1 // Copyright 2011 the V8 project authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #ifndef V8_HEAP_SPACES_H_ 6 #define V8_HEAP_SPACES_H_ 7 8 #include <list> 9 #include <memory> 10 #include <unordered_set> 11 12 #include "src/allocation.h" 13 #include "src/base/atomic-utils.h" 14 #include "src/base/atomicops.h" 15 #include "src/base/bits.h" 16 #include "src/base/hashmap.h" 17 #include "src/base/platform/mutex.h" 18 #include "src/flags.h" 19 #include "src/globals.h" 20 #include "src/heap/heap.h" 21 #include "src/heap/marking.h" 22 #include "src/list.h" 23 #include "src/objects.h" 24 #include "src/utils.h" 25 26 namespace v8 { 27 namespace internal { 28 29 class AllocationInfo; 30 class AllocationObserver; 31 class CompactionSpace; 32 class CompactionSpaceCollection; 33 class FreeList; 34 class Isolate; 35 class LocalArrayBufferTracker; 36 class MemoryAllocator; 37 class MemoryChunk; 38 class Page; 39 class PagedSpace; 40 class SemiSpace; 41 class SkipList; 42 class SlotsBuffer; 43 class SlotSet; 44 class TypedSlotSet; 45 class Space; 46 47 // ----------------------------------------------------------------------------- 48 // Heap structures: 49 // 50 // A JS heap consists of a young generation, an old generation, and a large 51 // object space. The young generation is divided into two semispaces. A 52 // scavenger implements Cheney's copying algorithm. The old generation is 53 // separated into a map space and an old object space. The map space contains 54 // all (and only) map objects, the rest of old objects go into the old space. 55 // The old generation is collected by a mark-sweep-compact collector. 56 // 57 // The semispaces of the young generation are contiguous. The old and map 58 // spaces consists of a list of pages. A page has a page header and an object 59 // area. 60 // 61 // There is a separate large object space for objects larger than 62 // kMaxRegularHeapObjectSize, so that they do not have to move during 63 // collection. The large object space is paged. Pages in large object space 64 // may be larger than the page size. 65 // 66 // A store-buffer based write barrier is used to keep track of intergenerational 67 // references. See heap/store-buffer.h. 68 // 69 // During scavenges and mark-sweep collections we sometimes (after a store 70 // buffer overflow) iterate intergenerational pointers without decoding heap 71 // object maps so if the page belongs to old space or large object space 72 // it is essential to guarantee that the page does not contain any 73 // garbage pointers to new space: every pointer aligned word which satisfies 74 // the Heap::InNewSpace() predicate must be a pointer to a live heap object in 75 // new space. Thus objects in old space and large object spaces should have a 76 // special layout (e.g. no bare integer fields). This requirement does not 77 // apply to map space which is iterated in a special fashion. However we still 78 // require pointer fields of dead maps to be cleaned. 79 // 80 // To enable lazy cleaning of old space pages we can mark chunks of the page 81 // as being garbage. Garbage sections are marked with a special map. These 82 // sections are skipped when scanning the page, even if we are otherwise 83 // scanning without regard for object boundaries. Garbage sections are chained 84 // together to form a free list after a GC. Garbage sections created outside 85 // of GCs by object trunctation etc. may not be in the free list chain. Very 86 // small free spaces are ignored, they need only be cleaned of bogus pointers 87 // into new space. 88 // 89 // Each page may have up to one special garbage section. The start of this 90 // section is denoted by the top field in the space. The end of the section 91 // is denoted by the limit field in the space. This special garbage section 92 // is not marked with a free space map in the data. The point of this section 93 // is to enable linear allocation without having to constantly update the byte 94 // array every time the top field is updated and a new object is created. The 95 // special garbage section is not in the chain of garbage sections. 96 // 97 // Since the top and limit fields are in the space, not the page, only one page 98 // has a special garbage section, and if the top and limit are equal then there 99 // is no special garbage section. 100 101 // Some assertion macros used in the debugging mode. 102 103 #define DCHECK_PAGE_ALIGNED(address) \ 104 DCHECK((OffsetFrom(address) & Page::kPageAlignmentMask) == 0) 105 106 #define DCHECK_OBJECT_ALIGNED(address) \ 107 DCHECK((OffsetFrom(address) & kObjectAlignmentMask) == 0) 108 109 #define DCHECK_OBJECT_SIZE(size) \ 110 DCHECK((0 < size) && (size <= kMaxRegularHeapObjectSize)) 111 112 #define DCHECK_CODEOBJECT_SIZE(size, code_space) \ 113 DCHECK((0 < size) && (size <= code_space->AreaSize())) 114 115 #define DCHECK_PAGE_OFFSET(offset) \ 116 DCHECK((Page::kObjectStartOffset <= offset) && (offset <= Page::kPageSize)) 117 118 enum FreeListCategoryType { 119 kTiniest, 120 kTiny, 121 kSmall, 122 kMedium, 123 kLarge, 124 kHuge, 125 126 kFirstCategory = kTiniest, 127 kLastCategory = kHuge, 128 kNumberOfCategories = kLastCategory + 1, 129 kInvalidCategory 130 }; 131 132 enum FreeMode { kLinkCategory, kDoNotLinkCategory }; 133 134 // A free list category maintains a linked list of free memory blocks. 135 class FreeListCategory { 136 public: 137 static const int kSize = kIntSize + // FreeListCategoryType type_ 138 kIntSize + // padding for type_ 139 kSizetSize + // size_t available_ 140 kPointerSize + // FreeSpace* top_ 141 kPointerSize + // FreeListCategory* prev_ 142 kPointerSize; // FreeListCategory* next_ 143 144 FreeListCategory() 145 : type_(kInvalidCategory), 146 available_(0), 147 top_(nullptr), 148 prev_(nullptr), 149 next_(nullptr) {} 150 151 void Initialize(FreeListCategoryType type) { 152 type_ = type; 153 available_ = 0; 154 top_ = nullptr; 155 prev_ = nullptr; 156 next_ = nullptr; 157 } 158 159 void Invalidate(); 160 161 void Reset(); 162 163 void ResetStats() { Reset(); } 164 165 void RepairFreeList(Heap* heap); 166 167 // Relinks the category into the currently owning free list. Requires that the 168 // category is currently unlinked. 169 void Relink(); 170 171 bool Free(FreeSpace* node, size_t size_in_bytes, FreeMode mode); 172 173 // Picks a node from the list and stores its size in |node_size|. Returns 174 // nullptr if the category is empty. 175 FreeSpace* PickNodeFromList(size_t* node_size); 176 177 // Performs a single try to pick a node of at least |minimum_size| from the 178 // category. Stores the actual size in |node_size|. Returns nullptr if no 179 // node is found. 180 FreeSpace* TryPickNodeFromList(size_t minimum_size, size_t* node_size); 181 182 // Picks a node of at least |minimum_size| from the category. Stores the 183 // actual size in |node_size|. Returns nullptr if no node is found. 184 FreeSpace* SearchForNodeInList(size_t minimum_size, size_t* node_size); 185 186 inline FreeList* owner(); 187 inline bool is_linked(); 188 bool is_empty() { return top() == nullptr; } 189 size_t available() const { return available_; } 190 191 #ifdef DEBUG 192 size_t SumFreeList(); 193 int FreeListLength(); 194 #endif 195 196 private: 197 // For debug builds we accurately compute free lists lengths up until 198 // {kVeryLongFreeList} by manually walking the list. 199 static const int kVeryLongFreeList = 500; 200 201 inline Page* page(); 202 203 FreeSpace* top() { return top_; } 204 void set_top(FreeSpace* top) { top_ = top; } 205 FreeListCategory* prev() { return prev_; } 206 void set_prev(FreeListCategory* prev) { prev_ = prev; } 207 FreeListCategory* next() { return next_; } 208 void set_next(FreeListCategory* next) { next_ = next; } 209 210 // |type_|: The type of this free list category. 211 FreeListCategoryType type_; 212 213 // |available_|: Total available bytes in all blocks of this free list 214 // category. 215 size_t available_; 216 217 // |top_|: Points to the top FreeSpace* in the free list category. 218 FreeSpace* top_; 219 220 FreeListCategory* prev_; 221 FreeListCategory* next_; 222 223 friend class FreeList; 224 friend class PagedSpace; 225 }; 226 227 // MemoryChunk represents a memory region owned by a specific space. 228 // It is divided into the header and the body. Chunk start is always 229 // 1MB aligned. Start of the body is aligned so it can accommodate 230 // any heap object. 231 class MemoryChunk { 232 public: 233 enum Flag { 234 NO_FLAGS = 0u, 235 IS_EXECUTABLE = 1u << 0, 236 POINTERS_TO_HERE_ARE_INTERESTING = 1u << 1, 237 POINTERS_FROM_HERE_ARE_INTERESTING = 1u << 2, 238 // A page in new space has one of the next to flags set. 239 IN_FROM_SPACE = 1u << 3, 240 IN_TO_SPACE = 1u << 4, 241 NEW_SPACE_BELOW_AGE_MARK = 1u << 5, 242 EVACUATION_CANDIDATE = 1u << 6, 243 NEVER_EVACUATE = 1u << 7, 244 245 // Large objects can have a progress bar in their page header. These object 246 // are scanned in increments and will be kept black while being scanned. 247 // Even if the mutator writes to them they will be kept black and a white 248 // to grey transition is performed in the value. 249 HAS_PROGRESS_BAR = 1u << 8, 250 251 // |PAGE_NEW_OLD_PROMOTION|: A page tagged with this flag has been promoted 252 // from new to old space during evacuation. 253 PAGE_NEW_OLD_PROMOTION = 1u << 9, 254 255 // |PAGE_NEW_NEW_PROMOTION|: A page tagged with this flag has been moved 256 // within the new space during evacuation. 257 PAGE_NEW_NEW_PROMOTION = 1u << 10, 258 259 // This flag is intended to be used for testing. Works only when both 260 // FLAG_stress_compaction and FLAG_manual_evacuation_candidates_selection 261 // are set. It forces the page to become an evacuation candidate at next 262 // candidates selection cycle. 263 FORCE_EVACUATION_CANDIDATE_FOR_TESTING = 1u << 11, 264 265 // This flag is intended to be used for testing. 266 NEVER_ALLOCATE_ON_PAGE = 1u << 12, 267 268 // The memory chunk is already logically freed, however the actual freeing 269 // still has to be performed. 270 PRE_FREED = 1u << 13, 271 272 // |POOLED|: When actually freeing this chunk, only uncommit and do not 273 // give up the reservation as we still reuse the chunk at some point. 274 POOLED = 1u << 14, 275 276 // |COMPACTION_WAS_ABORTED|: Indicates that the compaction in this page 277 // has been aborted and needs special handling by the sweeper. 278 COMPACTION_WAS_ABORTED = 1u << 15, 279 280 // |COMPACTION_WAS_ABORTED_FOR_TESTING|: During stress testing evacuation 281 // on pages is sometimes aborted. The flag is used to avoid repeatedly 282 // triggering on the same page. 283 COMPACTION_WAS_ABORTED_FOR_TESTING = 1u << 16, 284 285 // |ANCHOR|: Flag is set if page is an anchor. 286 ANCHOR = 1u << 17, 287 }; 288 typedef base::Flags<Flag, uintptr_t> Flags; 289 290 static const int kPointersToHereAreInterestingMask = 291 POINTERS_TO_HERE_ARE_INTERESTING; 292 293 static const int kPointersFromHereAreInterestingMask = 294 POINTERS_FROM_HERE_ARE_INTERESTING; 295 296 static const int kEvacuationCandidateMask = EVACUATION_CANDIDATE; 297 298 static const int kIsInNewSpaceMask = IN_FROM_SPACE | IN_TO_SPACE; 299 300 static const int kSkipEvacuationSlotsRecordingMask = 301 kEvacuationCandidateMask | kIsInNewSpaceMask; 302 303 // |kSweepingDone|: The page state when sweeping is complete or sweeping must 304 // not be performed on that page. Sweeper threads that are done with their 305 // work will set this value and not touch the page anymore. 306 // |kSweepingPending|: This page is ready for parallel sweeping. 307 // |kSweepingInProgress|: This page is currently swept by a sweeper thread. 308 enum ConcurrentSweepingState { 309 kSweepingDone, 310 kSweepingPending, 311 kSweepingInProgress, 312 }; 313 314 static const intptr_t kAlignment = 315 (static_cast<uintptr_t>(1) << kPageSizeBits); 316 317 static const intptr_t kAlignmentMask = kAlignment - 1; 318 319 static const intptr_t kSizeOffset = 0; 320 static const intptr_t kFlagsOffset = kSizeOffset + kSizetSize; 321 static const intptr_t kAreaStartOffset = kFlagsOffset + kIntptrSize; 322 static const intptr_t kAreaEndOffset = kAreaStartOffset + kPointerSize; 323 static const intptr_t kReservationOffset = kAreaEndOffset + kPointerSize; 324 static const intptr_t kOwnerOffset = kReservationOffset + 2 * kPointerSize; 325 326 static const size_t kMinHeaderSize = 327 kSizeOffset + kSizetSize // size_t size 328 + kIntptrSize // Flags flags_ 329 + kPointerSize // Address area_start_ 330 + kPointerSize // Address area_end_ 331 + 2 * kPointerSize // base::VirtualMemory reservation_ 332 + kPointerSize // Address owner_ 333 + kPointerSize // Heap* heap_ 334 + kIntSize // int progress_bar_ 335 + kIntSize // int live_bytes_count_ 336 + kPointerSize // SlotSet* old_to_new_slots_ 337 + kPointerSize // SlotSet* old_to_old_slots_ 338 + kPointerSize // TypedSlotSet* typed_old_to_new_slots_ 339 + kPointerSize // TypedSlotSet* typed_old_to_old_slots_ 340 + kPointerSize // SkipList* skip_list_ 341 + kPointerSize // AtomicValue high_water_mark_ 342 + kPointerSize // base::Mutex* mutex_ 343 + kPointerSize // base::AtomicWord concurrent_sweeping_ 344 + 2 * kSizetSize // AtomicNumber free-list statistics 345 + kPointerSize // AtomicValue next_chunk_ 346 + kPointerSize // AtomicValue prev_chunk_ 347 // FreeListCategory categories_[kNumberOfCategories] 348 + FreeListCategory::kSize * kNumberOfCategories + 349 kPointerSize; // LocalArrayBufferTracker* local_tracker_ 350 351 // We add some more space to the computed header size to amount for missing 352 // alignment requirements in our computation. 