v8/src/spaces.h

2836 lines
89 KiB
C
Raw Normal View History

// Copyright 2011 the V8 project authors. All rights reserved.
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
// * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#ifndef V8_SPACES_H_
#define V8_SPACES_H_
#include "allocation.h"
#include "hashmap.h"
#include "list.h"
#include "log.h"
namespace v8 {
namespace internal {
class Isolate;
// -----------------------------------------------------------------------------
// Heap structures:
//
// A JS heap consists of a young generation, an old generation, and a large
// object space. The young generation is divided into two semispaces. A
// scavenger implements Cheney's copying algorithm. The old generation is
// separated into a map space and an old object space. The map space contains
// all (and only) map objects, the rest of old objects go into the old space.
// The old generation is collected by a mark-sweep-compact collector.
//
// The semispaces of the young generation are contiguous. The old and map
// spaces consists of a list of pages. A page has a page header and an object
// area.
//
// There is a separate large object space for objects larger than
// Page::kMaxHeapObjectSize, so that they do not have to move during
// collection. The large object space is paged. Pages in large object space
// may be larger than the page size.
//
// A store-buffer based write barrier is used to keep track of intergenerational
// references. See store-buffer.h.
//
// During scavenges and mark-sweep collections we sometimes (after a store
// buffer overflow) iterate intergenerational pointers without decoding heap
// object maps so if the page belongs to old pointer space or large object
// space it is essential to guarantee that the page does not contain any
// garbage pointers to new space: every pointer aligned word which satisfies
// the Heap::InNewSpace() predicate must be a pointer to a live heap object in
// new space. Thus objects in old pointer and large object spaces should have a
// special layout (e.g. no bare integer fields). This requirement does not
// apply to map space which is iterated in a special fashion. However we still
// require pointer fields of dead maps to be cleaned.
//
// To enable lazy cleaning of old space pages we can mark chunks of the page
// as being garbage. Garbage sections are marked with a special map. These
// sections are skipped when scanning the page, even if we are otherwise
// scanning without regard for object boundaries. Garbage sections are chained
// together to form a free list after a GC. Garbage sections created outside
// of GCs by object trunctation etc. may not be in the free list chain. Very
// small free spaces are ignored, they need only be cleaned of bogus pointers
// into new space.
//
// Each page may have up to one special garbage section. The start of this
// section is denoted by the top field in the space. The end of the section
// is denoted by the limit field in the space. This special garbage section
// is not marked with a free space map in the data. The point of this section
// is to enable linear allocation without having to constantly update the byte
// array every time the top field is updated and a new object is created. The
// special garbage section is not in the chain of garbage sections.
//
// Since the top and limit fields are in the space, not the page, only one page
// has a special garbage section, and if the top and limit are equal then there
// is no special garbage section.
// Some assertion macros used in the debugging mode.
#define ASSERT_PAGE_ALIGNED(address) \
ASSERT((OffsetFrom(address) & Page::kPageAlignmentMask) == 0)
#define ASSERT_OBJECT_ALIGNED(address) \
ASSERT((OffsetFrom(address) & kObjectAlignmentMask) == 0)
#define ASSERT_OBJECT_SIZE(size) \
ASSERT((0 < size) && (size <= Page::kMaxNonCodeHeapObjectSize))
#define ASSERT_PAGE_OFFSET(offset) \
ASSERT((Page::kObjectStartOffset <= offset) \
&& (offset <= Page::kPageSize))
#define ASSERT_MAP_PAGE_INDEX(index) \
ASSERT((0 <= index) && (index <= MapSpace::kMaxMapPageIndex))
class PagedSpace;
class MemoryAllocator;
class AllocationInfo;
class Space;
class FreeList;
class MemoryChunk;
class MarkBit {
public:
typedef uint32_t CellType;
inline MarkBit(CellType* cell, CellType mask, bool data_only)
: cell_(cell), mask_(mask), data_only_(data_only) { }
inline CellType* cell() { return cell_; }
inline CellType mask() { return mask_; }
#ifdef DEBUG
bool operator==(const MarkBit& other) {
return cell_ == other.cell_ && mask_ == other.mask_;
}
#endif
inline void Set() { *cell_ |= mask_; }
inline bool Get() { return (*cell_ & mask_) != 0; }
inline void Clear() { *cell_ &= ~mask_; }
inline bool data_only() { return data_only_; }
inline MarkBit Next() {
CellType new_mask = mask_ << 1;
if (new_mask == 0) {
return MarkBit(cell_ + 1, 1, data_only_);
} else {
return MarkBit(cell_, new_mask, data_only_);
}
}
private:
CellType* cell_;
CellType mask_;
// This boolean indicates that the object is in a data-only space with no
// pointers. This enables some optimizations when marking.
// It is expected that this field is inlined and turned into control flow
// at the place where the MarkBit object is created.
bool data_only_;
};
// Bitmap is a sequence of cells each containing fixed number of bits.
class Bitmap {
public:
static const uint32_t kBitsPerCell = 32;
static const uint32_t kBitsPerCellLog2 = 5;
static const uint32_t kBitIndexMask = kBitsPerCell - 1;
static const uint32_t kBytesPerCell = kBitsPerCell / kBitsPerByte;
static const uint32_t kBytesPerCellLog2 = kBitsPerCellLog2 - kBitsPerByteLog2;
static const size_t kLength =
(1 << kPageSizeBits) >> (kPointerSizeLog2);
static const size_t kSize =
(1 << kPageSizeBits) >> (kPointerSizeLog2 + kBitsPerByteLog2);
static int CellsForLength(int length) {
return (length + kBitsPerCell - 1) >> kBitsPerCellLog2;
}
int CellsCount() {
return CellsForLength(kLength);
}
static int SizeFor(int cells_count) {
return sizeof(MarkBit::CellType) * cells_count;
}
INLINE(static uint32_t IndexToCell(uint32_t index)) {
return index >> kBitsPerCellLog2;
}
INLINE(static uint32_t CellToIndex(uint32_t index)) {
return index << kBitsPerCellLog2;
}
INLINE(static uint32_t CellAlignIndex(uint32_t index)) {
return (index + kBitIndexMask) & ~kBitIndexMask;
}
INLINE(MarkBit::CellType* cells()) {
return reinterpret_cast<MarkBit::CellType*>(this);
}
INLINE(Address address()) {
return reinterpret_cast<Address>(this);
}
INLINE(static Bitmap* FromAddress(Address addr)) {
return reinterpret_cast<Bitmap*>(addr);
}
inline MarkBit MarkBitFromIndex(uint32_t index, bool data_only = false) {
MarkBit::CellType mask = 1 << (index & kBitIndexMask);
MarkBit::CellType* cell = this->cells() + (index >> kBitsPerCellLog2);
return MarkBit(cell, mask, data_only);
}
static inline void Clear(MemoryChunk* chunk);
static void PrintWord(uint32_t word, uint32_t himask = 0) {
for (uint32_t mask = 1; mask != 0; mask <<= 1) {
if ((mask & himask) != 0) PrintF("[");
PrintF((mask & word) ? "1" : "0");
if ((mask & himask) != 0) PrintF("]");
}
}
class CellPrinter {
public:
CellPrinter() : seq_start(0), seq_type(0), seq_length(0) { }
void Print(uint32_t pos, uint32_t cell) {
if (cell == seq_type) {
seq_length++;
return;
}
Flush();
if (IsSeq(cell)) {
seq_start = pos;
seq_length = 0;
seq_type = cell;
return;
}
PrintF("%d: ", pos);
PrintWord(cell);
PrintF("\n");
}
void Flush() {
if (seq_length > 0) {
PrintF("%d: %dx%d\n",
seq_start,
seq_type == 0 ? 0 : 1,
seq_length * kBitsPerCell);
seq_length = 0;
}
}
static bool IsSeq(uint32_t cell) { return cell == 0 || cell == 0xFFFFFFFF; }
private:
uint32_t seq_start;
uint32_t seq_type;
uint32_t seq_length;
};
void Print() {
CellPrinter printer;
for (int i = 0; i < CellsCount(); i++) {
printer.Print(i, cells()[i]);
}
printer.Flush();
PrintF("\n");
}
bool IsClean() {
for (int i = 0; i < CellsCount(); i++) {
if (cells()[i] != 0) {
return false;
}
}
return true;
}
};
class SkipList;
class SlotsBuffer;
// MemoryChunk represents a memory region owned by a specific space.
// It is divided into the header and the body. Chunk start is always
// 1MB aligned. Start of the body is aligned so it can accommodate
// any heap object.
class MemoryChunk {
public:
// Only works if the pointer is in the first kPageSize of the MemoryChunk.
static MemoryChunk* FromAddress(Address a) {
return reinterpret_cast<MemoryChunk*>(OffsetFrom(a) & ~kAlignmentMask);
}
// Only works for addresses in pointer spaces, not data or code spaces.
static inline MemoryChunk* FromAnyPointerAddress(Address addr);
Address address() { return reinterpret_cast<Address>(this); }
bool is_valid() { return address() != NULL; }
MemoryChunk* next_chunk() const { return next_chunk_; }
MemoryChunk* prev_chunk() const { return prev_chunk_; }
void set_next_chunk(MemoryChunk* next) { next_chunk_ = next; }
void set_prev_chunk(MemoryChunk* prev) { prev_chunk_ = prev; }
Space* owner() const {
if ((reinterpret_cast<intptr_t>(owner_) & kFailureTagMask) ==
kFailureTag) {
return reinterpret_cast<Space*>(reinterpret_cast<intptr_t>(owner_) -
kFailureTag);
} else {
return NULL;
}
}
void set_owner(Space* space) {
ASSERT((reinterpret_cast<intptr_t>(space) & kFailureTagMask) == 0);
owner_ = reinterpret_cast<Address>(space) + kFailureTag;
ASSERT((reinterpret_cast<intptr_t>(owner_) & kFailureTagMask) ==
kFailureTag);
}
VirtualMemory* reserved_memory() {
return &reservation_;
}
void InitializeReservedMemory() {
reservation_.Reset();
}
void set_reserved_memory(VirtualMemory* reservation) {
ASSERT_NOT_NULL(reservation);
reservation_.TakeControl(reservation);
}
bool scan_on_scavenge() { return IsFlagSet(SCAN_ON_SCAVENGE); }
void initialize_scan_on_scavenge(bool scan) {
if (scan) {
SetFlag(SCAN_ON_SCAVENGE);
} else {
ClearFlag(SCAN_ON_SCAVENGE);
}
}
inline void set_scan_on_scavenge(bool scan);
int store_buffer_counter() { return store_buffer_counter_; }
void set_store_buffer_counter(int counter) {
store_buffer_counter_ = counter;
}
bool Contains(Address addr) {
return addr >= area_start() && addr < area_end();
}
// Checks whether addr can be a limit of addresses in this page.
