1 // Copyright 2011 the V8 project authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #ifndef V8_HEAP_SPACES_H_ 6 #define V8_HEAP_SPACES_H_ 7 8 #include <list> 9 10 #include "src/allocation.h" 11 #include "src/base/atomic-utils.h" 12 #include "src/base/atomicops.h" 13 #include "src/base/bits.h" 14 #include "src/base/hashmap.h" 15 #include "src/base/platform/mutex.h" 16 #include "src/flags.h" 17 #include "src/list.h" 18 #include "src/objects.h" 19 #include "src/utils.h" 20 21 namespace v8 { 22 namespace internal { 23 24 class AllocationInfo; 25 class AllocationObserver; 26 class CompactionSpace; 27 class CompactionSpaceCollection; 28 class FreeList; 29 class Isolate; 30 class LocalArrayBufferTracker; 31 class MemoryAllocator; 32 class MemoryChunk; 33 class Page; 34 class PagedSpace; 35 class SemiSpace; 36 class SkipList; 37 class SlotsBuffer; 38 class SlotSet; 39 class TypedSlotSet; 40 class Space; 41 42 // ----------------------------------------------------------------------------- 43 // Heap structures: 44 // 45 // A JS heap consists of a young generation, an old generation, and a large 46 // object space. The young generation is divided into two semispaces. A 47 // scavenger implements Cheney's copying algorithm. The old generation is 48 // separated into a map space and an old object space. The map space contains 49 // all (and only) map objects, the rest of old objects go into the old space. 50 // The old generation is collected by a mark-sweep-compact collector. 51 // 52 // The semispaces of the young generation are contiguous. The old and map 53 // spaces consists of a list of pages. A page has a page header and an object 54 // area. 55 // 56 // There is a separate large object space for objects larger than 57 // Page::kMaxRegularHeapObjectSize, so that they do not have to move during 58 // collection. The large object space is paged. Pages in large object space 59 // may be larger than the page size. 60 // 61 // A store-buffer based write barrier is used to keep track of intergenerational 62 // references. See heap/store-buffer.h. 63 // 64 // During scavenges and mark-sweep collections we sometimes (after a store 65 // buffer overflow) iterate intergenerational pointers without decoding heap 66 // object maps so if the page belongs to old space or large object space 67 // it is essential to guarantee that the page does not contain any 68 // garbage pointers to new space: every pointer aligned word which satisfies 69 // the Heap::InNewSpace() predicate must be a pointer to a live heap object in 70 // new space. Thus objects in old space and large object spaces should have a 71 // special layout (e.g. no bare integer fields). This requirement does not 72 // apply to map space which is iterated in a special fashion. However we still 73 // require pointer fields of dead maps to be cleaned. 74 // 75 // To enable lazy cleaning of old space pages we can mark chunks of the page 76 // as being garbage. Garbage sections are marked with a special map. These 77 // sections are skipped when scanning the page, even if we are otherwise 78 // scanning without regard for object boundaries. Garbage sections are chained 79 // together to form a free list after a GC. Garbage sections created outside 80 // of GCs by object trunctation etc. may not be in the free list chain. Very 81 // small free spaces are ignored, they need only be cleaned of bogus pointers 82 // into new space. 83 // 84 // Each page may have up to one special garbage section. The start of this 85 // section is denoted by the top field in the space. The end of the section 86 // is denoted by the limit field in the space. This special garbage section 87 // is not marked with a free space map in the data. The point of this section 88 // is to enable linear allocation without having to constantly update the byte 89 // array every time the top field is updated and a new object is created. The 90 // special garbage section is not in the chain of garbage sections. 91 // 92 // Since the top and limit fields are in the space, not the page, only one page 93 // has a special garbage section, and if the top and limit are equal then there 94 // is no special garbage section. 95 96 // Some assertion macros used in the debugging mode. 97 98 #define DCHECK_PAGE_ALIGNED(address) \ 99 DCHECK((OffsetFrom(address) & Page::kPageAlignmentMask) == 0) 100 101 #define DCHECK_OBJECT_ALIGNED(address) \ 102 DCHECK((OffsetFrom(address) & kObjectAlignmentMask) == 0) 103 104 #define DCHECK_OBJECT_SIZE(size) \ 105 DCHECK((0 < size) && (size <= Page::kMaxRegularHeapObjectSize)) 106 107 #define DCHECK_CODEOBJECT_SIZE(size, code_space) \ 108 DCHECK((0 < size) && (size <= code_space->AreaSize())) 109 110 #define DCHECK_PAGE_OFFSET(offset) \ 111 DCHECK((Page::kObjectStartOffset <= offset) && (offset <= Page::kPageSize)) 112 113 class MarkBit { 114 public: 115 typedef uint32_t CellType; 116 MarkBit(CellType * cell,CellType mask)117 inline MarkBit(CellType* cell, CellType mask) : cell_(cell), mask_(mask) {} 118 119 #ifdef DEBUG 120 bool operator==(const MarkBit& other) { 121 return cell_ == other.cell_ && mask_ == other.mask_; 122 } 123 #endif 124 125 private: cell()126 inline CellType* cell() { return cell_; } mask()127 inline CellType mask() { return mask_; } 128 Next()129 inline MarkBit Next() { 130 CellType new_mask = mask_ << 1; 131 if (new_mask == 0) { 132 return MarkBit(cell_ + 1, 1); 133 } else { 134 return MarkBit(cell_, new_mask); 135 } 136 } 137 Set()138 inline void Set() { *cell_ |= mask_; } Get()139 inline bool Get() { return (*cell_ & mask_) != 0; } Clear()140 inline void Clear() { *cell_ &= ~mask_; } 141 142 CellType* cell_; 143 CellType mask_; 144 145 friend class Marking; 146 }; 147 148 149 // Bitmap is a sequence of cells each containing fixed number of bits. 150 class Bitmap { 151 public: 152 static const uint32_t kBitsPerCell = 32; 153 static const uint32_t kBitsPerCellLog2 = 5; 154 static const uint32_t kBitIndexMask = kBitsPerCell - 1; 155 static const uint32_t kBytesPerCell = kBitsPerCell / kBitsPerByte; 156 static const uint32_t kBytesPerCellLog2 = kBitsPerCellLog2 - kBitsPerByteLog2; 157 158 static const size_t kLength = (1 << kPageSizeBits) >> (kPointerSizeLog2); 159 160 static const size_t kSize = 161 (1 << kPageSizeBits) >> (kPointerSizeLog2 + kBitsPerByteLog2); 162 163 CellsForLength(int length)164 static int CellsForLength(int length) { 165 return (length + kBitsPerCell - 1) >> kBitsPerCellLog2; 166 } 167 CellsCount()168 int CellsCount() { return CellsForLength(kLength); } 169 SizeFor(int cells_count)170 static int SizeFor(int cells_count) { 171 return sizeof(MarkBit::CellType) * cells_count; 172 } 173 INLINE(static uint32_t IndexToCell (uint32_t index))174 INLINE(static uint32_t IndexToCell(uint32_t index)) { 175 return index >> kBitsPerCellLog2; 176 } 177 IndexInCell(uint32_t index)178 V8_INLINE static uint32_t IndexInCell(uint32_t index) { 179 return index & kBitIndexMask; 180 } 181 INLINE(static uint32_t CellToIndex (uint32_t index))182 INLINE(static uint32_t CellToIndex(uint32_t index)) { 183 return index << kBitsPerCellLog2; 184 } 185 INLINE(static uint32_t CellAlignIndex (uint32_t index))186 INLINE(static uint32_t CellAlignIndex(uint32_t index)) { 187 return (index + kBitIndexMask) & ~kBitIndexMask; 188 } 189 INLINE(MarkBit::CellType * cells ())190 INLINE(MarkBit::CellType* cells()) { 191 return reinterpret_cast<MarkBit::CellType*>(this); 192 } 193 INLINE(Address address ())194 INLINE(Address address()) { return reinterpret_cast<Address>(this); } 195 INLINE(static Bitmap * FromAddress (Address addr))196 INLINE(static Bitmap* FromAddress(Address addr)) { 197 return reinterpret_cast<Bitmap*>(addr); 198 } 199 MarkBitFromIndex(uint32_t index)200 inline MarkBit MarkBitFromIndex(uint32_t index) { 201 MarkBit::CellType mask = 1u << IndexInCell(index); 202 MarkBit::CellType* cell = this->cells() + (index >> kBitsPerCellLog2); 203 return MarkBit(cell, mask); 204 } 205 206 static inline void Clear(MemoryChunk* chunk); 207 208 static inline void SetAllBits(MemoryChunk* chunk); 209 210 static void PrintWord(uint32_t word, uint32_t himask = 0) { 211 for (uint32_t mask = 1; mask != 0; mask <<= 1) { 212 if ((mask & himask) != 0) PrintF("["); 213 PrintF((mask & word) ? "1" : "0"); 214 if ((mask & himask) != 0) PrintF("]"); 215 } 216 } 217 218 class CellPrinter { 219 public: CellPrinter()220 CellPrinter() : seq_start(0), seq_type(0), seq_length(0) {} 221 Print(uint32_t pos,uint32_t cell)222 void Print(uint32_t pos, uint32_t cell) { 223 if (cell == seq_type) { 224 seq_length++; 225 return; 226 } 227 228 Flush(); 229 230 if (IsSeq(cell)) { 231 seq_start = pos; 232 seq_length = 0; 233 seq_type = cell; 234 return; 235 } 236 237 PrintF("%d: ", pos); 238 PrintWord(cell); 239 PrintF("\n"); 240 } 241 Flush()242 void Flush() { 243 if (seq_length > 0) { 244 PrintF("%d: %dx%d\n", seq_start, seq_type == 0 ? 0 : 1, 245 seq_length * kBitsPerCell); 246 seq_length = 0; 247 } 248 } 249 IsSeq(uint32_t cell)250 static bool IsSeq(uint32_t cell) { return cell == 0 || cell == 0xFFFFFFFF; } 251 252 private: 253 uint32_t seq_start; 254 uint32_t seq_type; 255 uint32_t seq_length; 256 }; 257 Print()258 void Print() { 259 CellPrinter printer; 260 for (int i = 0; i < CellsCount(); i++) { 261 printer.Print(i, cells()[i]); 262 } 263 printer.Flush(); 264 PrintF("\n"); 265 } 266 IsClean()267 bool IsClean() { 268 for (int i = 0; i < CellsCount(); i++) { 269 if (cells()[i] != 0) { 270 return false; 271 } 272 } 273 return true; 274 } 275 276 // Clears all bits starting from {cell_base_index} up to and excluding 277 // {index}. Note that {cell_base_index} is required to be cell aligned. ClearRange(uint32_t cell_base_index,uint32_t index)278 void ClearRange(uint32_t cell_base_index, uint32_t index) { 279 DCHECK_EQ(IndexInCell(cell_base_index), 0u); 280 DCHECK_GE(index, cell_base_index); 281 uint32_t start_cell_index = IndexToCell(cell_base_index); 282 uint32_t end_cell_index = IndexToCell(index); 283 DCHECK_GE(end_cell_index, start_cell_index); 284 // Clear all cells till the cell containing the last index. 285 for (uint32_t i = start_cell_index; i < end_cell_index; i++) { 286 cells()[i] = 0; 287 } 288 // Clear all bits in the last cell till the last bit before index. 289 uint32_t clear_mask = ~((1u << IndexInCell(index)) - 1); 290 cells()[end_cell_index] &= clear_mask; 291 } 292 }; 293 294 enum FreeListCategoryType { 295 kTiniest, 296 kTiny, 297 kSmall, 298 kMedium, 299 kLarge, 300 kHuge, 301 302 kFirstCategory = kTiniest, 303 kLastCategory = kHuge, 304 kNumberOfCategories = kLastCategory + 1, 305 kInvalidCategory 306 }; 307 308 enum FreeMode { kLinkCategory, kDoNotLinkCategory }; 309 310 // A free list category maintains a linked list of free memory blocks. 311 class FreeListCategory { 312 public: 313 static const int kSize = kIntSize + // FreeListCategoryType type_ 314 kIntSize + // int available_ 315 kPointerSize + // FreeSpace* top_ 316 kPointerSize + // FreeListCategory* prev_ 317 kPointerSize; // FreeListCategory* next_ 318 FreeListCategory()319 FreeListCategory() 320 : type_(kInvalidCategory), 321 available_(0), 322 top_(nullptr), 323 prev_(nullptr), 324 next_(nullptr) {} 325 Initialize(FreeListCategoryType type)326 void Initialize(FreeListCategoryType type) { 327 type_ = type; 328 available_ = 0; 329 top_ = nullptr; 330 prev_ = nullptr; 331 next_ = nullptr; 332 } 333 334 void Invalidate(); 335 336 void Reset(); 337 ResetStats()338 void ResetStats() { Reset(); } 339 340 void RepairFreeList(Heap* heap); 341 342 // Relinks the category into the currently owning free list. Requires that the 343 // category is currently unlinked. 344 void Relink(); 345 346 bool Free(FreeSpace* node, int size_in_bytes, FreeMode mode); 347 348 // Picks a node from the list and stores its size in |node_size|. Returns 349 // nullptr if the category is empty. 350 FreeSpace* PickNodeFromList(int* node_size); 351 352 // Performs a single try to pick a node of at least |minimum_size| from the 353 // category. Stores the actual size in |node_size|. Returns nullptr if no 354 // node is found. 355 FreeSpace* TryPickNodeFromList(int minimum_size, int* node_size); 356 357 // Picks a node of at least |minimum_size| from the category. Stores the 358 // actual size in |node_size|. Returns nullptr if no node is found. 359 FreeSpace* SearchForNodeInList(int minimum_size, int* node_size); 360 361 inline FreeList* owner(); 362 inline bool is_linked(); is_empty()363 bool is_empty() { return top() == nullptr; } available()364 int available() const { return available_; } 365 366 #ifdef DEBUG 367 intptr_t SumFreeList(); 368 int FreeListLength(); 369 #endif 370 371 private: 372 // For debug builds we accurately compute free lists lengths up until 373 // {kVeryLongFreeList} by manually walking the list. 374 static const int kVeryLongFreeList = 500; 375 376 inline Page* page(); 377 top()378 FreeSpace* top() { return top_; } set_top(FreeSpace * top)379 void set_top(FreeSpace* top) { top_ = top; } prev()380 FreeListCategory* prev() { return prev_; } set_prev(FreeListCategory * prev)381 void set_prev(FreeListCategory* prev) { prev_ = prev; } next()382 FreeListCategory* next() { return next_; } set_next(FreeListCategory * next)383 void set_next(FreeListCategory* next) { next_ = next; } 384 385 // |type_|: The type of this free list category. 386 FreeListCategoryType type_; 387 388 // |available_|: Total available bytes in all blocks of this free list 389 // category. 390 int available_; 391 392 // |top_|: Points to the top FreeSpace* in the free list category. 393 FreeSpace* top_; 394 395 FreeListCategory* prev_; 396 FreeListCategory* next_; 397 398 friend class FreeList; 399 friend class PagedSpace; 400 }; 401 402 // MemoryChunk represents a memory region owned by a specific space. 403 // It is divided into the header and the body. Chunk start is always 404 // 1MB aligned. Start of the body is aligned so it can accommodate 405 // any heap object. 406 class MemoryChunk { 407 public: 408 enum MemoryChunkFlags { 409 IS_EXECUTABLE, 410 POINTERS_TO_HERE_ARE_INTERESTING, 411 POINTERS_FROM_HERE_ARE_INTERESTING, 412 IN_FROM_SPACE, // Mutually exclusive with IN_TO_SPACE. 413 IN_TO_SPACE, // All pages in new space has one of these two set. 414 NEW_SPACE_BELOW_AGE_MARK, 415 EVACUATION_CANDIDATE, 416 NEVER_EVACUATE, // May contain immortal immutables. 417 418 // Large objects can have a progress bar in their page header. These object 419 // are scanned in increments and will be kept black while being scanned. 420 // Even if the mutator writes to them they will be kept black and a white 421 // to grey transition is performed in the value. 422 HAS_PROGRESS_BAR, 423 424 // |PAGE_NEW_OLD_PROMOTION|: A page tagged with this flag has been promoted 425 // from new to old space during evacuation. 426 PAGE_NEW_OLD_PROMOTION, 427 428 // |PAGE_NEW_NEW_PROMOTION|: A page tagged with this flag has been moved 429 // within the new space during evacuation. 