353 // Try to get kHeaderSize properly aligned on 32-bit and 64-bit machines. 354 static const size_t kHeaderSize = kMinHeaderSize; 355 356 static const int kBodyOffset = 357 CODE_POINTER_ALIGN(kHeaderSize + Bitmap::kSize); 358 359 // The start offset of the object area in a page. Aligned to both maps and 360 // code alignment to be suitable for both. Also aligned to 32 words because 361 // the marking bitmap is arranged in 32 bit chunks. 362 static const int kObjectStartAlignment = 32 * kPointerSize; 363 static const int kObjectStartOffset = 364 kBodyOffset - 1 + 365 (kObjectStartAlignment - (kBodyOffset - 1) % kObjectStartAlignment); 366 367 // Page size in bytes. This must be a multiple of the OS page size. 368 static const int kPageSize = 1 << kPageSizeBits; 369 static const intptr_t kPageAlignmentMask = (1 << kPageSizeBits) - 1; 370 371 static const int kAllocatableMemory = kPageSize - kObjectStartOffset; 372 373 static inline void IncrementLiveBytes(HeapObject* object, int by); 374 375 // Only works if the pointer is in the first kPageSize of the MemoryChunk. 376 static MemoryChunk* FromAddress(Address a) { 377 return reinterpret_cast<MemoryChunk*>(OffsetFrom(a) & ~kAlignmentMask); 378 } 379 380 static inline MemoryChunk* FromAnyPointerAddress(Heap* heap, Address addr); 381 382 static inline void UpdateHighWaterMark(Address mark) { 383 if (mark == nullptr) return; 384 // Need to subtract one from the mark because when a chunk is full the 385 // top points to the next address after the chunk, which effectively belongs 386 // to another chunk. See the comment to Page::FromTopOrLimit. 387 MemoryChunk* chunk = MemoryChunk::FromAddress(mark - 1); 388 intptr_t new_mark = static_cast<intptr_t>(mark - chunk->address()); 389 intptr_t old_mark = 0; 390 do { 391 old_mark = chunk->high_water_mark_.Value(); 392 } while ((new_mark > old_mark) && 393 !chunk->high_water_mark_.TrySetValue(old_mark, new_mark)); 394 } 395 396 static bool IsValid(MemoryChunk* chunk) { return chunk != nullptr; } 397 398 Address address() { return reinterpret_cast<Address>(this); } 399 400 base::Mutex* mutex() { return mutex_; } 401 402 bool Contains(Address addr) { 403 return addr >= area_start() && addr < area_end(); 404 } 405 406 // Checks whether |addr| can be a limit of addresses in this page. It's a 407 // limit if it's in the page, or if it's just after the last byte of the page. 408 bool ContainsLimit(Address addr) { 409 return addr >= area_start() && addr <= area_end(); 410 } 411 412 base::AtomicValue<ConcurrentSweepingState>& concurrent_sweeping_state() { 413 return concurrent_sweeping_; 414 } 415 416 bool SweepingDone() { 417 return concurrent_sweeping_state().Value() == kSweepingDone; 418 } 419 420 // Manage live byte count, i.e., count of bytes in black objects. 421 inline void ResetLiveBytes(); 422 inline void IncrementLiveBytes(int by); 423 424 int LiveBytes() { 425 DCHECK_LE(static_cast<unsigned>(live_byte_count_), size_); 426 return live_byte_count_; 427 } 428 429 void SetLiveBytes(int live_bytes) { 430 DCHECK_GE(live_bytes, 0); 431 DCHECK_LE(static_cast<size_t>(live_bytes), size_); 432 live_byte_count_ = live_bytes; 433 } 434 435 size_t size() const { return size_; } 436 void set_size(size_t size) { size_ = size; } 437 438 inline Heap* heap() const { return heap_; } 439 440 inline SkipList* skip_list() { return skip_list_; } 441 442 inline void set_skip_list(SkipList* skip_list) { skip_list_ = skip_list; } 443 444 inline SlotSet* old_to_new_slots() { return old_to_new_slots_.Value(); } 445 inline SlotSet* old_to_old_slots() { return old_to_old_slots_; } 446 inline TypedSlotSet* typed_old_to_new_slots() { 447 return typed_old_to_new_slots_.Value(); 448 } 449 inline TypedSlotSet* typed_old_to_old_slots() { 450 return typed_old_to_old_slots_; 451 } 452 inline LocalArrayBufferTracker* local_tracker() { return local_tracker_; } 453 454 V8_EXPORT_PRIVATE void AllocateOldToNewSlots(); 455 void ReleaseOldToNewSlots(); 456 V8_EXPORT_PRIVATE void AllocateOldToOldSlots(); 457 void ReleaseOldToOldSlots(); 458 void AllocateTypedOldToNewSlots(); 459 void ReleaseTypedOldToNewSlots(); 460 void AllocateTypedOldToOldSlots(); 461 void ReleaseTypedOldToOldSlots(); 462 void AllocateLocalTracker(); 463 void ReleaseLocalTracker(); 464 465 Address area_start() { return area_start_; } 466 Address area_end() { return area_end_; } 467 size_t area_size() { return static_cast<size_t>(area_end() - area_start()); } 468 469 bool CommitArea(size_t requested); 470 471 // Approximate amount of physical memory committed for this chunk. 472 size_t CommittedPhysicalMemory(); 473 474 Address HighWaterMark() { return address() + high_water_mark_.Value(); } 475 476 int progress_bar() { 477 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 478 return progress_bar_; 479 } 480 481 void set_progress_bar(int progress_bar) { 482 DCHECK(IsFlagSet(HAS_PROGRESS_BAR)); 483 progress_bar_ = progress_bar; 484 } 485 486 void ResetProgressBar() { 487 if (IsFlagSet(MemoryChunk::HAS_PROGRESS_BAR)) { 488 set_progress_bar(0); 489 } 490 } 491 492 inline Bitmap* markbits() { 493 return Bitmap::FromAddress(address() + kHeaderSize); 494 } 495 496 inline uint32_t AddressToMarkbitIndex(Address addr) { 497 return static_cast<uint32_t>(addr - this->address()) >> kPointerSizeLog2; 498 } 499 500 inline Address MarkbitIndexToAddress(uint32_t index) { 501 return this->address() + (index << kPointerSizeLog2); 502 } 503 504 void ClearLiveness(); 505 506 void PrintMarkbits() { markbits()->Print(); } 507 508 void SetFlag(Flag flag) { flags_ |= flag; } 509 void ClearFlag(Flag flag) { flags_ &= ~Flags(flag); } 510 bool IsFlagSet(Flag flag) { return flags_ & flag; } 511 512 // Set or clear multiple flags at a time. The flags in the mask are set to 513 // the value in "flags", the rest retain the current value in |flags_|. 514 void SetFlags(uintptr_t flags, uintptr_t mask) { 515 flags_ = (flags_ & ~Flags(mask)) | (Flags(flags) & Flags(mask)); 516 } 517 518 // Return all current flags. 519 uintptr_t GetFlags() { return flags_; } 520 521 bool NeverEvacuate() { return IsFlagSet(NEVER_EVACUATE); } 522 523 void MarkNeverEvacuate() { SetFlag(NEVER_EVACUATE); } 524 525 bool IsEvacuationCandidate() { 526 DCHECK(!(IsFlagSet(NEVER_EVACUATE) && IsFlagSet(EVACUATION_CANDIDATE))); 527 return IsFlagSet(EVACUATION_CANDIDATE); 528 } 529 530 bool CanAllocate() { 531 return !IsEvacuationCandidate() && !IsFlagSet(NEVER_ALLOCATE_ON_PAGE); 532 } 533 534 bool ShouldSkipEvacuationSlotRecording() { 535 return ((flags_ & kSkipEvacuationSlotsRecordingMask) != 0) && 536 !IsFlagSet(COMPACTION_WAS_ABORTED); 537 } 538 539 Executability executable() { 540 return IsFlagSet(IS_EXECUTABLE) ? EXECUTABLE : NOT_EXECUTABLE; 541 } 542 543 bool InNewSpace() { return (flags_ & kIsInNewSpaceMask) != 0; } 544 545 bool InToSpace() { return IsFlagSet(IN_TO_SPACE); } 546 547 bool InFromSpace() { return IsFlagSet(IN_FROM_SPACE); } 548 549 MemoryChunk* next_chunk() { return next_chunk_.Value(); } 550 551 MemoryChunk* prev_chunk() { return prev_chunk_.Value(); } 552 553 void set_next_chunk(MemoryChunk* next) { next_chunk_.SetValue(next); } 554 555 void set_prev_chunk(MemoryChunk* prev) { prev_chunk_.SetValue(prev); } 556 557 Space* owner() const { 558 intptr_t owner_value = base::NoBarrierAtomicValue<intptr_t>::FromAddress( 559 const_cast<Address*>(&owner_)) 560 ->Value(); 561 if ((owner_value & kPageHeaderTagMask) == kPageHeaderTag) { 562 return reinterpret_cast<Space*>(owner_value - kPageHeaderTag); 563 } else { 564 return nullptr; 565 } 566 } 567 568 void set_owner(Space* space) { 569 DCHECK((reinterpret_cast<intptr_t>(space) & kPageHeaderTagMask) == 0); 570 owner_ = reinterpret_cast<Address>(space) + kPageHeaderTag; 571 DCHECK((reinterpret_cast<intptr_t>(owner_) & kPageHeaderTagMask) == 572 kPageHeaderTag); 573 } 574 575 bool HasPageHeader() { return owner() != nullptr; } 576 577 void InsertAfter(MemoryChunk* other); 578 void Unlink(); 579 580 protected: 581 static MemoryChunk* Initialize(Heap* heap, Address base, size_t size, 582 Address area_start, Address area_end, 583 Executability executable, Space* owner, 584 base::VirtualMemory* reservation); 585 586 // Should be called when memory chunk is about to be freed. 587 void ReleaseAllocatedMemory(); 588 589 base::VirtualMemory* reserved_memory() { return &reservation_; } 590 591 size_t size_; 592 Flags flags_; 593 594 // Start and end of allocatable memory on this chunk. 595 Address area_start_; 596 Address area_end_; 597 598 // If the chunk needs to remember its memory reservation, it is stored here. 599 base::VirtualMemory reservation_; 600 601 // The identity of the owning space. This is tagged as a failure pointer, but 602 // no failure can be in an object, so this can be distinguished from any entry 603 // in a fixed array. 604 Address owner_; 605 606 Heap* heap_; 607 608 // Used by the incremental marker to keep track of the scanning progress in 609 // large objects that have a progress bar and are scanned in increments. 610 int progress_bar_; 611 612 // Count of bytes marked black on page. 613 int live_byte_count_; 614 615 // A single slot set for small pages (of size kPageSize) or an array of slot 616 // set for large pages. In the latter case the number of entries in the array 617 // is ceil(size() / kPageSize). 618 base::AtomicValue<SlotSet*> old_to_new_slots_; 619 SlotSet* old_to_old_slots_; 620 base::AtomicValue<TypedSlotSet*> typed_old_to_new_slots_; 621 TypedSlotSet* typed_old_to_old_slots_; 622 623 SkipList* skip_list_; 624 625 // Assuming the initial allocation on a page is sequential, 626 // count highest number of bytes ever allocated on the page. 627 base::AtomicValue<intptr_t> high_water_mark_; 628 629 base::Mutex* mutex_; 630 631 base::AtomicValue<ConcurrentSweepingState> concurrent_sweeping_; 632 633 // PagedSpace free-list statistics. 634 base::AtomicNumber<intptr_t> available_in_free_list_; 635 base::AtomicNumber<intptr_t> wasted_memory_; 636 637 // next_chunk_ holds a pointer of type MemoryChunk 638 base::AtomicValue<MemoryChunk*> next_chunk_; 639 // prev_chunk_ holds a pointer of type MemoryChunk 640 base::AtomicValue<MemoryChunk*> prev_chunk_; 641 642 FreeListCategory categories_[kNumberOfCategories]; 643 644 LocalArrayBufferTracker* local_tracker_; 645 646 private: 647 void InitializeReservedMemory() { reservation_.Reset(); } 648 649 friend class MemoryAllocator; 650 friend class MemoryChunkValidator; 651 }; 652 653 DEFINE_OPERATORS_FOR_FLAGS(MemoryChunk::Flags) 654 655 static_assert(kMaxRegularHeapObjectSize <= MemoryChunk::kAllocatableMemory, 656 "kMaxRegularHeapObjectSize <= MemoryChunk::kAllocatableMemory"); 657 658 // ----------------------------------------------------------------------------- 659 // A page is a memory chunk of a size 1MB. Large object pages may be larger. 660 // 661 // The only way to get a page pointer is by calling factory methods: 662 // Page* p = Page::FromAddress(addr); or 663 // Page* p = Page::FromTopOrLimit(top); 664 class Page : public MemoryChunk { 665 public: 666 static const intptr_t kCopyAllFlags = ~0; 667 668 // Page flags copied from from-space to to-space when flipping semispaces. 669 static const intptr_t kCopyOnFlipFlagsMask = 670 static_cast<intptr_t>(MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING) | 671 static_cast<intptr_t>(MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING); 672 673 static inline Page* ConvertNewToOld(Page* old_page); 674 675 // Returns the page containing a given address. The address ranges 676 // from [page_addr .. page_addr + kPageSize[. This only works if the object 677 // is in fact in a page. 678 static Page* FromAddress(Address addr) { 679 return reinterpret_cast<Page*>(OffsetFrom(addr) & ~kPageAlignmentMask); 680 } 681 682 // Returns the page containing the address provided. The address can 683 // potentially point righter after the page. To be also safe for tagged values 684 // we subtract a hole word. The valid address ranges from 685 // [page_addr + kObjectStartOffset .. page_addr + kPageSize + kPointerSize]. 686 static Page* FromAllocationAreaAddress(Address address) { 687 return Page::FromAddress(address - kPointerSize); 688 } 689 690 // Checks if address1 and address2 are on the same new space page. 691 static bool OnSamePage(Address address1, Address address2) { 692 return Page::FromAddress(address1) == Page::FromAddress(address2); 693 } 694 695 // Checks whether an address is page aligned. 696 static bool IsAlignedToPageSize(Address addr) { 697 return (OffsetFrom(addr) & kPageAlignmentMask) == 0; 698 } 699 700 static bool IsAtObjectStart(Address addr) { 701 return (reinterpret_cast<intptr_t>(addr) & kPageAlignmentMask) == 702 kObjectStartOffset; 703 } 704 705 inline static Page* FromAnyPointerAddress(Heap* heap, Address addr); 706 707 // Create a Page object that is only used as anchor for the doubly-linked 708 // list of real pages. 