1 // Copyright 2011 the V8 project authors. All rights reserved. 2 // Redistribution and use in source and binary forms, with or without 3 // modification, are permitted provided that the following conditions are 4 // met: 5 // 6 // * Redistributions of source code must retain the above copyright 7 // notice, this list of conditions and the following disclaimer. 8 // * Redistributions in binary form must reproduce the above 9 // copyright notice, this list of conditions and the following 10 // disclaimer in the documentation and/or other materials provided 11 // with the distribution. 12 // * Neither the name of Google Inc. nor the names of its 13 // contributors may be used to endorse or promote products derived 14 // from this software without specific prior written permission. 15 // 16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 20 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 21 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 22 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 26 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 28 #include "v8.h" 29 30 #include "macro-assembler.h" 31 #include "mark-compact.h" 32 #include "msan.h" 33 #include "platform.h" 34 35 namespace v8 { 36 namespace internal { 37 38 39 // ---------------------------------------------------------------------------- 40 // HeapObjectIterator 41 42 HeapObjectIterator::HeapObjectIterator(PagedSpace* space) { 43 // You can't actually iterate over the anchor page. It is not a real page, 44 // just an anchor for the double linked page list. Initialize as if we have 45 // reached the end of the anchor page, then the first iteration will move on 46 // to the first page. 47 Initialize(space, 48 NULL, 49 NULL, 50 kAllPagesInSpace, 51 NULL); 52 } 53 54 55 HeapObjectIterator::HeapObjectIterator(PagedSpace* space, 56 HeapObjectCallback size_func) { 57 // You can't actually iterate over the anchor page. It is not a real page, 58 // just an anchor for the double linked page list. Initialize the current 59 // address and end as NULL, then the first iteration will move on 60 // to the first page. 61 Initialize(space, 62 NULL, 63 NULL, 64 kAllPagesInSpace, 65 size_func); 66 } 67 68 69 HeapObjectIterator::HeapObjectIterator(Page* page, 70 HeapObjectCallback size_func) { 71 Space* owner = page->owner(); 72 ASSERT(owner == page->heap()->old_pointer_space() || 73 owner == page->heap()->old_data_space() || 74 owner == page->heap()->map_space() || 75 owner == page->heap()->cell_space() || 76 owner == page->heap()->property_cell_space() || 77 owner == page->heap()->code_space()); 78 Initialize(reinterpret_cast<PagedSpace*>(owner), 79 page->area_start(), 80 page->area_end(), 81 kOnePageOnly, 82 size_func); 83 ASSERT(page->WasSweptPrecisely()); 84 } 85 86 87 void HeapObjectIterator::Initialize(PagedSpace* space, 88 Address cur, Address end, 89 HeapObjectIterator::PageMode mode, 90 HeapObjectCallback size_f) { 91 // Check that we actually can iterate this space. 92 ASSERT(!space->was_swept_conservatively()); 93 94 space_ = space; 95 cur_addr_ = cur; 96 cur_end_ = end; 97 page_mode_ = mode; 98 size_func_ = size_f; 99 } 100 101 102 // We have hit the end of the page and should advance to the next block of 103 // objects. This happens at the end of the page. 104 bool HeapObjectIterator::AdvanceToNextPage() { 105 ASSERT(cur_addr_ == cur_end_); 106 if (page_mode_ == kOnePageOnly) return false; 107 Page* cur_page; 108 if (cur_addr_ == NULL) { 109 cur_page = space_->anchor(); 110 } else { 111 cur_page = Page::FromAddress(cur_addr_ - 1); 112 ASSERT(cur_addr_ == cur_page->area_end()); 113 } 114 cur_page = cur_page->next_page(); 115 if (cur_page == space_->anchor()) return false; 116 cur_addr_ = cur_page->area_start(); 117 cur_end_ = cur_page->area_end(); 118 ASSERT(cur_page->WasSweptPrecisely()); 119 return true; 120 } 121 122 123 // ----------------------------------------------------------------------------- 124 // CodeRange 125 126 127 CodeRange::CodeRange(Isolate* isolate) 128 : isolate_(isolate), 129 code_range_(NULL), 130 free_list_(0), 131 allocation_list_(0), 132 current_allocation_block_index_(0) { 133 } 134 135 136 bool CodeRange::SetUp(const size_t requested) { 137 ASSERT(code_range_ == NULL); 138 139 code_range_ = new VirtualMemory(requested); 140 CHECK(code_range_ != NULL); 141 if (!code_range_->IsReserved()) { 142 delete code_range_; 143 code_range_ = NULL; 144 return false; 145 } 146 147 // We are sure that we have mapped a block of requested addresses. 