// It's a limit if it's in the page, or if it's just after the
// last byte of the page.
bool ContainsLimit(Address addr) {
return addr >= area_start() && addr <= area_end();
}
// Every n write barrier invocations we go to runtime even though
// we could have handled it in generated code. This lets us check
// whether we have hit the limit and should do some more marking.
static const int kWriteBarrierCounterGranularity = 500;
enum MemoryChunkFlags {
IS_EXECUTABLE,
ABOUT_TO_BE_FREED,
POINTERS_TO_HERE_ARE_INTERESTING,
POINTERS_FROM_HERE_ARE_INTERESTING,
SCAN_ON_SCAVENGE,
IN_FROM_SPACE, // Mutually exclusive with IN_TO_SPACE.
IN_TO_SPACE, // All pages in new space has one of these two set.
NEW_SPACE_BELOW_AGE_MARK,
CONTAINS_ONLY_DATA,
EVACUATION_CANDIDATE,
RESCAN_ON_EVACUATION,
// Pages swept precisely can be iterated, hitting only the live objects.
// Whereas those swept conservatively cannot be iterated over. Both flags
// indicate that marking bits have been cleared by the sweeper, otherwise
// marking bits are still intact.
WAS_SWEPT_PRECISELY,
WAS_SWEPT_CONSERVATIVELY,
// Large objects can have a progress bar in their page header. These object
// are scanned in increments and will be kept black while being scanned.
// Even if the mutator writes to them they will be kept black and a white
// to grey transition is performed in the value.
HAS_PROGRESS_BAR,
// Last flag, keep at bottom.
NUM_MEMORY_CHUNK_FLAGS
};
static const int kPointersToHereAreInterestingMask =
1 << POINTERS_TO_HERE_ARE_INTERESTING;
static const int kPointersFromHereAreInterestingMask =
1 << POINTERS_FROM_HERE_ARE_INTERESTING;
static const int kEvacuationCandidateMask =
1 << EVACUATION_CANDIDATE;
static const int kSkipEvacuationSlotsRecordingMask =
(1 << EVACUATION_CANDIDATE) |
(1 << RESCAN_ON_EVACUATION) |
(1 << IN_FROM_SPACE) |
(1 << IN_TO_SPACE);
void SetFlag(int flag) {
flags_ |= static_cast<uintptr_t>(1) << flag;
}
void ClearFlag(int flag) {
flags_ &= ~(static_cast<uintptr_t>(1) << flag);
}
void SetFlagTo(int flag, bool value) {
if (value) {
SetFlag(flag);
} else {
ClearFlag(flag);
}
}
bool IsFlagSet(int flag) {
return (flags_ & (static_cast<uintptr_t>(1) << flag)) != 0;
}
// Set or clear multiple flags at a time. The flags in the mask
// are set to the value in "flags", the rest retain the current value
// in flags_.
void SetFlags(intptr_t flags, intptr_t mask) {
flags_ = (flags_ & ~mask) | (flags & mask);
}
// Return all current flags.
intptr_t GetFlags() { return flags_; }
intptr_t parallel_sweeping() const {
return parallel_sweeping_;
}
void set_parallel_sweeping(intptr_t state) {
parallel_sweeping_ = state;
}
bool TryParallelSweeping() {
return NoBarrier_CompareAndSwap(&parallel_sweeping_, 1, 0) == 1;
}
// Manage live byte count (count of bytes known to be live,
// because they are marked black).
void ResetLiveBytes() {
if (FLAG_gc_verbose) {
PrintF("ResetLiveBytes:%p:%x->0\n",
static_cast<void*>(this), live_byte_count_);
}
live_byte_count_ = 0;
}
void IncrementLiveBytes(int by) {
if (FLAG_gc_verbose) {
printf("UpdateLiveBytes:%p:%x%c=%x->%x\n",
static_cast<void*>(this), live_byte_count_,
((by < 0) ? '-' : '+'), ((by < 0) ? -by : by),
live_byte_count_ + by);
}
live_byte_count_ += by;
ASSERT_LE(static_cast<unsigned>(live_byte_count_), size_);
}
int LiveBytes() {
ASSERT(static_cast<unsigned>(live_byte_count_) <= size_);
return live_byte_count_;
}
int write_barrier_counter() {
return static_cast<int>(write_barrier_counter_);
}
void set_write_barrier_counter(int counter) {
write_barrier_counter_ = counter;
}
int progress_bar() {
ASSERT(IsFlagSet(HAS_PROGRESS_BAR));
return progress_bar_;
}
void set_progress_bar(int progress_bar) {
ASSERT(IsFlagSet(HAS_PROGRESS_BAR));
progress_bar_ = progress_bar;
}
void ResetProgressBar() {
if (IsFlagSet(MemoryChunk::HAS_PROGRESS_BAR)) {
set_progress_bar(0);
ClearFlag(MemoryChunk::HAS_PROGRESS_BAR);
}
}
bool IsLeftOfProgressBar(Object** slot) {
Address slot_address = reinterpret_cast<Address>(slot);
ASSERT(slot_address > this->address());
return (slot_address - (this->address() + kObjectStartOffset)) <
progress_bar();
}
static void IncrementLiveBytesFromGC(Address address, int by) {
MemoryChunk::FromAddress(address)->IncrementLiveBytes(by);
}
static void IncrementLiveBytesFromMutator(Address address, int by);
static const intptr_t kAlignment =
(static_cast<uintptr_t>(1) << kPageSizeBits);
static const intptr_t kAlignmentMask = kAlignment - 1;
static const intptr_t kSizeOffset = kPointerSize + kPointerSize;
static const intptr_t kLiveBytesOffset =
kSizeOffset + kPointerSize + kPointerSize + kPointerSize +
kPointerSize + kPointerSize +
kPointerSize + kPointerSize + kPointerSize + kIntSize;
static const size_t kSlotsBufferOffset = kLiveBytesOffset + kIntSize;
static const size_t kWriteBarrierCounterOffset =
kSlotsBufferOffset + kPointerSize + kPointerSize;
static const size_t kHeaderSize = kWriteBarrierCounterOffset + kPointerSize +
kIntSize + kIntSize + kPointerSize;
static const int kBodyOffset =
CODE_POINTER_ALIGN(kHeaderSize + Bitmap::kSize);
// The start offset of the object area in a page. Aligned to both maps and
// code alignment to be suitable for both. Also aligned to 32 words because
// the marking bitmap is arranged in 32 bit chunks.
static const int kObjectStartAlignment = 32 * kPointerSize;
static const int kObjectStartOffset = kBodyOffset - 1 +
(kObjectStartAlignment - (kBodyOffset - 1) % kObjectStartAlignment);
size_t size() const { return size_; }
void set_size(size_t size) {
size_ = size;
}
void SetArea(Address area_start, Address area_end) {
area_start_ = area_start;
area_end_ = area_end;
}
Executability executable() {
return IsFlagSet(IS_EXECUTABLE) ? EXECUTABLE : NOT_EXECUTABLE;
}
bool ContainsOnlyData() {
return IsFlagSet(CONTAINS_ONLY_DATA);
}
bool InNewSpace() {
return (flags_ & ((1 << IN_FROM_SPACE) | (1 << IN_TO_SPACE))) != 0;
}
bool InToSpace() {
return IsFlagSet(IN_TO_SPACE);
}
bool InFromSpace() {
return IsFlagSet(IN_FROM_SPACE);
}
// ---------------------------------------------------------------------
// Markbits support
inline Bitmap* markbits() {
return Bitmap::FromAddress(address() + kHeaderSize);
}
void PrintMarkbits() { markbits()->Print(); }
inline uint32_t AddressToMarkbitIndex(Address addr) {
return static_cast<uint32_t>(addr - this->address()) >> kPointerSizeLog2;
}
inline static uint32_t FastAddressToMarkbitIndex(Address addr) {
const intptr_t offset =
reinterpret_cast<intptr_t>(addr) & kAlignmentMask;
return static_cast<uint32_t>(offset) >> kPointerSizeLog2;
}
inline Address MarkbitIndexToAddress(uint32_t index) {
return this->address() + (index << kPointerSizeLog2);
}
void InsertAfter(MemoryChunk* other);
void Unlink();
inline Heap* heap() { return heap_; }
static const int kFlagsOffset = kPointerSize * 3;
bool IsEvacuationCandidate() { return IsFlagSet(EVACUATION_CANDIDATE); }
bool ShouldSkipEvacuationSlotRecording() {
return (flags_ & kSkipEvacuationSlotsRecordingMask) != 0;
}
inline SkipList* skip_list() {
return skip_list_;
}
inline void set_skip_list(SkipList* skip_list) {
skip_list_ = skip_list;
}
inline SlotsBuffer* slots_buffer() {
return slots_buffer_;
}
inline SlotsBuffer** slots_buffer_address() {
return &slots_buffer_;
}
void MarkEvacuationCandidate() {
ASSERT(slots_buffer_ == NULL);
SetFlag(EVACUATION_CANDIDATE);
}
void ClearEvacuationCandidate() {
ASSERT(slots_buffer_ == NULL);
ClearFlag(EVACUATION_CANDIDATE);
}
Address area_start() { return area_start_; }
Address area_end() { return area_end_; }
int area_size() {
return static_cast<int>(area_end() - area_start());
}
bool CommitArea(size_t requested);
// Approximate amount of physical memory committed for this chunk.
size_t CommittedPhysicalMemory() {
return high_water_mark_;
}
static inline void UpdateHighWaterMark(Address mark);
protected:
MemoryChunk* next_chunk_;
MemoryChunk* prev_chunk_;
size_t size_;
intptr_t flags_;
// Start and end of allocatable memory on this chunk.
Address area_start_;
Address area_end_;
// If the chunk needs to remember its memory reservation, it is stored here.
VirtualMemory reservation_;
// The identity of the owning space. This is tagged as a failure pointer, but
// no failure can be in an object, so this can be distinguished from any entry
// in a fixed array.
Address owner_;
Heap* heap_;
// Used by the store buffer to keep track of which pages to mark scan-on-
// scavenge.
int store_buffer_counter_;
// Count of bytes marked black on page.
int live_byte_count_;
SlotsBuffer* slots_buffer_;
SkipList* skip_list_;
intptr_t write_barrier_counter_;
// Used by the incremental marker to keep track of the scanning progress in
// large objects that have a progress bar and are scanned in increments.
int progress_bar_;
// Assuming the initial allocation on a page is sequential,
// count highest number of bytes ever allocated on the page.
int high_water_mark_;
intptr_t parallel_sweeping_;
static MemoryChunk* Initialize(Heap* heap,
Address base,
size_t size,
Address area_start,
Address area_end,
Executability executable,
Space* owner);
friend class MemoryAllocator;
};
STATIC_CHECK(sizeof(MemoryChunk) <= MemoryChunk::kHeaderSize);
// -----------------------------------------------------------------------------
// A page is a memory chunk of a size 1MB. Large object pages may be larger.
//
// The only way to get a page pointer is by calling factory methods:
// Page* p = Page::FromAddress(addr); or
// Page* p = Page::FromAllocationTop(top);
class Page : public MemoryChunk {
public:
// Returns the page containing a given address. The address ranges
// from [page_addr .. page_addr + kPageSize[
// This only works if the object is in fact in a page. See also MemoryChunk::
// FromAddress() and FromAnyAddress().