709 explicit Page(Space* owner) { InitializeAsAnchor(owner); } 710 711 inline void MarkNeverAllocateForTesting(); 712 inline void MarkEvacuationCandidate(); 713 inline void ClearEvacuationCandidate(); 714 715 Page* next_page() { return static_cast<Page*>(next_chunk()); } 716 Page* prev_page() { return static_cast<Page*>(prev_chunk()); } 717 void set_next_page(Page* page) { set_next_chunk(page); } 718 void set_prev_page(Page* page) { set_prev_chunk(page); } 719 720 template <typename Callback> 721 inline void ForAllFreeListCategories(Callback callback) { 722 for (int i = kFirstCategory; i < kNumberOfCategories; i++) { 723 callback(&categories_[i]); 724 } 725 } 726 727 // Returns the offset of a given address to this page. 728 inline size_t Offset(Address a) { return static_cast<size_t>(a - address()); } 729 730 // Returns the address for a given offset to the this page. 731 Address OffsetToAddress(size_t offset) { 732 DCHECK_PAGE_OFFSET(offset); 733 return address() + offset; 734 } 735 736 // WaitUntilSweepingCompleted only works when concurrent sweeping is in 737 // progress. In particular, when we know that right before this call a 738 // sweeper thread was sweeping this page. 739 void WaitUntilSweepingCompleted() { 740 mutex_->Lock(); 741 mutex_->Unlock(); 742 DCHECK(SweepingDone()); 743 } 744 745 void ResetFreeListStatistics(); 746 747 size_t AvailableInFreeList(); 748 749 size_t LiveBytesFromFreeList() { 750 DCHECK_GE(area_size(), wasted_memory() + available_in_free_list()); 751 return area_size() - wasted_memory() - available_in_free_list(); 752 } 753 754 FreeListCategory* free_list_category(FreeListCategoryType type) { 755 return &categories_[type]; 756 } 757 758 bool is_anchor() { return IsFlagSet(Page::ANCHOR); } 759 760 size_t wasted_memory() { return wasted_memory_.Value(); } 761 void add_wasted_memory(size_t waste) { wasted_memory_.Increment(waste); } 762 size_t available_in_free_list() { return available_in_free_list_.Value(); } 763 void add_available_in_free_list(size_t available) { 764 DCHECK_LE(available, area_size()); 765 available_in_free_list_.Increment(available); 766 } 767 void remove_available_in_free_list(size_t available) { 768 DCHECK_LE(available, area_size()); 769 DCHECK_GE(available_in_free_list(), available); 770 available_in_free_list_.Decrement(available); 771 } 772 773 size_t ShrinkToHighWaterMark(); 774 775 void CreateBlackArea(Address start, Address end); 776 777 #ifdef DEBUG 778 void Print(); 779 #endif // DEBUG 780 781 private: 782 enum InitializationMode { kFreeMemory, kDoNotFreeMemory }; 783 784 template <InitializationMode mode = kFreeMemory> 785 static inline Page* Initialize(Heap* heap, MemoryChunk* chunk, 786 Executability executable, PagedSpace* owner); 787 static inline Page* Initialize(Heap* heap, MemoryChunk* chunk, 788 Executability executable, SemiSpace* owner); 789 790 inline void InitializeFreeListCategories(); 791 792 void InitializeAsAnchor(Space* owner); 793 794 friend class MemoryAllocator; 795 }; 796 797 class LargePage : public MemoryChunk { 798 public: 799 HeapObject* GetObject() { return HeapObject::FromAddress(area_start()); } 800 801 inline LargePage* next_page() { 802 return static_cast<LargePage*>(next_chunk()); 803 } 804 805 inline void set_next_page(LargePage* page) { set_next_chunk(page); } 806 807 // Uncommit memory that is not in use anymore by the object. If the object 808 // cannot be shrunk 0 is returned. 809 Address GetAddressToShrink(); 810 811 void ClearOutOfLiveRangeSlots(Address free_start); 812 813 // A limit to guarantee that we do not overflow typed slot offset in 814 // the old to old remembered set. 815 // Note that this limit is higher than what assembler already imposes on 816 // x64 and ia32 architectures. 817 static const int kMaxCodePageSize = 512 * MB; 818 819 private: 820 static inline LargePage* Initialize(Heap* heap, MemoryChunk* chunk, 821 Executability executable, Space* owner); 822 823 friend class MemoryAllocator; 824 }; 825 826 827 // ---------------------------------------------------------------------------- 828 // Space is the abstract superclass for all allocation spaces. 829 class Space : public Malloced { 830 public: 831 Space(Heap* heap, AllocationSpace id, Executability executable) 832 : allocation_observers_(new List<AllocationObserver*>()), 833 allocation_observers_paused_(false), 834 heap_(heap), 835 id_(id), 836 executable_(executable), 837 committed_(0), 838 max_committed_(0) {} 839 840 virtual ~Space() {} 841 842 Heap* heap() const { return heap_; } 843 844 // Does the space need executable memory? 845 Executability executable() { return executable_; } 846 847 // Identity used in error reporting. 848 AllocationSpace identity() { return id_; } 849 850 virtual void AddAllocationObserver(AllocationObserver* observer) { 851 allocation_observers_->Add(observer); 852 } 853 854 virtual void RemoveAllocationObserver(AllocationObserver* observer) { 855 bool removed = allocation_observers_->RemoveElement(observer); 856 USE(removed); 857 DCHECK(removed); 858 } 859 860 virtual void PauseAllocationObservers() { 861 allocation_observers_paused_ = true; 862 } 863 864 virtual void ResumeAllocationObservers() { 865 allocation_observers_paused_ = false; 866 } 867 868 void AllocationStep(Address soon_object, int size); 869 870 // Return the total amount committed memory for this space, i.e., allocatable 871 // memory and page headers. 872 virtual size_t CommittedMemory() { return committed_; } 873 874 virtual size_t MaximumCommittedMemory() { return max_committed_; } 875 876 // Returns allocated size. 877 virtual size_t Size() = 0; 878 879 // Returns size of objects. Can differ from the allocated size 880 // (e.g. see LargeObjectSpace). 881 virtual size_t SizeOfObjects() { return Size(); } 882 883 // Approximate amount of physical memory committed for this space. 884 virtual size_t CommittedPhysicalMemory() = 0; 885 886 // Return the available bytes without growing. 887 virtual size_t Available() = 0; 888 889 virtual int RoundSizeDownToObjectAlignment(int size) { 890 if (id_ == CODE_SPACE) { 891 return RoundDown(size, kCodeAlignment); 892 } else { 893 return RoundDown(size, kPointerSize); 894 } 895 } 896 897 virtual std::unique_ptr<ObjectIterator> GetObjectIterator() = 0; 898 899 void AccountCommitted(size_t bytes) { 900 DCHECK_GE(committed_ + bytes, committed_); 901 committed_ += bytes; 902 if (committed_ > max_committed_) { 903 max_committed_ = committed_; 904 } 905 } 906 907 void AccountUncommitted(size_t bytes) { 908 DCHECK_GE(committed_, committed_ - bytes); 909 committed_ -= bytes; 910 } 911 912 #ifdef DEBUG 913 virtual void Print() = 0; 914 #endif 915 916 protected: 917 std::unique_ptr<List<AllocationObserver*>> allocation_observers_; 918 bool allocation_observers_paused_; 919 920 private: 921 Heap* heap_; 922 AllocationSpace id_; 923 Executability executable_; 924 925 // Keeps track of committed memory in a space. 926 size_t committed_; 927 size_t max_committed_; 928 929 DISALLOW_COPY_AND_ASSIGN(Space); 930 }; 931 932 933 class MemoryChunkValidator { 934 // Computed offsets should match the compiler generated ones. 935 STATIC_ASSERT(MemoryChunk::kSizeOffset == offsetof(MemoryChunk, size_)); 936 937 // Validate our estimates on the header size. 938 STATIC_ASSERT(sizeof(MemoryChunk) <= MemoryChunk::kHeaderSize); 939 STATIC_ASSERT(sizeof(LargePage) <= MemoryChunk::kHeaderSize); 940 STATIC_ASSERT(sizeof(Page) <= MemoryChunk::kHeaderSize); 941 }; 942 943 944 // ---------------------------------------------------------------------------- 945 // All heap objects containing executable code (code objects) must be allocated 946 // from a 2 GB range of memory, so that they can call each other using 32-bit 947 // displacements. This happens automatically on 32-bit platforms, where 32-bit 948 // displacements cover the entire 4GB virtual address space. On 64-bit 949 // platforms, we support this using the CodeRange object, which reserves and 950 // manages a range of virtual memory. 951 class CodeRange { 952 public: 953 explicit CodeRange(Isolate* isolate); 954 ~CodeRange() { TearDown(); } 955 956 // Reserves a range of virtual memory, but does not commit any of it. 957 // Can only be called once, at heap initialization time. 958 // Returns false on failure. 959 bool SetUp(size_t requested_size); 960 961 bool valid() { return code_range_ != NULL; } 962 Address start() { 963 DCHECK(valid()); 964 return static_cast<Address>(code_range_->address()); 965 } 966 size_t size() { 967 DCHECK(valid()); 968 return code_range_->size(); 969 } 970 bool contains(Address address) { 971 if (!valid()) return false; 972 Address start = static_cast<Address>(code_range_->address()); 973 return start <= address && address < start + code_range_->size(); 974 } 975 976 // Allocates a chunk of memory from the large-object portion of 977 // the code range. On platforms with no separate code range, should 978 // not be called. 979 MUST_USE_RESULT Address AllocateRawMemory(const size_t requested_size, 980 const size_t commit_size, 981 size_t* allocated); 982 bool CommitRawMemory(Address start, size_t length); 983 bool UncommitRawMemory(Address start, size_t length); 984 void FreeRawMemory(Address buf, size_t length); 985 986 private: 987 class FreeBlock { 988 public: 989 FreeBlock() : start(0), size(0) {} 990 FreeBlock(Address start_arg, size_t size_arg) 991 : start(start_arg), size(size_arg) { 992 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 993 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 994 } 995 FreeBlock(void* start_arg, size_t size_arg) 996 : start(static_cast<Address>(start_arg)), size(size_arg) { 997 DCHECK(IsAddressAligned(start, MemoryChunk::kAlignment)); 998 DCHECK(size >= static_cast<size_t>(Page::kPageSize)); 999 } 1000 1001 Address start; 1002 size_t size; 1003 }; 1004 1005 // Frees the range of virtual memory, and frees the data structures used to 1006 // manage it. 1007 void TearDown(); 1008 1009 // Finds a block on the allocation list that contains at least the 1010 // requested amount of memory. If none is found, sorts and merges 1011 // the existing free memory blocks, and searches again. 1012 // If none can be found, returns false. 1013 bool GetNextAllocationBlock(size_t requested); 1014 // Compares the start addresses of two free blocks. 1015 static int CompareFreeBlockAddress(const FreeBlock* left, 1016 const FreeBlock* right); 1017 bool ReserveBlock(const size_t requested_size, FreeBlock* block); 1018 void ReleaseBlock(const FreeBlock* block); 1019 1020 Isolate* isolate_; 1021 1022 // The reserved range of virtual memory that all code objects are put in. 1023 base::VirtualMemory* code_range_; 1024 1025 // The global mutex guards free_list_ and allocation_list_ as GC threads may 1026 // access both lists concurrently to the main thread. 1027 base::Mutex code_range_mutex_; 1028 1029 // Freed blocks of memory are added to the free list. When the allocation 1030 // list is exhausted, the free list is sorted and merged to make the new 1031 // allocation list. 1032 List<FreeBlock> free_list_; 1033 1034 // Memory is allocated from the free blocks on the allocation list. 1035 // The block at current_allocation_block_index_ is the current block. 1036 List<FreeBlock> allocation_list_; 1037 int current_allocation_block_index_; 1038 1039 DISALLOW_COPY_AND_ASSIGN(CodeRange); 1040 }; 1041 1042 1043 class SkipList { 1044 public: 1045 SkipList() { Clear(); } 1046 1047 void Clear() { 1048 for (int idx = 0; idx < kSize; idx++) { 1049 starts_[idx] = reinterpret_cast<Address>(-1); 1050 } 1051 } 1052 1053 Address StartFor(Address addr) { return starts_[RegionNumber(addr)]; } 1054 1055 void AddObject(Address addr, int size) { 1056 int start_region = RegionNumber(addr); 1057 int end_region = RegionNumber(addr + size - kPointerSize); 1058 for (int idx = start_region; idx <= end_region; idx++) { 1059 if (starts_[idx] > addr) { 1060 starts_[idx] = addr; 1061 } else { 1062 // In the first region, there may already be an object closer to the 1063 // start of the region. Do not change the start in that case. If this 1064 // is not the first region, you probably added overlapping objects. 1065 DCHECK_EQ(start_region, idx); 1066 } 1067 } 1068 } 1069 1070 static inline int RegionNumber(Address addr) { 1071 return (OffsetFrom(addr) & Page::kPageAlignmentMask) >> kRegionSizeLog2; 1072 } 1073 1074 static void Update(Address addr, int size) { 1075 Page* page = Page::FromAddress(addr); 1076 SkipList* list = page->skip_list(); 1077 if (list == NULL) { 1078 list = new SkipList(); 1079 page->set_skip_list(list); 1080 } 1081 1082 list->AddObject(addr, size); 1083 } 1084 1085 private: 1086 static const int kRegionSizeLog2 = 13; 1087 static const int kRegionSize = 1 << kRegionSizeLog2; 1088 static const int kSize = Page::kPageSize / kRegionSize; 1089 1090 STATIC_ASSERT(Page::kPageSize % kRegionSize == 0); 1091 1092 Address starts_[kSize]; 1093 }; 1094 1095 1096 // ---------------------------------------------------------------------------- 1097 // A space acquires chunks of memory from the operating system. The memory 1098 // allocator allocates and deallocates pages for the paged heap spaces and large 1099 // pages for large object space. 