// Copyright 2012 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/mark-compact.h"
#include "src/base/atomicops.h"
#include "src/base/bits.h"
#include "src/base/sys-info.h"
#include "src/code-stubs.h"
#include "src/compilation-cache.h"
#include "src/deoptimizer.h"
#include "src/execution.h"
#include "src/frames-inl.h"
#include "src/gdb-jit.h"
#include "src/global-handles.h"
#include "src/heap/array-buffer-tracker.h"
#include "src/heap/gc-tracer.h"
#include "src/heap/incremental-marking.h"
#include "src/heap/mark-compact-inl.h"
#include "src/heap/object-stats.h"
#include "src/heap/objects-visiting.h"
#include "src/heap/objects-visiting-inl.h"
#include "src/heap/slots-buffer.h"
#include "src/heap/spaces-inl.h"
#include "src/ic/ic.h"
#include "src/ic/stub-cache.h"
#include "src/profiler/cpu-profiler.h"
#include "src/v8.h"
namespace v8 {
namespace internal {
const char* Marking::kWhiteBitPattern = "00";
const char* Marking::kBlackBitPattern = "11";
const char* Marking::kGreyBitPattern = "10";
const char* Marking::kImpossibleBitPattern = "01";
// The following has to hold in order for {Marking::MarkBitFrom} to not produce
// invalid {kImpossibleBitPattern} in the marking bitmap by overlapping.
STATIC_ASSERT(Heap::kMinObjectSizeInWords >= 2);
// -------------------------------------------------------------------------
// MarkCompactCollector
MarkCompactCollector::MarkCompactCollector(Heap* heap)
: // NOLINT
#ifdef DEBUG
state_(IDLE),
#endif
marking_parity_(ODD_MARKING_PARITY),
was_marked_incrementally_(false),
evacuation_(false),
slots_buffer_allocator_(nullptr),
migration_slots_buffer_(nullptr),
heap_(heap),
marking_deque_memory_(NULL),
marking_deque_memory_committed_(0),
code_flusher_(nullptr),
have_code_to_deoptimize_(false),
compacting_(false),
sweeping_in_progress_(false),
compaction_in_progress_(false),
pending_sweeper_tasks_semaphore_(0),
pending_compaction_tasks_semaphore_(0) {
}
#ifdef VERIFY_HEAP
class VerifyMarkingVisitor : public ObjectVisitor {
public:
explicit VerifyMarkingVisitor(Heap* heap) : heap_(heap) {}
void VisitPointers(Object** start, Object** end) override {
for (Object** current = start; current < end; current++) {
if ((*current)->IsHeapObject()) {
HeapObject* object = HeapObject::cast(*current);
CHECK(heap_->mark_compact_collector()->IsMarked(object));
}
}
}
void VisitEmbeddedPointer(RelocInfo* rinfo) override {
DCHECK(rinfo->rmode() == RelocInfo::EMBEDDED_OBJECT);
if (!rinfo->host()->IsWeakObject(rinfo->target_object())) {
Object* p = rinfo->target_object();
VisitPointer(&p);
}
}
void VisitCell(RelocInfo* rinfo) override {
Code* code = rinfo->host();
DCHECK(rinfo->rmode() == RelocInfo::CELL);
if (!code->IsWeakObject(rinfo->target_cell())) {
ObjectVisitor::VisitCell(rinfo);
}
}
private:
Heap* heap_;
};
static void VerifyMarking(Heap* heap, Address bottom, Address top) {
VerifyMarkingVisitor visitor(heap);
HeapObject* object;
Address next_object_must_be_here_or_later = bottom;
for (Address current = bottom; current < top; current += kPointerSize) {
object = HeapObject::FromAddress(current);
if (MarkCompactCollector::IsMarked(object)) {
CHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
CHECK(current >= next_object_must_be_here_or_later);
object->Iterate(&visitor);
next_object_must_be_here_or_later = current + object->Size();
// The next word for sure belongs to the current object, jump over it.
current += kPointerSize;
}
}
}
static void VerifyMarking(NewSpace* space) {
Address end = space->top();
NewSpacePageIterator it(space->bottom(), end);
// The bottom position is at the start of its page. Allows us to use
// page->area_start() as start of range on all pages.
CHECK_EQ(space->bottom(),
NewSpacePage::FromAddress(space->bottom())->area_start());
while (it.has_next()) {
NewSpacePage* page = it.next();
Address limit = it.has_next() ? page->area_end() : end;
CHECK(limit == end || !page->Contains(end));
VerifyMarking(space->heap(), page->area_start(), limit);
}
}
static void VerifyMarking(PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
VerifyMarking(space->heap(), p->area_start(), p->area_end());
}
}
static void VerifyMarking(Heap* heap) {
VerifyMarking(heap->old_space());
VerifyMarking(heap->code_space());
VerifyMarking(heap->map_space());
VerifyMarking(heap->new_space());
VerifyMarkingVisitor visitor(heap);
LargeObjectIterator it(heap->lo_space());
for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) {
if (MarkCompactCollector::IsMarked(obj)) {
obj->Iterate(&visitor);
}
}
heap->IterateStrongRoots(&visitor, VISIT_ONLY_STRONG);
}
class VerifyEvacuationVisitor : public ObjectVisitor {
public:
void VisitPointers(Object** start, Object** end) override {
for (Object** current = start; current < end; current++) {
if ((*current)->IsHeapObject()) {
HeapObject* object = HeapObject::cast(*current);
CHECK(!MarkCompactCollector::IsOnEvacuationCandidate(object));
}
}
}
};
static void VerifyEvacuation(Page* page) {
VerifyEvacuationVisitor visitor;
HeapObjectIterator iterator(page);
for (HeapObject* heap_object = iterator.Next(); heap_object != NULL;
heap_object = iterator.Next()) {
// We skip free space objects.
if (!heap_object->IsFiller()) {
heap_object->Iterate(&visitor);
}
}
}
static void VerifyEvacuation(NewSpace* space) {
NewSpacePageIterator it(space->bottom(), space->top());
VerifyEvacuationVisitor visitor;
while (it.has_next()) {
NewSpacePage* page = it.next();
Address current = page->area_start();
Address limit = it.has_next() ? page->area_end() : space->top();
CHECK(limit == space->top() || !page->Contains(space->top()));
while (current < limit) {
HeapObject* object = HeapObject::FromAddress(current);
object->Iterate(&visitor);
current += object->Size();
}
}
}
static void VerifyEvacuation(Heap* heap, PagedSpace* space) {
if (FLAG_use_allocation_folding && (space == heap->old_space())) {
return;
}
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
if (p->IsEvacuationCandidate()) continue;
VerifyEvacuation(p);
}
}
static void VerifyEvacuation(Heap* heap) {
VerifyEvacuation(heap, heap->old_space());
VerifyEvacuation(heap, heap->code_space());
VerifyEvacuation(heap, heap->map_space());
VerifyEvacuation(heap->new_space());
VerifyEvacuationVisitor visitor;
heap->IterateStrongRoots(&visitor, VISIT_ALL);
}
#endif // VERIFY_HEAP
void MarkCompactCollector::SetUp() {
DCHECK(strcmp(Marking::kWhiteBitPattern, "00") == 0);
DCHECK(strcmp(Marking::kBlackBitPattern, "11") == 0);
DCHECK(strcmp(Marking::kGreyBitPattern, "10") == 0);
DCHECK(strcmp(Marking::kImpossibleBitPattern, "01") == 0);
free_list_old_space_.Reset(new FreeList(heap_->old_space()));
free_list_code_space_.Reset(new FreeList(heap_->code_space()));
free_list_map_space_.Reset(new FreeList(heap_->map_space()));
EnsureMarkingDequeIsReserved();
EnsureMarkingDequeIsCommitted(kMinMarkingDequeSize);
slots_buffer_allocator_ = new SlotsBufferAllocator();
if (FLAG_flush_code) {
code_flusher_ = new CodeFlusher(isolate());
if (FLAG_trace_code_flushing) {
PrintF("[code-flushing is now on]\n");
}
}
}
void MarkCompactCollector::TearDown() {
AbortCompaction();
delete marking_deque_memory_;
delete slots_buffer_allocator_;
delete code_flusher_;
}
void MarkCompactCollector::AddEvacuationCandidate(Page* p) {
DCHECK(!p->NeverEvacuate());
p->MarkEvacuationCandidate();
evacuation_candidates_.Add(p);
}
static void TraceFragmentation(PagedSpace* space) {
int number_of_pages = space->CountTotalPages();
intptr_t reserved = (number_of_pages * space->AreaSize());
intptr_t free = reserved - space->SizeOfObjects();
PrintF("[%s]: %d pages, %d (%.1f%%) free\n",
AllocationSpaceName(space->identity()), number_of_pages,
static_cast<int>(free), static_cast<double>(free) * 100 / reserved);
}
bool MarkCompactCollector::StartCompaction(CompactionMode mode) {
if (!compacting_) {
DCHECK(evacuation_candidates_.length() == 0);
CollectEvacuationCandidates(heap()->old_space());
if (FLAG_compact_code_space) {
CollectEvacuationCandidates(heap()->code_space());
} else if (FLAG_trace_fragmentation) {
TraceFragmentation(heap()->code_space());
}
if (FLAG_trace_fragmentation) {
TraceFragmentation(heap()->map_space());
}
heap()->old_space()->EvictEvacuationCandidatesFromLinearAllocationArea();
heap()->code_space()->EvictEvacuationCandidatesFromLinearAllocationArea();
compacting_ = evacuation_candidates_.length() > 0;
}
return compacting_;
}
void MarkCompactCollector::ClearInvalidStoreAndSlotsBufferEntries() {
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_STORE_BUFFER);
heap_->store_buffer()->ClearInvalidStoreBufferEntries();
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_SLOTS_BUFFER);
int number_of_pages = evacuation_candidates_.length();
for (int i = 0; i < number_of_pages; i++) {
Page* p = evacuation_candidates_[i];
SlotsBuffer::RemoveInvalidSlots(heap_, p->slots_buffer());
}
}
#ifdef VERIFY_HEAP
if (FLAG_verify_heap) {
VerifyValidStoreAndSlotsBufferEntries();
}
#endif
}
#ifdef VERIFY_HEAP
static void VerifyValidSlotsBufferEntries(Heap* heap, PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
SlotsBuffer::VerifySlots(heap, p->slots_buffer());
}
}
void MarkCompactCollector::VerifyValidStoreAndSlotsBufferEntries() {
heap()->store_buffer()->VerifyValidStoreBufferEntries();
VerifyValidSlotsBufferEntries(heap(), heap()->old_space());
VerifyValidSlotsBufferEntries(heap(), heap()->code_space());
VerifyValidSlotsBufferEntries(heap(), heap()->map_space());
LargeObjectIterator it(heap()->lo_space());
for (HeapObject* object = it.Next(); object != NULL; object = it.Next()) {
MemoryChunk* chunk = MemoryChunk::FromAddress(object->address());
SlotsBuffer::VerifySlots(heap(), chunk->slots_buffer());
}
}
#endif
void MarkCompactCollector::CollectGarbage() {
// Make sure that Prepare() has been called. The individual steps below will
// update the state as they proceed.
DCHECK(state_ == PREPARE_GC);
MarkLiveObjects();
DCHECK(heap_->incremental_marking()->IsStopped());
ClearNonLiveReferences();
#ifdef VERIFY_HEAP
if (FLAG_verify_heap) {
VerifyMarking(heap_);
}
#endif
SweepSpaces();
EvacuateNewSpaceAndCandidates();
Finish();
}
#ifdef VERIFY_HEAP
void MarkCompactCollector::VerifyMarkbitsAreClean(PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
CHECK(p->markbits()->IsClean());
CHECK_EQ(0, p->LiveBytes());
}
}
void MarkCompactCollector::VerifyMarkbitsAreClean(NewSpace* space) {
NewSpacePageIterator it(space->bottom(), space->top());
while (it.has_next()) {
NewSpacePage* p = it.next();
CHECK(p->markbits()->IsClean());
CHECK_EQ(0, p->LiveBytes());
}
}
void MarkCompactCollector::VerifyMarkbitsAreClean() {
VerifyMarkbitsAreClean(heap_->old_space());
VerifyMarkbitsAreClean(heap_->code_space());
VerifyMarkbitsAreClean(heap_->map_space());
VerifyMarkbitsAreClean(heap_->new_space());
LargeObjectIterator it(heap_->lo_space());
for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) {
MarkBit mark_bit = Marking::MarkBitFrom(obj);
CHECK(Marking::IsWhite(mark_bit));
CHECK_EQ(0, Page::FromAddress(obj->address())->LiveBytes());
}
}
void MarkCompactCollector::VerifyWeakEmbeddedObjectsInCode() {
HeapObjectIterator code_iterator(heap()->code_space());
for (HeapObject* obj = code_iterator.Next(); obj != NULL;
obj = code_iterator.Next()) {
Code* code = Code::cast(obj);
if (!code->is_optimized_code()) continue;
if (WillBeDeoptimized(code)) continue;
code->VerifyEmbeddedObjectsDependency();
}
}
void MarkCompactCollector::VerifyOmittedMapChecks() {
HeapObjectIterator iterator(heap()->map_space());
for (HeapObject* obj = iterator.Next(); obj != NULL; obj = iterator.Next()) {
Map* map = Map::cast(obj);
map->VerifyOmittedMapChecks();
}
}
#endif // VERIFY_HEAP
static void ClearMarkbitsInPagedSpace(PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Bitmap::Clear(it.next());
}
}
static void ClearMarkbitsInNewSpace(NewSpace* space) {
NewSpacePageIterator it(space->ToSpaceStart(), space->ToSpaceEnd());
while (it.has_next()) {
Bitmap::Clear(it.next());
}
}
void MarkCompactCollector::ClearMarkbits() {
ClearMarkbitsInPagedSpace(heap_->code_space());
ClearMarkbitsInPagedSpace(heap_->map_space());
ClearMarkbitsInPagedSpace(heap_->old_space());
ClearMarkbitsInNewSpace(heap_->new_space());
LargeObjectIterator it(heap_->lo_space());
for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) {
Marking::MarkWhite(Marking::MarkBitFrom(obj));
Page::FromAddress(obj->address())->ResetProgressBar();
Page::FromAddress(obj->address())->ResetLiveBytes();
}
}
class MarkCompactCollector::CompactionTask : public CancelableTask {
public:
explicit CompactionTask(Heap* heap, CompactionSpaceCollection* spaces)
: CancelableTask(heap->isolate()), spaces_(spaces) {}
virtual ~CompactionTask() {}
private:
// v8::internal::CancelableTask overrides.
void RunInternal() override {
MarkCompactCollector* mark_compact =
isolate()->heap()->mark_compact_collector();
SlotsBuffer* evacuation_slots_buffer = nullptr;
mark_compact->EvacuatePages(spaces_, &evacuation_slots_buffer);
mark_compact->AddEvacuationSlotsBufferSynchronized(evacuation_slots_buffer);
mark_compact->pending_compaction_tasks_semaphore_.Signal();
}
CompactionSpaceCollection* spaces_;
DISALLOW_COPY_AND_ASSIGN(CompactionTask);
};
class MarkCompactCollector::SweeperTask : public v8::Task {
public:
SweeperTask(Heap* heap, PagedSpace* space) : heap_(heap), space_(space) {}
virtual ~SweeperTask() {}
private:
// v8::Task overrides.
void Run() override {
heap_->mark_compact_collector()->SweepInParallel(space_, 0);
heap_->mark_compact_collector()->pending_sweeper_tasks_semaphore_.Signal();
}
Heap* heap_;
PagedSpace* space_;
DISALLOW_COPY_AND_ASSIGN(SweeperTask);
};
void MarkCompactCollector::StartSweeperThreads() {
DCHECK(free_list_old_space_.get()->IsEmpty());
DCHECK(free_list_code_space_.get()->IsEmpty());
DCHECK(free_list_map_space_.get()->IsEmpty());
V8::GetCurrentPlatform()->CallOnBackgroundThread(
new SweeperTask(heap(), heap()->old_space()),
v8::Platform::kShortRunningTask);
V8::GetCurrentPlatform()->CallOnBackgroundThread(
new SweeperTask(heap(), heap()->code_space()),
v8::Platform::kShortRunningTask);
V8::GetCurrentPlatform()->CallOnBackgroundThread(
new SweeperTask(heap(), heap()->map_space()),
v8::Platform::kShortRunningTask);
}
void MarkCompactCollector::SweepOrWaitUntilSweepingCompleted(Page* page) {
PagedSpace* owner = reinterpret_cast<PagedSpace*>(page->owner());
if (!page->SweepingCompleted()) {
SweepInParallel(page, owner);
if (!page->SweepingCompleted()) {
// We were not able to sweep that page, i.e., a concurrent
// sweeper thread currently owns this page. Wait for the sweeper
// thread to be done with this page.
page->WaitUntilSweepingCompleted();
}
}
}
void MarkCompactCollector::SweepAndRefill(CompactionSpace* space) {
if (FLAG_concurrent_sweeping && !IsSweepingCompleted()) {
SweepInParallel(heap()->paged_space(space->identity()), 0);
space->RefillFreeList();
}
}
void MarkCompactCollector::EnsureSweepingCompleted() {
DCHECK(sweeping_in_progress_ == true);
// If sweeping is not completed or not running at all, we try to complete it
// here.
if (!FLAG_concurrent_sweeping || !IsSweepingCompleted()) {
SweepInParallel(heap()->paged_space(OLD_SPACE), 0);
SweepInParallel(heap()->paged_space(CODE_SPACE), 0);
SweepInParallel(heap()->paged_space(MAP_SPACE), 0);
}
if (FLAG_concurrent_sweeping) {
pending_sweeper_tasks_semaphore_.Wait();
pending_sweeper_tasks_semaphore_.Wait();
pending_sweeper_tasks_semaphore_.Wait();
}
ParallelSweepSpacesComplete();
sweeping_in_progress_ = false;
heap()->old_space()->RefillFreeList();
heap()->code_space()->RefillFreeList();
heap()->map_space()->RefillFreeList();
#ifdef VERIFY_HEAP
if (FLAG_verify_heap && !evacuation()) {
VerifyEvacuation(heap_);
}
#endif
}
bool MarkCompactCollector::IsSweepingCompleted() {
if (!pending_sweeper_tasks_semaphore_.WaitFor(
base::TimeDelta::FromSeconds(0))) {
return false;
}
pending_sweeper_tasks_semaphore_.Signal();
return true;
}
void Marking::TransferMark(Heap* heap, Address old_start, Address new_start) {
// This is only used when resizing an object.
