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 #include "src/heap/spaces.h" 6 7 #include "src/base/bits.h" 8 #include "src/base/platform/platform.h" 9 #include "src/full-codegen/full-codegen.h" 10 #include "src/heap/slots-buffer.h" 11 #include "src/macro-assembler.h" 12 #include "src/msan.h" 13 #include "src/snapshot/snapshot.h" 14 15 namespace v8 { 16 namespace internal { 17 18 19 // ---------------------------------------------------------------------------- 20 // HeapObjectIterator 21 22 HeapObjectIterator::HeapObjectIterator(PagedSpace* space) { 23 // You can't actually iterate over the anchor page. It is not a real page, 24 // just an anchor for the double linked page list. Initialize as if we have 25 // reached the end of the anchor page, then the first iteration will move on 26 // to the first page. 27 Initialize(space, NULL, NULL, kAllPagesInSpace); 28 } 29 30 31 HeapObjectIterator::HeapObjectIterator(Page* page) { 32 Space* owner = page->owner(); 33 DCHECK(owner == page->heap()->old_space() || 34 owner == page->heap()->map_space() || 35 owner == page->heap()->code_space()); 36 Initialize(reinterpret_cast<PagedSpace*>(owner), page->area_start(), 37 page->area_end(), kOnePageOnly); 38 DCHECK(page->WasSwept() || page->SweepingCompleted()); 39 } 40 41 42 void HeapObjectIterator::Initialize(PagedSpace* space, Address cur, Address end, 43 HeapObjectIterator::PageMode mode) { 44 space_ = space; 45 cur_addr_ = cur; 46 cur_end_ = end; 47 page_mode_ = mode; 48 } 49 50 51 // We have hit the end of the page and should advance to the next block of 52 // objects. This happens at the end of the page. 53 bool HeapObjectIterator::AdvanceToNextPage() { 54 DCHECK(cur_addr_ == cur_end_); 55 if (page_mode_ == kOnePageOnly) return false; 56 Page* cur_page; 57 if (cur_addr_ == NULL) { 58 cur_page = space_->anchor(); 59 } else { 60 cur_page = Page::FromAddress(cur_addr_ - 1); 61 DCHECK(cur_addr_ == cur_page->area_end()); 62 } 63 cur_page = cur_page->next_page(); 64 if (cur_page == space_->anchor()) return false; 65 cur_page->heap()->mark_compact_collector()->SweepOrWaitUntilSweepingCompleted( 66 cur_page); 67 cur_addr_ = cur_page->area_start(); 68 cur_end_ = cur_page->area_end(); 69 DCHECK(cur_page->WasSwept() || cur_page->SweepingCompleted()); 70 return true; 71 } 72 73 74 // ----------------------------------------------------------------------------- 75 // CodeRange 76 77 78 CodeRange::CodeRange(Isolate* isolate) 79 : isolate_(isolate), 80 code_range_(NULL), 81 free_list_(0), 82 allocation_list_(0), 83 current_allocation_block_index_(0) {} 84 85 86 bool CodeRange::SetUp(size_t requested) { 87 DCHECK(code_range_ == NULL); 88 89 if (requested == 0) { 90 // When a target requires the code range feature, we put all code objects 91 // in a kMaximalCodeRangeSize range of virtual address space, so that 92 // they can call each other with near calls. 93 if (kRequiresCodeRange) { 94 requested = kMaximalCodeRangeSize; 95 } else { 96 return true; 97 } 98 } 99 100 if (requested <= kMinimumCodeRangeSize) { 101 requested = kMinimumCodeRangeSize; 102 } 103 104 DCHECK(!kRequiresCodeRange || requested <= kMaximalCodeRangeSize); 105 #ifdef V8_TARGET_ARCH_MIPS64 106 // To use pseudo-relative jumps such as j/jal instructions which have 28-bit 107 // encoded immediate, the addresses have to be in range of 256Mb aligned 108 // region. 109 code_range_ = new base::VirtualMemory(requested, kMaximalCodeRangeSize); 110 #else 111 code_range_ = new base::VirtualMemory(requested); 112 #endif 113 CHECK(code_range_ != NULL); 114 if (!code_range_->IsReserved()) { 115 delete code_range_; 116 code_range_ = NULL; 117 return false; 118 } 119 120 // We are sure that we have mapped a block of requested addresses. 121 DCHECK(code_range_->size() == requested); 122 Address base = reinterpret_cast<Address>(code_range_->address()); 123 124 // On some platforms, specifically Win64, we need to reserve some pages at 125 // the beginning of an executable space. 126 if (kReservedCodeRangePages) { 127 if (!code_range_->Commit( 128 base, kReservedCodeRangePages * base::OS::CommitPageSize(), true)) { 129 delete code_range_; 130 code_range_ = NULL; 131 return false; 132 } 133 base += kReservedCodeRangePages * base::OS::CommitPageSize(); 134 } 135 Address aligned_base = RoundUp(base, MemoryChunk::kAlignment); 136 size_t size = code_range_->size() - (aligned_base - base) - 137 kReservedCodeRangePages * base::OS::CommitPageSize(); 138 allocation_list_.Add(FreeBlock(aligned_base, size)); 139 current_allocation_block_index_ = 0; 140 141 LOG(isolate_, NewEvent("CodeRange", code_range_->address(), requested)); 142 return true; 143 } 144 145 146 int CodeRange::CompareFreeBlockAddress(const FreeBlock* left, 147 const FreeBlock* right) { 148 // The entire point of CodeRange is that the difference between two 149 // addresses in the range can be represented as a signed 32-bit int, 150 // so the cast is semantically correct. 151 return static_cast<int>(left->start - right->start); 152 } 153 154 155 bool CodeRange::GetNextAllocationBlock(size_t requested) { 156 for (current_allocation_block_index_++; 157 current_allocation_block_index_ < allocation_list_.length(); 158 current_allocation_block_index_++) { 159 if (requested <= allocation_list_[current_allocation_block_index_].size) { 160 return true; // Found a large enough allocation block. 161 } 162 } 163 164 // Sort and merge the free blocks on the free list and the allocation list. 165 free_list_.AddAll(allocation_list_); 166 allocation_list_.Clear(); 167 free_list_.Sort(&CompareFreeBlockAddress); 168 for (int i = 0; i < free_list_.length();) { 169 FreeBlock merged = free_list_[i]; 170 i++; 171 // Add adjacent free blocks to the current merged block. 172 while (i < free_list_.length() && 173 free_list_[i].start == merged.start + merged.size) { 174 merged.size += free_list_[i].size; 175 i++; 176 } 177 if (merged.size > 0) { 178 allocation_list_.Add(merged); 179 } 180 } 181 free_list_.Clear(); 182 183 for (current_allocation_block_index_ = 0; 184 current_allocation_block_index_ < allocation_list_.length(); 185 current_allocation_block_index_++) { 186 if (requested <= allocation_list_[current_allocation_block_index_].size) { 187 return true; // Found a large enough allocation block. 188 } 189 } 190 current_allocation_block_index_ = 0; 191 // Code range is full or too fragmented. 192 return false; 193 } 194 195 196 Address CodeRange::AllocateRawMemory(const size_t requested_size, 197 const size_t commit_size, 198 size_t* allocated) { 199 // request_size includes guards while committed_size does not. Make sure 200 // callers know about the invariant. 201 CHECK_LE(commit_size, 202 requested_size - 2 * MemoryAllocator::CodePageGuardSize()); 203 FreeBlock current; 204 if (!ReserveBlock(requested_size, ¤t)) { 205 *allocated = 0; 206 return NULL; 207 } 208 *allocated = current.size; 209 DCHECK(*allocated <= current.size); 210 DCHECK(IsAddressAligned(current.start, MemoryChunk::kAlignment)); 211 if (!isolate_->memory_allocator()->CommitExecutableMemory( 212 code_range_, current.start, commit_size, *allocated)) { 213 *allocated = 0; 214 ReleaseBlock(¤t); 215 return NULL; 216 } 217 return current.start; 218 } 219 220 221 bool CodeRange::CommitRawMemory(Address start, size_t length) { 222 return isolate_->memory_allocator()->CommitMemory(start, length, EXECUTABLE); 223 } 224 225 226 bool CodeRange::UncommitRawMemory(Address start, size_t length) { 227 return code_range_->Uncommit(start, length); 228 } 229 230 231 void CodeRange::FreeRawMemory(Address address, size_t length) { 232 DCHECK(IsAddressAligned(address, MemoryChunk::kAlignment)); 233 base::LockGuard<base::Mutex> guard(&code_range_mutex_); 234 free_list_.Add(FreeBlock(address, length)); 235 code_range_->Uncommit(address, length); 236 } 237 238 239 void CodeRange::TearDown() { 240 delete code_range_; // Frees all memory in the virtual memory range. 241 code_range_ = NULL; 242 base::LockGuard<base::Mutex> guard(&code_range_mutex_); 243 free_list_.Free(); 244 allocation_list_.Free(); 245 } 246 247 248 bool CodeRange::ReserveBlock(const size_t requested_size, FreeBlock* block) { 249 base::LockGuard<base::Mutex> guard(&code_range_mutex_); 250 DCHECK(allocation_list_.length() == 0 || 251 current_allocation_block_index_ < allocation_list_.length()); 252 if (allocation_list_.length() == 0 || 253 requested_size > allocation_list_[current_allocation_block_index_].size) { 254 // Find an allocation block large enough. 255 if (!GetNextAllocationBlock(requested_size)) return false; 256 } 257 // Commit the requested memory at the start of the current allocation block. 258 size_t aligned_requested = RoundUp(requested_size, MemoryChunk::kAlignment); 259 *block = allocation_list_[current_allocation_block_index_]; 260 // Don't leave a small free block, useless for a large object or chunk. 261 if (aligned_requested < (block->size - Page::kPageSize)) { 262 block->size = aligned_requested; 263 } 264 DCHECK(IsAddressAligned(block->start, MemoryChunk::kAlignment)); 265 allocation_list_[current_allocation_block_index_].start += block->size; 266 allocation_list_[current_allocation_block_index_].size -= block->size; 267 return true; 268 } 269 270 271 void CodeRange::ReleaseBlock(const FreeBlock* block) { 272 base::LockGuard<base::Mutex> guard(&code_range_mutex_); 273 free_list_.Add(*block); 274 } 275 276 277 // ----------------------------------------------------------------------------- 278 // MemoryAllocator 279 // 280 281 MemoryAllocator::MemoryAllocator(Isolate* isolate) 282 : isolate_(isolate), 283 capacity_(0), 284 capacity_executable_(0), 285 size_(0), 286 size_executable_(0), 287 lowest_ever_allocated_(reinterpret_cast<void*>(-1)), 288 highest_ever_allocated_(reinterpret_cast<void*>(0)) {} 289 290 291 bool MemoryAllocator::SetUp(intptr_t capacity, intptr_t capacity_executable) { 292 capacity_ = RoundUp(capacity, Page::kPageSize); 293 capacity_executable_ = RoundUp(capacity_executable, Page::kPageSize); 294 DCHECK_GE(capacity_, capacity_executable_); 295 296 size_ = 0; 297 size_executable_ = 0; 298 299 return true; 300 } 301 302 303 void MemoryAllocator::TearDown() { 304 // Check that spaces were torn down before MemoryAllocator. 305 DCHECK(size_.Value() == 0); 306 // TODO(gc) this will be true again when we fix FreeMemory. 307 // DCHECK(size_executable_ == 0); 308 capacity_ = 0; 309 capacity_executable_ = 0; 310 } 311 312 313 bool MemoryAllocator::CommitMemory(Address base, size_t size, 314 Executability executable) { 315 if (!base::VirtualMemory::CommitRegion(base, size, 316 executable == EXECUTABLE)) { 317 return false; 318 } 319 UpdateAllocatedSpaceLimits(base, base + size); 320 return true; 321 } 322 323 324 void MemoryAllocator::FreeNewSpaceMemory(Address addr, 325 base::VirtualMemory* reservation, 326 Executability executable) { 327 LOG(isolate_, DeleteEvent("NewSpace", addr)); 328 329 DCHECK(reservation->IsReserved()); 330 const intptr_t size = static_cast<intptr_t>(reservation->size()); 331 DCHECK(size_.Value() >= size); 332 size_.Increment(-size); 333 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 334 FreeMemory(reservation, NOT_EXECUTABLE); 335 } 336 337 338 void MemoryAllocator::FreeMemory(base::VirtualMemory* reservation, 339 Executability executable) { 340 // TODO(gc) make code_range part of memory allocator? 341 // Code which is part of the code-range does not have its own VirtualMemory. 342 DCHECK(isolate_->code_range() == NULL || 343 !isolate_->code_range()->contains( 344 static_cast<Address>(reservation->address()))); 345 DCHECK(executable == NOT_EXECUTABLE || isolate_->code_range() == NULL || 346 !isolate_->code_range()->valid() || 347 reservation->size() <= Page::kPageSize); 348 349 reservation->Release(); 350 } 351 352 353 void MemoryAllocator::FreeMemory(Address base, size_t size, 354 Executability executable) { 355 // TODO(gc) make code_range part of memory allocator? 356 if (isolate_->code_range() != NULL && 357 isolate_->code_range()->contains(static_cast<Address>(base))) { 358 DCHECK(executable == EXECUTABLE); 359 isolate_->code_range()->FreeRawMemory(base, size); 360 } else { 361 DCHECK(executable == NOT_EXECUTABLE || isolate_->code_range() == NULL || 362 !isolate_->code_range()->valid()); 363 bool result = base::VirtualMemory::ReleaseRegion(base, size); 364 USE(result); 365 DCHECK(result); 366 } 367 } 368 369 370 Address MemoryAllocator::ReserveAlignedMemory(size_t size, size_t alignment, 371 base::VirtualMemory* controller) { 372 base::VirtualMemory reservation(size, alignment); 373 374 if (!reservation.IsReserved()) return NULL; 375 size_.Increment(static_cast<intptr_t>(reservation.size())); 376 Address base = 377 RoundUp(static_cast<Address>(reservation.address()), alignment); 378 controller->TakeControl(&reservation); 379 return base; 380 } 381 382 383 Address MemoryAllocator::AllocateAlignedMemory( 384 size_t reserve_size, size_t commit_size, size_t alignment, 385 Executability executable, base::VirtualMemory* controller) { 386 DCHECK(commit_size <= reserve_size); 387 base::VirtualMemory reservation; 388 Address base = ReserveAlignedMemory(reserve_size, alignment, &reservation); 389 if (base == NULL) return NULL; 390 391 if (executable == EXECUTABLE) { 392 if (!CommitExecutableMemory(&reservation, base, commit_size, 393 reserve_size)) { 394 base = NULL; 395 } 396 } else { 397 if (reservation.Commit(base, commit_size, false)) { 398 UpdateAllocatedSpaceLimits(base, base + commit_size); 399 } else { 400 base = NULL; 401 } 402 } 403 404 if (base == NULL) { 405 // Failed to commit the body. Release the mapping and any partially 406 // commited regions inside it. 407 reservation.Release(); 408 return NULL; 409 } 410 411 controller->TakeControl(&reservation); 412 return base; 413 } 414 415 416 void Page::InitializeAsAnchor(PagedSpace* owner) { 417 set_owner(owner); 418 set_prev_page(this); 419 set_next_page(this); 420 } 421 422 423 NewSpacePage* NewSpacePage::Initialize(Heap* heap, Address start, 424 SemiSpace* semi_space) { 425 Address area_start = start + NewSpacePage::kObjectStartOffset; 426 Address area_end = start + Page::kPageSize; 427 428 MemoryChunk* chunk = 429 MemoryChunk::Initialize(heap, start, Page::kPageSize, area_start, 430 area_end, NOT_EXECUTABLE, semi_space); 431 chunk->initialize_scan_on_scavenge(true); 432 bool in_to_space = (semi_space->id() != kFromSpace); 433 chunk->SetFlag(in_to_space ? MemoryChunk::IN_TO_SPACE 434 : MemoryChunk::IN_FROM_SPACE); 435 DCHECK(!chunk->IsFlagSet(in_to_space ? MemoryChunk::IN_FROM_SPACE 436 : MemoryChunk::IN_TO_SPACE)); 437 NewSpacePage* page = static_cast<NewSpacePage*>(chunk); 438 heap->incremental_marking()->SetNewSpacePageFlags(page); 439 return page; 440 } 441 442 443 void NewSpacePage::InitializeAsAnchor(SemiSpace* semi_space) { 444 set_owner(semi_space); 445 set_next_chunk(this); 446 set_prev_chunk(this); 447 // Flags marks this invalid page as not being in new-space. 