |thread_|. Bind |thread_checker_| here to make // sure of that. DCHECK_CALLED_ON_VALID_THREAD(hang_watcher_thread_checker_); while (g_keep_monitoring.load(std::memory_order_relaxed)) { Wait(); if (!IsWatchListEmpty() && g_keep_monitoring.load(std::memory_order_relaxed)) { Monitor(); if (after_monitor_closure_for_testing_) { after_monitor_closure_for_testing_.Run(); } } } } // static HangWatcher* HangWatcher::GetInstance() { return g_instance; } // static void HangWatcher::RecordHang() { base::debug::DumpWithoutCrashing(); NO_CODE_FOLDING(); } ScopedClosureRunner HangWatcher::RegisterThreadInternal( ThreadType thread_type) { AutoLock auto_lock(watch_state_lock_); CHECK(base::FeatureList::GetInstance()); // Do not install a WatchState if the results would never be observable. if (!ThreadTypeLoggingLevelGreaterOrEqual(thread_type, LoggingLevel::kUmaOnly)) { return ScopedClosureRunner(base::DoNothing()); } watch_states_.push_back( internal::HangWatchState::CreateHangWatchStateForCurrentThread( thread_type)); return ScopedClosureRunner(BindOnce(&HangWatcher::UnregisterThread, Unretained(HangWatcher::GetInstance()))); } // static ScopedClosureRunner HangWatcher::RegisterThread(ThreadType thread_type) { if (!GetInstance()) { return ScopedClosureRunner(); } return GetInstance()->RegisterThreadInternal(thread_type); } base::TimeTicks HangWatcher::WatchStateSnapShot::GetHighestDeadline() const { DCHECK(IsActionable()); // Since entries are sorted in increasing order the last entry is the largest // one. return hung_watch_state_copies_.back().deadline; } HangWatcher::WatchStateSnapShot::WatchStateSnapShot() = default; void HangWatcher::WatchStateSnapShot::Init( const HangWatchStates& watch_states, base::TimeTicks deadline_ignore_threshold) { DCHECK(!initialized_); // No matter if the snapshot is actionable or not after this function // it will have been initialized. initialized_ = true; const base::TimeTicks now = base::TimeTicks::Now(); bool all_threads_marked = true; bool found_deadline_before_ignore_threshold = false; // Use an std::array to store the hang counts to avoid allocations. The // numerical values of the HangWatcher::ThreadType enum is used to index into // the array. A |kInvalidHangCount| is used to signify there were no threads // of the type found. constexpr size_t kHangCountArraySize = static_cast(base::HangWatcher::ThreadType::kMax) + 1; std::array hung_counts_per_thread_type; constexpr int kInvalidHangCount = -1; hung_counts_per_thread_type.fill(kInvalidHangCount); // Will be true if any of the hung threads has a logging level high enough, // as defined through finch params, to warant dumping a crash. bool any_hung_thread_has_dumping_enabled = false; // Copy hung thread information. for (const auto& watch_state : watch_states) { uint64_t flags; TimeTicks deadline; std::tie(flags, deadline) = watch_state->GetFlagsAndDeadline(); if (deadline <= deadline_ignore_threshold) { found_deadline_before_ignore_threshold = true; } if (internal::HangWatchDeadline::IsFlagSet( internal::HangWatchDeadline::Flag::kIgnoreCurrentWatchHangsInScope, flags)) { continue; } // If a thread type is monitored and did not hang it still needs to be // logged as a zero count; const size_t hang_count_index = static_cast(watch_state.get()->thread_type()); if (hung_counts_per_thread_type[hang_count_index] == kInvalidHangCount) { hung_counts_per_thread_type[hang_count_index] = 0; } // Only copy hung threads. if (deadline <= now) { ++hung_counts_per_thread_type[hang_count_index]; if (ThreadTypeLoggingLevelGreaterOrEqual(watch_state.get()->thread_type(), LoggingLevel::kUmaAndCrash)) { any_hung_thread_has_dumping_enabled = true; } #if BUILDFLAG(ENABLE_BASE_TRACING) // Emit trace events for monitored threads. if (ThreadTypeLoggingLevelGreaterOrEqual(watch_state.get()->thread_type(), LoggingLevel::kUmaOnly)) { const PlatformThreadId thread_id = watch_state.get()->GetThreadID(); const auto track = perfetto::Track::FromPointer( this, perfetto::ThreadTrack::ForThread(thread_id)); TRACE_EVENT_BEGIN("base", "HangWatcher::ThreadHung", track, deadline); TRACE_EVENT_END("base", track, now); // TODO(crbug.com/1021571): Remove this once fixed. PERFETTO_INTERNAL_ADD_EMPTY_EVENT(); } #endif // Attempt to mark the thread as needing to stay within its current // WatchHangsInScope until capture is complete. bool thread_marked = watch_state->SetShouldBlockOnHang(flags, deadline); // If marking some threads already failed the snapshot won't be kept so // there is no need to keep adding to it. The loop doesn't abort though // to keep marking the other threads. If these threads remain hung until // the next capture then they'll already be marked and will be included // in the capture at that time. if (thread_marked && all_threads_marked) { hung_watch_state_copies_.push_back( WatchStateCopy{deadline, watch_state.get()->GetThreadID()}); } else { all_threads_marked = false; } } } // Log the hung thread counts to histograms for each thread type if any thread // of the type were found. for (size_t i = 0; i < kHangCountArraySize; ++i) { const int hang_count = hung_counts_per_thread_type[i]; const HangWatcher::ThreadType thread_type = static_cast(i); if (hang_count != kInvalidHangCount && ThreadTypeLoggingLevelGreaterOrEqual(thread_type, LoggingLevel::kUmaOnly)) { LogHungThreadCountHistogram(thread_type, hang_count); } } // Three cases can invalidate this snapshot and prevent the capture of the // hang. // // 1. Some threads could not be marked for blocking so this snapshot isn't // actionable since marked threads could be hung because of unmarked ones. // If only the marked threads were captured the information would be // incomplete. // // 2. Any of the threads have a deadline before |deadline_ignore_threshold|. // If any thread is ignored it reduces the confidence in the whole state and // it's better to avoid capturing misleading data. // // 3. The hung threads found were all of types that are not configured through // Finch to trigger a crash dump. // if (!all_threads_marked || found_deadline_before_ignore_threshold || !any_hung_thread_has_dumping_enabled) { hung_watch_state_copies_.clear(); return; } // Sort |hung_watch_state_copies_| by order of decreasing hang severity so the // most severe hang is first in the list. ranges::sort(hung_watch_state_copies_, [](const WatchStateCopy& lhs, const WatchStateCopy& rhs) { return lhs.deadline < rhs.deadline; }); } void HangWatcher::WatchStateSnapShot::Clear() { hung_watch_state_copies_.clear(); initialized_ = false; } HangWatcher::WatchStateSnapShot::WatchStateSnapShot( const WatchStateSnapShot& other) = default; HangWatcher::WatchStateSnapShot::~WatchStateSnapShot() = default; std::string HangWatcher::WatchStateSnapShot::PrepareHungThreadListCrashKey() const { DCHECK(IsActionable()); // Build a crash key string that contains the ids of the hung threads. constexpr char kSeparator{'|'}; std::string list_of_hung_thread_ids; // Add as many thread ids to the crash key as possible. for (const WatchStateCopy& copy : hung_watch_state_copies_) { std::string fragment = base::NumberToString(copy.thread_id) + kSeparator; if (list_of_hung_thread_ids.size() + fragment.size() < static_cast(debug::CrashKeySize::Size256)) { list_of_hung_thread_ids += fragment; } else { // Respect the by priority ordering of thread ids in the crash key by // stopping the construction as soon as one does not fit. This avoids // including lesser priority ids while omitting more important ones. break; } } return list_of_hung_thread_ids; } bool HangWatcher::WatchStateSnapShot::IsActionable() const { DCHECK(initialized_); return !hung_watch_state_copies_.empty(); } HangWatcher::WatchStateSnapShot HangWatcher::GrabWatchStateSnapshotForTesting() const { WatchStateSnapShot snapshot; snapshot.Init(watch_states_, deadline_ignore_threshold_); return snapshot; } void HangWatcher::Monitor() { DCHECK_CALLED_ON_VALID_THREAD(hang_watcher_thread_checker_); AutoLock auto_lock(watch_state_lock_); // If all threads unregistered since this function was invoked there's // nothing to do anymore. if (watch_states_.empty()) return; watch_state_snapshot_.Init(watch_states_, deadline_ignore_threshold_); if (watch_state_snapshot_.IsActionable()) { DoDumpWithoutCrashing(watch_state_snapshot_); } watch_state_snapshot_.Clear(); } void HangWatcher::DoDumpWithoutCrashing( const WatchStateSnapShot& watch_state_snapshot) { TRACE_EVENT("base", "HangWatcher::DoDumpWithoutCrashing"); capture_in_progress_.store(true, std::memory_order_relaxed); base::AutoLock scope_lock(capture_lock_); #if !BUILDFLAG(IS_NACL) const