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 "src/allocation.h" 9 #include "src/atomic-utils.h" 10 #include "src/base/atomicops.h" 11 #include "src/base/bits.h" 12 #include "src/base/platform/mutex.h" 13 #include "src/flags.h" 14 #include "src/hashmap.h" 15 #include "src/list.h" 16 #include "src/objects.h" 17 #include "src/utils.h" 18 19 namespace v8 { 20 namespace internal { 21 22 class CompactionSpaceCollection; 23 class Isolate; 24 25 // ----------------------------------------------------------------------------- 26 // Heap structures: 27 // 28 // A JS heap consists of a young generation, an old generation, and a large 29 // object space. The young generation is divided into two semispaces. A 30 // scavenger implements Cheney's copying algorithm. The old generation is 31 // separated into a map space and an old object space. The map space contains 32 // all (and only) map objects, the rest of old objects go into the old space. 33 // The old generation is collected by a mark-sweep-compact collector. 34 // 35 // The semispaces of the young generation are contiguous. The old and map 36 // spaces consists of a list of pages. A page has a page header and an object 37 // area. 38 // 39 // There is a separate large object space for objects larger than 40 // Page::kMaxRegularHeapObjectSize, so that they do not have to move during 41 // collection. The large object space is paged. Pages in large object space 42 // may be larger than the page size. 43 // 44 // A store-buffer based write barrier is used to keep track of intergenerational 45 // references. See heap/store-buffer.h. 46 // 47 // During scavenges and mark-sweep collections we sometimes (after a store 48 // buffer overflow) iterate intergenerational pointers without decoding heap 49 // object maps so if the page belongs to old space or large object space 50 // it is essential to guarantee that the page does not contain any 51 // garbage pointers to new space: every pointer aligned word which satisfies 52 // the Heap::InNewSpace() predicate must be a pointer to a live heap object in 53 // new space. Thus objects in old space and large object spaces should have a 54 // special layout (e.g. no bare integer fields). This requirement does not 55 // apply to map space which is iterated in a special fashion. However we still 56 // require pointer fields of dead maps to be cleaned. 57 // 58 // To enable lazy cleaning of old space pages we can mark chunks of the page 59 // as being garbage. Garbage sections are marked with a special map. These 60 // sections are skipped when scanning the page, even if we are otherwise 61 // scanning without regard for object boundaries. Garbage sections are chained 62 // together to form a free list after a GC. Garbage sections created outside 63 // of GCs by object trunctation etc. may not be in the free list chain. Very 64 // small free spaces are ignored, they need only be cleaned of bogus pointers 65 // into new space. 66 // 67 // Each page may have up to one special garbage section. The start of this 68 // section is denoted by the top field in the space. The end of the section 69 // is denoted by the limit field in the space. This special garbage section 70 // is not marked with a free space map in the data. The point of this section 71 // is to enable linear allocation without having to constantly update the byte 72 // array every time the top field is updated and a new object is created. The 73 // special garbage section is not in the chain of garbage sections. 74 // 75 // Since the top and limit fields are in the space, not the page, only one page 76 // has a special garbage section, and if the top and limit are equal then there 77 // is no special garbage section. 78 79 // Some assertion macros used in the debugging mode. 80 81 #define DCHECK_PAGE_ALIGNED(address) \ 82 DCHECK((OffsetFrom(address) & Page::kPageAlignmentMask) == 0) 83 84 #define DCHECK_OBJECT_ALIGNED(address) \ 85 DCHECK((OffsetFrom(address) & kObjectAlignmentMask) == 0) 86 87 #define DCHECK_OBJECT_SIZE(size) \ 88 DCHECK((0 < size) && (size <= Page::kMaxRegularHeapObjectSize)) 89 90 #define DCHECK_CODEOBJECT_SIZE(size, code_space) \ 91 DCHECK((0 < size) && (size <= code_space->AreaSize())) 92 93 #define DCHECK_PAGE_OFFSET(offset) \ 94 DCHECK((Page::kObjectStartOffset <= offset) && (offset <= Page::kPageSize)) 95 96 #define DCHECK_MAP_PAGE_INDEX(index) \ 97 DCHECK((0 <= index) && (index <= MapSpace::kMaxMapPageIndex)) 98 99 class AllocationInfo; 100 class CompactionSpace; 101 class FreeList; 102 class MemoryAllocator; 103 class MemoryChunk; 104 class PagedSpace; 105 class Space; 106 107 class MarkBit { 108 public: 109 typedef uint32_t CellType; 110 111 inline MarkBit(CellType* cell, CellType mask) : cell_(cell), mask_(mask) {} 112 113 #ifdef DEBUG 114 bool operator==(const MarkBit& other) { 115 return cell_ == other.cell_ && mask_ == other.mask_; 116 } 117 #endif 118 119 private: 120 inline CellType* cell() { return cell_; } 121 inline CellType mask() { return mask_; } 122 123 inline MarkBit Next() { 124 CellType new_mask = mask_ << 1; 125 if (new_mask == 0) { 126 return MarkBit(cell_ + 1, 1); 127 } else { 128 return MarkBit(cell_, new_mask); 129 } 130 } 131 132 inline void Set() { *cell_ |= mask_; } 133 inline bool Get() { return (*cell_ & mask_) != 0; } 134 inline void Clear() { *cell_ &= ~mask_; } 135 136 CellType* cell_; 137 CellType mask_; 138 139 friend class Marking; 140 }; 141 142 143 // Bitmap is a sequence of cells each containing fixed number of bits. 144 class Bitmap { 145 public: 146 static const uint32_t kBitsPerCell = 32; 147 static const uint32_t kBitsPerCellLog2 = 5; 148 static const uint32_t kBitIndexMask = kBitsPerCell - 1; 149 static const uint32_t kBytesPerCell = kBitsPerCell / kBitsPerByte; 150 static const uint32_t kBytesPerCellLog2 = kBitsPerCellLog2 - kBitsPerByteLog2; 151 152 static const size_t kLength = (1 << kPageSizeBits) >> (kPointerSizeLog2); 153 154 static const size_t kSize = 155 (1 << kPageSizeBits) >> (kPointerSizeLog2 + kBitsPerByteLog2); 156 157 158 static int CellsForLength(int length) { 159 return (length + kBitsPerCell - 1) >> kBitsPerCellLog2; 160 } 161 162 int CellsCount() { return CellsForLength(kLength); } 163 164 static int SizeFor(int cells_count) { 165 return sizeof(MarkBit::CellType) * cells_count; 166 } 167 168 INLINE(static uint32_t IndexToCell(uint32_t index)) { 169 return index >> kBitsPerCellLog2; 170 } 171 172 V8_INLINE static uint32_t IndexInCell(uint32_t index) { 173 return index & kBitIndexMask; 174 } 175 176 INLINE(static uint32_t CellToIndex(uint32_t index)) { 177 return index << kBitsPerCellLog2; 178 } 179 180 INLINE(static uint32_t CellAlignIndex(uint32_t index)) { 181 return (index + kBitIndexMask) & ~kBitIndexMask; 182 } 183 184 INLINE(MarkBit::CellType* cells()) { 185 return reinterpret_cast<MarkBit::CellType*>(this); 186 } 187 188 INLINE(Address address()) { return reinterpret_cast<Address>(this); } 189 190 INLINE(static Bitmap* FromAddress(Address addr)) { 191 return reinterpret_cast<Bitmap*>(addr); 192 } 193 194 inline MarkBit MarkBitFromIndex(uint32_t index) { 195 MarkBit::CellType mask = 1u << IndexInCell(index); 196 MarkBit::CellType* cell = this->cells() + (index >> kBitsPerCellLog2); 197 return MarkBit(cell, mask); 198 } 199 200 static inline void Clear(MemoryChunk* chunk); 201 202 static void PrintWord(uint32_t word, uint32_t himask = 0) { 203 for (uint32_t mask = 1; mask != 0; mask <<= 1) { 204 if ((mask & himask) != 0) PrintF("["); 205 PrintF((mask & word) ? "1" : "0"); 206 if ((mask & himask) != 0) PrintF("]"); 207 } 208 } 209 210 class CellPrinter { 211 public: 212 CellPrinter() : seq_start(0), seq_type(0), seq_length(0) {} 213 214 void Print(uint32_t pos, uint32_t cell) { 215 if (cell == seq_type) { 216 seq_length++; 217 return; 218 } 219 220 Flush(); 221 222 if (IsSeq(cell)) { 223 seq_start = pos; 224 seq_length = 0; 225 seq_type = cell; 226 return; 227 } 228 229 PrintF("%d: ", pos); 230 PrintWord(cell); 231 PrintF("\n"); 232 } 233 234 void Flush() { 235 if (seq_length > 0) { 236 PrintF("%d: %dx%d\n", seq_start, seq_type == 0 ? 0 : 1, 237 seq_length * kBitsPerCell); 238 seq_length = 0; 239 } 240 } 241 242 static bool IsSeq(uint32_t cell) { return cell == 0 || cell == 0xFFFFFFFF; } 243 244 private: 245 uint32_t seq_start; 246 uint32_t seq_type; 247 uint32_t seq_length; 248 }; 249 250 void Print() { 251 CellPrinter printer; 252 for (int i = 0; i < CellsCount(); i++) { 253 printer.Print(i, cells()[i]); 254 } 255 printer.Flush(); 256 PrintF("\n"); 257 } 258 259 bool IsClean() { 260 for (int i = 0; i < CellsCount(); i++) { 261 if (cells()[i] != 0) { 262 return false; 263 } 264 } 265 return true; 266 } 267 268 // Clears all bits starting from {cell_base_index} up to and excluding 269 // {index}. Note that {cell_base_index} is required to be cell aligned. 270 void ClearRange(uint32_t cell_base_index, uint32_t index) { 271 DCHECK_EQ(IndexInCell(cell_base_index), 0u); 272 DCHECK_GE(index, cell_base_index); 273 uint32_t start_cell_index = IndexToCell(cell_base_index); 274 uint32_t end_cell_index = IndexToCell(index); 275 DCHECK_GE(end_cell_index, start_cell_index); 276 // Clear all cells till the cell containing the last index. 277 for (uint32_t i = start_cell_index; i < end_cell_index; i++) { 278 cells()[i] = 0; 279 } 280 // Clear all bits in the last cell till the last bit before index. 281 uint32_t clear_mask = ~((1u << IndexInCell(index)) - 1); 282 cells()[end_cell_index] &= clear_mask; 283 } 284 }; 285 286 287 class SkipList; 288 class SlotsBuffer; 289 290 // MemoryChunk represents a memory region owned by a specific space. 291 // It is divided into the header and the body. Chunk start is always 292 // 1MB aligned. Start of the body is aligned so it can accommodate 293 // any heap object. 294 class MemoryChunk { 295 public: 296 enum MemoryChunkFlags { 297 IS_EXECUTABLE, 298 ABOUT_TO_BE_FREED, 299 POINTERS_TO_HERE_ARE_INTERESTING, 300 POINTERS_FROM_HERE_ARE_INTERESTING, 301 SCAN_ON_SCAVENGE, 302 IN_FROM_SPACE, // Mutually exclusive with IN_TO_SPACE. 303 IN_TO_SPACE, // All pages in new space has one of these two set. 304 NEW_SPACE_BELOW_AGE_MARK, 305 EVACUATION_CANDIDATE, 306 RESCAN_ON_EVACUATION, 307 NEVER_EVACUATE, // May contain immortal immutables. 308 POPULAR_PAGE, // Slots buffer of this page overflowed on the previous GC. 309 310 // WAS_SWEPT indicates that marking bits have been cleared by the sweeper, 311 // otherwise marking bits are still intact. 312 WAS_SWEPT, 313 314 // Large objects can have a progress bar in their page header. These object 315 // are scanned in increments and will be kept black while being scanned. 316 // Even if the mutator writes to them they will be kept black and a white 317 // to grey transition is performed in the value. 318 HAS_PROGRESS_BAR, 319 320 // This flag is intended to be used for testing. Works only when both 321 // FLAG_stress_compaction and FLAG_manual_evacuation_candidates_selection 322 // are set. It forces the page to become an evacuation candidate at next 323 // candidates selection cycle. 324 FORCE_EVACUATION_CANDIDATE_FOR_TESTING, 325 326 // This flag is inteded to be used for testing. 327 NEVER_ALLOCATE_ON_PAGE, 328 329 // The memory chunk is already logically freed, however the actual freeing 330 // still has to be performed. 331 PRE_FREED, 332 333 // |COMPACTION_WAS_ABORTED|: Indicates that the compaction in this page 334 // has been aborted and needs special handling by the sweeper. 335 COMPACTION_WAS_ABORTED, 336 337 // Last flag, keep at bottom. 338 NUM_MEMORY_CHUNK_FLAGS 339 }; 340 341 // |kCompactionDone|: Initial compaction state of a |MemoryChunk|. 342 // |kCompactingInProgress|: Parallel compaction is currently in progress. 343 // |kCompactingFinalize|: Parallel compaction is done but the chunk needs to 344 // be finalized. 345 // |kCompactingAborted|: Parallel compaction has been aborted, which should 346 // for now only happen in OOM scenarios. 347 enum ParallelCompactingState { 348 kCompactingDone, 349 kCompactingInProgress, 350 kCompactingFinalize, 351 kCompactingAborted, 352 }; 353 354 // |kSweepingDone|: The page state when sweeping is complete or sweeping must 355 // not be performed on that page. 356 // |kSweepingFinalize|: A sweeper thread is done sweeping this page and will 357 // not touch the page memory anymore. 358 // |kSweepingInProgress|: This page is currently swept by a sweeper thread. 359 // |kSweepingPending|: This page is ready for parallel sweeping. 360 enum ParallelSweepingState { 361 kSweepingDone, 362 kSweepingFinalize, 363 kSweepingInProgress, 364 kSweepingPending 365 }; 366 367 // Every n write barrier invocations we go to runtime even though 368 // we could have handled it in generated code. This lets us check 369 // whether we have hit the limit and should do some more marking. 370 static const int kWriteBarrierCounterGranularity = 500; 371 372 static const int kPointersToHereAreInterestingMask = 373 1 << POINTERS_TO_HERE_ARE_INTERESTING; 374 375 static const int kPointersFromHereAreInterestingMask = 376 1 << POINTERS_FROM_HERE_ARE_INTERESTING; 377 378 static const int kEvacuationCandidateMask = 1 << EVACUATION_CANDIDATE; 379 380 static const int kSkipEvacuationSlotsRecordingMask = 381 (1 << EVACUATION_CANDIDATE) | (1 << RESCAN_ON_EVACUATION) | 382 (1 << IN_FROM_SPACE) | (1 << IN_TO_SPACE); 383 384 static const intptr_t kAlignment = 385 (static_cast<uintptr_t>(1) << kPageSizeBits); 386 387 static const intptr_t kAlignmentMask = kAlignment - 1; 388 389 static const intptr_t kSizeOffset = 0; 390 391 static const intptr_t kLiveBytesOffset = 392 kSizeOffset + kPointerSize // size_t size 393 + kIntptrSize // intptr_t flags_ 394 + kPointerSize // Address area_start_ 395 + kPointerSize // Address area_end_ 396 + 2 * kPointerSize // base::VirtualMemory reservation_ 397 + kPointerSize // Address owner_ 398 + kPointerSize // Heap* heap_ 399 + kIntSize; // int store_buffer_counter_ 400 401 static const size_t kSlotsBufferOffset = 402 kLiveBytesOffset + kIntSize; // int live_byte_count_ 403 404 static const size_t kWriteBarrierCounterOffset = 405 kSlotsBufferOffset + kPointerSize // SlotsBuffer* slots_buffer_; 406 + kPointerSize; // SkipList* skip_list_; 407 408 static const size_t kMinHeaderSize = 409 kWriteBarrierCounterOffset + 410 kIntptrSize // intptr_t write_barrier_counter_ 411 + kIntSize // int progress_bar_ 412 + kPointerSize // AtomicValue high_water_mark_ 413 + kPointerSize // base::Mutex* mutex_ 414 + kPointerSize // base::AtomicWord parallel_sweeping_ 415 + kPointerSize // AtomicValue parallel_compaction_ 416 + 5 * kPointerSize // AtomicNumber free-list statistics 417 + kPointerSize // AtomicValue next_chunk_ 418 + kPointerSize; // AtomicValue prev_chunk_ 419 420 // We add some more space to the computed header size to amount for missing 421 // alignment requirements in our computation. 