1100 class V8_EXPORT_PRIVATE MemoryAllocator { 1101 public: 1102 // Unmapper takes care of concurrently unmapping and uncommitting memory 1103 // chunks. 1104 class Unmapper { 1105 public: 1106 class UnmapFreeMemoryTask; 1107 1108 explicit Unmapper(MemoryAllocator* allocator) 1109 : allocator_(allocator), 1110 pending_unmapping_tasks_semaphore_(0), 1111 concurrent_unmapping_tasks_active_(0) { 1112 chunks_[kRegular].reserve(kReservedQueueingSlots); 1113 chunks_[kPooled].reserve(kReservedQueueingSlots); 1114 } 1115 1116 void AddMemoryChunkSafe(MemoryChunk* chunk) { 1117 if ((chunk->size() == Page::kPageSize) && 1118 (chunk->executable() != EXECUTABLE)) { 1119 AddMemoryChunkSafe<kRegular>(chunk); 1120 } else { 1121 AddMemoryChunkSafe<kNonRegular>(chunk); 1122 } 1123 } 1124 1125 MemoryChunk* TryGetPooledMemoryChunkSafe() { 1126 // Procedure: 1127 // (1) Try to get a chunk that was declared as pooled and already has 1128 // been uncommitted. 1129 // (2) Try to steal any memory chunk of kPageSize that would've been 1130 // unmapped. 1131 MemoryChunk* chunk = GetMemoryChunkSafe<kPooled>(); 1132 if (chunk == nullptr) { 1133 chunk = GetMemoryChunkSafe<kRegular>(); 1134 if (chunk != nullptr) { 1135 // For stolen chunks we need to manually free any allocated memory. 1136 chunk->ReleaseAllocatedMemory(); 1137 } 1138 } 1139 return chunk; 1140 } 1141 1142 void FreeQueuedChunks(); 1143 bool WaitUntilCompleted(); 1144 void TearDown(); 1145 1146 private: 1147 static const int kReservedQueueingSlots = 64; 1148 1149 enum ChunkQueueType { 1150 kRegular, // Pages of kPageSize that do not live in a CodeRange and 1151 // can thus be used for stealing. 1152 kNonRegular, // Large chunks and executable chunks. 1153 kPooled, // Pooled chunks, already uncommited and ready for reuse. 1154 kNumberOfChunkQueues, 1155 }; 1156 1157 enum class FreeMode { 1158 kUncommitPooled, 1159 kReleasePooled, 1160 }; 1161 1162 template <ChunkQueueType type> 1163 void AddMemoryChunkSafe(MemoryChunk* chunk) { 1164 base::LockGuard<base::Mutex> guard(&mutex_); 1165 if (type != kRegular || allocator_->CanFreeMemoryChunk(chunk)) { 1166 chunks_[type].push_back(chunk); 1167 } else { 1168 DCHECK_EQ(type, kRegular); 1169 delayed_regular_chunks_.push_back(chunk); 1170 } 1171 } 1172 1173 template <ChunkQueueType type> 1174 MemoryChunk* GetMemoryChunkSafe() { 1175 base::LockGuard<base::Mutex> guard(&mutex_); 1176 if (chunks_[type].empty()) return nullptr; 1177 MemoryChunk* chunk = chunks_[type].back(); 1178 chunks_[type].pop_back(); 1179 return chunk; 1180 } 1181 1182 void ReconsiderDelayedChunks(); 1183 template <FreeMode mode> 1184 void PerformFreeMemoryOnQueuedChunks(); 1185 1186 base::Mutex mutex_; 1187 MemoryAllocator* allocator_; 1188 std::vector<MemoryChunk*> chunks_[kNumberOfChunkQueues]; 1189 // Delayed chunks cannot be processed in the current unmapping cycle because 1190 // of dependencies such as an active sweeper. 1191 // See MemoryAllocator::CanFreeMemoryChunk. 1192 std::list<MemoryChunk*> delayed_regular_chunks_; 1193 base::Semaphore pending_unmapping_tasks_semaphore_; 1194 intptr_t concurrent_unmapping_tasks_active_; 1195 1196 friend class MemoryAllocator; 1197 }; 1198 1199 enum AllocationMode { 1200 kRegular, 1201 kPooled, 1202 }; 1203 1204 enum FreeMode { 1205 kFull, 1206 kAlreadyPooled, 1207 kPreFreeAndQueue, 1208 kPooledAndQueue, 1209 }; 1210 1211 static size_t CodePageGuardStartOffset(); 1212 1213 static size_t CodePageGuardSize(); 1214 1215 static size_t CodePageAreaStartOffset(); 1216 1217 static size_t CodePageAreaEndOffset(); 1218 1219 static size_t CodePageAreaSize() { 1220 return CodePageAreaEndOffset() - CodePageAreaStartOffset(); 1221 } 1222 1223 static size_t PageAreaSize(AllocationSpace space) { 1224 DCHECK_NE(LO_SPACE, space); 1225 return (space == CODE_SPACE) ? CodePageAreaSize() 1226 : Page::kAllocatableMemory; 1227 } 1228 1229 static intptr_t GetCommitPageSize(); 1230 1231 explicit MemoryAllocator(Isolate* isolate); 1232 1233 // Initializes its internal bookkeeping structures. 1234 // Max capacity of the total space and executable memory limit. 1235 bool SetUp(size_t max_capacity, size_t capacity_executable, 1236 size_t code_range_size); 1237 1238 void TearDown(); 1239 1240 // Allocates a Page from the allocator. AllocationMode is used to indicate 1241 // whether pooled allocation, which only works for MemoryChunk::kPageSize, 1242 // should be tried first. 1243 template <MemoryAllocator::AllocationMode alloc_mode = kRegular, 1244 typename SpaceType> 1245 Page* AllocatePage(size_t size, SpaceType* owner, Executability executable); 1246 1247 LargePage* AllocateLargePage(size_t size, LargeObjectSpace* owner, 1248 Executability executable); 1249 1250 template <MemoryAllocator::FreeMode mode = kFull> 1251 void Free(MemoryChunk* chunk); 1252 1253 bool CanFreeMemoryChunk(MemoryChunk* chunk); 1254 1255 // Returns allocated spaces in bytes. 1256 size_t Size() { return size_.Value(); } 1257 1258 // Returns allocated executable spaces in bytes. 1259 size_t SizeExecutable() { return size_executable_.Value(); } 1260 1261 // Returns the maximum available bytes of heaps. 1262 size_t Available() { 1263 const size_t size = Size(); 1264 return capacity_ < size ? 0 : capacity_ - size; 1265 } 1266 1267 // Returns the maximum available executable bytes of heaps. 1268 size_t AvailableExecutable() { 1269 const size_t executable_size = SizeExecutable(); 1270 if (capacity_executable_ < executable_size) return 0; 1271 return capacity_executable_ - executable_size; 1272 } 1273 1274 // Returns maximum available bytes that the old space can have. 1275 size_t MaxAvailable() { 1276 return (Available() / Page::kPageSize) * Page::kAllocatableMemory; 1277 } 1278 1279 // Returns an indication of whether a pointer is in a space that has 1280 // been allocated by this MemoryAllocator. 1281 V8_INLINE bool IsOutsideAllocatedSpace(const void* address) { 1282 return address < lowest_ever_allocated_.Value() || 1283 address >= highest_ever_allocated_.Value(); 1284 } 1285 1286 // Returns a MemoryChunk in which the memory region from commit_area_size to 1287 // reserve_area_size of the chunk area is reserved but not committed, it 1288 // could be committed later by calling MemoryChunk::CommitArea. 1289 MemoryChunk* AllocateChunk(size_t reserve_area_size, size_t commit_area_size, 1290 Executability executable, Space* space); 1291 1292 void ShrinkChunk(MemoryChunk* chunk, size_t bytes_to_shrink); 1293 1294 Address ReserveAlignedMemory(size_t requested, size_t alignment, 1295 base::VirtualMemory* controller); 1296 Address AllocateAlignedMemory(size_t reserve_size, size_t commit_size, 1297 size_t alignment, Executability executable, 1298 base::VirtualMemory* controller); 1299 1300 bool CommitMemory(Address addr, size_t size, Executability executable); 1301 1302 void FreeMemory(base::VirtualMemory* reservation, Executability executable); 1303 void PartialFreeMemory(MemoryChunk* chunk, Address start_free); 1304 void FreeMemory(Address addr, size_t size, Executability executable); 1305 1306 // Commit a contiguous block of memory from the initial chunk. Assumes that 1307 // the address is not NULL, the size is greater than zero, and that the 1308 // block is contained in the initial chunk. Returns true if it succeeded 1309 // and false otherwise. 1310 bool CommitBlock(Address start, size_t size, Executability executable); 1311 1312 // Uncommit a contiguous block of memory [start..(start+size)[. 1313 // start is not NULL, the size is greater than zero, and the 1314 // block is contained in the initial chunk. Returns true if it succeeded 1315 // and false otherwise. 1316 bool UncommitBlock(Address start, size_t size); 1317 1318 // Zaps a contiguous block of memory [start..(start+size)[ thus 1319 // filling it up with a recognizable non-NULL bit pattern. 1320 void ZapBlock(Address start, size_t size); 1321 1322 MUST_USE_RESULT bool CommitExecutableMemory(base::VirtualMemory* vm, 1323 Address start, size_t commit_size, 1324 size_t reserved_size); 1325 1326 CodeRange* code_range() { return code_range_; } 1327 Unmapper* unmapper() { return &unmapper_; } 1328 1329 #ifdef DEBUG 1330 // Reports statistic info of the space. 1331 void ReportStatistics(); 1332 #endif 1333 1334 private: 1335 // PreFree logically frees the object, i.e., it takes care of the size 1336 // bookkeeping and calls the allocation callback. 1337 void PreFreeMemory(MemoryChunk* chunk); 1338 1339 // FreeMemory can be called concurrently when PreFree was executed before. 1340 void PerformFreeMemory(MemoryChunk* chunk); 1341 1342 // See AllocatePage for public interface. Note that currently we only support 1343 // pools for NOT_EXECUTABLE pages of size MemoryChunk::kPageSize. 1344 template <typename SpaceType> 1345 MemoryChunk* AllocatePagePooled(SpaceType* owner); 1346 1347 // Initializes pages in a chunk. Returns the first page address. 1348 // This function and GetChunkId() are provided for the mark-compact 1349 // collector to rebuild page headers in the from space, which is 1350 // used as a marking stack and its page headers are destroyed. 1351 Page* InitializePagesInChunk(int chunk_id, int pages_in_chunk, 1352 PagedSpace* owner); 1353 1354 void UpdateAllocatedSpaceLimits(void* low, void* high) { 1355 // The use of atomic primitives does not guarantee correctness (wrt. 1356 // desired semantics) by default. The loop here ensures that we update the 1357 // values only if they did not change in between. 1358 void* ptr = nullptr; 1359 do { 1360 ptr = lowest_ever_allocated_.Value(); 1361 } while ((low < ptr) && !lowest_ever_allocated_.TrySetValue(ptr, low)); 1362 do { 1363 ptr = highest_ever_allocated_.Value(); 1364 } while ((high > ptr) && !highest_ever_allocated_.TrySetValue(ptr, high)); 1365 } 1366 1367 Isolate* isolate_; 1368 CodeRange* code_range_; 1369 1370 // Maximum space size in bytes. 1371 size_t capacity_; 1372 // Maximum subset of capacity_ that can be executable 1373 size_t capacity_executable_; 1374 1375 // Allocated space size in bytes. 1376 base::AtomicNumber<size_t> size_; 1377 // Allocated executable space size in bytes. 1378 base::AtomicNumber<size_t> size_executable_; 1379 1380 // We keep the lowest and highest addresses allocated as a quick way 1381 // of determining that pointers are outside the heap. The estimate is 1382 // conservative, i.e. not all addresses in 'allocated' space are allocated 1383 // to our heap. The range is [lowest, highest[, inclusive on the low end 1384 // and exclusive on the high end. 1385 base::AtomicValue<void*> lowest_ever_allocated_; 1386 base::AtomicValue<void*> highest_ever_allocated_; 1387 1388 base::VirtualMemory last_chunk_; 1389 Unmapper unmapper_; 1390 1391 friend class TestCodeRangeScope; 1392 1393 DISALLOW_IMPLICIT_CONSTRUCTORS(MemoryAllocator); 1394 }; 1395 1396 extern template Page* 1397 MemoryAllocator::AllocatePage<MemoryAllocator::kRegular, PagedSpace>( 1398 size_t size, PagedSpace* owner, Executability executable); 1399 extern template Page* 1400 MemoryAllocator::AllocatePage<MemoryAllocator::kRegular, SemiSpace>( 1401 size_t size, SemiSpace* owner, Executability executable); 1402 extern template Page* 1403 MemoryAllocator::AllocatePage<MemoryAllocator::kPooled, SemiSpace>( 1404 size_t size, SemiSpace* owner, Executability executable); 1405 1406 // ----------------------------------------------------------------------------- 1407 // Interface for heap object iterator to be implemented by all object space 1408 // object iterators. 1409 // 1410 // NOTE: The space specific object iterators also implements the own next() 1411 // method which is used to avoid using virtual functions 1412 // iterating a specific space. 1413 1414 class V8_EXPORT_PRIVATE ObjectIterator : public Malloced { 1415 public: 1416 virtual ~ObjectIterator() {} 1417 virtual HeapObject* Next() = 0; 1418 }; 1419 1420 template <class PAGE_TYPE> 1421 class PageIteratorImpl 1422 : public std::iterator<std::forward_iterator_tag, PAGE_TYPE> { 1423 public: 1424 explicit PageIteratorImpl(PAGE_TYPE* p) : p_(p) {} 1425 PageIteratorImpl(const PageIteratorImpl<PAGE_TYPE>& other) : p_(other.p_) {} 1426 PAGE_TYPE* operator*() { return p_; } 1427 bool operator==(const PageIteratorImpl<PAGE_TYPE>& rhs) { 1428 return rhs.p_ == p_; 1429 } 1430 bool operator!=(const PageIteratorImpl<PAGE_TYPE>& rhs) { 1431 return rhs.p_ != p_; 1432 } 1433 inline PageIteratorImpl<PAGE_TYPE>& operator++(); 1434 inline PageIteratorImpl<PAGE_TYPE> operator++(int); 1435 1436 private: 1437 PAGE_TYPE* p_; 1438 }; 1439 1440 typedef PageIteratorImpl<Page> PageIterator; 1441 typedef PageIteratorImpl<LargePage> LargePageIterator; 1442 1443 class PageRange { 1444 public: 1445 typedef PageIterator iterator; 1446 PageRange(Page* begin, Page* end) : begin_(begin), end_(end) {} 1447 explicit PageRange(Page* page) : PageRange(page, page->next_page()) {} 1448 inline PageRange(Address start, Address limit); 1449 1450 iterator begin() { return iterator(begin_); } 1451 iterator end() { return iterator(end_); } 1452 1453 private: 1454 Page* begin_; 1455 Page* end_; 1456 }; 1457 1458 // ----------------------------------------------------------------------------- 1459 // Heap object iterator in new/old/map spaces. 