std::string list_of_hung_thread_ids = watch_state_snapshot.PrepareHungThreadListCrashKey(); static debug::CrashKeyString* crash_key = AllocateCrashKeyString( "list-of-hung-threads", debug::CrashKeySize::Size256); const debug::ScopedCrashKeyString list_of_hung_threads_crash_key_string( crash_key, list_of_hung_thread_ids); const debug::ScopedCrashKeyString time_since_last_critical_memory_pressure_crash_key_string = GetTimeSinceLastCriticalMemoryPressureCrashKey(); SCOPED_CRASH_KEY_STRING32("HangWatcher", "seconds-since-last-resume", GetTimeSinceLastSystemPowerResumeCrashKeyValue()); #endif // To avoid capturing more than one hang that blames a subset of the same // threads it's necessary to keep track of what is the furthest deadline // that contributed to declaring a hang. Only once // all threads have deadlines past this point can we be sure that a newly // discovered hang is not directly related. // Example: // ********************************************************************** // Timeline A : L------1-------2----------3-------4----------N----------- // Timeline B : -------2----------3-------4----------L----5------N------- // Timeline C : L----------------------------5------6----7---8------9---N // ********************************************************************** // In the example when a Monitor() happens during timeline A // |deadline_ignore_threshold_| (L) is at time zero and deadlines (1-4) // are before Now() (N) . A hang is captured and L is updated. During // the next Monitor() (timeline B) a new deadline is over but we can't // capture a hang because deadlines 2-4 are still live and already counted // toward a hang. During a third monitor (timeline C) all live deadlines // are now after L and a second hang can be recorded. base::TimeTicks latest_expired_deadline = watch_state_snapshot.GetHighestDeadline(); if (on_hang_closure_for_testing_) on_hang_closure_for_testing_.Run(); else RecordHang(); // Update after running the actual capture. deadline_ignore_threshold_ = latest_expired_deadline; capture_in_progress_.store(false, std::memory_order_relaxed); } void HangWatcher::SetAfterMonitorClosureForTesting( base::RepeatingClosure closure) { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); after_monitor_closure_for_testing_ = std::move(closure); } void HangWatcher::SetOnHangClosureForTesting(base::RepeatingClosure closure) { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); on_hang_closure_for_testing_ = std::move(closure); } void HangWatcher::SetMonitoringPeriodForTesting(base::TimeDelta period) { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); monitor_period_ = period; } void HangWatcher::SetAfterWaitCallbackForTesting( RepeatingCallback callback) { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); after_wait_callback_ = callback; } void HangWatcher::SignalMonitorEventForTesting() { DCHECK_CALLED_ON_VALID_THREAD(constructing_thread_checker_); should_monitor_.Signal(); } // static void HangWatcher::StopMonitoringForTesting() { g_keep_monitoring.store(false, std::memory_order_relaxed); } void HangWatcher::SetTickClockForTesting(const base::TickClock* tick_clock) { tick_clock_ = tick_clock; } void HangWatcher::BlockIfCaptureInProgress() { // Makes a best-effort attempt to block execution if a hang is currently being // captured. Only block on |capture_lock| if |capture_in_progress_| hints that // it's already held to avoid serializing all threads on this function when no // hang capture is in-progress. if (capture_in_progress_.load(std::memory_order_relaxed)) base::AutoLock hang_lock(capture_lock_); } void HangWatcher::UnregisterThread() { AutoLock auto_lock(watch_state_lock_); auto it = ranges::find( watch_states_, internal::HangWatchState::GetHangWatchStateForCurrentThread(), &std::unique_ptr::get); // Thread should be registered to get unregistered. DCHECK(it != watch_states_.end()); watch_states_.erase(it); } namespace internal { namespace { constexpr uint64_t kOnlyDeadlineMask = 0x00FF'FFFF'FFFF'FFFFu; constexpr uint64_t kOnlyFlagsMask = ~kOnlyDeadlineMask; constexpr uint64_t kMaximumFlag = 0x8000'0000'0000'0000u; // Use as a mask to keep persistent flags and the deadline. constexpr uint64_t kPersistentFlagsAndDeadlineMask = kOnlyDeadlineMask | static_cast( HangWatchDeadline::Flag::kIgnoreCurrentWatchHangsInScope); } // namespace // Flag binary representation assertions. static_assert( static_cast(HangWatchDeadline::Flag::kMinValue) > kOnlyDeadlineMask, "Invalid numerical value for flag. Would interfere with bits of data."); static_assert(static_cast(HangWatchDeadline::Flag::kMaxValue) <= kMaximumFlag, "A flag can only set a single bit."); HangWatchDeadline::HangWatchDeadline() = default; HangWatchDeadline::~HangWatchDeadline() = default; std::pair HangWatchDeadline::GetFlagsAndDeadline() const { uint64_t bits = bits_.load(std::memory_order_relaxed); return std::make_pair(ExtractFlags(bits), DeadlineFromBits(ExtractDeadline((bits)))); } TimeTicks HangWatchDeadline::GetDeadline() const { return DeadlineFromBits( ExtractDeadline(bits_.load(std::memory_order_relaxed))); } // static TimeTicks HangWatchDeadline::Max() { // |kOnlyDeadlineMask| has all the bits reserved for the TimeTicks value // set. This means it also represents the highest representable value. return DeadlineFromBits(kOnlyDeadlineMask); } // static bool HangWatchDeadline::IsFlagSet(Flag flag, uint64_t flags) { return static_cast(flag) & flags; } void HangWatchDeadline::SetDeadline(TimeTicks new_deadline) { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); DCHECK(new_deadline <= Max()) << "Value too high to be represented."; DCHECK(new_deadline >= TimeTicks{}) << "Value cannot be negative."; if (switch_bits_callback_for_testing_) { const uint64_t switched_in_bits = SwitchBitsForTesting(); // If a concurrent deadline change is tested it cannot have a deadline or // persistent flag change since those always happen on the same thread. DCHECK((switched_in_bits & kPersistentFlagsAndDeadlineMask) == 0u); } // Discard all non-persistent flags and apply deadline change. const uint64_t old_bits = bits_.load(std::memory_order_relaxed); const uint64_t new_flags = ExtractFlags(old_bits & kPersistentFlagsAndDeadlineMask); bits_.store(new_flags | ExtractDeadline(static_cast( new_deadline.ToInternalValue())), std::memory_order_relaxed); } // TODO(crbug.com/1087026): Add flag DCHECKs here. bool HangWatchDeadline::SetShouldBlockOnHang(uint64_t old_flags, TimeTicks old_deadline) { DCHECK(old_deadline <= Max()) << "Value too high to be represented."; DCHECK(old_deadline >= TimeTicks{}) << "Value cannot be negative."; // Set the kShouldBlockOnHang flag only if |bits_| did not change since it was // read. kShouldBlockOnHang is the only non-persistent flag and should never // be set twice. Persistent flags and deadline changes are done from the same // thread so there is no risk of losing concurrently added information. uint64_t old_bits = old_flags | static_cast(old_deadline.ToInternalValue()); const uint64_t desired_bits = old_bits | static_cast(Flag::kShouldBlockOnHang); // If a test needs to simulate |bits_| changing since calling this function // this happens now. if (switch_bits_callback_for_testing_) { const uint64_t switched_in_bits = SwitchBitsForTesting(); // Injecting the flag being tested is invalid. DCHECK(!IsFlagSet(Flag::kShouldBlockOnHang, switched_in_bits)); } return bits_.compare_exchange_weak(old_bits, desired_bits, std::memory_order_relaxed, std::memory_order_relaxed); } void HangWatchDeadline::SetIgnoreCurrentWatchHangsInScope() { SetPersistentFlag(Flag::kIgnoreCurrentWatchHangsInScope); } void HangWatchDeadline::UnsetIgnoreCurrentWatchHangsInScope() { ClearPersistentFlag(Flag::kIgnoreCurrentWatchHangsInScope); } void HangWatchDeadline::SetPersistentFlag(Flag flag) { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); if (switch_bits_callback_for_testing_) SwitchBitsForTesting(); bits_.fetch_or(static_cast(flag), std::memory_order_relaxed); } void HangWatchDeadline::ClearPersistentFlag(Flag flag) { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); if (switch_bits_callback_for_testing_) SwitchBitsForTesting(); bits_.fetch_and(~(static_cast(flag)), std::memory_order_relaxed); } // static uint64_t HangWatchDeadline::ExtractFlags(uint64_t bits) { return bits & kOnlyFlagsMask; } // static uint64_t HangWatchDeadline::ExtractDeadline(uint64_t bits) { return bits & kOnlyDeadlineMask; } // static TimeTicks HangWatchDeadline::DeadlineFromBits(uint64_t bits) { // |kOnlyDeadlineMask| has all the deadline bits set to 1 so is the largest // representable value. DCHECK(bits <= kOnlyDeadlineMask) << "Flags bits are set. Remove them before returning deadline."; static_assert(kOnlyDeadlineMask <= std::numeric_limits::max()); return TimeTicks::FromInternalValue(static_cast(bits)); } bool HangWatchDeadline::IsFlagSet(Flag flag) const { return bits_.load(std::memory_order_relaxed) & static_cast(flag); } void HangWatchDeadline::SetSwitchBitsClosureForTesting( RepeatingCallback closure) { switch_bits_callback_for_testing_ = closure; } void HangWatchDeadline::ResetSwitchBitsClosureForTesting() { DCHECK(switch_bits_callback_for_testing_); switch_bits_callback_for_testing_.Reset(); } uint64_t HangWatchDeadline::SwitchBitsForTesting() { DCHECK(switch_bits_callback_for_testing_); const uint64_t old_bits = bits_.load(std::memory_order_relaxed); const uint64_t new_bits = switch_bits_callback_for_testing_.Run(); const uint64_t old_flags = ExtractFlags(old_bits); const uint64_t switched_in_bits = old_flags | new_bits; bits_.store(switched_in_bits, std::memory_order_relaxed); return switched_in_bits; } HangWatchState::HangWatchState(HangWatcher::ThreadType thread_type) : resetter_(&hang_watch_state, this, nullptr), thread_type_(thread_type) { // TODO(crbug.com/1223033): Remove this once macOS uses system-wide ids. // On macOS the thread ids used by CrashPad are not the same as the ones // provided by PlatformThread. Make sure to use the same for correct // attribution. #if BUILDFLAG(IS_MAC) uint64_t thread_id; pthread_threadid_np(pthread_self(), &thread_id); thread_id_ = checked_cast(thread_id); #else thread_id_ = PlatformThread::CurrentId(); #endif } HangWatchState::~HangWatchState() { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); DCHECK_EQ(GetHangWatchStateForCurrentThread(), this); #if DCHECK_IS_ON() // Destroying the HangWatchState should not be done if there are live // WatchHangsInScopes. DCHECK(!current_watch_hangs_in_scope_); #endif } // static std::unique_ptr HangWatchState::CreateHangWatchStateForCurrentThread( HangWatcher::ThreadType thread_type) { // Allocate a watch state object for this thread. std::unique_ptr hang_state = std::make_unique(thread_type); // Setting the thread local worked. DCHECK_EQ(GetHangWatchStateForCurrentThread(), hang_state.get()); // Transfer ownership to caller. return hang_state; } TimeTicks HangWatchState::GetDeadline() const { return deadline_.GetDeadline(); } std::pair HangWatchState::GetFlagsAndDeadline() const { return deadline_.GetFlagsAndDeadline(); } void HangWatchState::SetDeadline(TimeTicks deadline) { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); deadline_.SetDeadline(deadline); } bool HangWatchState::IsOverDeadline() const { return TimeTicks::Now() > deadline_.GetDeadline(); } void HangWatchState::SetIgnoreCurrentWatchHangsInScope() { deadline_.SetIgnoreCurrentWatchHangsInScope(); } void HangWatchState::UnsetIgnoreCurrentWatchHangsInScope() { deadline_.UnsetIgnoreCurrentWatchHangsInScope(); } bool HangWatchState::SetShouldBlockOnHang(uint64_t old_flags, TimeTicks old_deadline) { return deadline_.SetShouldBlockOnHang(old_flags, old_deadline); } bool HangWatchState::IsFlagSet(HangWatchDeadline::Flag flag) { return deadline_.IsFlagSet(flag); } #if DCHECK_IS_ON() void HangWatchState::SetCurrentWatchHangsInScope( WatchHangsInScope* current_hang_watch_scope_enable) { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); current_watch_hangs_in_scope_ = current_hang_watch_scope_enable; } WatchHangsInScope* HangWatchState::GetCurrentWatchHangsInScope() { DCHECK_CALLED_ON_VALID_THREAD(thread_checker_); return current_watch_hangs_in_scope_; } #endif HangWatchDeadline* HangWatchState::GetHangWatchDeadlineForTesting() { return &deadline_; } void HangWatchState::IncrementNestingLevel() { ++nesting_level_; } void HangWatchState::DecrementNestingLevel() { --nesting_level_; } // static HangWatchState* HangWatchState::GetHangWatchStateForCurrentThread() { // Workaround false-positive MSAN use-of-uninitialized-value on // thread_local storage for loaded libraries: // https://github.com/google/sanitizers/issues/1265 MSAN_UNPOISON(&hang_watch_state, sizeof(internal::HangWatchState*)); return hang_watch_state; } PlatformThreadId HangWatchState::GetThreadID() const { return thread_id_; } } // namespace internal } // namespace base