422 // Try to get kHeaderSize properly aligned on 32-bit and 64-bit machines. 423 static const size_t kHeaderSize = kMinHeaderSize + kIntSize; 424 425 static const int kBodyOffset = 426 CODE_POINTER_ALIGN(kHeaderSize + Bitmap::kSize); 427 428 // The start offset of the object area in a page. Aligned to both maps and 429 // code alignment to be suitable for both. Also aligned to 32 words because 430 // the marking bitmap is arranged in 32 bit chunks. 431 static const int kObjectStartAlignment = 32 * kPointerSize; 432 static const int kObjectStartOffset = 433 kBodyOffset - 1 + 434 (kObjectStartAlignment - (kBodyOffset - 1) % kObjectStartAlignment); 435 436 static const int kFlagsOffset = kPointerSize; 437 438 static void IncrementLiveBytesFromMutator(HeapObject* object, int by); 439 440 // Only works if the pointer is in the first kPageSize of the MemoryChunk. 441 static MemoryChunk* FromAddress(Address a) { 442 return reinterpret_cast<MemoryChunk*>(OffsetFrom(a) & ~kAlignmentMask); 443 } 444 445 static const MemoryChunk* FromAddress(const byte* a) { 446 return reinterpret_cast<const MemoryChunk*>(OffsetFrom(a) & 447 ~kAlignmentMask); 448 } 449 450 static void IncrementLiveBytesFromGC(HeapObject* object, int by) { 451 MemoryChunk::FromAddress(object->address())->IncrementLiveBytes(by); 452 } 453 454 // Only works for addresses in pointer spaces, not data or code spaces. 455 static inline MemoryChunk* FromAnyPointerAddress(Heap* heap, Address addr); 456 457 static inline uint32_t FastAddressToMarkbitIndex(Address addr) { 458 const intptr_t offset = reinterpret_cast<intptr_t>(addr) & kAlignmentMask; 459 return static_cast<uint32_t>(offset) >> kPointerSizeLog2; 460 } 461 462 static inline void UpdateHighWaterMark(Address mark) { 463 if (mark == nullptr) return; 464 // Need to subtract one from the mark because when a chunk is full the 465 // top points to the next address after the chunk, which effectively belongs 466 // to another chunk. See the comment to Page::FromAllocationTop. 467 MemoryChunk* chunk = MemoryChunk::FromAddress(mark - 1); 468 intptr_t new_mark = static_cast<intptr_t>(mark - chunk->address()); 469 intptr_t old_mark = 0; 470 do { 471 old_mark = chunk->high_water_mark_.Value(); 472 } while ((new_mark > old_mark) && 473 !chunk->high_water_mark_.TrySetValue(old_mark, new_mark)); 474 } 475 476 Address address() { return reinterpret_cast<Address>(this); } 477 478 bool is_valid() { return address() != NULL; } 479 480 MemoryChunk* next_chunk() { return next_chunk_.Value(); } 481 482 MemoryChunk* prev_chunk() { return prev_chunk_.Value(); } 483 484 void set_next_chunk(MemoryChunk* next) { next_chunk_.SetValue(next); } 485 486 void set_prev_chunk(MemoryChunk* prev) { prev_chunk_.SetValue(prev); } 487 488 Space* owner() const { 489 if ((reinterpret_cast<intptr_t>(owner_) & kPageHeaderTagMask) == 490 kPageHeaderTag) { 491 return reinterpret_cast<Space*>(reinterpret_cast<intptr_t>(owner_) - 492 kPageHeaderTag); 493 } else { 494 return NULL; 495 } 496 } 497 498 void set_owner(Space* space) { 499 DCHECK((reinterpret_cast<intptr_t>(space) & kPageHeaderTagMask) == 0); 500 owner_ = reinterpret_cast<Address>(space) + kPageHeaderTag; 501 DCHECK((reinterpret_cast<intptr_t>(owner_) & kPageHeaderTagMask) == 502 kPageHeaderTag); 503 } 504 505 base::VirtualMemory* reserved_memory() { return &reservation_; } 506 507 void set_reserved_memory(base::VirtualMemory* reservation) { 508 DCHECK_NOT_NULL(reservation); 509 reservation_.TakeControl(reservation); 510 } 511 512 bool scan_on_scavenge() { return IsFlagSet(SCAN_ON_SCAVENGE); } 513 void initialize_scan_on_scavenge(bool scan) { 514 if (scan) { 515 SetFlag(SCAN_ON_SCAVENGE); 516 } else { 517 ClearFlag(SCAN_ON_SCAVENGE); 518 } 519 } 520 inline void set_scan_on_scavenge(bool scan); 521 522 int store_buffer_counter() { return store_buffer_counter_; } 523 void set_store_buffer_counter(int counter) { 524 store_buffer_counter_ = counter; 525 } 526 527 bool Contains(Address addr) { 528 return addr >= area_start() && addr < area_end(); 529 } 530 531 // Checks whether addr can be a limit of addresses in this page. 532 // It's a limit if it's in the page, or if it's just after the 533 // last byte of the page. 534 bool ContainsLimit(Address addr) { 535 return addr >= area_start() && addr <= area_end(); 536 } 537 538 void SetFlag(int flag) { flags_ |= static_cast<uintptr_t>(1) << flag; } 539 540 void ClearFlag(int flag) { flags_ &= ~(static_cast<uintptr_t>(1) << flag); } 541 542 void SetFlagTo(int flag, bool value) { 543 if (value) { 544 SetFlag(flag); 545 } else { 546 ClearFlag(flag); 547 } 548 } 549 550 bool IsFlagSet(int flag) { 551 return (flags_ & (static_cast<uintptr_t>(1) << flag)) != 0; 552 } 553 554 // Set or clear multiple flags at a time. The flags in the mask 555 // are set to the value in "flags", the rest retain the current value 556 // in flags_. 557 void SetFlags(intptr_t flags, intptr_t mask) { 558 flags_ = (flags_ & ~mask) | (flags & mask); 559 } 560 561 // Return all current flags. 562 intptr_t GetFlags() { return flags_; } 563 564 AtomicValue<ParallelSweepingState>& parallel_sweeping_state() { 565 return parallel_sweeping_; 566 } 567 568 AtomicValue<ParallelCompactingState>& parallel_compaction_state() { 569 return parallel_compaction_; 570 } 571 572 bool TryLock() { return mutex_->TryLock(); } 573 574 base::Mutex* mutex() { return mutex_; } 575 576 // WaitUntilSweepingCompleted only works when concurrent sweeping is in 577 // progress. In particular, when we know that right before this call a 578 // sweeper thread was sweeping this page. 579 void WaitUntilSweepingCompleted() { 580 mutex_->Lock(); 581 mutex_->Unlock(); 582 DCHECK(SweepingCompleted()); 583 } 584 585 bool SweepingCompleted() { 586 return parallel_sweeping_state().Value() <= kSweepingFinalize; 587 } 588 589 // Manage live byte count (count of bytes known to be live, 590 // because they are marked black). 591 void ResetLiveBytes() { 592 if (FLAG_gc_verbose) { 593 PrintF("ResetLiveBytes:%p:%x->0\n", static_cast<void*>(this), 594 live_byte_count_); 595 } 596 live_byte_count_ = 0; 597 } 598 599 void IncrementLiveBytes(int by) { 600 if (FLAG_gc_verbose) { 601 printf("UpdateLiveBytes:%p:%x%c=%x->%x\n", static_cast<void*>(this), 602 live_byte_count_, ((by < 0) ? '-' : '+'), ((by < 0) ? -by : by), 603 live_byte_count_ + by); 604 } 605 live_byte_count_ += by; 606 DCHECK_GE(live_byte_count_, 0); 607 DCHECK_LE(static_cast<unsigned>(live_byte_count_), size_); 608 } 609 610 int LiveBytes() { 611 DCHECK_LE(static_cast<unsigned>(live_byte_count_), size_); 612 return live_byte_count_; 613 } 614 615 void SetLiveBytes(int live_bytes) { 616 DCHECK_GE(live_bytes, 0); 617 DCHECK_LE(static_cast<unsigned>(live_bytes), size_); 618 live_byte_count_ = live_bytes; 619 } 620 621 int write_barrier_counter() { 622 return static_cast<int>(write_barrier_counter_); 623 } 624 625 void set_write_barrier_counter(int counter) { 626 write_barrier_counter_ = counter; 627 } 628 629 int progress_bar() { 630 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 631 return progress_bar_; 632 } 633 634 void set_progress_bar(int progress_bar) { 635 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 636 progress_bar_ = progress_bar; 637 } 638 639 void ResetProgressBar() { 640 if (IsFlagSet(MemoryChunk::HAS_PROGRESS_BAR)) { 641 set_progress_bar(0); 642 ClearFlag(MemoryChunk::HAS_PROGRESS_BAR); 643 } 644 } 645 646 size_t size() const { return size_; } 647 648 void set_size(size_t size) { size_ = size; } 649 650 void SetArea(Address area_start, Address area_end) { 651 area_start_ = area_start; 652 area_end_ = area_end; 653 } 654 655 Executability executable() { 656 return IsFlagSet(IS_EXECUTABLE) ? EXECUTABLE : NOT_EXECUTABLE; 657 } 658 659 bool InNewSpace() { 660 return (flags_ & ((1 << IN_FROM_SPACE) | (1 << IN_TO_SPACE))) != 0; 661 } 662 663 bool InToSpace() { return IsFlagSet(IN_TO_SPACE); } 664 665 bool InFromSpace() { return IsFlagSet(IN_FROM_SPACE); } 666 667 // Markbits support 668 669 inline Bitmap* markbits() { 670 return Bitmap::FromAddress(address() + kHeaderSize); 671 } 672 673 void PrintMarkbits() { markbits()->Print(); } 674 675 inline uint32_t AddressToMarkbitIndex(Address addr) { 676 return static_cast<uint32_t>(addr - this->address()) >> kPointerSizeLog2; 677 } 678 679 inline Address MarkbitIndexToAddress(uint32_t index) { 680 return this->address() + (index << kPointerSizeLog2); 681 } 682 683 void InsertAfter(MemoryChunk* other); 684 void Unlink(); 685 686 inline Heap* heap() const { return heap_; } 687 688 bool NeverEvacuate() { return IsFlagSet(NEVER_EVACUATE); } 689 690 void MarkNeverEvacuate() { SetFlag(NEVER_EVACUATE); } 691 692 bool IsEvacuationCandidate() { 693 DCHECK(!(IsFlagSet(NEVER_EVACUATE) && IsFlagSet(EVACUATION_CANDIDATE))); 694 return IsFlagSet(EVACUATION_CANDIDATE); 695 } 696 697 bool CanAllocate() { 698 return !IsEvacuationCandidate() && !IsFlagSet(NEVER_ALLOCATE_ON_PAGE); 699 } 700 701 bool ShouldSkipEvacuationSlotRecording() { 702 return (flags_ & kSkipEvacuationSlotsRecordingMask) != 0; 703 } 704 705 inline SkipList* skip_list() { return skip_list_; } 706 707 inline void set_skip_list(SkipList* skip_list) { skip_list_ = skip_list; } 708 709 inline SlotsBuffer* slots_buffer() { return slots_buffer_; } 710 711 inline SlotsBuffer** slots_buffer_address() { return &slots_buffer_; } 712 713 void MarkEvacuationCandidate() { 714 DCHECK(!IsFlagSet(NEVER_EVACUATE)); 715 DCHECK(slots_buffer_ == NULL); 716 SetFlag(EVACUATION_CANDIDATE); 717 } 718 719 void ClearEvacuationCandidate() { 720 DCHECK(slots_buffer_ == NULL); 721 ClearFlag(EVACUATION_CANDIDATE); 722 } 723 724 Address area_start() { return area_start_; } 725 Address area_end() { return area_end_; } 726 int area_size() { return static_cast<int>(area_end() - area_start()); } 727 bool CommitArea(size_t requested); 728 729 // Approximate amount of physical memory committed for this chunk. 730 size_t CommittedPhysicalMemory() { return high_water_mark_.Value(); } 731 732 // Should be called when memory chunk is about to be freed. 733 void ReleaseAllocatedMemory(); 734 735 protected: 736 static MemoryChunk* Initialize(Heap* heap, Address base, size_t size, 737 Address area_start, Address area_end, 738 Executability executable, Space* owner); 739 740 size_t size_; 741 intptr_t flags_; 742 743 // Start and end of allocatable memory on this chunk. 744 Address area_start_; 745 Address area_end_; 746 747 // If the chunk needs to remember its memory reservation, it is stored here. 748 base::VirtualMemory reservation_; 749 // The identity of the owning space. This is tagged as a failure pointer, but 750 // no failure can be in an object, so this can be distinguished from any entry 751 // in a fixed array. 752 Address owner_; 753 Heap* heap_; 754 // Used by the store buffer to keep track of which pages to mark scan-on- 755 // scavenge. 756 int store_buffer_counter_; 757 // Count of bytes marked black on page. 758 int live_byte_count_; 759 SlotsBuffer* slots_buffer_; 760 SkipList* skip_list_; 761 intptr_t write_barrier_counter_; 762 // Used by the incremental marker to keep track of the scanning progress in 763 // large objects that have a progress bar and are scanned in increments. 764 int progress_bar_; 765 // Assuming the initial allocation on a page is sequential, 766 // count highest number of bytes ever allocated on the page. 767 AtomicValue<intptr_t> high_water_mark_; 768 769 base::Mutex* mutex_; 770 AtomicValue<ParallelSweepingState> parallel_sweeping_; 771 AtomicValue<ParallelCompactingState> parallel_compaction_; 772 773 // PagedSpace free-list statistics. 774 AtomicNumber<intptr_t> available_in_small_free_list_; 775 AtomicNumber<intptr_t> available_in_medium_free_list_; 776 AtomicNumber<intptr_t> available_in_large_free_list_; 777 AtomicNumber<intptr_t> available_in_huge_free_list_; 778 AtomicNumber<intptr_t> non_available_small_blocks_; 779 780 // next_chunk_ holds a pointer of type MemoryChunk 781 AtomicValue<MemoryChunk*> next_chunk_; 782 // prev_chunk_ holds a pointer of type MemoryChunk 783 AtomicValue<MemoryChunk*> prev_chunk_; 784 785 private: 786 void InitializeReservedMemory() { reservation_.Reset(); } 787 788 friend class MemoryAllocator; 789 friend class MemoryChunkValidator; 790 }; 791 792 793 enum FreeListCategoryType { kSmall, kMedium, kLarge, kHuge }; 794 795 796 // ----------------------------------------------------------------------------- 797 // A page is a memory chunk of a size 1MB. Large object pages may be larger. 798 // 799 // The only way to get a page pointer is by calling factory methods: 800 // Page* p = Page::FromAddress(addr); or 801 // Page* p = Page::FromAllocationTop(top); 802 class Page : public MemoryChunk { 803 public: 804 // Returns the page containing a given address. The address ranges 805 // from [page_addr .. page_addr + kPageSize[ 806 // This only works if the object is in fact in a page. See also MemoryChunk:: 807 // FromAddress() and FromAnyAddress(). 808 INLINE(static Page* FromAddress(Address a)) { 809 return reinterpret_cast<Page*>(OffsetFrom(a) & ~kPageAlignmentMask); 810 } 811 812 // Returns the page containing an allocation top. Because an allocation 813 // top address can be the upper bound of the page, we need to subtract 814 // it with kPointerSize first. The address ranges from 815 // [page_addr + kObjectStartOffset .. page_addr + kPageSize]. 816 INLINE(static Page* FromAllocationTop(Address top)) { 817 Page* p = FromAddress(top - kPointerSize); 818 return p; 819 } 820 821 // Returns the next page in the chain of pages owned by a space. 822 inline Page* next_page() { 823 DCHECK(next_chunk()->owner() == owner()); 824 return static_cast<Page*>(next_chunk()); 825 } 826 inline Page* prev_page() { 827 DCHECK(prev_chunk()->owner() == owner()); 828 return static_cast<Page*>(prev_chunk()); 829 } 830 inline void set_next_page(Page* page); 831 inline void set_prev_page(Page* page); 832 833 // Checks whether an address is page aligned. 