430 PAGE_NEW_NEW_PROMOTION, 431 432 // A black page has all mark bits set to 1 (black). A black page currently 433 // cannot be iterated because it is not swept. Moreover live bytes are also 434 // not updated. 435 BLACK_PAGE, 436 437 // This flag is intended to be used for testing. Works only when both 438 // FLAG_stress_compaction and FLAG_manual_evacuation_candidates_selection 439 // are set. It forces the page to become an evacuation candidate at next 440 // candidates selection cycle. 441 FORCE_EVACUATION_CANDIDATE_FOR_TESTING, 442 443 // This flag is intended to be used for testing. 444 NEVER_ALLOCATE_ON_PAGE, 445 446 // The memory chunk is already logically freed, however the actual freeing 447 // still has to be performed. 448 PRE_FREED, 449 450 // |POOLED|: When actually freeing this chunk, only uncommit and do not 451 // give up the reservation as we still reuse the chunk at some point. 452 POOLED, 453 454 // |COMPACTION_WAS_ABORTED|: Indicates that the compaction in this page 455 // has been aborted and needs special handling by the sweeper. 456 COMPACTION_WAS_ABORTED, 457 458 // |COMPACTION_WAS_ABORTED_FOR_TESTING|: During stress testing evacuation 459 // on pages is sometimes aborted. The flag is used to avoid repeatedly 460 // triggering on the same page. 461 COMPACTION_WAS_ABORTED_FOR_TESTING, 462 463 // |ANCHOR|: Flag is set if page is an anchor. 464 ANCHOR, 465 466 // Last flag, keep at bottom. 467 NUM_MEMORY_CHUNK_FLAGS 468 }; 469 470 // |kSweepingDone|: The page state when sweeping is complete or sweeping must 471 // not be performed on that page. Sweeper threads that are done with their 472 // work will set this value and not touch the page anymore. 473 // |kSweepingPending|: This page is ready for parallel sweeping. 474 // |kSweepingInProgress|: This page is currently swept by a sweeper thread. 475 enum ConcurrentSweepingState { 476 kSweepingDone, 477 kSweepingPending, 478 kSweepingInProgress, 479 }; 480 481 // Every n write barrier invocations we go to runtime even though 482 // we could have handled it in generated code. This lets us check 483 // whether we have hit the limit and should do some more marking. 484 static const int kWriteBarrierCounterGranularity = 500; 485 486 static const int kPointersToHereAreInterestingMask = 487 1 << POINTERS_TO_HERE_ARE_INTERESTING; 488 489 static const int kPointersFromHereAreInterestingMask = 490 1 << POINTERS_FROM_HERE_ARE_INTERESTING; 491 492 static const int kEvacuationCandidateMask = 1 << EVACUATION_CANDIDATE; 493 494 static const int kSkipEvacuationSlotsRecordingMask = 495 (1 << EVACUATION_CANDIDATE) | (1 << IN_FROM_SPACE) | (1 << IN_TO_SPACE); 496 497 static const intptr_t kAlignment = 498 (static_cast<uintptr_t>(1) << kPageSizeBits); 499 500 static const intptr_t kAlignmentMask = kAlignment - 1; 501 502 static const intptr_t kSizeOffset = 0; 503 504 static const intptr_t kFlagsOffset = kSizeOffset + kPointerSize; 505 506 static const intptr_t kLiveBytesOffset = 507 kSizeOffset + kPointerSize // size_t size 508 + kIntptrSize // intptr_t flags_ 509 + kPointerSize // Address area_start_ 510 + kPointerSize // Address area_end_ 511 + 2 * kPointerSize // base::VirtualMemory reservation_ 512 + kPointerSize // Address owner_ 513 + kPointerSize // Heap* heap_ 514 + kIntSize; // int progress_bar_ 515 516 static const size_t kOldToNewSlotsOffset = 517 kLiveBytesOffset + kIntSize; // int live_byte_count_ 518 519 static const size_t kWriteBarrierCounterOffset = 520 kOldToNewSlotsOffset + kPointerSize // SlotSet* old_to_new_slots_; 521 + kPointerSize // SlotSet* old_to_old_slots_; 522 + kPointerSize // TypedSlotSet* typed_old_to_new_slots_; 523 + kPointerSize // TypedSlotSet* typed_old_to_old_slots_; 524 + kPointerSize; // SkipList* skip_list_; 525 526 static const size_t kMinHeaderSize = 527 kWriteBarrierCounterOffset + 528 kIntptrSize // intptr_t write_barrier_counter_ 529 + kPointerSize // AtomicValue high_water_mark_ 530 + kPointerSize // base::Mutex* mutex_ 531 + kPointerSize // base::AtomicWord concurrent_sweeping_ 532 + 2 * kPointerSize // AtomicNumber free-list statistics 533 + kPointerSize // AtomicValue next_chunk_ 534 + kPointerSize // AtomicValue prev_chunk_ 535 // FreeListCategory categories_[kNumberOfCategories] 536 + FreeListCategory::kSize * kNumberOfCategories + 537 kPointerSize; // LocalArrayBufferTracker* local_tracker_; 538 539 // We add some more space to the computed header size to amount for missing 540 // alignment requirements in our computation. 541 // Try to get kHeaderSize properly aligned on 32-bit and 64-bit machines. 542 static const size_t kHeaderSize = kMinHeaderSize; 543 544 static const int kBodyOffset = 545 CODE_POINTER_ALIGN(kHeaderSize + Bitmap::kSize); 546 547 // The start offset of the object area in a page. Aligned to both maps and 548 // code alignment to be suitable for both. Also aligned to 32 words because 549 // the marking bitmap is arranged in 32 bit chunks. 550 static const int kObjectStartAlignment = 32 * kPointerSize; 551 static const int kObjectStartOffset = 552 kBodyOffset - 1 + 553 (kObjectStartAlignment - (kBodyOffset - 1) % kObjectStartAlignment); 554 555 // Page size in bytes. This must be a multiple of the OS page size. 556 static const int kPageSize = 1 << kPageSizeBits; 557 static const intptr_t kPageAlignmentMask = (1 << kPageSizeBits) - 1; 558 559 static const int kAllocatableMemory = kPageSize - kObjectStartOffset; 560 561 static inline void IncrementLiveBytesFromMutator(HeapObject* object, int by); 562 static inline void IncrementLiveBytesFromGC(HeapObject* object, int by); 563 564 // Only works if the pointer is in the first kPageSize of the MemoryChunk. FromAddress(Address a)565 static MemoryChunk* FromAddress(Address a) { 566 return reinterpret_cast<MemoryChunk*>(OffsetFrom(a) & ~kAlignmentMask); 567 } 568 569 static inline MemoryChunk* FromAnyPointerAddress(Heap* heap, Address addr); 570 UpdateHighWaterMark(Address mark)571 static inline void UpdateHighWaterMark(Address mark) { 572 if (mark == nullptr) return; 573 // Need to subtract one from the mark because when a chunk is full the 574 // top points to the next address after the chunk, which effectively belongs 575 // to another chunk. See the comment to Page::FromTopOrLimit. 576 MemoryChunk* chunk = MemoryChunk::FromAddress(mark - 1); 577 intptr_t new_mark = static_cast<intptr_t>(mark - chunk->address()); 578 intptr_t old_mark = 0; 579 do { 580 old_mark = chunk->high_water_mark_.Value(); 581 } while ((new_mark > old_mark) && 582 !chunk->high_water_mark_.TrySetValue(old_mark, new_mark)); 583 } 584 IsValid(MemoryChunk * chunk)585 static bool IsValid(MemoryChunk* chunk) { return chunk != nullptr; } 586 address()587 Address address() { return reinterpret_cast<Address>(this); } 588 mutex()589 base::Mutex* mutex() { return mutex_; } 590 Contains(Address addr)591 bool Contains(Address addr) { 592 return addr >= area_start() && addr < area_end(); 593 } 594 595 // Checks whether |addr| can be a limit of addresses in this page. It's a 596 // limit if it's in the page, or if it's just after the last byte of the page. ContainsLimit(Address addr)597 bool ContainsLimit(Address addr) { 598 return addr >= area_start() && addr <= area_end(); 599 } 600 concurrent_sweeping_state()601 base::AtomicValue<ConcurrentSweepingState>& concurrent_sweeping_state() { 602 return concurrent_sweeping_; 603 } 604 605 // Manage live byte count, i.e., count of bytes in black objects. 606 inline void ResetLiveBytes(); 607 inline void IncrementLiveBytes(int by); 608 LiveBytes()609 int LiveBytes() { 610 DCHECK_LE(static_cast<unsigned>(live_byte_count_), size_); 611 DCHECK(!IsFlagSet(BLACK_PAGE) || live_byte_count_ == 0); 612 return live_byte_count_; 613 } 614 SetLiveBytes(int live_bytes)615 void SetLiveBytes(int live_bytes) { 616 if (IsFlagSet(BLACK_PAGE)) return; 617 DCHECK_GE(live_bytes, 0); 618 DCHECK_LE(static_cast<size_t>(live_bytes), size_); 619 live_byte_count_ = live_bytes; 620 } 621 write_barrier_counter()622 int write_barrier_counter() { 623 return static_cast<int>(write_barrier_counter_); 624 } 625 set_write_barrier_counter(int counter)626 void set_write_barrier_counter(int counter) { 627 write_barrier_counter_ = counter; 628 } 629 size()630 size_t size() const { return size_; } 631 heap()632 inline Heap* heap() const { return heap_; } 633 skip_list()634 inline SkipList* skip_list() { return skip_list_; } 635 set_skip_list(SkipList * skip_list)636 inline void set_skip_list(SkipList* skip_list) { skip_list_ = skip_list; } 637 old_to_new_slots()638 inline SlotSet* old_to_new_slots() { return old_to_new_slots_; } old_to_old_slots()639 inline SlotSet* old_to_old_slots() { return old_to_old_slots_; } typed_old_to_new_slots()640 inline TypedSlotSet* typed_old_to_new_slots() { 641 return typed_old_to_new_slots_; 642 } typed_old_to_old_slots()643 inline TypedSlotSet* typed_old_to_old_slots() { 644 return typed_old_to_old_slots_; 645 } local_tracker()646 inline LocalArrayBufferTracker* local_tracker() { return local_tracker_; } 647 648 void AllocateOldToNewSlots(); 649 void ReleaseOldToNewSlots(); 650 void AllocateOldToOldSlots(); 651 void ReleaseOldToOldSlots(); 652 void AllocateTypedOldToNewSlots(); 653 void ReleaseTypedOldToNewSlots(); 654 void AllocateTypedOldToOldSlots(); 655 void ReleaseTypedOldToOldSlots(); 656 void AllocateLocalTracker(); 657 void ReleaseLocalTracker(); 658 area_start()659 Address area_start() { return area_start_; } area_end()660 Address area_end() { return area_end_; } area_size()661 int area_size() { return static_cast<int>(area_end() - area_start()); } 662 663 bool CommitArea(size_t requested); 664 665 // Approximate amount of physical memory committed for this chunk. CommittedPhysicalMemory()666 size_t CommittedPhysicalMemory() { return high_water_mark_.Value(); } 667 HighWaterMark()668 Address HighWaterMark() { return address() + high_water_mark_.Value(); } 669 progress_bar()670 int progress_bar() { 671 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 672 return progress_bar_; 673 } 674 set_progress_bar(int progress_bar)675 void set_progress_bar(int progress_bar) { 676 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 677 progress_bar_ = progress_bar; 678 } 679 ResetProgressBar()680 void ResetProgressBar() { 681 if (IsFlagSet(MemoryChunk::HAS_PROGRESS_BAR)) { 682 set_progress_bar(0); 683 ClearFlag(MemoryChunk::HAS_PROGRESS_BAR); 684 } 685 } 686 markbits()687 inline Bitmap* markbits() { 688 return Bitmap::FromAddress(address() + kHeaderSize); 689 } 690 AddressToMarkbitIndex(Address addr)691 inline uint32_t AddressToMarkbitIndex(Address addr) { 692 return static_cast<uint32_t>(addr - this->address()) >> kPointerSizeLog2; 693 } 694 MarkbitIndexToAddress(uint32_t index)695 inline Address MarkbitIndexToAddress(uint32_t index) { 696 return this->address() + (index << kPointerSizeLog2); 697 } 698 PrintMarkbits()699 void PrintMarkbits() { markbits()->Print(); } 700 SetFlag(int flag)701 void SetFlag(int flag) { flags_ |= static_cast<uintptr_t>(1) << flag; } 702 ClearFlag(int flag)703 void ClearFlag(int flag) { flags_ &= ~(static_cast<uintptr_t>(1) << flag); } 704 IsFlagSet(int flag)705 bool IsFlagSet(int flag) { 706 return (flags_ & (static_cast<uintptr_t>(1) << flag)) != 0; 707 } 708 709 // Set or clear multiple flags at a time. The flags in the mask are set to 710 // the value in "flags", the rest retain the current value in |flags_|. SetFlags(intptr_t flags,intptr_t mask)711 void SetFlags(intptr_t flags, intptr_t mask) { 712 flags_ = (flags_ & ~mask) | (flags & mask); 713 } 714 715 // Return all current flags. GetFlags()716 intptr_t GetFlags() { return flags_; } 717 NeverEvacuate()718 bool NeverEvacuate() { return IsFlagSet(NEVER_EVACUATE); } 719 MarkNeverEvacuate()720 void MarkNeverEvacuate() { SetFlag(NEVER_EVACUATE); } 721 IsEvacuationCandidate()722 bool IsEvacuationCandidate() { 723 DCHECK(!(IsFlagSet(NEVER_EVACUATE) && IsFlagSet(EVACUATION_CANDIDATE))); 724 return IsFlagSet(EVACUATION_CANDIDATE); 725 } 726 CanAllocate()727 bool CanAllocate() { 728 return !IsEvacuationCandidate() && !IsFlagSet(NEVER_ALLOCATE_ON_PAGE); 729 } 730 ShouldSkipEvacuationSlotRecording()731 bool ShouldSkipEvacuationSlotRecording() { 732 return ((flags_ & kSkipEvacuationSlotsRecordingMask) != 0) && 733 !IsFlagSet(COMPACTION_WAS_ABORTED); 734 } 735 executable()736 Executability executable() { 737 return IsFlagSet(IS_EXECUTABLE) ? EXECUTABLE : NOT_EXECUTABLE; 738 } 739 InNewSpace()740 bool InNewSpace() { 741 return (flags_ & ((1 << IN_FROM_SPACE) | (1 << IN_TO_SPACE))) != 0; 742 } 743 InToSpace()744 bool InToSpace() { return IsFlagSet(IN_TO_SPACE); } 745 InFromSpace()746 bool InFromSpace() { return IsFlagSet(IN_FROM_SPACE); } 747 next_chunk()748 MemoryChunk* next_chunk() { return next_chunk_.Value(); } 749 prev_chunk()750 MemoryChunk* prev_chunk() { return prev_chunk_.Value(); } 751 set_next_chunk(MemoryChunk * next)752 void set_next_chunk(MemoryChunk* next) { next_chunk_.SetValue(next); } 753 set_prev_chunk(MemoryChunk * prev)754 void set_prev_chunk(MemoryChunk* prev) { prev_chunk_.SetValue(prev); } 755 owner()756 Space* owner() const { 757 if ((reinterpret_cast<intptr_t>(owner_) & kPageHeaderTagMask) == 758 kPageHeaderTag) { 759 return reinterpret_cast<Space*>(reinterpret_cast<intptr_t>(owner_) - 760 kPageHeaderTag); 761 } else { 762 return nullptr; 763 } 764 } 765 set_owner(Space * space)766 void set_owner(Space* space) { 767 DCHECK((reinterpret_cast<intptr_t>(space) & kPageHeaderTagMask) == 0); 768 owner_ = reinterpret_cast<Address>(space) + kPageHeaderTag; 769 DCHECK((reinterpret_cast<intptr_t>(owner_) & kPageHeaderTagMask) == 770 kPageHeaderTag); 771 } 772 HasPageHeader()773 bool HasPageHeader() { return owner() != nullptr; } 774 775 void InsertAfter(MemoryChunk* other); 776 void Unlink(); 777 778 protected: 779 static MemoryChunk* Initialize(Heap* heap, Address base, size_t size, 780 Address area_start, Address area_end, 781 Executability executable, Space* owner, 782 base::VirtualMemory* reservation); 783 784 // Should be called when memory chunk is about to be freed. 785 void ReleaseAllocatedMemory(); 786 reserved_memory()787 base::VirtualMemory* reserved_memory() { return &reservation_; } 788 789 size_t size_; 790 intptr_t flags_; 791 792 // Start and end of allocatable memory on this chunk. 793 Address area_start_; 794 Address area_end_; 795 796 // If the chunk needs to remember its memory reservation, it is stored here. 797 base::VirtualMemory reservation_; 798 799 // The identity of the owning space. This is tagged as a failure pointer, but 800 // no failure can be in an object, so this can be distinguished from any entry 801 // in a fixed array. 802 Address owner_; 803 804 Heap* heap_; 805 806 // Used by the incremental marker to keep track of the scanning progress in 807 // large objects that have a progress bar and are scanned in increments. 