INLINE(static Page* FromAddress(Address a)) {
return reinterpret_cast<Page*>(OffsetFrom(a) & ~kPageAlignmentMask);
}
// Returns the page containing an allocation top. Because an allocation
// top address can be the upper bound of the page, we need to subtract
// it with kPointerSize first. The address ranges from
// [page_addr + kObjectStartOffset .. page_addr + kPageSize].
INLINE(static Page* FromAllocationTop(Address top)) {
Page* p = FromAddress(top - kPointerSize);
return p;
}
// Returns the next page in the chain of pages owned by a space.
inline Page* next_page();
inline Page* prev_page();
inline void set_next_page(Page* page);
inline void set_prev_page(Page* page);
// Checks whether an address is page aligned.
static bool IsAlignedToPageSize(Address a) {
return 0 == (OffsetFrom(a) & kPageAlignmentMask);
}
// Returns the offset of a given address to this page.
INLINE(int Offset(Address a)) {
int offset = static_cast<int>(a - address());
return offset;
}
// Returns the address for a given offset to the this page.
Address OffsetToAddress(int offset) {
ASSERT_PAGE_OFFSET(offset);
return address() + offset;
}
// ---------------------------------------------------------------------
// Page size in bytes. This must be a multiple of the OS page size.
static const int kPageSize = 1 << kPageSizeBits;
// Object area size in bytes.
static const int kNonCodeObjectAreaSize = kPageSize - kObjectStartOffset;
// Maximum object size that fits in a page.
static const int kMaxNonCodeHeapObjectSize = kNonCodeObjectAreaSize;
// Page size mask.
static const intptr_t kPageAlignmentMask = (1 << kPageSizeBits) - 1;
inline void ClearGCFields();
static inline Page* Initialize(Heap* heap,
MemoryChunk* chunk,
Executability executable,
PagedSpace* owner);
void InitializeAsAnchor(PagedSpace* owner);
bool WasSweptPrecisely() { return IsFlagSet(WAS_SWEPT_PRECISELY); }
bool WasSweptConservatively() { return IsFlagSet(WAS_SWEPT_CONSERVATIVELY); }
bool WasSwept() { return WasSweptPrecisely() || WasSweptConservatively(); }
void MarkSweptPrecisely() { SetFlag(WAS_SWEPT_PRECISELY); }
void MarkSweptConservatively() { SetFlag(WAS_SWEPT_CONSERVATIVELY); }
void ClearSweptPrecisely() { ClearFlag(WAS_SWEPT_PRECISELY); }
void ClearSweptConservatively() { ClearFlag(WAS_SWEPT_CONSERVATIVELY); }
#ifdef DEBUG
void Print();
#endif // DEBUG
friend class MemoryAllocator;
};
STATIC_CHECK(sizeof(Page) <= MemoryChunk::kHeaderSize);
class LargePage : public MemoryChunk {
public:
HeapObject* GetObject() {
return HeapObject::FromAddress(area_start());
}
inline LargePage* next_page() const {
return static_cast<LargePage*>(next_chunk());
}
inline void set_next_page(LargePage* page) {
set_next_chunk(page);
}
private:
static inline LargePage* Initialize(Heap* heap, MemoryChunk* chunk);
friend class MemoryAllocator;
};
STATIC_CHECK(sizeof(LargePage) <= MemoryChunk::kHeaderSize);
// ----------------------------------------------------------------------------
// Space is the abstract superclass for all allocation spaces.
class Space : public Malloced {
public:
Space(Heap* heap, AllocationSpace id, Executability executable)
: heap_(heap), id_(id), executable_(executable) {}
virtual ~Space() {}
Heap* heap() const { return heap_; }
// Does the space need executable memory?
Executability executable() { return executable_; }
// Identity used in error reporting.
AllocationSpace identity() { return id_; }
// Returns allocated size.
virtual intptr_t Size() = 0;
// Returns size of objects. Can differ from the allocated size
// (e.g. see LargeObjectSpace).
virtual intptr_t SizeOfObjects() { return Size(); }
virtual int RoundSizeDownToObjectAlignment(int size) {
if (id_ == CODE_SPACE) {
return RoundDown(size, kCodeAlignment);
} else {
return RoundDown(size, kPointerSize);
}
}
#ifdef DEBUG
virtual void Print() = 0;
#endif
private:
Heap* heap_;
AllocationSpace id_;
Executability executable_;
};
// ----------------------------------------------------------------------------
// All heap objects containing executable code (code objects) must be allocated
// from a 2 GB range of memory, so that they can call each other using 32-bit
// displacements. This happens automatically on 32-bit platforms, where 32-bit
// displacements cover the entire 4GB virtual address space. On 64-bit
// platforms, we support this using the CodeRange object, which reserves and
// manages a range of virtual memory.
class CodeRange {
public:
explicit CodeRange(Isolate* isolate);
~CodeRange() { TearDown(); }
// Reserves a range of virtual memory, but does not commit any of it.
// Can only be called once, at heap initialization time.
// Returns false on failure.
bool SetUp(const size_t requested_size);
// Frees the range of virtual memory, and frees the data structures used to
// manage it.
void TearDown();
bool exists() { return this != NULL && code_range_ != NULL; }
Address start() {
if (this == NULL || code_range_ == NULL) return NULL;
return static_cast<Address>(code_range_->address());
}
bool contains(Address address) {
if (this == NULL || code_range_ == NULL) return false;
Address start = static_cast<Address>(code_range_->address());
return start <= address && address < start + code_range_->size();
}
// Allocates a chunk of memory from the large-object portion of
// the code range. On platforms with no separate code range, should
// not be called.
MUST_USE_RESULT Address AllocateRawMemory(const size_t requested_size,
const size_t commit_size,
size_t* allocated);
bool CommitRawMemory(Address start, size_t length);
bool UncommitRawMemory(Address start, size_t length);
void FreeRawMemory(Address buf, size_t length);
private:
Isolate* isolate_;
// The reserved range of virtual memory that all code objects are put in.
VirtualMemory* code_range_;
// Plain old data class, just a struct plus a constructor.
class FreeBlock {
public:
FreeBlock(Address start_arg, size_t size_arg)
: start(start_arg), size(size_arg) {
ASSERT(IsAddressAligned(start, MemoryChunk::kAlignment));
ASSERT(size >= static_cast<size_t>(Page::kPageSize));
}
FreeBlock(void* start_arg, size_t size_arg)
: start(static_cast<Address>(start_arg)), size(size_arg) {
ASSERT(IsAddressAligned(start, MemoryChunk::kAlignment));
ASSERT(size >= static_cast<size_t>(Page::kPageSize));
}
Address start;
size_t size;
};
// Freed blocks of memory are added to the free list. When the allocation
// list is exhausted, the free list is sorted and merged to make the new
// allocation list.
List<FreeBlock> free_list_;
// Memory is allocated from the free blocks on the allocation list.
// The block at current_allocation_block_index_ is the current block.
List<FreeBlock> allocation_list_;
int current_allocation_block_index_;
// Finds a block on the allocation list that contains at least the
// requested amount of memory. If none is found, sorts and merges
// the existing free memory blocks, and searches again.
// If none can be found, terminates V8 with FatalProcessOutOfMemory.
void GetNextAllocationBlock(size_t requested);
// Compares the start addresses of two free blocks.
static int CompareFreeBlockAddress(const FreeBlock* left,
const FreeBlock* right);
DISALLOW_COPY_AND_ASSIGN(CodeRange);
};
class SkipList {
public:
SkipList() {
Clear();
}
void Clear() {
for (int idx = 0; idx < kSize; idx++) {
starts_[idx] = reinterpret_cast<Address>(-1);
}
}
Address StartFor(Address addr) {
return starts_[RegionNumber(addr)];
}
void AddObject(Address addr, int size) {
int start_region = RegionNumber(addr);
int end_region = RegionNumber(addr + size - kPointerSize);
for (int idx = start_region; idx <= end_region; idx++) {
if (starts_[idx] > addr) starts_[idx] = addr;
}
}
static inline int RegionNumber(Address addr) {
return (OffsetFrom(addr) & Page::kPageAlignmentMask) >> kRegionSizeLog2;
}
static void Update(Address addr, int size) {
Page* page = Page::FromAddress(addr);
SkipList* list = page->skip_list();
if (list == NULL) {
list = new SkipList();
page->set_skip_list(list);
}
list->AddObject(addr, size);
}
private:
static const int kRegionSizeLog2 = 13;
static const int kRegionSize = 1 << kRegionSizeLog2;
static const int kSize = Page::kPageSize / kRegionSize;
STATIC_ASSERT(Page::kPageSize % kRegionSize == 0);
Address starts_[kSize];
};
// ----------------------------------------------------------------------------
// A space acquires chunks of memory from the operating system. The memory
// allocator allocated and deallocates pages for the paged heap spaces and large
// pages for large object space.
//
// Each space has to manage it's own pages.
//
class MemoryAllocator {
public:
explicit MemoryAllocator(Isolate* isolate);
// Initializes its internal bookkeeping structures.
// Max capacity of the total space and executable memory limit.
bool SetUp(intptr_t max_capacity, intptr_t capacity_executable);
void TearDown();
Page* AllocatePage(
intptr_t size, PagedSpace* owner, Executability executable);
LargePage* AllocateLargePage(
intptr_t object_size, Space* owner, Executability executable);
void Free(MemoryChunk* chunk);
// Returns the maximum available bytes of heaps.
intptr_t Available() { return capacity_ < size_ ? 0 : capacity_ - size_; }
// Returns allocated spaces in bytes.
intptr_t Size() { return size_; }
// Returns the maximum available executable bytes of heaps.
intptr_t AvailableExecutable() {
if (capacity_executable_ < size_executable_) return 0;
return capacity_executable_ - size_executable_;
}
// Returns allocated executable spaces in bytes.
intptr_t SizeExecutable() { return size_executable_; }
// Returns maximum available bytes that the old space can have.
intptr_t MaxAvailable() {
return (Available() / Page::kPageSize) * Page::kMaxNonCodeHeapObjectSize;
}
#ifdef DEBUG
// Reports statistic info of the space.
void ReportStatistics();
#endif
// Returns a MemoryChunk in which the memory region from commit_area_size to
// reserve_area_size of the chunk area is reserved but not committed, it
// could be committed later by calling MemoryChunk::CommitArea.
MemoryChunk* AllocateChunk(intptr_t reserve_area_size,
intptr_t commit_area_size,
Executability executable,
Space* space);
Address ReserveAlignedMemory(size_t requested,
size_t alignment,
VirtualMemory* controller);
Address AllocateAlignedMemory(size_t reserve_size,
size_t commit_size,
size_t alignment,
Executability executable,
VirtualMemory* controller);
void FreeMemory(VirtualMemory* reservation, Executability executable);
void FreeMemory(Address addr, size_t size, Executability executable);
// Commit a contiguous block of memory from the initial chunk. Assumes that
// the address is not NULL, the size is greater than zero, and that the
// block is contained in the initial chunk. Returns true if it succeeded
// and false otherwise.
bool CommitBlock(Address start, size_t size, Executability executable);
// Uncommit a contiguous block of memory [start..(start+size)[.