448 // All real new-space pages will be in new-space. 449 SetFlags(0, ~0); 450 } 451 452 453 MemoryChunk* MemoryChunk::Initialize(Heap* heap, Address base, size_t size, 454 Address area_start, Address area_end, 455 Executability executable, Space* owner) { 456 MemoryChunk* chunk = FromAddress(base); 457 458 DCHECK(base == chunk->address()); 459 460 chunk->heap_ = heap; 461 chunk->size_ = size; 462 chunk->area_start_ = area_start; 463 chunk->area_end_ = area_end; 464 chunk->flags_ = 0; 465 chunk->set_owner(owner); 466 chunk->InitializeReservedMemory(); 467 chunk->slots_buffer_ = NULL; 468 chunk->skip_list_ = NULL; 469 chunk->write_barrier_counter_ = kWriteBarrierCounterGranularity; 470 chunk->progress_bar_ = 0; 471 chunk->high_water_mark_.SetValue(static_cast<intptr_t>(area_start - base)); 472 chunk->parallel_sweeping_state().SetValue(kSweepingDone); 473 chunk->parallel_compaction_state().SetValue(kCompactingDone); 474 chunk->mutex_ = NULL; 475 chunk->available_in_small_free_list_ = 0; 476 chunk->available_in_medium_free_list_ = 0; 477 chunk->available_in_large_free_list_ = 0; 478 chunk->available_in_huge_free_list_ = 0; 479 chunk->non_available_small_blocks_ = 0; 480 chunk->ResetLiveBytes(); 481 Bitmap::Clear(chunk); 482 chunk->initialize_scan_on_scavenge(false); 483 chunk->SetFlag(WAS_SWEPT); 484 chunk->set_next_chunk(nullptr); 485 chunk->set_prev_chunk(nullptr); 486 487 DCHECK(OFFSET_OF(MemoryChunk, flags_) == kFlagsOffset); 488 DCHECK(OFFSET_OF(MemoryChunk, live_byte_count_) == kLiveBytesOffset); 489 490 if (executable == EXECUTABLE) { 491 chunk->SetFlag(IS_EXECUTABLE); 492 } 493 494 return chunk; 495 } 496 497 498 // Commit MemoryChunk area to the requested size. 499 bool MemoryChunk::CommitArea(size_t requested) { 500 size_t guard_size = 501 IsFlagSet(IS_EXECUTABLE) ? MemoryAllocator::CodePageGuardSize() : 0; 502 size_t header_size = area_start() - address() - guard_size; 503 size_t commit_size = 504 RoundUp(header_size + requested, base::OS::CommitPageSize()); 505 size_t committed_size = RoundUp(header_size + (area_end() - area_start()), 506 base::OS::CommitPageSize()); 507 508 if (commit_size > committed_size) { 509 // Commit size should be less or equal than the reserved size. 510 DCHECK(commit_size <= size() - 2 * guard_size); 511 // Append the committed area. 512 Address start = address() + committed_size + guard_size; 513 size_t length = commit_size - committed_size; 514 if (reservation_.IsReserved()) { 515 Executability executable = 516 IsFlagSet(IS_EXECUTABLE) ? EXECUTABLE : NOT_EXECUTABLE; 517 if (!heap()->isolate()->memory_allocator()->CommitMemory(start, length, 518 executable)) { 519 return false; 520 } 521 } else { 522 CodeRange* code_range = heap_->isolate()->code_range(); 523 DCHECK(code_range != NULL && code_range->valid() && 524 IsFlagSet(IS_EXECUTABLE)); 525 if (!code_range->CommitRawMemory(start, length)) return false; 526 } 527 528 if (Heap::ShouldZapGarbage()) { 529 heap_->isolate()->memory_allocator()->ZapBlock(start, length); 530 } 531 } else if (commit_size < committed_size) { 532 DCHECK(commit_size > 0); 533 // Shrink the committed area. 534 size_t length = committed_size - commit_size; 535 Address start = address() + committed_size + guard_size - length; 536 if (reservation_.IsReserved()) { 537 if (!reservation_.Uncommit(start, length)) return false; 538 } else { 539 CodeRange* code_range = heap_->isolate()->code_range(); 540 DCHECK(code_range != NULL && code_range->valid() && 541 IsFlagSet(IS_EXECUTABLE)); 542 if (!code_range->UncommitRawMemory(start, length)) return false; 543 } 544 } 545 546 area_end_ = area_start_ + requested; 547 return true; 548 } 549 550 551 void MemoryChunk::InsertAfter(MemoryChunk* other) { 552 MemoryChunk* other_next = other->next_chunk(); 553 554 set_next_chunk(other_next); 555 set_prev_chunk(other); 556 other_next->set_prev_chunk(this); 557 other->set_next_chunk(this); 558 } 559 560 561 void MemoryChunk::Unlink() { 562 MemoryChunk* next_element = next_chunk(); 563 MemoryChunk* prev_element = prev_chunk(); 564 next_element->set_prev_chunk(prev_element); 565 prev_element->set_next_chunk(next_element); 566 set_prev_chunk(NULL); 567 set_next_chunk(NULL); 568 } 569 570 571 MemoryChunk* MemoryAllocator::AllocateChunk(intptr_t reserve_area_size, 572 intptr_t commit_area_size, 573 Executability executable, 574 Space* owner) { 575 DCHECK(commit_area_size <= reserve_area_size); 576 577 size_t chunk_size; 578 Heap* heap = isolate_->heap(); 579 Address base = NULL; 580 base::VirtualMemory reservation; 581 Address area_start = NULL; 582 Address area_end = NULL; 583 584 // 585 // MemoryChunk layout: 586 // 587 // Executable 588 // +----------------------------+<- base aligned with MemoryChunk::kAlignment 589 // | Header | 590 // +----------------------------+<- base + CodePageGuardStartOffset 591 // | Guard | 592 // +----------------------------+<- area_start_ 593 // | Area | 594 // +----------------------------+<- area_end_ (area_start + commit_area_size) 595 // | Committed but not used | 596 // +----------------------------+<- aligned at OS page boundary 597 // | Reserved but not committed | 598 // +----------------------------+<- aligned at OS page boundary 599 // | Guard | 600 // +----------------------------+<- base + chunk_size 601 // 602 // Non-executable 603 // +----------------------------+<- base aligned with MemoryChunk::kAlignment 604 // | Header | 605 // +----------------------------+<- area_start_ (base + kObjectStartOffset) 606 // | Area | 607 // +----------------------------+<- area_end_ (area_start + commit_area_size) 608 // | Committed but not used | 609 // +----------------------------+<- aligned at OS page boundary 610 // | Reserved but not committed | 611 // +----------------------------+<- base + chunk_size 612 // 613 614 if (executable == EXECUTABLE) { 615 chunk_size = RoundUp(CodePageAreaStartOffset() + reserve_area_size, 616 base::OS::CommitPageSize()) + 617 CodePageGuardSize(); 618 619 // Check executable memory limit. 620 if ((size_executable_.Value() + static_cast<intptr_t>(chunk_size)) > 621 capacity_executable_) { 622 LOG(isolate_, StringEvent("MemoryAllocator::AllocateRawMemory", 623 "V8 Executable Allocation capacity exceeded")); 624 return NULL; 625 } 626 627 // Size of header (not executable) plus area (executable). 628 size_t commit_size = RoundUp(CodePageGuardStartOffset() + commit_area_size, 629 base::OS::CommitPageSize()); 630 // Allocate executable memory either from code range or from the 631 // OS. 632 #ifdef V8_TARGET_ARCH_MIPS64 633 // Use code range only for large object space on mips64 to keep address 634 // range within 256-MB memory region. 635 if (isolate_->code_range() != NULL && isolate_->code_range()->valid() && 636 reserve_area_size > CodePageAreaSize()) { 637 #else 638 if (isolate_->code_range() != NULL && isolate_->code_range()->valid()) { 639 #endif 640 base = isolate_->code_range()->AllocateRawMemory(chunk_size, commit_size, 641 &chunk_size); 642 DCHECK( 643 IsAligned(reinterpret_cast<intptr_t>(base), MemoryChunk::kAlignment)); 644 if (base == NULL) return NULL; 645 size_.Increment(static_cast<intptr_t>(chunk_size)); 646 // Update executable memory size. 647 size_executable_.Increment(static_cast<intptr_t>(chunk_size)); 648 } else { 649 base = AllocateAlignedMemory(chunk_size, commit_size, 650 MemoryChunk::kAlignment, executable, 651 &reservation); 652 if (base == NULL) return NULL; 653 // Update executable memory size. 654 size_executable_.Increment(static_cast<intptr_t>(reservation.size())); 655 } 656 657 if (Heap::ShouldZapGarbage()) { 658 ZapBlock(base, CodePageGuardStartOffset()); 659 ZapBlock(base + CodePageAreaStartOffset(), commit_area_size); 660 } 661 662 area_start = base + CodePageAreaStartOffset(); 663 area_end = area_start + commit_area_size; 664 } else { 665 chunk_size = RoundUp(MemoryChunk::kObjectStartOffset + reserve_area_size, 666 base::OS::CommitPageSize()); 667 size_t commit_size = 668 RoundUp(MemoryChunk::kObjectStartOffset + commit_area_size, 669 base::OS::CommitPageSize()); 670 base = 671 AllocateAlignedMemory(chunk_size, commit_size, MemoryChunk::kAlignment, 672 executable, &reservation); 673 674 if (base == NULL) return NULL; 675 676 if (Heap::ShouldZapGarbage()) { 677 ZapBlock(base, Page::kObjectStartOffset + commit_area_size); 678 } 679 680 area_start = base + Page::kObjectStartOffset; 681 area_end = area_start + commit_area_size; 682 } 683 684 // Use chunk_size for statistics and callbacks because we assume that they 685 // treat reserved but not-yet committed memory regions of chunks as allocated. 686 isolate_->counters()->memory_allocated()->Increment( 687 static_cast<int>(chunk_size)); 688 689 LOG(isolate_, NewEvent("MemoryChunk", base, chunk_size)); 690 if (owner != NULL) { 691 ObjectSpace space = static_cast<ObjectSpace>(1 << owner->identity()); 692 PerformAllocationCallback(space, kAllocationActionAllocate, chunk_size); 693 } 694 695 MemoryChunk* result = MemoryChunk::Initialize( 696 heap, base, chunk_size, area_start, area_end, executable, owner); 697 result->set_reserved_memory(&reservation); 698 return result; 699 } 700 701 702 void Page::ResetFreeListStatistics() { 703 non_available_small_blocks_ = 0; 704 available_in_small_free_list_ = 0; 705 available_in_medium_free_list_ = 0; 706 available_in_large_free_list_ = 0; 707 available_in_huge_free_list_ = 0; 708 } 709 710 711 Page* MemoryAllocator::AllocatePage(intptr_t size, PagedSpace* owner, 712 Executability executable) { 713 MemoryChunk* chunk = AllocateChunk(size, size, executable, owner); 714 if (chunk == NULL) return NULL; 715 return Page::Initialize(isolate_->heap(), chunk, executable, owner); 716 } 717 718 719 LargePage* MemoryAllocator::AllocateLargePage(intptr_t object_size, 720 Space* owner, 721 Executability executable) { 722 MemoryChunk* chunk = 723 AllocateChunk(object_size, object_size, executable, owner); 724 if (chunk == NULL) return NULL; 725 return LargePage::Initialize(isolate_->heap(), chunk); 726 } 727 728 729 void MemoryAllocator::PreFreeMemory(MemoryChunk* chunk) { 730 DCHECK(!chunk->IsFlagSet(MemoryChunk::PRE_FREED)); 731 LOG(isolate_, DeleteEvent("MemoryChunk", chunk)); 732 if (chunk->owner() != NULL) { 733 ObjectSpace space = 734 static_cast<ObjectSpace>(1 << chunk->owner()->identity()); 735 PerformAllocationCallback(space, kAllocationActionFree, chunk->size()); 736 } 737 738 isolate_->heap()->RememberUnmappedPage(reinterpret_cast<Address>(chunk), 739 chunk->IsEvacuationCandidate()); 740 741 intptr_t size; 742 base::VirtualMemory* reservation = chunk->reserved_memory(); 743 if (reservation->IsReserved()) { 744 size = static_cast<intptr_t>(reservation->size()); 745 } else { 746 size = static_cast<intptr_t>(chunk->size()); 747 } 748 DCHECK(size_.Value() >= size); 749 size_.Increment(-size); 750 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 751 752 if (chunk->executable() == EXECUTABLE) { 753 DCHECK(size_executable_.Value() >= size); 754 size_executable_.Increment(-size); 755 } 756 757 chunk->SetFlag(MemoryChunk::PRE_FREED); 758 } 759 760 761 void MemoryAllocator::PerformFreeMemory(MemoryChunk* chunk) { 762 DCHECK(chunk->IsFlagSet(MemoryChunk::PRE_FREED)); 763 chunk->ReleaseAllocatedMemory(); 764 765 base::VirtualMemory* reservation = chunk->reserved_memory(); 766 if (reservation->IsReserved()) { 767 FreeMemory(reservation, chunk->executable()); 768 } else { 769 FreeMemory(chunk->address(), chunk->size(), chunk->executable()); 770 } 771 } 772 773 774 void MemoryAllocator::Free(MemoryChunk* chunk) { 775 PreFreeMemory(chunk); 776 PerformFreeMemory(chunk); 777 } 778 779 780 bool MemoryAllocator::CommitBlock(Address start, size_t size, 781 Executability executable) { 782 if (!CommitMemory(start, size, executable)) return false; 783 784 if (Heap::ShouldZapGarbage()) { 785 ZapBlock(start, size); 786 } 787 788 isolate_->counters()->memory_allocated()->Increment(static_cast<int>(size)); 789 return true; 790 } 791 792 793 bool MemoryAllocator::UncommitBlock(Address start, size_t size) { 794 if (!base::VirtualMemory::UncommitRegion(start, size)) return false; 795 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 796 return true; 797 } 798 799 800 void MemoryAllocator::ZapBlock(Address start, size_t size) { 801 for (size_t s = 0; s + kPointerSize <= size; s += kPointerSize) { 802 Memory::Address_at(start + s) = kZapValue; 803 } 804 } 805 806 807 void MemoryAllocator::PerformAllocationCallback(ObjectSpace space, 808 AllocationAction action, 809 size_t size) { 810 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 811 MemoryAllocationCallbackRegistration registration = 812 memory_allocation_callbacks_[i]; 813 if ((registration.space & space) == space && 814 (registration.action & action) == action) 815 registration.callback(space, action, static_cast<int>(size)); 816 } 817 } 818 819 820 bool MemoryAllocator::MemoryAllocationCallbackRegistered( 821 MemoryAllocationCallback callback) { 822 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 823 if (memory_allocation_callbacks_[i].callback == callback) return true; 824 } 825 return false; 826 } 827 828 829 void MemoryAllocator::AddMemoryAllocationCallback( 830 MemoryAllocationCallback callback, ObjectSpace space, 831 AllocationAction action) { 832 DCHECK(callback != NULL); 833 MemoryAllocationCallbackRegistration registration(callback, space, action); 834 DCHECK(!MemoryAllocator::MemoryAllocationCallbackRegistered(callback)); 835 return memory_allocation_callbacks_.Add(registration); 836 } 837 838 839 void MemoryAllocator::RemoveMemoryAllocationCallback( 840 MemoryAllocationCallback callback) { 841 DCHECK(callback != NULL); 842 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 843 if (memory_allocation_callbacks_[i].callback == callback) { 844 memory_allocation_callbacks_.Remove(i); 845 return; 846 } 847 } 848 UNREACHABLE(); 849 } 850 851 852 #ifdef DEBUG 853 void MemoryAllocator::ReportStatistics() { 854 intptr_t size = Size(); 855 float pct = static_cast<float>(capacity_ - size) / capacity_; 856 PrintF(" capacity: %" V8_PTR_PREFIX 857 "d" 858 ", used: %" V8_PTR_PREFIX 859 "d" 860 ", available: %%%d\n\n", 861 capacity_, size, static_cast<int>(pct * 100)); 862 } 863 #endif 864 865 866 int MemoryAllocator::CodePageGuardStartOffset() { 867 // We are guarding code pages: the first OS page after the header 868 // will be protected as non-writable. 