// Parallelism options. static constexpr bool kParallelCardScan = true; static constexpr bool kParallelRecursiveMark = true; // Don't attempt to parallelize mark stack processing unless the mark stack is at least n // elements. This is temporary until we reduce the overhead caused by allocating tasks, etc.. Not // having this can add overhead in ProcessReferences since we may end up doing many calls of // ProcessMarkStack with very small mark stacks. static constexpr size_t kMinimumParallelMarkStackSize = 128; static constexpr bool kParallelProcessMarkStack = true;
// Turn off kCheckLocks when profiling the GC since it slows the GC down by up to 40%. static constexpr bool kCheckLocks = kDebugLocking; static constexpr bool kVerifyRootsMarked = kIsDebugBuild;
// If true, revoke the rosalloc thread-local buffers at the // checkpoint, as opposed to during the pause. static constexpr bool kRevokeRosAllocThreadLocalBuffersAtCheckpoint = true;
void MarkSweep::BindBitmaps() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
WriterMutexLock mu(Thread::Current(), *Locks::heap_bitmap_lock_); // Mark all of the spaces we never collect as immune. for (constauto& space : GetHeap()->GetContinuousSpaces()) { if (space->GetGcRetentionPolicy() == space::kGcRetentionPolicyNeverCollect) {
immune_spaces_.AddSpace(space);
}
}
}
void MarkSweep::PausePhase() {
TimingLogger::ScopedTiming t("(Paused)PausePhase", GetTimings());
Thread* self = Thread::Current();
Locks::mutator_lock_->AssertExclusiveHeld(self); if (IsConcurrent()) { // Handle the dirty objects if we are a concurrent GC.
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_); // Re-mark root set.
ReMarkRoots(); // Scan dirty objects, this is only required if we are doing concurrent GC.
RecursiveMarkDirtyObjects(true, accounting::CardTable::kCardDirty);
}
{
TimingLogger::ScopedTiming t2("SwapStacks", GetTimings());
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
heap_->SwapStacks();
live_stack_freeze_size_ = heap_->GetLiveStack()->Size(); // Need to revoke all the thread local allocation stacks since we just swapped the allocation // stacks and don't want anybody to allocate into the live stack.
RevokeAllThreadLocalAllocationStacks(self);
}
heap_->PreSweepingGcVerification(this); // Disallow new system weaks to prevent a race which occurs when someone adds a new system // weak before we sweep them. Since this new system weak may not be marked, the GC may // incorrectly sweep it. This also fixes a race where interning may attempt to return a strong // reference to a string that is about to be swept.
Runtime::Current()->DisallowNewSystemWeaks(); // Enable the reference processing slow path, needs to be done with mutators paused since there // is no lock in the GetReferent fast path.
ReferenceProcessor* rp = GetHeap()->GetReferenceProcessor();
rp->Setup(self, this, /*concurrent=*/true, GetCurrentIteration()->GetClearSoftReferences());
rp->EnableSlowPath();
}
void MarkSweep::PreCleanCards() { // Don't do this for non concurrent GCs since they don't have any dirty cards. if (kPreCleanCards && IsConcurrent()) {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
Thread* self = Thread::Current();
CHECK(!Locks::mutator_lock_->IsExclusiveHeld(self)); // Process dirty cards and add dirty cards to mod union tables, also ages cards.
heap_->ProcessCards(GetTimings(), false, true, false); // The checkpoint root marking is required to avoid a race condition which occurs if the // following happens during a reference write: // 1. mutator dirties the card (write barrier) // 2. GC ages the card (the above ProcessCards call) // 3. GC scans the object (the RecursiveMarkDirtyObjects call below) // 4. mutator writes the value (corresponding to the write barrier in 1.) // This causes the GC to age the card but not necessarily mark the reference which the mutator // wrote into the object stored in the card. // Having the checkpoint fixes this issue since it ensures that the card mark and the // reference write are visible to the GC before the card is scanned (this is due to locks being // acquired / released in the checkpoint code). // The other roots are also marked to help reduce the pause.
MarkRootsCheckpoint(self, false);
MarkNonThreadRoots();
MarkConcurrentRoots( static_cast<VisitRootFlags>(kVisitRootFlagClearRootLog | kVisitRootFlagNewRoots)); // Process the newly aged cards.