808 int progress_bar_; 809 810 // Count of bytes marked black on page. 811 int live_byte_count_; 812 813 // A single slot set for small pages (of size kPageSize) or an array of slot 814 // set for large pages. In the latter case the number of entries in the array 815 // is ceil(size() / kPageSize). 816 SlotSet* old_to_new_slots_; 817 SlotSet* old_to_old_slots_; 818 TypedSlotSet* typed_old_to_new_slots_; 819 TypedSlotSet* typed_old_to_old_slots_; 820 821 SkipList* skip_list_; 822 823 intptr_t write_barrier_counter_; 824 825 // Assuming the initial allocation on a page is sequential, 826 // count highest number of bytes ever allocated on the page. 827 base::AtomicValue<intptr_t> high_water_mark_; 828 829 base::Mutex* mutex_; 830 831 base::AtomicValue<ConcurrentSweepingState> concurrent_sweeping_; 832 833 // PagedSpace free-list statistics. 834 base::AtomicNumber<intptr_t> available_in_free_list_; 835 base::AtomicNumber<intptr_t> wasted_memory_; 836 837 // next_chunk_ holds a pointer of type MemoryChunk 838 base::AtomicValue<MemoryChunk*> next_chunk_; 839 // prev_chunk_ holds a pointer of type MemoryChunk 840 base::AtomicValue<MemoryChunk*> prev_chunk_; 841 842 FreeListCategory categories_[kNumberOfCategories]; 843 844 LocalArrayBufferTracker* local_tracker_; 845 846 private: InitializeReservedMemory()847 void InitializeReservedMemory() { reservation_.Reset(); } 848 849 friend class MemoryAllocator; 850 friend class MemoryChunkValidator; 851 }; 852 853 // ----------------------------------------------------------------------------- 854 // A page is a memory chunk of a size 1MB. Large object pages may be larger. 855 // 856 // The only way to get a page pointer is by calling factory methods: 857 // Page* p = Page::FromAddress(addr); or 858 // Page* p = Page::FromTopOrLimit(top); 859 class Page : public MemoryChunk { 860 public: 861 static const intptr_t kCopyAllFlags = ~0; 862 863 // Page flags copied from from-space to to-space when flipping semispaces. 864 static const intptr_t kCopyOnFlipFlagsMask = 865 (1 << MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING) | 866 (1 << MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING); 867 868 // Maximum object size that gets allocated into regular pages. Objects larger 869 // than that size are allocated in large object space and are never moved in 870 // memory. This also applies to new space allocation, since objects are never 871 // migrated from new space to large object space. Takes double alignment into 872 // account. 873 // TODO(hpayer): This limit should be way smaller but we currently have 874 // short living objects >256K. 875 static const int kMaxRegularHeapObjectSize = 600 * KB; 876 877 static inline Page* ConvertNewToOld(Page* old_page, PagedSpace* new_owner); 878 879 // Returns the page containing a given address. The address ranges 880 // from [page_addr .. page_addr + kPageSize[. This only works if the object 881 // is in fact in a page. FromAddress(Address addr)882 static Page* FromAddress(Address addr) { 883 return reinterpret_cast<Page*>(OffsetFrom(addr) & ~kPageAlignmentMask); 884 } 885 886 // Returns the page containing the address provided. The address can 887 // potentially point righter after the page. To be also safe for tagged values 888 // we subtract a hole word. The valid address ranges from 889 // [page_addr + kObjectStartOffset .. page_addr + kPageSize + kPointerSize]. FromAllocationAreaAddress(Address address)890 static Page* FromAllocationAreaAddress(Address address) { 891 return Page::FromAddress(address - kPointerSize); 892 } 893 894 // Checks if address1 and address2 are on the same new space page. OnSamePage(Address address1,Address address2)895 static bool OnSamePage(Address address1, Address address2) { 896 return Page::FromAddress(address1) == Page::FromAddress(address2); 897 } 898 899 // Checks whether an address is page aligned. IsAlignedToPageSize(Address addr)900 static bool IsAlignedToPageSize(Address addr) { 901 return (OffsetFrom(addr) & kPageAlignmentMask) == 0; 902 } 903 IsAtObjectStart(Address addr)904 static bool IsAtObjectStart(Address addr) { 905 return (reinterpret_cast<intptr_t>(addr) & kPageAlignmentMask) == 906 kObjectStartOffset; 907 } 908 909 inline static Page* FromAnyPointerAddress(Heap* heap, Address addr); 910 911 // Create a Page object that is only used as anchor for the doubly-linked 912 // list of real pages. Page(Space * owner)913 explicit Page(Space* owner) { InitializeAsAnchor(owner); } 914 915 inline void MarkNeverAllocateForTesting(); 916 inline void MarkEvacuationCandidate(); 917 inline void ClearEvacuationCandidate(); 918 next_page()919 Page* next_page() { return static_cast<Page*>(next_chunk()); } prev_page()920 Page* prev_page() { return static_cast<Page*>(prev_chunk()); } set_next_page(Page * page)921 void set_next_page(Page* page) { set_next_chunk(page); } set_prev_page(Page * page)922 void set_prev_page(Page* page) { set_prev_chunk(page); } 923 924 template <typename Callback> ForAllFreeListCategories(Callback callback)925 inline void ForAllFreeListCategories(Callback callback) { 926 for (int i = kFirstCategory; i < kNumberOfCategories; i++) { 927 callback(&categories_[i]); 928 } 929 } 930 931 // Returns the offset of a given address to this page. Offset(Address a)932 inline int Offset(Address a) { 933 int offset = static_cast<int>(a - address()); 934 return offset; 935 } 936 937 // Returns the address for a given offset to the this page. OffsetToAddress(int offset)938 Address OffsetToAddress(int offset) { 939 DCHECK_PAGE_OFFSET(offset); 940 return address() + offset; 941 } 942 943 // WaitUntilSweepingCompleted only works when concurrent sweeping is in 944 // progress. In particular, when we know that right before this call a 945 // sweeper thread was sweeping this page. WaitUntilSweepingCompleted()946 void WaitUntilSweepingCompleted() { 947 mutex_->Lock(); 948 mutex_->Unlock(); 949 DCHECK(SweepingDone()); 950 } 951 SweepingDone()952 bool SweepingDone() { 953 return concurrent_sweeping_state().Value() == kSweepingDone; 954 } 955 956 void ResetFreeListStatistics(); 957 LiveBytesFromFreeList()958 int LiveBytesFromFreeList() { 959 return static_cast<int>(area_size() - wasted_memory() - 960 available_in_free_list()); 961 } 962 free_list_category(FreeListCategoryType type)963 FreeListCategory* free_list_category(FreeListCategoryType type) { 964 return &categories_[type]; 965 } 966 is_anchor()967 bool is_anchor() { return IsFlagSet(Page::ANCHOR); } 968 wasted_memory()969 intptr_t wasted_memory() { return wasted_memory_.Value(); } add_wasted_memory(intptr_t waste)970 void add_wasted_memory(intptr_t waste) { wasted_memory_.Increment(waste); } available_in_free_list()971 intptr_t available_in_free_list() { return available_in_free_list_.Value(); } add_available_in_free_list(intptr_t available)972 void add_available_in_free_list(intptr_t available) { 973 available_in_free_list_.Increment(available); 974 } 975 976 #ifdef DEBUG 977 void Print(); 978 #endif // DEBUG 979 980 private: 981 enum InitializationMode { kFreeMemory, kDoNotFreeMemory }; 982 983 template <InitializationMode mode = kFreeMemory> 984 static inline Page* Initialize(Heap* heap, MemoryChunk* chunk, 985 Executability executable, PagedSpace* owner); 986 static inline Page* Initialize(Heap* heap, MemoryChunk* chunk, 987 Executability executable, SemiSpace* owner); 988 989 inline void InitializeFreeListCategories(); 990 991 void InitializeAsAnchor(Space* owner); 992 993 friend class MemoryAllocator; 994 }; 995 996 class LargePage : public MemoryChunk { 997 public: GetObject()998 HeapObject* GetObject() { return HeapObject::FromAddress(area_start()); } 999 next_page()1000 inline LargePage* next_page() { 1001 return static_cast<LargePage*>(next_chunk()); 1002 } 1003 set_next_page(LargePage * page)1004 inline void set_next_page(LargePage* page) { set_next_chunk(page); } 1005 1006 // A limit to guarantee that we do not overflow typed slot offset in 1007 // the old to old remembered set. 1008 // Note that this limit is higher than what assembler already imposes on 1009 // x64 and ia32 architectures. 1010 static const int kMaxCodePageSize = 512 * MB; 1011 1012 private: 1013 static inline LargePage* Initialize(Heap* heap, MemoryChunk* chunk, 1014 Executability executable, Space* owner); 1015 1016 friend class MemoryAllocator; 1017 }; 1018 1019 1020 // ---------------------------------------------------------------------------- 1021 // Space is the abstract superclass for all allocation spaces. 1022 class Space : public Malloced { 1023 public: Space(Heap * heap,AllocationSpace id,Executability executable)1024 Space(Heap* heap, AllocationSpace id, Executability executable) 1025 : allocation_observers_(new List<AllocationObserver*>()), 1026 allocation_observers_paused_(false), 1027 heap_(heap), 1028 id_(id), 1029 executable_(executable), 1030 committed_(0), 1031 max_committed_(0) {} 1032 ~Space()1033 virtual ~Space() {} 1034 heap()1035 Heap* heap() const { return heap_; } 1036 1037 // Does the space need executable memory? executable()1038 Executability executable() { return executable_; } 1039 1040 // Identity used in error reporting. identity()1041 AllocationSpace identity() { return id_; } 1042 AddAllocationObserver(AllocationObserver * observer)1043 virtual void AddAllocationObserver(AllocationObserver* observer) { 1044 allocation_observers_->Add(observer); 1045 } 1046 RemoveAllocationObserver(AllocationObserver * observer)1047 virtual void RemoveAllocationObserver(AllocationObserver* observer) { 1048 bool removed = allocation_observers_->RemoveElement(observer); 1049 USE(removed); 1050 DCHECK(removed); 1051 } 1052 PauseAllocationObservers()1053 virtual void PauseAllocationObservers() { 1054 allocation_observers_paused_ = true; 1055 } 1056 ResumeAllocationObservers()1057 virtual void ResumeAllocationObservers() { 1058 allocation_observers_paused_ = false; 1059 } 1060 1061 void AllocationStep(Address soon_object, int size); 1062 1063 // Return the total amount committed memory for this space, i.e., allocatable 1064 // memory and page headers. CommittedMemory()1065 virtual intptr_t CommittedMemory() { return committed_; } 1066 MaximumCommittedMemory()1067 virtual intptr_t MaximumCommittedMemory() { return max_committed_; } 1068 1069 // Returns allocated size. 1070 virtual intptr_t Size() = 0; 1071 1072 // Returns size of objects. Can differ from the allocated size 1073 // (e.g. see LargeObjectSpace). SizeOfObjects()1074 virtual intptr_t SizeOfObjects() { return Size(); } 1075 1076 // Approximate amount of physical memory committed for this space. 1077 virtual size_t CommittedPhysicalMemory() = 0; 1078 1079 // Return the available bytes without growing. 1080 virtual intptr_t Available() = 0; 1081 RoundSizeDownToObjectAlignment(int size)1082 virtual int RoundSizeDownToObjectAlignment(int size) { 1083 if (id_ == CODE_SPACE) { 1084 return RoundDown(size, kCodeAlignment); 1085 } else { 1086 return RoundDown(size, kPointerSize); 1087 } 1088 } 1089 AccountCommitted(intptr_t bytes)1090 void AccountCommitted(intptr_t bytes) { 1091 DCHECK_GE(bytes, 0); 1092 committed_ += bytes; 1093 if (committed_ > max_committed_) { 1094 max_committed_ = committed_; 1095 } 1096 } 1097 AccountUncommitted(intptr_t bytes)1098 void AccountUncommitted(intptr_t bytes) { 1099 DCHECK_GE(bytes, 0); 1100 committed_ -= bytes; 1101 DCHECK_GE(committed_, 0); 1102 } 1103 1104 #ifdef DEBUG 1105 virtual void Print() = 0; 1106 #endif 1107 1108 protected: 1109 v8::base::SmartPointer<List<AllocationObserver*>> allocation_observers_; 1110 bool allocation_observers_paused_; 1111 1112 private: 1113 Heap* heap_; 1114 AllocationSpace id_; 1115 Executability executable_; 1116 1117 // Keeps track of committed memory in a space. 1118 intptr_t committed_; 1119 intptr_t max_committed_; 1120 }; 1121 1122 1123 class MemoryChunkValidator { 1124 // Computed offsets should match the compiler generated ones. 1125 STATIC_ASSERT(MemoryChunk::kSizeOffset == offsetof(MemoryChunk, size_)); 1126 STATIC_ASSERT(MemoryChunk::kLiveBytesOffset == 1127 offsetof(MemoryChunk, live_byte_count_)); 1128 STATIC_ASSERT(MemoryChunk::kOldToNewSlotsOffset == 1129 offsetof(MemoryChunk, old_to_new_slots_)); 1130 STATIC_ASSERT(MemoryChunk::kWriteBarrierCounterOffset == 1131 offsetof(MemoryChunk, write_barrier_counter_)); 1132 1133 // Validate our estimates on the header size. 1134 STATIC_ASSERT(sizeof(MemoryChunk) <= MemoryChunk::kHeaderSize); 1135 STATIC_ASSERT(sizeof(LargePage) <= MemoryChunk::kHeaderSize); 1136 STATIC_ASSERT(sizeof(Page) <= MemoryChunk::kHeaderSize); 1137 }; 1138 1139 1140 // ---------------------------------------------------------------------------- 1141 // All heap objects containing executable code (code objects) must be allocated 1142 // from a 2 GB range of memory, so that they can call each other using 32-bit 1143 // displacements. This happens automatically on 32-bit platforms, where 32-bit 1144 // displacements cover the entire 4GB virtual address space. On 64-bit 1145 // platforms, we support this using the CodeRange object, which reserves and 1146 // manages a range of virtual memory. 1147 class CodeRange { 1148 public: 1149 explicit CodeRange(Isolate* isolate); ~CodeRange()1150 ~CodeRange() { TearDown(); } 1151 1152 // Reserves a range of virtual memory, but does not commit any of it. 1153 // Can only be called once, at heap initialization time. 1154 // Returns false on failure. 1155 bool SetUp(size_t requested_size); 1156 valid()1157 bool valid() { return code_range_ != NULL; } start()1158 Address start() { 1159 DCHECK(valid()); 1160 return static_cast<Address>(code_range_->address()); 1161 } size()1162 size_t size() { 1163 DCHECK(valid()); 1164 return code_range_->size(); 1165 } contains(Address address)1166 bool contains(Address address) { 1167 if (!valid()) return false; 1168 Address start = static_cast<Address>(code_range_->address()); 1169 return start <= address && address < start + code_range_->size(); 1170 } 1171 1172 // Allocates a chunk of memory from the large-object portion of 1173 // the code range. On platforms with no separate code range, should 1174 // not be called. 1175 MUST_USE_RESULT Address AllocateRawMemory(const size_t requested_size, 1176 const size_t commit_size, 1177 size_t* allocated); 1178 bool CommitRawMemory(Address start, size_t length); 1179 bool UncommitRawMemory(Address start, size_t length); 1180 void FreeRawMemory(Address buf, size_t length); 1181 1182 private: 1183 class FreeBlock { 1184 public: FreeBlock()1185 FreeBlock() : start(0), size(0) {} FreeBlock(Address start_arg,size_t size_arg)1186 FreeBlock(Address start_arg, size_t size_arg) 1187 : start(start_arg), size(size_arg) { 1188 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 1189 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 1190 } FreeBlock(void * start_arg,size_t size_arg)1191 FreeBlock(void* start_arg, size_t size_arg) 1192 : start(static_cast<Address>(start_arg)), size(size_arg) { 1193 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 1194 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 1195 } 1196 1197 Address start; 1198 size_t size; 1199 }; 1200 1201 // Frees the range of virtual memory, and frees the data structures used to 1202 // manage it. 1203 void TearDown(); 1204 1205 // Finds a block on the allocation list that contains at least the 1206 // requested amount of memory. If none is found, sorts and merges 1207 // the existing free memory blocks, and searches again. 