// Copyright 2015 the V8 project authors. All rights reserved. // Use of this source code is governed by a BSD-style license that can be // found in the LICENSE file. #include "src/heap/memory-reducer.h" #include "src/flags.h" #include "src/heap/gc-tracer.h" #include "src/heap/heap-inl.h" #include "src/utils.h" #include "src/v8.h" namespace v8 { namespace internal { const int MemoryReducer::kLongDelayMs = 8000; const int MemoryReducer::kShortDelayMs = 500; const int MemoryReducer::kWatchdogDelayMs = 100000; const int MemoryReducer::kMaxNumberOfGCs = 3; const double MemoryReducer::kCommittedMemoryFactor = 1.1; const size_t MemoryReducer::kCommittedMemoryDelta = 10 * MB; MemoryReducer::TimerTask::TimerTask(MemoryReducer* memory_reducer) : CancelableTask(memory_reducer->heap()->isolate()), memory_reducer_(memory_reducer) {} void MemoryReducer::TimerTask::RunInternal() { Heap* heap = memory_reducer_->heap(); Event event; double time_ms = heap->MonotonicallyIncreasingTimeInMs(); heap->tracer()->SampleAllocation(time_ms, heap->NewSpaceAllocationCounter(), heap->OldGenerationAllocationCounter()); bool low_allocation_rate = heap->HasLowAllocationRate(); bool optimize_for_memory = heap->ShouldOptimizeForMemoryUsage(); if (FLAG_trace_gc_verbose) { heap->isolate()->PrintWithTimestamp( "Memory reducer: %s, %s\n", low_allocation_rate ? "low alloc" : "high alloc", optimize_for_memory ? "background" : "foreground"); } event.type = kTimer; event.time_ms = time_ms; // The memory reducer will start incremental markig if // 1) mutator is likely idle: js call rate is low and allocation rate is low. // 2) mutator is in background: optimize for memory flag is set. event.should_start_incremental_gc = low_allocation_rate || optimize_for_memory; event.can_start_incremental_gc = heap->incremental_marking()->IsStopped() && (heap->incremental_marking()->CanBeActivated() || optimize_for_memory); event.committed_memory = heap->CommittedOldGenerationMemory(); memory_reducer_->NotifyTimer(event); } void MemoryReducer::NotifyTimer(const Event& event) { DCHECK_EQ(kTimer, event.type); DCHECK_EQ(kWait, state_.action); state_ = Step(state_, event); if (state_.action == kRun) { DCHECK(heap()->incremental_marking()->IsStopped()); DCHECK(FLAG_incremental_marking); if (FLAG_trace_gc_verbose) { heap()->isolate()->PrintWithTimestamp("Memory reducer: started GC #%d\n", state_.started_gcs); } heap()->StartIdleIncrementalMarking( GarbageCollectionReason::kMemoryReducer); } else if (state_.action == kWait) { if (!heap()->incremental_marking()->IsStopped() && heap()->ShouldOptimizeForMemoryUsage()) { // Make progress with pending incremental marking if memory usage has // higher priority than latency. This is important for background tabs // that do not send idle notifications. const int kIncrementalMarkingDelayMs = 500; double deadline = heap()->MonotonicallyIncreasingTimeInMs() + kIncrementalMarkingDelayMs; heap()->incremental_marking()->AdvanceIncrementalMarking( deadline, IncrementalMarking::NO_GC_VIA_STACK_GUARD, IncrementalMarking::FORCE_COMPLETION, StepOrigin::kTask); heap()->FinalizeIncrementalMarkingIfComplete( GarbageCollectionReason::kFinalizeMarkingViaTask); } // Re-schedule the timer. ScheduleTimer(event.time_ms, state_.next_gc_start_ms - event.time_ms); if (FLAG_trace_gc_verbose) { heap()->isolate()->PrintWithTimestamp( "Memory reducer: waiting for %.f ms\n", state_.next_gc_start_ms - event.time_ms); } } } void MemoryReducer::NotifyMarkCompact(const Event& event) { DCHECK_EQ(kMarkCompact, event.type); Action old_action = state_.action; state_ = Step(state_, event); if (old_action != kWait && state_.action == kWait) { // If we are transitioning to the WAIT state, start the timer. ScheduleTimer(event.time_ms, state_.next_gc_start_ms - event.time_ms); } if (old_action == kRun) { if (FLAG_trace_gc_verbose) { heap()->isolate()->PrintWithTimestamp( "Memory reducer: finished GC #%d (%s)\n", state_.started_gcs, state_.action == kWait ? "will do more" : "done"); } } } void MemoryReducer::NotifyPossibleGarbage(const Event& event) { DCHECK_EQ(kPossibleGarbage, event.type); Action old_action = state_.action; state_ = Step(state_, event); if (old_action != kWait && state_.action == kWait) { // If we are transitioning to the WAIT state, start the timer. ScheduleTimer(event.time_ms, state_.next_gc_start_ms - event.time_ms); } } bool MemoryReducer::WatchdogGC(const State& state, const Event& event) { return state.last_gc_time_ms != 0 && event.time_ms > state.last_gc_time_ms + kWatchdogDelayMs; } // For specification of this function see the comment for MemoryReducer class. MemoryReducer::State MemoryReducer::Step(const State& state, const Event& event) { if (!FLAG_incremental_marking || !FLAG_memory_reducer) { return State(kDone, 0, 0, state.last_gc_time_ms, 0); } switch (state.action) { case kDone: if (event.type == kTimer) { return state; } else if (event.type == kMarkCompact) { if (event.committed_memory < Max(static_cast<size_t>(state.committed_memory_at_last_run * kCommittedMemoryFactor), state.committed_memory_at_last_run + kCommittedMemoryDelta)) { return state; } else { return State(kWait, 0, event.time_ms + kLongDelayMs, event.type == kMarkCompact ? event.time_ms : state.last_gc_time_ms, 0); } } else { DCHECK_EQ(kPossibleGarbage, event.type); return State( kWait, 0, event.time_ms + kLongDelayMs, event.type == kMarkCompact ? event.time_ms : state.last_gc_time_ms, 0); } case kWait: switch (event.type) { case kPossibleGarbage: return state; case kTimer: if (state.started_gcs >= kMaxNumberOfGCs) { return State(kDone, kMaxNumberOfGCs, 0.0, state.last_gc_time_ms, event.committed_memory); } else if (event.can_start_incremental_gc && (event.should_start_incremental_gc || WatchdogGC(state, event))) { if (state.next_gc_start_ms <= event.time_ms) { return State(kRun, state.started_gcs + 1, 0.0, state.last_gc_time_ms, 0); } else { return state; } } else { return State(kWait, state.started_gcs, event.time_ms + kLongDelayMs, state.last_gc_time_ms, 0); } case kMarkCompact: return State(kWait, state.started_gcs, event.time_ms + kLongDelayMs, event.time_ms, 0); } case kRun: if (event.type != kMarkCompact) { return state; } else { if (state.started_gcs < kMaxNumberOfGCs && (event.next_gc_likely_to_collect_more || state.started_gcs == 1)) { return State(kWait, state.started_gcs, event.time_ms + kShortDelayMs, event.time_ms, 0); } else { return State(kDone, kMaxNumberOfGCs, 0.0, event.time_ms, event.committed_memory); } } } UNREACHABLE(); return State(kDone, 0, 0, 0.0, 0); // Make the compiler happy. } void MemoryReducer::ScheduleTimer(double time_ms, double delay_ms) { DCHECK(delay_ms > 0); // Leave some room for precision error in task scheduler. const double kSlackMs = 100; v8::Isolate* isolate = reinterpret_cast<v8::Isolate*>(heap()->isolate()); auto timer_task = new MemoryReducer::TimerTask(this); V8::GetCurrentPlatform()->CallDelayedOnForegroundThread( isolate, timer_task, (delay_ms + kSlackMs) / 1000.0); } void MemoryReducer::TearDown() { state_ = State(kDone, 0, 0, 0.0, 0); } } // namespace internal } // namespace v8