RecursiveMarkDirtyObjects(false, accounting::CardTable::kCardDirty - 1); // TODO: Empty allocation stack to reduce the number of objects we need to test / mark as live // in the next GC.
}
}
void MarkSweep::MarkingPhase() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
Thread* self = Thread::Current();
BindBitmaps();
FindDefaultSpaceBitmap(); // Process dirty cards and add dirty cards to mod union tables. // If the GC type is non sticky, then we just clear the cards of the // alloc space instead of aging them. // // Note that it is fine to clear the cards of the alloc space here, // in the case of a concurrent (non-sticky) mark-sweep GC (whose // marking phase _is_ performed concurrently with mutator threads // running and possibly dirtying cards), as the whole alloc space // will be traced in that case, starting *after* this call to // Heap::ProcessCards (see calls to MarkSweep::MarkRoots and // MarkSweep::MarkReachableObjects). References held by objects on // cards that became dirty *after* the actual marking work started // will be marked in the pause (see MarkSweep::PausePhase), in a // *non-concurrent* way to prevent races with mutator threads. // // TODO: Do we need some sort of fence between the call to // Heap::ProcessCard and the calls to MarkSweep::MarkRoot / // MarkSweep::MarkReachableObjects below to make sure write // operations in the card table clearing the alloc space's dirty // cards (during the call to Heap::ProcessCard) are not reordered // *after* marking actually starts?
heap_->ProcessCards(GetTimings(), /* use_rem_sets= */ false, /* process_alloc_space_cards= */ true, /* clear_alloc_space_cards= */ GetGcType() != kGcTypeSticky);
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
MarkRoots(self);
MarkReachableObjects(); // Pre-clean dirtied cards to reduce pauses.
PreCleanCards();
}
void MarkSweep::UpdateAndMarkModUnion() { for (constauto& space : immune_spaces_.GetSpaces()) { constchar* name = space->IsZygoteSpace()
? "UpdateAndMarkZygoteModUnionTable"
: "UpdateAndMarkImageModUnionTable";
DCHECK(space->IsZygoteSpace() || space->IsImageSpace()) << *space;
TimingLogger::ScopedTiming t(name, GetTimings());
accounting::ModUnionTable* mod_union_table = heap_->FindModUnionTableFromSpace(space); if (mod_union_table != nullptr) {
mod_union_table->UpdateAndMarkReferences(this);
} else { // No mod-union table, scan all the live bits. This can only occur for app images.
space->GetLiveBitmap()->VisitMarkedRange(reinterpret_cast<uintptr_t>(space->Begin()), reinterpret_cast<uintptr_t>(space->End()),
ScanObjectVisitor(this));
}
}
}
void MarkSweep::MarkReachableObjects() {
UpdateAndMarkModUnion(); // Recursively mark all the non-image bits set in the mark bitmap.
RecursiveMark();
}
void MarkSweep::ReclaimPhase() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
Thread* const self = Thread::Current(); // Process the references concurrently.
ProcessReferences(self); // There is no need to sweep interpreter caches as this GC doesn't move // objects and hence would be a nop.
SweepSystemWeaks(self);
Runtime* const runtime = Runtime::Current();
runtime->AllowNewSystemWeaks(); // Clean up class loaders after system weaks are swept since that is how we know if class // unloading occurred.
runtime->GetClassLinker()->CleanupClassLoaders();
{
WriterMutexLock mu(self, *Locks::heap_bitmap_lock_);
GetHeap()->RecordFreeRevoke(); // Reclaim unmarked objects.
Sweep(false); // Swap the live and mark bitmaps for each space which we modified space. This is an // optimization that enables us to not clear live bits inside of the sweep. Only swaps unbound // bitmaps.
SwapBitmaps(); // Unbind the live and mark bitmaps.