834 static bool IsAlignedToPageSize(Address a) { 835 return 0 == (OffsetFrom(a) & kPageAlignmentMask); 836 } 837 838 // Returns the offset of a given address to this page. 839 INLINE(int Offset(Address a)) { 840 int offset = static_cast<int>(a - address()); 841 return offset; 842 } 843 844 // Returns the address for a given offset to the this page. 845 Address OffsetToAddress(int offset) { 846 DCHECK_PAGE_OFFSET(offset); 847 return address() + offset; 848 } 849 850 // --------------------------------------------------------------------- 851 852 // Page size in bytes. This must be a multiple of the OS page size. 853 static const int kPageSize = 1 << kPageSizeBits; 854 855 // Maximum object size that gets allocated into regular pages. Objects larger 856 // than that size are allocated in large object space and are never moved in 857 // memory. This also applies to new space allocation, since objects are never 858 // migrated from new space to large object space. Takes double alignment into 859 // account. 860 // TODO(hpayer): This limit should be way smaller but we currently have 861 // short living objects >256K. 862 static const int kMaxRegularHeapObjectSize = 600 * KB; 863 864 static const int kAllocatableMemory = kPageSize - kObjectStartOffset; 865 866 // Page size mask. 867 static const intptr_t kPageAlignmentMask = (1 << kPageSizeBits) - 1; 868 869 inline void ClearGCFields(); 870 871 static inline Page* Initialize(Heap* heap, MemoryChunk* chunk, 872 Executability executable, PagedSpace* owner); 873 874 void InitializeAsAnchor(PagedSpace* owner); 875 876 bool WasSwept() { return IsFlagSet(WAS_SWEPT); } 877 void SetWasSwept() { SetFlag(WAS_SWEPT); } 878 void ClearWasSwept() { ClearFlag(WAS_SWEPT); } 879 880 void ResetFreeListStatistics(); 881 882 int LiveBytesFromFreeList() { 883 return static_cast<int>( 884 area_size() - non_available_small_blocks() - 885 available_in_small_free_list() - available_in_medium_free_list() - 886 available_in_large_free_list() - available_in_huge_free_list()); 887 } 888 889 #define FRAGMENTATION_STATS_ACCESSORS(type, name) \ 890 type name() { return name##_.Value(); } \ 891 void set_##name(type name) { name##_.SetValue(name); } \ 892 void add_##name(type name) { name##_.Increment(name); } 893 894 FRAGMENTATION_STATS_ACCESSORS(intptr_t, non_available_small_blocks) 895 FRAGMENTATION_STATS_ACCESSORS(intptr_t, available_in_small_free_list) 896 FRAGMENTATION_STATS_ACCESSORS(intptr_t, available_in_medium_free_list) 897 FRAGMENTATION_STATS_ACCESSORS(intptr_t, available_in_large_free_list) 898 FRAGMENTATION_STATS_ACCESSORS(intptr_t, available_in_huge_free_list) 899 900 #undef FRAGMENTATION_STATS_ACCESSORS 901 902 void add_available_in_free_list(FreeListCategoryType type, intptr_t bytes) { 903 switch (type) { 904 case kSmall: 905 add_available_in_small_free_list(bytes); 906 break; 907 case kMedium: 908 add_available_in_medium_free_list(bytes); 909 break; 910 case kLarge: 911 add_available_in_large_free_list(bytes); 912 break; 913 case kHuge: 914 add_available_in_huge_free_list(bytes); 915 break; 916 default: 917 UNREACHABLE(); 918 } 919 } 920 921 intptr_t available_in_free_list(FreeListCategoryType type) { 922 switch (type) { 923 case kSmall: 924 return available_in_small_free_list(); 925 case kMedium: 926 return available_in_medium_free_list(); 927 case kLarge: 928 return available_in_large_free_list(); 929 case kHuge: 930 return available_in_huge_free_list(); 931 default: 932 UNREACHABLE(); 933 } 934 UNREACHABLE(); 935 return 0; 936 } 937 938 #ifdef DEBUG 939 void Print(); 940 #endif // DEBUG 941 942 friend class MemoryAllocator; 943 }; 944 945 946 class LargePage : public MemoryChunk { 947 public: 948 HeapObject* GetObject() { return HeapObject::FromAddress(area_start()); } 949 950 inline LargePage* next_page() { 951 return static_cast<LargePage*>(next_chunk()); 952 } 953 954 inline void set_next_page(LargePage* page) { set_next_chunk(page); } 955 956 private: 957 static inline LargePage* Initialize(Heap* heap, MemoryChunk* chunk); 958 959 friend class MemoryAllocator; 960 }; 961 962 963 // ---------------------------------------------------------------------------- 964 // Space is the abstract superclass for all allocation spaces. 965 class Space : public Malloced { 966 public: 967 Space(Heap* heap, AllocationSpace id, Executability executable) 968 : heap_(heap), 969 id_(id), 970 executable_(executable), 971 committed_(0), 972 max_committed_(0) {} 973 974 virtual ~Space() {} 975 976 Heap* heap() const { return heap_; } 977 978 // Does the space need executable memory? 979 Executability executable() { return executable_; } 980 981 // Identity used in error reporting. 982 AllocationSpace identity() { return id_; } 983 984 // Return the total amount committed memory for this space, i.e., allocatable 985 // memory and page headers. 986 virtual intptr_t CommittedMemory() { return committed_; } 987 988 virtual intptr_t MaximumCommittedMemory() { return max_committed_; } 989 990 // Returns allocated size. 991 virtual intptr_t Size() = 0; 992 993 // Returns size of objects. Can differ from the allocated size 994 // (e.g. see LargeObjectSpace). 995 virtual intptr_t SizeOfObjects() { return Size(); } 996 997 // Approximate amount of physical memory committed for this space. 998 virtual size_t CommittedPhysicalMemory() = 0; 999 1000 // Return the available bytes without growing. 1001 virtual intptr_t Available() = 0; 1002 1003 virtual int RoundSizeDownToObjectAlignment(int size) { 1004 if (id_ == CODE_SPACE) { 1005 return RoundDown(size, kCodeAlignment); 1006 } else { 1007 return RoundDown(size, kPointerSize); 1008 } 1009 } 1010 1011 #ifdef DEBUG 1012 virtual void Print() = 0; 1013 #endif 1014 1015 protected: 1016 void AccountCommitted(intptr_t bytes) { 1017 DCHECK_GE(bytes, 0); 1018 committed_ += bytes; 1019 if (committed_ > max_committed_) { 1020 max_committed_ = committed_; 1021 } 1022 } 1023 1024 void AccountUncommitted(intptr_t bytes) { 1025 DCHECK_GE(bytes, 0); 1026 committed_ -= bytes; 1027 DCHECK_GE(committed_, 0); 1028 } 1029 1030 private: 1031 Heap* heap_; 1032 AllocationSpace id_; 1033 Executability executable_; 1034 1035 // Keeps track of committed memory in a space. 1036 intptr_t committed_; 1037 intptr_t max_committed_; 1038 }; 1039 1040 1041 class MemoryChunkValidator { 1042 // Computed offsets should match the compiler generated ones. 1043 STATIC_ASSERT(MemoryChunk::kSizeOffset == offsetof(MemoryChunk, size_)); 1044 STATIC_ASSERT(MemoryChunk::kLiveBytesOffset == 1045 offsetof(MemoryChunk, live_byte_count_)); 1046 STATIC_ASSERT(MemoryChunk::kSlotsBufferOffset == 1047 offsetof(MemoryChunk, slots_buffer_)); 1048 STATIC_ASSERT(MemoryChunk::kWriteBarrierCounterOffset == 1049 offsetof(MemoryChunk, write_barrier_counter_)); 1050 1051 // Validate our estimates on the header size. 1052 STATIC_ASSERT(sizeof(MemoryChunk) <= MemoryChunk::kHeaderSize); 1053 STATIC_ASSERT(sizeof(LargePage) <= MemoryChunk::kHeaderSize); 1054 STATIC_ASSERT(sizeof(Page) <= MemoryChunk::kHeaderSize); 1055 }; 1056 1057 1058 // ---------------------------------------------------------------------------- 1059 // All heap objects containing executable code (code objects) must be allocated 1060 // from a 2 GB range of memory, so that they can call each other using 32-bit 1061 // displacements. This happens automatically on 32-bit platforms, where 32-bit 1062 // displacements cover the entire 4GB virtual address space. On 64-bit 1063 // platforms, we support this using the CodeRange object, which reserves and 1064 // manages a range of virtual memory. 1065 class CodeRange { 1066 public: 1067 explicit CodeRange(Isolate* isolate); 1068 ~CodeRange() { TearDown(); } 1069 1070 // Reserves a range of virtual memory, but does not commit any of it. 1071 // Can only be called once, at heap initialization time. 1072 // Returns false on failure. 1073 bool SetUp(size_t requested_size); 1074 1075 bool valid() { return code_range_ != NULL; } 1076 Address start() { 1077 DCHECK(valid()); 1078 return static_cast<Address>(code_range_->address()); 1079 } 1080 size_t size() { 1081 DCHECK(valid()); 1082 return code_range_->size(); 1083 } 1084 bool contains(Address address) { 1085 if (!valid()) return false; 1086 Address start = static_cast<Address>(code_range_->address()); 1087 return start <= address && address < start + code_range_->size(); 1088 } 1089 1090 // Allocates a chunk of memory from the large-object portion of 1091 // the code range. On platforms with no separate code range, should 1092 // not be called. 1093 MUST_USE_RESULT Address AllocateRawMemory(const size_t requested_size, 1094 const size_t commit_size, 1095 size_t* allocated); 1096 bool CommitRawMemory(Address start, size_t length); 1097 bool UncommitRawMemory(Address start, size_t length); 1098 void FreeRawMemory(Address buf, size_t length); 1099 1100 private: 1101 // Frees the range of virtual memory, and frees the data structures used to 1102 // manage it. 1103 void TearDown(); 1104 1105 Isolate* isolate_; 1106 1107 // The reserved range of virtual memory that all code objects are put in. 1108 base::VirtualMemory* code_range_; 1109 // Plain old data class, just a struct plus a constructor. 1110 class FreeBlock { 1111 public: 1112 FreeBlock() : start(0), size(0) {} 1113 FreeBlock(Address start_arg, size_t size_arg) 1114 : start(start_arg), size(size_arg) { 1115 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 1116 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 1117 } 1118 FreeBlock(void* start_arg, size_t size_arg) 1119 : start(static_cast<Address>(start_arg)), size(size_arg) { 1120 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 1121 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 1122 } 1123 1124 Address start; 1125 size_t size; 1126 }; 1127 1128 // The global mutex guards free_list_ and allocation_list_ as GC threads may 1129 // access both lists concurrently to the main thread. 1130 base::Mutex code_range_mutex_; 1131 1132 // Freed blocks of memory are added to the free list. When the allocation 1133 // list is exhausted, the free list is sorted and merged to make the new 1134 // allocation list. 1135 List<FreeBlock> free_list_; 1136 1137 // Memory is allocated from the free blocks on the allocation list. 1138 // The block at current_allocation_block_index_ is the current block. 1139 List<FreeBlock> allocation_list_; 1140 int current_allocation_block_index_; 1141 1142 // Finds a block on the allocation list that contains at least the 1143 // requested amount of memory. If none is found, sorts and merges 1144 // the existing free memory blocks, and searches again. 1145 // If none can be found, returns false. 1146 bool GetNextAllocationBlock(size_t requested); 1147 // Compares the start addresses of two free blocks. 1148 static int CompareFreeBlockAddress(const FreeBlock* left, 1149 const FreeBlock* right); 1150 bool ReserveBlock(const size_t requested_size, FreeBlock* block); 1151 void ReleaseBlock(const FreeBlock* block); 1152 1153 DISALLOW_COPY_AND_ASSIGN(CodeRange); 1154 }; 1155 1156 1157 class SkipList { 1158 public: 1159 SkipList() { Clear(); } 1160 1161 void Clear() { 1162 for (int idx = 0; idx < kSize; idx++) { 1163 starts_[idx] = reinterpret_cast<Address>(-1); 1164 } 1165 } 1166 1167 Address StartFor(Address addr) { return starts_[RegionNumber(addr)]; } 1168 1169 void AddObject(Address addr, int size) { 1170 int start_region = RegionNumber(addr); 1171 int end_region = RegionNumber(addr + size - kPointerSize); 1172 for (int idx = start_region; idx <= end_region; idx++) { 1173 if (starts_[idx] > addr) starts_[idx] = addr; 1174 } 1175 } 1176 1177 static inline int RegionNumber(Address addr) { 1178 return (OffsetFrom(addr) & Page::kPageAlignmentMask) >> kRegionSizeLog2; 1179 } 1180 1181 static void Update(Address addr, int size) { 1182 Page* page = Page::FromAddress(addr); 1183 SkipList* list = page->skip_list(); 1184 if (list == NULL) { 1185 list = new SkipList(); 1186 page->set_skip_list(list); 1187 } 1188 1189 list->AddObject(addr, size); 1190 } 1191 1192 private: 1193 static const int kRegionSizeLog2 = 13; 1194 static const int kRegionSize = 1 << kRegionSizeLog2; 1195 static const int kSize = Page::kPageSize / kRegionSize; 1196 1197 STATIC_ASSERT(Page::kPageSize % kRegionSize == 0); 1198 1199 Address starts_[kSize]; 1200 }; 1201 1202 1203 // ---------------------------------------------------------------------------- 1204 // A space acquires chunks of memory from the operating system. The memory 1205 // allocator allocated and deallocates pages for the paged heap spaces and large 1206 // pages for large object space. 1207 // 1208 // Each space has to manage it's own pages. 1209 // 1210 class MemoryAllocator { 1211 public: 1212 explicit MemoryAllocator(Isolate* isolate); 1213 1214 // Initializes its internal bookkeeping structures. 1215 // Max capacity of the total space and executable memory limit. 1216 bool SetUp(intptr_t max_capacity, intptr_t capacity_executable); 1217 1218 void TearDown(); 1219 1220 Page* AllocatePage(intptr_t size, PagedSpace* owner, 1221 Executability executable); 1222 1223 LargePage* AllocateLargePage(intptr_t object_size, Space* owner, 1224 Executability executable); 1225 1226 // PreFree logically frees the object, i.e., it takes care of the size 1227 // bookkeeping and calls the allocation callback. 1228 void PreFreeMemory(MemoryChunk* chunk); 1229 1230 // FreeMemory can be called concurrently when PreFree was executed before. 1231 void PerformFreeMemory(MemoryChunk* chunk); 1232 1233 // Free is a wrapper method, which calls PreFree and PerformFreeMemory 1234 // together. 1235 void Free(MemoryChunk* chunk); 1236 1237 // Returns allocated spaces in bytes. 1238 intptr_t Size() { return size_.Value(); } 1239 1240 // Returns allocated executable spaces in bytes. 1241 intptr_t SizeExecutable() { return size_executable_.Value(); } 1242 1243 // Returns the maximum available bytes of heaps. 1244 intptr_t Available() { 1245 intptr_t size = Size(); 1246 return capacity_ < size ? 0 : capacity_ - size; 1247 } 1248 1249 // Returns the maximum available executable bytes of heaps. 