869 return RoundUp(Page::kObjectStartOffset, base::OS::CommitPageSize()); 870 } 871 872 873 int MemoryAllocator::CodePageGuardSize() { 874 return static_cast<int>(base::OS::CommitPageSize()); 875 } 876 877 878 int MemoryAllocator::CodePageAreaStartOffset() { 879 // We are guarding code pages: the first OS page after the header 880 // will be protected as non-writable. 881 return CodePageGuardStartOffset() + CodePageGuardSize(); 882 } 883 884 885 int MemoryAllocator::CodePageAreaEndOffset() { 886 // We are guarding code pages: the last OS page will be protected as 887 // non-writable. 888 return Page::kPageSize - static_cast<int>(base::OS::CommitPageSize()); 889 } 890 891 892 bool MemoryAllocator::CommitExecutableMemory(base::VirtualMemory* vm, 893 Address start, size_t commit_size, 894 size_t reserved_size) { 895 // Commit page header (not executable). 896 Address header = start; 897 size_t header_size = CodePageGuardStartOffset(); 898 if (vm->Commit(header, header_size, false)) { 899 // Create guard page after the header. 900 if (vm->Guard(start + CodePageGuardStartOffset())) { 901 // Commit page body (executable). 902 Address body = start + CodePageAreaStartOffset(); 903 size_t body_size = commit_size - CodePageGuardStartOffset(); 904 if (vm->Commit(body, body_size, true)) { 905 // Create guard page before the end. 906 if (vm->Guard(start + reserved_size - CodePageGuardSize())) { 907 UpdateAllocatedSpaceLimits(start, start + CodePageAreaStartOffset() + 908 commit_size - 909 CodePageGuardStartOffset()); 910 return true; 911 } 912 vm->Uncommit(body, body_size); 913 } 914 } 915 vm->Uncommit(header, header_size); 916 } 917 return false; 918 } 919 920 921 // ----------------------------------------------------------------------------- 922 // MemoryChunk implementation 923 924 void MemoryChunk::IncrementLiveBytesFromMutator(HeapObject* object, int by) { 925 MemoryChunk* chunk = MemoryChunk::FromAddress(object->address()); 926 if (!chunk->InNewSpace() && !static_cast<Page*>(chunk)->WasSwept()) { 927 static_cast<PagedSpace*>(chunk->owner())->Allocate(by); 928 } 929 chunk->IncrementLiveBytes(by); 930 } 931 932 933 void MemoryChunk::ReleaseAllocatedMemory() { 934 delete slots_buffer_; 935 delete skip_list_; 936 delete mutex_; 937 } 938 939 940 // ----------------------------------------------------------------------------- 941 // PagedSpace implementation 942 943 STATIC_ASSERT(static_cast<ObjectSpace>(1 << AllocationSpace::NEW_SPACE) == 944 ObjectSpace::kObjectSpaceNewSpace); 945 STATIC_ASSERT(static_cast<ObjectSpace>(1 << AllocationSpace::OLD_SPACE) == 946 ObjectSpace::kObjectSpaceOldSpace); 947 STATIC_ASSERT(static_cast<ObjectSpace>(1 << AllocationSpace::CODE_SPACE) == 948 ObjectSpace::kObjectSpaceCodeSpace); 949 STATIC_ASSERT(static_cast<ObjectSpace>(1 << AllocationSpace::MAP_SPACE) == 950 ObjectSpace::kObjectSpaceMapSpace); 951 952 953 PagedSpace::PagedSpace(Heap* heap, AllocationSpace space, 954 Executability executable) 955 : Space(heap, space, executable), 956 free_list_(this), 957 end_of_unswept_pages_(NULL) { 958 area_size_ = MemoryAllocator::PageAreaSize(space); 959 accounting_stats_.Clear(); 960 961 allocation_info_.Reset(nullptr, nullptr); 962 963 anchor_.InitializeAsAnchor(this); 964 } 965 966 967 bool PagedSpace::SetUp() { return true; } 968 969 970 bool PagedSpace::HasBeenSetUp() { return true; } 971 972 973 void PagedSpace::TearDown() { 974 PageIterator iterator(this); 975 while (iterator.has_next()) { 976 heap()->isolate()->memory_allocator()->Free(iterator.next()); 977 } 978 anchor_.set_next_page(&anchor_); 979 anchor_.set_prev_page(&anchor_); 980 accounting_stats_.Clear(); 981 } 982 983 984 void PagedSpace::AddMemory(Address start, intptr_t size) { 985 accounting_stats_.ExpandSpace(static_cast<int>(size)); 986 Free(start, static_cast<int>(size)); 987 } 988 989 990 FreeSpace* PagedSpace::TryRemoveMemory(intptr_t size_in_bytes) { 991 FreeSpace* free_space = free_list()->TryRemoveMemory(size_in_bytes); 992 if (free_space != nullptr) { 993 accounting_stats_.DecreaseCapacity(free_space->size()); 994 } 995 return free_space; 996 } 997 998 999 void PagedSpace::DivideUponCompactionSpaces(CompactionSpaceCollection** other, 1000 int num, intptr_t limit) { 1001 DCHECK_GT(num, 0); 1002 DCHECK(other != nullptr); 1003 1004 if (limit == 0) limit = std::numeric_limits<intptr_t>::max(); 1005 1006 EmptyAllocationInfo(); 1007 1008 bool memory_available = true; 1009 bool spaces_need_memory = true; 1010 FreeSpace* node = nullptr; 1011 CompactionSpace* current_space = nullptr; 1012 // Iterate over spaces and memory as long as we have memory and there are 1013 // spaces in need of some. 1014 while (memory_available && spaces_need_memory) { 1015 spaces_need_memory = false; 1016 // Round-robin over all spaces. 1017 for (int i = 0; i < num; i++) { 1018 current_space = other[i]->Get(identity()); 1019 if (current_space->free_list()->Available() < limit) { 1020 // Space has not reached its limit. Try to get some memory. 1021 spaces_need_memory = true; 1022 node = TryRemoveMemory(limit - current_space->free_list()->Available()); 1023 if (node != nullptr) { 1024 CHECK(current_space->identity() == identity()); 1025 current_space->AddMemory(node->address(), node->size()); 1026 } else { 1027 memory_available = false; 1028 break; 1029 } 1030 } 1031 } 1032 } 1033 } 1034 1035 1036 void PagedSpace::RefillFreeList() { 1037 MarkCompactCollector* collector = heap()->mark_compact_collector(); 1038 FreeList* free_list = nullptr; 1039 if (this == heap()->old_space()) { 1040 free_list = collector->free_list_old_space().get(); 1041 } else if (this == heap()->code_space()) { 1042 free_list = collector->free_list_code_space().get(); 1043 } else if (this == heap()->map_space()) { 1044 free_list = collector->free_list_map_space().get(); 1045 } else { 1046 // Any PagedSpace might invoke RefillFreeList. We filter all but our old 1047 // generation spaces out. 1048 return; 1049 } 1050 DCHECK(free_list != nullptr); 1051 intptr_t added = free_list_.Concatenate(free_list); 1052 accounting_stats_.IncreaseCapacity(added); 1053 } 1054 1055 1056 void CompactionSpace::RefillFreeList() { 1057 MarkCompactCollector* collector = heap()->mark_compact_collector(); 1058 FreeList* free_list = nullptr; 1059 if (identity() == OLD_SPACE) { 1060 free_list = collector->free_list_old_space().get(); 1061 } else if (identity() == CODE_SPACE) { 1062 free_list = collector->free_list_code_space().get(); 1063 } else { 1064 // Compaction spaces only represent old or code space. 1065 UNREACHABLE(); 1066 } 1067 DCHECK(free_list != nullptr); 1068 intptr_t refilled = 0; 1069 while (refilled < kCompactionMemoryWanted) { 1070 FreeSpace* node = 1071 free_list->TryRemoveMemory(kCompactionMemoryWanted - refilled); 1072 if (node == nullptr) return; 1073 refilled += node->size(); 1074 AddMemory(node->address(), node->size()); 1075 } 1076 } 1077 1078 1079 void PagedSpace::MoveOverFreeMemory(PagedSpace* other) { 1080 DCHECK(identity() == other->identity()); 1081 // Destroy the linear allocation space of {other}. This is needed to 1082 // (a) not waste the memory and 1083 // (b) keep the rest of the chunk in an iterable state (filler is needed). 1084 other->EmptyAllocationInfo(); 1085 1086 // Move over the free list. Concatenate makes sure that the source free list 1087 // gets properly reset after moving over all nodes. 1088 intptr_t added = free_list_.Concatenate(other->free_list()); 1089 1090 // Moved memory is not recorded as allocated memory, but rather increases and 1091 // decreases capacity of the corresponding spaces. 1092 other->accounting_stats_.DecreaseCapacity(added); 1093 accounting_stats_.IncreaseCapacity(added); 1094 } 1095 1096 1097 void PagedSpace::MergeCompactionSpace(CompactionSpace* other) { 1098 // Unmerged fields: 1099 // area_size_ 1100 // anchor_ 1101 1102 MoveOverFreeMemory(other); 1103 1104 // Update and clear accounting statistics. 1105 accounting_stats_.Merge(other->accounting_stats_); 1106 other->accounting_stats_.Clear(); 1107 1108 // The linear allocation area of {other} should be destroyed now. 1109 DCHECK(other->top() == nullptr); 1110 DCHECK(other->limit() == nullptr); 1111 1112 DCHECK(other->end_of_unswept_pages_ == nullptr); 1113 1114 AccountCommitted(other->CommittedMemory()); 1115 1116 // Move over pages. 1117 PageIterator it(other); 1118 Page* p = nullptr; 1119 while (it.has_next()) { 1120 p = it.next(); 1121 p->Unlink(); 1122 p->set_owner(this); 1123 p->InsertAfter(anchor_.prev_page()); 1124 } 1125 } 1126 1127 1128 size_t PagedSpace::CommittedPhysicalMemory() { 1129 if (!base::VirtualMemory::HasLazyCommits()) return CommittedMemory(); 1130 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 1131 size_t size = 0; 1132 PageIterator it(this); 1133 while (it.has_next()) { 1134 size += it.next()->CommittedPhysicalMemory(); 1135 } 1136 return size; 1137 } 1138 1139 1140 bool PagedSpace::ContainsSafe(Address addr) { 1141 Page* p = Page::FromAddress(addr); 1142 PageIterator iterator(this); 1143 while (iterator.has_next()) { 1144 if (iterator.next() == p) return true; 1145 } 1146 return false; 1147 } 1148 1149 1150 Object* PagedSpace::FindObject(Address addr) { 1151 // Note: this function can only be called on iterable spaces. 1152 DCHECK(!heap()->mark_compact_collector()->in_use()); 1153 1154 if (!Contains(addr)) return Smi::FromInt(0); // Signaling not found. 1155 1156 Page* p = Page::FromAddress(addr); 1157 HeapObjectIterator it(p); 1158 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 1159 Address cur = obj->address(); 1160 Address next = cur + obj->Size(); 1161 if ((cur <= addr) && (addr < next)) return obj; 1162 } 1163 1164 UNREACHABLE(); 1165 return Smi::FromInt(0); 1166 } 1167 1168 1169 bool PagedSpace::CanExpand(size_t size) { 1170 DCHECK(heap()->mark_compact_collector()->is_compacting() || 1171 Capacity() <= heap()->MaxOldGenerationSize()); 1172 1173 // Are we going to exceed capacity for this space? At this point we can be 1174 // way over the maximum size because of AlwaysAllocate scopes and large 1175 // objects. 1176 if (!heap()->CanExpandOldGeneration(static_cast<int>(size))) return false; 1177 1178 return true; 1179 } 1180 1181 1182 bool PagedSpace::Expand() { 1183 intptr_t size = AreaSize(); 1184 if (snapshotable() && !HasPages()) { 1185 size = Snapshot::SizeOfFirstPage(heap()->isolate(), identity()); 1186 } 1187 1188 if (!CanExpand(size)) return false; 1189 1190 Page* p = heap()->isolate()->memory_allocator()->AllocatePage(size, this, 1191 executable()); 1192 if (p == NULL) return false; 1193 1194 AccountCommitted(static_cast<intptr_t>(p->size())); 1195 1196 // Pages created during bootstrapping may contain immortal immovable objects. 1197 if (!heap()->deserialization_complete()) p->MarkNeverEvacuate(); 1198 1199 DCHECK(Capacity() <= heap()->MaxOldGenerationSize()); 1200 1201 p->InsertAfter(anchor_.prev_page()); 1202 1203 return true; 1204 } 1205 1206 1207 int PagedSpace::CountTotalPages() { 1208 PageIterator it(this); 1209 int count = 0; 1210 while (it.has_next()) { 1211 it.next(); 1212 count++; 1213 } 1214 return count; 1215 } 1216 1217 1218 void PagedSpace::ResetFreeListStatistics() { 1219 PageIterator page_iterator(this); 1220 while (page_iterator.has_next()) { 1221 Page* page = page_iterator.next(); 1222 page->ResetFreeListStatistics(); 1223 } 1224 } 1225 1226 1227 void PagedSpace::IncreaseCapacity(int size) { 1228 accounting_stats_.ExpandSpace(size); 1229 } 1230 1231 1232 void PagedSpace::ReleasePage(Page* page) { 1233 DCHECK(page->LiveBytes() == 0); 1234 DCHECK(AreaSize() == page->area_size()); 1235 1236 if (page->WasSwept()) { 1237 intptr_t size = free_list_.EvictFreeListItems(page); 1238 accounting_stats_.AllocateBytes(size); 1239 DCHECK_EQ(AreaSize(), static_cast<int>(size)); 1240 } 1241 1242 if (page->IsFlagSet(MemoryChunk::SCAN_ON_SCAVENGE)) { 1243 heap()->decrement_scan_on_scavenge_pages(); 1244 page->ClearFlag(MemoryChunk::SCAN_ON_SCAVENGE); 1245 } 1246 1247 DCHECK(!free_list_.ContainsPageFreeListItems(page)); 1248 1249 if (Page::FromAllocationTop(allocation_info_.top()) == page) { 1250 allocation_info_.Reset(nullptr, nullptr); 1251 } 1252 1253 // If page is still in a list, unlink it from that list. 1254 if (page->next_chunk() != NULL) { 1255 DCHECK(page->prev_chunk() != NULL); 1256 page->Unlink(); 1257 } 1258 1259 AccountUncommitted(static_cast<intptr_t>(page->size())); 1260 heap()->QueueMemoryChunkForFree(page); 1261 1262 DCHECK(Capacity() > 0); 1263 accounting_stats_.ShrinkSpace(AreaSize()); 1264 } 1265 1266 1267 #ifdef DEBUG 1268 void PagedSpace::Print() {} 1269 #endif 1270 1271 #ifdef VERIFY_HEAP 1272 void PagedSpace::Verify(ObjectVisitor* visitor) { 1273 bool allocation_pointer_found_in_space = 1274 (allocation_info_.top() == allocation_info_.limit()); 1275 PageIterator page_iterator(this); 1276 while (page_iterator.has_next()) { 1277 Page* page = page_iterator.next(); 1278 CHECK(page->owner() == this); 1279 if (page == Page::FromAllocationTop(allocation_info_.top())) { 1280 allocation_pointer_found_in_space = true; 1281 } 1282 CHECK(page->WasSwept()); 1283 HeapObjectIterator it(page); 1284 Address end_of_previous_object = page->area_start(); 1285 Address top = page->area_end(); 1286 int black_size = 0; 1287 for (HeapObject* object = it.Next(); object != NULL; object = it.Next()) { 1288 CHECK(end_of_previous_object <= object->address()); 1289 1290 // The first word should be a map, and we expect all map pointers to 1291 // be in map space. 