GetHeap()->UnBindBitmaps();
}
}
void MarkSweep::FindDefaultSpaceBitmap() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); for (constauto& space : GetHeap()->GetContinuousSpaces()) {
accounting::ContinuousSpaceBitmap* bitmap = space->GetMarkBitmap(); // We want to have the main space instead of non moving if possible. if (bitmap != nullptr &&
space->GetGcRetentionPolicy() == space::kGcRetentionPolicyAlwaysCollect) {
current_space_bitmap_ = bitmap; // If we are not the non moving space exit the loop early since this will be good enough. if (space != heap_->GetNonMovingSpace()) { break;
}
}
}
CHECK(current_space_bitmap_ != nullptr) << "Could not find a default mark bitmap\n"
<< heap_->DumpSpaces();
}
void MarkSweep::ResizeMarkStack(size_t new_size) { // Rare case, no need to have Thread::Current be a parameter. if (UNLIKELY(mark_stack_->Size() < mark_stack_->Capacity())) { // Someone else acquired the lock and expanded the mark stack before us. return;
}
std::vector<StackReference<mirror::Object>> temp(mark_stack_->Begin(), mark_stack_->End());
CHECK_LE(mark_stack_->Size(), new_size);
mark_stack_->Resize(new_size); for (auto& obj : temp) {
mark_stack_->PushBack(obj.AsMirrorPtr());
}
}
inlinevoid MarkSweep::MarkObjectNonNullParallel(mirror::Object* obj) {
DCHECK(obj != nullptr); if (MarkObjectParallel(obj)) {
MutexLock mu(Thread::Current(), mark_stack_lock_); if (UNLIKELY(mark_stack_->Size() >= mark_stack_->Capacity())) {
ExpandMarkStack();
} // The object must be pushed on to the mark stack.
mark_stack_->PushBack(obj);
}
}
inlinevoid MarkSweep::MarkObjectNonNull(mirror::Object* obj,
mirror::Object* holder,
MemberOffset offset) {
DCHECK(obj != nullptr); if (kUseBakerReadBarrier) { // Verify all the objects have the correct state installed.
obj->AssertReadBarrierState();
} if (immune_spaces_.IsInImmuneRegion(obj)) { if (kCountMarkedObjects) {
++mark_immune_count_;
}
DCHECK(mark_bitmap_->Test(obj));
} elseif (LIKELY(current_space_bitmap_->HasAddress(obj))) { if (kCountMarkedObjects) {
++mark_fastpath_count_;
} if (UNLIKELY(!current_space_bitmap_->Set(obj))) {
PushOnMarkStack(obj); // This object was not previously marked.
}
} else { if (kCountMarkedObjects) {
++mark_slowpath_count_;
}
MarkObjectSlowPath visitor(this, holder, offset); // TODO: We already know that the object is not in the current_space_bitmap_ but MarkBitmap::Set // will check again. if (!mark_bitmap_->Set(obj, visitor)) {
PushOnMarkStack(obj); // Was not already marked, push.
}
}
}
inlinevoid MarkSweep::PushOnMarkStack(mirror::Object* obj) { if (UNLIKELY(mark_stack_->Size() >= mark_stack_->Capacity())) { // Lock is not needed but is here anyways to please annotalysis.
MutexLock mu(Thread::Current(), mark_stack_lock_);
ExpandMarkStack();
} // The object must be pushed on to the mark stack.
mark_stack_->PushBack(obj);
}
inlinebool MarkSweep::MarkObjectParallel(mirror::Object* obj) {
DCHECK(obj != nullptr); if (kUseBakerReadBarrier) { // Verify all the objects have the correct state installed.
obj->AssertReadBarrierState();
} if (immune_spaces_.IsInImmuneRegion(obj)) {
DCHECK(IsMarked(obj) != nullptr); returnfalse;
} // Try to take advantage of locality of references within a space, failing this find the space // the hard way.
accounting::ContinuousSpaceBitmap* object_bitmap = current_space_bitmap_; if (LIKELY(object_bitmap->HasAddress(obj))) { return !object_bitmap->AtomicTestAndSet(obj);
}
MarkObjectSlowPath visitor(this); return !mark_bitmap_->AtomicTestAndSet(obj, visitor);
}
// Used to mark objects when processing the mark stack. If an object is null, it is not marked. inlinevoid MarkSweep::MarkObject(mirror::Object* obj,
mirror::Object* holder,
MemberOffset offset) { if (obj != nullptr) {
MarkObjectNonNull(obj, holder, offset);
} elseif (kCountMarkedObjects) {
++mark_null_count_;
}
}
class MarkSweep::VerifyRootMarkedVisitor : public SingleRootVisitor { public: explicit VerifyRootMarkedVisitor(MarkSweep* collector) : collector_(collector) { }
void MarkSweep::MarkRoots(Thread* self) {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); if (Locks::mutator_lock_->IsExclusiveHeld(self)) { // If we exclusively hold the mutator lock, all threads must be suspended.