1460 // 1461 // A HeapObjectIterator iterates objects from the bottom of the given space 1462 // to its top or from the bottom of the given page to its top. 1463 // 1464 // If objects are allocated in the page during iteration the iterator may 1465 // or may not iterate over those objects. The caller must create a new 1466 // iterator in order to be sure to visit these new objects. 1467 class V8_EXPORT_PRIVATE HeapObjectIterator : public ObjectIterator { 1468 public: 1469 // Creates a new object iterator in a given space. 1470 explicit HeapObjectIterator(PagedSpace* space); 1471 explicit HeapObjectIterator(Page* page); 1472 1473 // Advance to the next object, skipping free spaces and other fillers and 1474 // skipping the special garbage section of which there is one per space. 1475 // Returns nullptr when the iteration has ended. 1476 inline HeapObject* Next() override; 1477 1478 private: 1479 // Fast (inlined) path of next(). 1480 inline HeapObject* FromCurrentPage(); 1481 1482 // Slow path of next(), goes into the next page. Returns false if the 1483 // iteration has ended. 1484 bool AdvanceToNextPage(); 1485 1486 Address cur_addr_; // Current iteration point. 1487 Address cur_end_; // End iteration point. 1488 PagedSpace* space_; 1489 PageRange page_range_; 1490 PageRange::iterator current_page_; 1491 }; 1492 1493 1494 // ----------------------------------------------------------------------------- 1495 // A space has a circular list of pages. The next page can be accessed via 1496 // Page::next_page() call. 1497 1498 // An abstraction of allocation and relocation pointers in a page-structured 1499 // space. 1500 class AllocationInfo { 1501 public: 1502 AllocationInfo() : original_top_(nullptr), top_(nullptr), limit_(nullptr) {} 1503 AllocationInfo(Address top, Address limit) 1504 : original_top_(top), top_(top), limit_(limit) {} 1505 1506 void Reset(Address top, Address limit) { 1507 original_top_ = top; 1508 set_top(top); 1509 set_limit(limit); 1510 } 1511 1512 Address original_top() { 1513 SLOW_DCHECK(top_ == NULL || 1514 (reinterpret_cast<intptr_t>(top_) & kHeapObjectTagMask) == 0); 1515 return original_top_; 1516 } 1517 1518 INLINE(void set_top(Address top)) { 1519 SLOW_DCHECK(top == NULL || 1520 (reinterpret_cast<intptr_t>(top) & kHeapObjectTagMask) == 0); 1521 top_ = top; 1522 } 1523 1524 INLINE(Address top()) const { 1525 SLOW_DCHECK(top_ == NULL || 1526 (reinterpret_cast<intptr_t>(top_) & kHeapObjectTagMask) == 0); 1527 return top_; 1528 } 1529 1530 Address* top_address() { return &top_; } 1531 1532 INLINE(void set_limit(Address limit)) { 1533 limit_ = limit; 1534 } 1535 1536 INLINE(Address limit()) const { 1537 return limit_; 1538 } 1539 1540 Address* limit_address() { return &limit_; } 1541 1542 #ifdef DEBUG 1543 bool VerifyPagedAllocation() { 1544 return (Page::FromAllocationAreaAddress(top_) == 1545 Page::FromAllocationAreaAddress(limit_)) && 1546 (top_ <= limit_); 1547 } 1548 #endif 1549 1550 private: 1551 // The original top address when the allocation info was initialized. 1552 Address original_top_; 1553 // Current allocation top. 1554 Address top_; 1555 // Current allocation limit. 1556 Address limit_; 1557 }; 1558 1559 1560 // An abstraction of the accounting statistics of a page-structured space. 1561 // 1562 // The stats are only set by functions that ensure they stay balanced. These 1563 // functions increase or decrease one of the non-capacity stats in conjunction 1564 // with capacity, or else they always balance increases and decreases to the 1565 // non-capacity stats. 1566 class AllocationStats BASE_EMBEDDED { 1567 public: 1568 AllocationStats() { Clear(); } 1569 1570 // Zero out all the allocation statistics (i.e., no capacity). 1571 void Clear() { 1572 capacity_ = 0; 1573 max_capacity_ = 0; 1574 size_ = 0; 1575 } 1576 1577 void ClearSize() { size_ = capacity_; } 1578 1579 // Accessors for the allocation statistics. 1580 size_t Capacity() { return capacity_; } 1581 size_t MaxCapacity() { return max_capacity_; } 1582 size_t Size() { return size_; } 1583 1584 // Grow the space by adding available bytes. They are initially marked as 1585 // being in use (part of the size), but will normally be immediately freed, 1586 // putting them on the free list and removing them from size_. 1587 void ExpandSpace(size_t bytes) { 1588 DCHECK_GE(size_ + bytes, size_); 1589 DCHECK_GE(capacity_ + bytes, capacity_); 1590 capacity_ += bytes; 1591 size_ += bytes; 1592 if (capacity_ > max_capacity_) { 1593 max_capacity_ = capacity_; 1594 } 1595 } 1596 1597 // Shrink the space by removing available bytes. Since shrinking is done 1598 // during sweeping, bytes have been marked as being in use (part of the size) 1599 // and are hereby freed. 1600 void ShrinkSpace(size_t bytes) { 1601 DCHECK_GE(capacity_, bytes); 1602 DCHECK_GE(size_, bytes); 1603 capacity_ -= bytes; 1604 size_ -= bytes; 1605 } 1606 1607 void AllocateBytes(size_t bytes) { 1608 DCHECK_GE(size_ + bytes, size_); 1609 size_ += bytes; 1610 } 1611 1612 void DeallocateBytes(size_t bytes) { 1613 DCHECK_GE(size_, bytes); 1614 size_ -= bytes; 1615 } 1616 1617 void DecreaseCapacity(size_t bytes) { 1618 DCHECK_GE(capacity_, bytes); 1619 DCHECK_GE(capacity_ - bytes, size_); 1620 capacity_ -= bytes; 1621 } 1622 1623 void IncreaseCapacity(size_t bytes) { 1624 DCHECK_GE(capacity_ + bytes, capacity_); 1625 capacity_ += bytes; 1626 } 1627 1628 // Merge |other| into |this|. 1629 void Merge(const AllocationStats& other) { 1630 DCHECK_GE(capacity_ + other.capacity_, capacity_); 1631 DCHECK_GE(size_ + other.size_, size_); 1632 capacity_ += other.capacity_; 1633 size_ += other.size_; 1634 if (other.max_capacity_ > max_capacity_) { 1635 max_capacity_ = other.max_capacity_; 1636 } 1637 } 1638 1639 private: 1640 // |capacity_|: The number of object-area bytes (i.e., not including page 1641 // bookkeeping structures) currently in the space. 1642 size_t capacity_; 1643 1644 // |max_capacity_|: The maximum capacity ever observed. 1645 size_t max_capacity_; 1646 1647 // |size_|: The number of allocated bytes. 1648 size_t size_; 1649 }; 1650 1651 // A free list maintaining free blocks of memory. The free list is organized in 1652 // a way to encourage objects allocated around the same time to be near each 1653 // other. The normal way to allocate is intended to be by bumping a 'top' 1654 // pointer until it hits a 'limit' pointer. When the limit is hit we need to 1655 // find a new space to allocate from. This is done with the free list, which is 1656 // divided up into rough categories to cut down on waste. Having finer 1657 // categories would scatter allocation more. 1658 1659 // The free list is organized in categories as follows: 1660 // kMinBlockSize-10 words (tiniest): The tiniest blocks are only used for 1661 // allocation, when categories >= small do not have entries anymore. 1662 // 11-31 words (tiny): The tiny blocks are only used for allocation, when 1663 // categories >= small do not have entries anymore. 1664 // 32-255 words (small): Used for allocating free space between 1-31 words in 1665 // size. 1666 // 256-2047 words (medium): Used for allocating free space between 32-255 words 1667 // in size. 1668 // 1048-16383 words (large): Used for allocating free space between 256-2047 1669 // words in size. 1670 // At least 16384 words (huge): This list is for objects of 2048 words or 1671 // larger. Empty pages are also added to this list. 1672 class V8_EXPORT_PRIVATE FreeList { 1673 public: 1674 // This method returns how much memory can be allocated after freeing 1675 // maximum_freed memory. 1676 static inline size_t GuaranteedAllocatable(size_t maximum_freed) { 1677 if (maximum_freed <= kTiniestListMax) { 1678 // Since we are not iterating over all list entries, we cannot guarantee 1679 // that we can find the maximum freed block in that free list. 1680 return 0; 1681 } else if (maximum_freed <= kTinyListMax) { 1682 return kTinyAllocationMax; 1683 } else if (maximum_freed <= kSmallListMax) { 1684 return kSmallAllocationMax; 1685 } else if (maximum_freed <= kMediumListMax) { 1686 return kMediumAllocationMax; 1687 } else if (maximum_freed <= kLargeListMax) { 1688 return kLargeAllocationMax; 1689 } 1690 return maximum_freed; 1691 } 1692 1693 explicit FreeList(PagedSpace* owner); 1694 1695 // Adds a node on the free list. The block of size {size_in_bytes} starting 1696 // at {start} is placed on the free list. The return value is the number of 1697 // bytes that were not added to the free list, because they freed memory block 1698 // was too small. Bookkeeping information will be written to the block, i.e., 1699 // its contents will be destroyed. The start address should be word aligned, 1700 // and the size should be a non-zero multiple of the word size. 1701 size_t Free(Address start, size_t size_in_bytes, FreeMode mode); 1702 1703 // Allocate a block of size {size_in_bytes} from the free list. The block is 1704 // unitialized. A failure is returned if no block is available. The size 1705 // should be a non-zero multiple of the word size. 1706 MUST_USE_RESULT HeapObject* Allocate(size_t size_in_bytes); 1707 1708 // Clear the free list. 1709 void Reset(); 1710 1711 void ResetStats() { 1712 wasted_bytes_.SetValue(0); 1713 ForAllFreeListCategories( 1714 [](FreeListCategory* category) { category->ResetStats(); }); 1715 } 1716 1717 // Return the number of bytes available on the free list. 1718 size_t Available() { 1719 size_t available = 0; 1720 ForAllFreeListCategories([&available](FreeListCategory* category) { 1721 available += category->available(); 1722 }); 1723 return available; 1724 } 1725 1726 bool IsEmpty() { 1727 bool empty = true; 1728 ForAllFreeListCategories([&empty](FreeListCategory* category) { 1729 if (!category->is_empty()) empty = false; 1730 }); 1731 return empty; 1732 } 1733 1734 // Used after booting the VM. 1735 void RepairLists(Heap* heap); 1736 1737 size_t EvictFreeListItems(Page* page); 1738 bool ContainsPageFreeListItems(Page* page); 1739 1740 PagedSpace* owner() { return owner_; } 1741 size_t wasted_bytes() { return wasted_bytes_.Value(); } 1742 1743 template <typename Callback> 1744 void ForAllFreeListCategories(FreeListCategoryType type, Callback callback) { 1745 FreeListCategory* current = categories_[type]; 1746 while (current != nullptr) { 1747 FreeListCategory* next = current->next(); 1748 callback(current); 1749 current = next; 1750 } 1751 } 1752 1753 template <typename Callback> 1754 void ForAllFreeListCategories(Callback callback) { 1755 for (int i = kFirstCategory; i < kNumberOfCategories; i++) { 1756 ForAllFreeListCategories(static_cast<FreeListCategoryType>(i), callback); 1757 } 1758 } 1759 1760 bool AddCategory(FreeListCategory* category); 1761 void RemoveCategory(FreeListCategory* category); 1762 void PrintCategories(FreeListCategoryType type); 1763 1764 #ifdef DEBUG 1765 size_t SumFreeLists(); 1766 bool IsVeryLong(); 1767 #endif 1768 1769 private: 1770 class FreeListCategoryIterator { 1771 public: 1772 FreeListCategoryIterator(FreeList* free_list, FreeListCategoryType type) 1773 : current_(free_list->categories_[type]) {} 1774 1775 bool HasNext() { return current_ != nullptr; } 1776 1777 FreeListCategory* Next() { 1778 DCHECK(HasNext()); 1779 FreeListCategory* tmp = current_; 1780 current_ = current_->next(); 1781 return tmp; 1782 } 1783 1784 private: 1785 FreeListCategory* current_; 1786 }; 1787 1788 // The size range of blocks, in bytes. 1789 static const size_t kMinBlockSize = 3 * kPointerSize; 1790 static const size_t kMaxBlockSize = Page::kAllocatableMemory; 1791 1792 static const size_t kTiniestListMax = 0xa * kPointerSize; 1793 static const size_t kTinyListMax = 0x1f * kPointerSize; 1794 static const size_t kSmallListMax = 0xff * kPointerSize; 1795 static const size_t kMediumListMax = 0x7ff * kPointerSize; 1796 static const size_t kLargeListMax = 0x3fff * kPointerSize; 1797 static const size_t kTinyAllocationMax = kTiniestListMax; 1798 static const size_t kSmallAllocationMax = kTinyListMax; 1799 static const size_t kMediumAllocationMax = kSmallListMax; 1800 static const size_t kLargeAllocationMax = kMediumListMax; 1801 1802 FreeSpace* FindNodeFor(size_t size_in_bytes, size_t* node_size); 1803 1804 // Walks all available categories for a given |type| and tries to retrieve 1805 // a node. Returns nullptr if the category is empty. 1806 FreeSpace* FindNodeIn(FreeListCategoryType type, size_t* node_size); 1807 1808 // Tries to retrieve a node from the first category in a given |type|. 1809 // Returns nullptr if the category is empty. 1810 FreeSpace* TryFindNodeIn(FreeListCategoryType type, size_t* node_size, 1811 size_t minimum_size); 1812 1813 // Searches a given |type| for a node of at least |minimum_size|. 1814 FreeSpace* SearchForNodeInList(FreeListCategoryType type, size_t* node_size, 1815 size_t minimum_size); 1816 1817 FreeListCategoryType SelectFreeListCategoryType(size_t size_in_bytes) { 1818 if (size_in_bytes <= kTiniestListMax) { 1819 return kTiniest; 1820 } else if (size_in_bytes <= kTinyListMax) { 1821 return kTiny; 1822 } else if (size_in_bytes <= kSmallListMax) { 1823 return kSmall; 1824 } else if (size_in_bytes <= kMediumListMax) { 1825 return kMedium; 1826 } else if (size_in_bytes <= kLargeListMax) { 1827 return kLarge; 1828 } 1829 return kHuge; 1830 } 1831 1832 // The tiny categories are not used for fast allocation. 