1292 Map* map = object->map(); 1293 CHECK(map->IsMap()); 1294 CHECK(heap()->map_space()->Contains(map)); 1295 1296 // Perform space-specific object verification. 1297 VerifyObject(object); 1298 1299 // The object itself should look OK. 1300 object->ObjectVerify(); 1301 1302 // All the interior pointers should be contained in the heap. 1303 int size = object->Size(); 1304 object->IterateBody(map->instance_type(), size, visitor); 1305 if (Marking::IsBlack(Marking::MarkBitFrom(object))) { 1306 black_size += size; 1307 } 1308 1309 CHECK(object->address() + size <= top); 1310 end_of_previous_object = object->address() + size; 1311 } 1312 CHECK_LE(black_size, page->LiveBytes()); 1313 } 1314 CHECK(allocation_pointer_found_in_space); 1315 } 1316 #endif // VERIFY_HEAP 1317 1318 // ----------------------------------------------------------------------------- 1319 // NewSpace implementation 1320 1321 1322 bool NewSpace::SetUp(int reserved_semispace_capacity, 1323 int maximum_semispace_capacity) { 1324 // Set up new space based on the preallocated memory block defined by 1325 // start and size. The provided space is divided into two semi-spaces. 1326 // To support fast containment testing in the new space, the size of 1327 // this chunk must be a power of two and it must be aligned to its size. 1328 int initial_semispace_capacity = heap()->InitialSemiSpaceSize(); 1329 1330 int target_semispace_capacity = heap()->TargetSemiSpaceSize(); 1331 1332 size_t size = 2 * reserved_semispace_capacity; 1333 Address base = heap()->isolate()->memory_allocator()->ReserveAlignedMemory( 1334 size, size, &reservation_); 1335 if (base == NULL) return false; 1336 1337 chunk_base_ = base; 1338 chunk_size_ = static_cast<uintptr_t>(size); 1339 LOG(heap()->isolate(), NewEvent("InitialChunk", chunk_base_, chunk_size_)); 1340 1341 DCHECK(initial_semispace_capacity <= maximum_semispace_capacity); 1342 DCHECK(base::bits::IsPowerOfTwo32(maximum_semispace_capacity)); 1343 1344 // Allocate and set up the histogram arrays if necessary. 1345 allocated_histogram_ = NewArray<HistogramInfo>(LAST_TYPE + 1); 1346 promoted_histogram_ = NewArray<HistogramInfo>(LAST_TYPE + 1); 1347 1348 #define SET_NAME(name) \ 1349 allocated_histogram_[name].set_name(#name); \ 1350 promoted_histogram_[name].set_name(#name); 1351 INSTANCE_TYPE_LIST(SET_NAME) 1352 #undef SET_NAME 1353 1354 DCHECK(reserved_semispace_capacity == heap()->ReservedSemiSpaceSize()); 1355 DCHECK(static_cast<intptr_t>(chunk_size_) >= 1356 2 * heap()->ReservedSemiSpaceSize()); 1357 DCHECK(IsAddressAligned(chunk_base_, 2 * reserved_semispace_capacity, 0)); 1358 1359 to_space_.SetUp(chunk_base_, initial_semispace_capacity, 1360 target_semispace_capacity, maximum_semispace_capacity); 1361 from_space_.SetUp(chunk_base_ + reserved_semispace_capacity, 1362 initial_semispace_capacity, target_semispace_capacity, 1363 maximum_semispace_capacity); 1364 if (!to_space_.Commit()) { 1365 return false; 1366 } 1367 DCHECK(!from_space_.is_committed()); // No need to use memory yet. 1368 1369 start_ = chunk_base_; 1370 address_mask_ = ~(2 * reserved_semispace_capacity - 1); 1371 object_mask_ = address_mask_ | kHeapObjectTagMask; 1372 object_expected_ = reinterpret_cast<uintptr_t>(start_) | kHeapObjectTag; 1373 1374 ResetAllocationInfo(); 1375 1376 return true; 1377 } 1378 1379 1380 void NewSpace::TearDown() { 1381 if (allocated_histogram_) { 1382 DeleteArray(allocated_histogram_); 1383 allocated_histogram_ = NULL; 1384 } 1385 if (promoted_histogram_) { 1386 DeleteArray(promoted_histogram_); 1387 promoted_histogram_ = NULL; 1388 } 1389 1390 start_ = NULL; 1391 allocation_info_.Reset(nullptr, nullptr); 1392 1393 1394 to_space_.TearDown(); 1395 from_space_.TearDown(); 1396 1397 heap()->isolate()->memory_allocator()->FreeNewSpaceMemory( 1398 chunk_base_, &reservation_, NOT_EXECUTABLE); 1399 1400 chunk_base_ = NULL; 1401 chunk_size_ = 0; 1402 } 1403 1404 1405 void NewSpace::Flip() { SemiSpace::Swap(&from_space_, &to_space_); } 1406 1407 1408 void NewSpace::Grow() { 1409 // Double the semispace size but only up to maximum capacity. 1410 DCHECK(TotalCapacity() < MaximumCapacity()); 1411 int new_capacity = 1412 Min(MaximumCapacity(), 1413 FLAG_semi_space_growth_factor * static_cast<int>(TotalCapacity())); 1414 if (to_space_.GrowTo(new_capacity)) { 1415 // Only grow from space if we managed to grow to-space. 1416 if (!from_space_.GrowTo(new_capacity)) { 1417 // If we managed to grow to-space but couldn't grow from-space, 1418 // attempt to shrink to-space. 1419 if (!to_space_.ShrinkTo(from_space_.TotalCapacity())) { 1420 // We are in an inconsistent state because we could not 1421 // commit/uncommit memory from new space. 1422 CHECK(false); 1423 } 1424 } 1425 } 1426 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1427 } 1428 1429 1430 bool NewSpace::GrowOnePage() { 1431 if (TotalCapacity() == MaximumCapacity()) return false; 1432 int new_capacity = static_cast<int>(TotalCapacity()) + Page::kPageSize; 1433 if (to_space_.GrowTo(new_capacity)) { 1434 // Only grow from space if we managed to grow to-space and the from space 1435 // is actually committed. 1436 if (from_space_.is_committed()) { 1437 if (!from_space_.GrowTo(new_capacity)) { 1438 // If we managed to grow to-space but couldn't grow from-space, 1439 // attempt to shrink to-space. 1440 if (!to_space_.ShrinkTo(from_space_.TotalCapacity())) { 1441 // We are in an inconsistent state because we could not 1442 // commit/uncommit memory from new space. 1443 CHECK(false); 1444 } 1445 return false; 1446 } 1447 } else { 1448 if (!from_space_.SetTotalCapacity(new_capacity)) { 1449 // Can't really happen, but better safe than sorry. 1450 CHECK(false); 1451 } 1452 } 1453 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1454 return true; 1455 } 1456 return false; 1457 } 1458 1459 1460 void NewSpace::Shrink() { 1461 int new_capacity = Max(InitialTotalCapacity(), 2 * SizeAsInt()); 1462 int rounded_new_capacity = RoundUp(new_capacity, Page::kPageSize); 1463 if (rounded_new_capacity < TotalCapacity() && 1464 to_space_.ShrinkTo(rounded_new_capacity)) { 1465 // Only shrink from-space if we managed to shrink to-space. 1466 from_space_.Reset(); 1467 if (!from_space_.ShrinkTo(rounded_new_capacity)) { 1468 // If we managed to shrink to-space but couldn't shrink from 1469 // space, attempt to grow to-space again. 1470 if (!to_space_.GrowTo(from_space_.TotalCapacity())) { 1471 // We are in an inconsistent state because we could not 1472 // commit/uncommit memory from new space. 1473 CHECK(false); 1474 } 1475 } 1476 } 1477 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1478 } 1479 1480 1481 void LocalAllocationBuffer::Close() { 1482 if (IsValid()) { 1483 heap_->CreateFillerObjectAt( 1484 allocation_info_.top(), 1485 static_cast<int>(allocation_info_.limit() - allocation_info_.top())); 1486 } 1487 } 1488 1489 1490 LocalAllocationBuffer::LocalAllocationBuffer(Heap* heap, 1491 AllocationInfo allocation_info) 1492 : heap_(heap), allocation_info_(allocation_info) { 1493 if (IsValid()) { 1494 heap_->CreateFillerObjectAt( 1495 allocation_info_.top(), 1496 static_cast<int>(allocation_info_.limit() - allocation_info_.top())); 1497 } 1498 } 1499 1500 1501 LocalAllocationBuffer::LocalAllocationBuffer( 1502 const LocalAllocationBuffer& other) { 1503 *this = other; 1504 } 1505 1506 1507 LocalAllocationBuffer& LocalAllocationBuffer::operator=( 1508 const LocalAllocationBuffer& other) { 1509 Close(); 1510 heap_ = other.heap_; 1511 allocation_info_ = other.allocation_info_; 1512 1513 // This is needed since we (a) cannot yet use move-semantics, and (b) want 1514 // to make the use of the class easy by it as value and (c) implicitly call 1515 // {Close} upon copy. 1516 const_cast<LocalAllocationBuffer&>(other) 1517 .allocation_info_.Reset(nullptr, nullptr); 1518 return *this; 1519 } 1520 1521 1522 void NewSpace::UpdateAllocationInfo() { 1523 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 1524 allocation_info_.Reset(to_space_.page_low(), to_space_.page_high()); 1525 UpdateInlineAllocationLimit(0); 1526 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1527 } 1528 1529 1530 void NewSpace::ResetAllocationInfo() { 1531 Address old_top = allocation_info_.top(); 1532 to_space_.Reset(); 1533 UpdateAllocationInfo(); 1534 pages_used_ = 0; 1535 // Clear all mark-bits in the to-space. 1536 NewSpacePageIterator it(&to_space_); 1537 while (it.has_next()) { 1538 Bitmap::Clear(it.next()); 1539 } 1540 InlineAllocationStep(old_top, allocation_info_.top(), nullptr, 0); 1541 } 1542 1543 1544 void NewSpace::UpdateInlineAllocationLimit(int size_in_bytes) { 1545 if (heap()->inline_allocation_disabled()) { 1546 // Lowest limit when linear allocation was disabled. 1547 Address high = to_space_.page_high(); 1548 Address new_top = allocation_info_.top() + size_in_bytes; 1549 allocation_info_.set_limit(Min(new_top, high)); 1550 } else if (inline_allocation_observers_paused_ || 1551 top_on_previous_step_ == 0) { 1552 // Normal limit is the end of the current page. 1553 allocation_info_.set_limit(to_space_.page_high()); 1554 } else { 1555 // Lower limit during incremental marking. 1556 Address high = to_space_.page_high(); 1557 Address new_top = allocation_info_.top() + size_in_bytes; 1558 Address new_limit = new_top + GetNextInlineAllocationStepSize() - 1; 1559 allocation_info_.set_limit(Min(new_limit, high)); 1560 } 1561 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1562 } 1563 1564 1565 bool NewSpace::AddFreshPage() { 1566 Address top = allocation_info_.top(); 1567 if (NewSpacePage::IsAtStart(top)) { 1568 // The current page is already empty. Don't try to make another. 1569 1570 // We should only get here if someone asks to allocate more 1571 // than what can be stored in a single page. 1572 // TODO(gc): Change the limit on new-space allocation to prevent this 1573 // from happening (all such allocations should go directly to LOSpace). 1574 return false; 1575 } 1576 if (!to_space_.AdvancePage()) { 1577 // Check if we reached the target capacity yet. If not, try to commit a page 1578 // and continue. 1579 if ((to_space_.TotalCapacity() < to_space_.TargetCapacity()) && 1580 GrowOnePage()) { 1581 if (!to_space_.AdvancePage()) { 1582 // It doesn't make sense that we managed to commit a page, but can't use 1583 // it. 1584 CHECK(false); 1585 } 1586 } else { 1587 // Failed to get a new page in to-space. 1588 return false; 1589 } 1590 } 1591 1592 // Clear remainder of current page. 1593 Address limit = NewSpacePage::FromLimit(top)->area_end(); 1594 if (heap()->gc_state() == Heap::SCAVENGE) { 1595 heap()->promotion_queue()->SetNewLimit(limit); 1596 } 1597 1598 int remaining_in_page = static_cast<int>(limit - top); 1599 heap()->CreateFillerObjectAt(top, remaining_in_page); 1600 pages_used_++; 1601 UpdateAllocationInfo(); 1602 1603 return true; 1604 } 1605 1606 1607 bool NewSpace::AddFreshPageSynchronized() { 1608 base::LockGuard<base::Mutex> guard(&mutex_); 1609 return AddFreshPage(); 1610 } 1611 1612 1613 bool NewSpace::EnsureAllocation(int size_in_bytes, 1614 AllocationAlignment alignment) { 1615 Address old_top = allocation_info_.top(); 1616 Address high = to_space_.page_high(); 1617 int filler_size = Heap::GetFillToAlign(old_top, alignment); 1618 int aligned_size_in_bytes = size_in_bytes + filler_size; 1619 1620 if (old_top + aligned_size_in_bytes >= high) { 1621 // Not enough room in the page, try to allocate a new one. 1622 if (!AddFreshPage()) { 1623 return false; 1624 } 1625 1626 InlineAllocationStep(old_top, allocation_info_.top(), nullptr, 0); 1627 1628 old_top = allocation_info_.top(); 1629 high = to_space_.page_high(); 1630 filler_size = Heap::GetFillToAlign(old_top, alignment); 1631 aligned_size_in_bytes = size_in_bytes + filler_size; 1632 } 1633 1634 DCHECK(old_top + aligned_size_in_bytes < high); 1635 1636 if (allocation_info_.limit() < high) { 1637 // Either the limit has been lowered because linear allocation was disabled 1638 // or because incremental marking wants to get a chance to do a step, 1639 // or because idle scavenge job wants to get a chance to post a task. 1640 // Set the new limit accordingly. 1641 Address new_top = old_top + aligned_size_in_bytes; 1642 Address soon_object = old_top + filler_size; 1643 InlineAllocationStep(new_top, new_top, soon_object, size_in_bytes); 1644 UpdateInlineAllocationLimit(aligned_size_in_bytes); 1645 } 1646 return true; 1647 } 1648 1649 1650 void NewSpace::StartNextInlineAllocationStep() { 1651 if (!inline_allocation_observers_paused_) { 1652 top_on_previous_step_ = 1653 inline_allocation_observers_.length() ? allocation_info_.top() : 0; 1654 UpdateInlineAllocationLimit(0); 1655 } 1656 } 1657 1658 1659 intptr_t NewSpace::GetNextInlineAllocationStepSize() { 1660 intptr_t next_step = 0; 1661 for (int i = 0; i < inline_allocation_observers_.length(); ++i) { 1662 InlineAllocationObserver* o = inline_allocation_observers_[i]; 1663 next_step = next_step ? Min(next_step, o->bytes_to_next_step()) 1664 : o->bytes_to_next_step(); 1665 } 1666 DCHECK(inline_allocation_observers_.length() == 0 || next_step != 0); 1667 return next_step; 1668 } 1669 1670 1671 void NewSpace::AddInlineAllocationObserver(InlineAllocationObserver* observer) { 1672 inline_allocation_observers_.Add(observer); 1673 StartNextInlineAllocationStep(); 1674 } 1675 1676 1677 void NewSpace::RemoveInlineAllocationObserver( 1678 InlineAllocationObserver* observer) { 1679 bool removed = inline_allocation_observers_.RemoveElement(observer); 1680 // Only used in assertion. Suppress unused variable warning. 