Runtime::Current()->VisitRoots(this);
RevokeAllThreadLocalAllocationStacks(self);
} else {
MarkRootsCheckpoint(self, kRevokeRosAllocThreadLocalBuffersAtCheckpoint); // At this point the live stack should no longer have any mutators which push into it.
MarkNonThreadRoots();
MarkConcurrentRoots( static_cast<VisitRootFlags>(kVisitRootFlagAllRoots | kVisitRootFlagStartLoggingNewRoots));
}
}
template <bool kUseFinger = false> class MarkSweep::MarkStackTask : public Task { public:
MarkStackTask(ThreadPool* thread_pool,
MarkSweep* mark_sweep,
size_t mark_stack_size,
StackReference<mirror::Object>* mark_stack)
: mark_sweep_(mark_sweep),
thread_pool_(thread_pool),
mark_stack_pos_(mark_stack_size) { // We may have to copy part of an existing mark stack when another mark stack overflows. if (mark_stack_size != 0) {
DCHECK(mark_stack != nullptr); // TODO: Check performance?
std::copy(mark_stack, mark_stack + mark_stack_size, mark_stack_);
} if (kCountTasks) {
++mark_sweep_->work_chunks_created_;
}
}
virtual ~MarkStackTask() { // Make sure that we have cleared our mark stack.
DCHECK_EQ(mark_stack_pos_, 0U); if (kCountTasks) {
++mark_sweep_->work_chunks_deleted_;
}
}
MarkSweep* const mark_sweep_;
ThreadPool* const thread_pool_; // Thread local mark stack for this task.
StackReference<mirror::Object> mark_stack_[kMaxSize]; // Mark stack position.
size_t mark_stack_pos_;
ALWAYS_INLINE void MarkStackPush(mirror::Object* obj)
REQUIRES_SHARED(Locks::mutator_lock_) { if (UNLIKELY(mark_stack_pos_ == kMaxSize)) { // Mark stack overflow, give 1/2 the stack to the thread pool as a new work task.
mark_stack_pos_ /= 2; auto* task = new MarkStackTask(thread_pool_,
mark_sweep_,
kMaxSize - mark_stack_pos_,
mark_stack_ + mark_stack_pos_);
thread_pool_->AddTask(Thread::Current(), task);
}
DCHECK(obj != nullptr);
DCHECK_LT(mark_stack_pos_, kMaxSize);
mark_stack_[mark_stack_pos_++].Assign(obj);
}
size_t MarkSweep::GetThreadCount(bool paused) const { // Use less threads if we are in a background state (non jank perceptible) since we want to leave // more CPU time for the foreground apps. if (heap_->GetThreadPool() == nullptr || !Runtime::Current()->InJankPerceptibleProcessState()) { return1;
} return (paused ? heap_->GetParallelGCThreadCount() : heap_->GetConcGCThreadCount()) + 1;
}
void MarkSweep::ScanGrayObjects(bool paused, uint8_t minimum_age) {
accounting::CardTable* card_table = GetHeap()->GetCardTable();
ThreadPool* thread_pool = GetHeap()->GetThreadPool();
size_t thread_count = GetThreadCount(paused); // The parallel version with only one thread is faster for card scanning, TODO: fix. if (kParallelCardScan && thread_count > 1) {
Thread* self = Thread::Current(); // Can't have a different split for each space since multiple spaces can have their cards being // scanned at the same time.
TimingLogger::ScopedTiming t(paused ? "(Paused)ScanGrayObjects" : __FUNCTION__,
GetTimings()); // Try to take some of the mark stack since we can pass this off to the worker tasks.
StackReference<mirror::Object>* mark_stack_begin = mark_stack_->Begin();
StackReference<mirror::Object>* mark_stack_end = mark_stack_->End(); const size_t mark_stack_size = mark_stack_end - mark_stack_begin; // Estimated number of work tasks we will create. const size_t mark_stack_tasks = GetHeap()->GetContinuousSpaces().size() * thread_count;
DCHECK_NE(mark_stack_tasks, 0U); const size_t mark_stack_delta = std::min(CardScanTask::kMaxSize / 2,
mark_stack_size / mark_stack_tasks + 1); for (constauto& space : GetHeap()->GetContinuousSpaces()) { if (space->GetMarkBitmap() == nullptr) { continue;
}
uint8_t* card_begin = space->Begin();
uint8_t* card_end = space->End(); // Align up the end address. For example, the image space's end // may not be card-size-aligned.