1208 // If none can be found, returns false. 1209 bool GetNextAllocationBlock(size_t requested); 1210 // Compares the start addresses of two free blocks. 1211 static int CompareFreeBlockAddress(const FreeBlock* left, 1212 const FreeBlock* right); 1213 bool ReserveBlock(const size_t requested_size, FreeBlock* block); 1214 void ReleaseBlock(const FreeBlock* block); 1215 1216 Isolate* isolate_; 1217 1218 // The reserved range of virtual memory that all code objects are put in. 1219 base::VirtualMemory* code_range_; 1220 1221 // The global mutex guards free_list_ and allocation_list_ as GC threads may 1222 // access both lists concurrently to the main thread. 1223 base::Mutex code_range_mutex_; 1224 1225 // Freed blocks of memory are added to the free list. When the allocation 1226 // list is exhausted, the free list is sorted and merged to make the new 1227 // allocation list. 1228 List<FreeBlock> free_list_; 1229 1230 // Memory is allocated from the free blocks on the allocation list. 1231 // The block at current_allocation_block_index_ is the current block. 1232 List<FreeBlock> allocation_list_; 1233 int current_allocation_block_index_; 1234 1235 DISALLOW_COPY_AND_ASSIGN(CodeRange); 1236 }; 1237 1238 1239 class SkipList { 1240 public: SkipList()1241 SkipList() { Clear(); } 1242 Clear()1243 void Clear() { 1244 for (int idx = 0; idx < kSize; idx++) { 1245 starts_[idx] = reinterpret_cast<Address>(-1); 1246 } 1247 } 1248 StartFor(Address addr)1249 Address StartFor(Address addr) { return starts_[RegionNumber(addr)]; } 1250 AddObject(Address addr,int size)1251 void AddObject(Address addr, int size) { 1252 int start_region = RegionNumber(addr); 1253 int end_region = RegionNumber(addr + size - kPointerSize); 1254 for (int idx = start_region; idx <= end_region; idx++) { 1255 if (starts_[idx] > addr) { 1256 starts_[idx] = addr; 1257 } else { 1258 // In the first region, there may already be an object closer to the 1259 // start of the region. Do not change the start in that case. If this 1260 // is not the first region, you probably added overlapping objects. 1261 DCHECK_EQ(start_region, idx); 1262 } 1263 } 1264 } 1265 RegionNumber(Address addr)1266 static inline int RegionNumber(Address addr) { 1267 return (OffsetFrom(addr) & Page::kPageAlignmentMask) >> kRegionSizeLog2; 1268 } 1269 Update(Address addr,int size)1270 static void Update(Address addr, int size) { 1271 Page* page = Page::FromAddress(addr); 1272 SkipList* list = page->skip_list(); 1273 if (list == NULL) { 1274 list = new SkipList(); 1275 page->set_skip_list(list); 1276 } 1277 1278 list->AddObject(addr, size); 1279 } 1280 1281 private: 1282 static const int kRegionSizeLog2 = 13; 1283 static const int kRegionSize = 1 << kRegionSizeLog2; 1284 static const int kSize = Page::kPageSize / kRegionSize; 1285 1286 STATIC_ASSERT(Page::kPageSize % kRegionSize == 0); 1287 1288 Address starts_[kSize]; 1289 }; 1290 1291 1292 // ---------------------------------------------------------------------------- 1293 // A space acquires chunks of memory from the operating system. The memory 1294 // allocator allocated and deallocates pages for the paged heap spaces and large 1295 // pages for large object space. 1296 class MemoryAllocator { 1297 public: 1298 // Unmapper takes care of concurrently unmapping and uncommitting memory 1299 // chunks. 1300 class Unmapper { 1301 public: 1302 class UnmapFreeMemoryTask; 1303 Unmapper(MemoryAllocator * allocator)1304 explicit Unmapper(MemoryAllocator* allocator) 1305 : allocator_(allocator), 1306 pending_unmapping_tasks_semaphore_(0), 1307 concurrent_unmapping_tasks_active_(0) {} 1308 AddMemoryChunkSafe(MemoryChunk * chunk)1309 void AddMemoryChunkSafe(MemoryChunk* chunk) { 1310 if ((chunk->size() == Page::kPageSize) && 1311 (chunk->executable() != EXECUTABLE)) { 1312 AddMemoryChunkSafe<kRegular>(chunk); 1313 } else { 1314 AddMemoryChunkSafe<kNonRegular>(chunk); 1315 } 1316 } 1317 TryGetPooledMemoryChunkSafe()1318 MemoryChunk* TryGetPooledMemoryChunkSafe() { 1319 // Procedure: 1320 // (1) Try to get a chunk that was declared as pooled and already has 1321 // been uncommitted. 1322 // (2) Try to steal any memory chunk of kPageSize that would've been 1323 // unmapped. 1324 MemoryChunk* chunk = GetMemoryChunkSafe<kPooled>(); 1325 if (chunk == nullptr) { 1326 chunk = GetMemoryChunkSafe<kRegular>(); 1327 if (chunk != nullptr) { 1328 // For stolen chunks we need to manually free any allocated memory. 1329 chunk->ReleaseAllocatedMemory(); 1330 } 1331 } 1332 return chunk; 1333 } 1334 1335 void FreeQueuedChunks(); 1336 bool WaitUntilCompleted(); 1337 1338 private: 1339 enum ChunkQueueType { 1340 kRegular, // Pages of kPageSize that do not live in a CodeRange and 1341 // can thus be used for stealing. 1342 kNonRegular, // Large chunks and executable chunks. 1343 kPooled, // Pooled chunks, already uncommited and ready for reuse. 1344 kNumberOfChunkQueues, 1345 }; 1346 1347 template <ChunkQueueType type> AddMemoryChunkSafe(MemoryChunk * chunk)1348 void AddMemoryChunkSafe(MemoryChunk* chunk) { 1349 base::LockGuard<base::Mutex> guard(&mutex_); 1350 if (type != kRegular || allocator_->CanFreeMemoryChunk(chunk)) { 1351 chunks_[type].push_back(chunk); 1352 } else { 1353 DCHECK_EQ(type, kRegular); 1354 delayed_regular_chunks_.push_back(chunk); 1355 } 1356 } 1357 1358 template <ChunkQueueType type> GetMemoryChunkSafe()1359 MemoryChunk* GetMemoryChunkSafe() { 1360 base::LockGuard<base::Mutex> guard(&mutex_); 1361 if (chunks_[type].empty()) return nullptr; 1362 MemoryChunk* chunk = chunks_[type].front(); 1363 chunks_[type].pop_front(); 1364 return chunk; 1365 } 1366 1367 void ReconsiderDelayedChunks(); 1368 void PerformFreeMemoryOnQueuedChunks(); 1369 1370 base::Mutex mutex_; 1371 MemoryAllocator* allocator_; 1372 std::list<MemoryChunk*> chunks_[kNumberOfChunkQueues]; 1373 // Delayed chunks cannot be processed in the current unmapping cycle because 1374 // of dependencies such as an active sweeper. 1375 // See MemoryAllocator::CanFreeMemoryChunk. 1376 std::list<MemoryChunk*> delayed_regular_chunks_; 1377 base::Semaphore pending_unmapping_tasks_semaphore_; 1378 intptr_t concurrent_unmapping_tasks_active_; 1379 1380 friend class MemoryAllocator; 1381 }; 1382 1383 enum AllocationMode { 1384 kRegular, 1385 kPooled, 1386 }; 1387 enum FreeMode { 1388 kFull, 1389 kPreFreeAndQueue, 1390 kPooledAndQueue, 1391 }; 1392 1393 explicit MemoryAllocator(Isolate* isolate); 1394 1395 // Initializes its internal bookkeeping structures. 1396 // Max capacity of the total space and executable memory limit. 1397 bool SetUp(intptr_t max_capacity, intptr_t capacity_executable, 1398 intptr_t code_range_size); 1399 1400 void TearDown(); 1401 1402 // Allocates a Page from the allocator. AllocationMode is used to indicate 1403 // whether pooled allocation, which only works for MemoryChunk::kPageSize, 1404 // should be tried first. 1405 template <MemoryAllocator::AllocationMode alloc_mode = kRegular, 1406 typename SpaceType> 1407 Page* AllocatePage(intptr_t size, SpaceType* owner, Executability executable); 1408 1409 LargePage* AllocateLargePage(intptr_t size, LargeObjectSpace* owner, 1410 Executability executable); 1411 1412 template <MemoryAllocator::FreeMode mode = kFull> 1413 void Free(MemoryChunk* chunk); 1414 1415 bool CanFreeMemoryChunk(MemoryChunk* chunk); 1416 1417 // Returns allocated spaces in bytes. Size()1418 intptr_t Size() { return size_.Value(); } 1419 1420 // Returns allocated executable spaces in bytes. SizeExecutable()1421 intptr_t SizeExecutable() { return size_executable_.Value(); } 1422 1423 // Returns the maximum available bytes of heaps. Available()1424 intptr_t Available() { 1425 intptr_t size = Size(); 1426 return capacity_ < size ? 0 : capacity_ - size; 1427 } 1428 1429 // Returns the maximum available executable bytes of heaps. AvailableExecutable()1430 intptr_t AvailableExecutable() { 1431 intptr_t executable_size = SizeExecutable(); 1432 if (capacity_executable_ < executable_size) return 0; 1433 return capacity_executable_ - executable_size; 1434 } 1435 1436 // Returns maximum available bytes that the old space can have. MaxAvailable()1437 intptr_t MaxAvailable() { 1438 return (Available() / Page::kPageSize) * Page::kAllocatableMemory; 1439 } 1440 1441 // Returns an indication of whether a pointer is in a space that has 1442 // been allocated by this MemoryAllocator. IsOutsideAllocatedSpace(const void * address)1443 V8_INLINE bool IsOutsideAllocatedSpace(const void* address) { 1444 return address < lowest_ever_allocated_.Value() || 1445 address >= highest_ever_allocated_.Value(); 1446 } 1447 1448 #ifdef DEBUG 1449 // Reports statistic info of the space. 1450 void ReportStatistics(); 1451 #endif 1452 1453 // Returns a MemoryChunk in which the memory region from commit_area_size to 1454 // reserve_area_size of the chunk area is reserved but not committed, it 1455 // could be committed later by calling MemoryChunk::CommitArea. 1456 MemoryChunk* AllocateChunk(intptr_t reserve_area_size, 1457 intptr_t commit_area_size, 1458 Executability executable, Space* space); 1459 1460 Address ReserveAlignedMemory(size_t requested, size_t alignment, 1461 base::VirtualMemory* controller); 1462 Address AllocateAlignedMemory(size_t reserve_size, size_t commit_size, 1463 size_t alignment, Executability executable, 1464 base::VirtualMemory* controller); 1465 1466 bool CommitMemory(Address addr, size_t size, Executability executable); 1467 1468 void FreeMemory(base::VirtualMemory* reservation, Executability executable); 1469 void FreeMemory(Address addr, size_t size, Executability executable); 1470 1471 // Commit a contiguous block of memory from the initial chunk. Assumes that 1472 // the address is not NULL, the size is greater than zero, and that the 1473 // block is contained in the initial chunk. Returns true if it succeeded 1474 // and false otherwise. 1475 bool CommitBlock(Address start, size_t size, Executability executable); 1476 1477 // Uncommit a contiguous block of memory [start..(start+size)[. 1478 // start is not NULL, the size is greater than zero, and the 1479 // block is contained in the initial chunk. Returns true if it succeeded 1480 // and false otherwise. 1481 bool UncommitBlock(Address start, size_t size); 1482 1483 // Zaps a contiguous block of memory [start..(start+size)[ thus 1484 // filling it up with a recognizable non-NULL bit pattern. 1485 void ZapBlock(Address start, size_t size); 1486 1487 static int CodePageGuardStartOffset(); 1488 1489 static int CodePageGuardSize(); 1490 1491 static int CodePageAreaStartOffset(); 1492 1493 static int CodePageAreaEndOffset(); 1494 CodePageAreaSize()1495 static int CodePageAreaSize() { 1496 return CodePageAreaEndOffset() - CodePageAreaStartOffset(); 1497 } 1498 PageAreaSize(AllocationSpace space)1499 static int PageAreaSize(AllocationSpace space) { 1500 DCHECK_NE(LO_SPACE, space); 1501 return (space == CODE_SPACE) ? CodePageAreaSize() 1502 : Page::kAllocatableMemory; 1503 } 1504 1505 MUST_USE_RESULT bool CommitExecutableMemory(base::VirtualMemory* vm, 1506 Address start, size_t commit_size, 1507 size_t reserved_size); 1508 code_range()1509 CodeRange* code_range() { return code_range_; } unmapper()1510 Unmapper* unmapper() { return &unmapper_; } 1511 1512 private: 1513 // PreFree logically frees the object, i.e., it takes care of the size 1514 // bookkeeping and calls the allocation callback. 1515 void PreFreeMemory(MemoryChunk* chunk); 1516 1517 // FreeMemory can be called concurrently when PreFree was executed before. 1518 void PerformFreeMemory(MemoryChunk* chunk); 1519 1520 // See AllocatePage for public interface. Note that currently we only support 1521 // pools for NOT_EXECUTABLE pages of size MemoryChunk::kPageSize. 1522 template <typename SpaceType> 1523 MemoryChunk* AllocatePagePooled(SpaceType* owner); 1524 1525 Isolate* isolate_; 1526 1527 CodeRange* code_range_; 1528 1529 // Maximum space size in bytes. 1530 intptr_t capacity_; 1531 // Maximum subset of capacity_ that can be executable 1532 intptr_t capacity_executable_; 1533 1534 // Allocated space size in bytes. 1535 base::AtomicNumber<intptr_t> size_; 1536 // Allocated executable space size in bytes. 1537 base::AtomicNumber<intptr_t> size_executable_; 1538 1539 // We keep the lowest and highest addresses allocated as a quick way 1540 // of determining that pointers are outside the heap. The estimate is 1541 // conservative, i.e. not all addrsses in 'allocated' space are allocated 1542 // to our heap. The range is [lowest, highest[, inclusive on the low end 1543 // and exclusive on the high end. 1544 base::AtomicValue<void*> lowest_ever_allocated_; 1545 base::AtomicValue<void*> highest_ever_allocated_; 1546 1547 // Initializes pages in a chunk. Returns the first page address. 1548 // This function and GetChunkId() are provided for the mark-compact 1549 // collector to rebuild page headers in the from space, which is 1550 // used as a marking stack and its page headers are destroyed. 1551 Page* InitializePagesInChunk(int chunk_id, int pages_in_chunk, 1552 PagedSpace* owner); 1553 UpdateAllocatedSpaceLimits(void * low,void * high)1554 void UpdateAllocatedSpaceLimits(void* low, void* high) { 1555 // The use of atomic primitives does not guarantee correctness (wrt. 1556 // desired semantics) by default. The loop here ensures that we update the 1557 // values only if they did not change in between. 1558 void* ptr = nullptr; 1559 do { 1560 ptr = lowest_ever_allocated_.Value(); 1561 } while ((low < ptr) && !lowest_ever_allocated_.TrySetValue(ptr, low)); 1562 do { 1563 ptr = highest_ever_allocated_.Value(); 1564 } while ((high > ptr) && !highest_ever_allocated_.TrySetValue(ptr, high)); 1565 } 1566 1567 base::VirtualMemory last_chunk_; 1568 Unmapper unmapper_; 1569 1570 friend class TestCodeRangeScope; 1571 1572 DISALLOW_IMPLICIT_CONSTRUCTORS(MemoryAllocator); 1573 }; 1574 1575 1576 // ----------------------------------------------------------------------------- 1577 // Interface for heap object iterator to be implemented by all object space 1578 // object iterators. 1579 // 1580 // NOTE: The space specific object iterators also implements the own next() 1581 // method which is used to avoid using virtual functions 1582 // iterating a specific space. 