1250 intptr_t AvailableExecutable() { 1251 intptr_t executable_size = SizeExecutable(); 1252 if (capacity_executable_ < executable_size) return 0; 1253 return capacity_executable_ - executable_size; 1254 } 1255 1256 // Returns maximum available bytes that the old space can have. 1257 intptr_t MaxAvailable() { 1258 return (Available() / Page::kPageSize) * Page::kAllocatableMemory; 1259 } 1260 1261 // Returns an indication of whether a pointer is in a space that has 1262 // been allocated by this MemoryAllocator. 1263 V8_INLINE bool IsOutsideAllocatedSpace(const void* address) { 1264 return address < lowest_ever_allocated_.Value() || 1265 address >= highest_ever_allocated_.Value(); 1266 } 1267 1268 #ifdef DEBUG 1269 // Reports statistic info of the space. 1270 void ReportStatistics(); 1271 #endif 1272 1273 // Returns a MemoryChunk in which the memory region from commit_area_size to 1274 // reserve_area_size of the chunk area is reserved but not committed, it 1275 // could be committed later by calling MemoryChunk::CommitArea. 1276 MemoryChunk* AllocateChunk(intptr_t reserve_area_size, 1277 intptr_t commit_area_size, 1278 Executability executable, Space* space); 1279 1280 Address ReserveAlignedMemory(size_t requested, size_t alignment, 1281 base::VirtualMemory* controller); 1282 Address AllocateAlignedMemory(size_t reserve_size, size_t commit_size, 1283 size_t alignment, Executability executable, 1284 base::VirtualMemory* controller); 1285 1286 bool CommitMemory(Address addr, size_t size, Executability executable); 1287 1288 void FreeNewSpaceMemory(Address addr, base::VirtualMemory* reservation, 1289 Executability executable); 1290 void FreeMemory(base::VirtualMemory* reservation, Executability executable); 1291 void FreeMemory(Address addr, size_t size, Executability executable); 1292 1293 // Commit a contiguous block of memory from the initial chunk. Assumes that 1294 // the address is not NULL, the size is greater than zero, and that the 1295 // block is contained in the initial chunk. Returns true if it succeeded 1296 // and false otherwise. 1297 bool CommitBlock(Address start, size_t size, Executability executable); 1298 1299 // Uncommit a contiguous block of memory [start..(start+size)[. 1300 // start is not NULL, the size is greater than zero, and the 1301 // block is contained in the initial chunk. Returns true if it succeeded 1302 // and false otherwise. 1303 bool UncommitBlock(Address start, size_t size); 1304 1305 // Zaps a contiguous block of memory [start..(start+size)[ thus 1306 // filling it up with a recognizable non-NULL bit pattern. 1307 void ZapBlock(Address start, size_t size); 1308 1309 void PerformAllocationCallback(ObjectSpace space, AllocationAction action, 1310 size_t size); 1311 1312 void AddMemoryAllocationCallback(MemoryAllocationCallback callback, 1313 ObjectSpace space, AllocationAction action); 1314 1315 void RemoveMemoryAllocationCallback(MemoryAllocationCallback callback); 1316 1317 bool MemoryAllocationCallbackRegistered(MemoryAllocationCallback callback); 1318 1319 static int CodePageGuardStartOffset(); 1320 1321 static int CodePageGuardSize(); 1322 1323 static int CodePageAreaStartOffset(); 1324 1325 static int CodePageAreaEndOffset(); 1326 1327 static int CodePageAreaSize() { 1328 return CodePageAreaEndOffset() - CodePageAreaStartOffset(); 1329 } 1330 1331 static int PageAreaSize(AllocationSpace space) { 1332 DCHECK_NE(LO_SPACE, space); 1333 return (space == CODE_SPACE) ? CodePageAreaSize() 1334 : Page::kAllocatableMemory; 1335 } 1336 1337 MUST_USE_RESULT bool CommitExecutableMemory(base::VirtualMemory* vm, 1338 Address start, size_t commit_size, 1339 size_t reserved_size); 1340 1341 private: 1342 Isolate* isolate_; 1343 1344 // Maximum space size in bytes. 1345 intptr_t capacity_; 1346 // Maximum subset of capacity_ that can be executable 1347 intptr_t capacity_executable_; 1348 1349 // Allocated space size in bytes. 1350 AtomicNumber<intptr_t> size_; 1351 // Allocated executable space size in bytes. 1352 AtomicNumber<intptr_t> size_executable_; 1353 1354 // We keep the lowest and highest addresses allocated as a quick way 1355 // of determining that pointers are outside the heap. The estimate is 1356 // conservative, i.e. not all addrsses in 'allocated' space are allocated 1357 // to our heap. The range is [lowest, highest[, inclusive on the low end 1358 // and exclusive on the high end. 1359 AtomicValue<void*> lowest_ever_allocated_; 1360 AtomicValue<void*> highest_ever_allocated_; 1361 1362 struct MemoryAllocationCallbackRegistration { 1363 MemoryAllocationCallbackRegistration(MemoryAllocationCallback callback, 1364 ObjectSpace space, 1365 AllocationAction action) 1366 : callback(callback), space(space), action(action) {} 1367 MemoryAllocationCallback callback; 1368 ObjectSpace space; 1369 AllocationAction action; 1370 }; 1371 1372 // A List of callback that are triggered when memory is allocated or free'd 1373 List<MemoryAllocationCallbackRegistration> memory_allocation_callbacks_; 1374 1375 // Initializes pages in a chunk. Returns the first page address. 1376 // This function and GetChunkId() are provided for the mark-compact 1377 // collector to rebuild page headers in the from space, which is 1378 // used as a marking stack and its page headers are destroyed. 1379 Page* InitializePagesInChunk(int chunk_id, int pages_in_chunk, 1380 PagedSpace* owner); 1381 1382 void UpdateAllocatedSpaceLimits(void* low, void* high) { 1383 // The use of atomic primitives does not guarantee correctness (wrt. 1384 // desired semantics) by default. The loop here ensures that we update the 1385 // values only if they did not change in between. 1386 void* ptr = nullptr; 1387 do { 1388 ptr = lowest_ever_allocated_.Value(); 1389 } while ((low < ptr) && !lowest_ever_allocated_.TrySetValue(ptr, low)); 1390 do { 1391 ptr = highest_ever_allocated_.Value(); 1392 } while ((high > ptr) && !highest_ever_allocated_.TrySetValue(ptr, high)); 1393 } 1394 1395 DISALLOW_IMPLICIT_CONSTRUCTORS(MemoryAllocator); 1396 }; 1397 1398 1399 // ----------------------------------------------------------------------------- 1400 // Interface for heap object iterator to be implemented by all object space 1401 // object iterators. 1402 // 1403 // NOTE: The space specific object iterators also implements the own next() 1404 // method which is used to avoid using virtual functions 1405 // iterating a specific space. 1406 1407 class ObjectIterator : public Malloced { 1408 public: 1409 virtual ~ObjectIterator() {} 1410 1411 virtual HeapObject* next_object() = 0; 1412 }; 1413 1414 1415 // ----------------------------------------------------------------------------- 1416 // Heap object iterator in new/old/map spaces. 1417 // 1418 // A HeapObjectIterator iterates objects from the bottom of the given space 1419 // to its top or from the bottom of the given page to its top. 1420 // 1421 // If objects are allocated in the page during iteration the iterator may 1422 // or may not iterate over those objects. The caller must create a new 1423 // iterator in order to be sure to visit these new objects. 1424 class HeapObjectIterator : public ObjectIterator { 1425 public: 1426 // Creates a new object iterator in a given space. 1427 explicit HeapObjectIterator(PagedSpace* space); 1428 explicit HeapObjectIterator(Page* page); 1429 1430 // Advance to the next object, skipping free spaces and other fillers and 1431 // skipping the special garbage section of which there is one per space. 1432 // Returns NULL when the iteration has ended. 1433 inline HeapObject* Next(); 1434 inline HeapObject* next_object() override; 1435 1436 private: 1437 enum PageMode { kOnePageOnly, kAllPagesInSpace }; 1438 1439 Address cur_addr_; // Current iteration point. 1440 Address cur_end_; // End iteration point. 1441 PagedSpace* space_; 1442 PageMode page_mode_; 1443 1444 // Fast (inlined) path of next(). 1445 inline HeapObject* FromCurrentPage(); 1446 1447 // Slow path of next(), goes into the next page. Returns false if the 1448 // iteration has ended. 1449 bool AdvanceToNextPage(); 1450 1451 // Initializes fields. 1452 inline void Initialize(PagedSpace* owner, Address start, Address end, 1453 PageMode mode); 1454 }; 1455 1456 1457 // ----------------------------------------------------------------------------- 1458 // A PageIterator iterates the pages in a paged space. 1459 1460 class PageIterator BASE_EMBEDDED { 1461 public: 1462 explicit inline PageIterator(PagedSpace* space); 1463 1464 inline bool has_next(); 1465 inline Page* next(); 1466 1467 private: 1468 PagedSpace* space_; 1469 Page* prev_page_; // Previous page returned. 1470 // Next page that will be returned. Cached here so that we can use this 1471 // iterator for operations that deallocate pages. 1472 Page* next_page_; 1473 }; 1474 1475 1476 // ----------------------------------------------------------------------------- 1477 // A space has a circular list of pages. The next page can be accessed via 1478 // Page::next_page() call. 1479 1480 // An abstraction of allocation and relocation pointers in a page-structured 1481 // space. 1482 class AllocationInfo { 1483 public: 1484 AllocationInfo() : top_(nullptr), limit_(nullptr) {} 1485 AllocationInfo(Address top, Address limit) : top_(top), limit_(limit) {} 1486 1487 void Reset(Address top, Address limit) { 1488 set_top(top); 1489 set_limit(limit); 1490 } 1491 1492 INLINE(void set_top(Address top)) { 1493 SLOW_DCHECK(top == NULL || 1494 (reinterpret_cast<intptr_t>(top) & kHeapObjectTagMask) == 0); 1495 top_ = top; 1496 } 1497 1498 INLINE(Address top()) const { 1499 SLOW_DCHECK(top_ == NULL || 1500 (reinterpret_cast<intptr_t>(top_) & kHeapObjectTagMask) == 0); 1501 return top_; 1502 } 1503 1504 Address* top_address() { return &top_; } 1505 1506 INLINE(void set_limit(Address limit)) { 1507 limit_ = limit; 1508 } 1509 1510 INLINE(Address limit()) const { 1511 return limit_; 1512 } 1513 1514 Address* limit_address() { return &limit_; } 1515 1516 #ifdef DEBUG 1517 bool VerifyPagedAllocation() { 1518 return (Page::FromAllocationTop(top_) == Page::FromAllocationTop(limit_)) && 1519 (top_ <= limit_); 1520 } 1521 #endif 1522 1523 private: 1524 // Current allocation top. 1525 Address top_; 1526 // Current allocation limit. 1527 Address limit_; 1528 }; 1529 1530 1531 // An abstraction of the accounting statistics of a page-structured space. 1532 // 1533 // The stats are only set by functions that ensure they stay balanced. These 1534 // functions increase or decrease one of the non-capacity stats in conjunction 1535 // with capacity, or else they always balance increases and decreases to the 1536 // non-capacity stats. 1537 class AllocationStats BASE_EMBEDDED { 1538 public: 1539 AllocationStats() { Clear(); } 1540 1541 // Zero out all the allocation statistics (i.e., no capacity). 1542 void Clear() { 1543 capacity_ = 0; 1544 max_capacity_ = 0; 1545 size_ = 0; 1546 } 1547 1548 void ClearSize() { size_ = capacity_; } 1549 1550 // Reset the allocation statistics (i.e., available = capacity with no wasted 1551 // or allocated bytes). 1552 void Reset() { 1553 size_ = 0; 1554 } 1555 1556 // Accessors for the allocation statistics. 1557 intptr_t Capacity() { return capacity_; } 1558 intptr_t MaxCapacity() { return max_capacity_; } 1559 intptr_t Size() { 1560 CHECK_GE(size_, 0); 1561 return size_; 1562 } 1563 1564 // Grow the space by adding available bytes. They are initially marked as 1565 // being in use (part of the size), but will normally be immediately freed, 1566 // putting them on the free list and removing them from size_. 1567 void ExpandSpace(int size_in_bytes) { 1568 capacity_ += size_in_bytes; 1569 size_ += size_in_bytes; 1570 if (capacity_ > max_capacity_) { 1571 max_capacity_ = capacity_; 1572 } 1573 CHECK(size_ >= 0); 1574 } 1575 1576 // Shrink the space by removing available bytes. Since shrinking is done 1577 // during sweeping, bytes have been marked as being in use (part of the size) 1578 // and are hereby freed. 1579 void ShrinkSpace(int size_in_bytes) { 1580 capacity_ -= size_in_bytes; 1581 size_ -= size_in_bytes; 1582 CHECK(size_ >= 0); 1583 } 1584 1585 // Allocate from available bytes (available -> size). 1586 void AllocateBytes(intptr_t size_in_bytes) { 1587 size_ += size_in_bytes; 1588 CHECK(size_ >= 0); 1589 } 1590 1591 // Free allocated bytes, making them available (size -> available). 1592 void DeallocateBytes(intptr_t size_in_bytes) { 1593 size_ -= size_in_bytes; 1594 CHECK_GE(size_, 0); 1595 } 1596 1597 // Merge {other} into {this}. 1598 void Merge(const AllocationStats& other) { 1599 capacity_ += other.capacity_; 1600 size_ += other.size_; 1601 if (other.max_capacity_ > max_capacity_) { 1602 max_capacity_ = other.max_capacity_; 1603 } 1604 CHECK_GE(size_, 0); 1605 } 1606 1607 void DecreaseCapacity(intptr_t size_in_bytes) { 1608 capacity_ -= size_in_bytes; 1609 CHECK_GE(capacity_, 0); 1610 CHECK_GE(capacity_, size_); 1611 } 1612 1613 void IncreaseCapacity(intptr_t size_in_bytes) { capacity_ += size_in_bytes; } 1614 1615 private: 1616 // |capacity_|: The number of object-area bytes (i.e., not including page 1617 // bookkeeping structures) currently in the space. 1618 intptr_t capacity_; 1619 1620 // |max_capacity_|: The maximum capacity ever observed. 1621 intptr_t max_capacity_; 1622 1623 // |size_|: The number of allocated bytes. 1624 intptr_t size_; 1625 }; 1626 1627 1628 // A free list category maintains a linked list of free memory blocks. 1629 class FreeListCategory { 1630 public: 1631 explicit FreeListCategory(FreeList* owner, FreeListCategoryType type) 1632 : type_(type), 1633 top_(nullptr), 1634 end_(nullptr), 1635 available_(0), 1636 owner_(owner) {} 1637 1638 // Concatenates {category} into {this}. 1639 // 1640 // Note: Thread-safe. 1641 intptr_t Concatenate(FreeListCategory* category); 1642 1643 void Reset(); 1644 1645 void Free(FreeSpace* node, int size_in_bytes); 1646 1647 // Pick a node from the list. 1648 FreeSpace* PickNodeFromList(int* node_size); 1649 1650 // Pick a node from the list and compare it against {size_in_bytes}. If the 1651 // node's size is greater or equal return the node and null otherwise. 1652 FreeSpace* PickNodeFromList(int size_in_bytes, int* node_size); 1653 1654 // Search for a node of size {size_in_bytes}. 