148 ASSERT(code_range_->size() == requested); 149 LOG(isolate_, NewEvent("CodeRange", code_range_->address(), requested)); 150 Address base = reinterpret_cast<Address>(code_range_->address()); 151 Address aligned_base = 152 RoundUp(reinterpret_cast<Address>(code_range_->address()), 153 MemoryChunk::kAlignment); 154 size_t size = code_range_->size() - (aligned_base - base); 155 allocation_list_.Add(FreeBlock(aligned_base, size)); 156 current_allocation_block_index_ = 0; 157 return true; 158 } 159 160 161 int CodeRange::CompareFreeBlockAddress(const FreeBlock* left, 162 const FreeBlock* right) { 163 // The entire point of CodeRange is that the difference between two 164 // addresses in the range can be represented as a signed 32-bit int, 165 // so the cast is semantically correct. 166 return static_cast<int>(left->start - right->start); 167 } 168 169 170 void CodeRange::GetNextAllocationBlock(size_t requested) { 171 for (current_allocation_block_index_++; 172 current_allocation_block_index_ < allocation_list_.length(); 173 current_allocation_block_index_++) { 174 if (requested <= allocation_list_[current_allocation_block_index_].size) { 175 return; // Found a large enough allocation block. 176 } 177 } 178 179 // Sort and merge the free blocks on the free list and the allocation list. 180 free_list_.AddAll(allocation_list_); 181 allocation_list_.Clear(); 182 free_list_.Sort(&CompareFreeBlockAddress); 183 for (int i = 0; i < free_list_.length();) { 184 FreeBlock merged = free_list_[i]; 185 i++; 186 // Add adjacent free blocks to the current merged block. 187 while (i < free_list_.length() && 188 free_list_[i].start == merged.start + merged.size) { 189 merged.size += free_list_[i].size; 190 i++; 191 } 192 if (merged.size > 0) { 193 allocation_list_.Add(merged); 194 } 195 } 196 free_list_.Clear(); 197 198 for (current_allocation_block_index_ = 0; 199 current_allocation_block_index_ < allocation_list_.length(); 200 current_allocation_block_index_++) { 201 if (requested <= allocation_list_[current_allocation_block_index_].size) { 202 return; // Found a large enough allocation block. 203 } 204 } 205 206 // Code range is full or too fragmented. 207 V8::FatalProcessOutOfMemory("CodeRange::GetNextAllocationBlock"); 208 } 209 210 211 Address CodeRange::AllocateRawMemory(const size_t requested_size, 212 const size_t commit_size, 213 size_t* allocated) { 214 ASSERT(commit_size <= requested_size); 215 ASSERT(current_allocation_block_index_ < allocation_list_.length()); 216 if (requested_size > allocation_list_[current_allocation_block_index_].size) { 217 // Find an allocation block large enough. This function call may 218 // call V8::FatalProcessOutOfMemory if it cannot find a large enough block. 219 GetNextAllocationBlock(requested_size); 220 } 221 // Commit the requested memory at the start of the current allocation block. 222 size_t aligned_requested = RoundUp(requested_size, MemoryChunk::kAlignment); 223 FreeBlock current = allocation_list_[current_allocation_block_index_]; 224 if (aligned_requested >= (current.size - Page::kPageSize)) { 225 // Don't leave a small free block, useless for a large object or chunk. 226 *allocated = current.size; 227 } else { 228 *allocated = aligned_requested; 229 } 230 ASSERT(*allocated <= current.size); 231 ASSERT(IsAddressAligned(current.start, MemoryChunk::kAlignment)); 232 if (!isolate_->memory_allocator()->CommitExecutableMemory(code_range_, 233 current.start, 234 commit_size, 235 *allocated)) { 236 *allocated = 0; 237 return NULL; 238 } 239 allocation_list_[current_allocation_block_index_].start += *allocated; 240 allocation_list_[current_allocation_block_index_].size -= *allocated; 241 if (*allocated == current.size) { 242 GetNextAllocationBlock(0); // This block is used up, get the next one. 243 } 244 return current.start; 245 } 246 247 248 bool CodeRange::CommitRawMemory(Address start, size_t length) { 249 return isolate_->memory_allocator()->CommitMemory(start, length, EXECUTABLE); 250 } 251 252 253 bool CodeRange::UncommitRawMemory(Address start, size_t length) { 254 return code_range_->Uncommit(start, length); 255 } 256 257 258 void CodeRange::FreeRawMemory(Address address, size_t length) { 259 ASSERT(IsAddressAligned(address, MemoryChunk::kAlignment)); 260 free_list_.Add(FreeBlock(address, length)); 261 code_range_->Uncommit(address, length); 262 } 263 264 265 void CodeRange::TearDown() { 266 delete code_range_; // Frees all memory in the virtual memory range. 