DCHECK(MemoryChunk::FromAddress(old_start) ==
MemoryChunk::FromAddress(new_start));
if (!heap->incremental_marking()->IsMarking()) return;
// If the mark doesn't move, we don't check the color of the object.
// It doesn't matter whether the object is black, since it hasn't changed
// size, so the adjustment to the live data count will be zero anyway.
if (old_start == new_start) return;
MarkBit new_mark_bit = MarkBitFrom(new_start);
MarkBit old_mark_bit = MarkBitFrom(old_start);
#ifdef DEBUG
ObjectColor old_color = Color(old_mark_bit);
#endif
if (Marking::IsBlack(old_mark_bit)) {
Marking::BlackToWhite(old_mark_bit);
Marking::MarkBlack(new_mark_bit);
return;
} else if (Marking::IsGrey(old_mark_bit)) {
Marking::GreyToWhite(old_mark_bit);
heap->incremental_marking()->WhiteToGreyAndPush(
HeapObject::FromAddress(new_start), new_mark_bit);
heap->incremental_marking()->RestartIfNotMarking();
}
#ifdef DEBUG
ObjectColor new_color = Color(new_mark_bit);
DCHECK(new_color == old_color);
#endif
}
const char* AllocationSpaceName(AllocationSpace space) {
switch (space) {
case NEW_SPACE:
return "NEW_SPACE";
case OLD_SPACE:
return "OLD_SPACE";
case CODE_SPACE:
return "CODE_SPACE";
case MAP_SPACE:
return "MAP_SPACE";
case LO_SPACE:
return "LO_SPACE";
default:
UNREACHABLE();
}
return NULL;
}
void MarkCompactCollector::ComputeEvacuationHeuristics(
int area_size, int* target_fragmentation_percent,
int* max_evacuated_bytes) {
// For memory reducing mode we directly define both constants.
const int kTargetFragmentationPercentForReduceMemory = 20;
const int kMaxEvacuatedBytesForReduceMemory = 12 * Page::kPageSize;
// For regular mode (which is latency critical) we define less aggressive
// defaults to start and switch to a trace-based (using compaction speed)
// approach as soon as we have enough samples.
const int kTargetFragmentationPercent = 70;
const int kMaxEvacuatedBytes = 4 * Page::kPageSize;
// Time to take for a single area (=payload of page). Used as soon as there
// exist enough compaction speed samples.
const int kTargetMsPerArea = 1;
if (heap()->ShouldReduceMemory()) {
*target_fragmentation_percent = kTargetFragmentationPercentForReduceMemory;
*max_evacuated_bytes = kMaxEvacuatedBytesForReduceMemory;
} else {
const intptr_t estimated_compaction_speed =
heap()->tracer()->CompactionSpeedInBytesPerMillisecond();
if (estimated_compaction_speed != 0) {
// Estimate the target fragmentation based on traced compaction speed
// and a goal for a single page.
const intptr_t estimated_ms_per_area =
1 + static_cast<intptr_t>(area_size) / estimated_compaction_speed;
*target_fragmentation_percent =
100 - 100 * kTargetMsPerArea / estimated_ms_per_area;
if (*target_fragmentation_percent <
kTargetFragmentationPercentForReduceMemory) {
*target_fragmentation_percent =
kTargetFragmentationPercentForReduceMemory;
}
} else {
*target_fragmentation_percent = kTargetFragmentationPercent;
}
*max_evacuated_bytes = kMaxEvacuatedBytes;
}
}
void MarkCompactCollector::CollectEvacuationCandidates(PagedSpace* space) {
DCHECK(space->identity() == OLD_SPACE || space->identity() == CODE_SPACE);
int number_of_pages = space->CountTotalPages();
int area_size = space->AreaSize();
// Pairs of (live_bytes_in_page, page).
typedef std::pair<int, Page*> LiveBytesPagePair;
std::vector<LiveBytesPagePair> pages;
pages.reserve(number_of_pages);
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
if (p->NeverEvacuate()) continue;
if (p->IsFlagSet(Page::POPULAR_PAGE)) {
// This page had slots buffer overflow on previous GC, skip it.
p->ClearFlag(Page::POPULAR_PAGE);
continue;
}
// Invariant: Evacuation candidates are just created when marking is
// started. At the end of a GC all evacuation candidates are cleared and
// their slot buffers are released.
CHECK(!p->IsEvacuationCandidate());
CHECK(p->slots_buffer() == NULL);
DCHECK(p->area_size() == area_size);
int live_bytes =
p->WasSwept() ? p->LiveBytesFromFreeList() : p->LiveBytes();
pages.push_back(std::make_pair(live_bytes, p));
}
int candidate_count = 0;
int total_live_bytes = 0;
const bool reduce_memory = heap()->ShouldReduceMemory();
if (FLAG_manual_evacuation_candidates_selection) {
for (size_t i = 0; i < pages.size(); i++) {
Page* p = pages[i].second;
if (p->IsFlagSet(MemoryChunk::FORCE_EVACUATION_CANDIDATE_FOR_TESTING)) {
candidate_count++;
total_live_bytes += pages[i].first;
p->ClearFlag(MemoryChunk::FORCE_EVACUATION_CANDIDATE_FOR_TESTING);
AddEvacuationCandidate(p);
}
}
} else if (FLAG_stress_compaction) {
for (size_t i = 0; i < pages.size(); i++) {
Page* p = pages[i].second;
if (i % 2 == 0) {
candidate_count++;
total_live_bytes += pages[i].first;
AddEvacuationCandidate(p);
}
}
} else {
// The following approach determines the pages that should be evacuated.
//
// We use two conditions to decide whether a page qualifies as an evacuation
// candidate, or not:
// * Target fragmentation: How fragmented is a page, i.e., how is the ratio
// between live bytes and capacity of this page (= area).
// * Evacuation quota: A global quota determining how much bytes should be
// compacted.
//
// The algorithm sorts all pages by live bytes and then iterates through
// them starting with the page with the most free memory, adding them to the
// set of evacuation candidates as long as both conditions (fragmentation
// and quota) hold.
int max_evacuated_bytes;
int target_fragmentation_percent;
ComputeEvacuationHeuristics(area_size, &target_fragmentation_percent,
&max_evacuated_bytes);
const intptr_t free_bytes_threshold =
target_fragmentation_percent * (area_size / 100);
// Sort pages from the most free to the least free, then select
// the first n pages for evacuation such that:
// - the total size of evacuated objects does not exceed the specified
// limit.
// - fragmentation of (n+1)-th page does not exceed the specified limit.
std::sort(pages.begin(), pages.end(),
[](const LiveBytesPagePair& a, const LiveBytesPagePair& b) {
return a.first < b.first;
});
for (size_t i = 0; i < pages.size(); i++) {
int live_bytes = pages[i].first;
int free_bytes = area_size - live_bytes;
if (FLAG_always_compact ||
((free_bytes >= free_bytes_threshold) &&
((total_live_bytes + live_bytes) <= max_evacuated_bytes))) {
candidate_count++;
total_live_bytes += live_bytes;
}
if (FLAG_trace_fragmentation_verbose) {
PrintIsolate(isolate(),
"compaction-selection-page: space=%s free_bytes_page=%d "
"fragmentation_limit_kb=%d fragmentation_limit_percent=%d "
"sum_compaction_kb=%d "
"compaction_limit_kb=%d\n",
AllocationSpaceName(space->identity()), free_bytes / KB,
free_bytes_threshold / KB, target_fragmentation_percent,
total_live_bytes / KB, max_evacuated_bytes / KB);
}
}
// How many pages we will allocated for the evacuated objects
// in the worst case: ceil(total_live_bytes / area_size)
int estimated_new_pages = (total_live_bytes + area_size - 1) / area_size;
DCHECK_LE(estimated_new_pages, candidate_count);
int estimated_released_pages = candidate_count - estimated_new_pages;
// Avoid (compact -> expand) cycles.
if ((estimated_released_pages == 0) && !FLAG_always_compact) {
candidate_count = 0;
}
for (int i = 0; i < candidate_count; i++) {
AddEvacuationCandidate(pages[i].second);
}
}
if (FLAG_trace_fragmentation) {
PrintIsolate(isolate(),
"compaction-selection: space=%s reduce_memory=%d pages=%d "
"total_live_bytes=%d\n",
AllocationSpaceName(space->identity()), reduce_memory,
candidate_count, total_live_bytes / KB);
}
}
void MarkCompactCollector::AbortCompaction() {
if (compacting_) {
int npages = evacuation_candidates_.length();
for (int i = 0; i < npages; i++) {
Page* p = evacuation_candidates_[i];
slots_buffer_allocator_->DeallocateChain(p->slots_buffer_address());
p->ClearEvacuationCandidate();
p->ClearFlag(MemoryChunk::RESCAN_ON_EVACUATION);
}
compacting_ = false;
evacuation_candidates_.Rewind(0);
}
DCHECK_EQ(0, evacuation_candidates_.length());
}
void MarkCompactCollector::Prepare() {
was_marked_incrementally_ = heap()->incremental_marking()->IsMarking();
#ifdef DEBUG
DCHECK(state_ == IDLE);
state_ = PREPARE_GC;
#endif
DCHECK(!FLAG_never_compact || !FLAG_always_compact);
if (sweeping_in_progress()) {
// Instead of waiting we could also abort the sweeper threads here.
EnsureSweepingCompleted();
}
// If concurrent unmapping tasks are still running, we should wait for
// them here.
heap()->WaitUntilUnmappingOfFreeChunksCompleted();
// Clear marking bits if incremental marking is aborted.
if (was_marked_incrementally_ && heap_->ShouldAbortIncrementalMarking()) {
heap()->incremental_marking()->Stop();
ClearMarkbits();
AbortWeakCollections();
AbortWeakCells();
AbortTransitionArrays();
AbortCompaction();
was_marked_incrementally_ = false;
}
// Don't start compaction if we are in the middle of incremental
// marking cycle. We did not collect any slots.
if (!FLAG_never_compact && !was_marked_incrementally_) {
StartCompaction(NON_INCREMENTAL_COMPACTION);
}
PagedSpaces spaces(heap());
for (PagedSpace* space = spaces.next(); space != NULL;
space = spaces.next()) {
space->PrepareForMarkCompact();
}
#ifdef VERIFY_HEAP
if (!was_marked_incrementally_ && FLAG_verify_heap) {
VerifyMarkbitsAreClean();
}
#endif
}
void MarkCompactCollector::Finish() {
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_FINISH);
// The hashing of weak_object_to_code_table is no longer valid.
heap()->weak_object_to_code_table()->Rehash(
heap()->isolate()->factory()->undefined_value());
// Clear the marking state of live large objects.
heap_->lo_space()->ClearMarkingStateOfLiveObjects();
#ifdef DEBUG
DCHECK(state_ == SWEEP_SPACES || state_ == RELOCATE_OBJECTS);
state_ = IDLE;
#endif
heap_->isolate()->inner_pointer_to_code_cache()->Flush();
// The stub cache is not traversed during GC; clear the cache to
// force lazy re-initialization of it. This must be done after the
// GC, because it relies on the new address of certain old space
// objects (empty string, illegal builtin).
isolate()->stub_cache()->Clear();
if (have_code_to_deoptimize_) {
// Some code objects were marked for deoptimization during the GC.
Deoptimizer::DeoptimizeMarkedCode(isolate());
have_code_to_deoptimize_ = false;
}
heap_->incremental_marking()->ClearIdleMarkingDelayCounter();
if (marking_parity_ == EVEN_MARKING_PARITY) {
marking_parity_ = ODD_MARKING_PARITY;
} else {
DCHECK(marking_parity_ == ODD_MARKING_PARITY);
marking_parity_ = EVEN_MARKING_PARITY;
}
}
// -------------------------------------------------------------------------
// Phase 1: tracing and marking live objects.
// before: all objects are in normal state.
// after: a live object's map pointer is marked as '00'.
// Marking all live objects in the heap as part of mark-sweep or mark-compact
// collection. Before marking, all objects are in their normal state. After
// marking, live objects' map pointers are marked indicating that the object
// has been found reachable.
//
// The marking algorithm is a (mostly) depth-first (because of possible stack
// overflow) traversal of the graph of objects reachable from the roots. It
// uses an explicit stack of pointers rather than recursion. The young
// generation's inactive ('from') space is used as a marking stack. The
// objects in the marking stack are the ones that have been reached and marked
// but their children have not yet been visited.
//
// The marking stack can overflow during traversal. In that case, we set an
// overflow flag. When the overflow flag is set, we continue marking objects
// reachable from the objects on the marking stack, but no longer push them on
// the marking stack. Instead, we mark them as both marked and overflowed.
// When the stack is in the overflowed state, objects marked as overflowed
// have been reached and marked but their children have not been visited yet.
// After emptying the marking stack, we clear the overflow flag and traverse
// the heap looking for objects marked as overflowed, push them on the stack,
// and continue with marking. This process repeats until all reachable
// objects have been marked.
void CodeFlusher::ProcessJSFunctionCandidates() {
Code* lazy_compile = isolate_->builtins()->builtin(Builtins::kCompileLazy);
Object* undefined = isolate_->heap()->undefined_value();
JSFunction* candidate = jsfunction_candidates_head_;
JSFunction* next_candidate;
while (candidate != NULL) {
next_candidate = GetNextCandidate(candidate);
ClearNextCandidate(candidate, undefined);
SharedFunctionInfo* shared = candidate->shared();
Code* code = shared->code();
MarkBit code_mark = Marking::MarkBitFrom(code);
if (Marking::IsWhite(code_mark)) {
if (FLAG_trace_code_flushing && shared->is_compiled()) {
PrintF("[code-flushing clears: ");
shared->ShortPrint();
PrintF(" - age: %d]\n", code->GetAge());
}
// Always flush the optimized code map if there is one.
if (!shared->OptimizedCodeMapIsCleared()) {
shared->ClearOptimizedCodeMap();
}
shared->set_code(lazy_compile);
candidate->set_code(lazy_compile);
} else {
DCHECK(Marking::IsBlack(code_mark));
candidate->set_code(code);
}
// We are in the middle of a GC cycle so the write barrier in the code
// setter did not record the slot update and we have to do that manually.
Address slot = candidate->address() + JSFunction::kCodeEntryOffset;
Code* target = Code::cast(Code::GetObjectFromEntryAddress(slot));
isolate_->heap()->mark_compact_collector()->RecordCodeEntrySlot(
candidate, slot, target);
Object** shared_code_slot =
HeapObject::RawField(shared, SharedFunctionInfo::kCodeOffset);
isolate_->heap()->mark_compact_collector()->RecordSlot(
shared, shared_code_slot, *shared_code_slot);
candidate = next_candidate;
}
jsfunction_candidates_head_ = NULL;
}
void CodeFlusher::ProcessSharedFunctionInfoCandidates() {
Code* lazy_compile = isolate_->builtins()->builtin(Builtins::kCompileLazy);
SharedFunctionInfo* candidate = shared_function_info_candidates_head_;
SharedFunctionInfo* next_candidate;
while (candidate != NULL) {
next_candidate = GetNextCandidate(candidate);
ClearNextCandidate(candidate);
Code* code = candidate->code();
MarkBit code_mark = Marking::MarkBitFrom(code);
if (Marking::IsWhite(code_mark)) {
if (FLAG_trace_code_flushing && candidate->is_compiled()) {
PrintF("[code-flushing clears: ");
candidate->ShortPrint();
PrintF(" - age: %d]\n", code->GetAge());
}
// Always flush the optimized code map if there is one.
if (!candidate->OptimizedCodeMapIsCleared()) {
candidate->ClearOptimizedCodeMap();
}
candidate->set_code(lazy_compile);
}
Object** code_slot =
HeapObject::RawField(candidate, SharedFunctionInfo::kCodeOffset);
isolate_->heap()->mark_compact_collector()->RecordSlot(candidate, code_slot,
*code_slot);
candidate = next_candidate;
}
shared_function_info_candidates_head_ = NULL;
}
void CodeFlusher::EvictCandidate(SharedFunctionInfo* shared_info) {
// Make sure previous flushing decisions are revisited.
isolate_->heap()->incremental_marking()->RecordWrites(shared_info);
if (FLAG_trace_code_flushing) {
PrintF("[code-flushing abandons function-info: ");
shared_info->ShortPrint();
PrintF("]\n");
}
SharedFunctionInfo* candidate = shared_function_info_candidates_head_;
SharedFunctionInfo* next_candidate;
if (candidate == shared_info) {
next_candidate = GetNextCandidate(shared_info);
shared_function_info_candidates_head_ = next_candidate;
ClearNextCandidate(shared_info);
} else {
while (candidate != NULL) {
next_candidate = GetNextCandidate(candidate);
if (next_candidate == shared_info) {
next_candidate = GetNextCandidate(shared_info);
SetNextCandidate(candidate, next_candidate);
ClearNextCandidate(shared_info);
break;
}
candidate = next_candidate;
}
}
}
void CodeFlusher::EvictCandidate(JSFunction* function) {
DCHECK(!function->next_function_link()->IsUndefined());
Object* undefined = isolate_->heap()->undefined_value();
// Make sure previous flushing decisions are revisited.
isolate_->heap()->incremental_marking()->RecordWrites(function);
isolate_->heap()->incremental_marking()->RecordWrites(function->shared());
if (FLAG_trace_code_flushing) {
PrintF("[code-flushing abandons closure: ");
function->shared()->ShortPrint();
PrintF("]\n");
}
JSFunction* candidate = jsfunction_candidates_head_;
JSFunction* next_candidate;
if (candidate == function) {
next_candidate = GetNextCandidate(function);
jsfunction_candidates_head_ = next_candidate;
ClearNextCandidate(function, undefined);
} else {
while (candidate != NULL) {
next_candidate = GetNextCandidate(candidate);
if (next_candidate == function) {
next_candidate = GetNextCandidate(function);
SetNextCandidate(candidate, next_candidate);
ClearNextCandidate(function, undefined);
break;
}
candidate = next_candidate;
}
}
}
void CodeFlusher::IteratePointersToFromSpace(ObjectVisitor* v) {
Heap* heap = isolate_->heap();
JSFunction** slot = &jsfunction_candidates_head_;
JSFunction* candidate = jsfunction_candidates_head_;
while (candidate != NULL) {
if (heap->InFromSpace(candidate)) {
v->VisitPointer(reinterpret_cast<Object**>(slot));
}
candidate = GetNextCandidate(*slot);
slot = GetNextCandidateSlot(*slot);
}
}
class MarkCompactMarkingVisitor
: public StaticMarkingVisitor<MarkCompactMarkingVisitor> {
public:
static void Initialize();
INLINE(static void VisitPointer(Heap* heap, HeapObject* object, Object** p)) {
MarkObjectByPointer(heap->mark_compact_collector(), object, p);
}
INLINE(static void VisitPointers(Heap* heap, HeapObject* object,
Object** start, Object** end)) {
// Mark all objects pointed to in [start, end).
const int kMinRangeForMarkingRecursion = 64;
if (end - start >= kMinRangeForMarkingRecursion) {
if (VisitUnmarkedObjects(heap, object, start, end)) return;
// We are close to a stack overflow, so just mark the objects.