1833 FreeListCategoryType SelectFastAllocationFreeListCategoryType( 1834 size_t size_in_bytes) { 1835 if (size_in_bytes <= kSmallAllocationMax) { 1836 return kSmall; 1837 } else if (size_in_bytes <= kMediumAllocationMax) { 1838 return kMedium; 1839 } else if (size_in_bytes <= kLargeAllocationMax) { 1840 return kLarge; 1841 } 1842 return kHuge; 1843 } 1844 1845 FreeListCategory* top(FreeListCategoryType type) { return categories_[type]; } 1846 1847 PagedSpace* owner_; 1848 base::AtomicNumber<size_t> wasted_bytes_; 1849 FreeListCategory* categories_[kNumberOfCategories]; 1850 1851 friend class FreeListCategory; 1852 1853 DISALLOW_IMPLICIT_CONSTRUCTORS(FreeList); 1854 }; 1855 1856 // LocalAllocationBuffer represents a linear allocation area that is created 1857 // from a given {AllocationResult} and can be used to allocate memory without 1858 // synchronization. 1859 // 1860 // The buffer is properly closed upon destruction and reassignment. 1861 // Example: 1862 // { 1863 // AllocationResult result = ...; 1864 // LocalAllocationBuffer a(heap, result, size); 1865 // LocalAllocationBuffer b = a; 1866 // CHECK(!a.IsValid()); 1867 // CHECK(b.IsValid()); 1868 // // {a} is invalid now and cannot be used for further allocations. 1869 // } 1870 // // Since {b} went out of scope, the LAB is closed, resulting in creating a 1871 // // filler object for the remaining area. 1872 class LocalAllocationBuffer { 1873 public: 1874 // Indicates that a buffer cannot be used for allocations anymore. Can result 1875 // from either reassigning a buffer, or trying to construct it from an 1876 // invalid {AllocationResult}. 1877 static inline LocalAllocationBuffer InvalidBuffer(); 1878 1879 // Creates a new LAB from a given {AllocationResult}. Results in 1880 // InvalidBuffer if the result indicates a retry. 1881 static inline LocalAllocationBuffer FromResult(Heap* heap, 1882 AllocationResult result, 1883 intptr_t size); 1884 1885 ~LocalAllocationBuffer() { Close(); } 1886 1887 // Convert to C++11 move-semantics once allowed by the style guide. 1888 LocalAllocationBuffer(const LocalAllocationBuffer& other); 1889 LocalAllocationBuffer& operator=(const LocalAllocationBuffer& other); 1890 1891 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 1892 int size_in_bytes, AllocationAlignment alignment); 1893 1894 inline bool IsValid() { return allocation_info_.top() != nullptr; } 1895 1896 // Try to merge LABs, which is only possible when they are adjacent in memory. 1897 // Returns true if the merge was successful, false otherwise. 1898 inline bool TryMerge(LocalAllocationBuffer* other); 1899 1900 private: 1901 LocalAllocationBuffer(Heap* heap, AllocationInfo allocation_info); 1902 1903 void Close(); 1904 1905 Heap* heap_; 1906 AllocationInfo allocation_info_; 1907 }; 1908 1909 class V8_EXPORT_PRIVATE PagedSpace : NON_EXPORTED_BASE(public Space) { 1910 public: 1911 typedef PageIterator iterator; 1912 1913 static const intptr_t kCompactionMemoryWanted = 500 * KB; 1914 1915 // Creates a space with an id. 1916 PagedSpace(Heap* heap, AllocationSpace id, Executability executable); 1917 1918 ~PagedSpace() override { TearDown(); } 1919 1920 // Set up the space using the given address range of virtual memory (from 1921 // the memory allocator's initial chunk) if possible. If the block of 1922 // addresses is not big enough to contain a single page-aligned page, a 1923 // fresh chunk will be allocated. 1924 bool SetUp(); 1925 1926 // Returns true if the space has been successfully set up and not 1927 // subsequently torn down. 1928 bool HasBeenSetUp(); 1929 1930 // Checks whether an object/address is in this space. 1931 inline bool Contains(Address a); 1932 inline bool Contains(Object* o); 1933 bool ContainsSlow(Address addr); 1934 1935 // During boot the free_space_map is created, and afterwards we may need 1936 // to write it into the free list nodes that were already created. 1937 void RepairFreeListsAfterDeserialization(); 1938 1939 // Prepares for a mark-compact GC. 1940 void PrepareForMarkCompact(); 1941 1942 // Current capacity without growing (Size() + Available()). 1943 size_t Capacity() { return accounting_stats_.Capacity(); } 1944 1945 // Approximate amount of physical memory committed for this space. 1946 size_t CommittedPhysicalMemory() override; 1947 1948 void ResetFreeListStatistics(); 1949 1950 // Sets the capacity, the available space and the wasted space to zero. 1951 // The stats are rebuilt during sweeping by adding each page to the 1952 // capacity and the size when it is encountered. As free spaces are 1953 // discovered during the sweeping they are subtracted from the size and added 1954 // to the available and wasted totals. 1955 void ClearStats() { 1956 accounting_stats_.ClearSize(); 1957 free_list_.ResetStats(); 1958 ResetFreeListStatistics(); 1959 } 1960 1961 // Available bytes without growing. These are the bytes on the free list. 1962 // The bytes in the linear allocation area are not included in this total 1963 // because updating the stats would slow down allocation. New pages are 1964 // immediately added to the free list so they show up here. 1965 size_t Available() override { return free_list_.Available(); } 1966 1967 // Allocated bytes in this space. Garbage bytes that were not found due to 1968 // concurrent sweeping are counted as being allocated! The bytes in the 1969 // current linear allocation area (between top and limit) are also counted 1970 // here. 1971 size_t Size() override { return accounting_stats_.Size(); } 1972 1973 // As size, but the bytes in lazily swept pages are estimated and the bytes 1974 // in the current linear allocation area are not included. 1975 size_t SizeOfObjects() override; 1976 1977 // Wasted bytes in this space. These are just the bytes that were thrown away 1978 // due to being too small to use for allocation. 1979 virtual size_t Waste() { return free_list_.wasted_bytes(); } 1980 1981 // Returns the allocation pointer in this space. 1982 Address top() { return allocation_info_.top(); } 1983 Address limit() { return allocation_info_.limit(); } 1984 1985 // The allocation top address. 1986 Address* allocation_top_address() { return allocation_info_.top_address(); } 1987 1988 // The allocation limit address. 1989 Address* allocation_limit_address() { 1990 return allocation_info_.limit_address(); 1991 } 1992 1993 enum UpdateSkipList { UPDATE_SKIP_LIST, IGNORE_SKIP_LIST }; 1994 1995 // Allocate the requested number of bytes in the space if possible, return a 1996 // failure object if not. Only use IGNORE_SKIP_LIST if the skip list is going 1997 // to be manually updated later. 1998 MUST_USE_RESULT inline AllocationResult AllocateRawUnaligned( 1999 int size_in_bytes, UpdateSkipList update_skip_list = UPDATE_SKIP_LIST); 2000 2001 MUST_USE_RESULT inline AllocationResult AllocateRawUnalignedSynchronized( 2002 int size_in_bytes); 2003 2004 // Allocate the requested number of bytes in the space double aligned if 2005 // possible, return a failure object if not. 2006 MUST_USE_RESULT inline AllocationResult AllocateRawAligned( 2007 int size_in_bytes, AllocationAlignment alignment); 2008 2009 // Allocate the requested number of bytes in the space and consider allocation 2010 // alignment if needed. 2011 MUST_USE_RESULT inline AllocationResult AllocateRaw( 2012 int size_in_bytes, AllocationAlignment alignment); 2013 2014 // Give a block of memory to the space's free list. It might be added to 2015 // the free list or accounted as waste. 2016 // If add_to_freelist is false then just accounting stats are updated and 2017 // no attempt to add area to free list is made. 2018 size_t Free(Address start, size_t size_in_bytes) { 2019 size_t wasted = free_list_.Free(start, size_in_bytes, kLinkCategory); 2020 accounting_stats_.DeallocateBytes(size_in_bytes); 2021 DCHECK_GE(size_in_bytes, wasted); 2022 return size_in_bytes - wasted; 2023 } 2024 2025 size_t UnaccountedFree(Address start, size_t size_in_bytes) { 2026 size_t wasted = free_list_.Free(start, size_in_bytes, kDoNotLinkCategory); 2027 DCHECK_GE(size_in_bytes, wasted); 2028 return size_in_bytes - wasted; 2029 } 2030 2031 void ResetFreeList() { free_list_.Reset(); } 2032 2033 // Set space allocation info. 2034 void SetTopAndLimit(Address top, Address limit) { 2035 DCHECK(top == limit || 2036 Page::FromAddress(top) == Page::FromAddress(limit - 1)); 2037 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 2038 allocation_info_.Reset(top, limit); 2039 } 2040 2041 void SetAllocationInfo(Address top, Address limit); 2042 2043 // Empty space allocation info, returning unused area to free list. 2044 void EmptyAllocationInfo(); 2045 2046 void MarkAllocationInfoBlack(); 2047 2048 void AccountAllocatedBytes(size_t bytes) { 2049 accounting_stats_.AllocateBytes(bytes); 2050 } 2051 2052 void IncreaseCapacity(size_t bytes); 2053 2054 // Releases an unused page and shrinks the space. 2055 void ReleasePage(Page* page); 2056 2057 // The dummy page that anchors the linked list of pages. 2058 Page* anchor() { return &anchor_; } 2059 2060 2061 #ifdef VERIFY_HEAP 2062 // Verify integrity of this space. 2063 virtual void Verify(ObjectVisitor* visitor); 2064 2065 // Overridden by subclasses to verify space-specific object 2066 // properties (e.g., only maps or free-list nodes are in map space). 2067 virtual void VerifyObject(HeapObject* obj) {} 2068 #endif 2069 2070 #ifdef DEBUG 2071 // Print meta info and objects in this space. 2072 void Print() override; 2073 2074 // Reports statistics for the space 2075 void ReportStatistics(); 2076 2077 // Report code object related statistics 2078 static void ReportCodeStatistics(Isolate* isolate); 2079 static void ResetCodeStatistics(Isolate* isolate); 2080 #endif 2081 2082 Page* FirstPage() { return anchor_.next_page(); } 2083 Page* LastPage() { return anchor_.prev_page(); } 2084 2085 bool CanExpand(size_t size); 2086 2087 // Returns the number of total pages in this space. 2088 int CountTotalPages(); 2089 2090 // Return size of allocatable area on a page in this space. 2091 inline int AreaSize() { return static_cast<int>(area_size_); } 2092 2093 virtual bool is_local() { return false; } 2094 2095 // Merges {other} into the current space. Note that this modifies {other}, 2096 // e.g., removes its bump pointer area and resets statistics. 2097 void MergeCompactionSpace(CompactionSpace* other); 2098 2099 // Refills the free list from the corresponding free list filled by the 2100 // sweeper. 2101 virtual void RefillFreeList(); 2102 2103 FreeList* free_list() { return &free_list_; } 2104 2105 base::Mutex* mutex() { return &space_mutex_; } 2106 2107 inline void UnlinkFreeListCategories(Page* page); 2108 inline intptr_t RelinkFreeListCategories(Page* page); 2109 2110 iterator begin() { return iterator(anchor_.next_page()); } 2111 iterator end() { return iterator(&anchor_); } 2112 2113 // Shrink immortal immovable pages of the space to be exactly the size needed 2114 // using the high water mark. 2115 void ShrinkImmortalImmovablePages(); 2116 2117 std::unique_ptr<ObjectIterator> GetObjectIterator() override; 2118 2119 protected: 2120 // PagedSpaces that should be included in snapshots have different, i.e., 2121 // smaller, initial pages. 2122 virtual bool snapshotable() { return true; } 2123 2124 bool HasPages() { return anchor_.next_page() != &anchor_; } 2125 2126 // Cleans up the space, frees all pages in this space except those belonging 2127 // to the initial chunk, uncommits addresses in the initial chunk. 2128 void TearDown(); 2129 2130 // Expands the space by allocating a fixed number of pages. Returns false if 2131 // it cannot allocate requested number of pages from OS, or if the hard heap 2132 // size limit has been hit. 2133 bool Expand(); 2134 2135 // Generic fast case allocation function that tries linear allocation at the 2136 // address denoted by top in allocation_info_. 2137 inline HeapObject* AllocateLinearly(int size_in_bytes); 2138 2139 // Generic fast case allocation function that tries aligned linear allocation 2140 // at the address denoted by top in allocation_info_. Writes the aligned 2141 // allocation size, which includes the filler size, to size_in_bytes. 2142 inline HeapObject* AllocateLinearlyAligned(int* size_in_bytes, 2143 AllocationAlignment alignment); 2144 2145 // If sweeping is still in progress try to sweep unswept pages. If that is 2146 // not successful, wait for the sweeper threads and re-try free-list 2147 // allocation. 2148 MUST_USE_RESULT virtual HeapObject* SweepAndRetryAllocation( 2149 int size_in_bytes); 2150 2151 // Slow path of AllocateRaw. This function is space-dependent. 2152 MUST_USE_RESULT HeapObject* SlowAllocateRaw(int size_in_bytes); 2153 2154 size_t area_size_; 2155 2156 // Accounting information for this space. 2157 AllocationStats accounting_stats_; 2158 2159 // The dummy page that anchors the double linked list of pages. 2160 Page anchor_; 2161 2162 // The space's free list. 2163 FreeList free_list_; 2164 2165 // Normal allocation information. 