1681 static_cast<void>(removed); 1682 DCHECK(removed); 1683 StartNextInlineAllocationStep(); 1684 } 1685 1686 1687 void NewSpace::PauseInlineAllocationObservers() { 1688 // Do a step to account for memory allocated so far. 1689 InlineAllocationStep(top(), top(), nullptr, 0); 1690 inline_allocation_observers_paused_ = true; 1691 top_on_previous_step_ = 0; 1692 UpdateInlineAllocationLimit(0); 1693 } 1694 1695 1696 void NewSpace::ResumeInlineAllocationObservers() { 1697 DCHECK(top_on_previous_step_ == 0); 1698 inline_allocation_observers_paused_ = false; 1699 StartNextInlineAllocationStep(); 1700 } 1701 1702 1703 void NewSpace::InlineAllocationStep(Address top, Address new_top, 1704 Address soon_object, size_t size) { 1705 if (top_on_previous_step_) { 1706 int bytes_allocated = static_cast<int>(top - top_on_previous_step_); 1707 for (int i = 0; i < inline_allocation_observers_.length(); ++i) { 1708 inline_allocation_observers_[i]->InlineAllocationStep(bytes_allocated, 1709 soon_object, size); 1710 } 1711 top_on_previous_step_ = new_top; 1712 } 1713 } 1714 1715 #ifdef VERIFY_HEAP 1716 // We do not use the SemiSpaceIterator because verification doesn't assume 1717 // that it works (it depends on the invariants we are checking). 1718 void NewSpace::Verify() { 1719 // The allocation pointer should be in the space or at the very end. 1720 DCHECK_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1721 1722 // There should be objects packed in from the low address up to the 1723 // allocation pointer. 1724 Address current = to_space_.first_page()->area_start(); 1725 CHECK_EQ(current, to_space_.space_start()); 1726 1727 while (current != top()) { 1728 if (!NewSpacePage::IsAtEnd(current)) { 1729 // The allocation pointer should not be in the middle of an object. 1730 CHECK(!NewSpacePage::FromLimit(current)->ContainsLimit(top()) || 1731 current < top()); 1732 1733 HeapObject* object = HeapObject::FromAddress(current); 1734 1735 // The first word should be a map, and we expect all map pointers to 1736 // be in map space. 1737 Map* map = object->map(); 1738 CHECK(map->IsMap()); 1739 CHECK(heap()->map_space()->Contains(map)); 1740 1741 // The object should not be code or a map. 1742 CHECK(!object->IsMap()); 1743 CHECK(!object->IsCode()); 1744 1745 // The object itself should look OK. 1746 object->ObjectVerify(); 1747 1748 // All the interior pointers should be contained in the heap. 1749 VerifyPointersVisitor visitor; 1750 int size = object->Size(); 1751 object->IterateBody(map->instance_type(), size, &visitor); 1752 1753 current += size; 1754 } else { 1755 // At end of page, switch to next page. 1756 NewSpacePage* page = NewSpacePage::FromLimit(current)->next_page(); 1757 // Next page should be valid. 1758 CHECK(!page->is_anchor()); 1759 current = page->area_start(); 1760 } 1761 } 1762 1763 // Check semi-spaces. 1764 CHECK_EQ(from_space_.id(), kFromSpace); 1765 CHECK_EQ(to_space_.id(), kToSpace); 1766 from_space_.Verify(); 1767 to_space_.Verify(); 1768 } 1769 #endif 1770 1771 // ----------------------------------------------------------------------------- 1772 // SemiSpace implementation 1773 1774 void SemiSpace::SetUp(Address start, int initial_capacity, int target_capacity, 1775 int maximum_capacity) { 1776 // Creates a space in the young generation. The constructor does not 1777 // allocate memory from the OS. A SemiSpace is given a contiguous chunk of 1778 // memory of size 'capacity' when set up, and does not grow or shrink 1779 // otherwise. In the mark-compact collector, the memory region of the from 1780 // space is used as the marking stack. It requires contiguous memory 1781 // addresses. 1782 DCHECK(maximum_capacity >= Page::kPageSize); 1783 DCHECK(initial_capacity <= target_capacity); 1784 DCHECK(target_capacity <= maximum_capacity); 1785 initial_total_capacity_ = RoundDown(initial_capacity, Page::kPageSize); 1786 total_capacity_ = initial_capacity; 1787 target_capacity_ = RoundDown(target_capacity, Page::kPageSize); 1788 maximum_total_capacity_ = RoundDown(maximum_capacity, Page::kPageSize); 1789 committed_ = false; 1790 start_ = start; 1791 address_mask_ = ~(maximum_capacity - 1); 1792 object_mask_ = address_mask_ | kHeapObjectTagMask; 1793 object_expected_ = reinterpret_cast<uintptr_t>(start) | kHeapObjectTag; 1794 age_mark_ = start_ + NewSpacePage::kObjectStartOffset; 1795 } 1796 1797 1798 void SemiSpace::TearDown() { 1799 start_ = NULL; 1800 total_capacity_ = 0; 1801 } 1802 1803 1804 bool SemiSpace::Commit() { 1805 DCHECK(!is_committed()); 1806 int pages = total_capacity_ / Page::kPageSize; 1807 if (!heap()->isolate()->memory_allocator()->CommitBlock( 1808 start_, total_capacity_, executable())) { 1809 return false; 1810 } 1811 AccountCommitted(total_capacity_); 1812 1813 NewSpacePage* current = anchor(); 1814 for (int i = 0; i < pages; i++) { 1815 NewSpacePage* new_page = 1816 NewSpacePage::Initialize(heap(), start_ + i * Page::kPageSize, this); 1817 new_page->InsertAfter(current); 1818 current = new_page; 1819 } 1820 1821 SetCapacity(total_capacity_); 1822 committed_ = true; 1823 Reset(); 1824 return true; 1825 } 1826 1827 1828 bool SemiSpace::Uncommit() { 1829 DCHECK(is_committed()); 1830 Address start = start_ + maximum_total_capacity_ - total_capacity_; 1831 if (!heap()->isolate()->memory_allocator()->UncommitBlock(start, 1832 total_capacity_)) { 1833 return false; 1834 } 1835 AccountUncommitted(total_capacity_); 1836 1837 anchor()->set_next_page(anchor()); 1838 anchor()->set_prev_page(anchor()); 1839 1840 committed_ = false; 1841 return true; 1842 } 1843 1844 1845 size_t SemiSpace::CommittedPhysicalMemory() { 1846 if (!is_committed()) return 0; 1847 size_t size = 0; 1848 NewSpacePageIterator it(this); 1849 while (it.has_next()) { 1850 size += it.next()->CommittedPhysicalMemory(); 1851 } 1852 return size; 1853 } 1854 1855 1856 bool SemiSpace::GrowTo(int new_capacity) { 1857 if (!is_committed()) { 1858 if (!Commit()) return false; 1859 } 1860 DCHECK((new_capacity & Page::kPageAlignmentMask) == 0); 1861 DCHECK(new_capacity <= maximum_total_capacity_); 1862 DCHECK(new_capacity > total_capacity_); 1863 int pages_before = total_capacity_ / Page::kPageSize; 1864 int pages_after = new_capacity / Page::kPageSize; 1865 1866 size_t delta = new_capacity - total_capacity_; 1867 1868 DCHECK(IsAligned(delta, base::OS::AllocateAlignment())); 1869 if (!heap()->isolate()->memory_allocator()->CommitBlock( 1870 start_ + total_capacity_, delta, executable())) { 1871 return false; 1872 } 1873 AccountCommitted(static_cast<intptr_t>(delta)); 1874 SetCapacity(new_capacity); 1875 NewSpacePage* last_page = anchor()->prev_page(); 1876 DCHECK(last_page != anchor()); 1877 for (int i = pages_before; i < pages_after; i++) { 1878 Address page_address = start_ + i * Page::kPageSize; 1879 NewSpacePage* new_page = 1880 NewSpacePage::Initialize(heap(), page_address, this); 1881 new_page->InsertAfter(last_page); 1882 Bitmap::Clear(new_page); 1883 // Duplicate the flags that was set on the old page. 1884 new_page->SetFlags(last_page->GetFlags(), 1885 NewSpacePage::kCopyOnFlipFlagsMask); 1886 last_page = new_page; 1887 } 1888 return true; 1889 } 1890 1891 1892 bool SemiSpace::ShrinkTo(int new_capacity) { 1893 DCHECK((new_capacity & Page::kPageAlignmentMask) == 0); 1894 DCHECK(new_capacity >= initial_total_capacity_); 1895 DCHECK(new_capacity < total_capacity_); 1896 if (is_committed()) { 1897 size_t delta = total_capacity_ - new_capacity; 1898 DCHECK(IsAligned(delta, base::OS::AllocateAlignment())); 1899 1900 MemoryAllocator* allocator = heap()->isolate()->memory_allocator(); 1901 if (!allocator->UncommitBlock(start_ + new_capacity, delta)) { 1902 return false; 1903 } 1904 AccountUncommitted(static_cast<intptr_t>(delta)); 1905 1906 int pages_after = new_capacity / Page::kPageSize; 1907 NewSpacePage* new_last_page = 1908 NewSpacePage::FromAddress(start_ + (pages_after - 1) * Page::kPageSize); 1909 new_last_page->set_next_page(anchor()); 1910 anchor()->set_prev_page(new_last_page); 1911 DCHECK((current_page_ >= first_page()) && (current_page_ <= new_last_page)); 1912 } 1913 1914 SetCapacity(new_capacity); 1915 1916 return true; 1917 } 1918 1919 1920 bool SemiSpace::SetTotalCapacity(int new_capacity) { 1921 CHECK(!is_committed()); 1922 if (new_capacity >= initial_total_capacity_ && 1923 new_capacity <= maximum_total_capacity_) { 1924 total_capacity_ = new_capacity; 1925 return true; 1926 } 1927 return false; 1928 } 1929 1930 1931 void SemiSpace::FlipPages(intptr_t flags, intptr_t mask) { 1932 anchor_.set_owner(this); 1933 // Fixup back-pointers to anchor. Address of anchor changes 1934 // when we swap. 1935 anchor_.prev_page()->set_next_page(&anchor_); 1936 anchor_.next_page()->set_prev_page(&anchor_); 1937 1938 bool becomes_to_space = (id_ == kFromSpace); 1939 id_ = becomes_to_space ? kToSpace : kFromSpace; 1940 NewSpacePage* page = anchor_.next_page(); 1941 while (page != &anchor_) { 1942 page->set_owner(this); 1943 page->SetFlags(flags, mask); 1944 if (becomes_to_space) { 1945 page->ClearFlag(MemoryChunk::IN_FROM_SPACE); 1946 page->SetFlag(MemoryChunk::IN_TO_SPACE); 1947 page->ClearFlag(MemoryChunk::NEW_SPACE_BELOW_AGE_MARK); 1948 page->ResetLiveBytes(); 1949 } else { 1950 page->SetFlag(MemoryChunk::IN_FROM_SPACE); 1951 page->ClearFlag(MemoryChunk::IN_TO_SPACE); 1952 } 1953 DCHECK(page->IsFlagSet(MemoryChunk::SCAN_ON_SCAVENGE)); 1954 DCHECK(page->IsFlagSet(MemoryChunk::IN_TO_SPACE) || 1955 page->IsFlagSet(MemoryChunk::IN_FROM_SPACE)); 1956 page = page->next_page(); 1957 } 1958 } 1959 1960 1961 void SemiSpace::Reset() { 1962 DCHECK(anchor_.next_page() != &anchor_); 1963 current_page_ = anchor_.next_page(); 1964 } 1965 1966 1967 void SemiSpace::Swap(SemiSpace* from, SemiSpace* to) { 1968 // We won't be swapping semispaces without data in them. 1969 DCHECK(from->anchor_.next_page() != &from->anchor_); 1970 DCHECK(to->anchor_.next_page() != &to->anchor_); 1971 1972 // Swap bits. 1973 SemiSpace tmp = *from; 1974 *from = *to; 1975 *to = tmp; 1976 1977 // Fixup back-pointers to the page list anchor now that its address 1978 // has changed. 1979 // Swap to/from-space bits on pages. 1980 // Copy GC flags from old active space (from-space) to new (to-space). 1981 intptr_t flags = from->current_page()->GetFlags(); 1982 to->FlipPages(flags, NewSpacePage::kCopyOnFlipFlagsMask); 1983 1984 from->FlipPages(0, 0); 1985 } 1986 1987 1988 void SemiSpace::SetCapacity(int new_capacity) { 1989 total_capacity_ = new_capacity; 1990 } 1991 1992 1993 void SemiSpace::set_age_mark(Address mark) { 1994 DCHECK(NewSpacePage::FromLimit(mark)->semi_space() == this); 1995 age_mark_ = mark; 1996 // Mark all pages up to the one containing mark. 1997 NewSpacePageIterator it(space_start(), mark); 1998 while (it.has_next()) { 1999 it.next()->SetFlag(MemoryChunk::NEW_SPACE_BELOW_AGE_MARK); 2000 } 2001 } 2002 2003 2004 #ifdef DEBUG 2005 void SemiSpace::Print() {} 2006 #endif 2007 2008 #ifdef VERIFY_HEAP 2009 void SemiSpace::Verify() { 2010 bool is_from_space = (id_ == kFromSpace); 2011 NewSpacePage* page = anchor_.next_page(); 2012 CHECK(anchor_.semi_space() == this); 2013 while (page != &anchor_) { 2014 CHECK(page->semi_space() == this); 2015 CHECK(page->InNewSpace()); 2016 CHECK(page->IsFlagSet(is_from_space ? MemoryChunk::IN_FROM_SPACE 2017 : MemoryChunk::IN_TO_SPACE)); 2018 CHECK(!page->IsFlagSet(is_from_space ? MemoryChunk::IN_TO_SPACE 2019 : MemoryChunk::IN_FROM_SPACE)); 2020 CHECK(page->IsFlagSet(MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING)); 2021 if (!is_from_space) { 2022 // The pointers-from-here-are-interesting flag isn't updated dynamically 2023 // on from-space pages, so it might be out of sync with the marking state. 2024 if (page->heap()->incremental_marking()->IsMarking()) { 2025 CHECK(page->IsFlagSet(MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING)); 2026 } else { 2027 CHECK( 2028 !page->IsFlagSet(MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING)); 2029 } 2030 // TODO(gc): Check that the live_bytes_count_ field matches the 2031 // black marking on the page (if we make it match in new-space). 2032 } 2033 CHECK(page->IsFlagSet(MemoryChunk::SCAN_ON_SCAVENGE)); 2034 CHECK(page->prev_page()->next_page() == page); 2035 page = page->next_page(); 2036 } 2037 } 2038 #endif 2039 2040 #ifdef DEBUG 2041 void SemiSpace::AssertValidRange(Address start, Address end) { 2042 // Addresses belong to same semi-space 2043 NewSpacePage* page = NewSpacePage::FromLimit(start); 2044 NewSpacePage* end_page = NewSpacePage::FromLimit(end); 2045 SemiSpace* space = page->semi_space(); 2046 CHECK_EQ(space, end_page->semi_space()); 2047 // Start address is before end address, either on same page, 2048 // or end address is on a later page in the linked list of 2049 // semi-space pages. 2050 if (page == end_page) { 2051 CHECK(start <= end); 2052 } else { 2053 while (page != end_page) { 2054 page = page->next_page(); 2055 CHECK_NE(page, space->anchor()); 2056 } 2057 } 2058 } 2059 #endif 2060 2061 2062 // ----------------------------------------------------------------------------- 2063 // SemiSpaceIterator implementation. 2064 2065 SemiSpaceIterator::SemiSpaceIterator(NewSpace* space) { 2066 Initialize(space->bottom(), space->top()); 2067 } 2068 2069 2070 void SemiSpaceIterator::Initialize(Address start, Address end) { 2071 SemiSpace::AssertValidRange(start, end); 2072 current_ = start; 2073 limit_ = end; 2074 } 2075 2076 2077 #ifdef DEBUG 2078 // heap_histograms is shared, always clear it before using it. 2079 static void ClearHistograms(Isolate* isolate) { 2080 // We reset the name each time, though it hasn't changed. 2081 #define DEF_TYPE_NAME(name) isolate->heap_histograms()[name].set_name(#name); 2082 INSTANCE_TYPE_LIST(DEF_TYPE_NAME) 2083 #undef DEF_TYPE_NAME 2084 2085 #define CLEAR_HISTOGRAM(name) isolate->heap_histograms()[name].clear(); 2086 INSTANCE_TYPE_LIST(CLEAR_HISTOGRAM) 2087 #undef CLEAR_HISTOGRAM 2088 2089 isolate->js_spill_information()->Clear(); 2090 } 2091 2092 2093 static void ClearCodeKindStatistics(int* code_kind_statistics) { 2094 for (int i = 0; i < Code::NUMBER_OF_KINDS; i++) { 2095 code_kind_statistics[i] = 0; 2096 } 2097 } 2098 2099 2100 static void ReportCodeKindStatistics(int* code_kind_statistics) { 2101 PrintF("\n Code kind histograms: \n"); 2102 for (int i = 0; i < Code::NUMBER_OF_KINDS; i++) { 2103 if (code_kind_statistics[i] > 0) { 2104 PrintF(" %-20s: %10d bytes\n", 2105 Code::Kind2String(static_cast<Code::Kind>(i)), 2106 code_kind_statistics[i]); 2107 } 2108 } 2109 PrintF("\n"); 2110 } 2111 2112 2113 static int CollectHistogramInfo(HeapObject* obj) { 2114 Isolate* isolate = obj->GetIsolate(); 2115 InstanceType type = obj->map()->instance_type(); 2116 DCHECK(0 <= type && type <= LAST_TYPE); 2117 DCHECK(isolate->heap_histograms()[type].name() != NULL); 2118 isolate->heap_histograms()[type].increment_number(1); 2119 isolate->heap_histograms()[type].increment_bytes(obj->Size()); 2120 2121 if (FLAG_collect_heap_spill_statistics && obj->IsJSObject()) { 2122 JSObject::cast(obj) 2123 ->IncrementSpillStatistics(isolate->js_spill_information()); 2124 } 2125 2126 return obj->Size(); 2127 } 2128 2129 2130 static void ReportHistogram(Isolate* isolate, bool print_spill) { 2131 PrintF("\n Object Histogram:\n"); 2132 for (int i = 0; i <= LAST_TYPE; i++) { 2133 if (isolate->heap_histograms()[i].number() > 0) { 2134 PrintF(" %-34s%10d (%10d bytes)\n", 2135 isolate->heap_histograms()[i].name(), 2136 isolate->heap_histograms()[i].number(), 2137 isolate->heap_histograms()[i].bytes()); 2138 } 2139 } 2140 PrintF("\n"); 2141 2142 // Summarize string types. 