card_end = AlignUp(card_end, accounting::CardTable::kCardSize);
DCHECK_ALIGNED(card_begin, accounting::CardTable::kCardSize);
DCHECK_ALIGNED(card_end, accounting::CardTable::kCardSize); // Calculate how many bytes of heap we will scan, const size_t address_range = card_end - card_begin; // Calculate how much address range each task gets. const size_t card_delta = RoundUp(address_range / thread_count + 1,
accounting::CardTable::kCardSize); // If paused and the space is neither zygote nor image space, we could clear the dirty // cards to avoid accumulating them to increase card scanning load in the following GC // cycles. We need to keep dirty cards of image space and zygote space in order to track // references to the other spaces. bool clear_card = paused && !space->IsZygoteSpace() && !space->IsImageSpace(); // Create the worker tasks for this space. while (card_begin != card_end) { // Add a range of cards.
size_t addr_remaining = card_end - card_begin;
size_t card_increment = std::min(card_delta, addr_remaining); // Take from the back of the mark stack.
size_t mark_stack_remaining = mark_stack_end - mark_stack_begin;
size_t mark_stack_increment = std::min(mark_stack_delta, mark_stack_remaining);
mark_stack_end -= mark_stack_increment;
mark_stack_->PopBackCount(static_cast<int32_t>(mark_stack_increment));
DCHECK_EQ(mark_stack_end, mark_stack_->End()); // Add the new task to the thread pool. auto* task = new CardScanTask(thread_pool, this,
space->GetMarkBitmap(),
card_begin,
card_begin + card_increment,
minimum_age,
mark_stack_increment,
mark_stack_end,
clear_card);
thread_pool->AddTask(self, task);
card_begin += card_increment;
}
}
// Note: the card scan below may dirty new cards (and scan them) // as a side effect when a Reference object is encountered and // queued during the marking. See b/11465268.
thread_pool->SetMaxActiveWorkers(thread_count - 1);
thread_pool->StartWorkers(self);
thread_pool->Wait(self, true, true);
thread_pool->StopWorkers(self);
} else { for (constauto& space : GetHeap()->GetContinuousSpaces()) { if (space->GetMarkBitmap() != nullptr) { // Image spaces are handled properly since live == marked for them. constchar* name = nullptr; switch (space->GetGcRetentionPolicy()) { case space::kGcRetentionPolicyNeverCollect:
name = paused ? "(Paused)ScanGrayImageSpaceObjects" : "ScanGrayImageSpaceObjects"; break; case space::kGcRetentionPolicyFullCollect:
name = paused ? "(Paused)ScanGrayZygoteSpaceObjects" : "ScanGrayZygoteSpaceObjects"; break; case space::kGcRetentionPolicyAlwaysCollect:
name = paused ? "(Paused)ScanGrayAllocSpaceObjects" : "ScanGrayAllocSpaceObjects"; break;
}
TimingLogger::ScopedTiming t(name, GetTimings());
ScanObjectVisitor visitor(this); bool clear_card = paused && !space->IsZygoteSpace() && !space->IsImageSpace(); if (clear_card) {
card_table->Scan<true>(space->GetMarkBitmap(),
space->Begin(),
space->End(),
visitor,
minimum_age);
} else {
card_table->Scan<false>(space->GetMarkBitmap(),
space->Begin(),
space->End(),
visitor,
minimum_age);
}
}
}
}
}
// Scans all of the objects void Run(Thread* self) override NO_THREAD_SAFETY_ANALYSIS {
ScanObjectParallelVisitor visitor(this);
bitmap_->VisitMarkedRange(begin_, end_, visitor); // Finish by emptying our local mark stack.