1655 FreeSpace* SearchForNodeInList(int size_in_bytes, int* node_size); 1656 1657 intptr_t EvictFreeListItemsInList(Page* p); 1658 bool ContainsPageFreeListItemsInList(Page* p); 1659 1660 void RepairFreeList(Heap* heap); 1661 1662 bool IsEmpty() { return top() == nullptr; } 1663 1664 FreeList* owner() { return owner_; } 1665 int available() const { return available_; } 1666 1667 #ifdef DEBUG 1668 intptr_t SumFreeList(); 1669 int FreeListLength(); 1670 bool IsVeryLong(); 1671 #endif 1672 1673 private: 1674 // For debug builds we accurately compute free lists lengths up until 1675 // {kVeryLongFreeList} by manually walking the list. 1676 static const int kVeryLongFreeList = 500; 1677 1678 FreeSpace* top() { return top_.Value(); } 1679 void set_top(FreeSpace* top) { top_.SetValue(top); } 1680 1681 FreeSpace* end() const { return end_; } 1682 void set_end(FreeSpace* end) { end_ = end; } 1683 1684 // |type_|: The type of this free list category. 1685 FreeListCategoryType type_; 1686 1687 // |top_|: Points to the top FreeSpace* in the free list category. 1688 AtomicValue<FreeSpace*> top_; 1689 1690 // |end_|: Points to the end FreeSpace* in the free list category. 1691 FreeSpace* end_; 1692 1693 // |available_|: Total available bytes in all blocks of this free list 1694 // category. 1695 int available_; 1696 1697 // |owner_|: The owning free list of this category. 1698 FreeList* owner_; 1699 }; 1700 1701 // A free list maintaining free blocks of memory. The free list is organized in 1702 // a way to encourage objects allocated around the same time to be near each 1703 // other. The normal way to allocate is intended to be by bumping a 'top' 1704 // pointer until it hits a 'limit' pointer. When the limit is hit we need to 1705 // find a new space to allocate from. This is done with the free list, which is 1706 // divided up into rough categories to cut down on waste. Having finer 1707 // categories would scatter allocation more. 1708 1709 // The free list is organized in categories as follows: 1710 // 1-31 words (too small): Such small free areas are discarded for efficiency 1711 // reasons. They can be reclaimed by the compactor. However the distance 1712 // between top and limit may be this small. 1713 // 32-255 words (small): Used for allocating free space between 1-31 words in 1714 // size. 1715 // 256-2047 words (medium): Used for allocating free space between 32-255 words 1716 // in size. 1717 // 1048-16383 words (large): Used for allocating free space between 256-2047 1718 // words in size. 1719 // At least 16384 words (huge): This list is for objects of 2048 words or 1720 // larger. Empty pages are also added to this list. 1721 class FreeList { 1722 public: 1723 // This method returns how much memory can be allocated after freeing 1724 // maximum_freed memory. 1725 static inline int GuaranteedAllocatable(int maximum_freed) { 1726 if (maximum_freed <= kSmallListMin) { 1727 return 0; 1728 } else if (maximum_freed <= kSmallListMax) { 1729 return kSmallAllocationMax; 1730 } else if (maximum_freed <= kMediumListMax) { 1731 return kMediumAllocationMax; 1732 } else if (maximum_freed <= kLargeListMax) { 1733 return kLargeAllocationMax; 1734 } 1735 return maximum_freed; 1736 } 1737 1738 explicit FreeList(PagedSpace* owner); 1739 1740 // The method concatenates {other} into {this} and returns the added bytes, 1741 // including waste. 1742 // 1743 // Note: Thread-safe. 1744 intptr_t Concatenate(FreeList* other); 1745 1746 // Adds a node on the free list. The block of size {size_in_bytes} starting 1747 // at {start} is placed on the free list. The return value is the number of 1748 // bytes that were not added to the free list, because they freed memory block 1749 // was too small. Bookkeeping information will be written to the block, i.e., 1750 // its contents will be destroyed. The start address should be word aligned, 1751 // and the size should be a non-zero multiple of the word size. 1752 int Free(Address start, int size_in_bytes); 1753 1754 // Allocate a block of size {size_in_bytes} from the free list. The block is 1755 // unitialized. A failure is returned if no block is available. The size 1756 // should be a non-zero multiple of the word size. 1757 MUST_USE_RESULT HeapObject* Allocate(int size_in_bytes); 1758 1759 // Clear the free list. 1760 void Reset(); 1761 1762 void ResetStats() { wasted_bytes_ = 0; } 1763 1764 // Return the number of bytes available on the free list. 1765 intptr_t Available() { 1766 return small_list_.available() + medium_list_.available() + 1767 large_list_.available() + huge_list_.available(); 1768 } 1769 1770 // The method tries to find a {FreeSpace} node of at least {size_in_bytes} 1771 // size in the free list category exactly matching the size. If no suitable 1772 // node could be found, the method falls back to retrieving a {FreeSpace} 1773 // from the large or huge free list category. 1774 // 1775 // Can be used concurrently. 1776 MUST_USE_RESULT FreeSpace* TryRemoveMemory(intptr_t hint_size_in_bytes); 1777 1778 bool IsEmpty() { 1779 return small_list_.IsEmpty() && medium_list_.IsEmpty() && 1780 large_list_.IsEmpty() && huge_list_.IsEmpty(); 1781 } 1782 1783 // Used after booting the VM. 1784 void RepairLists(Heap* heap); 1785 1786 intptr_t EvictFreeListItems(Page* p); 1787 bool ContainsPageFreeListItems(Page* p); 1788 1789 PagedSpace* owner() { return owner_; } 1790 intptr_t wasted_bytes() { return wasted_bytes_; } 1791 base::Mutex* mutex() { return &mutex_; } 1792 1793 #ifdef DEBUG 1794 void Zap(); 1795 intptr_t SumFreeLists(); 1796 bool IsVeryLong(); 1797 #endif 1798 1799 private: 1800 // The size range of blocks, in bytes. 1801 static const int kMinBlockSize = 3 * kPointerSize; 1802 static const int kMaxBlockSize = Page::kAllocatableMemory; 1803 1804 static const int kSmallListMin = 0x1f * kPointerSize; 1805 static const int kSmallListMax = 0xff * kPointerSize; 1806 static const int kMediumListMax = 0x7ff * kPointerSize; 1807 static const int kLargeListMax = 0x3fff * kPointerSize; 1808 static const int kSmallAllocationMax = kSmallListMin; 1809 static const int kMediumAllocationMax = kSmallListMax; 1810 static const int kLargeAllocationMax = kMediumListMax; 1811 1812 FreeSpace* FindNodeFor(int size_in_bytes, int* node_size); 1813 FreeSpace* FindNodeIn(FreeListCategoryType category, int* node_size); 1814 1815 FreeListCategory* GetFreeListCategory(FreeListCategoryType category) { 1816 switch (category) { 1817 case kSmall: 1818 return &small_list_; 1819 case kMedium: 1820 return &medium_list_; 1821 case kLarge: 1822 return &large_list_; 1823 case kHuge: 1824 return &huge_list_; 1825 default: 1826 UNREACHABLE(); 1827 } 1828 UNREACHABLE(); 1829 return nullptr; 1830 } 1831 1832 PagedSpace* owner_; 1833 base::Mutex mutex_; 1834 intptr_t wasted_bytes_; 1835 FreeListCategory small_list_; 1836 FreeListCategory medium_list_; 1837 FreeListCategory large_list_; 1838 FreeListCategory huge_list_; 1839 1840 DISALLOW_IMPLICIT_CONSTRUCTORS(FreeList); 1841 }; 1842 1843 1844 class AllocationResult { 1845 public: 1846 // Implicit constructor from Object*. 1847 AllocationResult(Object* object) // NOLINT 1848 : object_(object) { 1849 // AllocationResults can't return Smis, which are used to represent 1850 // failure and the space to retry in. 1851 CHECK(!object->IsSmi()); 1852 } 1853 1854 AllocationResult() : object_(Smi::FromInt(NEW_SPACE)) {} 1855 1856 static inline AllocationResult Retry(AllocationSpace space = NEW_SPACE) { 1857 return AllocationResult(space); 1858 } 1859 1860 inline bool IsRetry() { return object_->IsSmi(); } 1861 1862 template <typename T> 1863 bool To(T** obj) { 1864 if (IsRetry()) return false; 1865 *obj = T::cast(object_); 1866 return true; 1867 } 1868 1869 Object* ToObjectChecked() { 1870 CHECK(!IsRetry()); 1871 return object_; 1872 } 1873 1874 inline AllocationSpace RetrySpace(); 1875 1876 private: 1877 explicit AllocationResult(AllocationSpace space) 1878 : object_(Smi::FromInt(static_cast<int>(space))) {} 1879 1880 Object* object_; 1881 }; 1882 1883 1884 STATIC_ASSERT(sizeof(AllocationResult) == kPointerSize); 1885 1886 1887 // LocalAllocationBuffer represents a linear allocation area that is created 1888 // from a given {AllocationResult} and can be used to allocate memory without 1889 // synchronization. 1890 // 1891 // The buffer is properly closed upon destruction and reassignment. 1892 // Example: 1893 // { 1894 // AllocationResult result = ...; 1895 // LocalAllocationBuffer a(heap, result, size); 1896 // LocalAllocationBuffer b = a; 1897 // CHECK(!a.IsValid()); 1898 // CHECK(b.IsValid()); 1899 // // {a} is invalid now and cannot be used for further allocations. 1900 // } 1901 // // Since {b} went out of scope, the LAB is closed, resulting in creating a 1902 // // filler object for the remaining area. 1903 class LocalAllocationBuffer { 1904 public: 1905 // Indicates that a buffer cannot be used for allocations anymore. Can result 1906 // from either reassigning a buffer, or trying to construct it from an 1907 // invalid {AllocationResult}. 1908 static inline LocalAllocationBuffer InvalidBuffer(); 1909 1910 // Creates a new LAB from a given {AllocationResult}. Results in 1911 // InvalidBuffer if the result indicates a retry. 1912 static inline LocalAllocationBuffer FromResult(Heap* heap, 1913 AllocationResult result, 1914 intptr_t size); 1915 1916 ~LocalAllocationBuffer() { Close(); } 1917 1918 // Convert to C++11 move-semantics once allowed by the style guide. 1919 LocalAllocationBuffer(const LocalAllocationBuffer& other); 1920 LocalAllocationBuffer& operator=(const LocalAllocationBuffer& other); 1921 1922 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 1923 int size_in_bytes, AllocationAlignment alignment); 1924 1925 inline bool IsValid() { return allocation_info_.top() != nullptr; } 1926 1927 // Try to merge LABs, which is only possible when they are adjacent in memory. 1928 // Returns true if the merge was successful, false otherwise. 1929 inline bool TryMerge(LocalAllocationBuffer* other); 1930 1931 private: 1932 LocalAllocationBuffer(Heap* heap, AllocationInfo allocation_info); 1933 1934 void Close(); 1935 1936 Heap* heap_; 1937 AllocationInfo allocation_info_; 1938 }; 1939 1940 1941 class PagedSpace : public Space { 1942 public: 1943 static const intptr_t kCompactionMemoryWanted = 500 * KB; 1944 1945 // Creates a space with an id. 1946 PagedSpace(Heap* heap, AllocationSpace id, Executability executable); 1947 1948 ~PagedSpace() override { TearDown(); } 1949 1950 // Set up the space using the given address range of virtual memory (from 1951 // the memory allocator's initial chunk) if possible. If the block of 1952 // addresses is not big enough to contain a single page-aligned page, a 1953 // fresh chunk will be allocated. 1954 bool SetUp(); 1955 1956 // Returns true if the space has been successfully set up and not 1957 // subsequently torn down. 1958 bool HasBeenSetUp(); 1959 1960 // Checks whether an object/address is in this space. 1961 inline bool Contains(Address a); 1962 inline bool Contains(HeapObject* o); 1963 // Unlike Contains() methods it is safe to call this one even for addresses 1964 // of unmapped memory. 1965 bool ContainsSafe(Address addr); 1966 1967 // Given an address occupied by a live object, return that object if it is 1968 // in this space, or a Smi if it is not. The implementation iterates over 1969 // objects in the page containing the address, the cost is linear in the 1970 // number of objects in the page. It may be slow. 1971 Object* FindObject(Address addr); 1972 1973 // During boot the free_space_map is created, and afterwards we may need 1974 // to write it into the free list nodes that were already created. 1975 void RepairFreeListsAfterDeserialization(); 1976 1977 // Prepares for a mark-compact GC. 1978 void PrepareForMarkCompact(); 1979 1980 // Current capacity without growing (Size() + Available()). 1981 intptr_t Capacity() { return accounting_stats_.Capacity(); } 1982 1983 // Approximate amount of physical memory committed for this space. 1984 size_t CommittedPhysicalMemory() override; 1985 1986 void ResetFreeListStatistics(); 1987 1988 // Sets the capacity, the available space and the wasted space to zero. 1989 // The stats are rebuilt during sweeping by adding each page to the 1990 // capacity and the size when it is encountered. As free spaces are 1991 // discovered during the sweeping they are subtracted from the size and added 1992 // to the available and wasted totals. 1993 void ClearStats() { 1994 accounting_stats_.ClearSize(); 1995 free_list_.ResetStats(); 1996 ResetFreeListStatistics(); 1997 } 1998 1999 // Increases the number of available bytes of that space. 2000 void AddToAccountingStats(intptr_t bytes) { 2001 accounting_stats_.DeallocateBytes(bytes); 2002 } 2003 2004 // Available bytes without growing. These are the bytes on the free list. 2005 // The bytes in the linear allocation area are not included in this total 2006 // because updating the stats would slow down allocation. New pages are 2007 // immediately added to the free list so they show up here. 2008 intptr_t Available() override { return free_list_.Available(); } 2009 2010 // Allocated bytes in this space. Garbage bytes that were not found due to 2011 // concurrent sweeping are counted as being allocated! The bytes in the 2012 // current linear allocation area (between top and limit) are also counted 2013 // here. 2014 intptr_t Size() override { return accounting_stats_.Size(); } 2015 2016 // As size, but the bytes in lazily swept pages are estimated and the bytes 2017 // in the current linear allocation area are not included. 2018 intptr_t SizeOfObjects() override; 2019 2020 // Wasted bytes in this space. These are just the bytes that were thrown away 2021 // due to being too small to use for allocation. 2022 virtual intptr_t Waste() { return free_list_.wasted_bytes(); } 2023 2024 // Returns the allocation pointer in this space. 2025 Address top() { return allocation_info_.top(); } 2026 Address limit() { return allocation_info_.limit(); } 2027 2028 // The allocation top address. 2029 Address* allocation_top_address() { return allocation_info_.top_address(); } 2030 2031 // The allocation limit address. 