1583 1584 class ObjectIterator : public Malloced { 1585 public: ~ObjectIterator()1586 virtual ~ObjectIterator() {} 1587 virtual HeapObject* Next() = 0; 1588 }; 1589 1590 template <class PAGE_TYPE> 1591 class PageIteratorImpl 1592 : public std::iterator<std::forward_iterator_tag, PAGE_TYPE> { 1593 public: PageIteratorImpl(PAGE_TYPE * p)1594 explicit PageIteratorImpl(PAGE_TYPE* p) : p_(p) {} PageIteratorImpl(const PageIteratorImpl<PAGE_TYPE> & other)1595 PageIteratorImpl(const PageIteratorImpl<PAGE_TYPE>& other) : p_(other.p_) {} 1596 PAGE_TYPE* operator*() { return p_; } 1597 bool operator==(const PageIteratorImpl<PAGE_TYPE>& rhs) { 1598 return rhs.p_ == p_; 1599 } 1600 bool operator!=(const PageIteratorImpl<PAGE_TYPE>& rhs) { 1601 return rhs.p_ != p_; 1602 } 1603 inline PageIteratorImpl<PAGE_TYPE>& operator++(); 1604 inline PageIteratorImpl<PAGE_TYPE> operator++(int); 1605 1606 private: 1607 PAGE_TYPE* p_; 1608 }; 1609 1610 typedef PageIteratorImpl<Page> PageIterator; 1611 typedef PageIteratorImpl<LargePage> LargePageIterator; 1612 1613 class PageRange { 1614 public: 1615 typedef PageIterator iterator; PageRange(Page * begin,Page * end)1616 PageRange(Page* begin, Page* end) : begin_(begin), end_(end) {} PageRange(Page * page)1617 explicit PageRange(Page* page) : PageRange(page, page->next_page()) {} begin()1618 iterator begin() { return iterator(begin_); } end()1619 iterator end() { return iterator(end_); } 1620 1621 private: 1622 Page* begin_; 1623 Page* end_; 1624 }; 1625 1626 // ----------------------------------------------------------------------------- 1627 // Heap object iterator in new/old/map spaces. 1628 // 1629 // A HeapObjectIterator iterates objects from the bottom of the given space 1630 // to its top or from the bottom of the given page to its top. 1631 // 1632 // If objects are allocated in the page during iteration the iterator may 1633 // or may not iterate over those objects. The caller must create a new 1634 // iterator in order to be sure to visit these new objects. 1635 class HeapObjectIterator : public ObjectIterator { 1636 public: 1637 // Creates a new object iterator in a given space. 1638 explicit HeapObjectIterator(PagedSpace* space); 1639 explicit HeapObjectIterator(Page* page); 1640 1641 // Advance to the next object, skipping free spaces and other fillers and 1642 // skipping the special garbage section of which there is one per space. 1643 // Returns nullptr when the iteration has ended. 1644 inline HeapObject* Next() override; 1645 1646 private: 1647 // Fast (inlined) path of next(). 1648 inline HeapObject* FromCurrentPage(); 1649 1650 // Slow path of next(), goes into the next page. Returns false if the 1651 // iteration has ended. 1652 bool AdvanceToNextPage(); 1653 1654 Address cur_addr_; // Current iteration point. 1655 Address cur_end_; // End iteration point. 1656 PagedSpace* space_; 1657 PageRange page_range_; 1658 PageRange::iterator current_page_; 1659 }; 1660 1661 1662 // ----------------------------------------------------------------------------- 1663 // A space has a circular list of pages. The next page can be accessed via 1664 // Page::next_page() call. 1665 1666 // An abstraction of allocation and relocation pointers in a page-structured 1667 // space. 1668 class AllocationInfo { 1669 public: AllocationInfo()1670 AllocationInfo() : top_(nullptr), limit_(nullptr) {} AllocationInfo(Address top,Address limit)1671 AllocationInfo(Address top, Address limit) : top_(top), limit_(limit) {} 1672 Reset(Address top,Address limit)1673 void Reset(Address top, Address limit) { 1674 set_top(top); 1675 set_limit(limit); 1676 } 1677 INLINE(void set_top (Address top))1678 INLINE(void set_top(Address top)) { 1679 SLOW_DCHECK(top == NULL || 1680 (reinterpret_cast<intptr_t>(top) & kHeapObjectTagMask) == 0); 1681 top_ = top; 1682 } 1683 INLINE(Address top ())1684 INLINE(Address top()) const { 1685 SLOW_DCHECK(top_ == NULL || 1686 (reinterpret_cast<intptr_t>(top_) & kHeapObjectTagMask) == 0); 1687 return top_; 1688 } 1689 top_address()1690 Address* top_address() { return &top_; } 1691 INLINE(void set_limit (Address limit))1692 INLINE(void set_limit(Address limit)) { 1693 limit_ = limit; 1694 } 1695 INLINE(Address limit ())1696 INLINE(Address limit()) const { 1697 return limit_; 1698 } 1699 limit_address()1700 Address* limit_address() { return &limit_; } 1701 1702 #ifdef DEBUG VerifyPagedAllocation()1703 bool VerifyPagedAllocation() { 1704 return (Page::FromAllocationAreaAddress(top_) == 1705 Page::FromAllocationAreaAddress(limit_)) && 1706 (top_ <= limit_); 1707 } 1708 #endif 1709 1710 private: 1711 // Current allocation top. 1712 Address top_; 1713 // Current allocation limit. 1714 Address limit_; 1715 }; 1716 1717 1718 // An abstraction of the accounting statistics of a page-structured space. 1719 // 1720 // The stats are only set by functions that ensure they stay balanced. These 1721 // functions increase or decrease one of the non-capacity stats in conjunction 1722 // with capacity, or else they always balance increases and decreases to the 1723 // non-capacity stats. 1724 class AllocationStats BASE_EMBEDDED { 1725 public: AllocationStats()1726 AllocationStats() { Clear(); } 1727 1728 // Zero out all the allocation statistics (i.e., no capacity). Clear()1729 void Clear() { 1730 capacity_ = 0; 1731 max_capacity_ = 0; 1732 size_ = 0; 1733 } 1734 ClearSize()1735 void ClearSize() { size_ = capacity_; } 1736 1737 // Accessors for the allocation statistics. Capacity()1738 intptr_t Capacity() { return capacity_; } MaxCapacity()1739 intptr_t MaxCapacity() { return max_capacity_; } Size()1740 intptr_t Size() { 1741 CHECK_GE(size_, 0); 1742 return size_; 1743 } 1744 1745 // Grow the space by adding available bytes. They are initially marked as 1746 // being in use (part of the size), but will normally be immediately freed, 1747 // putting them on the free list and removing them from size_. ExpandSpace(int size_in_bytes)1748 void ExpandSpace(int size_in_bytes) { 1749 capacity_ += size_in_bytes; 1750 size_ += size_in_bytes; 1751 if (capacity_ > max_capacity_) { 1752 max_capacity_ = capacity_; 1753 } 1754 CHECK(size_ >= 0); 1755 } 1756 1757 // Shrink the space by removing available bytes. Since shrinking is done 1758 // during sweeping, bytes have been marked as being in use (part of the size) 1759 // and are hereby freed. ShrinkSpace(int size_in_bytes)1760 void ShrinkSpace(int size_in_bytes) { 1761 capacity_ -= size_in_bytes; 1762 size_ -= size_in_bytes; 1763 CHECK_GE(size_, 0); 1764 } 1765 1766 // Allocate from available bytes (available -> size). AllocateBytes(intptr_t size_in_bytes)1767 void AllocateBytes(intptr_t size_in_bytes) { 1768 size_ += size_in_bytes; 1769 CHECK_GE(size_, 0); 1770 } 1771 1772 // Free allocated bytes, making them available (size -> available). DeallocateBytes(intptr_t size_in_bytes)1773 void DeallocateBytes(intptr_t size_in_bytes) { 1774 size_ -= size_in_bytes; 1775 CHECK_GE(size_, 0); 1776 } 1777 1778 // Merge {other} into {this}. Merge(const AllocationStats & other)1779 void Merge(const AllocationStats& other) { 1780 capacity_ += other.capacity_; 1781 size_ += other.size_; 1782 if (other.max_capacity_ > max_capacity_) { 1783 max_capacity_ = other.max_capacity_; 1784 } 1785 CHECK_GE(size_, 0); 1786 } 1787 DecreaseCapacity(intptr_t size_in_bytes)1788 void DecreaseCapacity(intptr_t size_in_bytes) { 1789 capacity_ -= size_in_bytes; 1790 CHECK_GE(capacity_, 0); 1791 CHECK_GE(capacity_, size_); 1792 } 1793 IncreaseCapacity(intptr_t size_in_bytes)1794 void IncreaseCapacity(intptr_t size_in_bytes) { capacity_ += size_in_bytes; } 1795 1796 private: 1797 // |capacity_|: The number of object-area bytes (i.e., not including page 1798 // bookkeeping structures) currently in the space. 1799 intptr_t capacity_; 1800 1801 // |max_capacity_|: The maximum capacity ever observed. 1802 intptr_t max_capacity_; 1803 1804 // |size_|: The number of allocated bytes. 1805 intptr_t size_; 1806 }; 1807 1808 // A free list maintaining free blocks of memory. The free list is organized in 1809 // a way to encourage objects allocated around the same time to be near each 1810 // other. The normal way to allocate is intended to be by bumping a 'top' 1811 // pointer until it hits a 'limit' pointer. When the limit is hit we need to 1812 // find a new space to allocate from. This is done with the free list, which is 1813 // divided up into rough categories to cut down on waste. Having finer 1814 // categories would scatter allocation more. 1815 1816 // The free list is organized in categories as follows: 1817 // kMinBlockSize-10 words (tiniest): The tiniest blocks are only used for 1818 // allocation, when categories >= small do not have entries anymore. 1819 // 11-31 words (tiny): The tiny blocks are only used for allocation, when 1820 // categories >= small do not have entries anymore. 1821 // 32-255 words (small): Used for allocating free space between 1-31 words in 1822 // size. 1823 // 256-2047 words (medium): Used for allocating free space between 32-255 words 1824 // in size. 1825 // 1048-16383 words (large): Used for allocating free space between 256-2047 1826 // words in size. 1827 // At least 16384 words (huge): This list is for objects of 2048 words or 1828 // larger. Empty pages are also added to this list. 1829 class FreeList { 1830 public: 1831 // This method returns how much memory can be allocated after freeing 1832 // maximum_freed memory. GuaranteedAllocatable(int maximum_freed)1833 static inline int GuaranteedAllocatable(int maximum_freed) { 1834 if (maximum_freed <= kTiniestListMax) { 1835 // Since we are not iterating over all list entries, we cannot guarantee 1836 // that we can find the maximum freed block in that free list. 1837 return 0; 1838 } else if (maximum_freed <= kTinyListMax) { 1839 return kTinyAllocationMax; 1840 } else if (maximum_freed <= kSmallListMax) { 1841 return kSmallAllocationMax; 1842 } else if (maximum_freed <= kMediumListMax) { 1843 return kMediumAllocationMax; 1844 } else if (maximum_freed <= kLargeListMax) { 1845 return kLargeAllocationMax; 1846 } 1847 return maximum_freed; 1848 } 1849 1850 explicit FreeList(PagedSpace* owner); 1851 1852 // Adds a node on the free list. The block of size {size_in_bytes} starting 1853 // at {start} is placed on the free list. The return value is the number of 1854 // bytes that were not added to the free list, because they freed memory block 1855 // was too small. Bookkeeping information will be written to the block, i.e., 1856 // its contents will be destroyed. The start address should be word aligned, 1857 // and the size should be a non-zero multiple of the word size. 1858 int Free(Address start, int size_in_bytes, FreeMode mode); 1859 1860 // Allocate a block of size {size_in_bytes} from the free list. The block is 1861 // unitialized. A failure is returned if no block is available. The size 1862 // should be a non-zero multiple of the word size. 1863 MUST_USE_RESULT HeapObject* Allocate(int size_in_bytes); 1864 1865 // Clear the free list. 1866 void Reset(); 1867 ResetStats()1868 void ResetStats() { 1869 wasted_bytes_.SetValue(0); 1870 ForAllFreeListCategories( 1871 [](FreeListCategory* category) { category->ResetStats(); }); 1872 } 1873 1874 // Return the number of bytes available on the free list. Available()1875 intptr_t Available() { 1876 intptr_t available = 0; 1877 ForAllFreeListCategories([&available](FreeListCategory* category) { 1878 available += category->available(); 1879 }); 1880 return available; 1881 } 1882 IsEmpty()1883 bool IsEmpty() { 1884 bool empty = true; 1885 ForAllFreeListCategories([&empty](FreeListCategory* category) { 1886 if (!category->is_empty()) empty = false; 1887 }); 1888 return empty; 1889 } 1890 1891 // Used after booting the VM. 1892 void RepairLists(Heap* heap); 1893 1894 intptr_t EvictFreeListItems(Page* page); 1895 bool ContainsPageFreeListItems(Page* page); 1896 owner()1897 PagedSpace* owner() { return owner_; } wasted_bytes()1898 intptr_t wasted_bytes() { return wasted_bytes_.Value(); } 1899 1900 template <typename Callback> ForAllFreeListCategories(FreeListCategoryType type,Callback callback)1901 void ForAllFreeListCategories(FreeListCategoryType type, Callback callback) { 1902 FreeListCategory* current = categories_[type]; 1903 while (current != nullptr) { 1904 FreeListCategory* next = current->next(); 1905 callback(current); 1906 current = next; 1907 } 1908 } 1909 1910 template <typename Callback> ForAllFreeListCategories(Callback callback)1911 void ForAllFreeListCategories(Callback callback) { 1912 for (int i = kFirstCategory; i < kNumberOfCategories; i++) { 1913 ForAllFreeListCategories(static_cast<FreeListCategoryType>(i), callback); 1914 } 1915 } 1916 1917 bool AddCategory(FreeListCategory* category); 1918 void RemoveCategory(FreeListCategory* category); 1919 void PrintCategories(FreeListCategoryType type); 1920 1921 #ifdef DEBUG 1922 intptr_t SumFreeLists(); 1923 bool IsVeryLong(); 1924 #endif 1925 1926 private: 1927 class FreeListCategoryIterator { 1928 public: FreeListCategoryIterator(FreeList * free_list,FreeListCategoryType type)1929 FreeListCategoryIterator(FreeList* free_list, FreeListCategoryType type) 1930 : current_(free_list->categories_[type]) {} 1931 HasNext()1932 bool HasNext() { return current_ != nullptr; } 1933 Next()1934 FreeListCategory* Next() { 1935 DCHECK(HasNext()); 1936 FreeListCategory* tmp = current_; 1937 current_ = current_->next(); 1938 return tmp; 1939 } 1940 1941 private: 1942 FreeListCategory* current_; 1943 }; 1944 1945 // The size range of blocks, in bytes. 1946 static const int kMinBlockSize = 3 * kPointerSize; 1947 static const int kMaxBlockSize = Page::kAllocatableMemory; 1948 1949 static const int kTiniestListMax = 0xa * kPointerSize; 1950 static const int kTinyListMax = 0x1f * kPointerSize; 1951 static const int kSmallListMax = 0xff * kPointerSize; 1952 static const int kMediumListMax = 0x7ff * kPointerSize; 1953 static const int kLargeListMax = 0x3fff * kPointerSize; 1954 static const int kTinyAllocationMax = kTiniestListMax; 1955 static const int kSmallAllocationMax = kTinyListMax; 1956 static const int kMediumAllocationMax = kSmallListMax; 1957 static const int kLargeAllocationMax = kMediumListMax; 1958 1959 FreeSpace* FindNodeFor(int size_in_bytes, int* node_size); 1960 1961 // Walks all available categories for a given |type| and tries to retrieve 1962 // a node. Returns nullptr if the category is empty. 1963 FreeSpace* FindNodeIn(FreeListCategoryType type, int* node_size); 1964 1965 // Tries to retrieve a node from the first category in a given |type|. 1966 // Returns nullptr if the category is empty. 1967 FreeSpace* TryFindNodeIn(FreeListCategoryType type, int* node_size, 1968 int minimum_size); 1969 1970 // Searches a given |type| for a node of at least |minimum_size|. 