MarkStackTask::Run(self);
}
};
// Populates the mark stack based on the set of marked objects and // recursively marks until the mark stack is emptied. void MarkSweep::RecursiveMark() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); // RecursiveMark will build the lists of known instances of the Reference classes. See // DelayReferenceReferent for details. if (kUseRecursiveMark) { constbool partial = GetGcType() == kGcTypePartial;
ScanObjectVisitor scan_visitor(this); auto* self = Thread::Current();
ThreadPool* thread_pool = heap_->GetThreadPool();
size_t thread_count = GetThreadCount(false); constbool parallel = kParallelRecursiveMark && thread_count > 1;
mark_stack_->Reset(); for (constauto& space : GetHeap()->GetContinuousSpaces()) { if ((space->GetGcRetentionPolicy() == space::kGcRetentionPolicyAlwaysCollect) ||
(!partial && space->GetGcRetentionPolicy() == space::kGcRetentionPolicyFullCollect)) {
current_space_bitmap_ = space->GetMarkBitmap(); if (current_space_bitmap_ == nullptr) { continue;
} if (parallel) { // We will use the mark stack the future. // CHECK(mark_stack_->IsEmpty()); // This function does not handle heap end increasing, so we must use the space end.
uintptr_t begin = reinterpret_cast<uintptr_t>(space->Begin());
uintptr_t end = reinterpret_cast<uintptr_t>(space->End());
atomic_finger_.store(AtomicInteger::MaxValue(), std::memory_order_relaxed);
// Create a few worker tasks. const size_t n = thread_count * 2; while (begin != end) {
uintptr_t start = begin;
uintptr_t delta = (end - begin) / n;
delta = RoundUp(delta, KB); if (delta < 16 * KB) delta = end - begin;
begin += delta; auto* task = new RecursiveMarkTask(thread_pool, this,
current_space_bitmap_,
start,
begin);
thread_pool->AddTask(self, task);
}
thread_pool->SetMaxActiveWorkers(thread_count - 1);
thread_pool->StartWorkers(self);
thread_pool->Wait(self, true, true);
thread_pool->StopWorkers(self);
} else { // This function does not handle heap end increasing, so we must use the space end.
uintptr_t begin = reinterpret_cast<uintptr_t>(space->Begin());
uintptr_t end = reinterpret_cast<uintptr_t>(space->End());
current_space_bitmap_->VisitMarkedRange(begin, end, scan_visitor);
}
}
}
}
ProcessMarkStack(false);
}
void MarkSweep::VerifyIsLive(const mirror::Object* obj) { if (!heap_->GetLiveBitmap()->Test(obj)) { // TODO: Consider live stack? Has this code bitrotted?
CHECK(!heap_->allocation_stack_->Contains(obj))
<< "Found dead object " << obj << "\n" << heap_->DumpSpaces();
}
}
void MarkSweep::VerifySystemWeaks() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); // Verify system weaks, uses a special object visitor which returns the input object.
VerifySystemWeakVisitor visitor(this);
Runtime* runtime = Runtime::Current();
runtime->SweepSystemWeaks(&visitor);
}
class MarkSweep::CheckpointMarkThreadRoots : public Closure, public RootVisitor { public:
CheckpointMarkThreadRoots(MarkSweep* mark_sweep, bool revoke_ros_alloc_thread_local_buffers_at_checkpoint)
: mark_sweep_(mark_sweep),
revoke_ros_alloc_thread_local_buffers_at_checkpoint_(
revoke_ros_alloc_thread_local_buffers_at_checkpoint) {
}
void VisitRoots(mirror::Object*** roots,
size_t count,
[[maybe_unused]] const RootInfo& info) override
REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(Locks::heap_bitmap_lock_) { for (size_t i = 0; i < count; ++i) {
mark_sweep_->MarkObjectNonNullParallel(*roots[i]);
}
}
void VisitRoots(mirror::CompressedReference<mirror::Object>** roots,
size_t count,
[[maybe_unused]] const RootInfo& info) override
REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(Locks::heap_bitmap_lock_) { for (size_t i = 0; i < count; ++i) {
mark_sweep_->MarkObjectNonNullParallel(roots[i]->AsMirrorPtr());
}
}
void Run(Thread* thread) override NO_THREAD_SAFETY_ANALYSIS {
ScopedTrace trace("Marking thread roots"); // Note: self is not necessarily equal to thread since thread may be suspended.