2032 Address* allocation_limit_address() { 2033 return allocation_info_.limit_address(); 2034 } 2035 2036 // Allocate the requested number of bytes in the space if possible, return a 2037 // failure object if not. 2038 MUST_USE_RESULT inline AllocationResult AllocateRawUnaligned( 2039 int size_in_bytes); 2040 2041 MUST_USE_RESULT inline AllocationResult AllocateRawUnalignedSynchronized( 2042 int size_in_bytes); 2043 2044 // Allocate the requested number of bytes in the space double aligned if 2045 // possible, return a failure object if not. 2046 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 2047 int size_in_bytes, AllocationAlignment alignment); 2048 2049 // Allocate the requested number of bytes in the space and consider allocation 2050 // alignment if needed. 2051 MUST_USE_RESULT inline AllocationResult AllocateRaw( 2052 int size_in_bytes, AllocationAlignment alignment); 2053 2054 // Give a block of memory to the space's free list. It might be added to 2055 // the free list or accounted as waste. 2056 // If add_to_freelist is false then just accounting stats are updated and 2057 // no attempt to add area to free list is made. 2058 int Free(Address start, int size_in_bytes) { 2059 int wasted = free_list_.Free(start, size_in_bytes); 2060 accounting_stats_.DeallocateBytes(size_in_bytes); 2061 return size_in_bytes - wasted; 2062 } 2063 2064 void ResetFreeList() { free_list_.Reset(); } 2065 2066 // Set space allocation info. 2067 void SetTopAndLimit(Address top, Address limit) { 2068 DCHECK(top == limit || 2069 Page::FromAddress(top) == Page::FromAddress(limit - 1)); 2070 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 2071 allocation_info_.Reset(top, limit); 2072 } 2073 2074 // Empty space allocation info, returning unused area to free list. 2075 void EmptyAllocationInfo() { 2076 // Mark the old linear allocation area with a free space map so it can be 2077 // skipped when scanning the heap. 2078 int old_linear_size = static_cast<int>(limit() - top()); 2079 Free(top(), old_linear_size); 2080 SetTopAndLimit(NULL, NULL); 2081 } 2082 2083 void Allocate(int bytes) { accounting_stats_.AllocateBytes(bytes); } 2084 2085 void IncreaseCapacity(int size); 2086 2087 // Releases an unused page and shrinks the space. 2088 void ReleasePage(Page* page); 2089 2090 // The dummy page that anchors the linked list of pages. 2091 Page* anchor() { return &anchor_; } 2092 2093 #ifdef VERIFY_HEAP 2094 // Verify integrity of this space. 2095 virtual void Verify(ObjectVisitor* visitor); 2096 2097 // Overridden by subclasses to verify space-specific object 2098 // properties (e.g., only maps or free-list nodes are in map space). 2099 virtual void VerifyObject(HeapObject* obj) {} 2100 #endif 2101 2102 #ifdef DEBUG 2103 // Print meta info and objects in this space. 2104 void Print() override; 2105 2106 // Reports statistics for the space 2107 void ReportStatistics(); 2108 2109 // Report code object related statistics 2110 void CollectCodeStatistics(); 2111 static void ReportCodeStatistics(Isolate* isolate); 2112 static void ResetCodeStatistics(Isolate* isolate); 2113 #endif 2114 2115 // Evacuation candidates are swept by evacuator. Needs to return a valid 2116 // result before _and_ after evacuation has finished. 2117 static bool ShouldBeSweptBySweeperThreads(Page* p) { 2118 return !p->IsEvacuationCandidate() && 2119 !p->IsFlagSet(Page::RESCAN_ON_EVACUATION) && !p->WasSwept(); 2120 } 2121 2122 // This function tries to steal size_in_bytes memory from the sweeper threads 2123 // free-lists. If it does not succeed stealing enough memory, it will wait 2124 // for the sweeper threads to finish sweeping. 2125 // It returns true when sweeping is completed and false otherwise. 2126 bool EnsureSweeperProgress(intptr_t size_in_bytes); 2127 2128 void set_end_of_unswept_pages(Page* page) { end_of_unswept_pages_ = page; } 2129 2130 Page* end_of_unswept_pages() { return end_of_unswept_pages_; } 2131 2132 Page* FirstPage() { return anchor_.next_page(); } 2133 Page* LastPage() { return anchor_.prev_page(); } 2134 2135 void EvictEvacuationCandidatesFromLinearAllocationArea(); 2136 2137 bool CanExpand(size_t size); 2138 2139 // Returns the number of total pages in this space. 2140 int CountTotalPages(); 2141 2142 // Return size of allocatable area on a page in this space. 2143 inline int AreaSize() { return area_size_; } 2144 2145 virtual bool is_local() { return false; } 2146 2147 // Merges {other} into the current space. Note that this modifies {other}, 2148 // e.g., removes its bump pointer area and resets statistics. 2149 void MergeCompactionSpace(CompactionSpace* other); 2150 2151 void DivideUponCompactionSpaces(CompactionSpaceCollection** other, int num, 2152 intptr_t limit = kCompactionMemoryWanted); 2153 2154 // Refills the free list from the corresponding free list filled by the 2155 // sweeper. 2156 virtual void RefillFreeList(); 2157 2158 protected: 2159 void AddMemory(Address start, intptr_t size); 2160 2161 FreeSpace* TryRemoveMemory(intptr_t size_in_bytes); 2162 2163 void MoveOverFreeMemory(PagedSpace* other); 2164 2165 // PagedSpaces that should be included in snapshots have different, i.e., 2166 // smaller, initial pages. 2167 virtual bool snapshotable() { return true; } 2168 2169 FreeList* free_list() { return &free_list_; } 2170 2171 bool HasPages() { return anchor_.next_page() != &anchor_; } 2172 2173 // Cleans up the space, frees all pages in this space except those belonging 2174 // to the initial chunk, uncommits addresses in the initial chunk. 2175 void TearDown(); 2176 2177 // Expands the space by allocating a fixed number of pages. Returns false if 2178 // it cannot allocate requested number of pages from OS, or if the hard heap 2179 // size limit has been hit. 2180 bool Expand(); 2181 2182 // Generic fast case allocation function that tries linear allocation at the 2183 // address denoted by top in allocation_info_. 2184 inline HeapObject* AllocateLinearly(int size_in_bytes); 2185 2186 // Generic fast case allocation function that tries aligned linear allocation 2187 // at the address denoted by top in allocation_info_. Writes the aligned 2188 // allocation size, which includes the filler size, to size_in_bytes. 2189 inline HeapObject* AllocateLinearlyAligned(int* size_in_bytes, 2190 AllocationAlignment alignment); 2191 2192 // If sweeping is still in progress try to sweep unswept pages. If that is 2193 // not successful, wait for the sweeper threads and re-try free-list 2194 // allocation. 2195 MUST_USE_RESULT virtual HeapObject* SweepAndRetryAllocation( 2196 int size_in_bytes); 2197 2198 // Slow path of AllocateRaw. This function is space-dependent. 2199 MUST_USE_RESULT HeapObject* SlowAllocateRaw(int size_in_bytes); 2200 2201 int area_size_; 2202 2203 // Accounting information for this space. 2204 AllocationStats accounting_stats_; 2205 2206 // The dummy page that anchors the double linked list of pages. 2207 Page anchor_; 2208 2209 // The space's free list. 2210 FreeList free_list_; 2211 2212 // Normal allocation information. 2213 AllocationInfo allocation_info_; 2214 2215 // The sweeper threads iterate over the list of pointer and data space pages 2216 // and sweep these pages concurrently. They will stop sweeping after the 2217 // end_of_unswept_pages_ page. 2218 Page* end_of_unswept_pages_; 2219 2220 // Mutex guarding any concurrent access to the space. 2221 base::Mutex space_mutex_; 2222 2223 friend class MarkCompactCollector; 2224 friend class PageIterator; 2225 2226 // Used in cctest. 2227 friend class HeapTester; 2228 }; 2229 2230 2231 class NumberAndSizeInfo BASE_EMBEDDED { 2232 public: 2233 NumberAndSizeInfo() : number_(0), bytes_(0) {} 2234 2235 int number() const { return number_; } 2236 void increment_number(int num) { number_ += num; } 2237 2238 int bytes() const { return bytes_; } 2239 void increment_bytes(int size) { bytes_ += size; } 2240 2241 void clear() { 2242 number_ = 0; 2243 bytes_ = 0; 2244 } 2245 2246 private: 2247 int number_; 2248 int bytes_; 2249 }; 2250 2251 2252 // HistogramInfo class for recording a single "bar" of a histogram. This 2253 // class is used for collecting statistics to print to the log file. 2254 class HistogramInfo : public NumberAndSizeInfo { 2255 public: 2256 HistogramInfo() : NumberAndSizeInfo() {} 2257 2258 const char* name() { return name_; } 2259 void set_name(const char* name) { name_ = name; } 2260 2261 private: 2262 const char* name_; 2263 }; 2264 2265 2266 enum SemiSpaceId { kFromSpace = 0, kToSpace = 1 }; 2267 2268 2269 class SemiSpace; 2270 2271 2272 class NewSpacePage : public MemoryChunk { 2273 public: 2274 // GC related flags copied from from-space to to-space when 2275 // flipping semispaces. 2276 static const intptr_t kCopyOnFlipFlagsMask = 2277 (1 << MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING) | 2278 (1 << MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING) | 2279 (1 << MemoryChunk::SCAN_ON_SCAVENGE); 2280 2281 static const int kAreaSize = Page::kAllocatableMemory; 2282 2283 inline NewSpacePage* next_page() { 2284 return static_cast<NewSpacePage*>(next_chunk()); 2285 } 2286 2287 inline void set_next_page(NewSpacePage* page) { set_next_chunk(page); } 2288 2289 inline NewSpacePage* prev_page() { 2290 return static_cast<NewSpacePage*>(prev_chunk()); 2291 } 2292 2293 inline void set_prev_page(NewSpacePage* page) { set_prev_chunk(page); } 2294 2295 SemiSpace* semi_space() { return reinterpret_cast<SemiSpace*>(owner()); } 2296 2297 bool is_anchor() { return !this->InNewSpace(); } 2298 2299 static bool IsAtStart(Address addr) { 2300 return (reinterpret_cast<intptr_t>(addr) & Page::kPageAlignmentMask) == 2301 kObjectStartOffset; 2302 } 2303 2304 static bool IsAtEnd(Address addr) { 2305 return (reinterpret_cast<intptr_t>(addr) & Page::kPageAlignmentMask) == 0; 2306 } 2307 2308 Address address() { return reinterpret_cast<Address>(this); } 2309 2310 // Finds the NewSpacePage containing the given address. 2311 static inline NewSpacePage* FromAddress(Address address_in_page) { 2312 Address page_start = 2313 reinterpret_cast<Address>(reinterpret_cast<uintptr_t>(address_in_page) & 2314 ~Page::kPageAlignmentMask); 2315 NewSpacePage* page = reinterpret_cast<NewSpacePage*>(page_start); 2316 return page; 2317 } 2318 2319 // Find the page for a limit address. A limit address is either an address 2320 // inside a page, or the address right after the last byte of a page. 2321 static inline NewSpacePage* FromLimit(Address address_limit) { 2322 return NewSpacePage::FromAddress(address_limit - 1); 2323 } 2324 2325 // Checks if address1 and address2 are on the same new space page. 2326 static inline bool OnSamePage(Address address1, Address address2) { 2327 return NewSpacePage::FromAddress(address1) == 2328 NewSpacePage::FromAddress(address2); 2329 } 2330 2331 private: 2332 // Create a NewSpacePage object that is only used as anchor 2333 // for the doubly-linked list of real pages. 2334 explicit NewSpacePage(SemiSpace* owner) { InitializeAsAnchor(owner); } 2335 2336 static NewSpacePage* Initialize(Heap* heap, Address start, 2337 SemiSpace* semi_space); 2338 2339 // Intialize a fake NewSpacePage used as sentinel at the ends 2340 // of a doubly-linked list of real NewSpacePages. 2341 // Only uses the prev/next links, and sets flags to not be in new-space. 2342 void InitializeAsAnchor(SemiSpace* owner); 2343 2344 friend class SemiSpace; 2345 friend class SemiSpaceIterator; 2346 }; 2347 2348 2349 // ----------------------------------------------------------------------------- 2350 // SemiSpace in young generation 2351 // 2352 // A semispace is a contiguous chunk of memory holding page-like memory 2353 // chunks. The mark-compact collector uses the memory of the first page in 2354 // the from space as a marking stack when tracing live objects. 2355 2356 class SemiSpace : public Space { 2357 public: 2358 // Constructor. 2359 SemiSpace(Heap* heap, SemiSpaceId semispace) 2360 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2361 start_(NULL), 2362 age_mark_(NULL), 2363 id_(semispace), 2364 anchor_(this), 2365 current_page_(NULL) {} 2366 2367 // Sets up the semispace using the given chunk. 2368 void SetUp(Address start, int initial_capacity, int target_capacity, 2369 int maximum_capacity); 2370 2371 // Tear down the space. Heap memory was not allocated by the space, so it 2372 // is not deallocated here. 2373 void TearDown(); 2374 2375 // True if the space has been set up but not torn down. 2376 bool HasBeenSetUp() { return start_ != NULL; } 2377 2378 // Grow the semispace to the new capacity. The new capacity 2379 // requested must be larger than the current capacity and less than 2380 // the maximum capacity. 2381 bool GrowTo(int new_capacity); 2382 2383 // Shrinks the semispace to the new capacity. The new capacity 2384 // requested must be more than the amount of used memory in the 2385 // semispace and less than the current capacity. 2386 bool ShrinkTo(int new_capacity); 2387 2388 // Sets the total capacity. Only possible when the space is not committed. 2389 bool SetTotalCapacity(int new_capacity); 2390 2391 // Returns the start address of the first page of the space. 2392 Address space_start() { 2393 DCHECK(anchor_.next_page() != &anchor_); 2394 return anchor_.next_page()->area_start(); 2395 } 2396 2397 // Returns the start address of the current page of the space. 2398 Address page_low() { return current_page_->area_start(); } 2399 2400 // Returns one past the end address of the space. 2401 Address space_end() { return anchor_.prev_page()->area_end(); } 2402 2403 // Returns one past the end address of the current page of the space. 2404 Address page_high() { return current_page_->area_end(); } 2405 2406 bool AdvancePage() { 2407 NewSpacePage* next_page = current_page_->next_page(); 2408 if (next_page == anchor()) return false; 2409 current_page_ = next_page; 2410 return true; 2411 } 2412 2413 // Resets the space to using the first page. 2414 void Reset(); 2415 2416 // Age mark accessors. 2417 Address age_mark() { return age_mark_; } 2418 void set_age_mark(Address mark); 2419 2420 // True if the address is in the address range of this semispace (not 2421 // necessarily below the allocation pointer). 2422 bool Contains(Address a) { 2423 return (reinterpret_cast<uintptr_t>(a) & address_mask_) == 2424 reinterpret_cast<uintptr_t>(start_); 2425 } 2426 2427 // True if the object is a heap object in the address range of this 2428 // semispace (not necessarily below the allocation pointer). 