2143 int string_number = 0; 2144 int string_bytes = 0; 2145 #define INCREMENT(type, size, name, camel_name) \ 2146 string_number += isolate->heap_histograms()[type].number(); \ 2147 string_bytes += isolate->heap_histograms()[type].bytes(); 2148 STRING_TYPE_LIST(INCREMENT) 2149 #undef INCREMENT 2150 if (string_number > 0) { 2151 PrintF(" %-34s%10d (%10d bytes)\n\n", "STRING_TYPE", string_number, 2152 string_bytes); 2153 } 2154 2155 if (FLAG_collect_heap_spill_statistics && print_spill) { 2156 isolate->js_spill_information()->Print(); 2157 } 2158 } 2159 #endif // DEBUG 2160 2161 2162 // Support for statistics gathering for --heap-stats and --log-gc. 2163 void NewSpace::ClearHistograms() { 2164 for (int i = 0; i <= LAST_TYPE; i++) { 2165 allocated_histogram_[i].clear(); 2166 promoted_histogram_[i].clear(); 2167 } 2168 } 2169 2170 2171 // Because the copying collector does not touch garbage objects, we iterate 2172 // the new space before a collection to get a histogram of allocated objects. 2173 // This only happens when --log-gc flag is set. 2174 void NewSpace::CollectStatistics() { 2175 ClearHistograms(); 2176 SemiSpaceIterator it(this); 2177 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) 2178 RecordAllocation(obj); 2179 } 2180 2181 2182 static void DoReportStatistics(Isolate* isolate, HistogramInfo* info, 2183 const char* description) { 2184 LOG(isolate, HeapSampleBeginEvent("NewSpace", description)); 2185 // Lump all the string types together. 2186 int string_number = 0; 2187 int string_bytes = 0; 2188 #define INCREMENT(type, size, name, camel_name) \ 2189 string_number += info[type].number(); \ 2190 string_bytes += info[type].bytes(); 2191 STRING_TYPE_LIST(INCREMENT) 2192 #undef INCREMENT 2193 if (string_number > 0) { 2194 LOG(isolate, 2195 HeapSampleItemEvent("STRING_TYPE", string_number, string_bytes)); 2196 } 2197 2198 // Then do the other types. 2199 for (int i = FIRST_NONSTRING_TYPE; i <= LAST_TYPE; ++i) { 2200 if (info[i].number() > 0) { 2201 LOG(isolate, HeapSampleItemEvent(info[i].name(), info[i].number(), 2202 info[i].bytes())); 2203 } 2204 } 2205 LOG(isolate, HeapSampleEndEvent("NewSpace", description)); 2206 } 2207 2208 2209 void NewSpace::ReportStatistics() { 2210 #ifdef DEBUG 2211 if (FLAG_heap_stats) { 2212 float pct = static_cast<float>(Available()) / TotalCapacity(); 2213 PrintF(" capacity: %" V8_PTR_PREFIX 2214 "d" 2215 ", available: %" V8_PTR_PREFIX "d, %%%d\n", 2216 TotalCapacity(), Available(), static_cast<int>(pct * 100)); 2217 PrintF("\n Object Histogram:\n"); 2218 for (int i = 0; i <= LAST_TYPE; i++) { 2219 if (allocated_histogram_[i].number() > 0) { 2220 PrintF(" %-34s%10d (%10d bytes)\n", allocated_histogram_[i].name(), 2221 allocated_histogram_[i].number(), 2222 allocated_histogram_[i].bytes()); 2223 } 2224 } 2225 PrintF("\n"); 2226 } 2227 #endif // DEBUG 2228 2229 if (FLAG_log_gc) { 2230 Isolate* isolate = heap()->isolate(); 2231 DoReportStatistics(isolate, allocated_histogram_, "allocated"); 2232 DoReportStatistics(isolate, promoted_histogram_, "promoted"); 2233 } 2234 } 2235 2236 2237 void NewSpace::RecordAllocation(HeapObject* obj) { 2238 InstanceType type = obj->map()->instance_type(); 2239 DCHECK(0 <= type && type <= LAST_TYPE); 2240 allocated_histogram_[type].increment_number(1); 2241 allocated_histogram_[type].increment_bytes(obj->Size()); 2242 } 2243 2244 2245 void NewSpace::RecordPromotion(HeapObject* obj) { 2246 InstanceType type = obj->map()->instance_type(); 2247 DCHECK(0 <= type && type <= LAST_TYPE); 2248 promoted_histogram_[type].increment_number(1); 2249 promoted_histogram_[type].increment_bytes(obj->Size()); 2250 } 2251 2252 2253 size_t NewSpace::CommittedPhysicalMemory() { 2254 if (!base::VirtualMemory::HasLazyCommits()) return CommittedMemory(); 2255 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 2256 size_t size = to_space_.CommittedPhysicalMemory(); 2257 if (from_space_.is_committed()) { 2258 size += from_space_.CommittedPhysicalMemory(); 2259 } 2260 return size; 2261 } 2262 2263 2264 // ----------------------------------------------------------------------------- 2265 // Free lists for old object spaces implementation 2266 2267 intptr_t FreeListCategory::Concatenate(FreeListCategory* category) { 2268 intptr_t free_bytes = 0; 2269 if (category->top() != NULL) { 2270 DCHECK(category->end_ != NULL); 2271 free_bytes = category->available(); 2272 if (end_ == NULL) { 2273 end_ = category->end(); 2274 } else { 2275 category->end()->set_next(top()); 2276 } 2277 set_top(category->top()); 2278 available_ += category->available(); 2279 category->Reset(); 2280 } 2281 return free_bytes; 2282 } 2283 2284 2285 void FreeListCategory::Reset() { 2286 set_top(nullptr); 2287 set_end(nullptr); 2288 available_ = 0; 2289 } 2290 2291 2292 intptr_t FreeListCategory::EvictFreeListItemsInList(Page* p) { 2293 intptr_t sum = 0; 2294 FreeSpace* prev_node = nullptr; 2295 for (FreeSpace* cur_node = top(); cur_node != nullptr; 2296 cur_node = cur_node->next()) { 2297 Page* page_for_node = Page::FromAddress(cur_node->address()); 2298 if (page_for_node == p) { 2299 // FreeSpace node on eviction page found, unlink it. 2300 int size = cur_node->size(); 2301 sum += size; 2302 DCHECK((prev_node != nullptr) || (top() == cur_node)); 2303 if (cur_node == top()) { 2304 set_top(cur_node->next()); 2305 } 2306 if (cur_node == end()) { 2307 set_end(prev_node); 2308 } 2309 if (prev_node != nullptr) { 2310 prev_node->set_next(cur_node->next()); 2311 } 2312 continue; 2313 } 2314 prev_node = cur_node; 2315 } 2316 DCHECK_EQ(p->available_in_free_list(type_), sum); 2317 p->add_available_in_free_list(type_, -sum); 2318 available_ -= sum; 2319 return sum; 2320 } 2321 2322 2323 bool FreeListCategory::ContainsPageFreeListItemsInList(Page* p) { 2324 FreeSpace* node = top(); 2325 while (node != NULL) { 2326 if (Page::FromAddress(node->address()) == p) return true; 2327 node = node->next(); 2328 } 2329 return false; 2330 } 2331 2332 2333 FreeSpace* FreeListCategory::PickNodeFromList(int* node_size) { 2334 FreeSpace* node = top(); 2335 if (node == nullptr) return nullptr; 2336 2337 Page* page = Page::FromAddress(node->address()); 2338 while ((node != nullptr) && !page->CanAllocate()) { 2339 available_ -= node->size(); 2340 page->add_available_in_free_list(type_, -(node->Size())); 2341 node = node->next(); 2342 } 2343 2344 if (node != nullptr) { 2345 set_top(node->next()); 2346 *node_size = node->Size(); 2347 available_ -= *node_size; 2348 } else { 2349 set_top(nullptr); 2350 } 2351 2352 if (top() == nullptr) { 2353 set_end(nullptr); 2354 } 2355 2356 return node; 2357 } 2358 2359 2360 FreeSpace* FreeListCategory::PickNodeFromList(int size_in_bytes, 2361 int* node_size) { 2362 FreeSpace* node = PickNodeFromList(node_size); 2363 if ((node != nullptr) && (*node_size < size_in_bytes)) { 2364 Free(node, *node_size); 2365 *node_size = 0; 2366 return nullptr; 2367 } 2368 return node; 2369 } 2370 2371 2372 FreeSpace* FreeListCategory::SearchForNodeInList(int size_in_bytes, 2373 int* node_size) { 2374 FreeSpace* prev_non_evac_node = nullptr; 2375 for (FreeSpace* cur_node = top(); cur_node != nullptr; 2376 cur_node = cur_node->next()) { 2377 int size = cur_node->size(); 2378 Page* page_for_node = Page::FromAddress(cur_node->address()); 2379 2380 if ((size >= size_in_bytes) || !page_for_node->CanAllocate()) { 2381 // The node is either large enough or contained in an evacuation 2382 // candidate. In both cases we need to unlink it from the list. 2383 available_ -= size; 2384 if (cur_node == top()) { 2385 set_top(cur_node->next()); 2386 } 2387 if (cur_node == end()) { 2388 set_end(prev_non_evac_node); 2389 } 2390 if (prev_non_evac_node != nullptr) { 2391 prev_non_evac_node->set_next(cur_node->next()); 2392 } 2393 // For evacuation candidates we continue. 2394 if (!page_for_node->CanAllocate()) { 2395 page_for_node->add_available_in_free_list(type_, -size); 2396 continue; 2397 } 2398 // Otherwise we have a large enough node and can return. 2399 *node_size = size; 2400 return cur_node; 2401 } 2402 2403 prev_non_evac_node = cur_node; 2404 } 2405 return nullptr; 2406 } 2407 2408 2409 void FreeListCategory::Free(FreeSpace* free_space, int size_in_bytes) { 2410 free_space->set_next(top()); 2411 set_top(free_space); 2412 if (end_ == NULL) { 2413 end_ = free_space; 2414 } 2415 available_ += size_in_bytes; 2416 } 2417 2418 2419 void FreeListCategory::RepairFreeList(Heap* heap) { 2420 FreeSpace* n = top(); 2421 while (n != NULL) { 2422 Map** map_location = reinterpret_cast<Map**>(n->address()); 2423 if (*map_location == NULL) { 2424 *map_location = heap->free_space_map(); 2425 } else { 2426 DCHECK(*map_location == heap->free_space_map()); 2427 } 2428 n = n->next(); 2429 } 2430 } 2431 2432 2433 FreeList::FreeList(PagedSpace* owner) 2434 : owner_(owner), 2435 wasted_bytes_(0), 2436 small_list_(this, kSmall), 2437 medium_list_(this, kMedium), 2438 large_list_(this, kLarge), 2439 huge_list_(this, kHuge) { 2440 Reset(); 2441 } 2442 2443 2444 intptr_t FreeList::Concatenate(FreeList* other) { 2445 intptr_t usable_bytes = 0; 2446 intptr_t wasted_bytes = 0; 2447 2448 // This is safe (not going to deadlock) since Concatenate operations 2449 // are never performed on the same free lists at the same time in 2450 // reverse order. Furthermore, we only lock if the PagedSpace containing 2451 // the free list is know to be globally available, i.e., not local. 2452 if (!owner()->is_local()) mutex_.Lock(); 2453 if (!other->owner()->is_local()) other->mutex()->Lock(); 2454 2455 wasted_bytes = other->wasted_bytes_; 2456 wasted_bytes_ += wasted_bytes; 2457 other->wasted_bytes_ = 0; 2458 2459 usable_bytes += small_list_.Concatenate(other->GetFreeListCategory(kSmall)); 2460 usable_bytes += medium_list_.Concatenate(other->GetFreeListCategory(kMedium)); 2461 usable_bytes += large_list_.Concatenate(other->GetFreeListCategory(kLarge)); 2462 usable_bytes += huge_list_.Concatenate(other->GetFreeListCategory(kHuge)); 2463 2464 if (!other->owner()->is_local()) other->mutex()->Unlock(); 2465 if (!owner()->is_local()) mutex_.Unlock(); 2466 return usable_bytes + wasted_bytes; 2467 } 2468 2469 2470 void FreeList::Reset() { 2471 small_list_.Reset(); 2472 medium_list_.Reset(); 2473 large_list_.Reset(); 2474 huge_list_.Reset(); 2475 ResetStats(); 2476 } 2477 2478 2479 int FreeList::Free(Address start, int size_in_bytes) { 2480 if (size_in_bytes == 0) return 0; 2481 2482 owner()->heap()->CreateFillerObjectAt(start, size_in_bytes); 2483 2484 Page* page = Page::FromAddress(start); 2485 2486 // Early return to drop too-small blocks on the floor. 2487 if (size_in_bytes <= kSmallListMin) { 2488 page->add_non_available_small_blocks(size_in_bytes); 2489 wasted_bytes_ += size_in_bytes; 2490 return size_in_bytes; 2491 } 2492 2493 FreeSpace* free_space = FreeSpace::cast(HeapObject::FromAddress(start)); 2494 // Insert other blocks at the head of a free list of the appropriate 2495 // magnitude. 2496 if (size_in_bytes <= kSmallListMax) { 2497 small_list_.Free(free_space, size_in_bytes); 2498 page->add_available_in_small_free_list(size_in_bytes); 2499 } else if (size_in_bytes <= kMediumListMax) { 2500 medium_list_.Free(free_space, size_in_bytes); 2501 page->add_available_in_medium_free_list(size_in_bytes); 2502 } else if (size_in_bytes <= kLargeListMax) { 2503 large_list_.Free(free_space, size_in_bytes); 2504 page->add_available_in_large_free_list(size_in_bytes); 2505 } else { 2506 huge_list_.Free(free_space, size_in_bytes); 2507 page->add_available_in_huge_free_list(size_in_bytes); 2508 } 2509 2510 DCHECK(IsVeryLong() || Available() == SumFreeLists()); 2511 return 0; 2512 } 2513 2514 2515 FreeSpace* FreeList::FindNodeIn(FreeListCategoryType category, int* node_size) { 2516 FreeSpace* node = GetFreeListCategory(category)->PickNodeFromList(node_size); 2517 if (node != nullptr) { 2518 Page::FromAddress(node->address()) 2519 ->add_available_in_free_list(category, -(*node_size)); 2520 DCHECK(IsVeryLong() || Available() == SumFreeLists()); 2521 } 2522 return node; 2523 } 2524 2525 2526 FreeSpace* FreeList::FindNodeFor(int size_in_bytes, int* node_size) { 2527 FreeSpace* node = nullptr; 2528 Page* page = nullptr; 2529 2530 if (size_in_bytes <= kSmallAllocationMax) { 2531 node = FindNodeIn(kSmall, node_size); 2532 if (node != nullptr) return node; 2533 } 2534 2535 if (size_in_bytes <= kMediumAllocationMax) { 2536 node = FindNodeIn(kMedium, node_size); 2537 if (node != nullptr) return node; 2538 } 2539 2540 if (size_in_bytes <= kLargeAllocationMax) { 2541 node = FindNodeIn(kLarge, node_size); 2542 if (node != nullptr) return node; 2543 } 2544 2545 node = huge_list_.SearchForNodeInList(size_in_bytes, node_size); 2546 if (node != nullptr) { 2547 page = Page::FromAddress(node->address()); 2548 page->add_available_in_large_free_list(-(*node_size)); 2549 DCHECK(IsVeryLong() || Available() == SumFreeLists()); 2550 return node; 2551 } 2552 2553 if (size_in_bytes <= kSmallListMax) { 2554 node = small_list_.PickNodeFromList(size_in_bytes, node_size); 2555 if (node != NULL) { 2556 DCHECK(size_in_bytes <= *node_size); 2557 page = Page::FromAddress(node->address()); 2558 page->add_available_in_small_free_list(-(*node_size)); 2559 } 2560 } else if (size_in_bytes <= kMediumListMax) { 2561 node = medium_list_.PickNodeFromList(size_in_bytes, node_size); 2562 if (node != NULL) { 2563 DCHECK(size_in_bytes <= *node_size); 2564 page = Page::FromAddress(node->address()); 2565 page->add_available_in_medium_free_list(-(*node_size)); 2566 } 2567 } else if (size_in_bytes <= kLargeListMax) { 2568 node = large_list_.PickNodeFromList(size_in_bytes, node_size); 2569 if (node != NULL) { 2570 DCHECK(size_in_bytes <= *node_size); 2571 page = Page::FromAddress(node->address()); 2572 page->add_available_in_large_free_list(-(*node_size)); 2573 } 2574 } 2575 2576 DCHECK(IsVeryLong() || Available() == SumFreeLists()); 2577 return node; 2578 } 2579 2580 2581 FreeSpace* FreeList::TryRemoveMemory(intptr_t hint_size_in_bytes) { 2582 hint_size_in_bytes = RoundDown(hint_size_in_bytes, kPointerSize); 2583 base::LockGuard<base::Mutex> guard(&mutex_); 2584 FreeSpace* node = nullptr; 2585 int node_size = 0; 2586 // Try to find a node that fits exactly. 