// start is not NULL, the size is greater than zero, and the
// block is contained in the initial chunk. Returns true if it succeeded
// and false otherwise.
bool UncommitBlock(Address start, size_t size);
// Zaps a contiguous block of memory [start..(start+size)[ thus
// filling it up with a recognizable non-NULL bit pattern.
void ZapBlock(Address start, size_t size);
void PerformAllocationCallback(ObjectSpace space,
AllocationAction action,
size_t size);
void AddMemoryAllocationCallback(MemoryAllocationCallback callback,
ObjectSpace space,
AllocationAction action);
void RemoveMemoryAllocationCallback(
MemoryAllocationCallback callback);
bool MemoryAllocationCallbackRegistered(
MemoryAllocationCallback callback);
static int CodePageGuardStartOffset();
static int CodePageGuardSize();
static int CodePageAreaStartOffset();
static int CodePageAreaEndOffset();
static int CodePageAreaSize() {
return CodePageAreaEndOffset() - CodePageAreaStartOffset();
}
MUST_USE_RESULT static bool CommitExecutableMemory(VirtualMemory* vm,
Address start,
size_t commit_size,
size_t reserved_size);
private:
Isolate* isolate_;
// Maximum space size in bytes.
size_t capacity_;
// Maximum subset of capacity_ that can be executable
size_t capacity_executable_;
// Allocated space size in bytes.
size_t size_;
// Allocated executable space size in bytes.
size_t size_executable_;
struct MemoryAllocationCallbackRegistration {
MemoryAllocationCallbackRegistration(MemoryAllocationCallback callback,
ObjectSpace space,
AllocationAction action)
: callback(callback), space(space), action(action) {
}
MemoryAllocationCallback callback;
ObjectSpace space;
AllocationAction action;
};
// A List of callback that are triggered when memory is allocated or free'd
List<MemoryAllocationCallbackRegistration>
memory_allocation_callbacks_;
// Initializes pages in a chunk. Returns the first page address.
// This function and GetChunkId() are provided for the mark-compact
// collector to rebuild page headers in the from space, which is
// used as a marking stack and its page headers are destroyed.
Page* InitializePagesInChunk(int chunk_id, int pages_in_chunk,
PagedSpace* owner);
DISALLOW_IMPLICIT_CONSTRUCTORS(MemoryAllocator);
};
// -----------------------------------------------------------------------------
// Interface for heap object iterator to be implemented by all object space
// object iterators.
//
// NOTE: The space specific object iterators also implements the own next()
// method which is used to avoid using virtual functions
// iterating a specific space.
class ObjectIterator : public Malloced {
public:
virtual ~ObjectIterator() { }
virtual HeapObject* next_object() = 0;
};
// -----------------------------------------------------------------------------
// Heap object iterator in new/old/map spaces.
//
// A HeapObjectIterator iterates objects from the bottom of the given space
// to its top or from the bottom of the given page to its top.
//
// If objects are allocated in the page during iteration the iterator may
// or may not iterate over those objects. The caller must create a new
// iterator in order to be sure to visit these new objects.
class HeapObjectIterator: public ObjectIterator {
public:
// Creates a new object iterator in a given space.
// If the size function is not given, the iterator calls the default
// Object::Size().
explicit HeapObjectIterator(PagedSpace* space);
HeapObjectIterator(PagedSpace* space, HeapObjectCallback size_func);
HeapObjectIterator(Page* page, HeapObjectCallback size_func);
// Advance to the next object, skipping free spaces and other fillers and
// skipping the special garbage section of which there is one per space.
// Returns NULL when the iteration has ended.
inline HeapObject* Next() {
do {
HeapObject* next_obj = FromCurrentPage();
if (next_obj != NULL) return next_obj;
} while (AdvanceToNextPage());
return NULL;
}
virtual HeapObject* next_object() {
return Next();
}
private:
enum PageMode { kOnePageOnly, kAllPagesInSpace };
Address cur_addr_; // Current iteration point.
Address cur_end_; // End iteration point.
HeapObjectCallback size_func_; // Size function or NULL.
PagedSpace* space_;
PageMode page_mode_;
// Fast (inlined) path of next().
inline HeapObject* FromCurrentPage();
// Slow path of next(), goes into the next page. Returns false if the
// iteration has ended.
bool AdvanceToNextPage();
// Initializes fields.
inline void Initialize(PagedSpace* owner,
Address start,
Address end,
PageMode mode,
HeapObjectCallback size_func);
};
// -----------------------------------------------------------------------------
// A PageIterator iterates the pages in a paged space.
class PageIterator BASE_EMBEDDED {
public:
explicit inline PageIterator(PagedSpace* space);
inline bool has_next();
inline Page* next();
private:
PagedSpace* space_;
Page* prev_page_; // Previous page returned.
// Next page that will be returned. Cached here so that we can use this
// iterator for operations that deallocate pages.
Page* next_page_;
};
// -----------------------------------------------------------------------------
// A space has a circular list of pages. The next page can be accessed via
// Page::next_page() call.
// An abstraction of allocation and relocation pointers in a page-structured
// space.
class AllocationInfo {
public:
AllocationInfo() : top(NULL), limit(NULL) {
}
Address top; // Current allocation top.
Address limit; // Current allocation limit.
#ifdef DEBUG
bool VerifyPagedAllocation() {
return (Page::FromAllocationTop(top) == Page::FromAllocationTop(limit))
&& (top <= limit);
}
#endif
};
// An abstraction of the accounting statistics of a page-structured space.
// The 'capacity' of a space is the number of object-area bytes (i.e., not
// including page bookkeeping structures) currently in the space. The 'size'
// of a space is the number of allocated bytes, the 'waste' in the space is
// the number of bytes that are not allocated and not available to
// allocation without reorganizing the space via a GC (e.g. small blocks due
// to internal fragmentation, top of page areas in map space), and the bytes
// 'available' is the number of unallocated bytes that are not waste. The
// capacity is the sum of size, waste, and available.
//
// The stats are only set by functions that ensure they stay balanced. These
// functions increase or decrease one of the non-capacity stats in
// conjunction with capacity, or else they always balance increases and
// decreases to the non-capacity stats.
class AllocationStats BASE_EMBEDDED {
public:
AllocationStats() { Clear(); }
// Zero out all the allocation statistics (i.e., no capacity).
void Clear() {
capacity_ = 0;
size_ = 0;
waste_ = 0;
}
void ClearSizeWaste() {
size_ = capacity_;
waste_ = 0;
}
// Reset the allocation statistics (i.e., available = capacity with no
// wasted or allocated bytes).
void Reset() {
size_ = 0;
waste_ = 0;
}
// Accessors for the allocation statistics.
intptr_t Capacity() { return capacity_; }
intptr_t Size() { return size_; }
intptr_t Waste() { return waste_; }
// Grow the space by adding available bytes. They are initially marked as
// being in use (part of the size), but will normally be immediately freed,
// putting them on the free list and removing them from size_.
void ExpandSpace(int size_in_bytes) {
capacity_ += size_in_bytes;
size_ += size_in_bytes;
ASSERT(size_ >= 0);
}
// Shrink the space by removing available bytes. Since shrinking is done
// during sweeping, bytes have been marked as being in use (part of the size)
// and are hereby freed.
void ShrinkSpace(int size_in_bytes) {
capacity_ -= size_in_bytes;
size_ -= size_in_bytes;
ASSERT(size_ >= 0);
}
// Allocate from available bytes (available -> size).
void AllocateBytes(intptr_t size_in_bytes) {
size_ += size_in_bytes;
ASSERT(size_ >= 0);
}
// Free allocated bytes, making them available (size -> available).
void DeallocateBytes(intptr_t size_in_bytes) {
size_ -= size_in_bytes;
ASSERT(size_ >= 0);
}
// Waste free bytes (available -> waste).
void WasteBytes(int size_in_bytes) {
size_ -= size_in_bytes;
waste_ += size_in_bytes;
ASSERT(size_ >= 0);
}
private:
intptr_t capacity_;
intptr_t size_;
intptr_t waste_;
};
// -----------------------------------------------------------------------------
// Free lists for old object spaces
//
// Free-list nodes are free blocks in the heap. They look like heap objects
// (free-list node pointers have the heap object tag, and they have a map like
// a heap object). They have a size and a next pointer. The next pointer is
// the raw address of the next free list node (or NULL).
class FreeListNode: public HeapObject {
public:
// Obtain a free-list node from a raw address. This is not a cast because
// it does not check nor require that the first word at the address is a map
// pointer.
static FreeListNode* FromAddress(Address address) {
return reinterpret_cast<FreeListNode*>(HeapObject::FromAddress(address));
}
static inline bool IsFreeListNode(HeapObject* object);
// Set the size in bytes, which can be read with HeapObject::Size(). This
// function also writes a map to the first word of the block so that it
// looks like a heap object to the garbage collector and heap iteration
// functions.
void set_size(Heap* heap, int size_in_bytes);
// Accessors for the next field.
inline FreeListNode* next();
inline FreeListNode** next_address();
inline void set_next(FreeListNode* next);
inline void Zap();
Refactoring of snapshots. This simplifies and improves the speed of deserializing code. The current startup time improvement for V8 is around 6%, but code deserialization is speeded up disproportionately, and we will soon have more code in the snapshot. * Removed support for deserializing into large object space. The regular pages are 1Mbyte now and that is plenty. This is a big simplification. * Instead of reserving space for the snapshot we actually allocate it now. This removes some special casing from the memory management and simplifies deserialization since we are just bumping a pointer rather than calling the normal allocation routines during deserialization. * Record in the snapshot how much we need to boot up and allocate it instead of just assuming that allocations in a new VM will always be linear. * In the snapshot we always address an object as a negative offset from the current allocation point. We used to sometimes address from the start of the deserialized data, but this is less useful now that we have good support for roots and repetitions in the deserialization data. * Code objects were previously deserialized (like other objects) by alternating raw data (deserialized with memcpy) and pointers (to external references, other objects, etc.). Now we deserialize code objects with a single memcpy, followed by a series of skips and pointers that partially overwrite the code we memcopied out of the snapshot. The skips are sometimes merged into the following instruction in the deserialization data to reduce dispatch time. * Integers in the snapshot were stored in a variable length format that gives a compact representation for small positive integers. This is still the case, but the new encoding can be decoded without branches or conditional instructions, which is faster on a modern CPU. Review URL: https://chromiumcodereview.appspot.com/10918067 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12505 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-09-14 11:16:56 +00:00
static inline FreeListNode* cast(MaybeObject* maybe) {
ASSERT(!maybe->IsFailure());
return reinterpret_cast<FreeListNode*>(maybe);
}
private:
static const int kNextOffset = POINTER_SIZE_ALIGN(FreeSpace::kHeaderSize);
DISALLOW_IMPLICIT_CONSTRUCTORS(FreeListNode);
};
// The free list category holds a pointer to the top element and a pointer to
// the end element of the linked list of free memory blocks.
class FreeListCategory {
public:
FreeListCategory() :
top_(NULL),
end_(NULL),
mutex_(OS::CreateMutex()),
available_(0) {}
~FreeListCategory() {
delete mutex_;
}
intptr_t Concatenate(FreeListCategory* category);
void Reset();
void Free(FreeListNode* node, int size_in_bytes);
FreeListNode* PickNodeFromList(int *node_size);
intptr_t CountFreeListItemsInList(Page* p);
intptr_t EvictFreeListItemsInList(Page* p);
void RepairFreeList(Heap* heap);
FreeListNode** GetTopAddress() { return &top_; }
FreeListNode* top() const { return top_; }
void set_top(FreeListNode* top) { top_ = top; }
FreeListNode** GetEndAddress() { return &end_; }
FreeListNode* end() const { return end_; }
void set_end(FreeListNode* end) { end_ = end; }
int* GetAvailableAddress() { return &available_; }
int available() const { return available_; }
void set_available(int available) { available_ = available; }
Mutex* mutex() { return mutex_; }
#ifdef DEBUG
intptr_t SumFreeList();
int FreeListLength();
#endif
private:
FreeListNode* top_;
FreeListNode* end_;
Mutex* mutex_;
// Total available bytes in all blocks of this free list category.
int available_;
};
// The free list for the old space. The free list is organized in such a way
// as to encourage objects allocated around the same time to be near each
// other. The normal way to allocate is intended to be by bumping a 'top'
// pointer until it hits a 'limit' pointer. When the limit is hit we need to
// find a new space to allocate from. This is done with the free list, which
// is divided up into rough categories to cut down on waste. Having finer
// categories would scatter allocation more.