2429 bool Contains(Object* o) { 2430 return (reinterpret_cast<uintptr_t>(o) & object_mask_) == object_expected_; 2431 } 2432 2433 // If we don't have these here then SemiSpace will be abstract. However 2434 // they should never be called: 2435 2436 intptr_t Size() override { 2437 UNREACHABLE(); 2438 return 0; 2439 } 2440 2441 intptr_t SizeOfObjects() override { return Size(); } 2442 2443 intptr_t Available() override { 2444 UNREACHABLE(); 2445 return 0; 2446 } 2447 2448 2449 bool is_committed() { return committed_; } 2450 bool Commit(); 2451 bool Uncommit(); 2452 2453 NewSpacePage* first_page() { return anchor_.next_page(); } 2454 NewSpacePage* current_page() { return current_page_; } 2455 2456 #ifdef VERIFY_HEAP 2457 virtual void Verify(); 2458 #endif 2459 2460 #ifdef DEBUG 2461 void Print() override; 2462 // Validate a range of of addresses in a SemiSpace. 2463 // The "from" address must be on a page prior to the "to" address, 2464 // in the linked page order, or it must be earlier on the same page. 2465 static void AssertValidRange(Address from, Address to); 2466 #else 2467 // Do nothing. 2468 inline static void AssertValidRange(Address from, Address to) {} 2469 #endif 2470 2471 // Returns the current total capacity of the semispace. 2472 int TotalCapacity() { return total_capacity_; } 2473 2474 // Returns the target for total capacity of the semispace. 2475 int TargetCapacity() { return target_capacity_; } 2476 2477 // Returns the maximum total capacity of the semispace. 2478 int MaximumTotalCapacity() { return maximum_total_capacity_; } 2479 2480 // Returns the initial capacity of the semispace. 2481 int InitialTotalCapacity() { return initial_total_capacity_; } 2482 2483 SemiSpaceId id() { return id_; } 2484 2485 static void Swap(SemiSpace* from, SemiSpace* to); 2486 2487 // Approximate amount of physical memory committed for this space. 2488 size_t CommittedPhysicalMemory() override; 2489 2490 private: 2491 // Flips the semispace between being from-space and to-space. 2492 // Copies the flags into the masked positions on all pages in the space. 2493 void FlipPages(intptr_t flags, intptr_t flag_mask); 2494 2495 // Updates Capacity and MaximumCommitted based on new capacity. 2496 void SetCapacity(int new_capacity); 2497 2498 NewSpacePage* anchor() { return &anchor_; } 2499 2500 // The current and maximum total capacity of the space. 2501 int total_capacity_; 2502 int target_capacity_; 2503 int maximum_total_capacity_; 2504 int initial_total_capacity_; 2505 2506 // The start address of the space. 2507 Address start_; 2508 // Used to govern object promotion during mark-compact collection. 2509 Address age_mark_; 2510 2511 // Masks and comparison values to test for containment in this semispace. 2512 uintptr_t address_mask_; 2513 uintptr_t object_mask_; 2514 uintptr_t object_expected_; 2515 2516 bool committed_; 2517 SemiSpaceId id_; 2518 2519 NewSpacePage anchor_; 2520 NewSpacePage* current_page_; 2521 2522 friend class SemiSpaceIterator; 2523 friend class NewSpacePageIterator; 2524 }; 2525 2526 2527 // A SemiSpaceIterator is an ObjectIterator that iterates over the active 2528 // semispace of the heap's new space. It iterates over the objects in the 2529 // semispace from a given start address (defaulting to the bottom of the 2530 // semispace) to the top of the semispace. New objects allocated after the 2531 // iterator is created are not iterated. 2532 class SemiSpaceIterator : public ObjectIterator { 2533 public: 2534 // Create an iterator over the allocated objects in the given to-space. 2535 explicit SemiSpaceIterator(NewSpace* space); 2536 2537 inline HeapObject* Next(); 2538 2539 // Implementation of the ObjectIterator functions. 2540 inline HeapObject* next_object() override; 2541 2542 private: 2543 void Initialize(Address start, Address end); 2544 2545 // The current iteration point. 2546 Address current_; 2547 // The end of iteration. 2548 Address limit_; 2549 }; 2550 2551 2552 // ----------------------------------------------------------------------------- 2553 // A PageIterator iterates the pages in a semi-space. 2554 class NewSpacePageIterator BASE_EMBEDDED { 2555 public: 2556 // Make an iterator that runs over all pages in to-space. 2557 explicit inline NewSpacePageIterator(NewSpace* space); 2558 2559 // Make an iterator that runs over all pages in the given semispace, 2560 // even those not used in allocation. 2561 explicit inline NewSpacePageIterator(SemiSpace* space); 2562 2563 // Make iterator that iterates from the page containing start 2564 // to the page that contains limit in the same semispace. 2565 inline NewSpacePageIterator(Address start, Address limit); 2566 2567 inline bool has_next(); 2568 inline NewSpacePage* next(); 2569 2570 private: 2571 NewSpacePage* prev_page_; // Previous page returned. 2572 // Next page that will be returned. Cached here so that we can use this 2573 // iterator for operations that deallocate pages. 2574 NewSpacePage* next_page_; 2575 // Last page returned. 2576 NewSpacePage* last_page_; 2577 }; 2578 2579 // ----------------------------------------------------------------------------- 2580 // Allows observation of inline allocation in the new space. 2581 class InlineAllocationObserver { 2582 public: 2583 explicit InlineAllocationObserver(intptr_t step_size) 2584 : step_size_(step_size), bytes_to_next_step_(step_size) { 2585 DCHECK(step_size >= kPointerSize); 2586 } 2587 virtual ~InlineAllocationObserver() {} 2588 2589 private: 2590 intptr_t step_size() const { return step_size_; } 2591 intptr_t bytes_to_next_step() const { return bytes_to_next_step_; } 2592 2593 // Pure virtual method provided by the subclasses that gets called when at 2594 // least step_size bytes have been allocated. soon_object is the address just 2595 // allocated (but not yet initialized.) size is the size of the object as 2596 // requested (i.e. w/o the alignment fillers). Some complexities to be aware 2597 // of: 2598 // 1) soon_object will be nullptr in cases where we end up observing an 2599 // allocation that happens to be a filler space (e.g. page boundaries.) 2600 // 2) size is the requested size at the time of allocation. Right-trimming 2601 // may change the object size dynamically. 2602 // 3) soon_object may actually be the first object in an allocation-folding 2603 // group. In such a case size is the size of the group rather than the 2604 // first object. 2605 virtual void Step(int bytes_allocated, Address soon_object, size_t size) = 0; 2606 2607 // Called each time the new space does an inline allocation step. This may be 2608 // more frequently than the step_size we are monitoring (e.g. when there are 2609 // multiple observers, or when page or space boundary is encountered.) 2610 void InlineAllocationStep(int bytes_allocated, Address soon_object, 2611 size_t size) { 2612 bytes_to_next_step_ -= bytes_allocated; 2613 if (bytes_to_next_step_ <= 0) { 2614 Step(static_cast<int>(step_size_ - bytes_to_next_step_), soon_object, 2615 size); 2616 bytes_to_next_step_ = step_size_; 2617 } 2618 } 2619 2620 intptr_t step_size_; 2621 intptr_t bytes_to_next_step_; 2622 2623 friend class NewSpace; 2624 2625 DISALLOW_COPY_AND_ASSIGN(InlineAllocationObserver); 2626 }; 2627 2628 // ----------------------------------------------------------------------------- 2629 // The young generation space. 2630 // 2631 // The new space consists of a contiguous pair of semispaces. It simply 2632 // forwards most functions to the appropriate semispace. 2633 2634 class NewSpace : public Space { 2635 public: 2636 // Constructor. 2637 explicit NewSpace(Heap* heap) 2638 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2639 to_space_(heap, kToSpace), 2640 from_space_(heap, kFromSpace), 2641 reservation_(), 2642 top_on_previous_step_(0), 2643 inline_allocation_observers_paused_(false) {} 2644 2645 // Sets up the new space using the given chunk. 2646 bool SetUp(int reserved_semispace_size_, int max_semi_space_size); 2647 2648 // Tears down the space. Heap memory was not allocated by the space, so it 2649 // is not deallocated here. 2650 void TearDown(); 2651 2652 // True if the space has been set up but not torn down. 2653 bool HasBeenSetUp() { 2654 return to_space_.HasBeenSetUp() && from_space_.HasBeenSetUp(); 2655 } 2656 2657 // Flip the pair of spaces. 2658 void Flip(); 2659 2660 // Grow the capacity of the semispaces. Assumes that they are not at 2661 // their maximum capacity. 2662 void Grow(); 2663 2664 // Grow the capacity of the semispaces by one page. 2665 bool GrowOnePage(); 2666 2667 // Shrink the capacity of the semispaces. 2668 void Shrink(); 2669 2670 // True if the address or object lies in the address range of either 2671 // semispace (not necessarily below the allocation pointer). 2672 bool Contains(Address a) { 2673 return (reinterpret_cast<uintptr_t>(a) & address_mask_) == 2674 reinterpret_cast<uintptr_t>(start_); 2675 } 2676 2677 bool Contains(Object* o) { 2678 Address a = reinterpret_cast<Address>(o); 2679 return (reinterpret_cast<uintptr_t>(a) & object_mask_) == object_expected_; 2680 } 2681 2682 // Return the allocated bytes in the active semispace. 2683 intptr_t Size() override { 2684 return pages_used_ * NewSpacePage::kAreaSize + 2685 static_cast<int>(top() - to_space_.page_low()); 2686 } 2687 2688 // The same, but returning an int. We have to have the one that returns 2689 // intptr_t because it is inherited, but if we know we are dealing with the 2690 // new space, which can't get as big as the other spaces then this is useful: 2691 int SizeAsInt() { return static_cast<int>(Size()); } 2692 2693 // Return the allocatable capacity of a semispace. 2694 intptr_t Capacity() { 2695 SLOW_DCHECK(to_space_.TotalCapacity() == from_space_.TotalCapacity()); 2696 return (to_space_.TotalCapacity() / Page::kPageSize) * 2697 NewSpacePage::kAreaSize; 2698 } 2699 2700 // Return the current size of a semispace, allocatable and non-allocatable 2701 // memory. 2702 intptr_t TotalCapacity() { 2703 DCHECK(to_space_.TotalCapacity() == from_space_.TotalCapacity()); 2704 return to_space_.TotalCapacity(); 2705 } 2706 2707 // Committed memory for NewSpace is the committed memory of both semi-spaces 2708 // combined. 2709 intptr_t CommittedMemory() override { 2710 return from_space_.CommittedMemory() + to_space_.CommittedMemory(); 2711 } 2712 2713 intptr_t MaximumCommittedMemory() override { 2714 return from_space_.MaximumCommittedMemory() + 2715 to_space_.MaximumCommittedMemory(); 2716 } 2717 2718 // Approximate amount of physical memory committed for this space. 2719 size_t CommittedPhysicalMemory() override; 2720 2721 // Return the available bytes without growing. 2722 intptr_t Available() override { return Capacity() - Size(); } 2723 2724 intptr_t PagesFromStart(Address addr) { 2725 return static_cast<intptr_t>(addr - bottom()) / Page::kPageSize; 2726 } 2727 2728 size_t AllocatedSinceLastGC() { 2729 intptr_t allocated = top() - to_space_.age_mark(); 2730 if (allocated < 0) { 2731 // Runtime has lowered the top below the age mark. 2732 return 0; 2733 } 2734 // Correctly account for non-allocatable regions at the beginning of 2735 // each page from the age_mark() to the top(). 2736 intptr_t pages = 2737 PagesFromStart(top()) - PagesFromStart(to_space_.age_mark()); 2738 allocated -= pages * (NewSpacePage::kObjectStartOffset); 2739 DCHECK(0 <= allocated && allocated <= Size()); 2740 return static_cast<size_t>(allocated); 2741 } 2742 2743 // Return the maximum capacity of a semispace. 2744 int MaximumCapacity() { 2745 DCHECK(to_space_.MaximumTotalCapacity() == 2746 from_space_.MaximumTotalCapacity()); 2747 return to_space_.MaximumTotalCapacity(); 2748 } 2749 2750 bool IsAtMaximumCapacity() { return TotalCapacity() == MaximumCapacity(); } 2751 2752 // Returns the initial capacity of a semispace. 2753 int InitialTotalCapacity() { 2754 DCHECK(to_space_.InitialTotalCapacity() == 2755 from_space_.InitialTotalCapacity()); 2756 return to_space_.InitialTotalCapacity(); 2757 } 2758 2759 // Return the address of the allocation pointer in the active semispace. 2760 Address top() { 2761 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.top())); 2762 return allocation_info_.top(); 2763 } 2764 2765 // Return the address of the allocation pointer limit in the active semispace. 2766 Address limit() { 2767 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.limit())); 2768 return allocation_info_.limit(); 2769 } 2770 2771 // Return the address of the first object in the active semispace. 2772 Address bottom() { return to_space_.space_start(); } 2773 2774 // Get the age mark of the inactive semispace. 2775 Address age_mark() { return from_space_.age_mark(); } 2776 // Set the age mark in the active semispace. 2777 void set_age_mark(Address mark) { to_space_.set_age_mark(mark); } 2778 2779 // The start address of the space and a bit mask. Anding an address in the 2780 // new space with the mask will result in the start address. 2781 Address start() { return start_; } 2782 uintptr_t mask() { return address_mask_; } 2783 2784 INLINE(uint32_t AddressToMarkbitIndex(Address addr)) { 2785 DCHECK(Contains(addr)); 2786 DCHECK(IsAligned(OffsetFrom(addr), kPointerSize) || 2787 IsAligned(OffsetFrom(addr) - 1, kPointerSize)); 2788 return static_cast<uint32_t>(addr - start_) >> kPointerSizeLog2; 2789 } 2790 2791 INLINE(Address MarkbitIndexToAddress(uint32_t index)) { 2792 return reinterpret_cast<Address>(index << kPointerSizeLog2); 2793 } 2794 2795 // The allocation top and limit address. 2796 Address* allocation_top_address() { return allocation_info_.top_address(); } 2797 2798 // The allocation limit address. 2799 Address* allocation_limit_address() { 2800 return allocation_info_.limit_address(); 2801 } 2802 2803 MUST_USE_RESULT INLINE(AllocationResult AllocateRawAligned( 2804 int size_in_bytes, AllocationAlignment alignment)); 2805 2806 MUST_USE_RESULT INLINE( 2807 AllocationResult AllocateRawUnaligned(int size_in_bytes)); 2808 2809 MUST_USE_RESULT INLINE(AllocationResult AllocateRaw( 2810 int size_in_bytes, AllocationAlignment alignment)); 2811 2812 MUST_USE_RESULT inline AllocationResult AllocateRawSynchronized( 2813 int size_in_bytes, AllocationAlignment alignment); 2814 2815 // Reset the allocation pointer to the beginning of the active semispace. 2816 void ResetAllocationInfo(); 2817 2818 void UpdateInlineAllocationLimit(int size_in_bytes); 2819 2820 // Allows observation of inline allocation. The observer->Step() method gets 2821 // called after every step_size bytes have been allocated (approximately). 