}
MarkCompactCollector* collector = heap->mark_compact_collector();
for (Object** p = start; p < end; p++) {
MarkObjectByPointer(collector, object, p);
}
}
// Marks the object black and pushes it on the marking stack.
INLINE(static void MarkObject(Heap* heap, HeapObject* object)) {
MarkBit mark = Marking::MarkBitFrom(object);
heap->mark_compact_collector()->MarkObject(object, mark);
}
// Marks the object black without pushing it on the marking stack.
// Returns true if object needed marking and false otherwise.
INLINE(static bool MarkObjectWithoutPush(Heap* heap, HeapObject* object)) {
MarkBit mark_bit = Marking::MarkBitFrom(object);
if (Marking::IsWhite(mark_bit)) {
heap->mark_compact_collector()->SetMark(object, mark_bit);
return true;
}
return false;
}
// Mark object pointed to by p.
INLINE(static void MarkObjectByPointer(MarkCompactCollector* collector,
HeapObject* object, Object** p)) {
if (!(*p)->IsHeapObject()) return;
HeapObject* target_object = HeapObject::cast(*p);
collector->RecordSlot(object, p, target_object);
MarkBit mark = Marking::MarkBitFrom(target_object);
collector->MarkObject(target_object, mark);
}
// Visit an unmarked object.
INLINE(static void VisitUnmarkedObject(MarkCompactCollector* collector,
HeapObject* obj)) {
#ifdef DEBUG
DCHECK(collector->heap()->Contains(obj));
DCHECK(!collector->heap()->mark_compact_collector()->IsMarked(obj));
#endif
Map* map = obj->map();
Heap* heap = obj->GetHeap();
MarkBit mark = Marking::MarkBitFrom(obj);
heap->mark_compact_collector()->SetMark(obj, mark);
// Mark the map pointer and the body.
MarkBit map_mark = Marking::MarkBitFrom(map);
heap->mark_compact_collector()->MarkObject(map, map_mark);
IterateBody(map, obj);
}
// Visit all unmarked objects pointed to by [start, end).
// Returns false if the operation fails (lack of stack space).
INLINE(static bool VisitUnmarkedObjects(Heap* heap, HeapObject* object,
Object** start, Object** end)) {
// Return false is we are close to the stack limit.
StackLimitCheck check(heap->isolate());
if (check.HasOverflowed()) return false;
MarkCompactCollector* collector = heap->mark_compact_collector();
// Visit the unmarked objects.
for (Object** p = start; p < end; p++) {
Object* o = *p;
if (!o->IsHeapObject()) continue;
collector->RecordSlot(object, p, o);
HeapObject* obj = HeapObject::cast(o);
MarkBit mark = Marking::MarkBitFrom(obj);
if (Marking::IsBlackOrGrey(mark)) continue;
VisitUnmarkedObject(collector, obj);
}
return true;
}
private:
template <int id>
static inline void TrackObjectStatsAndVisit(Map* map, HeapObject* obj);
// Code flushing support.
static const int kRegExpCodeThreshold = 5;
static void UpdateRegExpCodeAgeAndFlush(Heap* heap, JSRegExp* re,
bool is_one_byte) {
// Make sure that the fixed array is in fact initialized on the RegExp.
// We could potentially trigger a GC when initializing the RegExp.
if (HeapObject::cast(re->data())->map()->instance_type() !=
FIXED_ARRAY_TYPE)
return;
// Make sure this is a RegExp that actually contains code.
if (re->TypeTag() != JSRegExp::IRREGEXP) return;
Object* code = re->DataAt(JSRegExp::code_index(is_one_byte));
if (!code->IsSmi() &&
HeapObject::cast(code)->map()->instance_type() == CODE_TYPE) {
// Save a copy that can be reinstated if we need the code again.
re->SetDataAt(JSRegExp::saved_code_index(is_one_byte), code);
// Saving a copy might create a pointer into compaction candidate
// that was not observed by marker. This might happen if JSRegExp data
// was marked through the compilation cache before marker reached JSRegExp
// object.
FixedArray* data = FixedArray::cast(re->data());
Object** slot =
data->data_start() + JSRegExp::saved_code_index(is_one_byte);
heap->mark_compact_collector()->RecordSlot(data, slot, code);
// Set a number in the 0-255 range to guarantee no smi overflow.
re->SetDataAt(JSRegExp::code_index(is_one_byte),
Smi::FromInt(heap->ms_count() & 0xff));
} else if (code->IsSmi()) {
int value = Smi::cast(code)->value();
// The regexp has not been compiled yet or there was a compilation error.
if (value == JSRegExp::kUninitializedValue ||
value == JSRegExp::kCompilationErrorValue) {
return;
}
// Check if we should flush now.
if (value == ((heap->ms_count() - kRegExpCodeThreshold) & 0xff)) {
re->SetDataAt(JSRegExp::code_index(is_one_byte),
Smi::FromInt(JSRegExp::kUninitializedValue));
re->SetDataAt(JSRegExp::saved_code_index(is_one_byte),
Smi::FromInt(JSRegExp::kUninitializedValue));
}
}
}
// Works by setting the current sweep_generation (as a smi) in the
// code object place in the data array of the RegExp and keeps a copy
// around that can be reinstated if we reuse the RegExp before flushing.
// If we did not use the code for kRegExpCodeThreshold mark sweep GCs
// we flush the code.
static void VisitRegExpAndFlushCode(Map* map, HeapObject* object) {
Heap* heap = map->GetHeap();
MarkCompactCollector* collector = heap->mark_compact_collector();
if (!collector->is_code_flushing_enabled()) {
VisitJSRegExp(map, object);
return;
}
JSRegExp* re = reinterpret_cast<JSRegExp*>(object);
// Flush code or set age on both one byte and two byte code.
UpdateRegExpCodeAgeAndFlush(heap, re, true);
UpdateRegExpCodeAgeAndFlush(heap, re, false);
// Visit the fields of the RegExp, including the updated FixedArray.
VisitJSRegExp(map, object);
}
};
void MarkCompactMarkingVisitor::Initialize() {
StaticMarkingVisitor<MarkCompactMarkingVisitor>::Initialize();
table_.Register(kVisitJSRegExp, &VisitRegExpAndFlushCode);
if (FLAG_track_gc_object_stats) {
ObjectStatsVisitor::Initialize(&table_);
}
}
class CodeMarkingVisitor : public ThreadVisitor {
public:
explicit CodeMarkingVisitor(MarkCompactCollector* collector)
: collector_(collector) {}
void VisitThread(Isolate* isolate, ThreadLocalTop* top) {
collector_->PrepareThreadForCodeFlushing(isolate, top);
}
private:
MarkCompactCollector* collector_;
};
class SharedFunctionInfoMarkingVisitor : public ObjectVisitor {
public:
explicit SharedFunctionInfoMarkingVisitor(MarkCompactCollector* collector)
: collector_(collector) {}
void VisitPointers(Object** start, Object** end) override {
for (Object** p = start; p < end; p++) VisitPointer(p);
}
void VisitPointer(Object** slot) override {
Object* obj = *slot;
if (obj->IsSharedFunctionInfo()) {
SharedFunctionInfo* shared = reinterpret_cast<SharedFunctionInfo*>(obj);
MarkBit shared_mark = Marking::MarkBitFrom(shared);
MarkBit code_mark = Marking::MarkBitFrom(shared->code());
collector_->MarkObject(shared->code(), code_mark);
collector_->MarkObject(shared, shared_mark);
}
}
private:
MarkCompactCollector* collector_;
};
void MarkCompactCollector::PrepareThreadForCodeFlushing(Isolate* isolate,
ThreadLocalTop* top) {
for (StackFrameIterator it(isolate, top); !it.done(); it.Advance()) {
// Note: for the frame that has a pending lazy deoptimization
// StackFrame::unchecked_code will return a non-optimized code object for
// the outermost function and StackFrame::LookupCode will return
// actual optimized code object.
StackFrame* frame = it.frame();
Code* code = frame->unchecked_code();
MarkBit code_mark = Marking::MarkBitFrom(code);
MarkObject(code, code_mark);
if (frame->is_optimized()) {
Code* optimized_code = frame->LookupCode();
MarkBit optimized_code_mark = Marking::MarkBitFrom(optimized_code);
MarkObject(optimized_code, optimized_code_mark);
}
}
}
void MarkCompactCollector::PrepareForCodeFlushing() {
// If code flushing is disabled, there is no need to prepare for it.
if (!is_code_flushing_enabled()) return;
// Ensure that empty descriptor array is marked. Method MarkDescriptorArray
// relies on it being marked before any other descriptor array.
HeapObject* descriptor_array = heap()->empty_descriptor_array();
MarkBit descriptor_array_mark = Marking::MarkBitFrom(descriptor_array);
MarkObject(descriptor_array, descriptor_array_mark);
// Make sure we are not referencing the code from the stack.
DCHECK(this == heap()->mark_compact_collector());
PrepareThreadForCodeFlushing(heap()->isolate(),
heap()->isolate()->thread_local_top());
// Iterate the archived stacks in all threads to check if
// the code is referenced.
CodeMarkingVisitor code_marking_visitor(this);
heap()->isolate()->thread_manager()->IterateArchivedThreads(
&code_marking_visitor);
SharedFunctionInfoMarkingVisitor visitor(this);
heap()->isolate()->compilation_cache()->IterateFunctions(&visitor);
heap()->isolate()->handle_scope_implementer()->Iterate(&visitor);
ProcessMarkingDeque();
}
// Visitor class for marking heap roots.
class RootMarkingVisitor : public ObjectVisitor {
public:
explicit RootMarkingVisitor(Heap* heap)
: collector_(heap->mark_compact_collector()) {}
void VisitPointer(Object** p) override { MarkObjectByPointer(p); }
void VisitPointers(Object** start, Object** end) override {
for (Object** p = start; p < end; p++) MarkObjectByPointer(p);
}
// Skip the weak next code link in a code object, which is visited in
// ProcessTopOptimizedFrame.
void VisitNextCodeLink(Object** p) override {}
private:
void MarkObjectByPointer(Object** p) {
if (!(*p)->IsHeapObject()) return;
// Replace flat cons strings in place.
HeapObject* object = HeapObject::cast(*p);
MarkBit mark_bit = Marking::MarkBitFrom(object);
if (Marking::IsBlackOrGrey(mark_bit)) return;
Map* map = object->map();
// Mark the object.
collector_->SetMark(object, mark_bit);
// Mark the map pointer and body, and push them on the marking stack.
MarkBit map_mark = Marking::MarkBitFrom(map);
collector_->MarkObject(map, map_mark);
MarkCompactMarkingVisitor::IterateBody(map, object);
// Mark all the objects reachable from the map and body. May leave
// overflowed objects in the heap.
collector_->EmptyMarkingDeque();
}
MarkCompactCollector* collector_;
};
// Helper class for pruning the string table.
template <bool finalize_external_strings>
class StringTableCleaner : public ObjectVisitor {
public:
explicit StringTableCleaner(Heap* heap) : heap_(heap), pointers_removed_(0) {}
void VisitPointers(Object** start, Object** end) override {
// Visit all HeapObject pointers in [start, end).
for (Object** p = start; p < end; p++) {
Object* o = *p;
if (o->IsHeapObject() &&
Marking::IsWhite(Marking::MarkBitFrom(HeapObject::cast(o)))) {
if (finalize_external_strings) {
DCHECK(o->IsExternalString());
heap_->FinalizeExternalString(String::cast(*p));
} else {
pointers_removed_++;
}
// Set the entry to the_hole_value (as deleted).
*p = heap_->the_hole_value();
}
}
}
int PointersRemoved() {
DCHECK(!finalize_external_strings);
return pointers_removed_;
}
private:
Heap* heap_;
int pointers_removed_;
};
typedef StringTableCleaner<false> InternalizedStringTableCleaner;
typedef StringTableCleaner<true> ExternalStringTableCleaner;
// Implementation of WeakObjectRetainer for mark compact GCs. All marked objects
// are retained.
class MarkCompactWeakObjectRetainer : public WeakObjectRetainer {
public:
virtual Object* RetainAs(Object* object) {
MarkBit mark_bit = Marking::MarkBitFrom(HeapObject::cast(object));
DCHECK(!Marking::IsGrey(mark_bit));
if (Marking::IsBlack(mark_bit)) {
return object;
} else if (object->IsAllocationSite() &&
!(AllocationSite::cast(object)->IsZombie())) {
// "dead" AllocationSites need to live long enough for a traversal of new
// space. These sites get a one-time reprieve.
AllocationSite* site = AllocationSite::cast(object);
site->MarkZombie();
site->GetHeap()->mark_compact_collector()->MarkAllocationSite(site);
return object;
} else {
return NULL;
}
}
};
// Fill the marking stack with overflowed objects returned by the given
// iterator. Stop when the marking stack is filled or the end of the space
// is reached, whichever comes first.
template <class T>
void MarkCompactCollector::DiscoverGreyObjectsWithIterator(T* it) {
// The caller should ensure that the marking stack is initially not full,
// so that we don't waste effort pointlessly scanning for objects.
DCHECK(!marking_deque()->IsFull());
Map* filler_map = heap()->one_pointer_filler_map();
for (HeapObject* object = it->Next(); object != NULL; object = it->Next()) {
MarkBit markbit = Marking::MarkBitFrom(object);
if ((object->map() != filler_map) && Marking::IsGrey(markbit)) {
Marking::GreyToBlack(markbit);
PushBlack(object);
if (marking_deque()->IsFull()) return;
}
}
}
void MarkCompactCollector::DiscoverGreyObjectsOnPage(MemoryChunk* p) {
DCHECK(!marking_deque()->IsFull());
LiveObjectIterator<kGreyObjects> it(p);
HeapObject* object = NULL;
while ((object = it.Next()) != NULL) {
MarkBit markbit = Marking::MarkBitFrom(object);
DCHECK(Marking::IsGrey(markbit));
Marking::GreyToBlack(markbit);
PushBlack(object);
if (marking_deque()->IsFull()) return;
}
}
class MarkCompactCollector::HeapObjectVisitor {
public:
virtual ~HeapObjectVisitor() {}
virtual bool Visit(HeapObject* object) = 0;
};
class MarkCompactCollector::EvacuateVisitorBase
: public MarkCompactCollector::HeapObjectVisitor {
public:
EvacuateVisitorBase(Heap* heap, SlotsBuffer** evacuation_slots_buffer)
: heap_(heap), evacuation_slots_buffer_(evacuation_slots_buffer) {}
bool TryEvacuateObject(PagedSpace* target_space, HeapObject* object,
HeapObject** target_object) {
int size = object->Size();
AllocationAlignment alignment = object->RequiredAlignment();
AllocationResult allocation = target_space->AllocateRaw(size, alignment);
if (allocation.To(target_object)) {
heap_->mark_compact_collector()->MigrateObject(
*target_object, object, size, target_space->identity(),
evacuation_slots_buffer_);
return true;
}
return false;
}
protected:
Heap* heap_;
SlotsBuffer** evacuation_slots_buffer_;
};
class MarkCompactCollector::EvacuateNewSpaceVisitor final
: public MarkCompactCollector::EvacuateVisitorBase {
public:
static const intptr_t kLabSize = 4 * KB;
static const intptr_t kMaxLabObjectSize = 256;
explicit EvacuateNewSpaceVisitor(Heap* heap,
SlotsBuffer** evacuation_slots_buffer,
HashMap* local_pretenuring_feedback)
: EvacuateVisitorBase(heap, evacuation_slots_buffer),
buffer_(LocalAllocationBuffer::InvalidBuffer()),
space_to_allocate_(NEW_SPACE),
promoted_size_(0),
semispace_copied_size_(0),
local_pretenuring_feedback_(local_pretenuring_feedback) {}
bool Visit(HeapObject* object) override {
heap_->UpdateAllocationSite(object, local_pretenuring_feedback_);
int size = object->Size();
HeapObject* target_object = nullptr;
if (heap_->ShouldBePromoted(object->address(), size) &&
TryEvacuateObject(heap_->old_space(), object, &target_object)) {
// If we end up needing more special cases, we should factor this out.