267 code_range_ = NULL; 268 free_list_.Free(); 269 allocation_list_.Free(); 270 } 271 272 273 // ----------------------------------------------------------------------------- 274 // MemoryAllocator 275 // 276 277 MemoryAllocator::MemoryAllocator(Isolate* isolate) 278 : isolate_(isolate), 279 capacity_(0), 280 capacity_executable_(0), 281 size_(0), 282 size_executable_(0), 283 lowest_ever_allocated_(reinterpret_cast<void*>(-1)), 284 highest_ever_allocated_(reinterpret_cast<void*>(0)) { 285 } 286 287 288 bool MemoryAllocator::SetUp(intptr_t capacity, intptr_t capacity_executable) { 289 capacity_ = RoundUp(capacity, Page::kPageSize); 290 capacity_executable_ = RoundUp(capacity_executable, Page::kPageSize); 291 ASSERT_GE(capacity_, capacity_executable_); 292 293 size_ = 0; 294 size_executable_ = 0; 295 296 return true; 297 } 298 299 300 void MemoryAllocator::TearDown() { 301 // Check that spaces were torn down before MemoryAllocator. 302 ASSERT(size_ == 0); 303 // TODO(gc) this will be true again when we fix FreeMemory. 304 // ASSERT(size_executable_ == 0); 305 capacity_ = 0; 306 capacity_executable_ = 0; 307 } 308 309 310 bool MemoryAllocator::CommitMemory(Address base, 311 size_t size, 312 Executability executable) { 313 if (!VirtualMemory::CommitRegion(base, size, executable == EXECUTABLE)) { 314 return false; 315 } 316 UpdateAllocatedSpaceLimits(base, base + size); 317 return true; 318 } 319 320 321 void MemoryAllocator::FreeMemory(VirtualMemory* reservation, 322 Executability executable) { 323 // TODO(gc) make code_range part of memory allocator? 324 ASSERT(reservation->IsReserved()); 325 size_t size = reservation->size(); 326 ASSERT(size_ >= size); 327 size_ -= size; 328 329 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 330 331 if (executable == EXECUTABLE) { 332 ASSERT(size_executable_ >= size); 333 size_executable_ -= size; 334 } 335 // Code which is part of the code-range does not have its own VirtualMemory. 336 ASSERT(!isolate_->code_range()->contains( 337 static_cast<Address>(reservation->address()))); 338 ASSERT(executable == NOT_EXECUTABLE || !isolate_->code_range()->exists()); 339 reservation->Release(); 340 } 341 342 343 void MemoryAllocator::FreeMemory(Address base, 344 size_t size, 345 Executability executable) { 346 // TODO(gc) make code_range part of memory allocator? 347 ASSERT(size_ >= size); 348 size_ -= size; 349 350 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 351 352 if (executable == EXECUTABLE) { 353 ASSERT(size_executable_ >= size); 354 size_executable_ -= size; 355 } 356 if (isolate_->code_range()->contains(static_cast<Address>(base))) { 357 ASSERT(executable == EXECUTABLE); 358 isolate_->code_range()->FreeRawMemory(base, size); 359 } else { 360 ASSERT(executable == NOT_EXECUTABLE || !isolate_->code_range()->exists()); 361 bool result = VirtualMemory::ReleaseRegion(base, size); 362 USE(result); 363 ASSERT(result); 364 } 365 } 366 367 368 Address MemoryAllocator::ReserveAlignedMemory(size_t size, 369 size_t alignment, 370 VirtualMemory* controller) { 371 VirtualMemory reservation(size, alignment); 372 373 if (!reservation.IsReserved()) return NULL; 374 size_ += reservation.size(); 375 Address base = RoundUp(static_cast<Address>(reservation.address()), 376 alignment); 377 controller->TakeControl(&reservation); 378 return base; 379 } 380 381 382 Address MemoryAllocator::AllocateAlignedMemory(size_t reserve_size, 383 size_t commit_size, 384 size_t alignment, 385 Executability executable, 386 VirtualMemory* controller) { 387 ASSERT(commit_size <= reserve_size); 388 VirtualMemory reservation; 389 Address base = ReserveAlignedMemory(reserve_size, alignment, &reservation); 390 if (base == NULL) return NULL; 391 392 if (executable == EXECUTABLE) { 393 if (!CommitExecutableMemory(&reservation, 394 base, 395 commit_size, 396 reserve_size)) { 397 base = NULL; 398 } 399 } else { 400 if (reservation.Commit(base, commit_size, false)) { 401 UpdateAllocatedSpaceLimits(base, base + commit_size); 402 } else { 403 base = NULL; 404 } 405 } 406 407 if (base == NULL) { 408 // Failed to commit the body. Release the mapping and any partially 409 // commited regions inside it. 410 reservation.Release(); 411 return NULL; 412 } 413 414 controller->TakeControl(&reservation); 415 return base; 416 } 417 418 419 void Page::InitializeAsAnchor(PagedSpace* owner) { 420 set_owner(owner); 421 set_prev_page(this); 422 set_next_page(this); 423 } 424 425 426 NewSpacePage* NewSpacePage::Initialize(Heap* heap, 427 Address start, 428 SemiSpace* semi_space) { 429 Address area_start = start + NewSpacePage::kObjectStartOffset; 430 Address area_end = start + Page::kPageSize; 431 432 MemoryChunk* chunk = MemoryChunk::Initialize(heap, 433 start, 434 Page::kPageSize, 435 area_start, 436 area_end, 437 NOT_EXECUTABLE, 438 semi_space); 439 chunk->set_next_chunk(NULL); 