2822 // This works by adjusting the allocation limit to a lower value and adjusting 2823 // it after each step. 2824 void AddInlineAllocationObserver(InlineAllocationObserver* observer); 2825 2826 // Removes a previously installed observer. 2827 void RemoveInlineAllocationObserver(InlineAllocationObserver* observer); 2828 2829 void DisableInlineAllocationSteps() { 2830 top_on_previous_step_ = 0; 2831 UpdateInlineAllocationLimit(0); 2832 } 2833 2834 // Get the extent of the inactive semispace (for use as a marking stack, 2835 // or to zap it). Notice: space-addresses are not necessarily on the 2836 // same page, so FromSpaceStart() might be above FromSpaceEnd(). 2837 Address FromSpacePageLow() { return from_space_.page_low(); } 2838 Address FromSpacePageHigh() { return from_space_.page_high(); } 2839 Address FromSpaceStart() { return from_space_.space_start(); } 2840 Address FromSpaceEnd() { return from_space_.space_end(); } 2841 2842 // Get the extent of the active semispace's pages' memory. 2843 Address ToSpaceStart() { return to_space_.space_start(); } 2844 Address ToSpaceEnd() { return to_space_.space_end(); } 2845 2846 inline bool ToSpaceContains(Address address) { 2847 return to_space_.Contains(address); 2848 } 2849 inline bool FromSpaceContains(Address address) { 2850 return from_space_.Contains(address); 2851 } 2852 2853 // True if the object is a heap object in the address range of the 2854 // respective semispace (not necessarily below the allocation pointer of the 2855 // semispace). 2856 inline bool ToSpaceContains(Object* o) { return to_space_.Contains(o); } 2857 inline bool FromSpaceContains(Object* o) { return from_space_.Contains(o); } 2858 2859 // Try to switch the active semispace to a new, empty, page. 2860 // Returns false if this isn't possible or reasonable (i.e., there 2861 // are no pages, or the current page is already empty), or true 2862 // if successful. 2863 bool AddFreshPage(); 2864 bool AddFreshPageSynchronized(); 2865 2866 #ifdef VERIFY_HEAP 2867 // Verify the active semispace. 2868 virtual void Verify(); 2869 #endif 2870 2871 #ifdef DEBUG 2872 // Print the active semispace. 2873 void Print() override { to_space_.Print(); } 2874 #endif 2875 2876 // Iterates the active semispace to collect statistics. 2877 void CollectStatistics(); 2878 // Reports previously collected statistics of the active semispace. 2879 void ReportStatistics(); 2880 // Clears previously collected statistics. 2881 void ClearHistograms(); 2882 2883 // Record the allocation or promotion of a heap object. Note that we don't 2884 // record every single allocation, but only those that happen in the 2885 // to space during a scavenge GC. 2886 void RecordAllocation(HeapObject* obj); 2887 void RecordPromotion(HeapObject* obj); 2888 2889 // Return whether the operation succeded. 2890 bool CommitFromSpaceIfNeeded() { 2891 if (from_space_.is_committed()) return true; 2892 return from_space_.Commit(); 2893 } 2894 2895 bool UncommitFromSpace() { 2896 if (!from_space_.is_committed()) return true; 2897 return from_space_.Uncommit(); 2898 } 2899 2900 bool IsFromSpaceCommitted() { return from_space_.is_committed(); } 2901 2902 SemiSpace* active_space() { return &to_space_; } 2903 2904 private: 2905 // Update allocation info to match the current to-space page. 2906 void UpdateAllocationInfo(); 2907 2908 base::Mutex mutex_; 2909 2910 Address chunk_base_; 2911 uintptr_t chunk_size_; 2912 2913 // The semispaces. 2914 SemiSpace to_space_; 2915 SemiSpace from_space_; 2916 base::VirtualMemory reservation_; 2917 int pages_used_; 2918 2919 // Start address and bit mask for containment testing. 2920 Address start_; 2921 uintptr_t address_mask_; 2922 uintptr_t object_mask_; 2923 uintptr_t object_expected_; 2924 2925 // Allocation pointer and limit for normal allocation and allocation during 2926 // mark-compact collection. 2927 AllocationInfo allocation_info_; 2928 2929 // When inline allocation stepping is active, either because of incremental 2930 // marking or because of idle scavenge, we 'interrupt' inline allocation every 2931 // once in a while. This is done by setting allocation_info_.limit to be lower 2932 // than the actual limit and and increasing it in steps to guarantee that the 2933 // observers are notified periodically. 2934 List<InlineAllocationObserver*> inline_allocation_observers_; 2935 Address top_on_previous_step_; 2936 bool inline_allocation_observers_paused_; 2937 2938 HistogramInfo* allocated_histogram_; 2939 HistogramInfo* promoted_histogram_; 2940 2941 bool EnsureAllocation(int size_in_bytes, AllocationAlignment alignment); 2942 2943 // If we are doing inline allocation in steps, this method performs the 'step' 2944 // operation. top is the memory address of the bump pointer at the last 2945 // inline allocation (i.e. it determines the numbers of bytes actually 2946 // allocated since the last step.) new_top is the address of the bump pointer 2947 // where the next byte is going to be allocated from. top and new_top may be 2948 // different when we cross a page boundary or reset the space. 2949 void InlineAllocationStep(Address top, Address new_top, Address soon_object, 2950 size_t size); 2951 intptr_t GetNextInlineAllocationStepSize(); 2952 void StartNextInlineAllocationStep(); 2953 void PauseInlineAllocationObservers(); 2954 void ResumeInlineAllocationObservers(); 2955 2956 friend class PauseInlineAllocationObserversScope; 2957 friend class SemiSpaceIterator; 2958 }; 2959 2960 class PauseInlineAllocationObserversScope { 2961 public: 2962 explicit PauseInlineAllocationObserversScope(NewSpace* new_space) 2963 : new_space_(new_space) { 2964 new_space_->PauseInlineAllocationObservers(); 2965 } 2966 ~PauseInlineAllocationObserversScope() { 2967 new_space_->ResumeInlineAllocationObservers(); 2968 } 2969 2970 private: 2971 NewSpace* new_space_; 2972 DISALLOW_COPY_AND_ASSIGN(PauseInlineAllocationObserversScope); 2973 }; 2974 2975 // ----------------------------------------------------------------------------- 2976 // Compaction space that is used temporarily during compaction. 2977 2978 class CompactionSpace : public PagedSpace { 2979 public: 2980 CompactionSpace(Heap* heap, AllocationSpace id, Executability executable) 2981 : PagedSpace(heap, id, executable) {} 2982 2983 // Adds external memory starting at {start} of {size_in_bytes} to the space. 2984 void AddExternalMemory(Address start, int size_in_bytes) { 2985 IncreaseCapacity(size_in_bytes); 2986 Free(start, size_in_bytes); 2987 } 2988 2989 bool is_local() override { return true; } 2990 2991 void RefillFreeList() override; 2992 2993 protected: 2994 // The space is temporary and not included in any snapshots. 2995 bool snapshotable() override { return false; } 2996 2997 MUST_USE_RESULT HeapObject* SweepAndRetryAllocation( 2998 int size_in_bytes) override; 2999 }; 3000 3001 3002 // A collection of |CompactionSpace|s used by a single compaction task. 3003 class CompactionSpaceCollection : public Malloced { 3004 public: 3005 explicit CompactionSpaceCollection(Heap* heap) 3006 : old_space_(heap, OLD_SPACE, Executability::NOT_EXECUTABLE), 3007 code_space_(heap, CODE_SPACE, Executability::EXECUTABLE), 3008 duration_(0.0), 3009 bytes_compacted_(0) {} 3010 3011 CompactionSpace* Get(AllocationSpace space) { 3012 switch (space) { 3013 case OLD_SPACE: 3014 return &old_space_; 3015 case CODE_SPACE: 3016 return &code_space_; 3017 default: 3018 UNREACHABLE(); 3019 } 3020 UNREACHABLE(); 3021 return nullptr; 3022 } 3023 3024 void ReportCompactionProgress(double duration, intptr_t bytes_compacted) { 3025 duration_ += duration; 3026 bytes_compacted_ += bytes_compacted; 3027 } 3028 3029 double duration() const { return duration_; } 3030 intptr_t bytes_compacted() const { return bytes_compacted_; } 3031 3032 private: 3033 CompactionSpace old_space_; 3034 CompactionSpace code_space_; 3035 3036 // Book keeping. 3037 double duration_; 3038 intptr_t bytes_compacted_; 3039 }; 3040 3041 3042 // ----------------------------------------------------------------------------- 3043 // Old object space (includes the old space of objects and code space) 3044 3045 class OldSpace : public PagedSpace { 3046 public: 3047 // Creates an old space object. The constructor does not allocate pages 3048 // from OS. 3049 OldSpace(Heap* heap, AllocationSpace id, Executability executable) 3050 : PagedSpace(heap, id, executable) {} 3051 }; 3052 3053 3054 // For contiguous spaces, top should be in the space (or at the end) and limit 3055 // should be the end of the space. 3056 #define DCHECK_SEMISPACE_ALLOCATION_INFO(info, space) \ 3057 SLOW_DCHECK((space).page_low() <= (info).top() && \ 3058 (info).top() <= (space).page_high() && \ 3059 (info).limit() <= (space).page_high()) 3060 3061 3062 // ----------------------------------------------------------------------------- 3063 // Old space for all map objects 3064 3065 class MapSpace : public PagedSpace { 3066 public: 3067 // Creates a map space object. 3068 MapSpace(Heap* heap, AllocationSpace id) 3069 : PagedSpace(heap, id, NOT_EXECUTABLE), 3070 max_map_space_pages_(kMaxMapPageIndex - 1) {} 3071 3072 // Given an index, returns the page address. 3073 // TODO(1600): this limit is artifical just to keep code compilable 3074 static const int kMaxMapPageIndex = 1 << 16; 3075 3076 int RoundSizeDownToObjectAlignment(int size) override { 3077 if (base::bits::IsPowerOfTwo32(Map::kSize)) { 3078 return RoundDown(size, Map::kSize); 3079 } else { 3080 return (size / Map::kSize) * Map::kSize; 3081 } 3082 } 3083 3084 #ifdef VERIFY_HEAP 3085 void VerifyObject(HeapObject* obj) override; 3086 #endif 3087 3088 private: 3089 static const int kMapsPerPage = Page::kAllocatableMemory / Map::kSize; 3090 3091 // Do map space compaction if there is a page gap. 3092 int CompactionThreshold() { 3093 return kMapsPerPage * (max_map_space_pages_ - 1); 3094 } 3095 3096 const int max_map_space_pages_; 3097 }; 3098 3099 3100 // ----------------------------------------------------------------------------- 3101 // Large objects ( > Page::kMaxRegularHeapObjectSize ) are allocated and 3102 // managed by the large object space. A large object is allocated from OS 3103 // heap with extra padding bytes (Page::kPageSize + Page::kObjectStartOffset). 3104 // A large object always starts at Page::kObjectStartOffset to a page. 3105 // Large objects do not move during garbage collections. 3106 3107 class LargeObjectSpace : public Space { 3108 public: 3109 LargeObjectSpace(Heap* heap, AllocationSpace id); 3110 virtual ~LargeObjectSpace(); 3111 3112 // Initializes internal data structures. 3113 bool SetUp(); 3114 3115 // Releases internal resources, frees objects in this space. 3116 void TearDown(); 3117 3118 static intptr_t ObjectSizeFor(intptr_t chunk_size) { 3119 if (chunk_size <= (Page::kPageSize + Page::kObjectStartOffset)) return 0; 3120 return chunk_size - Page::kPageSize - Page::kObjectStartOffset; 3121 } 3122 3123 // Shared implementation of AllocateRaw, AllocateRawCode and 3124 // AllocateRawFixedArray. 3125 MUST_USE_RESULT AllocationResult 3126 AllocateRaw(int object_size, Executability executable); 3127 3128 // Available bytes for objects in this space. 3129 inline intptr_t Available() override; 3130 3131 intptr_t Size() override { return size_; } 3132 3133 intptr_t SizeOfObjects() override { return objects_size_; } 3134 3135 // Approximate amount of physical memory committed for this space. 3136 size_t CommittedPhysicalMemory() override; 3137 3138 int PageCount() { return page_count_; } 3139 3140 // Finds an object for a given address, returns a Smi if it is not found. 3141 // The function iterates through all objects in this space, may be slow. 3142 Object* FindObject(Address a); 3143 3144 // Finds a large object page containing the given address, returns NULL 3145 // if such a page doesn't exist. 3146 LargePage* FindPage(Address a); 3147 3148 // Clears the marking state of live objects. 3149 void ClearMarkingStateOfLiveObjects(); 3150 3151 // Frees unmarked objects. 3152 void FreeUnmarkedObjects(); 3153 3154 // Checks whether a heap object is in this space; O(1). 3155 bool Contains(HeapObject* obj); 3156 bool Contains(Address address); 3157 3158 // Checks whether the space is empty. 3159 bool IsEmpty() { return first_page_ == NULL; } 3160 3161 LargePage* first_page() { return first_page_; } 3162 3163 #ifdef VERIFY_HEAP 3164 virtual void Verify(); 3165 #endif 3166 3167 #ifdef DEBUG 3168 void Print() override; 3169 void ReportStatistics(); 3170 void CollectCodeStatistics(); 3171 #endif 3172 // Checks whether an address is in the object area in this space. It 3173 // iterates all objects in the space. May be slow. 3174 bool SlowContains(Address addr) { return FindObject(addr)->IsHeapObject(); } 3175 3176 private: 3177 // The head of the linked list of large object chunks. 3178 LargePage* first_page_; 3179 intptr_t size_; // allocated bytes 3180 int page_count_; // number of chunks 3181 intptr_t objects_size_; // size of objects 3182 // Map MemoryChunk::kAlignment-aligned chunks to large pages covering them 3183 HashMap chunk_map_; 3184 3185 friend class LargeObjectIterator; 3186 }; 3187 3188 3189 class LargeObjectIterator : public ObjectIterator { 3190 public: 3191 explicit LargeObjectIterator(LargeObjectSpace* space); 3192 3193 HeapObject* Next(); 3194 3195 // implementation of ObjectIterator. 3196 virtual HeapObject* next_object() { return Next(); } 3197 3198 private: 3199 LargePage* current_; 3200 }; 3201 3202 3203 // Iterates over the chunks (pages and large object pages) that can contain 3204 // pointers to new space. 3205 class PointerChunkIterator BASE_EMBEDDED { 3206 public: 3207 inline explicit PointerChunkIterator(Heap* heap); 3208 3209 // Return NULL when the iterator is done. 3210 inline MemoryChunk* next(); 3211 3212 private: 3213 enum State { kOldSpaceState, kMapState, kLargeObjectState, kFinishedState }; 3214 State state_; 3215 PageIterator old_iterator_; 3216 PageIterator map_iterator_; 3217 LargeObjectIterator lo_iterator_; 3218 }; 3219 3220 3221 #ifdef DEBUG 3222 struct CommentStatistic { 3223 const char* comment; 3224 int size; 3225 int count; 3226 void Clear() { 3227 comment = NULL; 3228 size = 0; 3229 count = 0; 3230 } 3231 // Must be small, since an iteration is used for lookup. 3232 static const int kMaxComments = 64; 3233 }; 3234 #endif 3235 } // namespace internal 3236 } // namespace v8 3237 3238 #endif // V8_HEAP_SPACES_H_ 3239