2166 AllocationInfo allocation_info_; 2167 2168 // Mutex guarding any concurrent access to the space. 2169 base::Mutex space_mutex_; 2170 2171 friend class IncrementalMarking; 2172 friend class MarkCompactCollector; 2173 2174 // Used in cctest. 2175 friend class HeapTester; 2176 }; 2177 2178 enum SemiSpaceId { kFromSpace = 0, kToSpace = 1 }; 2179 2180 // ----------------------------------------------------------------------------- 2181 // SemiSpace in young generation 2182 // 2183 // A SemiSpace is a contiguous chunk of memory holding page-like memory chunks. 2184 // The mark-compact collector uses the memory of the first page in the from 2185 // space as a marking stack when tracing live objects. 2186 class SemiSpace : public Space { 2187 public: 2188 typedef PageIterator iterator; 2189 2190 static void Swap(SemiSpace* from, SemiSpace* to); 2191 2192 SemiSpace(Heap* heap, SemiSpaceId semispace) 2193 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2194 current_capacity_(0), 2195 maximum_capacity_(0), 2196 minimum_capacity_(0), 2197 age_mark_(nullptr), 2198 committed_(false), 2199 id_(semispace), 2200 anchor_(this), 2201 current_page_(nullptr), 2202 pages_used_(0) {} 2203 2204 inline bool Contains(HeapObject* o); 2205 inline bool Contains(Object* o); 2206 inline bool ContainsSlow(Address a); 2207 2208 void SetUp(size_t initial_capacity, size_t maximum_capacity); 2209 void TearDown(); 2210 bool HasBeenSetUp() { return maximum_capacity_ != 0; } 2211 2212 bool Commit(); 2213 bool Uncommit(); 2214 bool is_committed() { return committed_; } 2215 2216 // Grow the semispace to the new capacity. The new capacity requested must 2217 // be larger than the current capacity and less than the maximum capacity. 2218 bool GrowTo(size_t new_capacity); 2219 2220 // Shrinks the semispace to the new capacity. The new capacity requested 2221 // must be more than the amount of used memory in the semispace and less 2222 // than the current capacity. 2223 bool ShrinkTo(size_t new_capacity); 2224 2225 bool EnsureCurrentCapacity(); 2226 2227 // Returns the start address of the first page of the space. 2228 Address space_start() { 2229 DCHECK_NE(anchor_.next_page(), anchor()); 2230 return anchor_.next_page()->area_start(); 2231 } 2232 2233 Page* first_page() { return anchor_.next_page(); } 2234 Page* current_page() { return current_page_; } 2235 int pages_used() { return pages_used_; } 2236 2237 // Returns one past the end address of the space. 2238 Address space_end() { return anchor_.prev_page()->area_end(); } 2239 2240 // Returns the start address of the current page of the space. 2241 Address page_low() { return current_page_->area_start(); } 2242 2243 // Returns one past the end address of the current page of the space. 2244 Address page_high() { return current_page_->area_end(); } 2245 2246 bool AdvancePage() { 2247 Page* next_page = current_page_->next_page(); 2248 // We cannot expand if we reached the maximum number of pages already. Note 2249 // that we need to account for the next page already for this check as we 2250 // could potentially fill the whole page after advancing. 2251 const bool reached_max_pages = (pages_used_ + 1) == max_pages(); 2252 if (next_page == anchor() || reached_max_pages) { 2253 return false; 2254 } 2255 current_page_ = next_page; 2256 pages_used_++; 2257 return true; 2258 } 2259 2260 // Resets the space to using the first page. 2261 void Reset(); 2262 2263 void RemovePage(Page* page); 2264 void PrependPage(Page* page); 2265 2266 // Age mark accessors. 2267 Address age_mark() { return age_mark_; } 2268 void set_age_mark(Address mark); 2269 2270 // Returns the current capacity of the semispace. 2271 size_t current_capacity() { return current_capacity_; } 2272 2273 // Returns the maximum capacity of the semispace. 2274 size_t maximum_capacity() { return maximum_capacity_; } 2275 2276 // Returns the initial capacity of the semispace. 2277 size_t minimum_capacity() { return minimum_capacity_; } 2278 2279 SemiSpaceId id() { return id_; } 2280 2281 // Approximate amount of physical memory committed for this space. 2282 size_t CommittedPhysicalMemory() override; 2283 2284 // If we don't have these here then SemiSpace will be abstract. However 2285 // they should never be called: 2286 2287 size_t Size() override { 2288 UNREACHABLE(); 2289 return 0; 2290 } 2291 2292 size_t SizeOfObjects() override { return Size(); } 2293 2294 size_t Available() override { 2295 UNREACHABLE(); 2296 return 0; 2297 } 2298 2299 iterator begin() { return iterator(anchor_.next_page()); } 2300 iterator end() { return iterator(anchor()); } 2301 2302 std::unique_ptr<ObjectIterator> GetObjectIterator() override; 2303 2304 #ifdef DEBUG 2305 void Print() override; 2306 // Validate a range of of addresses in a SemiSpace. 2307 // The "from" address must be on a page prior to the "to" address, 2308 // in the linked page order, or it must be earlier on the same page. 2309 static void AssertValidRange(Address from, Address to); 2310 #else 2311 // Do nothing. 2312 inline static void AssertValidRange(Address from, Address to) {} 2313 #endif 2314 2315 #ifdef VERIFY_HEAP 2316 virtual void Verify(); 2317 #endif 2318 2319 private: 2320 void RewindPages(Page* start, int num_pages); 2321 2322 inline Page* anchor() { return &anchor_; } 2323 inline int max_pages() { 2324 return static_cast<int>(current_capacity_ / Page::kPageSize); 2325 } 2326 2327 // Copies the flags into the masked positions on all pages in the space. 2328 void FixPagesFlags(intptr_t flags, intptr_t flag_mask); 2329 2330 // The currently committed space capacity. 2331 size_t current_capacity_; 2332 2333 // The maximum capacity that can be used by this space. A space cannot grow 2334 // beyond that size. 2335 size_t maximum_capacity_; 2336 2337 // The minimum capacity for the space. A space cannot shrink below this size. 2338 size_t minimum_capacity_; 2339 2340 // Used to govern object promotion during mark-compact collection. 2341 Address age_mark_; 2342 2343 bool committed_; 2344 SemiSpaceId id_; 2345 2346 Page anchor_; 2347 Page* current_page_; 2348 int pages_used_; 2349 2350 friend class NewSpace; 2351 friend class SemiSpaceIterator; 2352 }; 2353 2354 2355 // A SemiSpaceIterator is an ObjectIterator that iterates over the active 2356 // semispace of the heap's new space. It iterates over the objects in the 2357 // semispace from a given start address (defaulting to the bottom of the 2358 // semispace) to the top of the semispace. New objects allocated after the 2359 // iterator is created are not iterated. 2360 class SemiSpaceIterator : public ObjectIterator { 2361 public: 2362 // Create an iterator over the allocated objects in the given to-space. 2363 explicit SemiSpaceIterator(NewSpace* space); 2364 2365 inline HeapObject* Next() override; 2366 2367 private: 2368 void Initialize(Address start, Address end); 2369 2370 // The current iteration point. 2371 Address current_; 2372 // The end of iteration. 2373 Address limit_; 2374 }; 2375 2376 // ----------------------------------------------------------------------------- 2377 // The young generation space. 2378 // 2379 // The new space consists of a contiguous pair of semispaces. It simply 2380 // forwards most functions to the appropriate semispace. 2381 2382 class NewSpace : public Space { 2383 public: 2384 typedef PageIterator iterator; 2385 2386 explicit NewSpace(Heap* heap) 2387 : Space(heap, NEW_SPACE, NOT_EXECUTABLE), 2388 to_space_(heap, kToSpace), 2389 from_space_(heap, kFromSpace), 2390 reservation_(), 2391 top_on_previous_step_(0), 2392 allocated_histogram_(nullptr), 2393 promoted_histogram_(nullptr) {} 2394 2395 inline bool Contains(HeapObject* o); 2396 inline bool ContainsSlow(Address a); 2397 inline bool Contains(Object* o); 2398 2399 bool SetUp(size_t initial_semispace_capacity, size_t max_semispace_capacity); 2400 2401 // Tears down the space. Heap memory was not allocated by the space, so it 2402 // is not deallocated here. 2403 void TearDown(); 2404 2405 // True if the space has been set up but not torn down. 2406 bool HasBeenSetUp() { 2407 return to_space_.HasBeenSetUp() && from_space_.HasBeenSetUp(); 2408 } 2409 2410 // Flip the pair of spaces. 2411 void Flip(); 2412 2413 // Grow the capacity of the semispaces. Assumes that they are not at 2414 // their maximum capacity. 2415 void Grow(); 2416 2417 // Shrink the capacity of the semispaces. 2418 void Shrink(); 2419 2420 // Return the allocated bytes in the active semispace. 2421 size_t Size() override { 2422 DCHECK_GE(top(), to_space_.page_low()); 2423 return to_space_.pages_used() * Page::kAllocatableMemory + 2424 static_cast<size_t>(top() - to_space_.page_low()); 2425 } 2426 2427 size_t SizeOfObjects() override { return Size(); } 2428 2429 // Return the allocatable capacity of a semispace. 2430 size_t Capacity() { 2431 SLOW_DCHECK(to_space_.current_capacity() == from_space_.current_capacity()); 2432 return (to_space_.current_capacity() / Page::kPageSize) * 2433 Page::kAllocatableMemory; 2434 } 2435 2436 // Return the current size of a semispace, allocatable and non-allocatable 2437 // memory. 2438 size_t TotalCapacity() { 2439 DCHECK(to_space_.current_capacity() == from_space_.current_capacity()); 2440 return to_space_.current_capacity(); 2441 } 2442 2443 // Committed memory for NewSpace is the committed memory of both semi-spaces 2444 // combined. 2445 size_t CommittedMemory() override { 2446 return from_space_.CommittedMemory() + to_space_.CommittedMemory(); 2447 } 2448 2449 size_t MaximumCommittedMemory() override { 2450 return from_space_.MaximumCommittedMemory() + 2451 to_space_.MaximumCommittedMemory(); 2452 } 2453 2454 // Approximate amount of physical memory committed for this space. 2455 size_t CommittedPhysicalMemory() override; 2456 2457 // Return the available bytes without growing. 2458 size_t Available() override { 2459 DCHECK_GE(Capacity(), Size()); 2460 return Capacity() - Size(); 2461 } 2462 2463 size_t AllocatedSinceLastGC() { 2464 const Address age_mark = to_space_.age_mark(); 2465 DCHECK_NOT_NULL(age_mark); 2466 DCHECK_NOT_NULL(top()); 2467 Page* const age_mark_page = Page::FromAllocationAreaAddress(age_mark); 2468 Page* const last_page = Page::FromAllocationAreaAddress(top()); 2469 Page* current_page = age_mark_page; 2470 size_t allocated = 0; 2471 if (current_page != last_page) { 2472 DCHECK_EQ(current_page, age_mark_page); 2473 DCHECK_GE(age_mark_page->area_end(), age_mark); 2474 allocated += age_mark_page->area_end() - age_mark; 2475 current_page = current_page->next_page(); 2476 } else { 2477 DCHECK_GE(top(), age_mark); 2478 return top() - age_mark; 2479 } 2480 while (current_page != last_page) { 2481 DCHECK_NE(current_page, age_mark_page); 2482 allocated += Page::kAllocatableMemory; 2483 current_page = current_page->next_page(); 2484 } 2485 DCHECK_GE(top(), current_page->area_start()); 2486 allocated += top() - current_page->area_start(); 2487 DCHECK_LE(allocated, Size()); 2488 return allocated; 2489 } 2490 2491 void MovePageFromSpaceToSpace(Page* page) { 2492 DCHECK(page->InFromSpace()); 2493 from_space_.RemovePage(page); 2494 to_space_.PrependPage(page); 2495 } 2496 2497 bool Rebalance(); 2498 2499 // Return the maximum capacity of a semispace. 2500 size_t MaximumCapacity() { 2501 DCHECK(to_space_.maximum_capacity() == from_space_.maximum_capacity()); 2502 return to_space_.maximum_capacity(); 2503 } 2504 2505 bool IsAtMaximumCapacity() { return TotalCapacity() == MaximumCapacity(); } 2506 2507 // Returns the initial capacity of a semispace. 2508 size_t InitialTotalCapacity() { 2509 DCHECK(to_space_.minimum_capacity() == from_space_.minimum_capacity()); 2510 return to_space_.minimum_capacity(); 2511 } 2512 2513 // Return the address of the allocation pointer in the active semispace. 2514 Address top() { 2515 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.top())); 2516 return allocation_info_.top(); 2517 } 2518 2519 // Return the address of the allocation pointer limit in the active semispace. 2520 Address limit() { 2521 DCHECK(to_space_.current_page()->ContainsLimit(allocation_info_.limit())); 2522 return allocation_info_.limit(); 2523 } 2524 2525 // Return the address of the first object in the active semispace. 2526 Address bottom() { return to_space_.space_start(); } 2527 2528 // Get the age mark of the inactive semispace. 2529 Address age_mark() { return from_space_.age_mark(); } 2530 // Set the age mark in the active semispace. 2531 void set_age_mark(Address mark) { to_space_.set_age_mark(mark); } 2532 2533 // The allocation top and limit address. 2534 Address* allocation_top_address() { return allocation_info_.top_address(); } 2535 2536 // The allocation limit address. 2537 Address* allocation_limit_address() { 2538 return allocation_info_.limit_address(); 2539 } 2540 2541 MUST_USE_RESULT INLINE(AllocationResult AllocateRawAligned( 2542 int size_in_bytes, AllocationAlignment alignment)); 2543 2544 MUST_USE_RESULT INLINE( 2545 AllocationResult AllocateRawUnaligned(int size_in_bytes)); 2546 2547 MUST_USE_RESULT INLINE(AllocationResult AllocateRaw( 2548 int size_in_bytes, AllocationAlignment alignment)); 2549 2550 MUST_USE_RESULT inline AllocationResult AllocateRawSynchronized( 2551 int size_in_bytes, AllocationAlignment alignment); 2552 2553 // Reset the allocation pointer to the beginning of the active semispace. 