440 chunk->set_prev_chunk(NULL); 441 chunk->initialize_scan_on_scavenge(true); 442 bool in_to_space = (semi_space->id() != kFromSpace); 443 chunk->SetFlag(in_to_space ? MemoryChunk::IN_TO_SPACE 444 : MemoryChunk::IN_FROM_SPACE); 445 ASSERT(!chunk->IsFlagSet(in_to_space ? MemoryChunk::IN_FROM_SPACE 446 : MemoryChunk::IN_TO_SPACE)); 447 NewSpacePage* page = static_cast<NewSpacePage*>(chunk); 448 heap->incremental_marking()->SetNewSpacePageFlags(page); 449 return page; 450 } 451 452 453 void NewSpacePage::InitializeAsAnchor(SemiSpace* semi_space) { 454 set_owner(semi_space); 455 set_next_chunk(this); 456 set_prev_chunk(this); 457 // Flags marks this invalid page as not being in new-space. 458 // All real new-space pages will be in new-space. 459 SetFlags(0, ~0); 460 } 461 462 463 MemoryChunk* MemoryChunk::Initialize(Heap* heap, 464 Address base, 465 size_t size, 466 Address area_start, 467 Address area_end, 468 Executability executable, 469 Space* owner) { 470 MemoryChunk* chunk = FromAddress(base); 471 472 ASSERT(base == chunk->address()); 473 474 chunk->heap_ = heap; 475 chunk->size_ = size; 476 chunk->area_start_ = area_start; 477 chunk->area_end_ = area_end; 478 chunk->flags_ = 0; 479 chunk->set_owner(owner); 480 chunk->InitializeReservedMemory(); 481 chunk->slots_buffer_ = NULL; 482 chunk->skip_list_ = NULL; 483 chunk->write_barrier_counter_ = kWriteBarrierCounterGranularity; 484 chunk->progress_bar_ = 0; 485 chunk->high_water_mark_ = static_cast<int>(area_start - base); 486 chunk->parallel_sweeping_ = 0; 487 chunk->available_in_small_free_list_ = 0; 488 chunk->available_in_medium_free_list_ = 0; 489 chunk->available_in_large_free_list_ = 0; 490 chunk->available_in_huge_free_list_ = 0; 491 chunk->non_available_small_blocks_ = 0; 492 chunk->ResetLiveBytes(); 493 Bitmap::Clear(chunk); 494 chunk->initialize_scan_on_scavenge(false); 495 chunk->SetFlag(WAS_SWEPT_PRECISELY); 496 497 ASSERT(OFFSET_OF(MemoryChunk, flags_) == kFlagsOffset); 498 ASSERT(OFFSET_OF(MemoryChunk, live_byte_count_) == kLiveBytesOffset); 499 500 if (executable == EXECUTABLE) { 501 chunk->SetFlag(IS_EXECUTABLE); 502 } 503 504 if (owner == heap->old_data_space()) { 505 chunk->SetFlag(CONTAINS_ONLY_DATA); 506 } 507 508 return chunk; 509 } 510 511 512 // Commit MemoryChunk area to the requested size. 513 bool MemoryChunk::CommitArea(size_t requested) { 514 size_t guard_size = IsFlagSet(IS_EXECUTABLE) ? 515 MemoryAllocator::CodePageGuardSize() : 0; 516 size_t header_size = area_start() - address() - guard_size; 517 size_t commit_size = RoundUp(header_size + requested, OS::CommitPageSize()); 518 size_t committed_size = RoundUp(header_size + (area_end() - area_start()), 519 OS::CommitPageSize()); 520 521 if (commit_size > committed_size) { 522 // Commit size should be less or equal than the reserved size. 523 ASSERT(commit_size <= size() - 2 * guard_size); 524 // Append the committed area. 525 Address start = address() + committed_size + guard_size; 526 size_t length = commit_size - committed_size; 527 if (reservation_.IsReserved()) { 528 Executability executable = IsFlagSet(IS_EXECUTABLE) 529 ? EXECUTABLE : NOT_EXECUTABLE; 530 if (!heap()->isolate()->memory_allocator()->CommitMemory( 531 start, length, executable)) { 532 return false; 533 } 534 } else { 535 CodeRange* code_range = heap_->isolate()->code_range(); 536 ASSERT(code_range->exists() && IsFlagSet(IS_EXECUTABLE)); 537 if (!code_range->CommitRawMemory(start, length)) return false; 538 } 539 540 if (Heap::ShouldZapGarbage()) { 541 heap_->isolate()->memory_allocator()->ZapBlock(start, length); 542 } 543 } else if (commit_size < committed_size) { 544 ASSERT(commit_size > 0); 545 // Shrink the committed area. 546 size_t length = committed_size - commit_size; 547 Address start = address() + committed_size + guard_size - length; 548 if (reservation_.IsReserved()) { 549 if (!reservation_.Uncommit(start, length)) return false; 550 } else { 551 CodeRange* code_range = heap_->isolate()->code_range(); 552 ASSERT(code_range->exists() && IsFlagSet(IS_EXECUTABLE)); 553 if (!code_range->UncommitRawMemory(start, length)) return false; 554 } 555 } 556 557 area_end_ = area_start_ + requested; 558 return true; 559 } 560 561 562 void MemoryChunk::InsertAfter(MemoryChunk* other) { 563 next_chunk_ = other->next_chunk_; 564 prev_chunk_ = other; 565 566 // This memory barrier is needed since concurrent sweeper threads may iterate 567 // over the list of pages while a new page is inserted. 