2587 node = FindNodeFor(static_cast<int>(hint_size_in_bytes), &node_size); 2588 // If no node could be found get as much memory as possible. 2589 if (node == nullptr) node = FindNodeIn(kHuge, &node_size); 2590 if (node == nullptr) node = FindNodeIn(kLarge, &node_size); 2591 if (node != nullptr) { 2592 // We round up the size to (kSmallListMin + kPointerSize) to (a) have a 2593 // size larger then the minimum size required for FreeSpace, and (b) to get 2594 // a block that can actually be freed into some FreeList later on. 2595 if (hint_size_in_bytes <= kSmallListMin) { 2596 hint_size_in_bytes = kSmallListMin + kPointerSize; 2597 } 2598 // Give back left overs that were not required by {size_in_bytes}. 2599 intptr_t left_over = node_size - hint_size_in_bytes; 2600 2601 // Do not bother to return anything below {kSmallListMin} as it would be 2602 // immediately discarded anyways. 2603 if (left_over > kSmallListMin) { 2604 Free(node->address() + hint_size_in_bytes, static_cast<int>(left_over)); 2605 node->set_size(static_cast<int>(hint_size_in_bytes)); 2606 } 2607 } 2608 return node; 2609 } 2610 2611 2612 // Allocation on the old space free list. If it succeeds then a new linear 2613 // allocation space has been set up with the top and limit of the space. If 2614 // the allocation fails then NULL is returned, and the caller can perform a GC 2615 // or allocate a new page before retrying. 2616 HeapObject* FreeList::Allocate(int size_in_bytes) { 2617 DCHECK(0 < size_in_bytes); 2618 DCHECK(size_in_bytes <= kMaxBlockSize); 2619 DCHECK(IsAligned(size_in_bytes, kPointerSize)); 2620 // Don't free list allocate if there is linear space available. 2621 DCHECK(owner_->limit() - owner_->top() < size_in_bytes); 2622 2623 int old_linear_size = static_cast<int>(owner_->limit() - owner_->top()); 2624 // Mark the old linear allocation area with a free space map so it can be 2625 // skipped when scanning the heap. This also puts it back in the free list 2626 // if it is big enough. 2627 owner_->Free(owner_->top(), old_linear_size); 2628 owner_->SetTopAndLimit(nullptr, nullptr); 2629 2630 owner_->heap()->incremental_marking()->OldSpaceStep(size_in_bytes - 2631 old_linear_size); 2632 2633 int new_node_size = 0; 2634 FreeSpace* new_node = FindNodeFor(size_in_bytes, &new_node_size); 2635 if (new_node == nullptr) return nullptr; 2636 2637 int bytes_left = new_node_size - size_in_bytes; 2638 DCHECK(bytes_left >= 0); 2639 2640 #ifdef DEBUG 2641 for (int i = 0; i < size_in_bytes / kPointerSize; i++) { 2642 reinterpret_cast<Object**>(new_node->address())[i] = 2643 Smi::FromInt(kCodeZapValue); 2644 } 2645 #endif 2646 2647 // The old-space-step might have finished sweeping and restarted marking. 2648 // Verify that it did not turn the page of the new node into an evacuation 2649 // candidate. 2650 DCHECK(!MarkCompactCollector::IsOnEvacuationCandidate(new_node)); 2651 2652 const int kThreshold = IncrementalMarking::kAllocatedThreshold; 2653 2654 // Memory in the linear allocation area is counted as allocated. We may free 2655 // a little of this again immediately - see below. 2656 owner_->Allocate(new_node_size); 2657 2658 if (owner_->heap()->inline_allocation_disabled()) { 2659 // Keep the linear allocation area empty if requested to do so, just 2660 // return area back to the free list instead. 2661 owner_->Free(new_node->address() + size_in_bytes, bytes_left); 2662 DCHECK(owner_->top() == NULL && owner_->limit() == NULL); 2663 } else if (bytes_left > kThreshold && 2664 owner_->heap()->incremental_marking()->IsMarkingIncomplete() && 2665 FLAG_incremental_marking) { 2666 int linear_size = owner_->RoundSizeDownToObjectAlignment(kThreshold); 2667 // We don't want to give too large linear areas to the allocator while 2668 // incremental marking is going on, because we won't check again whether 2669 // we want to do another increment until the linear area is used up. 2670 owner_->Free(new_node->address() + size_in_bytes + linear_size, 2671 new_node_size - size_in_bytes - linear_size); 2672 owner_->SetTopAndLimit(new_node->address() + size_in_bytes, 2673 new_node->address() + size_in_bytes + linear_size); 2674 } else if (bytes_left > 0) { 2675 // Normally we give the rest of the node to the allocator as its new 2676 // linear allocation area. 2677 owner_->SetTopAndLimit(new_node->address() + size_in_bytes, 2678 new_node->address() + new_node_size); 2679 } 2680 2681 return new_node; 2682 } 2683 2684 2685 intptr_t FreeList::EvictFreeListItems(Page* p) { 2686 intptr_t sum = huge_list_.EvictFreeListItemsInList(p); 2687 if (sum < p->area_size()) { 2688 sum += small_list_.EvictFreeListItemsInList(p) + 2689 medium_list_.EvictFreeListItemsInList(p) + 2690 large_list_.EvictFreeListItemsInList(p); 2691 } 2692 return sum; 2693 } 2694 2695 2696 bool FreeList::ContainsPageFreeListItems(Page* p) { 2697 return huge_list_.EvictFreeListItemsInList(p) || 2698 small_list_.EvictFreeListItemsInList(p) || 2699 medium_list_.EvictFreeListItemsInList(p) || 2700 large_list_.EvictFreeListItemsInList(p); 2701 } 2702 2703 2704 void FreeList::RepairLists(Heap* heap) { 2705 small_list_.RepairFreeList(heap); 2706 medium_list_.RepairFreeList(heap); 2707 large_list_.RepairFreeList(heap); 2708 huge_list_.RepairFreeList(heap); 2709 } 2710 2711 2712 #ifdef DEBUG 2713 intptr_t FreeListCategory::SumFreeList() { 2714 intptr_t sum = 0; 2715 FreeSpace* cur = top(); 2716 while (cur != NULL) { 2717 DCHECK(cur->map() == cur->GetHeap()->root(Heap::kFreeSpaceMapRootIndex)); 2718 sum += cur->nobarrier_size(); 2719 cur = cur->next(); 2720 } 2721 return sum; 2722 } 2723 2724 2725 int FreeListCategory::FreeListLength() { 2726 int length = 0; 2727 FreeSpace* cur = top(); 2728 while (cur != NULL) { 2729 length++; 2730 cur = cur->next(); 2731 if (length == kVeryLongFreeList) return length; 2732 } 2733 return length; 2734 } 2735 2736 2737 bool FreeListCategory::IsVeryLong() { 2738 return FreeListLength() == kVeryLongFreeList; 2739 } 2740 2741 2742 bool FreeList::IsVeryLong() { 2743 return small_list_.IsVeryLong() || medium_list_.IsVeryLong() || 2744 large_list_.IsVeryLong() || huge_list_.IsVeryLong(); 2745 } 2746 2747 2748 // This can take a very long time because it is linear in the number of entries 2749 // on the free list, so it should not be called if FreeListLength returns 2750 // kVeryLongFreeList. 2751 intptr_t FreeList::SumFreeLists() { 2752 intptr_t sum = small_list_.SumFreeList(); 2753 sum += medium_list_.SumFreeList(); 2754 sum += large_list_.SumFreeList(); 2755 sum += huge_list_.SumFreeList(); 2756 return sum; 2757 } 2758 #endif 2759 2760 2761 // ----------------------------------------------------------------------------- 2762 // OldSpace implementation 2763 2764 void PagedSpace::PrepareForMarkCompact() { 2765 // We don't have a linear allocation area while sweeping. It will be restored 2766 // on the first allocation after the sweep. 2767 EmptyAllocationInfo(); 2768 2769 // Clear the free list before a full GC---it will be rebuilt afterward. 2770 free_list_.Reset(); 2771 } 2772 2773 2774 intptr_t PagedSpace::SizeOfObjects() { 2775 const intptr_t size = Size() - (limit() - top()); 2776 CHECK_GE(limit(), top()); 2777 CHECK_GE(size, 0); 2778 USE(size); 2779 return size; 2780 } 2781 2782 2783 // After we have booted, we have created a map which represents free space 2784 // on the heap. If there was already a free list then the elements on it 2785 // were created with the wrong FreeSpaceMap (normally NULL), so we need to 2786 // fix them. 2787 void PagedSpace::RepairFreeListsAfterDeserialization() { 2788 free_list_.RepairLists(heap()); 2789 // Each page may have a small free space that is not tracked by a free list. 2790 // Update the maps for those free space objects. 2791 PageIterator iterator(this); 2792 while (iterator.has_next()) { 2793 Page* page = iterator.next(); 2794 int size = static_cast<int>(page->non_available_small_blocks()); 2795 if (size == 0) continue; 2796 Address address = page->OffsetToAddress(Page::kPageSize - size); 2797 heap()->CreateFillerObjectAt(address, size); 2798 } 2799 } 2800 2801 2802 void PagedSpace::EvictEvacuationCandidatesFromLinearAllocationArea() { 2803 if (allocation_info_.top() >= allocation_info_.limit()) return; 2804 2805 if (!Page::FromAllocationTop(allocation_info_.top())->CanAllocate()) { 2806 // Create filler object to keep page iterable if it was iterable. 2807 int remaining = 2808 static_cast<int>(allocation_info_.limit() - allocation_info_.top()); 2809 heap()->CreateFillerObjectAt(allocation_info_.top(), remaining); 2810 allocation_info_.Reset(nullptr, nullptr); 2811 } 2812 } 2813 2814 2815 HeapObject* PagedSpace::SweepAndRetryAllocation(int size_in_bytes) { 2816 MarkCompactCollector* collector = heap()->mark_compact_collector(); 2817 if (collector->sweeping_in_progress()) { 2818 // Wait for the sweeper threads here and complete the sweeping phase. 2819 collector->EnsureSweepingCompleted(); 2820 2821 // After waiting for the sweeper threads, there may be new free-list 2822 // entries. 2823 return free_list_.Allocate(size_in_bytes); 2824 } 2825 return nullptr; 2826 } 2827 2828 2829 HeapObject* CompactionSpace::SweepAndRetryAllocation(int size_in_bytes) { 2830 MarkCompactCollector* collector = heap()->mark_compact_collector(); 2831 if (collector->sweeping_in_progress()) { 2832 collector->SweepAndRefill(this); 2833 return free_list_.Allocate(size_in_bytes); 2834 } 2835 return nullptr; 2836 } 2837 2838 2839 HeapObject* PagedSpace::SlowAllocateRaw(int size_in_bytes) { 2840 // Allocation in this space has failed. 2841 2842 MarkCompactCollector* collector = heap()->mark_compact_collector(); 2843 // Sweeping is still in progress. 2844 if (collector->sweeping_in_progress()) { 2845 // First try to refill the free-list, concurrent sweeper threads 2846 // may have freed some objects in the meantime. 2847 RefillFreeList(); 2848 2849 // Retry the free list allocation. 2850 HeapObject* object = free_list_.Allocate(size_in_bytes); 2851 if (object != NULL) return object; 2852 2853 // If sweeping is still in progress try to sweep pages on the main thread. 2854 collector->SweepInParallel(heap()->paged_space(identity()), size_in_bytes); 2855 RefillFreeList(); 2856 object = free_list_.Allocate(size_in_bytes); 2857 if (object != nullptr) return object; 2858 } 2859 2860 // Free list allocation failed and there is no next page. Fail if we have 2861 // hit the old generation size limit that should cause a garbage 2862 // collection. 2863 if (!heap()->always_allocate() && 2864 heap()->OldGenerationAllocationLimitReached()) { 2865 // If sweeper threads are active, wait for them at that point and steal 2866 // elements form their free-lists. 2867 HeapObject* object = SweepAndRetryAllocation(size_in_bytes); 2868 return object; 2869 } 2870 2871 // Try to expand the space and allocate in the new next page. 2872 if (Expand()) { 2873 DCHECK((CountTotalPages() > 1) || 2874 (size_in_bytes <= free_list_.Available())); 2875 return free_list_.Allocate(size_in_bytes); 2876 } 2877 2878 // If sweeper threads are active, wait for them at that point and steal 2879 // elements form their free-lists. Allocation may still fail their which 2880 // would indicate that there is not enough memory for the given allocation. 2881 return SweepAndRetryAllocation(size_in_bytes); 2882 } 2883 2884 2885 #ifdef DEBUG 2886 void PagedSpace::ReportCodeStatistics(Isolate* isolate) { 2887 CommentStatistic* comments_statistics = 2888 isolate->paged_space_comments_statistics(); 2889 ReportCodeKindStatistics(isolate->code_kind_statistics()); 2890 PrintF( 2891 "Code comment statistics (\" [ comment-txt : size/ " 2892 "count (average)\"):\n"); 2893 for (int i = 0; i <= CommentStatistic::kMaxComments; i++) { 2894 const CommentStatistic& cs = comments_statistics[i]; 2895 if (cs.size > 0) { 2896 PrintF(" %-30s: %10d/%6d (%d)\n", cs.comment, cs.size, cs.count, 2897 cs.size / cs.count); 2898 } 2899 } 2900 PrintF("\n"); 2901 } 2902 2903 2904 void PagedSpace::ResetCodeStatistics(Isolate* isolate) { 2905 CommentStatistic* comments_statistics = 2906 isolate->paged_space_comments_statistics(); 2907 ClearCodeKindStatistics(isolate->code_kind_statistics()); 2908 for (int i = 0; i < CommentStatistic::kMaxComments; i++) { 2909 comments_statistics[i].Clear(); 2910 } 2911 comments_statistics[CommentStatistic::kMaxComments].comment = "Unknown"; 2912 comments_statistics[CommentStatistic::kMaxComments].size = 0; 2913 comments_statistics[CommentStatistic::kMaxComments].count = 0; 2914 } 2915 2916 2917 // Adds comment to 'comment_statistics' table. Performance OK as long as 2918 // 'kMaxComments' is small 2919 static void EnterComment(Isolate* isolate, const char* comment, int delta) { 2920 CommentStatistic* comments_statistics = 2921 isolate->paged_space_comments_statistics(); 2922 // Do not count empty comments 2923 if (delta <= 0) return; 2924 CommentStatistic* cs = &comments_statistics[CommentStatistic::kMaxComments]; 2925 // Search for a free or matching entry in 'comments_statistics': 'cs' 2926 // points to result. 