2554 void ResetAllocationInfo(); 2555 2556 // When inline allocation stepping is active, either because of incremental 2557 // marking, idle scavenge, or allocation statistics gathering, we 'interrupt' 2558 // inline allocation every once in a while. This is done by setting 2559 // allocation_info_.limit to be lower than the actual limit and and increasing 2560 // it in steps to guarantee that the observers are notified periodically. 2561 void UpdateInlineAllocationLimit(int size_in_bytes); 2562 2563 void DisableInlineAllocationSteps() { 2564 top_on_previous_step_ = 0; 2565 UpdateInlineAllocationLimit(0); 2566 } 2567 2568 // Allows observation of inline allocation. The observer->Step() method gets 2569 // called after every step_size bytes have been allocated (approximately). 2570 // This works by adjusting the allocation limit to a lower value and adjusting 2571 // it after each step. 2572 void AddAllocationObserver(AllocationObserver* observer) override; 2573 2574 void RemoveAllocationObserver(AllocationObserver* observer) override; 2575 2576 // Get the extent of the inactive semispace (for use as a marking stack, 2577 // or to zap it). Notice: space-addresses are not necessarily on the 2578 // same page, so FromSpaceStart() might be above FromSpaceEnd(). 2579 Address FromSpacePageLow() { return from_space_.page_low(); } 2580 Address FromSpacePageHigh() { return from_space_.page_high(); } 2581 Address FromSpaceStart() { return from_space_.space_start(); } 2582 Address FromSpaceEnd() { return from_space_.space_end(); } 2583 2584 // Get the extent of the active semispace's pages' memory. 2585 Address ToSpaceStart() { return to_space_.space_start(); } 2586 Address ToSpaceEnd() { return to_space_.space_end(); } 2587 2588 inline bool ToSpaceContainsSlow(Address a); 2589 inline bool FromSpaceContainsSlow(Address a); 2590 inline bool ToSpaceContains(Object* o); 2591 inline bool FromSpaceContains(Object* o); 2592 2593 // Try to switch the active semispace to a new, empty, page. 2594 // Returns false if this isn't possible or reasonable (i.e., there 2595 // are no pages, or the current page is already empty), or true 2596 // if successful. 2597 bool AddFreshPage(); 2598 bool AddFreshPageSynchronized(); 2599 2600 #ifdef VERIFY_HEAP 2601 // Verify the active semispace. 2602 virtual void Verify(); 2603 #endif 2604 2605 #ifdef DEBUG 2606 // Print the active semispace. 2607 void Print() override { to_space_.Print(); } 2608 #endif 2609 2610 // Iterates the active semispace to collect statistics. 2611 void CollectStatistics(); 2612 // Reports previously collected statistics of the active semispace. 2613 void ReportStatistics(); 2614 // Clears previously collected statistics. 2615 void ClearHistograms(); 2616 2617 // Record the allocation or promotion of a heap object. Note that we don't 2618 // record every single allocation, but only those that happen in the 2619 // to space during a scavenge GC. 2620 void RecordAllocation(HeapObject* obj); 2621 void RecordPromotion(HeapObject* obj); 2622 2623 // Return whether the operation succeded. 2624 bool CommitFromSpaceIfNeeded() { 2625 if (from_space_.is_committed()) return true; 2626 return from_space_.Commit(); 2627 } 2628 2629 bool UncommitFromSpace() { 2630 if (!from_space_.is_committed()) return true; 2631 return from_space_.Uncommit(); 2632 } 2633 2634 bool IsFromSpaceCommitted() { return from_space_.is_committed(); } 2635 2636 SemiSpace* active_space() { return &to_space_; } 2637 2638 void PauseAllocationObservers() override; 2639 void ResumeAllocationObservers() override; 2640 2641 iterator begin() { return to_space_.begin(); } 2642 iterator end() { return to_space_.end(); } 2643 2644 std::unique_ptr<ObjectIterator> GetObjectIterator() override; 2645 2646 private: 2647 // Update allocation info to match the current to-space page. 2648 void UpdateAllocationInfo(); 2649 2650 base::Mutex mutex_; 2651 2652 // The semispaces. 2653 SemiSpace to_space_; 2654 SemiSpace from_space_; 2655 base::VirtualMemory reservation_; 2656 2657 // Allocation pointer and limit for normal allocation and allocation during 2658 // mark-compact collection. 2659 AllocationInfo allocation_info_; 2660 2661 Address top_on_previous_step_; 2662 2663 HistogramInfo* allocated_histogram_; 2664 HistogramInfo* promoted_histogram_; 2665 2666 bool EnsureAllocation(int size_in_bytes, AllocationAlignment alignment); 2667 2668 // If we are doing inline allocation in steps, this method performs the 'step' 2669 // operation. top is the memory address of the bump pointer at the last 2670 // inline allocation (i.e. it determines the numbers of bytes actually 2671 // allocated since the last step.) new_top is the address of the bump pointer 2672 // where the next byte is going to be allocated from. top and new_top may be 2673 // different when we cross a page boundary or reset the space. 2674 void InlineAllocationStep(Address top, Address new_top, Address soon_object, 2675 size_t size); 2676 intptr_t GetNextInlineAllocationStepSize(); 2677 void StartNextInlineAllocationStep(); 2678 2679 friend class SemiSpaceIterator; 2680 }; 2681 2682 class PauseAllocationObserversScope { 2683 public: 2684 explicit PauseAllocationObserversScope(Heap* heap); 2685 ~PauseAllocationObserversScope(); 2686 2687 private: 2688 Heap* heap_; 2689 DISALLOW_COPY_AND_ASSIGN(PauseAllocationObserversScope); 2690 }; 2691 2692 // ----------------------------------------------------------------------------- 2693 // Compaction space that is used temporarily during compaction. 2694 2695 class CompactionSpace : public PagedSpace { 2696 public: 2697 CompactionSpace(Heap* heap, AllocationSpace id, Executability executable) 2698 : PagedSpace(heap, id, executable) {} 2699 2700 bool is_local() override { return true; } 2701 2702 protected: 2703 // The space is temporary and not included in any snapshots. 2704 bool snapshotable() override { return false; } 2705 2706 MUST_USE_RESULT HeapObject* SweepAndRetryAllocation( 2707 int size_in_bytes) override; 2708 }; 2709 2710 2711 // A collection of |CompactionSpace|s used by a single compaction task. 2712 class CompactionSpaceCollection : public Malloced { 2713 public: 2714 explicit CompactionSpaceCollection(Heap* heap) 2715 : old_space_(heap, OLD_SPACE, Executability::NOT_EXECUTABLE), 2716 code_space_(heap, CODE_SPACE, Executability::EXECUTABLE) {} 2717 2718 CompactionSpace* Get(AllocationSpace space) { 2719 switch (space) { 2720 case OLD_SPACE: 2721 return &old_space_; 2722 case CODE_SPACE: 2723 return &code_space_; 2724 default: 2725 UNREACHABLE(); 2726 } 2727 UNREACHABLE(); 2728 return nullptr; 2729 } 2730 2731 private: 2732 CompactionSpace old_space_; 2733 CompactionSpace code_space_; 2734 }; 2735 2736 2737 // ----------------------------------------------------------------------------- 2738 // Old object space (includes the old space of objects and code space) 2739 2740 class OldSpace : public PagedSpace { 2741 public: 2742 // Creates an old space object. The constructor does not allocate pages 2743 // from OS. 2744 OldSpace(Heap* heap, AllocationSpace id, Executability executable) 2745 : PagedSpace(heap, id, executable) {} 2746 }; 2747 2748 2749 // For contiguous spaces, top should be in the space (or at the end) and limit 2750 // should be the end of the space. 2751 #define DCHECK_SEMISPACE_ALLOCATION_INFO(info, space) \ 2752 SLOW_DCHECK((space).page_low() <= (info).top() && \ 2753 (info).top() <= (space).page_high() && \ 2754 (info).limit() <= (space).page_high()) 2755 2756 2757 // ----------------------------------------------------------------------------- 2758 // Old space for all map objects 2759 2760 class MapSpace : public PagedSpace { 2761 public: 2762 // Creates a map space object. 2763 MapSpace(Heap* heap, AllocationSpace id) 2764 : PagedSpace(heap, id, NOT_EXECUTABLE) {} 2765 2766 int RoundSizeDownToObjectAlignment(int size) override { 2767 if (base::bits::IsPowerOfTwo32(Map::kSize)) { 2768 return RoundDown(size, Map::kSize); 2769 } else { 2770 return (size / Map::kSize) * Map::kSize; 2771 } 2772 } 2773 2774 #ifdef VERIFY_HEAP 2775 void VerifyObject(HeapObject* obj) override; 2776 #endif 2777 }; 2778 2779 2780 // ----------------------------------------------------------------------------- 2781 // Large objects ( > kMaxRegularHeapObjectSize ) are allocated and 2782 // managed by the large object space. A large object is allocated from OS 2783 // heap with extra padding bytes (Page::kPageSize + Page::kObjectStartOffset). 2784 // A large object always starts at Page::kObjectStartOffset to a page. 2785 // Large objects do not move during garbage collections. 2786 2787 class LargeObjectSpace : public Space { 2788 public: 2789 typedef LargePageIterator iterator; 2790 2791 LargeObjectSpace(Heap* heap, AllocationSpace id); 2792 virtual ~LargeObjectSpace(); 2793 2794 // Initializes internal data structures. 2795 bool SetUp(); 2796 2797 // Releases internal resources, frees objects in this space. 2798 void TearDown(); 2799 2800 static size_t ObjectSizeFor(size_t chunk_size) { 2801 if (chunk_size <= (Page::kPageSize + Page::kObjectStartOffset)) return 0; 2802 return chunk_size - Page::kPageSize - Page::kObjectStartOffset; 2803 } 2804 2805 // Shared implementation of AllocateRaw, AllocateRawCode and 2806 // AllocateRawFixedArray. 2807 MUST_USE_RESULT AllocationResult 2808 AllocateRaw(int object_size, Executability executable); 2809 2810 // Available bytes for objects in this space. 2811 inline size_t Available() override; 2812 2813 size_t Size() override { return size_; } 2814 2815 size_t SizeOfObjects() override { return objects_size_; } 2816 2817 // Approximate amount of physical memory committed for this space. 2818 size_t CommittedPhysicalMemory() override; 2819 2820 int PageCount() { return page_count_; } 2821 2822 // Finds an object for a given address, returns a Smi if it is not found. 2823 // The function iterates through all objects in this space, may be slow. 2824 Object* FindObject(Address a); 2825 2826 // Takes the chunk_map_mutex_ and calls FindPage after that. 2827 LargePage* FindPageThreadSafe(Address a); 2828 2829 // Finds a large object page containing the given address, returns NULL 2830 // if such a page doesn't exist. 2831 LargePage* FindPage(Address a); 2832 2833 // Clears the marking state of live objects. 2834 void ClearMarkingStateOfLiveObjects(); 2835 2836 // Frees unmarked objects. 2837 void FreeUnmarkedObjects(); 2838 2839 void InsertChunkMapEntries(LargePage* page); 2840 void RemoveChunkMapEntries(LargePage* page); 2841 void RemoveChunkMapEntries(LargePage* page, Address free_start); 2842 2843 // Checks whether a heap object is in this space; O(1). 2844 bool Contains(HeapObject* obj); 2845 // Checks whether an address is in the object area in this space. Iterates 2846 // all objects in the space. May be slow. 2847 bool ContainsSlow(Address addr) { return FindObject(addr)->IsHeapObject(); } 2848 2849 // Checks whether the space is empty. 2850 bool IsEmpty() { return first_page_ == NULL; } 2851 2852 void AdjustLiveBytes(int by) { objects_size_ += by; } 2853 2854 LargePage* first_page() { return first_page_; } 2855 2856 // Collect code statistics. 2857 void CollectCodeStatistics(); 2858 2859 iterator begin() { return iterator(first_page_); } 2860 iterator end() { return iterator(nullptr); } 2861 2862 std::unique_ptr<ObjectIterator> GetObjectIterator() override; 2863 2864 #ifdef VERIFY_HEAP 2865 virtual void Verify(); 2866 #endif 2867 2868 #ifdef DEBUG 2869 void Print() override; 2870 void ReportStatistics(); 2871 #endif 2872 2873 private: 2874 // The head of the linked list of large object chunks. 2875 LargePage* first_page_; 2876 size_t size_; // allocated bytes 2877 int page_count_; // number of chunks 2878 size_t objects_size_; // size of objects 2879 // The chunk_map_mutex_ has to be used when the chunk map is accessed 2880 // concurrently. 2881 base::Mutex chunk_map_mutex_; 2882 // Map MemoryChunk::kAlignment-aligned chunks to large pages covering them 2883 base::HashMap chunk_map_; 2884 2885 friend class LargeObjectIterator; 2886 }; 2887 2888 2889 class LargeObjectIterator : public ObjectIterator { 2890 public: 2891 explicit LargeObjectIterator(LargeObjectSpace* space); 2892 2893 HeapObject* Next() override; 2894 2895 private: 2896 LargePage* current_; 2897 }; 2898 2899 // Iterates over the chunks (pages and large object pages) that can contain 2900 // pointers to new space or to evacuation candidates. 2901 class MemoryChunkIterator BASE_EMBEDDED { 2902 public: 2903 inline explicit MemoryChunkIterator(Heap* heap); 2904 2905 // Return NULL when the iterator is done. 2906 inline MemoryChunk* next(); 2907 2908 private: 2909 enum State { 2910 kOldSpaceState, 2911 kMapState, 2912 kCodeState, 2913 kLargeObjectState, 2914 kFinishedState 2915 }; 2916 Heap* heap_; 2917 State state_; 2918 PageIterator old_iterator_; 2919 PageIterator code_iterator_; 2920 PageIterator map_iterator_; 2921 LargePageIterator lo_iterator_; 2922 }; 2923 2924 } // namespace internal 2925 } // namespace v8 2926 2927 #endif // V8_HEAP_SPACES_H_ 2928