568 // TODO(hpayer): find a cleaner way to guarantee that the page list can be 569 // expanded concurrently 570 MemoryBarrier(); 571 572 // The following two write operations can take effect in arbitrary order 573 // since pages are always iterated by the sweeper threads in LIFO order, i.e, 574 // the inserted page becomes visible for the sweeper threads after 575 // other->next_chunk_ = this; 576 other->next_chunk_->prev_chunk_ = this; 577 other->next_chunk_ = this; 578 } 579 580 581 void MemoryChunk::Unlink() { 582 if (!InNewSpace() && IsFlagSet(SCAN_ON_SCAVENGE)) { 583 heap_->decrement_scan_on_scavenge_pages(); 584 ClearFlag(SCAN_ON_SCAVENGE); 585 } 586 next_chunk_->prev_chunk_ = prev_chunk_; 587 prev_chunk_->next_chunk_ = next_chunk_; 588 prev_chunk_ = NULL; 589 next_chunk_ = NULL; 590 } 591 592 593 MemoryChunk* MemoryAllocator::AllocateChunk(intptr_t reserve_area_size, 594 intptr_t commit_area_size, 595 Executability executable, 596 Space* owner) { 597 ASSERT(commit_area_size <= reserve_area_size); 598 599 size_t chunk_size; 600 Heap* heap = isolate_->heap(); 601 Address base = NULL; 602 VirtualMemory reservation; 603 Address area_start = NULL; 604 Address area_end = NULL; 605 606 // 607 // MemoryChunk layout: 608 // 609 // Executable 610 // +----------------------------+<- base aligned with MemoryChunk::kAlignment 611 // | Header | 612 // +----------------------------+<- base + CodePageGuardStartOffset 613 // | Guard | 614 // +----------------------------+<- area_start_ 615 // | Area | 616 // +----------------------------+<- area_end_ (area_start + commit_area_size) 617 // | Committed but not used | 618 // +----------------------------+<- aligned at OS page boundary 619 // | Reserved but not committed | 620 // +----------------------------+<- aligned at OS page boundary 621 // | Guard | 622 // +----------------------------+<- base + chunk_size 623 // 624 // Non-executable 625 // +----------------------------+<- base aligned with MemoryChunk::kAlignment 626 // | Header | 627 // +----------------------------+<- area_start_ (base + kObjectStartOffset) 628 // | Area | 629 // +----------------------------+<- area_end_ (area_start + commit_area_size) 630 // | Committed but not used | 631 // +----------------------------+<- aligned at OS page boundary 632 // | Reserved but not committed | 633 // +----------------------------+<- base + chunk_size 634 // 635 636 if (executable == EXECUTABLE) { 637 chunk_size = RoundUp(CodePageAreaStartOffset() + reserve_area_size, 638 OS::CommitPageSize()) + CodePageGuardSize(); 639 640 // Check executable memory limit. 641 if (size_executable_ + chunk_size > capacity_executable_) { 642 LOG(isolate_, 643 StringEvent("MemoryAllocator::AllocateRawMemory", 644 "V8 Executable Allocation capacity exceeded")); 645 return NULL; 646 } 647 648 // Size of header (not executable) plus area (executable). 649 size_t commit_size = RoundUp(CodePageGuardStartOffset() + commit_area_size, 650 OS::CommitPageSize()); 651 // Allocate executable memory either from code range or from the 652 // OS. 653 if (isolate_->code_range()->exists()) { 654 base = isolate_->code_range()->AllocateRawMemory(chunk_size, 655 commit_size, 656 &chunk_size); 657 ASSERT(IsAligned(reinterpret_cast<intptr_t>(base), 658 MemoryChunk::kAlignment)); 659 if (base == NULL) return NULL; 660 size_ += chunk_size; 661 // Update executable memory size. 662 size_executable_ += chunk_size; 663 } else { 664 base = AllocateAlignedMemory(chunk_size, 665 commit_size, 666 MemoryChunk::kAlignment, 667 executable, 668 &reservation); 669 if (base == NULL) return NULL; 670 // Update executable memory size. 671 size_executable_ += reservation.size(); 672 } 673 674 if (Heap::ShouldZapGarbage()) { 675 ZapBlock(base, CodePageGuardStartOffset()); 676 ZapBlock(base + CodePageAreaStartOffset(), commit_area_size); 677 } 678 679 area_start = base + CodePageAreaStartOffset(); 680 area_end = area_start + commit_area_size; 681 } else { 682 chunk_size = RoundUp(MemoryChunk::kObjectStartOffset + reserve_area_size, 683 OS::CommitPageSize()); 684 size_t commit_size = RoundUp(MemoryChunk::kObjectStartOffset + 685 commit_area_size, OS::CommitPageSize()); 686 base = AllocateAlignedMemory(chunk_size, 687 commit_size, 688 MemoryChunk::kAlignment, 689 executable, 690 &reservation); 691 692 if (base == NULL) return NULL; 693 694 if (Heap::ShouldZapGarbage()) { 695 ZapBlock(base, Page::kObjectStartOffset + commit_area_size); 696 } 697 698 area_start = base + Page::kObjectStartOffset; 699 area_end = area_start + commit_area_size; 700 } 701 702 // Use chunk_size for statistics and callbacks because we assume that they 703 // treat reserved but not-yet committed memory regions of chunks as allocated. 