if (V8_UNLIKELY(target_object->IsJSArrayBuffer())) {
heap_->array_buffer_tracker()->Promote(
JSArrayBuffer::cast(target_object));
}
promoted_size_ += size;
return true;
}
HeapObject* target = nullptr;
AllocationSpace space = AllocateTargetObject(object, &target);
heap_->mark_compact_collector()->MigrateObject(
HeapObject::cast(target), object, size, space,
(space == NEW_SPACE) ? nullptr : evacuation_slots_buffer_);
if (V8_UNLIKELY(target->IsJSArrayBuffer())) {
heap_->array_buffer_tracker()->MarkLive(JSArrayBuffer::cast(target));
}
semispace_copied_size_ += size;
return true;
}
intptr_t promoted_size() { return promoted_size_; }
intptr_t semispace_copied_size() { return semispace_copied_size_; }
private:
enum NewSpaceAllocationMode {
kNonstickyBailoutOldSpace,
kStickyBailoutOldSpace,
};
inline AllocationSpace AllocateTargetObject(HeapObject* old_object,
HeapObject** target_object) {
const int size = old_object->Size();
AllocationAlignment alignment = old_object->RequiredAlignment();
AllocationResult allocation;
if (space_to_allocate_ == NEW_SPACE) {
if (size > kMaxLabObjectSize) {
allocation =
AllocateInNewSpace(size, alignment, kNonstickyBailoutOldSpace);
} else {
allocation = AllocateInLab(size, alignment);
}
}
if (allocation.IsRetry() || (space_to_allocate_ == OLD_SPACE)) {
allocation = AllocateInOldSpace(size, alignment);
}
bool ok = allocation.To(target_object);
DCHECK(ok);
USE(ok);
return space_to_allocate_;
}
inline bool NewLocalAllocationBuffer() {
AllocationResult result =
AllocateInNewSpace(kLabSize, kWordAligned, kStickyBailoutOldSpace);
LocalAllocationBuffer saved_old_buffer = buffer_;
buffer_ = LocalAllocationBuffer::FromResult(heap_, result, kLabSize);
if (buffer_.IsValid()) {
buffer_.TryMerge(&saved_old_buffer);
return true;
}
return false;
}
inline AllocationResult AllocateInNewSpace(int size_in_bytes,
AllocationAlignment alignment,
NewSpaceAllocationMode mode) {
AllocationResult allocation =
heap_->new_space()->AllocateRawSynchronized(size_in_bytes, alignment);
if (allocation.IsRetry()) {
if (!heap_->new_space()->AddFreshPageSynchronized()) {
if (mode == kStickyBailoutOldSpace) space_to_allocate_ = OLD_SPACE;
} else {
allocation = heap_->new_space()->AllocateRawSynchronized(size_in_bytes,
alignment);
if (allocation.IsRetry()) {
if (mode == kStickyBailoutOldSpace) space_to_allocate_ = OLD_SPACE;
}
}
}
return allocation;
}
inline AllocationResult AllocateInOldSpace(int size_in_bytes,
AllocationAlignment alignment) {
AllocationResult allocation =
heap_->old_space()->AllocateRaw(size_in_bytes, alignment);
if (allocation.IsRetry()) {
FatalProcessOutOfMemory(
"MarkCompactCollector: semi-space copy, fallback in old gen\n");
}
return allocation;
}
inline AllocationResult AllocateInLab(int size_in_bytes,
AllocationAlignment alignment) {
AllocationResult allocation;
if (!buffer_.IsValid()) {
if (!NewLocalAllocationBuffer()) {
space_to_allocate_ = OLD_SPACE;
return AllocationResult::Retry(OLD_SPACE);
}
}
allocation = buffer_.AllocateRawAligned(size_in_bytes, alignment);
if (allocation.IsRetry()) {
if (!NewLocalAllocationBuffer()) {
space_to_allocate_ = OLD_SPACE;
return AllocationResult::Retry(OLD_SPACE);
} else {
allocation = buffer_.AllocateRawAligned(size_in_bytes, alignment);
if (allocation.IsRetry()) {
space_to_allocate_ = OLD_SPACE;
return AllocationResult::Retry(OLD_SPACE);
}
}
}
return allocation;
}
LocalAllocationBuffer buffer_;
AllocationSpace space_to_allocate_;
intptr_t promoted_size_;
intptr_t semispace_copied_size_;
HashMap* local_pretenuring_feedback_;
};
class MarkCompactCollector::EvacuateOldSpaceVisitor final
: public MarkCompactCollector::EvacuateVisitorBase {
public:
EvacuateOldSpaceVisitor(Heap* heap,
CompactionSpaceCollection* compaction_spaces,
SlotsBuffer** evacuation_slots_buffer)
: EvacuateVisitorBase(heap, evacuation_slots_buffer),
compaction_spaces_(compaction_spaces) {}
bool Visit(HeapObject* object) override {
CompactionSpace* target_space = compaction_spaces_->Get(
Page::FromAddress(object->address())->owner()->identity());
HeapObject* target_object = nullptr;
if (TryEvacuateObject(target_space, object, &target_object)) {
DCHECK(object->map_word().IsForwardingAddress());
return true;
}
return false;
}
private:
CompactionSpaceCollection* compaction_spaces_;
};
void MarkCompactCollector::DiscoverGreyObjectsInSpace(PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
DiscoverGreyObjectsOnPage(p);
if (marking_deque()->IsFull()) return;
}
}
void MarkCompactCollector::DiscoverGreyObjectsInNewSpace() {
NewSpace* space = heap()->new_space();
NewSpacePageIterator it(space->bottom(), space->top());
while (it.has_next()) {
NewSpacePage* page = it.next();
DiscoverGreyObjectsOnPage(page);
if (marking_deque()->IsFull()) return;
}
}
bool MarkCompactCollector::IsUnmarkedHeapObject(Object** p) {
Object* o = *p;
if (!o->IsHeapObject()) return false;
HeapObject* heap_object = HeapObject::cast(o);
MarkBit mark = Marking::MarkBitFrom(heap_object);
return Marking::IsWhite(mark);
}
bool MarkCompactCollector::IsUnmarkedHeapObjectWithHeap(Heap* heap,
Object** p) {
Object* o = *p;
DCHECK(o->IsHeapObject());
HeapObject* heap_object = HeapObject::cast(o);
MarkBit mark = Marking::MarkBitFrom(heap_object);
return Marking::IsWhite(mark);
}
void MarkCompactCollector::MarkStringTable(RootMarkingVisitor* visitor) {
StringTable* string_table = heap()->string_table();
// Mark the string table itself.
MarkBit string_table_mark = Marking::MarkBitFrom(string_table);
if (Marking::IsWhite(string_table_mark)) {
// String table could have already been marked by visiting the handles list.
SetMark(string_table, string_table_mark);
}
// Explicitly mark the prefix.
string_table->IteratePrefix(visitor);
ProcessMarkingDeque();
}
void MarkCompactCollector::MarkAllocationSite(AllocationSite* site) {
MarkBit mark_bit = Marking::MarkBitFrom(site);
SetMark(site, mark_bit);
}
void MarkCompactCollector::MarkRoots(RootMarkingVisitor* visitor) {
// Mark the heap roots including global variables, stack variables,
// etc., and all objects reachable from them.
heap()->IterateStrongRoots(visitor, VISIT_ONLY_STRONG);
// Handle the string table specially.
MarkStringTable(visitor);
// There may be overflowed objects in the heap. Visit them now.
while (marking_deque_.overflowed()) {
RefillMarkingDeque();
EmptyMarkingDeque();
}
}
void MarkCompactCollector::MarkImplicitRefGroups(
MarkObjectFunction mark_object) {
List<ImplicitRefGroup*>* ref_groups =
isolate()->global_handles()->implicit_ref_groups();
int last = 0;
for (int i = 0; i < ref_groups->length(); i++) {
ImplicitRefGroup* entry = ref_groups->at(i);
DCHECK(entry != NULL);
if (!IsMarked(*entry->parent)) {
(*ref_groups)[last++] = entry;
continue;
}
Object*** children = entry->children;
// A parent object is marked, so mark all child heap objects.
for (size_t j = 0; j < entry->length; ++j) {
if ((*children[j])->IsHeapObject()) {
mark_object(heap(), HeapObject::cast(*children[j]));
}
}
// Once the entire group has been marked, dispose it because it's
// not needed anymore.
delete entry;
}
ref_groups->Rewind(last);
}
// Mark all objects reachable from the objects on the marking stack.
// Before: the marking stack contains zero or more heap object pointers.
// After: the marking stack is empty, and all objects reachable from the
// marking stack have been marked, or are overflowed in the heap.
void MarkCompactCollector::EmptyMarkingDeque() {
Map* filler_map = heap_->one_pointer_filler_map();
while (!marking_deque_.IsEmpty()) {
HeapObject* object = marking_deque_.Pop();
// Explicitly skip one word fillers. Incremental markbit patterns are
// correct only for objects that occupy at least two words.
Map* map = object->map();
if (map == filler_map) continue;
DCHECK(object->IsHeapObject());
DCHECK(heap()->Contains(object));
DCHECK(!Marking::IsWhite(Marking::MarkBitFrom(object)));
MarkBit map_mark = Marking::MarkBitFrom(map);
MarkObject(map, map_mark);
MarkCompactMarkingVisitor::IterateBody(map, object);
}
}
// Sweep the heap for overflowed objects, clear their overflow bits, and
// push them on the marking stack. Stop early if the marking stack fills
// before sweeping completes. If sweeping completes, there are no remaining
// overflowed objects in the heap so the overflow flag on the markings stack
// is cleared.
void MarkCompactCollector::RefillMarkingDeque() {
isolate()->CountUsage(v8::Isolate::UseCounterFeature::kMarkDequeOverflow);
DCHECK(marking_deque_.overflowed());
DiscoverGreyObjectsInNewSpace();
if (marking_deque_.IsFull()) return;
DiscoverGreyObjectsInSpace(heap()->old_space());
if (marking_deque_.IsFull()) return;
DiscoverGreyObjectsInSpace(heap()->code_space());
if (marking_deque_.IsFull()) return;
DiscoverGreyObjectsInSpace(heap()->map_space());
if (marking_deque_.IsFull()) return;
LargeObjectIterator lo_it(heap()->lo_space());
DiscoverGreyObjectsWithIterator(&lo_it);
if (marking_deque_.IsFull()) return;
marking_deque_.ClearOverflowed();
}
// Mark all objects reachable (transitively) from objects on the marking
// stack. Before: the marking stack contains zero or more heap object
// pointers. After: the marking stack is empty and there are no overflowed
// objects in the heap.
void MarkCompactCollector::ProcessMarkingDeque() {
EmptyMarkingDeque();
while (marking_deque_.overflowed()) {
RefillMarkingDeque();
EmptyMarkingDeque();
}
}
// Mark all objects reachable (transitively) from objects on the marking
// stack including references only considered in the atomic marking pause.
void MarkCompactCollector::ProcessEphemeralMarking(
ObjectVisitor* visitor, bool only_process_harmony_weak_collections) {
bool work_to_do = true;
DCHECK(marking_deque_.IsEmpty() && !marking_deque_.overflowed());
while (work_to_do) {
if (!only_process_harmony_weak_collections) {
isolate()->global_handles()->IterateObjectGroups(
visitor, &IsUnmarkedHeapObjectWithHeap);
MarkImplicitRefGroups(&MarkCompactMarkingVisitor::MarkObject);
}
ProcessWeakCollections();
work_to_do = !marking_deque_.IsEmpty();
ProcessMarkingDeque();
}
}
void MarkCompactCollector::ProcessTopOptimizedFrame(ObjectVisitor* visitor) {
for (StackFrameIterator it(isolate(), isolate()->thread_local_top());
!it.done(); it.Advance()) {
if (it.frame()->type() == StackFrame::JAVA_SCRIPT) {
return;
}
if (it.frame()->type() == StackFrame::OPTIMIZED) {
Code* code = it.frame()->LookupCode();
if (!code->CanDeoptAt(it.frame()->pc())) {
Code::BodyDescriptor::IterateBody(code, visitor);
}
ProcessMarkingDeque();
return;
}
}
}
void MarkCompactCollector::EnsureMarkingDequeIsReserved() {
DCHECK(!marking_deque_.in_use());
if (marking_deque_memory_ == NULL) {
marking_deque_memory_ = new base::VirtualMemory(kMaxMarkingDequeSize);
marking_deque_memory_committed_ = 0;
}
if (marking_deque_memory_ == NULL) {
V8::FatalProcessOutOfMemory("EnsureMarkingDequeIsReserved");
}
}
void MarkCompactCollector::EnsureMarkingDequeIsCommitted(size_t max_size) {
// If the marking deque is too small, we try to allocate a bigger one.
// If that fails, make do with a smaller one.
CHECK(!marking_deque_.in_use());
for (size_t size = max_size; size >= kMinMarkingDequeSize; size >>= 1) {
base::VirtualMemory* memory = marking_deque_memory_;
size_t currently_committed = marking_deque_memory_committed_;
if (currently_committed == size) return;
if (currently_committed > size) {
bool success = marking_deque_memory_->Uncommit(
reinterpret_cast<Address>(marking_deque_memory_->address()) + size,
currently_committed - size);
if (success) {
marking_deque_memory_committed_ = size;
return;
}
UNREACHABLE();
}
bool success = memory->Commit(
reinterpret_cast<Address>(memory->address()) + currently_committed,
size - currently_committed,
false); // Not executable.
if (success) {
marking_deque_memory_committed_ = size;
return;
}
}
V8::FatalProcessOutOfMemory("EnsureMarkingDequeIsCommitted");
}
void MarkCompactCollector::InitializeMarkingDeque() {
DCHECK(!marking_deque_.in_use());
DCHECK(marking_deque_memory_committed_ > 0);
Address addr = static_cast<Address>(marking_deque_memory_->address());
size_t size = marking_deque_memory_committed_;
if (FLAG_force_marking_deque_overflows) size = 64 * kPointerSize;
marking_deque_.Initialize(addr, addr + size);
}
void MarkingDeque::Initialize(Address low, Address high) {
DCHECK(!in_use_);
HeapObject** obj_low = reinterpret_cast<HeapObject**>(low);
HeapObject** obj_high = reinterpret_cast<HeapObject**>(high);
array_ = obj_low;
mask_ = base::bits::RoundDownToPowerOfTwo32(
static_cast<uint32_t>(obj_high - obj_low)) -
1;
top_ = bottom_ = 0;
overflowed_ = false;
in_use_ = true;
}
void MarkingDeque::Uninitialize(bool aborting) {
if (!aborting) {
DCHECK(IsEmpty());
DCHECK(!overflowed_);
}
DCHECK(in_use_);
top_ = bottom_ = 0xdecbad;
in_use_ = false;
}
void MarkCompactCollector::MarkLiveObjects() {
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_MARK);
double start_time = 0.0;
if (FLAG_print_cumulative_gc_stat) {
start_time = heap_->MonotonicallyIncreasingTimeInMs();
}
// The recursive GC marker detects when it is nearing stack overflow,
// and switches to a different marking system. JS interrupts interfere
// with the C stack limit check.
PostponeInterruptsScope postpone(isolate());
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_MARK_FINISH_INCREMENTAL);
IncrementalMarking* incremental_marking = heap_->incremental_marking();
if (was_marked_incrementally_) {
incremental_marking->Finalize();
} else {
// Abort any pending incremental activities e.g. incremental sweeping.
incremental_marking->Stop();
if (marking_deque_.in_use()) {
marking_deque_.Uninitialize(true);
}
}
}
#ifdef DEBUG
DCHECK(state_ == PREPARE_GC);
state_ = MARK_LIVE_OBJECTS;
#endif
EnsureMarkingDequeIsCommittedAndInitialize(
MarkCompactCollector::kMaxMarkingDequeSize);
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_MARK_PREPARE_CODE_FLUSH);
PrepareForCodeFlushing();
}
RootMarkingVisitor root_visitor(heap());
{
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_MARK_ROOTS);
MarkRoots(&root_visitor);
ProcessTopOptimizedFrame(&root_visitor);
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_MARK_WEAK_CLOSURE);
// The objects reachable from the roots are marked, yet unreachable
// objects are unmarked. Mark objects reachable due to host
// application specific logic or through Harmony weak maps.
ProcessEphemeralMarking(&root_visitor, false);
// The objects reachable from the roots, weak maps or object groups
// are marked. Objects pointed to only by weak global handles cannot be
// immediately reclaimed. Instead, we have to mark them as pending and mark
// objects reachable from them.
//
// First we identify nonlive weak handles and mark them as pending
// destruction.
heap()->isolate()->global_handles()->IdentifyWeakHandles(
&IsUnmarkedHeapObject);
// Then we mark the objects.
heap()->isolate()->global_handles()->IterateWeakRoots(&root_visitor);
ProcessMarkingDeque();
// Repeat Harmony weak maps marking to mark unmarked objects reachable from
// the weak roots we just marked as pending destruction.
//
// We only process harmony collections, as all object groups have been fully
// processed and no weakly reachable node can discover new objects groups.
ProcessEphemeralMarking(&root_visitor, true);
}
if (FLAG_print_cumulative_gc_stat) {
heap_->tracer()->AddMarkingTime(heap_->MonotonicallyIncreasingTimeInMs() -
start_time);
}
if (FLAG_track_gc_object_stats) {
if (FLAG_trace_gc_object_stats) {
heap()->object_stats_->TraceObjectStats();
}
heap()->object_stats_->CheckpointObjectStats();
}
}
void MarkCompactCollector::ClearNonLiveReferences() {
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_CLEAR);
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_STRING_TABLE);
// Prune the string table removing all strings only pointed to by the
// string table. Cannot use string_table() here because the string
// table is marked.