// The old space free list is organized in categories.
// 1-31 words: Such small free areas are discarded for efficiency reasons.
// They can be reclaimed by the compactor. However the distance between top
// and limit may be this small.
// 32-255 words: There is a list of spaces this large. It is used for top and
// limit when the object we need to allocate is 1-31 words in size. These
// spaces are called small.
// 256-2047 words: There is a list of spaces this large. It is used for top and
// limit when the object we need to allocate is 32-255 words in size. These
// spaces are called medium.
// 1048-16383 words: There is a list of spaces this large. It is used for top
// and limit when the object we need to allocate is 256-2047 words in size.
// These spaces are call large.
// At least 16384 words. This list is for objects of 2048 words or larger.
// Empty pages are added to this list. These spaces are called huge.
class FreeList BASE_EMBEDDED {
public:
explicit FreeList(PagedSpace* owner);
intptr_t Concatenate(FreeList* free_list);
// Clear the free list.
void Reset();
// Return the number of bytes available on the free list.
intptr_t available() {
return small_list_.available() + medium_list_.available() +
large_list_.available() + huge_list_.available();
}
// Place a node on the free list. The block of size 'size_in_bytes'
// starting at 'start' is placed on the free list. The return value is the
// number of bytes that have been lost due to internal fragmentation by
// freeing the block. Bookkeeping information will be written to the block,
// i.e., its contents will be destroyed. The start address should be word
// aligned, and the size should be a non-zero multiple of the word size.
int Free(Address start, int size_in_bytes);
// Allocate a block of size 'size_in_bytes' from the free list. The block
// is unitialized. A failure is returned if no block is available. The
// number of bytes lost to fragmentation is returned in the output parameter
// 'wasted_bytes'. The size should be a non-zero multiple of the word size.
MUST_USE_RESULT HeapObject* Allocate(int size_in_bytes);
#ifdef DEBUG
void Zap();
intptr_t SumFreeLists();
bool IsVeryLong();
#endif
Refactoring of snapshots. This simplifies and improves the speed of deserializing code. The current startup time improvement for V8 is around 6%, but code deserialization is speeded up disproportionately, and we will soon have more code in the snapshot. * Removed support for deserializing into large object space. The regular pages are 1Mbyte now and that is plenty. This is a big simplification. * Instead of reserving space for the snapshot we actually allocate it now. This removes some special casing from the memory management and simplifies deserialization since we are just bumping a pointer rather than calling the normal allocation routines during deserialization. * Record in the snapshot how much we need to boot up and allocate it instead of just assuming that allocations in a new VM will always be linear. * In the snapshot we always address an object as a negative offset from the current allocation point. We used to sometimes address from the start of the deserialized data, but this is less useful now that we have good support for roots and repetitions in the deserialization data. * Code objects were previously deserialized (like other objects) by alternating raw data (deserialized with memcpy) and pointers (to external references, other objects, etc.). Now we deserialize code objects with a single memcpy, followed by a series of skips and pointers that partially overwrite the code we memcopied out of the snapshot. The skips are sometimes merged into the following instruction in the deserialization data to reduce dispatch time. * Integers in the snapshot were stored in a variable length format that gives a compact representation for small positive integers. This is still the case, but the new encoding can be decoded without branches or conditional instructions, which is faster on a modern CPU. Review URL: https://chromiumcodereview.appspot.com/10918067 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12505 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-09-14 11:16:56 +00:00
// Used after booting the VM.
void RepairLists(Heap* heap);
struct SizeStats {
intptr_t Total() {
return small_size_ + medium_size_ + large_size_ + huge_size_;
}
intptr_t small_size_;
intptr_t medium_size_;
intptr_t large_size_;
intptr_t huge_size_;
};
void CountFreeListItems(Page* p, SizeStats* sizes);
intptr_t EvictFreeListItems(Page* p);
FreeListCategory* small_list() { return &small_list_; }
FreeListCategory* medium_list() { return &medium_list_; }
FreeListCategory* large_list() { return &large_list_; }
FreeListCategory* huge_list() { return &huge_list_; }
private:
// The size range of blocks, in bytes.
static const int kMinBlockSize = 3 * kPointerSize;
static const int kMaxBlockSize = Page::kMaxNonCodeHeapObjectSize;
FreeListNode* FindNodeFor(int size_in_bytes, int* node_size);
PagedSpace* owner_;
Heap* heap_;
static const int kSmallListMin = 0x20 * kPointerSize;
static const int kSmallListMax = 0xff * kPointerSize;
static const int kMediumListMax = 0x7ff * kPointerSize;
static const int kLargeListMax = 0x3fff * kPointerSize;
static const int kSmallAllocationMax = kSmallListMin - kPointerSize;
static const int kMediumAllocationMax = kSmallListMax;
static const int kLargeAllocationMax = kMediumListMax;
FreeListCategory small_list_;
FreeListCategory medium_list_;
FreeListCategory large_list_;
FreeListCategory huge_list_;
DISALLOW_IMPLICIT_CONSTRUCTORS(FreeList);
};
class PagedSpace : public Space {
public:
// Creates a space with a maximum capacity, and an id.
PagedSpace(Heap* heap,
intptr_t max_capacity,
AllocationSpace id,
Executability executable);
virtual ~PagedSpace() {}
// Set up the space using the given address range of virtual memory (from
// the memory allocator's initial chunk) if possible. If the block of
// addresses is not big enough to contain a single page-aligned page, a
// fresh chunk will be allocated.
bool SetUp();
// Returns true if the space has been successfully set up and not
// subsequently torn down.
bool HasBeenSetUp();
// Cleans up the space, frees all pages in this space except those belonging
// to the initial chunk, uncommits addresses in the initial chunk.
void TearDown();
// Checks whether an object/address is in this space.
inline bool Contains(Address a);
bool Contains(HeapObject* o) { return Contains(o->address()); }
// Given an address occupied by a live object, return that object if it is
// in this space, or Failure::Exception() if it is not. The implementation
// iterates over objects in the page containing the address, the cost is
// linear in the number of objects in the page. It may be slow.
MUST_USE_RESULT MaybeObject* FindObject(Address addr);
Refactoring of snapshots. This simplifies and improves the speed of deserializing code. The current startup time improvement for V8 is around 6%, but code deserialization is speeded up disproportionately, and we will soon have more code in the snapshot. * Removed support for deserializing into large object space. The regular pages are 1Mbyte now and that is plenty. This is a big simplification. * Instead of reserving space for the snapshot we actually allocate it now. This removes some special casing from the memory management and simplifies deserialization since we are just bumping a pointer rather than calling the normal allocation routines during deserialization. * Record in the snapshot how much we need to boot up and allocate it instead of just assuming that allocations in a new VM will always be linear. * In the snapshot we always address an object as a negative offset from the current allocation point. We used to sometimes address from the start of the deserialized data, but this is less useful now that we have good support for roots and repetitions in the deserialization data. * Code objects were previously deserialized (like other objects) by alternating raw data (deserialized with memcpy) and pointers (to external references, other objects, etc.). Now we deserialize code objects with a single memcpy, followed by a series of skips and pointers that partially overwrite the code we memcopied out of the snapshot. The skips are sometimes merged into the following instruction in the deserialization data to reduce dispatch time. * Integers in the snapshot were stored in a variable length format that gives a compact representation for small positive integers. This is still the case, but the new encoding can be decoded without branches or conditional instructions, which is faster on a modern CPU. Review URL: https://chromiumcodereview.appspot.com/10918067 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12505 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-09-14 11:16:56 +00:00
// During boot the free_space_map is created, and afterwards we may need
// to write it into the free list nodes that were already created.
virtual void RepairFreeListsAfterBoot();
// Prepares for a mark-compact GC.
virtual void PrepareForMarkCompact();
// Current capacity without growing (Size() + Available()).
intptr_t Capacity() { return accounting_stats_.Capacity(); }
// Total amount of memory committed for this space. For paged
// spaces this equals the capacity.
intptr_t CommittedMemory() { return Capacity(); }
// Approximate amount of physical memory committed for this space.
size_t CommittedPhysicalMemory();
// Sets the capacity, the available space and the wasted space to zero.
// The stats are rebuilt during sweeping by adding each page to the
// capacity and the size when it is encountered. As free spaces are
// discovered during the sweeping they are subtracted from the size and added
// to the available and wasted totals.
void ClearStats() {
accounting_stats_.ClearSizeWaste();
}
// Increases the number of available bytes of that space.
void AddToAccountingStats(intptr_t bytes) {
accounting_stats_.DeallocateBytes(bytes);
}
// Available bytes without growing. These are the bytes on the free list.
// The bytes in the linear allocation area are not included in this total
// because updating the stats would slow down allocation. New pages are
// immediately added to the free list so they show up here.
intptr_t Available() { return free_list_.available(); }
// Allocated bytes in this space. Garbage bytes that were not found due to
// lazy sweeping are counted as being allocated! The bytes in the current
// linear allocation area (between top and limit) are also counted here.
virtual intptr_t Size() { return accounting_stats_.Size(); }
// As size, but the bytes in lazily swept pages are estimated and the bytes
// in the current linear allocation area are not included.
virtual intptr_t SizeOfObjects();
// Wasted bytes in this space. These are just the bytes that were thrown away
// due to being too small to use for allocation. They do not include the
// free bytes that were not found at all due to lazy sweeping.
virtual intptr_t Waste() { return accounting_stats_.Waste(); }
// Returns the allocation pointer in this space.
Address top() { return allocation_info_.top; }
Address limit() { return allocation_info_.limit; }
// The allocation top and limit addresses.
Address* allocation_top_address() { return &allocation_info_.top; }
Address* allocation_limit_address() { return &allocation_info_.limit; }
// Allocate the requested number of bytes in the space if possible, return a
// failure object if not.