704 isolate_->counters()->memory_allocated()-> 705 Increment(static_cast<int>(chunk_size)); 706 707 LOG(isolate_, NewEvent("MemoryChunk", base, chunk_size)); 708 if (owner != NULL) { 709 ObjectSpace space = static_cast<ObjectSpace>(1 << owner->identity()); 710 PerformAllocationCallback(space, kAllocationActionAllocate, chunk_size); 711 } 712 713 MemoryChunk* result = MemoryChunk::Initialize(heap, 714 base, 715 chunk_size, 716 area_start, 717 area_end, 718 executable, 719 owner); 720 result->set_reserved_memory(&reservation); 721 MSAN_MEMORY_IS_INITIALIZED(base, chunk_size); 722 return result; 723 } 724 725 726 void Page::ResetFreeListStatistics() { 727 non_available_small_blocks_ = 0; 728 available_in_small_free_list_ = 0; 729 available_in_medium_free_list_ = 0; 730 available_in_large_free_list_ = 0; 731 available_in_huge_free_list_ = 0; 732 } 733 734 735 Page* MemoryAllocator::AllocatePage(intptr_t size, 736 PagedSpace* owner, 737 Executability executable) { 738 MemoryChunk* chunk = AllocateChunk(size, size, executable, owner); 739 740 if (chunk == NULL) return NULL; 741 742 return Page::Initialize(isolate_->heap(), chunk, executable, owner); 743 } 744 745 746 LargePage* MemoryAllocator::AllocateLargePage(intptr_t object_size, 747 Space* owner, 748 Executability executable) { 749 MemoryChunk* chunk = AllocateChunk(object_size, 750 object_size, 751 executable, 752 owner); 753 if (chunk == NULL) return NULL; 754 return LargePage::Initialize(isolate_->heap(), chunk); 755 } 756 757 758 void MemoryAllocator::Free(MemoryChunk* chunk) { 759 LOG(isolate_, DeleteEvent("MemoryChunk", chunk)); 760 if (chunk->owner() != NULL) { 761 ObjectSpace space = 762 static_cast<ObjectSpace>(1 << chunk->owner()->identity()); 763 PerformAllocationCallback(space, kAllocationActionFree, chunk->size()); 764 } 765 766 isolate_->heap()->RememberUnmappedPage( 767 reinterpret_cast<Address>(chunk), chunk->IsEvacuationCandidate()); 768 769 delete chunk->slots_buffer(); 770 delete chunk->skip_list(); 771 772 VirtualMemory* reservation = chunk->reserved_memory(); 773 if (reservation->IsReserved()) { 774 FreeMemory(reservation, chunk->executable()); 775 } else { 776 FreeMemory(chunk->address(), 777 chunk->size(), 778 chunk->executable()); 779 } 780 } 781 782 783 bool MemoryAllocator::CommitBlock(Address start, 784 size_t size, 785 Executability executable) { 786 if (!CommitMemory(start, size, executable)) return false; 787 788 if (Heap::ShouldZapGarbage()) { 789 ZapBlock(start, size); 790 } 791 792 isolate_->counters()->memory_allocated()->Increment(static_cast<int>(size)); 793 return true; 794 } 795 796 797 bool MemoryAllocator::UncommitBlock(Address start, size_t size) { 798 if (!VirtualMemory::UncommitRegion(start, size)) return false; 799 isolate_->counters()->memory_allocated()->Decrement(static_cast<int>(size)); 800 return true; 801 } 802 803 804 void MemoryAllocator::ZapBlock(Address start, size_t size) { 805 for (size_t s = 0; s + kPointerSize <= size; s += kPointerSize) { 806 Memory::Address_at(start + s) = kZapValue; 807 } 808 } 809 810 811 void MemoryAllocator::PerformAllocationCallback(ObjectSpace space, 812 AllocationAction action, 813 size_t size) { 814 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 815 MemoryAllocationCallbackRegistration registration = 816 memory_allocation_callbacks_[i]; 817 if ((registration.space & space) == space && 818 (registration.action & action) == action) 819 registration.callback(space, action, static_cast<int>(size)); 820 } 821 } 822 823 824 bool MemoryAllocator::MemoryAllocationCallbackRegistered( 825 MemoryAllocationCallback callback) { 826 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 827 if (memory_allocation_callbacks_[i].callback == callback) return true; 828 } 829 return false; 830 } 831 832 833 void MemoryAllocator::AddMemoryAllocationCallback( 834 MemoryAllocationCallback callback, 835 ObjectSpace space, 836 AllocationAction action) { 837 ASSERT(callback != NULL); 838 MemoryAllocationCallbackRegistration registration(callback, space, action); 839 ASSERT(!MemoryAllocator::MemoryAllocationCallbackRegistered(callback)); 840 return memory_allocation_callbacks_.Add(registration); 841 } 842 843 844 void MemoryAllocator::RemoveMemoryAllocationCallback( 845 MemoryAllocationCallback callback) { 846 ASSERT(callback != NULL); 847 for (int i = 0; i < memory_allocation_callbacks_.length(); ++i) { 848 if (memory_allocation_callbacks_[i].callback == callback) { 849 memory_allocation_callbacks_.Remove(i); 850 return; 851 } 852 } 853 UNREACHABLE(); 854 } 855 856 857 #ifdef DEBUG 858 void MemoryAllocator::ReportStatistics() { 859 float pct = static_cast<float>(capacity_ - size_) / capacity_; 860 PrintF(" capacity: %" V8_PTR_PREFIX "d" 861 ", used: %" V8_PTR_PREFIX "d" 862 ", available: %%%d\n\n", 863 capacity_, size_, static_cast<int>(pct*100)); 864 } 865 #endif 866 867 868 int MemoryAllocator::CodePageGuardStartOffset() { 869 // We are guarding code pages: the first OS page after the header 870 // will be protected as non-writable. 