StringTable* string_table = heap()->string_table();
InternalizedStringTableCleaner internalized_visitor(heap());
string_table->IterateElements(&internalized_visitor);
string_table->ElementsRemoved(internalized_visitor.PointersRemoved());
ExternalStringTableCleaner external_visitor(heap());
heap()->external_string_table_.Iterate(&external_visitor);
heap()->external_string_table_.CleanUp();
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_WEAK_LISTS);
// Process the weak references.
MarkCompactWeakObjectRetainer mark_compact_object_retainer;
heap()->ProcessAllWeakReferences(&mark_compact_object_retainer);
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_GLOBAL_HANDLES);
// Remove object groups after marking phase.
heap()->isolate()->global_handles()->RemoveObjectGroups();
heap()->isolate()->global_handles()->RemoveImplicitRefGroups();
}
// Flush code from collected candidates.
if (is_code_flushing_enabled()) {
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_CODE_FLUSH);
code_flusher_->ProcessCandidates();
}
DependentCode* dependent_code_list;
Object* non_live_map_list;
ClearWeakCells(&non_live_map_list, &dependent_code_list);
{
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_CLEAR_MAPS);
ClearSimpleMapTransitions(non_live_map_list);
ClearFullMapTransitions();
}
MarkDependentCodeForDeoptimization(dependent_code_list);
ClearWeakCollections();
ClearInvalidStoreAndSlotsBufferEntries();
}
void MarkCompactCollector::MarkDependentCodeForDeoptimization(
DependentCode* list_head) {
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_DEPENDENT_CODE);
Isolate* isolate = this->isolate();
DependentCode* current = list_head;
while (current->length() > 0) {
have_code_to_deoptimize_ |= current->MarkCodeForDeoptimization(
isolate, DependentCode::kWeakCodeGroup);
current = current->next_link();
}
WeakHashTable* table = heap_->weak_object_to_code_table();
uint32_t capacity = table->Capacity();
for (uint32_t i = 0; i < capacity; i++) {
uint32_t key_index = table->EntryToIndex(i);
Object* key = table->get(key_index);
if (!table->IsKey(key)) continue;
uint32_t value_index = table->EntryToValueIndex(i);
Object* value = table->get(value_index);
DCHECK(key->IsWeakCell());
if (WeakCell::cast(key)->cleared()) {
have_code_to_deoptimize_ |=
DependentCode::cast(value)->MarkCodeForDeoptimization(
isolate, DependentCode::kWeakCodeGroup);
table->set(key_index, heap_->the_hole_value());
table->set(value_index, heap_->the_hole_value());
table->ElementRemoved();
}
}
}
void MarkCompactCollector::ClearSimpleMapTransitions(
Object* non_live_map_list) {
Object* the_hole_value = heap()->the_hole_value();
Object* weak_cell_obj = non_live_map_list;
while (weak_cell_obj != Smi::FromInt(0)) {
WeakCell* weak_cell = WeakCell::cast(weak_cell_obj);
Map* map = Map::cast(weak_cell->value());
DCHECK(Marking::IsWhite(Marking::MarkBitFrom(map)));
Object* potential_parent = map->constructor_or_backpointer();
if (potential_parent->IsMap()) {
Map* parent = Map::cast(potential_parent);
if (Marking::IsBlackOrGrey(Marking::MarkBitFrom(parent)) &&
parent->raw_transitions() == weak_cell) {
ClearSimpleMapTransition(parent, map);
}
}
weak_cell->clear();
weak_cell_obj = weak_cell->next();
weak_cell->clear_next(the_hole_value);
}
}
void MarkCompactCollector::ClearSimpleMapTransition(Map* map,
Map* dead_transition) {
// A previously existing simple transition (stored in a WeakCell) is going
// to be cleared. Clear the useless cell pointer, and take ownership
// of the descriptor array.
map->set_raw_transitions(Smi::FromInt(0));
int number_of_own_descriptors = map->NumberOfOwnDescriptors();
DescriptorArray* descriptors = map->instance_descriptors();
if (descriptors == dead_transition->instance_descriptors() &&
number_of_own_descriptors > 0) {
TrimDescriptorArray(map, descriptors);
DCHECK(descriptors->number_of_descriptors() == number_of_own_descriptors);
map->set_owns_descriptors(true);
}
}
void MarkCompactCollector::ClearFullMapTransitions() {
HeapObject* undefined = heap()->undefined_value();
Object* obj = heap()->encountered_transition_arrays();
while (obj != Smi::FromInt(0)) {
TransitionArray* array = TransitionArray::cast(obj);
int num_transitions = array->number_of_entries();
DCHECK_EQ(TransitionArray::NumberOfTransitions(array), num_transitions);
if (num_transitions > 0) {
Map* map = array->GetTarget(0);
Map* parent = Map::cast(map->constructor_or_backpointer());
bool parent_is_alive =
Marking::IsBlackOrGrey(Marking::MarkBitFrom(parent));
DescriptorArray* descriptors =
parent_is_alive ? parent->instance_descriptors() : nullptr;
bool descriptors_owner_died =
CompactTransitionArray(parent, array, descriptors);
if (descriptors_owner_died) {
TrimDescriptorArray(parent, descriptors);
}
}
obj = array->next_link();
array->set_next_link(undefined, SKIP_WRITE_BARRIER);
}
heap()->set_encountered_transition_arrays(Smi::FromInt(0));
}
bool MarkCompactCollector::CompactTransitionArray(
Map* map, TransitionArray* transitions, DescriptorArray* descriptors) {
int num_transitions = transitions->number_of_entries();
bool descriptors_owner_died = false;
int transition_index = 0;
// Compact all live transitions to the left.
for (int i = 0; i < num_transitions; ++i) {
Map* target = transitions->GetTarget(i);
DCHECK_EQ(target->constructor_or_backpointer(), map);
if (Marking::IsWhite(Marking::MarkBitFrom(target))) {
if (descriptors != nullptr &&
target->instance_descriptors() == descriptors) {
descriptors_owner_died = true;
}
} else {
if (i != transition_index) {
Name* key = transitions->GetKey(i);
transitions->SetKey(transition_index, key);
Object** key_slot = transitions->GetKeySlot(transition_index);
RecordSlot(transitions, key_slot, key);
// Target slots do not need to be recorded since maps are not compacted.
transitions->SetTarget(transition_index, transitions->GetTarget(i));
}
transition_index++;
}
}
// If there are no transitions to be cleared, return.
if (transition_index == num_transitions) {
DCHECK(!descriptors_owner_died);
return false;
}
// Note that we never eliminate a transition array, though we might right-trim
// such that number_of_transitions() == 0. If this assumption changes,
// TransitionArray::Insert() will need to deal with the case that a transition
// array disappeared during GC.
int trim = TransitionArray::Capacity(transitions) - transition_index;
if (trim > 0) {
heap_->RightTrimFixedArray<Heap::SEQUENTIAL_TO_SWEEPER>(
transitions, trim * TransitionArray::kTransitionSize);
transitions->SetNumberOfTransitions(transition_index);
}
return descriptors_owner_died;
}
void MarkCompactCollector::TrimDescriptorArray(Map* map,
DescriptorArray* descriptors) {
int number_of_own_descriptors = map->NumberOfOwnDescriptors();
if (number_of_own_descriptors == 0) {
DCHECK(descriptors == heap_->empty_descriptor_array());
return;
}
int number_of_descriptors = descriptors->number_of_descriptors_storage();
int to_trim = number_of_descriptors - number_of_own_descriptors;
if (to_trim > 0) {
heap_->RightTrimFixedArray<Heap::SEQUENTIAL_TO_SWEEPER>(
descriptors, to_trim * DescriptorArray::kDescriptorSize);
descriptors->SetNumberOfDescriptors(number_of_own_descriptors);
if (descriptors->HasEnumCache()) TrimEnumCache(map, descriptors);
descriptors->Sort();
if (FLAG_unbox_double_fields) {
LayoutDescriptor* layout_descriptor = map->layout_descriptor();
layout_descriptor = layout_descriptor->Trim(heap_, map, descriptors,
number_of_own_descriptors);
SLOW_DCHECK(layout_descriptor->IsConsistentWithMap(map, true));
}
}
DCHECK(descriptors->number_of_descriptors() == number_of_own_descriptors);
map->set_owns_descriptors(true);
}
void MarkCompactCollector::TrimEnumCache(Map* map,
DescriptorArray* descriptors) {
int live_enum = map->EnumLength();
if (live_enum == kInvalidEnumCacheSentinel) {
live_enum =
map->NumberOfDescribedProperties(OWN_DESCRIPTORS, ENUMERABLE_STRINGS);
}
if (live_enum == 0) return descriptors->ClearEnumCache();
FixedArray* enum_cache = descriptors->GetEnumCache();
int to_trim = enum_cache->length() - live_enum;
if (to_trim <= 0) return;
heap_->RightTrimFixedArray<Heap::SEQUENTIAL_TO_SWEEPER>(
descriptors->GetEnumCache(), to_trim);
if (!descriptors->HasEnumIndicesCache()) return;
FixedArray* enum_indices_cache = descriptors->GetEnumIndicesCache();
heap_->RightTrimFixedArray<Heap::SEQUENTIAL_TO_SWEEPER>(enum_indices_cache,
to_trim);
}
void MarkCompactCollector::ProcessWeakCollections() {
Object* weak_collection_obj = heap()->encountered_weak_collections();
while (weak_collection_obj != Smi::FromInt(0)) {
JSWeakCollection* weak_collection =
reinterpret_cast<JSWeakCollection*>(weak_collection_obj);
DCHECK(MarkCompactCollector::IsMarked(weak_collection));
if (weak_collection->table()->IsHashTable()) {
ObjectHashTable* table = ObjectHashTable::cast(weak_collection->table());
for (int i = 0; i < table->Capacity(); i++) {
if (MarkCompactCollector::IsMarked(HeapObject::cast(table->KeyAt(i)))) {
Object** key_slot =
table->RawFieldOfElementAt(ObjectHashTable::EntryToIndex(i));
RecordSlot(table, key_slot, *key_slot);
Object** value_slot =
table->RawFieldOfElementAt(ObjectHashTable::EntryToValueIndex(i));
MarkCompactMarkingVisitor::MarkObjectByPointer(this, table,
value_slot);
}
}
}
weak_collection_obj = weak_collection->next();
}
}
void MarkCompactCollector::ClearWeakCollections() {
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_CLEAR_WEAK_COLLECTIONS);
Object* weak_collection_obj = heap()->encountered_weak_collections();
while (weak_collection_obj != Smi::FromInt(0)) {
JSWeakCollection* weak_collection =
reinterpret_cast<JSWeakCollection*>(weak_collection_obj);
DCHECK(MarkCompactCollector::IsMarked(weak_collection));
if (weak_collection->table()->IsHashTable()) {
ObjectHashTable* table = ObjectHashTable::cast(weak_collection->table());
for (int i = 0; i < table->Capacity(); i++) {
HeapObject* key = HeapObject::cast(table->KeyAt(i));
if (!MarkCompactCollector::IsMarked(key)) {
table->RemoveEntry(i);
}
}
}
weak_collection_obj = weak_collection->next();
weak_collection->set_next(heap()->undefined_value());
}
heap()->set_encountered_weak_collections(Smi::FromInt(0));
}
void MarkCompactCollector::AbortWeakCollections() {
Object* weak_collection_obj = heap()->encountered_weak_collections();
while (weak_collection_obj != Smi::FromInt(0)) {
JSWeakCollection* weak_collection =
reinterpret_cast<JSWeakCollection*>(weak_collection_obj);
weak_collection_obj = weak_collection->next();
weak_collection->set_next(heap()->undefined_value());
}
heap()->set_encountered_weak_collections(Smi::FromInt(0));
}
void MarkCompactCollector::ClearWeakCells(Object** non_live_map_list,
DependentCode** dependent_code_list) {
Heap* heap = this->heap();
GCTracer::Scope gc_scope(heap->tracer(),
GCTracer::Scope::MC_CLEAR_WEAK_CELLS);
Object* weak_cell_obj = heap->encountered_weak_cells();
Object* the_hole_value = heap->the_hole_value();
DependentCode* dependent_code_head =
DependentCode::cast(heap->empty_fixed_array());
Object* non_live_map_head = Smi::FromInt(0);
while (weak_cell_obj != Smi::FromInt(0)) {
WeakCell* weak_cell = reinterpret_cast<WeakCell*>(weak_cell_obj);
Object* next_weak_cell = weak_cell->next();
bool clear_value = true;
bool clear_next = true;
// We do not insert cleared weak cells into the list, so the value
// cannot be a Smi here.
HeapObject* value = HeapObject::cast(weak_cell->value());
if (!MarkCompactCollector::IsMarked(value)) {
// Cells for new-space objects embedded in optimized code are wrapped in
// WeakCell and put into Heap::weak_object_to_code_table.
// Such cells do not have any strong references but we want to keep them
// alive as long as the cell value is alive.
// TODO(ulan): remove this once we remove Heap::weak_object_to_code_table.
if (value->IsCell()) {
Object* cell_value = Cell::cast(value)->value();
if (cell_value->IsHeapObject() &&
MarkCompactCollector::IsMarked(HeapObject::cast(cell_value))) {
// Resurrect the cell.
MarkBit mark = Marking::MarkBitFrom(value);
SetMark(value, mark);
Object** slot = HeapObject::RawField(value, Cell::kValueOffset);
RecordSlot(value, slot, *slot);
slot = HeapObject::RawField(weak_cell, WeakCell::kValueOffset);
RecordSlot(weak_cell, slot, *slot);
clear_value = false;
}
}
if (value->IsMap()) {
// The map is non-live.
Map* map = Map::cast(value);
// Add dependent code to the dependent_code_list.
DependentCode* candidate = map->dependent_code();
// We rely on the fact that the weak code group comes first.
STATIC_ASSERT(DependentCode::kWeakCodeGroup == 0);
if (candidate->length() > 0 &&
candidate->group() == DependentCode::kWeakCodeGroup) {
candidate->set_next_link(dependent_code_head);
dependent_code_head = candidate;
}
// Add the weak cell to the non_live_map list.
weak_cell->set_next(non_live_map_head);
non_live_map_head = weak_cell;
clear_value = false;
clear_next = false;
}
} else {
// The value of the weak cell is alive.
Object** slot = HeapObject::RawField(weak_cell, WeakCell::kValueOffset);
RecordSlot(weak_cell, slot, *slot);
clear_value = false;
}
if (clear_value) {
weak_cell->clear();
}
if (clear_next) {
weak_cell->clear_next(the_hole_value);
}
weak_cell_obj = next_weak_cell;
}
heap->set_encountered_weak_cells(Smi::FromInt(0));
*non_live_map_list = non_live_map_head;
*dependent_code_list = dependent_code_head;
}
void MarkCompactCollector::AbortWeakCells() {
Object* the_hole_value = heap()->the_hole_value();
Object* weak_cell_obj = heap()->encountered_weak_cells();
while (weak_cell_obj != Smi::FromInt(0)) {
WeakCell* weak_cell = reinterpret_cast<WeakCell*>(weak_cell_obj);
weak_cell_obj = weak_cell->next();
weak_cell->clear_next(the_hole_value);
}
heap()->set_encountered_weak_cells(Smi::FromInt(0));
}
void MarkCompactCollector::AbortTransitionArrays() {
HeapObject* undefined = heap()->undefined_value();
Object* obj = heap()->encountered_transition_arrays();
while (obj != Smi::FromInt(0)) {
TransitionArray* array = TransitionArray::cast(obj);
obj = array->next_link();
array->set_next_link(undefined, SKIP_WRITE_BARRIER);
}
heap()->set_encountered_transition_arrays(Smi::FromInt(0));
}
void MarkCompactCollector::RecordMigratedSlot(
Object* value, Address slot, SlotsBuffer** evacuation_slots_buffer) {
// When parallel compaction is in progress, store and slots buffer entries
// require synchronization.
if (heap_->InNewSpace(value)) {
if (compaction_in_progress_) {
heap_->store_buffer()->MarkSynchronized(slot);
} else {
heap_->store_buffer()->Mark(slot);
}
} else if (value->IsHeapObject() && IsOnEvacuationCandidate(value)) {
SlotsBuffer::AddTo(slots_buffer_allocator_, evacuation_slots_buffer,
reinterpret_cast<Object**>(slot),
SlotsBuffer::IGNORE_OVERFLOW);
}
}
void MarkCompactCollector::RecordMigratedCodeEntrySlot(
Address code_entry, Address code_entry_slot,
SlotsBuffer** evacuation_slots_buffer) {
if (Page::FromAddress(code_entry)->IsEvacuationCandidate()) {
SlotsBuffer::AddTo(slots_buffer_allocator_, evacuation_slots_buffer,
SlotsBuffer::CODE_ENTRY_SLOT, code_entry_slot,
SlotsBuffer::IGNORE_OVERFLOW);
}
}
void MarkCompactCollector::RecordMigratedCodeObjectSlot(
Address code_object, SlotsBuffer** evacuation_slots_buffer) {
SlotsBuffer::AddTo(slots_buffer_allocator_, evacuation_slots_buffer,
SlotsBuffer::RELOCATED_CODE_OBJECT, code_object,
SlotsBuffer::IGNORE_OVERFLOW);
}
static inline SlotsBuffer::SlotType SlotTypeForRMode(RelocInfo::Mode rmode) {
if (RelocInfo::IsCodeTarget(rmode)) {
return SlotsBuffer::CODE_TARGET_SLOT;
} else if (RelocInfo::IsCell(rmode)) {
return SlotsBuffer::CELL_TARGET_SLOT;
} else if (RelocInfo::IsEmbeddedObject(rmode)) {
return SlotsBuffer::EMBEDDED_OBJECT_SLOT;
} else if (RelocInfo::IsDebugBreakSlot(rmode)) {
return SlotsBuffer::DEBUG_TARGET_SLOT;
}
UNREACHABLE();
return SlotsBuffer::NUMBER_OF_SLOT_TYPES;
}
static inline SlotsBuffer::SlotType DecodeSlotType(
SlotsBuffer::ObjectSlot slot) {
return static_cast<SlotsBuffer::SlotType>(reinterpret_cast<intptr_t>(slot));
}
void MarkCompactCollector::RecordRelocSlot(RelocInfo* rinfo, Object* target) {
Page* target_page = Page::FromAddress(reinterpret_cast<Address>(target));
RelocInfo::Mode rmode = rinfo->rmode();
if (target_page->IsEvacuationCandidate() &&
(rinfo->host() == NULL ||
!ShouldSkipEvacuationSlotRecording(rinfo->host()))) {
Address addr = rinfo->pc();
SlotsBuffer::SlotType slot_type = SlotTypeForRMode(rmode);
if (rinfo->IsInConstantPool()) {
addr = rinfo->constant_pool_entry_address();
if (RelocInfo::IsCodeTarget(rmode)) {
slot_type = SlotsBuffer::CODE_ENTRY_SLOT;
} else {
DCHECK(RelocInfo::IsEmbeddedObject(rmode));
slot_type = SlotsBuffer::OBJECT_SLOT;
}
}
bool success = SlotsBuffer::AddTo(
slots_buffer_allocator_, target_page->slots_buffer_address(), slot_type,
addr, SlotsBuffer::FAIL_ON_OVERFLOW);
if (!success) {
EvictPopularEvacuationCandidate(target_page);
}
}
}
class RecordMigratedSlotVisitor final : public ObjectVisitor {
public:
RecordMigratedSlotVisitor(MarkCompactCollector* collector,
SlotsBuffer** evacuation_slots_buffer)
: collector_(collector),
evacuation_slots_buffer_(evacuation_slots_buffer) {}
V8_INLINE void VisitPointer(Object** p) override {
collector_->RecordMigratedSlot(*p, reinterpret_cast<Address>(p),
evacuation_slots_buffer_);
}
V8_INLINE void VisitPointers(Object** start, Object** end) override {
while (start < end) {
collector_->RecordMigratedSlot(*start, reinterpret_cast<Address>(start),
evacuation_slots_buffer_);
++start;
}
}
V8_INLINE void VisitCodeEntry(Address code_entry_slot) override {
if (collector_->compacting_) {
Address code_entry = Memory::Address_at(code_entry_slot);
collector_->RecordMigratedCodeEntrySlot(code_entry, code_entry_slot,
evacuation_slots_buffer_);
}
}
private:
MarkCompactCollector* collector_;
SlotsBuffer** evacuation_slots_buffer_;
};
// We scavenge new space simultaneously with sweeping. This is done in two
// passes.