MUST_USE_RESULT inline MaybeObject* AllocateRaw(int size_in_bytes);
virtual bool ReserveSpace(int bytes);
// Give a block of memory to the space's free list. It might be added to
// the free list or accounted as waste.
// If add_to_freelist is false then just accounting stats are updated and
// no attempt to add area to free list is made.
int Free(Address start, int size_in_bytes) {
int wasted = free_list_.Free(start, size_in_bytes);
accounting_stats_.DeallocateBytes(size_in_bytes - wasted);
return size_in_bytes - wasted;
}
void ResetFreeList() {
free_list_.Reset();
}
// Set space allocation info.
void SetTop(Address top, Address limit) {
ASSERT(top == limit ||
Page::FromAddress(top) == Page::FromAddress(limit - 1));
MemoryChunk::UpdateHighWaterMark(allocation_info_.top);
allocation_info_.top = top;
allocation_info_.limit = limit;
}
void Allocate(int bytes) {
accounting_stats_.AllocateBytes(bytes);
}
void IncreaseCapacity(int size) {
accounting_stats_.ExpandSpace(size);
}
// Releases an unused page and shrinks the space.
void ReleasePage(Page* page, bool unlink);
// The dummy page that anchors the linked list of pages.
Page* anchor() { return &anchor_; }
#ifdef VERIFY_HEAP
// Verify integrity of this space.
virtual void Verify(ObjectVisitor* visitor);
// Overridden by subclasses to verify space-specific object
// properties (e.g., only maps or free-list nodes are in map space).
virtual void VerifyObject(HeapObject* obj) {}
#endif
#ifdef DEBUG
// Print meta info and objects in this space.
virtual void Print();
// Reports statistics for the space
void ReportStatistics();
// Report code object related statistics
void CollectCodeStatistics();
static void ReportCodeStatistics();
static void ResetCodeStatistics();
#endif
bool was_swept_conservatively() { return was_swept_conservatively_; }
void set_was_swept_conservatively(bool b) { was_swept_conservatively_ = b; }
// Evacuation candidates are swept by evacuator. Needs to return a valid
// result before _and_ after evacuation has finished.
static bool ShouldBeSweptLazily(Page* p) {
return !p->IsEvacuationCandidate() &&
!p->IsFlagSet(Page::RESCAN_ON_EVACUATION) &&
!p->WasSweptPrecisely();
}
void SetPagesToSweep(Page* first) {
ASSERT(unswept_free_bytes_ == 0);
if (first == &anchor_) first = NULL;
first_unswept_page_ = first;
}
void IncrementUnsweptFreeBytes(intptr_t by) {
unswept_free_bytes_ += by;
}
void IncreaseUnsweptFreeBytes(Page* p) {
ASSERT(ShouldBeSweptLazily(p));
unswept_free_bytes_ += (p->area_size() - p->LiveBytes());
}
void DecrementUnsweptFreeBytes(intptr_t by) {
unswept_free_bytes_ -= by;
}
void DecreaseUnsweptFreeBytes(Page* p) {
ASSERT(ShouldBeSweptLazily(p));
unswept_free_bytes_ -= (p->area_size() - p->LiveBytes());
}
void ResetUnsweptFreeBytes() {
unswept_free_bytes_ = 0;
}
bool AdvanceSweeper(intptr_t bytes_to_sweep);
// When parallel sweeper threads are active and the main thread finished
// its sweeping phase, this function waits for them to complete, otherwise
// AdvanceSweeper with size_in_bytes is called.
bool EnsureSweeperProgress(intptr_t size_in_bytes);
bool IsLazySweepingComplete() {
return !first_unswept_page_->is_valid();
}
Page* FirstPage() { return anchor_.next_page(); }
Page* LastPage() { return anchor_.prev_page(); }
void CountFreeListItems(Page* p, FreeList::SizeStats* sizes) {
free_list_.CountFreeListItems(p, sizes);
}
void EvictEvacuationCandidatesFromFreeLists();
bool CanExpand();
// Returns the number of total pages in this space.
int CountTotalPages();
// Return size of allocatable area on a page in this space.
inline int AreaSize() {
return area_size_;
}
protected:
FreeList* free_list() { return &free_list_; }
int area_size_;
// Maximum capacity of this space.
intptr_t max_capacity_;
intptr_t SizeOfFirstPage();
// Accounting information for this space.
AllocationStats accounting_stats_;
// The dummy page that anchors the double linked list of pages.
Page anchor_;
// The space's free list.
FreeList free_list_;
// Normal allocation information.
AllocationInfo allocation_info_;
// Bytes of each page that cannot be allocated. Possibly non-zero
// for pages in spaces with only fixed-size objects. Always zero
// for pages in spaces with variable sized objects (those pages are
// padded with free-list nodes).
int page_extra_;
bool was_swept_conservatively_;
// The first page to be swept when the lazy sweeper advances. Is set
// to NULL when all pages have been swept.
Page* first_unswept_page_;
// The number of free bytes which could be reclaimed by advancing the
// lazy sweeper. This is only an estimation because lazy sweeping is
// done conservatively.
intptr_t unswept_free_bytes_;
// Expands the space by allocating a fixed number of pages. Returns false if
// it cannot allocate requested number of pages from OS, or if the hard heap
// size limit has been hit.
bool Expand();
// Generic fast case allocation function that tries linear allocation at the
// address denoted by top in allocation_info_.
inline HeapObject* AllocateLinearly(int size_in_bytes);
// Slow path of AllocateRaw. This function is space-dependent.
MUST_USE_RESULT virtual HeapObject* SlowAllocateRaw(int size_in_bytes);
friend class PageIterator;
friend class SweeperThread;
};
class NumberAndSizeInfo BASE_EMBEDDED {
public:
NumberAndSizeInfo() : number_(0), bytes_(0) {}
int number() const { return number_; }
void increment_number(int num) { number_ += num; }
int bytes() const { return bytes_; }
void increment_bytes(int size) { bytes_ += size; }
void clear() {
number_ = 0;
bytes_ = 0;
}
private:
int number_;
int bytes_;
};
// HistogramInfo class for recording a single "bar" of a histogram. This
// class is used for collecting statistics to print to the log file.
class HistogramInfo: public NumberAndSizeInfo {
public:
HistogramInfo() : NumberAndSizeInfo() {}
const char* name() { return name_; }
void set_name(const char* name) { name_ = name; }
private:
const char* name_;
};
enum SemiSpaceId {
kFromSpace = 0,
kToSpace = 1
};
class SemiSpace;
class NewSpacePage : public MemoryChunk {
public:
// GC related flags copied from from-space to to-space when
// flipping semispaces.
static const intptr_t kCopyOnFlipFlagsMask =
(1 << MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING) |
(1 << MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING) |
(1 << MemoryChunk::SCAN_ON_SCAVENGE);
static const int kAreaSize = Page::kNonCodeObjectAreaSize;
inline NewSpacePage* next_page() const {
return static_cast<NewSpacePage*>(next_chunk());
}
inline void set_next_page(NewSpacePage* page) {
set_next_chunk(page);
}
inline NewSpacePage* prev_page() const {
return static_cast<NewSpacePage*>(prev_chunk());
}
inline void set_prev_page(NewSpacePage* page) {
set_prev_chunk(page);
}
SemiSpace* semi_space() {
return reinterpret_cast<SemiSpace*>(owner());
}
bool is_anchor() { return !this->InNewSpace(); }
static bool IsAtStart(Address addr) {
return (reinterpret_cast<intptr_t>(addr) & Page::kPageAlignmentMask)
== kObjectStartOffset;
}
static bool IsAtEnd(Address addr) {
return (reinterpret_cast<intptr_t>(addr) & Page::kPageAlignmentMask) == 0;
}
Address address() {
return reinterpret_cast<Address>(this);
}
// Finds the NewSpacePage containg the given address.
static inline NewSpacePage* FromAddress(Address address_in_page) {
Address page_start =
reinterpret_cast<Address>(reinterpret_cast<uintptr_t>(address_in_page) &
~Page::kPageAlignmentMask);
NewSpacePage* page = reinterpret_cast<NewSpacePage*>(page_start);
return page;
}
// Find the page for a limit address. A limit address is either an address
// inside a page, or the address right after the last byte of a page.
static inline NewSpacePage* FromLimit(Address address_limit) {
return NewSpacePage::FromAddress(address_limit - 1);
}
private:
// Create a NewSpacePage object that is only used as anchor
// for the doubly-linked list of real pages.
explicit NewSpacePage(SemiSpace* owner) {
InitializeAsAnchor(owner);
}
static NewSpacePage* Initialize(Heap* heap,
Address start,
SemiSpace* semi_space);
// Intialize a fake NewSpacePage used as sentinel at the ends
// of a doubly-linked list of real NewSpacePages.
// Only uses the prev/next links, and sets flags to not be in new-space.
void InitializeAsAnchor(SemiSpace* owner);
friend class SemiSpace;
friend class SemiSpaceIterator;
};
// -----------------------------------------------------------------------------
// SemiSpace in young generation
//
// A semispace is a contiguous chunk of memory holding page-like memory
// chunks. The mark-compact collector uses the memory of the first page in
// the from space as a marking stack when tracing live objects.
class SemiSpace : public Space {
public:
// Constructor.
SemiSpace(Heap* heap, SemiSpaceId semispace)
: Space(heap, NEW_SPACE, NOT_EXECUTABLE),
start_(NULL),
age_mark_(NULL),
id_(semispace),
anchor_(this),
current_page_(NULL) { }
// Sets up the semispace using the given chunk.
void SetUp(Address start, int initial_capacity, int maximum_capacity);
// Tear down the space. Heap memory was not allocated by the space, so it
// is not deallocated here.
void TearDown();
// True if the space has been set up but not torn down.
bool HasBeenSetUp() { return start_ != NULL; }
// Grow the semispace to the new capacity. The new capacity
// requested must be larger than the current capacity and less than
// the maximum capacity.
bool GrowTo(int new_capacity);
// Shrinks the semispace to the new capacity. The new capacity
// requested must be more than the amount of used memory in the
// semispace and less than the current capacity.
bool ShrinkTo(int new_capacity);
// Returns the start address of the first page of the space.
Address space_start() {
ASSERT(anchor_.next_page() != &anchor_);
return anchor_.next_page()->area_start();
}
// Returns the start address of the current page of the space.
Address page_low() {
return current_page_->area_start();
}
// Returns one past the end address of the space.
Address space_end() {
return anchor_.prev_page()->area_end();
}
// Returns one past the end address of the current page of the space.
Address page_high() {
return current_page_->area_end();
}
bool AdvancePage() {
NewSpacePage* next_page = current_page_->next_page();
if (next_page == anchor()) return false;
current_page_ = next_page;
return true;
}
// Resets the space to using the first page.
void Reset();
// Age mark accessors.