871 return RoundUp(Page::kObjectStartOffset, OS::CommitPageSize()); 872 } 873 874 875 int MemoryAllocator::CodePageGuardSize() { 876 return static_cast<int>(OS::CommitPageSize()); 877 } 878 879 880 int MemoryAllocator::CodePageAreaStartOffset() { 881 // We are guarding code pages: the first OS page after the header 882 // will be protected as non-writable. 883 return CodePageGuardStartOffset() + CodePageGuardSize(); 884 } 885 886 887 int MemoryAllocator::CodePageAreaEndOffset() { 888 // We are guarding code pages: the last OS page will be protected as 889 // non-writable. 890 return Page::kPageSize - static_cast<int>(OS::CommitPageSize()); 891 } 892 893 894 bool MemoryAllocator::CommitExecutableMemory(VirtualMemory* vm, 895 Address start, 896 size_t commit_size, 897 size_t reserved_size) { 898 // Commit page header (not executable). 899 if (!vm->Commit(start, 900 CodePageGuardStartOffset(), 901 false)) { 902 return false; 903 } 904 905 // Create guard page after the header. 906 if (!vm->Guard(start + CodePageGuardStartOffset())) { 907 return false; 908 } 909 910 // Commit page body (executable). 911 if (!vm->Commit(start + CodePageAreaStartOffset(), 912 commit_size - CodePageGuardStartOffset(), 913 true)) { 914 return false; 915 } 916 917 // Create guard page before the end. 918 if (!vm->Guard(start + reserved_size - CodePageGuardSize())) { 919 return false; 920 } 921 922 UpdateAllocatedSpaceLimits(start, 923 start + CodePageAreaStartOffset() + 924 commit_size - CodePageGuardStartOffset()); 925 return true; 926 } 927 928 929 // ----------------------------------------------------------------------------- 930 // MemoryChunk implementation 931 932 void MemoryChunk::IncrementLiveBytesFromMutator(Address address, int by) { 933 MemoryChunk* chunk = MemoryChunk::FromAddress(address); 934 if (!chunk->InNewSpace() && !static_cast<Page*>(chunk)->WasSwept()) { 935 static_cast<PagedSpace*>(chunk->owner())->IncrementUnsweptFreeBytes(-by); 936 } 937 chunk->IncrementLiveBytes(by); 938 } 939 940 941 // ----------------------------------------------------------------------------- 942 // PagedSpace implementation 943 944 PagedSpace::PagedSpace(Heap* heap, 945 intptr_t max_capacity, 946 AllocationSpace id, 947 Executability executable) 948 : Space(heap, id, executable), 949 free_list_(this), 950 was_swept_conservatively_(false), 951 first_unswept_page_(Page::FromAddress(NULL)), 952 unswept_free_bytes_(0) { 953 if (id == CODE_SPACE) { 954 area_size_ = heap->isolate()->memory_allocator()-> 955 CodePageAreaSize(); 956 } else { 957 area_size_ = Page::kPageSize - Page::kObjectStartOffset; 958 } 959 max_capacity_ = (RoundDown(max_capacity, Page::kPageSize) / Page::kPageSize) 960 * AreaSize(); 961 accounting_stats_.Clear(); 962 963 allocation_info_.set_top(NULL); 964 allocation_info_.set_limit(NULL); 965 966 anchor_.InitializeAsAnchor(this); 967 } 968 969 970 bool PagedSpace::SetUp() { 971 return true; 972 } 973 974 975 bool PagedSpace::HasBeenSetUp() { 976 return true; 977 } 978 979 980 void PagedSpace::TearDown() { 981 PageIterator iterator(this); 982 while (iterator.has_next()) { 983 heap()->isolate()->memory_allocator()->Free(iterator.next()); 984 } 985 anchor_.set_next_page(&anchor_); 986 anchor_.set_prev_page(&anchor_); 987 accounting_stats_.Clear(); 988 } 989 990 991 size_t PagedSpace::CommittedPhysicalMemory() { 992 if (!VirtualMemory::HasLazyCommits()) return CommittedMemory(); 993 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 994 size_t size = 0; 995 PageIterator it(this); 996 while (it.has_next()) { 997 size += it.next()->CommittedPhysicalMemory(); 998 } 999 return size; 1000 } 1001 1002 1003 MaybeObject* PagedSpace::FindObject(Address addr) { 1004 // Note: this function can only be called on precisely swept spaces. 1005 ASSERT(!heap()->mark_compact_collector()->in_use()); 1006 1007 if (!Contains(addr)) return Failure::Exception(); 1008 1009 Page* p = Page::FromAddress(addr); 1010 HeapObjectIterator it(p, NULL); 1011 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 1012 Address cur = obj->address(); 1013 Address next = cur + obj->Size(); 1014 if ((cur <= addr) && (addr < next)) return obj; 1015 } 1016 1017 UNREACHABLE(); 1018 return Failure::Exception(); 1019 } 1020 1021 1022 bool PagedSpace::CanExpand() { 1023 ASSERT(max_capacity_ % AreaSize() == 0); 1024 1025 if (Capacity() == max_capacity_) return false; 1026 1027 ASSERT(Capacity() < max_capacity_); 1028 1029 // Are we going to exceed capacity for this space? 