2927 for (int i = 0; i < CommentStatistic::kMaxComments; i++) { 2928 if (comments_statistics[i].comment == NULL) { 2929 cs = &comments_statistics[i]; 2930 cs->comment = comment; 2931 break; 2932 } else if (strcmp(comments_statistics[i].comment, comment) == 0) { 2933 cs = &comments_statistics[i]; 2934 break; 2935 } 2936 } 2937 // Update entry for 'comment' 2938 cs->size += delta; 2939 cs->count += 1; 2940 } 2941 2942 2943 // Call for each nested comment start (start marked with '[ xxx', end marked 2944 // with ']'. RelocIterator 'it' must point to a comment reloc info. 2945 static void CollectCommentStatistics(Isolate* isolate, RelocIterator* it) { 2946 DCHECK(!it->done()); 2947 DCHECK(it->rinfo()->rmode() == RelocInfo::COMMENT); 2948 const char* tmp = reinterpret_cast<const char*>(it->rinfo()->data()); 2949 if (tmp[0] != '[') { 2950 // Not a nested comment; skip 2951 return; 2952 } 2953 2954 // Search for end of nested comment or a new nested comment 2955 const char* const comment_txt = 2956 reinterpret_cast<const char*>(it->rinfo()->data()); 2957 const byte* prev_pc = it->rinfo()->pc(); 2958 int flat_delta = 0; 2959 it->next(); 2960 while (true) { 2961 // All nested comments must be terminated properly, and therefore exit 2962 // from loop. 2963 DCHECK(!it->done()); 2964 if (it->rinfo()->rmode() == RelocInfo::COMMENT) { 2965 const char* const txt = 2966 reinterpret_cast<const char*>(it->rinfo()->data()); 2967 flat_delta += static_cast<int>(it->rinfo()->pc() - prev_pc); 2968 if (txt[0] == ']') break; // End of nested comment 2969 // A new comment 2970 CollectCommentStatistics(isolate, it); 2971 // Skip code that was covered with previous comment 2972 prev_pc = it->rinfo()->pc(); 2973 } 2974 it->next(); 2975 } 2976 EnterComment(isolate, comment_txt, flat_delta); 2977 } 2978 2979 2980 // Collects code size statistics: 2981 // - by code kind 2982 // - by code comment 2983 void PagedSpace::CollectCodeStatistics() { 2984 Isolate* isolate = heap()->isolate(); 2985 HeapObjectIterator obj_it(this); 2986 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) { 2987 if (obj->IsCode()) { 2988 Code* code = Code::cast(obj); 2989 isolate->code_kind_statistics()[code->kind()] += code->Size(); 2990 RelocIterator it(code); 2991 int delta = 0; 2992 const byte* prev_pc = code->instruction_start(); 2993 while (!it.done()) { 2994 if (it.rinfo()->rmode() == RelocInfo::COMMENT) { 2995 delta += static_cast<int>(it.rinfo()->pc() - prev_pc); 2996 CollectCommentStatistics(isolate, &it); 2997 prev_pc = it.rinfo()->pc(); 2998 } 2999 it.next(); 3000 } 3001 3002 DCHECK(code->instruction_start() <= prev_pc && 3003 prev_pc <= code->instruction_end()); 3004 delta += static_cast<int>(code->instruction_end() - prev_pc); 3005 EnterComment(isolate, "NoComment", delta); 3006 } 3007 } 3008 } 3009 3010 3011 void PagedSpace::ReportStatistics() { 3012 int pct = static_cast<int>(Available() * 100 / Capacity()); 3013 PrintF(" capacity: %" V8_PTR_PREFIX 3014 "d" 3015 ", waste: %" V8_PTR_PREFIX 3016 "d" 3017 ", available: %" V8_PTR_PREFIX "d, %%%d\n", 3018 Capacity(), Waste(), Available(), pct); 3019 3020 if (heap()->mark_compact_collector()->sweeping_in_progress()) { 3021 heap()->mark_compact_collector()->EnsureSweepingCompleted(); 3022 } 3023 ClearHistograms(heap()->isolate()); 3024 HeapObjectIterator obj_it(this); 3025 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) 3026 CollectHistogramInfo(obj); 3027 ReportHistogram(heap()->isolate(), true); 3028 } 3029 #endif 3030 3031 3032 // ----------------------------------------------------------------------------- 3033 // MapSpace implementation 3034 3035 #ifdef VERIFY_HEAP 3036 void MapSpace::VerifyObject(HeapObject* object) { CHECK(object->IsMap()); } 3037 #endif 3038 3039 3040 // ----------------------------------------------------------------------------- 3041 // LargeObjectIterator 3042 3043 LargeObjectIterator::LargeObjectIterator(LargeObjectSpace* space) { 3044 current_ = space->first_page_; 3045 } 3046 3047 3048 HeapObject* LargeObjectIterator::Next() { 3049 if (current_ == NULL) return NULL; 3050 3051 HeapObject* object = current_->GetObject(); 3052 current_ = current_->next_page(); 3053 return object; 3054 } 3055 3056 3057 // ----------------------------------------------------------------------------- 3058 // LargeObjectSpace 3059 3060 3061 LargeObjectSpace::LargeObjectSpace(Heap* heap, AllocationSpace id) 3062 : Space(heap, id, NOT_EXECUTABLE), // Managed on a per-allocation basis 3063 first_page_(NULL), 3064 size_(0), 3065 page_count_(0), 3066 objects_size_(0), 3067 chunk_map_(HashMap::PointersMatch, 1024) {} 3068 3069 3070 LargeObjectSpace::~LargeObjectSpace() {} 3071 3072 3073 bool LargeObjectSpace::SetUp() { 3074 first_page_ = NULL; 3075 size_ = 0; 3076 page_count_ = 0; 3077 objects_size_ = 0; 3078 chunk_map_.Clear(); 3079 return true; 3080 } 3081 3082 3083 void LargeObjectSpace::TearDown() { 3084 while (first_page_ != NULL) { 3085 LargePage* page = first_page_; 3086 first_page_ = first_page_->next_page(); 3087 LOG(heap()->isolate(), DeleteEvent("LargeObjectChunk", page->address())); 3088 3089 ObjectSpace space = static_cast<ObjectSpace>(1 << identity()); 3090 heap()->isolate()->memory_allocator()->PerformAllocationCallback( 3091 space, kAllocationActionFree, page->size()); 3092 heap()->isolate()->memory_allocator()->Free(page); 3093 } 3094 SetUp(); 3095 } 3096 3097 3098 AllocationResult LargeObjectSpace::AllocateRaw(int object_size, 3099 Executability executable) { 3100 // Check if we want to force a GC before growing the old space further. 3101 // If so, fail the allocation. 3102 if (!heap()->CanExpandOldGeneration(object_size)) { 3103 return AllocationResult::Retry(identity()); 3104 } 3105 3106 LargePage* page = heap()->isolate()->memory_allocator()->AllocateLargePage( 3107 object_size, this, executable); 3108 if (page == NULL) return AllocationResult::Retry(identity()); 3109 DCHECK(page->area_size() >= object_size); 3110 3111 size_ += static_cast<int>(page->size()); 3112 AccountCommitted(static_cast<intptr_t>(page->size())); 3113 objects_size_ += object_size; 3114 page_count_++; 3115 page->set_next_page(first_page_); 3116 first_page_ = page; 3117 3118 // Register all MemoryChunk::kAlignment-aligned chunks covered by 3119 // this large page in the chunk map. 3120 uintptr_t base = reinterpret_cast<uintptr_t>(page) / MemoryChunk::kAlignment; 3121 uintptr_t limit = base + (page->size() - 1) / MemoryChunk::kAlignment; 3122 for (uintptr_t key = base; key <= limit; key++) { 3123 HashMap::Entry* entry = chunk_map_.LookupOrInsert( 3124 reinterpret_cast<void*>(key), static_cast<uint32_t>(key)); 3125 DCHECK(entry != NULL); 3126 entry->value = page; 3127 } 3128 3129 HeapObject* object = page->GetObject(); 3130 3131 MSAN_ALLOCATED_UNINITIALIZED_MEMORY(object->address(), object_size); 3132 3133 if (Heap::ShouldZapGarbage()) { 3134 // Make the object consistent so the heap can be verified in OldSpaceStep. 3135 // We only need to do this in debug builds or if verify_heap is on. 3136 reinterpret_cast<Object**>(object->address())[0] = 3137 heap()->fixed_array_map(); 3138 reinterpret_cast<Object**>(object->address())[1] = Smi::FromInt(0); 3139 } 3140 3141 heap()->incremental_marking()->OldSpaceStep(object_size); 3142 return object; 3143 } 3144 3145 3146 size_t LargeObjectSpace::CommittedPhysicalMemory() { 3147 if (!base::VirtualMemory::HasLazyCommits()) return CommittedMemory(); 3148 size_t size = 0; 3149 LargePage* current = first_page_; 3150 while (current != NULL) { 3151 size += current->CommittedPhysicalMemory(); 3152 current = current->next_page(); 3153 } 3154 return size; 3155 } 3156 3157 3158 // GC support 3159 Object* LargeObjectSpace::FindObject(Address a) { 3160 LargePage* page = FindPage(a); 3161 if (page != NULL) { 3162 return page->GetObject(); 3163 } 3164 return Smi::FromInt(0); // Signaling not found. 3165 } 3166 3167 3168 LargePage* LargeObjectSpace::FindPage(Address a) { 3169 uintptr_t key = reinterpret_cast<uintptr_t>(a) / MemoryChunk::kAlignment; 3170 HashMap::Entry* e = chunk_map_.Lookup(reinterpret_cast<void*>(key), 3171 static_cast<uint32_t>(key)); 3172 if (e != NULL) { 3173 DCHECK(e->value != NULL); 3174 LargePage* page = reinterpret_cast<LargePage*>(e->value); 3175 DCHECK(page->is_valid()); 3176 if (page->Contains(a)) { 3177 return page; 3178 } 3179 } 3180 return NULL; 3181 } 3182 3183 3184 void LargeObjectSpace::ClearMarkingStateOfLiveObjects() { 3185 LargePage* current = first_page_; 3186 while (current != NULL) { 3187 HeapObject* object = current->GetObject(); 3188 MarkBit mark_bit = Marking::MarkBitFrom(object); 3189 DCHECK(Marking::IsBlack(mark_bit)); 3190 Marking::BlackToWhite(mark_bit); 3191 Page::FromAddress(object->address())->ResetProgressBar(); 3192 Page::FromAddress(object->address())->ResetLiveBytes(); 3193 current = current->next_page(); 3194 } 3195 } 3196 3197 3198 void LargeObjectSpace::FreeUnmarkedObjects() { 3199 LargePage* previous = NULL; 3200 LargePage* current = first_page_; 3201 while (current != NULL) { 3202 HeapObject* object = current->GetObject(); 3203 MarkBit mark_bit = Marking::MarkBitFrom(object); 3204 DCHECK(!Marking::IsGrey(mark_bit)); 3205 if (Marking::IsBlack(mark_bit)) { 3206 previous = current; 3207 current = current->next_page(); 3208 } else { 3209 LargePage* page = current; 3210 // Cut the chunk out from the chunk list. 3211 current = current->next_page(); 3212 if (previous == NULL) { 3213 first_page_ = current; 3214 } else { 3215 previous->set_next_page(current); 3216 } 3217 3218 // Free the chunk. 3219 heap()->mark_compact_collector()->ReportDeleteIfNeeded(object, 3220 heap()->isolate()); 3221 size_ -= static_cast<int>(page->size()); 3222 AccountUncommitted(static_cast<intptr_t>(page->size())); 3223 objects_size_ -= object->Size(); 3224 page_count_--; 3225 3226 // Remove entries belonging to this page. 3227 // Use variable alignment to help pass length check (<= 80 characters) 3228 // of single line in tools/presubmit.py. 3229 const intptr_t alignment = MemoryChunk::kAlignment; 3230 uintptr_t base = reinterpret_cast<uintptr_t>(page) / alignment; 3231 uintptr_t limit = base + (page->size() - 1) / alignment; 3232 for (uintptr_t key = base; key <= limit; key++) { 3233 chunk_map_.Remove(reinterpret_cast<void*>(key), 3234 static_cast<uint32_t>(key)); 3235 } 3236 3237 heap()->QueueMemoryChunkForFree(page); 3238 } 3239 } 3240 } 3241 3242 3243 bool LargeObjectSpace::Contains(HeapObject* object) { 3244 Address address = object->address(); 3245 MemoryChunk* chunk = MemoryChunk::FromAddress(address); 3246 3247 bool owned = (chunk->owner() == this); 3248 3249 SLOW_DCHECK(!owned || FindObject(address)->IsHeapObject()); 3250 3251 return owned; 3252 } 3253 3254 3255 bool LargeObjectSpace::Contains(Address address) { 3256 return FindPage(address) != NULL; 3257 } 3258 3259 3260 #ifdef VERIFY_HEAP 3261 // We do not assume that the large object iterator works, because it depends 3262 // on the invariants we are checking during verification. 3263 void LargeObjectSpace::Verify() { 3264 for (LargePage* chunk = first_page_; chunk != NULL; 3265 chunk = chunk->next_page()) { 3266 // Each chunk contains an object that starts at the large object page's 3267 // object area start. 3268 HeapObject* object = chunk->GetObject(); 3269 Page* page = Page::FromAddress(object->address()); 3270 CHECK(object->address() == page->area_start()); 3271 3272 // The first word should be a map, and we expect all map pointers to be 3273 // in map space. 3274 Map* map = object->map(); 3275 CHECK(map->IsMap()); 3276 CHECK(heap()->map_space()->Contains(map)); 3277 3278 // We have only code, sequential strings, external strings 3279 // (sequential strings that have been morphed into external 3280 // strings), fixed arrays, byte arrays, and constant pool arrays in the 3281 // large object space. 3282 CHECK(object->IsCode() || object->IsSeqString() || 3283 object->IsExternalString() || object->IsFixedArray() || 3284 object->IsFixedDoubleArray() || object->IsByteArray()); 3285 3286 // The object itself should look OK. 3287 object->ObjectVerify(); 3288 3289 // Byte arrays and strings don't have interior pointers. 3290 if (object->IsCode()) { 3291 VerifyPointersVisitor code_visitor; 3292 object->IterateBody(map->instance_type(), object->Size(), &code_visitor); 3293 } else if (object->IsFixedArray()) { 3294 FixedArray* array = FixedArray::cast(object); 3295 for (int j = 0; j < array->length(); j++) { 3296 Object* element = array->get(j); 3297 if (element->IsHeapObject()) { 3298 HeapObject* element_object = HeapObject::cast(element); 3299 CHECK(heap()->Contains(element_object)); 3300 CHECK(element_object->map()->IsMap()); 3301 } 3302 } 3303 } 3304 } 3305 } 3306 #endif 3307 3308 3309 #ifdef DEBUG 3310 void LargeObjectSpace::Print() { 3311 OFStream os(stdout); 3312 LargeObjectIterator it(this); 3313 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 3314 obj->Print(os); 3315 } 3316 } 3317 3318 3319 void LargeObjectSpace::ReportStatistics() { 3320 PrintF(" size: %" V8_PTR_PREFIX "d\n", size_); 3321 int num_objects = 0; 3322 ClearHistograms(heap()->isolate()); 3323 LargeObjectIterator it(this); 3324 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 3325 num_objects++; 3326 CollectHistogramInfo(obj); 3327 } 3328 3329 PrintF( 3330 " number of objects %d, " 3331 "size of objects %" V8_PTR_PREFIX "d\n", 3332 num_objects, objects_size_); 3333 if (num_objects > 0) ReportHistogram(heap()->isolate(), false); 3334 } 3335 3336 3337 void LargeObjectSpace::CollectCodeStatistics() { 3338 Isolate* isolate = heap()->isolate(); 3339 LargeObjectIterator obj_it(this); 3340 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) { 3341 if (obj->IsCode()) { 3342 Code* code = Code::cast(obj); 3343 isolate->code_kind_statistics()[code->kind()] += code->Size(); 3344 } 3345 } 3346 } 3347 3348 3349 void Page::Print() { 3350 // Make a best-effort to print the objects in the page. 3351 PrintF("Page@%p in %s\n", this->address(), 3352 AllocationSpaceName(this->owner()->identity())); 3353 printf(" --------------------------------------\n"); 3354 HeapObjectIterator objects(this); 3355 unsigned mark_size = 0; 3356 for (HeapObject* object = objects.Next(); object != NULL; 3357 object = objects.Next()) { 3358 bool is_marked = Marking::IsBlackOrGrey(Marking::MarkBitFrom(object)); 3359 PrintF(" %c ", (is_marked ? '!' : ' ')); // Indent a little. 3360 if (is_marked) { 3361 mark_size += object->Size(); 3362 } 3363 object->ShortPrint(); 3364 PrintF("\n"); 3365 } 3366 printf(" --------------------------------------\n"); 3367 printf(" Marked: %x, LiveCount: %x\n", mark_size, LiveBytes()); 3368 } 3369 3370 #endif // DEBUG 3371 } // namespace internal 3372 } // namespace v8 3373