Address age_mark() { return age_mark_; }
void set_age_mark(Address mark);
// True if the address is in the address range of this semispace (not
// necessarily below the allocation pointer).
bool Contains(Address a) {
return (reinterpret_cast<uintptr_t>(a) & address_mask_)
== reinterpret_cast<uintptr_t>(start_);
}
// True if the object is a heap object in the address range of this
// semispace (not necessarily below the allocation pointer).
bool Contains(Object* o) {
return (reinterpret_cast<uintptr_t>(o) & object_mask_) == object_expected_;
}
// If we don't have these here then SemiSpace will be abstract. However
// they should never be called.
virtual intptr_t Size() {
UNREACHABLE();
return 0;
}
virtual bool ReserveSpace(int bytes) {
UNREACHABLE();
return false;
}
bool is_committed() { return committed_; }
bool Commit();
bool Uncommit();
NewSpacePage* first_page() { return anchor_.next_page(); }
NewSpacePage* current_page() { return current_page_; }
#ifdef VERIFY_HEAP
virtual void Verify();
#endif
#ifdef DEBUG
virtual void Print();
// Validate a range of of addresses in a SemiSpace.
// The "from" address must be on a page prior to the "to" address,
// in the linked page order, or it must be earlier on the same page.
static void AssertValidRange(Address from, Address to);
#else
// Do nothing.
inline static void AssertValidRange(Address from, Address to) {}
#endif
// Returns the current capacity of the semi space.
int Capacity() { return capacity_; }
// Returns the maximum capacity of the semi space.
int MaximumCapacity() { return maximum_capacity_; }
// Returns the initial capacity of the semi space.
int InitialCapacity() { return initial_capacity_; }
SemiSpaceId id() { return id_; }
static void Swap(SemiSpace* from, SemiSpace* to);
// Approximate amount of physical memory committed for this space.
size_t CommittedPhysicalMemory();
private:
// Flips the semispace between being from-space and to-space.
// Copies the flags into the masked positions on all pages in the space.
void FlipPages(intptr_t flags, intptr_t flag_mask);
NewSpacePage* anchor() { return &anchor_; }
// The current and maximum capacity of the space.
int capacity_;
int maximum_capacity_;
int initial_capacity_;
// The start address of the space.
Address start_;
// Used to govern object promotion during mark-compact collection.
Address age_mark_;
// Masks and comparison values to test for containment in this semispace.
uintptr_t address_mask_;
uintptr_t object_mask_;
uintptr_t object_expected_;
bool committed_;
SemiSpaceId id_;
NewSpacePage anchor_;
NewSpacePage* current_page_;
friend class SemiSpaceIterator;
friend class NewSpacePageIterator;
public:
TRACK_MEMORY("SemiSpace")
};
// A SemiSpaceIterator is an ObjectIterator that iterates over the active
// semispace of the heap's new space. It iterates over the objects in the
// semispace from a given start address (defaulting to the bottom of the
// semispace) to the top of the semispace. New objects allocated after the
// iterator is created are not iterated.
class SemiSpaceIterator : public ObjectIterator {
public:
// Create an iterator over the objects in the given space. If no start
// address is given, the iterator starts from the bottom of the space. If
// no size function is given, the iterator calls Object::Size().
// Iterate over all of allocated to-space.
explicit SemiSpaceIterator(NewSpace* space);
// Iterate over all of allocated to-space, with a custome size function.
SemiSpaceIterator(NewSpace* space, HeapObjectCallback size_func);
// Iterate over part of allocated to-space, from start to the end
// of allocation.
SemiSpaceIterator(NewSpace* space, Address start);
// Iterate from one address to another in the same semi-space.
SemiSpaceIterator(Address from, Address to);
HeapObject* Next() {
if (current_ == limit_) return NULL;
if (NewSpacePage::IsAtEnd(current_)) {
NewSpacePage* page = NewSpacePage::FromLimit(current_);
page = page->next_page();
ASSERT(!page->is_anchor());
current_ = page->area_start();
if (current_ == limit_) return NULL;
}
HeapObject* object = HeapObject::FromAddress(current_);
int size = (size_func_ == NULL) ? object->Size() : size_func_(object);
current_ += size;
return object;
}
// Implementation of the ObjectIterator functions.
virtual HeapObject* next_object() { return Next(); }
private:
void Initialize(Address start,
Address end,
HeapObjectCallback size_func);
// The current iteration point.
Address current_;
// The end of iteration.
Address limit_;
// The callback function.
HeapObjectCallback size_func_;
};
// -----------------------------------------------------------------------------
// A PageIterator iterates the pages in a semi-space.
class NewSpacePageIterator BASE_EMBEDDED {
public:
// Make an iterator that runs over all pages in to-space.
explicit inline NewSpacePageIterator(NewSpace* space);
// Make an iterator that runs over all pages in the given semispace,
// even those not used in allocation.
explicit inline NewSpacePageIterator(SemiSpace* space);
// Make iterator that iterates from the page containing start
// to the page that contains limit in the same semispace.
inline NewSpacePageIterator(Address start, Address limit);
inline bool has_next();
inline NewSpacePage* next();
private:
NewSpacePage* prev_page_; // Previous page returned.
// Next page that will be returned. Cached here so that we can use this
// iterator for operations that deallocate pages.
NewSpacePage* next_page_;
// Last page returned.
NewSpacePage* last_page_;
};
// -----------------------------------------------------------------------------
// The young generation space.
//
// The new space consists of a contiguous pair of semispaces. It simply
// forwards most functions to the appropriate semispace.
class NewSpace : public Space {
public:
// Constructor.
explicit NewSpace(Heap* heap)
: Space(heap, NEW_SPACE, NOT_EXECUTABLE),
to_space_(heap, kToSpace),
from_space_(heap, kFromSpace),
reservation_(),
inline_allocation_limit_step_(0) {}
// Sets up the new space using the given chunk.
bool SetUp(int reserved_semispace_size_, int max_semispace_size);
// Tears down the space. Heap memory was not allocated by the space, so it
// is not deallocated here.
void TearDown();
// True if the space has been set up but not torn down.
bool HasBeenSetUp() {
return to_space_.HasBeenSetUp() && from_space_.HasBeenSetUp();
}
// Flip the pair of spaces.
void Flip();
// Grow the capacity of the semispaces. Assumes that they are not at
// their maximum capacity.
void Grow();
// Shrink the capacity of the semispaces.
void Shrink();
// True if the address or object lies in the address range of either
// semispace (not necessarily below the allocation pointer).
bool Contains(Address a) {
return (reinterpret_cast<uintptr_t>(a) & address_mask_)
== reinterpret_cast<uintptr_t>(start_);
}
bool Contains(Object* o) {
Address a = reinterpret_cast<Address>(o);
return (reinterpret_cast<uintptr_t>(a) & object_mask_) == object_expected_;
}
// Return the allocated bytes in the active semispace.
virtual intptr_t Size() {
return pages_used_ * NewSpacePage::kAreaSize +
static_cast<int>(top() - to_space_.page_low());
}
// The same, but returning an int. We have to have the one that returns
// intptr_t because it is inherited, but if we know we are dealing with the
// new space, which can't get as big as the other spaces then this is useful:
int SizeAsInt() { return static_cast<int>(Size()); }
// Return the current capacity of a semispace.
intptr_t EffectiveCapacity() {
SLOW_ASSERT(to_space_.Capacity() == from_space_.Capacity());
return (to_space_.Capacity() / Page::kPageSize) * NewSpacePage::kAreaSize;
}
// Return the current capacity of a semispace.
intptr_t Capacity() {
ASSERT(to_space_.Capacity() == from_space_.Capacity());
return to_space_.Capacity();
}
// Return the total amount of memory committed for new space.
intptr_t CommittedMemory() {
if (from_space_.is_committed()) return 2 * Capacity();
return Capacity();
}
// Approximate amount of physical memory committed for this space.
size_t CommittedPhysicalMemory();
// Return the available bytes without growing.
intptr_t Available() {
return Capacity() - Size();
}
// Return the maximum capacity of a semispace.
int MaximumCapacity() {
ASSERT(to_space_.MaximumCapacity() == from_space_.MaximumCapacity());
return to_space_.MaximumCapacity();
}
// Returns the initial capacity of a semispace.
int InitialCapacity() {
ASSERT(to_space_.InitialCapacity() == from_space_.InitialCapacity());
return to_space_.InitialCapacity();
}
// Return the address of the allocation pointer in the active semispace.
Address top() {
ASSERT(to_space_.current_page()->ContainsLimit(allocation_info_.top));
return allocation_info_.top;
}
// Return the address of the first object in the active semispace.
Address bottom() { return to_space_.space_start(); }
// Get the age mark of the inactive semispace.
Address age_mark() { return from_space_.age_mark(); }
// Set the age mark in the active semispace.
void set_age_mark(Address mark) { to_space_.set_age_mark(mark); }
// The start address of the space and a bit mask. Anding an address in the
// new space with the mask will result in the start address.
Address start() { return start_; }
uintptr_t mask() { return address_mask_; }
INLINE(uint32_t AddressToMarkbitIndex(Address addr)) {
ASSERT(Contains(addr));
ASSERT(IsAligned(OffsetFrom(addr), kPointerSize) ||
IsAligned(OffsetFrom(addr) - 1, kPointerSize));
return static_cast<uint32_t>(addr - start_) >> kPointerSizeLog2;
}
INLINE(Address MarkbitIndexToAddress(uint32_t index)) {
return reinterpret_cast<Address>(index << kPointerSizeLog2);
}
// The allocation top and limit addresses.
Address* allocation_top_address() { return &allocation_info_.top; }
Address* allocation_limit_address() { return &allocation_info_.limit; }
MUST_USE_RESULT INLINE(MaybeObject* AllocateRaw(int size_in_bytes));
// Reset the allocation pointer to the beginning of the active semispace.
void ResetAllocationInfo();
void LowerInlineAllocationLimit(intptr_t step) {
inline_allocation_limit_step_ = step;
if (step == 0) {
allocation_info_.limit = to_space_.page_high();
} else {
allocation_info_.limit = Min(
allocation_info_.top + inline_allocation_limit_step_,
allocation_info_.limit);
}
top_on_previous_step_ = allocation_info_.top;
}
// Get the extent of the inactive semispace (for use as a marking stack,
// or to zap it). Notice: space-addresses are not necessarily on the
// same page, so FromSpaceStart() might be above FromSpaceEnd().
Address FromSpacePageLow() { return from_space_.page_low(); }
Address FromSpacePageHigh() { return from_space_.page_high(); }
Address FromSpaceStart() { return from_space_.space_start(); }
Address FromSpaceEnd() { return from_space_.space_end(); }
// Get the extent of the active semispace's pages' memory.
Address ToSpaceStart() { return to_space_.space_start(); }
Address ToSpaceEnd() { return to_space_.space_end(); }
inline bool ToSpaceContains(Address address) {
return to_space_.Contains(address);
}
inline bool FromSpaceContains(Address address) {
return from_space_.Contains(address);
}
// True if the object is a heap object in the address range of the
// respective semispace (not necessarily below the allocation pointer of the
// semispace).
inline bool ToSpaceContains(Object* o) { return to_space_.Contains(o); }
inline bool FromSpaceContains(Object* o) { return from_space_.Contains(o); }
// Try to switch the active semispace to a new, empty, page.