Thread* const self = Thread::Current();
CHECK(thread == self ||
thread->IsSuspended() ||
thread->GetState() == ThreadState::kWaitingPerformingGc)
<< thread->GetState() << " thread " << thread << " self " << self;
thread->VisitRoots(this, kVisitRootFlagAllRoots); if (revoke_ros_alloc_thread_local_buffers_at_checkpoint_) {
ScopedTrace trace2("RevokeRosAllocThreadLocalBuffers");
mark_sweep_->GetHeap()->RevokeRosAllocThreadLocalBuffers(thread);
} // If thread is a running mutator, then act on behalf of the garbage collector. // See the code in ThreadList::RunCheckpoint.
mark_sweep_->GetBarrier().Pass(self);
}
void MarkSweep::MarkRootsCheckpoint(Thread* self, bool revoke_ros_alloc_thread_local_buffers_at_checkpoint) {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
CheckpointMarkThreadRoots check_point(this, revoke_ros_alloc_thread_local_buffers_at_checkpoint);
ThreadList* thread_list = Runtime::Current()->GetThreadList(); // Request the check point is run on all threads returning a count of the threads that must // run through the barrier including self.
size_t barrier_count = thread_list->RunCheckpoint(&check_point); // Release locks then wait for all mutator threads to pass the barrier. // If there are no threads to wait which implys that all the checkpoint functions are finished, // then no need to release locks. if (barrier_count == 0) { return;
}
Locks::heap_bitmap_lock_->ExclusiveUnlock(self);
Locks::mutator_lock_->SharedUnlock(self);
{
ScopedThreadStateChange tsc(self, ThreadState::kWaitingForCheckPointsToRun);
gc_barrier_->Increment(self, barrier_count);
}
Locks::mutator_lock_->SharedLock(self);
Locks::heap_bitmap_lock_->ExclusiveLock(self);
}
void MarkSweep::SweepArray(accounting::ObjectStack* obj_arr, bool swap_bitmaps) {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); // Change the order to ensure that the non-moving space last swept as an optimization.
std::vector<space::ContinuousSpace*> sweep_spaces;
space::ContinuousSpace* non_moving_space = nullptr; for (space::ContinuousSpace* space : heap_->GetContinuousSpaces()) { if (space->IsAllocSpace() &&
!immune_spaces_.ContainsSpace(space) &&
space->GetLiveBitmap() != nullptr) { if (space == heap_->GetNonMovingSpace()) {
non_moving_space = space;
} else {
sweep_spaces.push_back(space);
}
}
} // Unlikely to sweep a significant amount of non_movable objects, so we do these after // the other alloc spaces as an optimization. if (non_moving_space != nullptr) {
sweep_spaces.push_back(non_moving_space);
}
GarbageCollector::SweepArray(obj_arr, swap_bitmaps, &sweep_spaces);
}
void MarkSweep::Sweep(bool swap_bitmaps) {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); // Ensure that nobody inserted items in the live stack after we swapped the stacks.
CHECK_GE(live_stack_freeze_size_, GetHeap()->GetLiveStack()->Size());
{
TimingLogger::ScopedTiming t2("MarkAllocStackAsLive", GetTimings()); // Mark everything allocated since the last GC as live so that we can sweep concurrently, // knowing that new allocations won't be marked as live.
accounting::ObjectStack* live_stack = heap_->GetLiveStack();
heap_->MarkAllocStackAsLive(live_stack);
live_stack->Reset();
DCHECK(mark_stack_->IsEmpty());
} for (constauto& space : GetHeap()->GetContinuousSpaces()) { if (space->IsContinuousMemMapAllocSpace()) {
space::ContinuousMemMapAllocSpace* alloc_space = space->AsContinuousMemMapAllocSpace();
TimingLogger::ScopedTiming split(
alloc_space->IsZygoteSpace() ? "SweepZygoteSpace" : "SweepMallocSpace",
GetTimings());
RecordFree(alloc_space->Sweep(swap_bitmaps));
}
}
SweepLargeObjects(swap_bitmaps);
}
void MarkSweep::SweepLargeObjects(bool swap_bitmaps) {
space::LargeObjectSpace* los = heap_->GetLargeObjectsSpace(); if (los != nullptr) {
TimingLogger::ScopedTiming split(__FUNCTION__, GetTimings());
RecordFreeLOS(los->Sweep(swap_bitmaps));
}
}
// Process the "referent" field lin a java.lang.ref.Reference. If the referent has not yet been // marked, put it on the appropriate list in the heap for later processing. void MarkSweep::DelayReferenceReferent(ObjPtr<mirror::Class> klass, ObjPtr<mirror::Reference> ref) {