1971 FreeSpace* SearchForNodeInList(FreeListCategoryType type, int* node_size, 1972 int minimum_size); 1973 SelectFreeListCategoryType(size_t size_in_bytes)1974 FreeListCategoryType SelectFreeListCategoryType(size_t size_in_bytes) { 1975 if (size_in_bytes <= kTiniestListMax) { 1976 return kTiniest; 1977 } else if (size_in_bytes <= kTinyListMax) { 1978 return kTiny; 1979 } else if (size_in_bytes <= kSmallListMax) { 1980 return kSmall; 1981 } else if (size_in_bytes <= kMediumListMax) { 1982 return kMedium; 1983 } else if (size_in_bytes <= kLargeListMax) { 1984 return kLarge; 1985 } 1986 return kHuge; 1987 } 1988 1989 // The tiny categories are not used for fast allocation. SelectFastAllocationFreeListCategoryType(size_t size_in_bytes)1990 FreeListCategoryType SelectFastAllocationFreeListCategoryType( 1991 size_t size_in_bytes) { 1992 if (size_in_bytes <= kSmallAllocationMax) { 1993 return kSmall; 1994 } else if (size_in_bytes <= kMediumAllocationMax) { 1995 return kMedium; 1996 } else if (size_in_bytes <= kLargeAllocationMax) { 1997 return kLarge; 1998 } 1999 return kHuge; 2000 } 2001 top(FreeListCategoryType type)2002 FreeListCategory* top(FreeListCategoryType type) { return categories_[type]; } 2003 2004 PagedSpace* owner_; 2005 base::AtomicNumber<intptr_t> wasted_bytes_; 2006 FreeListCategory* categories_[kNumberOfCategories]; 2007 2008 friend class FreeListCategory; 2009 2010 DISALLOW_IMPLICIT_CONSTRUCTORS(FreeList); 2011 }; 2012 2013 2014 class AllocationResult { 2015 public: 2016 // Implicit constructor from Object*. AllocationResult(Object * object)2017 AllocationResult(Object* object) // NOLINT 2018 : object_(object) { 2019 // AllocationResults can't return Smis, which are used to represent 2020 // failure and the space to retry in. 2021 CHECK(!object->IsSmi()); 2022 } 2023 AllocationResult()2024 AllocationResult() : object_(Smi::FromInt(NEW_SPACE)) {} 2025 2026 static inline AllocationResult Retry(AllocationSpace space = NEW_SPACE) { 2027 return AllocationResult(space); 2028 } 2029 IsRetry()2030 inline bool IsRetry() { return object_->IsSmi(); } 2031 2032 template <typename T> To(T ** obj)2033 bool To(T** obj) { 2034 if (IsRetry()) return false; 2035 *obj = T::cast(object_); 2036 return true; 2037 } 2038 ToObjectChecked()2039 Object* ToObjectChecked() { 2040 CHECK(!IsRetry()); 2041 return object_; 2042 } 2043 2044 inline AllocationSpace RetrySpace(); 2045 2046 private: AllocationResult(AllocationSpace space)2047 explicit AllocationResult(AllocationSpace space) 2048 : object_(Smi::FromInt(static_cast<int>(space))) {} 2049 2050 Object* object_; 2051 }; 2052 2053 2054 STATIC_ASSERT(sizeof(AllocationResult) == kPointerSize); 2055 2056 2057 // LocalAllocationBuffer represents a linear allocation area that is created 2058 // from a given {AllocationResult} and can be used to allocate memory without 2059 // synchronization. 2060 // 2061 // The buffer is properly closed upon destruction and reassignment. 2062 // Example: 2063 // { 2064 // AllocationResult result = ...; 2065 // LocalAllocationBuffer a(heap, result, size); 2066 // LocalAllocationBuffer b = a; 2067 // CHECK(!a.IsValid()); 2068 // CHECK(b.IsValid()); 2069 // // {a} is invalid now and cannot be used for further allocations. 2070 // } 2071 // // Since {b} went out of scope, the LAB is closed, resulting in creating a 2072 // // filler object for the remaining area. 2073 class LocalAllocationBuffer { 2074 public: 2075 // Indicates that a buffer cannot be used for allocations anymore. Can result 2076 // from either reassigning a buffer, or trying to construct it from an 2077 // invalid {AllocationResult}. 2078 static inline LocalAllocationBuffer InvalidBuffer(); 2079 2080 // Creates a new LAB from a given {AllocationResult}. Results in 2081 // InvalidBuffer if the result indicates a retry. 2082 static inline LocalAllocationBuffer FromResult(Heap* heap, 2083 AllocationResult result, 2084 intptr_t size); 2085 ~LocalAllocationBuffer()2086 ~LocalAllocationBuffer() { Close(); } 2087 2088 // Convert to C++11 move-semantics once allowed by the style guide. 2089 LocalAllocationBuffer(const LocalAllocationBuffer& other); 2090 LocalAllocationBuffer& operator=(const LocalAllocationBuffer& other); 2091 2092 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 2093 int size_in_bytes, AllocationAlignment alignment); 2094 IsValid()2095 inline bool IsValid() { return allocation_info_.top() != nullptr; } 2096 2097 // Try to merge LABs, which is only possible when they are adjacent in memory. 2098 // Returns true if the merge was successful, false otherwise. 2099 inline bool TryMerge(LocalAllocationBuffer* other); 2100 2101 private: 2102 LocalAllocationBuffer(Heap* heap, AllocationInfo allocation_info); 2103 2104 void Close(); 2105 2106 Heap* heap_; 2107 AllocationInfo allocation_info_; 2108 }; 2109 2110 class NewSpacePageRange { 2111 public: 2112 typedef PageRange::iterator iterator; 2113 inline NewSpacePageRange(Address start, Address limit); begin()2114 iterator begin() { return range_.begin(); } end()2115 iterator end() { return range_.end(); } 2116 2117 private: 2118 PageRange range_; 2119 }; 2120 2121 class PagedSpace : public Space { 2122 public: 2123 typedef PageIterator iterator; 2124 2125 static const intptr_t kCompactionMemoryWanted = 500 * KB; 2126 2127 // Creates a space with an id. 2128 PagedSpace(Heap* heap, AllocationSpace id, Executability executable); 2129 ~PagedSpace()2130 ~PagedSpace() override { TearDown(); } 2131 2132 // Set up the space using the given address range of virtual memory (from 2133 // the memory allocator's initial chunk) if possible. If the block of 2134 // addresses is not big enough to contain a single page-aligned page, a 2135 // fresh chunk will be allocated. 2136 bool SetUp(); 2137 2138 // Returns true if the space has been successfully set up and not 2139 // subsequently torn down. 2140 bool HasBeenSetUp(); 2141 2142 // Checks whether an object/address is in this space. 2143 inline bool Contains(Address a); 2144 inline bool Contains(Object* o); 2145 bool ContainsSlow(Address addr); 2146 2147 // Given an address occupied by a live object, return that object if it is 2148 // in this space, or a Smi if it is not. The implementation iterates over 2149 // objects in the page containing the address, the cost is linear in the 2150 // number of objects in the page. It may be slow. 2151 Object* FindObject(Address addr); 2152 2153 // During boot the free_space_map is created, and afterwards we may need 2154 // to write it into the free list nodes that were already created. 2155 void RepairFreeListsAfterDeserialization(); 2156 2157 // Prepares for a mark-compact GC. 2158 void PrepareForMarkCompact(); 2159 2160 // Current capacity without growing (Size() + Available()). Capacity()2161 intptr_t Capacity() { return accounting_stats_.Capacity(); } 2162 2163 // Approximate amount of physical memory committed for this space. 2164 size_t CommittedPhysicalMemory() override; 2165 2166 void ResetFreeListStatistics(); 2167 2168 // Sets the capacity, the available space and the wasted space to zero. 2169 // The stats are rebuilt during sweeping by adding each page to the 2170 // capacity and the size when it is encountered. As free spaces are 2171 // discovered during the sweeping they are subtracted from the size and added 2172 // to the available and wasted totals. ClearStats()2173 void ClearStats() { 2174 accounting_stats_.ClearSize(); 2175 free_list_.ResetStats(); 2176 ResetFreeListStatistics(); 2177 } 2178 2179 // Available bytes without growing. These are the bytes on the free list. 2180 // The bytes in the linear allocation area are not included in this total 2181 // because updating the stats would slow down allocation. New pages are 2182 // immediately added to the free list so they show up here. Available()2183 intptr_t Available() override { return free_list_.Available(); } 2184 2185 // Allocated bytes in this space. Garbage bytes that were not found due to 2186 // concurrent sweeping are counted as being allocated! The bytes in the 2187 // current linear allocation area (between top and limit) are also counted 2188 // here. Size()2189 intptr_t Size() override { return accounting_stats_.Size(); } 2190 2191 // As size, but the bytes in lazily swept pages are estimated and the bytes 2192 // in the current linear allocation area are not included. 2193 intptr_t SizeOfObjects() override; 2194 2195 // Wasted bytes in this space. These are just the bytes that were thrown away 2196 // due to being too small to use for allocation. Waste()2197 virtual intptr_t Waste() { return free_list_.wasted_bytes(); } 2198 2199 // Returns the allocation pointer in this space. top()2200 Address top() { return allocation_info_.top(); } limit()2201 Address limit() { return allocation_info_.limit(); } 2202 2203 // The allocation top address. allocation_top_address()2204 Address* allocation_top_address() { return allocation_info_.top_address(); } 2205 2206 // The allocation limit address. allocation_limit_address()2207 Address* allocation_limit_address() { 2208 return allocation_info_.limit_address(); 2209 } 2210 2211 enum UpdateSkipList { UPDATE_SKIP_LIST, IGNORE_SKIP_LIST }; 2212 2213 // Allocate the requested number of bytes in the space if possible, return a 2214 // failure object if not. Only use IGNORE_SKIP_LIST if the skip list is going 2215 // to be manually updated later. 2216 MUST_USE_RESULT inline AllocationResult AllocateRawUnaligned( 2217 int size_in_bytes, UpdateSkipList update_skip_list = UPDATE_SKIP_LIST); 2218 2219 MUST_USE_RESULT inline AllocationResult AllocateRawUnalignedSynchronized( 2220 int size_in_bytes); 2221 2222 // Allocate the requested number of bytes in the space double aligned if 2223 // possible, return a failure object if not. 2224 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 2225 int size_in_bytes, AllocationAlignment alignment); 2226 2227 // Allocate the requested number of bytes in the space and consider allocation 2228 // alignment if needed. 2229 MUST_USE_RESULT inline AllocationResult AllocateRaw( 2230 int size_in_bytes, AllocationAlignment alignment); 2231 2232 // Give a block of memory to the space's free list. It might be added to 2233 // the free list or accounted as waste. 2234 // If add_to_freelist is false then just accounting stats are updated and 2235 // no attempt to add area to free list is made. Free(Address start,int size_in_bytes)2236 int Free(Address start, int size_in_bytes) { 2237 int wasted = free_list_.Free(start, size_in_bytes, kLinkCategory); 2238 accounting_stats_.DeallocateBytes(size_in_bytes); 2239 return size_in_bytes - wasted; 2240 } 2241 UnaccountedFree(Address start,int size_in_bytes)2242 int UnaccountedFree(Address start, int size_in_bytes) { 2243 int wasted = free_list_.Free(start, size_in_bytes, kDoNotLinkCategory); 2244 return size_in_bytes - wasted; 2245 } 2246 ResetFreeList()2247 void ResetFreeList() { free_list_.Reset(); } 2248 2249 // Set space allocation info. SetTopAndLimit(Address top,Address limit)2250 void SetTopAndLimit(Address top, Address limit) { 2251 DCHECK(top == limit || 2252 Page::FromAddress(top) == Page::FromAddress(limit - 1)); 2253 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 2254 allocation_info_.Reset(top, limit); 2255 } 2256 2257 // Empty space allocation info, returning unused area to free list. EmptyAllocationInfo()2258 void EmptyAllocationInfo() { 2259 // Mark the old linear allocation area with a free space map so it can be 2260 // skipped when scanning the heap. 2261 int old_linear_size = static_cast<int>(limit() - top()); 2262 Free(top(), old_linear_size); 2263 SetTopAndLimit(NULL, NULL); 2264 } 2265 Allocate(int bytes)2266 void Allocate(int bytes) { accounting_stats_.AllocateBytes(bytes); } 2267 2268 void IncreaseCapacity(int size); 2269 2270 // Releases an unused page and shrinks the space. 2271 void ReleasePage(Page* page); 2272 2273 // The dummy page that anchors the linked list of pages. anchor()2274 Page* anchor() { return &anchor_; } 2275 2276 // Collect code size related statistics 2277 void CollectCodeStatistics(); 2278 2279 // Reset code size related statistics 2280 static void ResetCodeAndMetadataStatistics(Isolate* isolate); 2281 2282 #ifdef VERIFY_HEAP 2283 // Verify integrity of this space. 2284 virtual void Verify(ObjectVisitor* visitor); 2285 2286 // Overridden by subclasses to verify space-specific object 2287 // properties (e.g., only maps or free-list nodes are in map space). VerifyObject(HeapObject * obj)2288 virtual void VerifyObject(HeapObject* obj) {} 2289 #endif 2290 2291 #ifdef DEBUG 2292 // Print meta info and objects in this space. 2293 void Print() override; 2294 2295 // Reports statistics for the space 2296 void ReportStatistics(); 2297 2298 // Report code object related statistics 2299 static void ReportCodeStatistics(Isolate* isolate); 2300 static void ResetCodeStatistics(Isolate* isolate); 2301 #endif 2302 FirstPage()2303 Page* FirstPage() { return anchor_.next_page(); } LastPage()2304 Page* LastPage() { return anchor_.prev_page(); } 2305 2306 void EvictEvacuationCandidatesFromLinearAllocationArea(); 2307 2308 bool CanExpand(size_t size); 2309 2310 // Returns the number of total pages in this space. 2311 int CountTotalPages(); 2312 2313 // Return size of allocatable area on a page in this space. AreaSize()2314 inline int AreaSize() { return area_size_; } 2315 is_local()2316 virtual bool is_local() { return false; } 2317 2318 // Merges {other} into the current space. Note that this modifies {other}, 2319 // e.g., removes its bump pointer area and resets statistics. 2320 void MergeCompactionSpace(CompactionSpace* other); 2321 2322 // Refills the free list from the corresponding free list filled by the 2323 // sweeper. 2324 virtual void RefillFreeList(); 2325 free_list()2326 FreeList* free_list() { return &free_list_; } 2327 mutex()2328 base::Mutex* mutex() { return &space_mutex_; } 2329 2330 inline void UnlinkFreeListCategories(Page* page); 2331 inline intptr_t RelinkFreeListCategories(Page* page); 2332 begin()2333 iterator begin() { return iterator(anchor_.next_page()); } end()2334 iterator end() { return iterator(&anchor_); } 2335 2336 protected: 2337 // PagedSpaces that should be included in snapshots have different, i.e., 2338 // smaller, initial pages. snapshotable()2339 virtual bool snapshotable() { return true; } 2340 HasPages()2341 bool HasPages() { return anchor_.next_page() != &anchor_; } 2342 2343 // Cleans up the space, frees all pages in this space except those belonging 2344 // to the initial chunk, uncommits addresses in the initial chunk. 2345 void TearDown(); 2346 2347 // Expands the space by allocating a fixed number of pages. Returns false if 2348 // it cannot allocate requested number of pages from OS, or if the hard heap 2349 // size limit has been hit. 2350 bool Expand(); 2351 2352 // Generic fast case allocation function that tries linear allocation at the 2353 // address denoted by top in allocation_info_. 