// Returns false if this isn't possible or reasonable (i.e., there
// are no pages, or the current page is already empty), or true
// if successful.
bool AddFreshPage();
virtual bool ReserveSpace(int bytes);
#ifdef VERIFY_HEAP
// Verify the active semispace.
virtual void Verify();
#endif
#ifdef DEBUG
// Print the active semispace.
virtual void Print() { to_space_.Print(); }
#endif
// Iterates the active semispace to collect statistics.
void CollectStatistics();
// Reports previously collected statistics of the active semispace.
void ReportStatistics();
// Clears previously collected statistics.
void ClearHistograms();
// Record the allocation or promotion of a heap object. Note that we don't
// record every single allocation, but only those that happen in the
// to space during a scavenge GC.
void RecordAllocation(HeapObject* obj);
void RecordPromotion(HeapObject* obj);
// Return whether the operation succeded.
bool CommitFromSpaceIfNeeded() {
if (from_space_.is_committed()) return true;
return from_space_.Commit();
}
bool UncommitFromSpace() {
if (!from_space_.is_committed()) return true;
return from_space_.Uncommit();
}
inline intptr_t inline_allocation_limit_step() {
return inline_allocation_limit_step_;
}
SemiSpace* active_space() { return &to_space_; }
private:
// Update allocation info to match the current to-space page.
void UpdateAllocationInfo();
Address chunk_base_;
uintptr_t chunk_size_;
// The semispaces.
SemiSpace to_space_;
SemiSpace from_space_;
VirtualMemory reservation_;
int pages_used_;
// Start address and bit mask for containment testing.
Address start_;
uintptr_t address_mask_;
uintptr_t object_mask_;
uintptr_t object_expected_;
// Allocation pointer and limit for normal allocation and allocation during
// mark-compact collection.
AllocationInfo allocation_info_;
// When incremental marking is active we will set allocation_info_.limit
// to be lower than actual limit and then will gradually increase it
// in steps to guarantee that we do incremental marking steps even
// when all allocation is performed from inlined generated code.
intptr_t inline_allocation_limit_step_;
Address top_on_previous_step_;
HistogramInfo* allocated_histogram_;
HistogramInfo* promoted_histogram_;
MUST_USE_RESULT MaybeObject* SlowAllocateRaw(int size_in_bytes);
friend class SemiSpaceIterator;
public:
TRACK_MEMORY("NewSpace")
};
// -----------------------------------------------------------------------------
// Old object space (excluding map objects)
class OldSpace : public PagedSpace {
public:
// Creates an old space object with a given maximum capacity.
// The constructor does not allocate pages from OS.
OldSpace(Heap* heap,
intptr_t max_capacity,
AllocationSpace id,
Executability executable)
: PagedSpace(heap, max_capacity, id, executable) {
page_extra_ = 0;
}
// The limit of allocation for a page in this space.
virtual Address PageAllocationLimit(Page* page) {
return page->area_end();
}
public:
TRACK_MEMORY("OldSpace")
};
// For contiguous spaces, top should be in the space (or at the end) and limit
// should be the end of the space.
#define ASSERT_SEMISPACE_ALLOCATION_INFO(info, space) \
SLOW_ASSERT((space).page_low() <= (info).top \
&& (info).top <= (space).page_high() \
&& (info).limit <= (space).page_high())
// -----------------------------------------------------------------------------
// Old space for objects of a fixed size
class FixedSpace : public PagedSpace {
public:
FixedSpace(Heap* heap,
intptr_t max_capacity,
AllocationSpace id,
int object_size_in_bytes)
: PagedSpace(heap, max_capacity, id, NOT_EXECUTABLE),
object_size_in_bytes_(object_size_in_bytes) {
page_extra_ = Page::kNonCodeObjectAreaSize % object_size_in_bytes;
}
// The limit of allocation for a page in this space.
virtual Address PageAllocationLimit(Page* page) {
return page->area_end() - page_extra_;
}
int object_size_in_bytes() { return object_size_in_bytes_; }
// Prepares for a mark-compact GC.
virtual void PrepareForMarkCompact();
private:
// The size of objects in this space.
int object_size_in_bytes_;
};
// -----------------------------------------------------------------------------
// Old space for all map objects
class MapSpace : public FixedSpace {
public:
// Creates a map space object with a maximum capacity.
MapSpace(Heap* heap, intptr_t max_capacity, AllocationSpace id)
: FixedSpace(heap, max_capacity, id, Map::kSize),
max_map_space_pages_(kMaxMapPageIndex - 1) {
}
// Given an index, returns the page address.
// TODO(1600): this limit is artifical just to keep code compilable
static const int kMaxMapPageIndex = 1 << 16;
virtual int RoundSizeDownToObjectAlignment(int size) {
if (IsPowerOf2(Map::kSize)) {
return RoundDown(size, Map::kSize);
} else {
return (size / Map::kSize) * Map::kSize;
}
}
protected:
virtual void VerifyObject(HeapObject* obj);
private:
static const int kMapsPerPage = Page::kNonCodeObjectAreaSize / Map::kSize;
// Do map space compaction if there is a page gap.
int CompactionThreshold() {
return kMapsPerPage * (max_map_space_pages_ - 1);
}
const int max_map_space_pages_;
public:
TRACK_MEMORY("MapSpace")
};
// -----------------------------------------------------------------------------
// Old space for all global object property cell objects
class CellSpace : public FixedSpace {
public:
// Creates a property cell space object with a maximum capacity.
CellSpace(Heap* heap, intptr_t max_capacity, AllocationSpace id)
: FixedSpace(heap, max_capacity, id, JSGlobalPropertyCell::kSize)
{}
virtual int RoundSizeDownToObjectAlignment(int size) {
if (IsPowerOf2(JSGlobalPropertyCell::kSize)) {
return RoundDown(size, JSGlobalPropertyCell::kSize);
} else {
return (size / JSGlobalPropertyCell::kSize) * JSGlobalPropertyCell::kSize;
}
}
protected:
virtual void VerifyObject(HeapObject* obj);
public:
TRACK_MEMORY("CellSpace")
};
// -----------------------------------------------------------------------------
// Large objects ( > Page::kMaxHeapObjectSize ) are allocated and managed by
// the large object space. A large object is allocated from OS heap with
// extra padding bytes (Page::kPageSize + Page::kObjectStartOffset).
// A large object always starts at Page::kObjectStartOffset to a page.
// Large objects do not move during garbage collections.
class LargeObjectSpace : public Space {
public:
LargeObjectSpace(Heap* heap, intptr_t max_capacity, AllocationSpace id);
virtual ~LargeObjectSpace() {}
// Initializes internal data structures.
bool SetUp();
// Releases internal resources, frees objects in this space.
void TearDown();
static intptr_t ObjectSizeFor(intptr_t chunk_size) {
if (chunk_size <= (Page::kPageSize + Page::kObjectStartOffset)) return 0;
return chunk_size - Page::kPageSize - Page::kObjectStartOffset;
}
// Shared implementation of AllocateRaw, AllocateRawCode and
// AllocateRawFixedArray.
MUST_USE_RESULT MaybeObject* AllocateRaw(int object_size,
Executability executable);
// Available bytes for objects in this space.
inline intptr_t Available();
virtual intptr_t Size() {
return size_;
}
virtual intptr_t SizeOfObjects() {
return objects_size_;
}
intptr_t CommittedMemory() {
return Size();
}
// Approximate amount of physical memory committed for this space.
size_t CommittedPhysicalMemory();
int PageCount() {
return page_count_;
}
// Finds an object for a given address, returns Failure::Exception()
// if it is not found. The function iterates through all objects in this
// space, may be slow.
MaybeObject* FindObject(Address a);
// Finds a large object page containing the given address, returns NULL
// if such a page doesn't exist.
LargePage* FindPage(Address a);
// Frees unmarked objects.
void FreeUnmarkedObjects();
// Checks whether a heap object is in this space; O(1).
bool Contains(HeapObject* obj);
// Checks whether the space is empty.
bool IsEmpty() { return first_page_ == NULL; }
// See the comments for ReserveSpace in the Space class. This has to be
// called after ReserveSpace has been called on the paged spaces, since they
// may use some memory, leaving less for large objects.
virtual bool ReserveSpace(int bytes);
LargePage* first_page() { return first_page_; }
#ifdef VERIFY_HEAP
virtual void Verify();
#endif
#ifdef DEBUG
virtual void Print();
void ReportStatistics();
void CollectCodeStatistics();
#endif
// Checks whether an address is in the object area in this space. It
// iterates all objects in the space. May be slow.
bool SlowContains(Address addr) { return !FindObject(addr)->IsFailure(); }
private:
intptr_t max_capacity_;
// The head of the linked list of large object chunks.
LargePage* first_page_;
intptr_t size_; // allocated bytes
int page_count_; // number of chunks
intptr_t objects_size_; // size of objects
// Map MemoryChunk::kAlignment-aligned chunks to large pages covering them
HashMap chunk_map_;
friend class LargeObjectIterator;
public:
TRACK_MEMORY("LargeObjectSpace")
};
class LargeObjectIterator: public ObjectIterator {
public:
explicit LargeObjectIterator(LargeObjectSpace* space);
LargeObjectIterator(LargeObjectSpace* space, HeapObjectCallback size_func);
HeapObject* Next();
// implementation of ObjectIterator.
virtual HeapObject* next_object() { return Next(); }
private:
LargePage* current_;
HeapObjectCallback size_func_;
};
// Iterates over the chunks (pages and large object pages) that can contain
// pointers to new space.
class PointerChunkIterator BASE_EMBEDDED {
public:
inline explicit PointerChunkIterator(Heap* heap);
// Return NULL when the iterator is done.
MemoryChunk* next() {
switch (state_) {
case kOldPointerState: {
if (old_pointer_iterator_.has_next()) {
return old_pointer_iterator_.next();
}
state_ = kMapState;
// Fall through.
}
case kMapState: {
if (map_iterator_.has_next()) {
return map_iterator_.next();
}
state_ = kLargeObjectState;
// Fall through.
}
case kLargeObjectState: {
HeapObject* heap_object;
do {
heap_object = lo_iterator_.Next();
if (heap_object == NULL) {
state_ = kFinishedState;
return NULL;
}
// Fixed arrays are the only pointer-containing objects in large
// object space.
} while (!heap_object->IsFixedArray());
MemoryChunk* answer = MemoryChunk::FromAddress(heap_object->address());
return answer;
}
case kFinishedState:
return NULL;
default:
break;
}
UNREACHABLE();
return NULL;
}
private:
enum State {
kOldPointerState,
kMapState,
kLargeObjectState,
kFinishedState
};
State state_;
PageIterator old_pointer_iterator_;
PageIterator map_iterator_;
LargeObjectIterator lo_iterator_;
};
#ifdef DEBUG
struct CommentStatistic {
const char* comment;
int size;
int count;
void Clear() {
comment = NULL;
size = 0;
count = 0;
}
// Must be small, since an iteration is used for lookup.
static const int kMaxComments = 64;
};
#endif
} } // namespace v8::internal
#endif // V8_SPACES_H_