//
// The first pass migrates all alive objects from one semispace to another or
// promotes them to old space. Forwarding address is written directly into
// first word of object without any encoding. If object is dead we write
// NULL as a forwarding address.
//
// The second pass updates pointers to new space in all spaces. It is possible
// to encounter pointers to dead new space objects during traversal of pointers
// to new space. We should clear them to avoid encountering them during next
// pointer iteration. This is an issue if the store buffer overflows and we
// have to scan the entire old space, including dead objects, looking for
// pointers to new space.
void MarkCompactCollector::MigrateObject(
HeapObject* dst, HeapObject* src, int size, AllocationSpace dest,
SlotsBuffer** evacuation_slots_buffer) {
Address dst_addr = dst->address();
Address src_addr = src->address();
DCHECK(heap()->AllowedToBeMigrated(src, dest));
DCHECK(dest != LO_SPACE);
if (dest == OLD_SPACE) {
DCHECK_OBJECT_SIZE(size);
DCHECK(evacuation_slots_buffer != nullptr);
DCHECK(IsAligned(size, kPointerSize));
heap()->MoveBlock(dst->address(), src->address(), size);
RecordMigratedSlotVisitor visitor(this, evacuation_slots_buffer);
dst->IterateBody(&visitor);
} else if (dest == CODE_SPACE) {
DCHECK_CODEOBJECT_SIZE(size, heap()->code_space());
DCHECK(evacuation_slots_buffer != nullptr);
PROFILE(isolate(), CodeMoveEvent(src_addr, dst_addr));
heap()->MoveBlock(dst_addr, src_addr, size);
RecordMigratedCodeObjectSlot(dst_addr, evacuation_slots_buffer);
Code::cast(dst)->Relocate(dst_addr - src_addr);
} else {
DCHECK_OBJECT_SIZE(size);
DCHECK(evacuation_slots_buffer == nullptr);
DCHECK(dest == NEW_SPACE);
heap()->MoveBlock(dst_addr, src_addr, size);
}
heap()->OnMoveEvent(dst, src, size);
Memory::Address_at(src_addr) = dst_addr;
}
static inline void UpdateSlot(Isolate* isolate, ObjectVisitor* v,
SlotsBuffer::SlotType slot_type, Address addr) {
switch (slot_type) {
case SlotsBuffer::CODE_TARGET_SLOT: {
RelocInfo rinfo(isolate, addr, RelocInfo::CODE_TARGET, 0, NULL);
rinfo.Visit(isolate, v);
break;
}
case SlotsBuffer::CELL_TARGET_SLOT: {
RelocInfo rinfo(isolate, addr, RelocInfo::CELL, 0, NULL);
rinfo.Visit(isolate, v);
break;
}
case SlotsBuffer::CODE_ENTRY_SLOT: {
v->VisitCodeEntry(addr);
break;
}
case SlotsBuffer::RELOCATED_CODE_OBJECT: {
HeapObject* obj = HeapObject::FromAddress(addr);
Code::BodyDescriptor::IterateBody(obj, v);
break;
}
case SlotsBuffer::DEBUG_TARGET_SLOT: {
RelocInfo rinfo(isolate, addr, RelocInfo::DEBUG_BREAK_SLOT_AT_POSITION, 0,
NULL);
if (rinfo.IsPatchedDebugBreakSlotSequence()) rinfo.Visit(isolate, v);
break;
}
case SlotsBuffer::EMBEDDED_OBJECT_SLOT: {
RelocInfo rinfo(isolate, addr, RelocInfo::EMBEDDED_OBJECT, 0, NULL);
rinfo.Visit(isolate, v);
break;
}
case SlotsBuffer::OBJECT_SLOT: {
v->VisitPointer(reinterpret_cast<Object**>(addr));
break;
}
default:
UNREACHABLE();
break;
}
}
// Visitor for updating pointers from live objects in old spaces to new space.
// It does not expect to encounter pointers to dead objects.
class PointersUpdatingVisitor : public ObjectVisitor {
public:
explicit PointersUpdatingVisitor(Heap* heap) : heap_(heap) {}
void VisitPointer(Object** p) override { UpdatePointer(p); }
void VisitPointers(Object** start, Object** end) override {
for (Object** p = start; p < end; p++) UpdatePointer(p);
}
void VisitCell(RelocInfo* rinfo) override {
DCHECK(rinfo->rmode() == RelocInfo::CELL);
Object* cell = rinfo->target_cell();
Object* old_cell = cell;
VisitPointer(&cell);
if (cell != old_cell) {
rinfo->set_target_cell(reinterpret_cast<Cell*>(cell));
}
}
void VisitEmbeddedPointer(RelocInfo* rinfo) override {
DCHECK(rinfo->rmode() == RelocInfo::EMBEDDED_OBJECT);
Object* target = rinfo->target_object();
Object* old_target = target;
VisitPointer(&target);
// Avoid unnecessary changes that might unnecessary flush the instruction
// cache.
if (target != old_target) {
rinfo->set_target_object(target);
}
}
void VisitCodeTarget(RelocInfo* rinfo) override {
DCHECK(RelocInfo::IsCodeTarget(rinfo->rmode()));
Object* target = Code::GetCodeFromTargetAddress(rinfo->target_address());
Object* old_target = target;
VisitPointer(&target);
if (target != old_target) {
rinfo->set_target_address(Code::cast(target)->instruction_start());
}
}
void VisitCodeAgeSequence(RelocInfo* rinfo) override {
DCHECK(RelocInfo::IsCodeAgeSequence(rinfo->rmode()));
Object* stub = rinfo->code_age_stub();
DCHECK(stub != NULL);
VisitPointer(&stub);
if (stub != rinfo->code_age_stub()) {
rinfo->set_code_age_stub(Code::cast(stub));
}
}
void VisitDebugTarget(RelocInfo* rinfo) override {
DCHECK(RelocInfo::IsDebugBreakSlot(rinfo->rmode()) &&
rinfo->IsPatchedDebugBreakSlotSequence());
Object* target =
Code::GetCodeFromTargetAddress(rinfo->debug_call_address());
VisitPointer(&target);
rinfo->set_debug_call_address(Code::cast(target)->instruction_start());
}
static inline void UpdateSlot(Heap* heap, Object** slot) {
Object* obj = reinterpret_cast<Object*>(
base::NoBarrier_Load(reinterpret_cast<base::AtomicWord*>(slot)));
if (!obj->IsHeapObject()) return;
HeapObject* heap_obj = HeapObject::cast(obj);
MapWord map_word = heap_obj->map_word();
if (map_word.IsForwardingAddress()) {
DCHECK(heap->InFromSpace(heap_obj) ||
MarkCompactCollector::IsOnEvacuationCandidate(heap_obj) ||
Page::FromAddress(heap_obj->address())
->IsFlagSet(Page::COMPACTION_WAS_ABORTED));
HeapObject* target = map_word.ToForwardingAddress();
base::NoBarrier_CompareAndSwap(
reinterpret_cast<base::AtomicWord*>(slot),
reinterpret_cast<base::AtomicWord>(obj),
reinterpret_cast<base::AtomicWord>(target));
DCHECK(!heap->InFromSpace(target) &&
!MarkCompactCollector::IsOnEvacuationCandidate(target));
}
}
private:
inline void UpdatePointer(Object** p) { UpdateSlot(heap_, p); }
Heap* heap_;
};
void MarkCompactCollector::UpdateSlots(SlotsBuffer* buffer) {
PointersUpdatingVisitor v(heap_);
size_t buffer_size = buffer->Size();
for (size_t slot_idx = 0; slot_idx < buffer_size; ++slot_idx) {
SlotsBuffer::ObjectSlot slot = buffer->Get(slot_idx);
if (!SlotsBuffer::IsTypedSlot(slot)) {
PointersUpdatingVisitor::UpdateSlot(heap_, slot);
} else {
++slot_idx;
DCHECK(slot_idx < buffer_size);
UpdateSlot(heap_->isolate(), &v, DecodeSlotType(slot),
reinterpret_cast<Address>(buffer->Get(slot_idx)));
}
}
}
void MarkCompactCollector::UpdateSlotsRecordedIn(SlotsBuffer* buffer) {
while (buffer != NULL) {
UpdateSlots(buffer);
buffer = buffer->next();
}
}
static void UpdatePointer(HeapObject** address, HeapObject* object) {
MapWord map_word = object->map_word();
// The store buffer can still contain stale pointers in dead large objects.
// Ignore these pointers here.
DCHECK(map_word.IsForwardingAddress() ||
object->GetHeap()->lo_space()->FindPage(
reinterpret_cast<Address>(address)) != NULL);
if (map_word.IsForwardingAddress()) {
// Update the corresponding slot.
*address = map_word.ToForwardingAddress();
}
}
static String* UpdateReferenceInExternalStringTableEntry(Heap* heap,
Object** p) {
MapWord map_word = HeapObject::cast(*p)->map_word();
if (map_word.IsForwardingAddress()) {
return String::cast(map_word.ToForwardingAddress());
}
return String::cast(*p);
}
bool MarkCompactCollector::IsSlotInBlackObject(Page* p, Address slot,
HeapObject** out_object) {
Space* owner = p->owner();
if (owner == heap_->lo_space() || owner == NULL) {
Object* large_object = heap_->lo_space()->FindObject(slot);
// This object has to exist, otherwise we would not have recorded a slot
// for it.
CHECK(large_object->IsHeapObject());
HeapObject* large_heap_object = HeapObject::cast(large_object);
if (IsMarked(large_heap_object)) {
*out_object = large_heap_object;
return true;
}
return false;
}
uint32_t mark_bit_index = p->AddressToMarkbitIndex(slot);
unsigned int cell_index = mark_bit_index >> Bitmap::kBitsPerCellLog2;
MarkBit::CellType index_mask = 1u << Bitmap::IndexInCell(mark_bit_index);
MarkBit::CellType* cells = p->markbits()->cells();
Address base_address = p->area_start();
unsigned int base_address_cell_index = Bitmap::IndexToCell(
Bitmap::CellAlignIndex(p->AddressToMarkbitIndex(base_address)));
// Check if the slot points to the start of an object. This can happen e.g.
// when we left trim a fixed array. Such slots are invalid and we can remove
// them.
if (index_mask > 1) {
if ((cells[cell_index] & index_mask) != 0 &&
(cells[cell_index] & (index_mask >> 1)) == 0) {
return false;
}
} else {
// Left trimming moves the mark bits so we cannot be in the very first cell.
DCHECK(cell_index != base_address_cell_index);
if ((cells[cell_index] & index_mask) != 0 &&
(cells[cell_index - 1] & (1u << Bitmap::kBitIndexMask)) == 0) {
return false;
}
}
// Check if the object is in the current cell.
MarkBit::CellType slot_mask;
if ((cells[cell_index] == 0) ||
(base::bits::CountTrailingZeros32(cells[cell_index]) >
base::bits::CountTrailingZeros32(cells[cell_index] | index_mask))) {
// If we are already in the first cell, there is no live object.
if (cell_index == base_address_cell_index) return false;
// If not, find a cell in a preceding cell slot that has a mark bit set.
do {
cell_index--;
} while (cell_index > base_address_cell_index && cells[cell_index] == 0);
// The slot must be in a dead object if there are no preceding cells that
// have mark bits set.
if (cells[cell_index] == 0) {
return false;
}
// The object is in a preceding cell. Set the mask to find any object.
slot_mask = ~0u;
} else {
// We are interested in object mark bits right before the slot.
slot_mask = index_mask + (index_mask - 1);
}
MarkBit::CellType current_cell = cells[cell_index];
CHECK(current_cell != 0);
// Find the last live object in the cell.
unsigned int leading_zeros =
base::bits::CountLeadingZeros32(current_cell & slot_mask);
CHECK(leading_zeros != Bitmap::kBitsPerCell);
int offset = static_cast<int>(Bitmap::kBitIndexMask - leading_zeros) - 1;
base_address += (cell_index - base_address_cell_index) *
Bitmap::kBitsPerCell * kPointerSize;
Address address = base_address + offset * kPointerSize;
HeapObject* object = HeapObject::FromAddress(address);
CHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
CHECK(object->address() < reinterpret_cast<Address>(slot));
if ((object->address() + kPointerSize) <= slot &&
(object->address() + object->Size()) > slot) {
// If the slot is within the last found object in the cell, the slot is
// in a live object.
// Slots pointing to the first word of an object are invalid and removed.
// This can happen when we move the object header while left trimming.
*out_object = object;
return true;
}
return false;
}
bool MarkCompactCollector::IsSlotInBlackObjectSlow(Page* p, Address slot) {
// This function does not support large objects right now.
Space* owner = p->owner();
if (owner == heap_->lo_space() || owner == NULL) {
Object* large_object = heap_->lo_space()->FindObject(slot);
// This object has to exist, otherwise we would not have recorded a slot
// for it.
CHECK(large_object->IsHeapObject());
HeapObject* large_heap_object = HeapObject::cast(large_object);
if (IsMarked(large_heap_object)) {
return true;
}
return false;
}
LiveObjectIterator<kBlackObjects> it(p);
HeapObject* object = NULL;
while ((object = it.Next()) != NULL) {
int size = object->Size();
if (object->address() > slot) return false;
if (object->address() <= slot && slot < (object->address() + size)) {
return true;
}
}
return false;
}
bool MarkCompactCollector::IsSlotInLiveObject(Address slot) {
HeapObject* object = NULL;
// The target object is black but we don't know if the source slot is black.
// The source object could have died and the slot could be part of a free
// space. Find out based on mark bits if the slot is part of a live object.
if (!IsSlotInBlackObject(Page::FromAddress(slot), slot, &object)) {
return false;
}
DCHECK(object != NULL);
int offset = static_cast<int>(slot - object->address());
return object->IsValidSlot(offset);
}
void MarkCompactCollector::VerifyIsSlotInLiveObject(Address slot,
HeapObject* object) {
// The target object has to be black.
CHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
// The target object is black but we don't know if the source slot is black.
// The source object could have died and the slot could be part of a free
// space. Use the mark bit iterator to find out about liveness of the slot.
CHECK(IsSlotInBlackObjectSlow(Page::FromAddress(slot), slot));
}
void MarkCompactCollector::EvacuateNewSpacePrologue() {
// There are soft limits in the allocation code, designed trigger a mark
// sweep collection by failing allocations. But since we are already in
// a mark-sweep allocation, there is no sense in trying to trigger one.
AlwaysAllocateScope scope(isolate());
NewSpace* new_space = heap()->new_space();
// Store allocation range before flipping semispaces.