2354 inline HeapObject* AllocateLinearly(int size_in_bytes); 2355 2356 // Generic fast case allocation function that tries aligned linear allocation 2357 // at the address denoted by top in allocation_info_. Writes the aligned 2358 // allocation size, which includes the filler size, to size_in_bytes. 2359 inline HeapObject* AllocateLinearlyAligned(int* size_in_bytes, 2360 AllocationAlignment alignment); 2361 2362 // If sweeping is still in progress try to sweep unswept pages. If that is 2363 // not successful, wait for the sweeper threads and re-try free-list 2364 // allocation. 2365 MUST_USE_RESULT virtual HeapObject* SweepAndRetryAllocation( 2366 int size_in_bytes); 2367 2368 // Slow path of AllocateRaw. This function is space-dependent. 2369 MUST_USE_RESULT HeapObject* SlowAllocateRaw(int size_in_bytes); 2370 2371 int area_size_; 2372 2373 // Accounting information for this space. 2374 AllocationStats accounting_stats_; 2375 2376 // The dummy page that anchors the double linked list of pages. 2377 Page anchor_; 2378 2379 // The space's free list. 2380 FreeList free_list_; 2381 2382 // Normal allocation information. 2383 AllocationInfo allocation_info_; 2384 2385 // Mutex guarding any concurrent access to the space. 2386 base::Mutex space_mutex_; 2387 2388 friend class IncrementalMarking; 2389 friend class MarkCompactCollector; 2390 2391 // Used in cctest. 2392 friend class HeapTester; 2393 }; 2394 2395 2396 class NumberAndSizeInfo BASE_EMBEDDED { 2397 public: NumberAndSizeInfo()2398 NumberAndSizeInfo() : number_(0), bytes_(0) {} 2399 number()2400 int number() const { return number_; } increment_number(int num)2401 void increment_number(int num) { number_ += num; } 2402 bytes()2403 int bytes() const { return bytes_; } increment_bytes(int size)2404 void increment_bytes(int size) { bytes_ += size; } 2405 clear()2406 void clear() { 2407 number_ = 0; 2408 bytes_ = 0; 2409 } 2410 2411 private: 2412 int number_; 2413 int bytes_; 2414 }; 2415 2416 2417 // HistogramInfo class for recording a single "bar" of a histogram. This 2418 // class is used for collecting statistics to print to the log file. 2419 class HistogramInfo : public NumberAndSizeInfo { 2420 public: HistogramInfo()2421 HistogramInfo() : NumberAndSizeInfo() {} 2422 name()2423 const char* name() { return name_; } set_name(const char * name)2424 void set_name(const char* name) { name_ = name; } 2425 2426 private: 2427 const char* name_; 2428 }; 2429 2430 enum SemiSpaceId { kFromSpace = 0, kToSpace = 1 }; 2431 2432 // ----------------------------------------------------------------------------- 2433 // SemiSpace in young generation 2434 // 2435 // A SemiSpace is a contiguous chunk of memory holding page-like memory chunks. 2436 // The mark-compact collector uses the memory of the first page in the from 2437 // space as a marking stack when tracing live objects. 2438 class SemiSpace : public Space { 2439 public: 2440 typedef PageIterator iterator; 2441 2442 static void Swap(SemiSpace* from, SemiSpace* to); 2443 SemiSpace(Heap * heap,SemiSpaceId semispace)2444 SemiSpace(Heap* heap, SemiSpaceId semispace) 2445 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2446 current_capacity_(0), 2447 maximum_capacity_(0), 2448 minimum_capacity_(0), 2449 age_mark_(nullptr), 2450 committed_(false), 2451 id_(semispace), 2452 anchor_(this), 2453 current_page_(nullptr), 2454 pages_used_(0) {} 2455 2456 inline bool Contains(HeapObject* o); 2457 inline bool Contains(Object* o); 2458 inline bool ContainsSlow(Address a); 2459 2460 void SetUp(int initial_capacity, int maximum_capacity); 2461 void TearDown(); HasBeenSetUp()2462 bool HasBeenSetUp() { return maximum_capacity_ != 0; } 2463 2464 bool Commit(); 2465 bool Uncommit(); is_committed()2466 bool is_committed() { return committed_; } 2467 2468 // Grow the semispace to the new capacity. The new capacity requested must 2469 // be larger than the current capacity and less than the maximum capacity. 2470 bool GrowTo(int new_capacity); 2471 2472 // Shrinks the semispace to the new capacity. The new capacity requested 2473 // must be more than the amount of used memory in the semispace and less 2474 // than the current capacity. 2475 bool ShrinkTo(int new_capacity); 2476 2477 bool EnsureCurrentCapacity(); 2478 2479 // Returns the start address of the first page of the space. space_start()2480 Address space_start() { 2481 DCHECK_NE(anchor_.next_page(), anchor()); 2482 return anchor_.next_page()->area_start(); 2483 } 2484 first_page()2485 Page* first_page() { return anchor_.next_page(); } current_page()2486 Page* current_page() { return current_page_; } pages_used()2487 int pages_used() { return pages_used_; } 2488 2489 // Returns one past the end address of the space. space_end()2490 Address space_end() { return anchor_.prev_page()->area_end(); } 2491 2492 // Returns the start address of the current page of the space. page_low()2493 Address page_low() { return current_page_->area_start(); } 2494 2495 // Returns one past the end address of the current page of the space. page_high()2496 Address page_high() { return current_page_->area_end(); } 2497 AdvancePage()2498 bool AdvancePage() { 2499 Page* next_page = current_page_->next_page(); 2500 // We cannot expand if we reached the maximum number of pages already. Note 2501 // that we need to account for the next page already for this check as we 2502 // could potentially fill the whole page after advancing. 2503 const bool reached_max_pages = (pages_used_ + 1) == max_pages(); 2504 if (next_page == anchor() || reached_max_pages) { 2505 return false; 2506 } 2507 current_page_ = next_page; 2508 pages_used_++; 2509 return true; 2510 } 2511 2512 // Resets the space to using the first page. 2513 void Reset(); 2514 2515 void RemovePage(Page* page); 2516 void PrependPage(Page* page); 2517 2518 // Age mark accessors. age_mark()2519 Address age_mark() { return age_mark_; } 2520 void set_age_mark(Address mark); 2521 2522 // Returns the current capacity of the semispace. current_capacity()2523 int current_capacity() { return current_capacity_; } 2524 2525 // Returns the maximum capacity of the semispace. maximum_capacity()2526 int maximum_capacity() { return maximum_capacity_; } 2527 2528 // Returns the initial capacity of the semispace. minimum_capacity()2529 int minimum_capacity() { return minimum_capacity_; } 2530 id()2531 SemiSpaceId id() { return id_; } 2532 2533 // Approximate amount of physical memory committed for this space. 2534 size_t CommittedPhysicalMemory() override; 2535 2536 // If we don't have these here then SemiSpace will be abstract. However 2537 // they should never be called: 2538 Size()2539 intptr_t Size() override { 2540 UNREACHABLE(); 2541 return 0; 2542 } 2543 SizeOfObjects()2544 intptr_t SizeOfObjects() override { return Size(); } 2545 Available()2546 intptr_t Available() override { 2547 UNREACHABLE(); 2548 return 0; 2549 } 2550 2551 #ifdef DEBUG 2552 void Print() override; 2553 // Validate a range of of addresses in a SemiSpace. 2554 // The "from" address must be on a page prior to the "to" address, 2555 // in the linked page order, or it must be earlier on the same page. 2556 static void AssertValidRange(Address from, Address to); 2557 #else 2558 // Do nothing. AssertValidRange(Address from,Address to)2559 inline static void AssertValidRange(Address from, Address to) {} 2560 #endif 2561 2562 #ifdef VERIFY_HEAP 2563 virtual void Verify(); 2564 #endif 2565 begin()2566 iterator begin() { return iterator(anchor_.next_page()); } end()2567 iterator end() { return iterator(anchor()); } 2568 2569 private: 2570 void RewindPages(Page* start, int num_pages); 2571 anchor()2572 inline Page* anchor() { return &anchor_; } max_pages()2573 inline int max_pages() { return current_capacity_ / Page::kPageSize; } 2574 2575 // Copies the flags into the masked positions on all pages in the space. 2576 void FixPagesFlags(intptr_t flags, intptr_t flag_mask); 2577 2578 // The currently committed space capacity. 2579 int current_capacity_; 2580 2581 // The maximum capacity that can be used by this space. A space cannot grow 2582 // beyond that size. 2583 int maximum_capacity_; 2584 2585 // The minimum capacity for the space. A space cannot shrink below this size. 2586 int minimum_capacity_; 2587 2588 // Used to govern object promotion during mark-compact collection. 2589 Address age_mark_; 2590 2591 bool committed_; 2592 SemiSpaceId id_; 2593 2594 Page anchor_; 2595 Page* current_page_; 2596 int pages_used_; 2597 2598 friend class NewSpace; 2599 friend class SemiSpaceIterator; 2600 }; 2601 2602 2603 // A SemiSpaceIterator is an ObjectIterator that iterates over the active 2604 // semispace of the heap's new space. It iterates over the objects in the 2605 // semispace from a given start address (defaulting to the bottom of the 2606 // semispace) to the top of the semispace. New objects allocated after the 2607 // iterator is created are not iterated. 2608 class SemiSpaceIterator : public ObjectIterator { 2609 public: 2610 // Create an iterator over the allocated objects in the given to-space. 2611 explicit SemiSpaceIterator(NewSpace* space); 2612 2613 inline HeapObject* Next() override; 2614 2615 private: 2616 void Initialize(Address start, Address end); 2617 2618 // The current iteration point. 2619 Address current_; 2620 // The end of iteration. 2621 Address limit_; 2622 }; 2623 2624 // ----------------------------------------------------------------------------- 2625 // The young generation space. 2626 // 2627 // The new space consists of a contiguous pair of semispaces. It simply 2628 // forwards most functions to the appropriate semispace. 2629 2630 class NewSpace : public Space { 2631 public: 2632 typedef PageIterator iterator; 2633 NewSpace(Heap * heap)2634 explicit NewSpace(Heap* heap) 2635 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2636 to_space_(heap, kToSpace), 2637 from_space_(heap, kFromSpace), 2638 reservation_(), 2639 top_on_previous_step_(0), 2640 allocated_histogram_(nullptr), 2641 promoted_histogram_(nullptr) {} 2642 2643 inline bool Contains(HeapObject* o); 2644 inline bool ContainsSlow(Address a); 2645 inline bool Contains(Object* o); 2646 2647 bool SetUp(int initial_semispace_capacity, int max_semispace_capacity); 2648 2649 // Tears down the space. Heap memory was not allocated by the space, so it 2650 // is not deallocated here. 2651 void TearDown(); 2652 2653 // True if the space has been set up but not torn down. HasBeenSetUp()2654 bool HasBeenSetUp() { 2655 return to_space_.HasBeenSetUp() && from_space_.HasBeenSetUp(); 2656 } 2657 2658 // Flip the pair of spaces. 2659 void Flip(); 2660 2661 // Grow the capacity of the semispaces. Assumes that they are not at 2662 // their maximum capacity. 2663 void Grow(); 2664 2665 // Shrink the capacity of the semispaces. 2666 void Shrink(); 2667 2668 // Return the allocated bytes in the active semispace. Size()2669 intptr_t Size() override { 2670 return to_space_.pages_used() * Page::kAllocatableMemory + 2671 static_cast<int>(top() - to_space_.page_low()); 2672 } 2673 2674 // The same, but returning an int. We have to have the one that returns 2675 // intptr_t because it is inherited, but if we know we are dealing with the 2676 // new space, which can't get as big as the other spaces then this is useful: SizeAsInt()2677 int SizeAsInt() { return static_cast<int>(Size()); } 2678 2679 // Return the allocatable capacity of a semispace. Capacity()2680 intptr_t Capacity() { 2681 SLOW_DCHECK(to_space_.current_capacity() == from_space_.current_capacity()); 2682 return (to_space_.current_capacity() / Page::kPageSize) * 2683 Page::kAllocatableMemory; 2684 } 2685 2686 // Return the current size of a semispace, allocatable and non-allocatable 2687 // memory. TotalCapacity()2688 intptr_t TotalCapacity() { 2689 DCHECK(to_space_.current_capacity() == from_space_.current_capacity()); 2690 return to_space_.current_capacity(); 2691 } 2692 2693 // Committed memory for NewSpace is the committed memory of both semi-spaces 2694 // combined. CommittedMemory()2695 intptr_t CommittedMemory() override { 2696 return from_space_.CommittedMemory() + to_space_.CommittedMemory(); 2697 } 2698 MaximumCommittedMemory()2699 intptr_t MaximumCommittedMemory() override { 2700 return from_space_.MaximumCommittedMemory() + 2701 to_space_.MaximumCommittedMemory(); 2702 } 2703 2704 // Approximate amount of physical memory committed for this space. 2705 size_t CommittedPhysicalMemory() override; 2706 2707 // Return the available bytes without growing. Available()2708 intptr_t Available() override { return Capacity() - Size(); } 2709 AllocatedSinceLastGC()2710 size_t AllocatedSinceLastGC() { 2711 bool seen_age_mark = false; 2712 Address age_mark = to_space_.age_mark(); 2713 Page* current_page = to_space_.first_page(); 2714 Page* age_mark_page = Page::FromAddress(age_mark); 2715 Page* last_page = Page::FromAddress(top() - kPointerSize); 2716 if (age_mark_page == last_page) { 2717 if (top() - age_mark >= 0) { 2718 return top() - age_mark; 2719 } 2720 // Top was reset at some point, invalidating this metric. 2721 return 0; 2722 } 2723 while (current_page != last_page) { 2724 if (current_page == age_mark_page) { 2725 seen_age_mark = true; 2726 break; 2727 } 2728 current_page = current_page->next_page(); 2729 } 2730 if (!seen_age_mark) { 2731 // Top was reset at some point, invalidating this metric. 2732 return 0; 2733 } 2734 intptr_t allocated = age_mark_page->area_end() - age_mark; 2735 DCHECK_EQ(current_page, age_mark_page); 2736 current_page = age_mark_page->next_page(); 2737 while (current_page != last_page) { 2738 allocated += Page::kAllocatableMemory; 2739 current_page = current_page->next_page(); 2740 } 2741 allocated += top() - current_page->area_start(); 2742 DCHECK_LE(0, allocated); 2743 DCHECK_LE(allocated, Size()); 2744 return static_cast<size_t>(allocated); 2745 } 2746 MovePageFromSpaceToSpace(Page * page)2747 void MovePageFromSpaceToSpace(Page* page) { 2748 DCHECK(page->InFromSpace()); 2749 from_space_.RemovePage(page); 2750 to_space_.PrependPage(page); 2751 } 2752 2753 bool Rebalance(); 2754 2755 // Return the maximum capacity of a semispace. MaximumCapacity()2756 int MaximumCapacity() { 2757 DCHECK(to_space_.maximum_capacity() == from_space_.maximum_capacity()); 2758 return to_space_.maximum_capacity(); 2759 } 2760 IsAtMaximumCapacity()2761 bool IsAtMaximumCapacity() { return TotalCapacity() == MaximumCapacity(); } 2762 2763 // Returns the initial capacity of a semispace. InitialTotalCapacity()2764 int InitialTotalCapacity() { 2765 DCHECK(to_space_.minimum_capacity() == from_space_.minimum_capacity()); 2766 return to_space_.minimum_capacity(); 2767 } 2768 2769 // Return the address of the allocation pointer in the active semispace. top()2770 Address top() { 2771 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.top())); 2772 return allocation_info_.top(); 2773 } 2774 2775 // Return the address of the allocation pointer limit in the active semispace. limit()2776 Address limit() { 2777 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.limit())); 2778 return allocation_info_.limit(); 2779 } 2780 2781 // Return the address of the first object in the active semispace. bottom()2782 Address bottom() { return to_space_.space_start(); } 2783 2784 // Get the age mark of the inactive semispace. age_mark()2785 Address age_mark() { return from_space_.age_mark(); } 2786 // Set the age mark in the active semispace. set_age_mark(Address mark)2787 void set_age_mark(Address mark) { to_space_.set_age_mark(mark); } 2788 2789 // The allocation top and limit address. allocation_top_address()2790 Address* allocation_top_address() { return allocation_info_.top_address(); } 2791 2792 // The allocation limit address. allocation_limit_address()2793 Address* allocation_limit_address() { 2794 return allocation_info_.limit_address(); 2795 } 2796 2797 MUST_USE_RESULT INLINE(AllocationResult AllocateRawAligned( 2798 int size_in_bytes, AllocationAlignment alignment)); 2799 2800 MUST_USE_RESULT INLINE( 2801 AllocationResult AllocateRawUnaligned(int size_in_bytes)); 2802 2803 MUST_USE_RESULT INLINE(AllocationResult AllocateRaw( 2804 int size_in_bytes, AllocationAlignment alignment)); 2805 2806 MUST_USE_RESULT inline AllocationResult AllocateRawSynchronized( 2807 int size_in_bytes, AllocationAlignment alignment); 2808 2809 // Reset the allocation pointer to the beginning of the active semispace. 