heap_->GetReferenceProcessor()->DelayReferenceReferent(klass, ref, this);
}
// Scans an object reference. Determines the type of the reference // and dispatches to a specialized scanning routine. void MarkSweep::ScanObject(mirror::Object* obj) {
MarkVisitor mark_visitor(this);
DelayReferenceReferentVisitor ref_visitor(this);
ScanObjectVisit(obj, mark_visitor, ref_visitor);
}
void MarkSweep::ProcessMarkStackParallel(size_t thread_count) {
Thread* self = Thread::Current();
ThreadPool* thread_pool = GetHeap()->GetThreadPool(); const size_t chunk_size = std::min(mark_stack_->Size() / thread_count + 1, static_cast<size_t>(MarkStackTask<false>::kMaxSize));
CHECK_GT(chunk_size, 0U); // Split the current mark stack up into work tasks. for (auto* it = mark_stack_->Begin(), *end = mark_stack_->End(); it < end; ) { const size_t delta = std::min(static_cast<size_t>(end - it), chunk_size);
thread_pool->AddTask(self, new MarkStackTask<false>(thread_pool, this, delta, it));
it += delta;
}
thread_pool->SetMaxActiveWorkers(thread_count - 1);
thread_pool->StartWorkers(self);
thread_pool->Wait(self, true, true);
thread_pool->StopWorkers(self);
mark_stack_->Reset();
CHECK_EQ(work_chunks_created_.load(std::memory_order_seq_cst),
work_chunks_deleted_.load(std::memory_order_seq_cst))
<< " some of the work chunks were leaked";
}
inline mirror::Object* MarkSweep::IsMarked(mirror::Object* object) { if (immune_spaces_.IsInImmuneRegion(object)) { return object;
} if (current_space_bitmap_->HasAddress(object)) { return current_space_bitmap_->Test(object) ? object : nullptr;
} // This function returns nullptr for objects allocated after marking phase as // they are not marked in the bitmap. return mark_bitmap_->Test(object) ? object : nullptr;
}
void MarkSweep::FinishPhase() {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings()); if (kCountScannedTypes) {
VLOG(gc)
<< "MarkSweep scanned"
<< " no reference objects=" << no_reference_class_count_.load(std::memory_order_relaxed)
<< " normal objects=" << normal_count_.load(std::memory_order_relaxed)
<< " classes=" << class_count_.load(std::memory_order_relaxed)
<< " object arrays=" << object_array_count_.load(std::memory_order_relaxed)
<< " references=" << reference_count_.load(std::memory_order_relaxed)
<< " other=" << other_count_.load(std::memory_order_relaxed);
} if (kCountTasks) {
VLOG(gc)
<< "Total number of work chunks allocated: "
<< work_chunks_created_.load(std::memory_order_relaxed);
} if (kMeasureOverhead) {
VLOG(gc) << "Overhead time " << PrettyDuration(overhead_time_.load(std::memory_order_relaxed));
} if (kProfileLargeObjects) {
VLOG(gc)
<< "Large objects tested " << large_object_test_.load(std::memory_order_relaxed)
<< " marked " << large_object_mark_.load(std::memory_order_relaxed);
} if (kCountMarkedObjects) {
VLOG(gc)
<< "Marked: null=" << mark_null_count_.load(std::memory_order_relaxed)
<< " immune=" << mark_immune_count_.load(std::memory_order_relaxed)
<< " fastpath=" << mark_fastpath_count_.load(std::memory_order_relaxed)
<< " slowpath=" << mark_slowpath_count_.load(std::memory_order_relaxed);
}
CHECK(mark_stack_->IsEmpty()); // Ensure that the mark stack is empty.
mark_stack_->Reset();
Thread* const self = Thread::Current();
ReaderMutexLock mu(self, *Locks::mutator_lock_);
WriterMutexLock mu2(self, *Locks::heap_bitmap_lock_);
heap_->ClearMarkedObjects();
}
void MarkSweep::RevokeAllThreadLocalBuffers() { if (kRevokeRosAllocThreadLocalBuffersAtCheckpoint && IsConcurrent()) { // If concurrent, rosalloc thread-local buffers are revoked at the // thread checkpoint. Bump pointer space thread-local buffers must // not be in use.
GetHeap()->AssertAllBumpPointerSpaceThreadLocalBuffersAreRevoked();
} else {
TimingLogger::ScopedTiming t(__FUNCTION__, GetTimings());
GetHeap()->RevokeAllThreadLocalBuffers();
}
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.