Address from_bottom = new_space->bottom();
Address from_top = new_space->top();
// Flip the semispaces. After flipping, to space is empty, from space has
// live objects.
new_space->Flip();
new_space->ResetAllocationInfo();
newspace_evacuation_candidates_.Clear();
NewSpacePageIterator it(from_bottom, from_top);
while (it.has_next()) {
newspace_evacuation_candidates_.Add(it.next());
}
}
HashMap* MarkCompactCollector::EvacuateNewSpaceInParallel() {
HashMap* local_pretenuring_feedback = new HashMap(
HashMap::PointersMatch, kInitialLocalPretenuringFeedbackCapacity);
EvacuateNewSpaceVisitor new_space_visitor(heap(), &migration_slots_buffer_,
local_pretenuring_feedback);
// First pass: traverse all objects in inactive semispace, remove marks,
// migrate live objects and write forwarding addresses. This stage puts
// new entries in the store buffer and may cause some pages to be marked
// scan-on-scavenge.
for (int i = 0; i < newspace_evacuation_candidates_.length(); i++) {
NewSpacePage* p =
reinterpret_cast<NewSpacePage*>(newspace_evacuation_candidates_[i]);
bool ok = VisitLiveObjects(p, &new_space_visitor, kClearMarkbits);
USE(ok);
DCHECK(ok);
}
heap_->IncrementPromotedObjectsSize(
static_cast<int>(new_space_visitor.promoted_size()));
heap_->IncrementSemiSpaceCopiedObjectSize(
static_cast<int>(new_space_visitor.semispace_copied_size()));
heap_->IncrementYoungSurvivorsCounter(
static_cast<int>(new_space_visitor.promoted_size()) +
static_cast<int>(new_space_visitor.semispace_copied_size()));
return local_pretenuring_feedback;
}
void MarkCompactCollector::AddEvacuationSlotsBufferSynchronized(
SlotsBuffer* evacuation_slots_buffer) {
base::LockGuard<base::Mutex> lock_guard(&evacuation_slots_buffers_mutex_);
evacuation_slots_buffers_.Add(evacuation_slots_buffer);
}
int MarkCompactCollector::NumberOfParallelCompactionTasks() {
if (!FLAG_parallel_compaction) return 1;
// Compute the number of needed tasks based on a target compaction time, the
// profiled compaction speed and marked live memory.
//
// The number of parallel compaction tasks is limited by:
// - #evacuation pages
// - (#cores - 1)
// - a hard limit
const double kTargetCompactionTimeInMs = 1;
const int kMaxCompactionTasks = 8;
intptr_t compaction_speed =
heap()->tracer()->CompactionSpeedInBytesPerMillisecond();
if (compaction_speed == 0) return 1;
intptr_t live_bytes = 0;
for (Page* page : evacuation_candidates_) {
live_bytes += page->LiveBytes();
}
const int cores = Max(1, base::SysInfo::NumberOfProcessors() - 1);
const int tasks =
1 + static_cast<int>(static_cast<double>(live_bytes) / compaction_speed /
kTargetCompactionTimeInMs);
const int tasks_capped_pages = Min(evacuation_candidates_.length(), tasks);
const int tasks_capped_cores = Min(cores, tasks_capped_pages);
const int tasks_capped_hard = Min(kMaxCompactionTasks, tasks_capped_cores);
return tasks_capped_hard;
}
void MarkCompactCollector::EvacuatePagesInParallel() {
const int num_pages = evacuation_candidates_.length();
if (num_pages == 0) return;
// Used for trace summary.
intptr_t live_bytes = 0;
intptr_t compaction_speed = 0;
if (FLAG_trace_fragmentation) {
for (Page* page : evacuation_candidates_) {
live_bytes += page->LiveBytes();
}
compaction_speed = heap()->tracer()->CompactionSpeedInBytesPerMillisecond();
}
const int num_tasks = NumberOfParallelCompactionTasks();
// Set up compaction spaces.
CompactionSpaceCollection** compaction_spaces_for_tasks =
new CompactionSpaceCollection*[num_tasks];
for (int i = 0; i < num_tasks; i++) {
compaction_spaces_for_tasks[i] = new CompactionSpaceCollection(heap());
}
heap()->old_space()->DivideUponCompactionSpaces(compaction_spaces_for_tasks,
num_tasks);
heap()->code_space()->DivideUponCompactionSpaces(compaction_spaces_for_tasks,
num_tasks);
uint32_t* task_ids = new uint32_t[num_tasks - 1];
// Kick off parallel tasks.
StartParallelCompaction(compaction_spaces_for_tasks, task_ids, num_tasks);
// Wait for unfinished and not-yet-started tasks.
WaitUntilCompactionCompleted(task_ids, num_tasks - 1);
delete[] task_ids;
double compaction_duration = 0.0;
intptr_t compacted_memory = 0;
// Merge back memory (compacted and unused) from compaction spaces.
for (int i = 0; i < num_tasks; i++) {
heap()->old_space()->MergeCompactionSpace(
compaction_spaces_for_tasks[i]->Get(OLD_SPACE));
heap()->code_space()->MergeCompactionSpace(
compaction_spaces_for_tasks[i]->Get(CODE_SPACE));
compacted_memory += compaction_spaces_for_tasks[i]->bytes_compacted();
compaction_duration += compaction_spaces_for_tasks[i]->duration();
delete compaction_spaces_for_tasks[i];
}
delete[] compaction_spaces_for_tasks;
heap()->tracer()->AddCompactionEvent(compaction_duration, compacted_memory);
// Finalize sequentially.
int abandoned_pages = 0;
for (int i = 0; i < num_pages; i++) {
Page* p = evacuation_candidates_[i];
switch (p->parallel_compaction_state().Value()) {
case MemoryChunk::ParallelCompactingState::kCompactingAborted:
// We have partially compacted the page, i.e., some objects may have
// moved, others are still in place.
// We need to:
// - Leave the evacuation candidate flag for later processing of
// slots buffer entries.
// - Leave the slots buffer there for processing of entries added by
// the write barrier.
// - Rescan the page as slot recording in the migration buffer only
// happens upon moving (which we potentially didn't do).
// - Leave the page in the list of pages of a space since we could not
// fully evacuate it.
// - Mark them for rescanning for store buffer entries as we otherwise
// might have stale store buffer entries that become "valid" again
// after reusing the memory. Note that all existing store buffer
// entries of such pages are filtered before rescanning.
DCHECK(p->IsEvacuationCandidate());
p->SetFlag(Page::COMPACTION_WAS_ABORTED);
p->set_scan_on_scavenge(true);
abandoned_pages++;
break;
case MemoryChunk::kCompactingFinalize:
DCHECK(p->IsEvacuationCandidate());
p->SetWasSwept();
p->Unlink();
break;
case MemoryChunk::kCompactingDone:
DCHECK(p->IsFlagSet(Page::POPULAR_PAGE));
DCHECK(p->IsFlagSet(Page::RESCAN_ON_EVACUATION));
break;
default:
// We should not observe kCompactingInProgress, or kCompactingDone.
UNREACHABLE();
}
p->parallel_compaction_state().SetValue(MemoryChunk::kCompactingDone);
}
if (FLAG_trace_fragmentation) {
PrintIsolate(isolate(),
"%8.0f ms: compaction: parallel=%d pages=%d aborted=%d "
"tasks=%d cores=%d live_bytes=%" V8_PTR_PREFIX
"d compaction_speed=%" V8_PTR_PREFIX "d\n",
isolate()->time_millis_since_init(), FLAG_parallel_compaction,
num_pages, abandoned_pages, num_tasks,
base::SysInfo::NumberOfProcessors(), live_bytes,
compaction_speed);
}
}
void MarkCompactCollector::StartParallelCompaction(
CompactionSpaceCollection** compaction_spaces, uint32_t* task_ids,
int len) {
compaction_in_progress_ = true;
for (int i = 1; i < len; i++) {
CompactionTask* task = new CompactionTask(heap(), compaction_spaces[i]);
task_ids[i - 1] = task->id();
V8::GetCurrentPlatform()->CallOnBackgroundThread(
task, v8::Platform::kShortRunningTask);
}
// Contribute in main thread.
EvacuatePages(compaction_spaces[0], &migration_slots_buffer_);
}
void MarkCompactCollector::WaitUntilCompactionCompleted(uint32_t* task_ids,
int len) {
// Try to cancel compaction tasks that have not been run (as they might be
// stuck in a worker queue). Tasks that cannot be canceled, have either
// already completed or are still running, hence we need to wait for their
// semaphore signal.
for (int i = 0; i < len; i++) {
if (!heap()->isolate()->cancelable_task_manager()->TryAbort(task_ids[i])) {
pending_compaction_tasks_semaphore_.Wait();
}
}
compaction_in_progress_ = false;
}
void MarkCompactCollector::EvacuatePages(
CompactionSpaceCollection* compaction_spaces,
SlotsBuffer** evacuation_slots_buffer) {
EvacuateOldSpaceVisitor visitor(heap(), compaction_spaces,
evacuation_slots_buffer);
for (int i = 0; i < evacuation_candidates_.length(); i++) {
Page* p = evacuation_candidates_[i];
DCHECK(p->IsEvacuationCandidate() ||
p->IsFlagSet(Page::RESCAN_ON_EVACUATION));
DCHECK(static_cast<int>(p->parallel_sweeping_state().Value()) ==
MemoryChunk::kSweepingDone);
if (p->parallel_compaction_state().TrySetValue(
MemoryChunk::kCompactingDone, MemoryChunk::kCompactingInProgress)) {
if (p->IsEvacuationCandidate()) {
DCHECK_EQ(p->parallel_compaction_state().Value(),
MemoryChunk::kCompactingInProgress);
double start = heap()->MonotonicallyIncreasingTimeInMs();
intptr_t live_bytes = p->LiveBytes();
AlwaysAllocateScope always_allocate(isolate());
if (VisitLiveObjects(p, &visitor, kClearMarkbits)) {
p->ResetLiveBytes();
p->parallel_compaction_state().SetValue(
MemoryChunk::kCompactingFinalize);
compaction_spaces->ReportCompactionProgress(
heap()->MonotonicallyIncreasingTimeInMs() - start, live_bytes);
} else {
p->parallel_compaction_state().SetValue(
MemoryChunk::kCompactingAborted);
}
} else {
// There could be popular pages in the list of evacuation candidates
// which we do compact.
p->parallel_compaction_state().SetValue(MemoryChunk::kCompactingDone);
}
}
}
}
class EvacuationWeakObjectRetainer : public WeakObjectRetainer {
public:
virtual Object* RetainAs(Object* object) {
if (object->IsHeapObject()) {
HeapObject* heap_object = HeapObject::cast(object);
MapWord map_word = heap_object->map_word();
if (map_word.IsForwardingAddress()) {
return map_word.ToForwardingAddress();
}
}
return object;
}
};
enum SweepingMode { SWEEP_ONLY, SWEEP_AND_VISIT_LIVE_OBJECTS };
enum SkipListRebuildingMode { REBUILD_SKIP_LIST, IGNORE_SKIP_LIST };
enum FreeSpaceTreatmentMode { IGNORE_FREE_SPACE, ZAP_FREE_SPACE };
template <MarkCompactCollector::SweepingParallelism mode>
static intptr_t Free(PagedSpace* space, FreeList* free_list, Address start,
int size) {
if (mode == MarkCompactCollector::SWEEP_ON_MAIN_THREAD) {
DCHECK(free_list == NULL);
return space->Free(start, size);
} else {
return size - free_list->Free(start, size);
}
}
// Sweeps a page. After sweeping the page can be iterated.
// Slots in live objects pointing into evacuation candidates are updated
// if requested.
// Returns the size of the biggest continuous freed memory chunk in bytes.
template <SweepingMode sweeping_mode,
MarkCompactCollector::SweepingParallelism parallelism,
SkipListRebuildingMode skip_list_mode,
FreeSpaceTreatmentMode free_space_mode>
static int Sweep(PagedSpace* space, FreeList* free_list, Page* p,
ObjectVisitor* v) {
DCHECK(!p->IsEvacuationCandidate() && !p->WasSwept());
DCHECK_EQ(skip_list_mode == REBUILD_SKIP_LIST,
space->identity() == CODE_SPACE);
DCHECK((p->skip_list() == NULL) || (skip_list_mode == REBUILD_SKIP_LIST));
DCHECK(parallelism == MarkCompactCollector::SWEEP_ON_MAIN_THREAD ||
sweeping_mode == SWEEP_ONLY);
Address free_start = p->area_start();
DCHECK(reinterpret_cast<intptr_t>(free_start) % (32 * kPointerSize) == 0);
// If we use the skip list for code space pages, we have to lock the skip
// list because it could be accessed concurrently by the runtime or the
// deoptimizer.
SkipList* skip_list = p->skip_list();
if ((skip_list_mode == REBUILD_SKIP_LIST) && skip_list) {
skip_list->Clear();
}
intptr_t freed_bytes = 0;
intptr_t max_freed_bytes = 0;
int curr_region = -1;
LiveObjectIterator<kBlackObjects> it(p);
HeapObject* object = NULL;
while ((object = it.Next()) != NULL) {
DCHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
Address free_end = object->address();
if (free_end != free_start) {
int size = static_cast<int>(free_end - free_start);
if (free_space_mode == ZAP_FREE_SPACE) {
memset(free_start, 0xcc, size);
}
freed_bytes = Free<parallelism>(space, free_list, free_start, size);
max_freed_bytes = Max(freed_bytes, max_freed_bytes);
}
Map* map = object->synchronized_map();
int size = object->SizeFromMap(map);
if (sweeping_mode == SWEEP_AND_VISIT_LIVE_OBJECTS) {
object->IterateBody(map->instance_type(), size, v);
}
if ((skip_list_mode == REBUILD_SKIP_LIST) && skip_list != NULL) {
int new_region_start = SkipList::RegionNumber(free_end);
int new_region_end =
SkipList::RegionNumber(free_end + size - kPointerSize);
if (new_region_start != curr_region || new_region_end != curr_region) {
skip_list->AddObject(free_end, size);
curr_region = new_region_end;
}
}
free_start = free_end + size;
}
// Clear the mark bits of that page and reset live bytes count.
Bitmap::Clear(p);
if (free_start != p->area_end()) {
int size = static_cast<int>(p->area_end() - free_start);
if (free_space_mode == ZAP_FREE_SPACE) {
memset(free_start, 0xcc, size);
}
freed_bytes = Free<parallelism>(space, free_list, free_start, size);
max_freed_bytes = Max(freed_bytes, max_freed_bytes);
}
if (parallelism == MarkCompactCollector::SWEEP_IN_PARALLEL) {
// When concurrent sweeping is active, the page will be marked after
// sweeping by the main thread.
p->parallel_sweeping_state().SetValue(MemoryChunk::kSweepingFinalize);
} else {
p->SetWasSwept();
}
return FreeList::GuaranteedAllocatable(static_cast<int>(max_freed_bytes));
}
void MarkCompactCollector::InvalidateCode(Code* code) {
if (heap_->incremental_marking()->IsCompacting() &&
!ShouldSkipEvacuationSlotRecording(code)) {
DCHECK(compacting_);
// If the object is white than no slots were recorded on it yet.
MarkBit mark_bit = Marking::MarkBitFrom(code);
if (Marking::IsWhite(mark_bit)) return;
// Ignore all slots that might have been recorded in the body of the
// deoptimized code object. Assumption: no slots will be recorded for
// this object after invalidating it.
RemoveObjectSlots(code->instruction_start(),
code->address() + code->Size());
}
}
// Return true if the given code is deoptimized or will be deoptimized.
bool MarkCompactCollector::WillBeDeoptimized(Code* code) {
return code->is_optimized_code() && code->marked_for_deoptimization();
}
void MarkCompactCollector::RemoveObjectSlots(Address start_slot,
Address end_slot) {
// Remove entries by replacing them with an old-space slot containing a smi
// that is located in an unmovable page.
int npages = evacuation_candidates_.length();
for (int i = 0; i < npages; i++) {
Page* p = evacuation_candidates_[i];
DCHECK(p->IsEvacuationCandidate() ||
p->IsFlagSet(Page::RESCAN_ON_EVACUATION));
if (p->IsEvacuationCandidate()) {
SlotsBuffer::RemoveObjectSlots(heap_, p->slots_buffer(), start_slot,
end_slot);
}
}
}
#ifdef VERIFY_HEAP
static void VerifyAllBlackObjects(MemoryChunk* page) {
LiveObjectIterator<kAllLiveObjects> it(page);
HeapObject* object = NULL;
while ((object = it.Next()) != NULL) {
CHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
}
}
#endif // VERIFY_HEAP
bool MarkCompactCollector::VisitLiveObjects(MemoryChunk* page,
HeapObjectVisitor* visitor,
IterationMode mode) {
#ifdef VERIFY_HEAP
VerifyAllBlackObjects(page);
#endif // VERIFY_HEAP
LiveObjectIterator<kBlackObjects> it(page);
HeapObject* object = nullptr;
while ((object = it.Next()) != nullptr) {
DCHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
if (!visitor->Visit(object)) {
if (mode == kClearMarkbits) {
page->markbits()->ClearRange(
page->AddressToMarkbitIndex(page->area_start()),
page->AddressToMarkbitIndex(object->address()));
RecomputeLiveBytes(page);
}
return false;
}
}
if (mode == kClearMarkbits) {
Bitmap::Clear(page);
}
return true;
}
void MarkCompactCollector::RecomputeLiveBytes(MemoryChunk* page) {
LiveObjectIterator<kBlackObjects> it(page);
int new_live_size = 0;
HeapObject* object = nullptr;
while ((object = it.Next()) != nullptr) {
new_live_size += object->Size();
}
page->SetLiveBytes(new_live_size);
}
void MarkCompactCollector::VisitLiveObjectsBody(Page* page,
ObjectVisitor* visitor) {
#ifdef VERIFY_HEAP
VerifyAllBlackObjects(page);
#endif // VERIFY_HEAP
LiveObjectIterator<kBlackObjects> it(page);
HeapObject* object = NULL;
while ((object = it.Next()) != NULL) {
DCHECK(Marking::IsBlack(Marking::MarkBitFrom(object)));
Map* map = object->synchronized_map();
int size = object->SizeFromMap(map);
object->IterateBody(map->instance_type(), size, visitor);
}
}
void MarkCompactCollector::SweepAbortedPages() {
// Second pass on aborted pages.
for (int i = 0; i < evacuation_candidates_.length(); i++) {
Page* p = evacuation_candidates_[i];
if (p->IsFlagSet(Page::COMPACTION_WAS_ABORTED)) {
p->ClearFlag(MemoryChunk::COMPACTION_WAS_ABORTED);
PagedSpace* space = static_cast<PagedSpace*>(p->owner());
switch (space->identity()) {
case OLD_SPACE:
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, IGNORE_SKIP_LIST,
IGNORE_FREE_SPACE>(space, nullptr, p, nullptr);
break;
case CODE_SPACE:
if (FLAG_zap_code_space) {
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, REBUILD_SKIP_LIST,
ZAP_FREE_SPACE>(space, NULL, p, nullptr);
} else {
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, REBUILD_SKIP_LIST,
IGNORE_FREE_SPACE>(space, NULL, p, nullptr);
}
break;
default:
UNREACHABLE();
break;
}
}
}
}
void MarkCompactCollector::EvacuateNewSpaceAndCandidates() {
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_EVACUATE);
Heap::RelocationLock relocation_lock(heap());
HashMap* local_pretenuring_feedback = nullptr;
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_NEW_SPACE);
EvacuationScope evacuation_scope(this);
EvacuateNewSpacePrologue();
local_pretenuring_feedback = EvacuateNewSpaceInParallel();
heap_->new_space()->set_age_mark(heap_->new_space()->top());
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_CANDIDATES);
EvacuationScope evacuation_scope(this);
EvacuatePagesInParallel();
}
{
heap_->MergeAllocationSitePretenuringFeedback(*local_pretenuring_feedback);
delete local_pretenuring_feedback;
}
UpdatePointersAfterEvacuation();
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_CLEAN_UP);
// After updating all pointers, we can finally sweep the aborted pages,
// effectively overriding any forward pointers.