2810 void ResetAllocationInfo(); 2811 2812 // When inline allocation stepping is active, either because of incremental 2813 // marking, idle scavenge, or allocation statistics gathering, we 'interrupt' 2814 // inline allocation every once in a while. This is done by setting 2815 // allocation_info_.limit to be lower than the actual limit and and increasing 2816 // it in steps to guarantee that the observers are notified periodically. 2817 void UpdateInlineAllocationLimit(int size_in_bytes); 2818 DisableInlineAllocationSteps()2819 void DisableInlineAllocationSteps() { 2820 top_on_previous_step_ = 0; 2821 UpdateInlineAllocationLimit(0); 2822 } 2823 2824 // Allows observation of inline allocation. The observer->Step() method gets 2825 // called after every step_size bytes have been allocated (approximately). 2826 // This works by adjusting the allocation limit to a lower value and adjusting 2827 // it after each step. 2828 void AddAllocationObserver(AllocationObserver* observer) override; 2829 2830 void RemoveAllocationObserver(AllocationObserver* observer) override; 2831 2832 // Get the extent of the inactive semispace (for use as a marking stack, 2833 // or to zap it). Notice: space-addresses are not necessarily on the 2834 // same page, so FromSpaceStart() might be above FromSpaceEnd(). FromSpacePageLow()2835 Address FromSpacePageLow() { return from_space_.page_low(); } FromSpacePageHigh()2836 Address FromSpacePageHigh() { return from_space_.page_high(); } FromSpaceStart()2837 Address FromSpaceStart() { return from_space_.space_start(); } FromSpaceEnd()2838 Address FromSpaceEnd() { return from_space_.space_end(); } 2839 2840 // Get the extent of the active semispace's pages' memory. ToSpaceStart()2841 Address ToSpaceStart() { return to_space_.space_start(); } ToSpaceEnd()2842 Address ToSpaceEnd() { return to_space_.space_end(); } 2843 2844 inline bool ToSpaceContainsSlow(Address a); 2845 inline bool FromSpaceContainsSlow(Address a); 2846 inline bool ToSpaceContains(Object* o); 2847 inline bool FromSpaceContains(Object* o); 2848 2849 // Try to switch the active semispace to a new, empty, page. 2850 // Returns false if this isn't possible or reasonable (i.e., there 2851 // are no pages, or the current page is already empty), or true 2852 // if successful. 2853 bool AddFreshPage(); 2854 bool AddFreshPageSynchronized(); 2855 2856 #ifdef VERIFY_HEAP 2857 // Verify the active semispace. 2858 virtual void Verify(); 2859 #endif 2860 2861 #ifdef DEBUG 2862 // Print the active semispace. Print()2863 void Print() override { to_space_.Print(); } 2864 #endif 2865 2866 // Iterates the active semispace to collect statistics. 2867 void CollectStatistics(); 2868 // Reports previously collected statistics of the active semispace. 2869 void ReportStatistics(); 2870 // Clears previously collected statistics. 2871 void ClearHistograms(); 2872 2873 // Record the allocation or promotion of a heap object. Note that we don't 2874 // record every single allocation, but only those that happen in the 2875 // to space during a scavenge GC. 2876 void RecordAllocation(HeapObject* obj); 2877 void RecordPromotion(HeapObject* obj); 2878 2879 // Return whether the operation succeded. CommitFromSpaceIfNeeded()2880 bool CommitFromSpaceIfNeeded() { 2881 if (from_space_.is_committed()) return true; 2882 return from_space_.Commit(); 2883 } 2884 UncommitFromSpace()2885 bool UncommitFromSpace() { 2886 if (!from_space_.is_committed()) return true; 2887 return from_space_.Uncommit(); 2888 } 2889 IsFromSpaceCommitted()2890 bool IsFromSpaceCommitted() { return from_space_.is_committed(); } 2891 active_space()2892 SemiSpace* active_space() { return &to_space_; } 2893 2894 void PauseAllocationObservers() override; 2895 void ResumeAllocationObservers() override; 2896 begin()2897 iterator begin() { return to_space_.begin(); } end()2898 iterator end() { return to_space_.end(); } 2899 2900 private: 2901 // Update allocation info to match the current to-space page. 2902 void UpdateAllocationInfo(); 2903 2904 base::Mutex mutex_; 2905 2906 // The semispaces. 2907 SemiSpace to_space_; 2908 SemiSpace from_space_; 2909 base::VirtualMemory reservation_; 2910 2911 // Allocation pointer and limit for normal allocation and allocation during 2912 // mark-compact collection. 2913 AllocationInfo allocation_info_; 2914 2915 Address top_on_previous_step_; 2916 2917 HistogramInfo* allocated_histogram_; 2918 HistogramInfo* promoted_histogram_; 2919 2920 bool EnsureAllocation(int size_in_bytes, AllocationAlignment alignment); 2921 2922 // If we are doing inline allocation in steps, this method performs the 'step' 2923 // operation. top is the memory address of the bump pointer at the last 2924 // inline allocation (i.e. it determines the numbers of bytes actually 2925 // allocated since the last step.) new_top is the address of the bump pointer 2926 // where the next byte is going to be allocated from. top and new_top may be 2927 // different when we cross a page boundary or reset the space. 2928 void InlineAllocationStep(Address top, Address new_top, Address soon_object, 2929 size_t size); 2930 intptr_t GetNextInlineAllocationStepSize(); 2931 void StartNextInlineAllocationStep(); 2932 2933 friend class SemiSpaceIterator; 2934 }; 2935 2936 class PauseAllocationObserversScope { 2937 public: 2938 explicit PauseAllocationObserversScope(Heap* heap); 2939 ~PauseAllocationObserversScope(); 2940 2941 private: 2942 Heap* heap_; 2943 DISALLOW_COPY_AND_ASSIGN(PauseAllocationObserversScope); 2944 }; 2945 2946 // ----------------------------------------------------------------------------- 2947 // Compaction space that is used temporarily during compaction. 2948 2949 class CompactionSpace : public PagedSpace { 2950 public: CompactionSpace(Heap * heap,AllocationSpace id,Executability executable)2951 CompactionSpace(Heap* heap, AllocationSpace id, Executability executable) 2952 : PagedSpace(heap, id, executable) {} 2953 is_local()2954 bool is_local() override { return true; } 2955 2956 protected: 2957 // The space is temporary and not included in any snapshots. snapshotable()2958 bool snapshotable() override { return false; } 2959 2960 MUST_USE_RESULT HeapObject* SweepAndRetryAllocation( 2961 int size_in_bytes) override; 2962 }; 2963 2964 2965 // A collection of |CompactionSpace|s used by a single compaction task. 2966 class CompactionSpaceCollection : public Malloced { 2967 public: CompactionSpaceCollection(Heap * heap)2968 explicit CompactionSpaceCollection(Heap* heap) 2969 : old_space_(heap, OLD_SPACE, Executability::NOT_EXECUTABLE), 2970 code_space_(heap, CODE_SPACE, Executability::EXECUTABLE) {} 2971 Get(AllocationSpace space)2972 CompactionSpace* Get(AllocationSpace space) { 2973 switch (space) { 2974 case OLD_SPACE: 2975 return &old_space_; 2976 case CODE_SPACE: 2977 return &code_space_; 2978 default: 2979 UNREACHABLE(); 2980 } 2981 UNREACHABLE(); 2982 return nullptr; 2983 } 2984 2985 private: 2986 CompactionSpace old_space_; 2987 CompactionSpace code_space_; 2988 }; 2989 2990 2991 // ----------------------------------------------------------------------------- 2992 // Old object space (includes the old space of objects and code space) 2993 2994 class OldSpace : public PagedSpace { 2995 public: 2996 // Creates an old space object. The constructor does not allocate pages 2997 // from OS. OldSpace(Heap * heap,AllocationSpace id,Executability executable)2998 OldSpace(Heap* heap, AllocationSpace id, Executability executable) 2999 : PagedSpace(heap, id, executable) {} 3000 }; 3001 3002 3003 // For contiguous spaces, top should be in the space (or at the end) and limit 3004 // should be the end of the space. 3005 #define DCHECK_SEMISPACE_ALLOCATION_INFO(info, space) \ 3006 SLOW_DCHECK((space).page_low() <= (info).top() && \ 3007 (info).top() <= (space).page_high() && \ 3008 (info).limit() <= (space).page_high()) 3009 3010 3011 // ----------------------------------------------------------------------------- 3012 // Old space for all map objects 3013 3014 class MapSpace : public PagedSpace { 3015 public: 3016 // Creates a map space object. MapSpace(Heap * heap,AllocationSpace id)3017 MapSpace(Heap* heap, AllocationSpace id) 3018 : PagedSpace(heap, id, NOT_EXECUTABLE) {} 3019 RoundSizeDownToObjectAlignment(int size)3020 int RoundSizeDownToObjectAlignment(int size) override { 3021 if (base::bits::IsPowerOfTwo32(Map::kSize)) { 3022 return RoundDown(size, Map::kSize); 3023 } else { 3024 return (size / Map::kSize) * Map::kSize; 3025 } 3026 } 3027 3028 #ifdef VERIFY_HEAP 3029 void VerifyObject(HeapObject* obj) override; 3030 #endif 3031 }; 3032 3033 3034 // ----------------------------------------------------------------------------- 3035 // Large objects ( > Page::kMaxRegularHeapObjectSize ) are allocated and 3036 // managed by the large object space. A large object is allocated from OS 3037 // heap with extra padding bytes (Page::kPageSize + Page::kObjectStartOffset). 3038 // A large object always starts at Page::kObjectStartOffset to a page. 3039 // Large objects do not move during garbage collections. 3040 3041 class LargeObjectSpace : public Space { 3042 public: 3043 typedef LargePageIterator iterator; 3044 3045 LargeObjectSpace(Heap* heap, AllocationSpace id); 3046 virtual ~LargeObjectSpace(); 3047 3048 // Initializes internal data structures. 3049 bool SetUp(); 3050 3051 // Releases internal resources, frees objects in this space. 3052 void TearDown(); 3053 ObjectSizeFor(intptr_t chunk_size)3054 static intptr_t ObjectSizeFor(intptr_t chunk_size) { 3055 if (chunk_size <= (Page::kPageSize + Page::kObjectStartOffset)) return 0; 3056 return chunk_size - Page::kPageSize - Page::kObjectStartOffset; 3057 } 3058 3059 // Shared implementation of AllocateRaw, AllocateRawCode and 3060 // AllocateRawFixedArray. 3061 MUST_USE_RESULT AllocationResult 3062 AllocateRaw(int object_size, Executability executable); 3063 3064 // Available bytes for objects in this space. 3065 inline intptr_t Available() override; 3066 Size()3067 intptr_t Size() override { return size_; } 3068 SizeOfObjects()3069 intptr_t SizeOfObjects() override { return objects_size_; } 3070 3071 // Approximate amount of physical memory committed for this space. 3072 size_t CommittedPhysicalMemory() override; 3073 PageCount()3074 int PageCount() { return page_count_; } 3075 3076 // Finds an object for a given address, returns a Smi if it is not found. 3077 // The function iterates through all objects in this space, may be slow. 3078 Object* FindObject(Address a); 3079 3080 // Finds a large object page containing the given address, returns NULL 3081 // if such a page doesn't exist. 3082 LargePage* FindPage(Address a); 3083 3084 // Clears the marking state of live objects. 3085 void ClearMarkingStateOfLiveObjects(); 3086 3087 // Frees unmarked objects. 3088 void FreeUnmarkedObjects(); 3089 3090 // Checks whether a heap object is in this space; O(1). 3091 bool Contains(HeapObject* obj); 3092 // Checks whether an address is in the object area in this space. Iterates 3093 // all objects in the space. May be slow. ContainsSlow(Address addr)3094 bool ContainsSlow(Address addr) { return FindObject(addr)->IsHeapObject(); } 3095 3096 // Checks whether the space is empty. IsEmpty()3097 bool IsEmpty() { return first_page_ == NULL; } 3098 AdjustLiveBytes(int by)3099 void AdjustLiveBytes(int by) { objects_size_ += by; } 3100 first_page()3101 LargePage* first_page() { return first_page_; } 3102 3103 // Collect code statistics. 3104 void CollectCodeStatistics(); 3105 begin()3106 iterator begin() { return iterator(first_page_); } end()3107 iterator end() { return iterator(nullptr); } 3108 3109 #ifdef VERIFY_HEAP 3110 virtual void Verify(); 3111 #endif 3112 3113 #ifdef DEBUG 3114 void Print() override; 3115 void ReportStatistics(); 3116 #endif 3117 3118 private: 3119 // The head of the linked list of large object chunks. 3120 LargePage* first_page_; 3121 intptr_t size_; // allocated bytes 3122 int page_count_; // number of chunks 3123 intptr_t objects_size_; // size of objects 3124 // Map MemoryChunk::kAlignment-aligned chunks to large pages covering them 3125 base::HashMap chunk_map_; 3126 3127 friend class LargeObjectIterator; 3128 }; 3129 3130 3131 class LargeObjectIterator : public ObjectIterator { 3132 public: 3133 explicit LargeObjectIterator(LargeObjectSpace* space); 3134 3135 HeapObject* Next() override; 3136 3137 private: 3138 LargePage* current_; 3139 }; 3140 3141 // Iterates over the chunks (pages and large object pages) that can contain 3142 // pointers to new space or to evacuation candidates. 3143 class MemoryChunkIterator BASE_EMBEDDED { 3144 public: 3145 inline explicit MemoryChunkIterator(Heap* heap); 3146 3147 // Return NULL when the iterator is done. 3148 inline MemoryChunk* next(); 3149 3150 private: 3151 enum State { 3152 kOldSpaceState, 3153 kMapState, 3154 kCodeState, 3155 kLargeObjectState, 3156 kFinishedState 3157 }; 3158 Heap* heap_; 3159 State state_; 3160 PageIterator old_iterator_; 3161 PageIterator code_iterator_; 3162 PageIterator map_iterator_; 3163 LargePageIterator lo_iterator_; 3164 }; 3165 3166 #ifdef DEBUG 3167 struct CommentStatistic { 3168 const char* comment; 3169 int size; 3170 int count; ClearCommentStatistic3171 void Clear() { 3172 comment = NULL; 3173 size = 0; 3174 count = 0; 3175 } 3176 // Must be small, since an iteration is used for lookup. 3177 static const int kMaxComments = 64; 3178 }; 3179 #endif 3180 } // namespace internal 3181 } // namespace v8 3182 3183 #endif // V8_HEAP_SPACES_H_ 3184