SweepAbortedPages();
// EvacuateNewSpaceAndCandidates iterates over new space objects and for
// ArrayBuffers either re-registers them as live or promotes them. This is
// needed to properly free them.
heap()->array_buffer_tracker()->FreeDead(false);
// Deallocate evacuated candidate pages.
ReleaseEvacuationCandidates();
}
#ifdef VERIFY_HEAP
if (FLAG_verify_heap && !sweeping_in_progress_) {
VerifyEvacuation(heap());
}
#endif
}
void MarkCompactCollector::UpdatePointersAfterEvacuation() {
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_UPDATE_POINTERS);
{
GCTracer::Scope gc_scope(
heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_UPDATE_POINTERS_TO_EVACUATED);
UpdateSlotsRecordedIn(migration_slots_buffer_);
if (FLAG_trace_fragmentation_verbose) {
PrintF(" migration slots buffer: %d\n",
SlotsBuffer::SizeOfChain(migration_slots_buffer_));
}
slots_buffer_allocator_->DeallocateChain(&migration_slots_buffer_);
DCHECK(migration_slots_buffer_ == NULL);
// TODO(hpayer): Process the slots buffers in parallel. This has to be done
// after evacuation of all pages finishes.
int buffers = evacuation_slots_buffers_.length();
for (int i = 0; i < buffers; i++) {
SlotsBuffer* buffer = evacuation_slots_buffers_[i];
UpdateSlotsRecordedIn(buffer);
slots_buffer_allocator_->DeallocateChain(&buffer);
}
evacuation_slots_buffers_.Rewind(0);
}
// Second pass: find pointers to new space and update them.
PointersUpdatingVisitor updating_visitor(heap());
{
GCTracer::Scope gc_scope(
heap()->tracer(), GCTracer::Scope::MC_EVACUATE_UPDATE_POINTERS_TO_NEW);
// Update pointers in to space.
SemiSpaceIterator to_it(heap()->new_space());
for (HeapObject* object = to_it.Next(); object != NULL;
object = to_it.Next()) {
Map* map = object->map();
object->IterateBody(map->instance_type(), object->SizeFromMap(map),
&updating_visitor);
}
// Update roots.
heap_->IterateRoots(&updating_visitor, VISIT_ALL_IN_SWEEP_NEWSPACE);
StoreBufferRebuildScope scope(heap_, heap_->store_buffer(),
&Heap::ScavengeStoreBufferCallback);
heap_->store_buffer()->IteratePointersToNewSpace(&UpdatePointer);
}
int npages = evacuation_candidates_.length();
{
GCTracer::Scope gc_scope(
heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_UPDATE_POINTERS_BETWEEN_EVACUATED);
for (int i = 0; i < npages; i++) {
Page* p = evacuation_candidates_[i];
DCHECK(p->IsEvacuationCandidate() ||
p->IsFlagSet(Page::RESCAN_ON_EVACUATION));
if (p->IsEvacuationCandidate()) {
UpdateSlotsRecordedIn(p->slots_buffer());
if (FLAG_trace_fragmentation_verbose) {
PrintF(" page %p slots buffer: %d\n", reinterpret_cast<void*>(p),
SlotsBuffer::SizeOfChain(p->slots_buffer()));
}
slots_buffer_allocator_->DeallocateChain(p->slots_buffer_address());
// Important: skip list should be cleared only after roots were updated
// because root iteration traverses the stack and might have to find
// code objects from non-updated pc pointing into evacuation candidate.
SkipList* list = p->skip_list();
if (list != NULL) list->Clear();
// First pass on aborted pages, fixing up all live objects.
if (p->IsFlagSet(Page::COMPACTION_WAS_ABORTED)) {
p->ClearEvacuationCandidate();
VisitLiveObjectsBody(p, &updating_visitor);
}
}
if (p->IsFlagSet(Page::RESCAN_ON_EVACUATION)) {
if (FLAG_gc_verbose) {
PrintF("Sweeping 0x%" V8PRIxPTR " during evacuation.\n",
reinterpret_cast<intptr_t>(p));
}
PagedSpace* space = static_cast<PagedSpace*>(p->owner());
p->ClearFlag(MemoryChunk::RESCAN_ON_EVACUATION);
switch (space->identity()) {
case OLD_SPACE:
Sweep<SWEEP_AND_VISIT_LIVE_OBJECTS, SWEEP_ON_MAIN_THREAD,
IGNORE_SKIP_LIST, IGNORE_FREE_SPACE>(space, NULL, p,
&updating_visitor);
break;
case CODE_SPACE:
if (FLAG_zap_code_space) {
Sweep<SWEEP_AND_VISIT_LIVE_OBJECTS, SWEEP_ON_MAIN_THREAD,
REBUILD_SKIP_LIST, ZAP_FREE_SPACE>(space, NULL, p,
&updating_visitor);
} else {
Sweep<SWEEP_AND_VISIT_LIVE_OBJECTS, SWEEP_ON_MAIN_THREAD,
REBUILD_SKIP_LIST, IGNORE_FREE_SPACE>(space, NULL, p,
&updating_visitor);
}
break;
default:
UNREACHABLE();
break;
}
}
}
}
{
GCTracer::Scope gc_scope(heap()->tracer(),
GCTracer::Scope::MC_EVACUATE_UPDATE_POINTERS_WEAK);
heap_->string_table()->Iterate(&updating_visitor);
// Update pointers from external string table.
heap_->UpdateReferencesInExternalStringTable(
&UpdateReferenceInExternalStringTableEntry);
EvacuationWeakObjectRetainer evacuation_object_retainer;
heap()->ProcessAllWeakReferences(&evacuation_object_retainer);
}
}
void MarkCompactCollector::MoveEvacuationCandidatesToEndOfPagesList() {
int npages = evacuation_candidates_.length();
for (int i = 0; i < npages; i++) {
Page* p = evacuation_candidates_[i];
if (!p->IsEvacuationCandidate()) continue;
p->Unlink();
PagedSpace* space = static_cast<PagedSpace*>(p->owner());
p->InsertAfter(space->LastPage());
}
}
void MarkCompactCollector::ReleaseEvacuationCandidates() {
int npages = evacuation_candidates_.length();
for (int i = 0; i < npages; i++) {
Page* p = evacuation_candidates_[i];
if (!p->IsEvacuationCandidate()) continue;
PagedSpace* space = static_cast<PagedSpace*>(p->owner());
space->Free(p->area_start(), p->area_size());
p->set_scan_on_scavenge(false);
p->ResetLiveBytes();
CHECK(p->WasSwept());
space->ReleasePage(p);
}
evacuation_candidates_.Rewind(0);
compacting_ = false;
heap()->FilterStoreBufferEntriesOnAboutToBeFreedPages();
heap()->FreeQueuedChunks();
}
int MarkCompactCollector::SweepInParallel(PagedSpace* space,
int required_freed_bytes) {
int max_freed = 0;
int max_freed_overall = 0;
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
max_freed = SweepInParallel(p, space);
DCHECK(max_freed >= 0);
if (required_freed_bytes > 0 && max_freed >= required_freed_bytes) {
return max_freed;
}
max_freed_overall = Max(max_freed, max_freed_overall);
if (p == space->end_of_unswept_pages()) break;
}
return max_freed_overall;
}
int MarkCompactCollector::SweepInParallel(Page* page, PagedSpace* space) {
int max_freed = 0;
if (page->TryLock()) {
// If this page was already swept in the meantime, we can return here.
if (page->parallel_sweeping_state().Value() !=
MemoryChunk::kSweepingPending) {
page->mutex()->Unlock();
return 0;
}
page->parallel_sweeping_state().SetValue(MemoryChunk::kSweepingInProgress);
FreeList* free_list;
FreeList private_free_list(space);
if (space->identity() == OLD_SPACE) {
free_list = free_list_old_space_.get();
max_freed =
Sweep<SWEEP_ONLY, SWEEP_IN_PARALLEL, IGNORE_SKIP_LIST,
IGNORE_FREE_SPACE>(space, &private_free_list, page, NULL);
} else if (space->identity() == CODE_SPACE) {
free_list = free_list_code_space_.get();
max_freed =
Sweep<SWEEP_ONLY, SWEEP_IN_PARALLEL, REBUILD_SKIP_LIST,
IGNORE_FREE_SPACE>(space, &private_free_list, page, NULL);
} else {
free_list = free_list_map_space_.get();
max_freed =
Sweep<SWEEP_ONLY, SWEEP_IN_PARALLEL, IGNORE_SKIP_LIST,
IGNORE_FREE_SPACE>(space, &private_free_list, page, NULL);
}
free_list->Concatenate(&private_free_list);
page->mutex()->Unlock();
}
return max_freed;
}
void MarkCompactCollector::StartSweepSpace(PagedSpace* space) {
space->ClearStats();
// We defensively initialize end_of_unswept_pages_ here with the first page
// of the pages list.
space->set_end_of_unswept_pages(space->FirstPage());
PageIterator it(space);
int pages_swept = 0;
bool unused_page_present = false;
bool parallel_sweeping_active = false;
while (it.has_next()) {
Page* p = it.next();
DCHECK(p->parallel_sweeping_state().Value() == MemoryChunk::kSweepingDone);
// Clear sweeping flags indicating that marking bits are still intact.
p->ClearWasSwept();
if (p->IsFlagSet(Page::RESCAN_ON_EVACUATION) ||
p->IsEvacuationCandidate()) {
// Will be processed in EvacuateNewSpaceAndCandidates.
DCHECK(evacuation_candidates_.length() > 0);
continue;
}
if (p->IsFlagSet(Page::NEVER_ALLOCATE_ON_PAGE)) {
// We need to sweep the page to get it into an iterable state again. Note
// that this adds unusable memory into the free list that is later on
// (in the free list) dropped again. Since we only use the flag for
// testing this is fine.
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, IGNORE_SKIP_LIST,
IGNORE_FREE_SPACE>(space, nullptr, p, nullptr);
continue;
}
// One unused page is kept, all further are released before sweeping them.
if (p->LiveBytes() == 0) {
if (unused_page_present) {
if (FLAG_gc_verbose) {
PrintIsolate(isolate(), "sweeping: released page: %p", p);
}
space->ReleasePage(p);
continue;
}
unused_page_present = true;
}
if (!parallel_sweeping_active) {
if (FLAG_gc_verbose) {
PrintIsolate(isolate(), "sweeping: %p", p);
}
if (space->identity() == CODE_SPACE) {
if (FLAG_zap_code_space) {
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, REBUILD_SKIP_LIST,
ZAP_FREE_SPACE>(space, NULL, p, NULL);
} else {
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, REBUILD_SKIP_LIST,
IGNORE_FREE_SPACE>(space, NULL, p, NULL);
}
} else {
Sweep<SWEEP_ONLY, SWEEP_ON_MAIN_THREAD, IGNORE_SKIP_LIST,
IGNORE_FREE_SPACE>(space, NULL, p, NULL);
}
pages_swept++;
parallel_sweeping_active = true;
} else {
if (FLAG_gc_verbose) {
PrintIsolate(isolate(), "sweeping: initialized for parallel: %p", p);
}
p->parallel_sweeping_state().SetValue(MemoryChunk::kSweepingPending);
int to_sweep = p->area_size() - p->LiveBytes();
space->accounting_stats_.ShrinkSpace(to_sweep);
}
space->set_end_of_unswept_pages(p);
}
if (FLAG_gc_verbose) {
PrintIsolate(isolate(), "sweeping: space=%s pages_swept=%d",
AllocationSpaceName(space->identity()), pages_swept);
}
}
void MarkCompactCollector::SweepSpaces() {
GCTracer::Scope gc_scope(heap()->tracer(), GCTracer::Scope::MC_SWEEP);
double start_time = 0.0;
if (FLAG_print_cumulative_gc_stat) {
start_time = heap_->MonotonicallyIncreasingTimeInMs();
}
#ifdef DEBUG
state_ = SWEEP_SPACES;
#endif
MoveEvacuationCandidatesToEndOfPagesList();
{
sweeping_in_progress_ = true;
{
GCTracer::Scope sweep_scope(heap()->tracer(),
GCTracer::Scope::MC_SWEEP_OLD);
StartSweepSpace(heap()->old_space());
}
{
GCTracer::Scope sweep_scope(heap()->tracer(),
GCTracer::Scope::MC_SWEEP_CODE);
StartSweepSpace(heap()->code_space());
}
{
GCTracer::Scope sweep_scope(heap()->tracer(),
GCTracer::Scope::MC_SWEEP_MAP);
StartSweepSpace(heap()->map_space());
}
if (FLAG_concurrent_sweeping) {
StartSweeperThreads();
}
}
// Deallocate unmarked large objects.
heap_->lo_space()->FreeUnmarkedObjects();
// Give pages that are queued to be freed back to the OS. Invalid store
// buffer entries are already filter out. We can just release the memory.
heap()->FreeQueuedChunks();
if (FLAG_print_cumulative_gc_stat) {
heap_->tracer()->AddSweepingTime(heap_->MonotonicallyIncreasingTimeInMs() -
start_time);
}
}
void MarkCompactCollector::ParallelSweepSpaceComplete(PagedSpace* space) {
PageIterator it(space);
while (it.has_next()) {
Page* p = it.next();
if (p->parallel_sweeping_state().Value() ==
MemoryChunk::kSweepingFinalize) {
p->parallel_sweeping_state().SetValue(MemoryChunk::kSweepingDone);
p->SetWasSwept();
}
DCHECK(p->parallel_sweeping_state().Value() == MemoryChunk::kSweepingDone);
}
}
void MarkCompactCollector::ParallelSweepSpacesComplete() {
ParallelSweepSpaceComplete(heap()->old_space());
ParallelSweepSpaceComplete(heap()->code_space());
ParallelSweepSpaceComplete(heap()->map_space());
}
// TODO(1466) ReportDeleteIfNeeded is not called currently.
// Our profiling tools do not expect intersections between
// code objects. We should either reenable it or change our tools.
void MarkCompactCollector::ReportDeleteIfNeeded(HeapObject* obj,
Isolate* isolate) {
if (obj->IsCode()) {
PROFILE(isolate, CodeDeleteEvent(obj->address()));
}
}
Isolate* MarkCompactCollector::isolate() const { return heap_->isolate(); }
void MarkCompactCollector::Initialize() {
MarkCompactMarkingVisitor::Initialize();
IncrementalMarking::Initialize();
}
void MarkCompactCollector::EvictPopularEvacuationCandidate(Page* page) {
if (FLAG_trace_fragmentation) {
PrintF("Page %p is too popular. Disabling evacuation.\n",
reinterpret_cast<void*>(page));
}
isolate()->CountUsage(v8::Isolate::UseCounterFeature::kSlotsBufferOverflow);
// TODO(gc) If all evacuation candidates are too popular we
// should stop slots recording entirely.
page->ClearEvacuationCandidate();
DCHECK(!page->IsFlagSet(Page::POPULAR_PAGE));
page->SetFlag(Page::POPULAR_PAGE);
// We were not collecting slots on this page that point
// to other evacuation candidates thus we have to
// rescan the page after evacuation to discover and update all
// pointers to evacuated objects.
page->SetFlag(Page::RESCAN_ON_EVACUATION);
}
void MarkCompactCollector::RecordCodeEntrySlot(HeapObject* object, Address slot,
Code* target) {
Page* target_page = Page::FromAddress(reinterpret_cast<Address>(target));
if (target_page->IsEvacuationCandidate() &&
!ShouldSkipEvacuationSlotRecording(object)) {
if (!SlotsBuffer::AddTo(slots_buffer_allocator_,
target_page->slots_buffer_address(),
SlotsBuffer::CODE_ENTRY_SLOT, slot,
SlotsBuffer::FAIL_ON_OVERFLOW)) {
EvictPopularEvacuationCandidate(target_page);
}
}
}
void MarkCompactCollector::RecordCodeTargetPatch(Address pc, Code* target) {
DCHECK(heap()->gc_state() == Heap::MARK_COMPACT);
if (is_compacting()) {
Code* host =
isolate()->inner_pointer_to_code_cache()->GcSafeFindCodeForInnerPointer(
pc);
MarkBit mark_bit = Marking::MarkBitFrom(host);
if (Marking::IsBlack(mark_bit)) {
RelocInfo rinfo(isolate(), pc, RelocInfo::CODE_TARGET, 0, host);
RecordRelocSlot(&rinfo, target);
}
}
}
} // namespace internal
} // namespace v8