1030 if ((Capacity() + Page::kPageSize) > max_capacity_) return false; 1031 1032 return true; 1033 } 1034 1035 1036 bool PagedSpace::Expand() { 1037 if (!CanExpand()) return false; 1038 1039 intptr_t size = AreaSize(); 1040 1041 if (anchor_.next_page() == &anchor_) { 1042 size = SizeOfFirstPage(); 1043 } 1044 1045 Page* p = heap()->isolate()->memory_allocator()->AllocatePage( 1046 size, this, executable()); 1047 if (p == NULL) return false; 1048 1049 ASSERT(Capacity() <= max_capacity_); 1050 1051 p->InsertAfter(anchor_.prev_page()); 1052 1053 return true; 1054 } 1055 1056 1057 intptr_t PagedSpace::SizeOfFirstPage() { 1058 int size = 0; 1059 switch (identity()) { 1060 case OLD_POINTER_SPACE: 1061 size = 72 * kPointerSize * KB; 1062 break; 1063 case OLD_DATA_SPACE: 1064 size = 192 * KB; 1065 break; 1066 case MAP_SPACE: 1067 size = 16 * kPointerSize * KB; 1068 break; 1069 case CELL_SPACE: 1070 size = 16 * kPointerSize * KB; 1071 break; 1072 case PROPERTY_CELL_SPACE: 1073 size = 8 * kPointerSize * KB; 1074 break; 1075 case CODE_SPACE: 1076 if (heap()->isolate()->code_range()->exists()) { 1077 // When code range exists, code pages are allocated in a special way 1078 // (from the reserved code range). That part of the code is not yet 1079 // upgraded to handle small pages. 1080 size = AreaSize(); 1081 } else { 1082 #if V8_TARGET_ARCH_MIPS 1083 // TODO(plind): Investigate larger code stubs size on MIPS. 1084 size = 480 * KB; 1085 #else 1086 size = 416 * KB; 1087 #endif 1088 } 1089 break; 1090 default: 1091 UNREACHABLE(); 1092 } 1093 return Min(size, AreaSize()); 1094 } 1095 1096 1097 int PagedSpace::CountTotalPages() { 1098 PageIterator it(this); 1099 int count = 0; 1100 while (it.has_next()) { 1101 it.next(); 1102 count++; 1103 } 1104 return count; 1105 } 1106 1107 1108 void PagedSpace::ObtainFreeListStatistics(Page* page, SizeStats* sizes) { 1109 sizes->huge_size_ = page->available_in_huge_free_list(); 1110 sizes->small_size_ = page->available_in_small_free_list(); 1111 sizes->medium_size_ = page->available_in_medium_free_list(); 1112 sizes->large_size_ = page->available_in_large_free_list(); 1113 } 1114 1115 1116 void PagedSpace::ResetFreeListStatistics() { 1117 PageIterator page_iterator(this); 1118 while (page_iterator.has_next()) { 1119 Page* page = page_iterator.next(); 1120 page->ResetFreeListStatistics(); 1121 } 1122 } 1123 1124 1125 void PagedSpace::IncreaseCapacity(int size) { 1126 accounting_stats_.ExpandSpace(size); 1127 } 1128 1129 1130 void PagedSpace::ReleasePage(Page* page, bool unlink) { 1131 ASSERT(page->LiveBytes() == 0); 1132 ASSERT(AreaSize() == page->area_size()); 1133 1134 // Adjust list of unswept pages if the page is the head of the list. 1135 if (first_unswept_page_ == page) { 1136 first_unswept_page_ = page->next_page(); 1137 if (first_unswept_page_ == anchor()) { 1138 first_unswept_page_ = Page::FromAddress(NULL); 1139 } 1140 } 1141 1142 if (page->WasSwept()) { 1143 intptr_t size = free_list_.EvictFreeListItems(page); 1144 accounting_stats_.AllocateBytes(size); 1145 ASSERT_EQ(AreaSize(), static_cast<int>(size)); 1146 } else { 1147 DecreaseUnsweptFreeBytes(page); 1148 } 1149 1150 if (Page::FromAllocationTop(allocation_info_.top()) == page) { 1151 allocation_info_.set_top(NULL); 1152 allocation_info_.set_limit(NULL); 1153 } 1154 1155 if (unlink) { 1156 page->Unlink(); 1157 } 1158 if (page->IsFlagSet(MemoryChunk::CONTAINS_ONLY_DATA)) { 1159 heap()->isolate()->memory_allocator()->Free(page); 1160 } else { 1161 heap()->QueueMemoryChunkForFree(page); 1162 } 1163 1164 ASSERT(Capacity() > 0); 1165 accounting_stats_.ShrinkSpace(AreaSize()); 1166 } 1167 1168 1169 #ifdef DEBUG 1170 void PagedSpace::Print() { } 1171 #endif 1172 1173 #ifdef VERIFY_HEAP 1174 void PagedSpace::Verify(ObjectVisitor* visitor) { 1175 // We can only iterate over the pages if they were swept precisely. 1176 if (was_swept_conservatively_) return; 1177 1178 bool allocation_pointer_found_in_space = 1179 (allocation_info_.top() == allocation_info_.limit()); 1180 PageIterator page_iterator(this); 1181 while (page_iterator.has_next()) { 1182 Page* page = page_iterator.next(); 1183 CHECK(page->owner() == this); 1184 if (page == Page::FromAllocationTop(allocation_info_.top())) { 1185 allocation_pointer_found_in_space = true; 1186 } 1187 CHECK(page->WasSweptPrecisely()); 1188 HeapObjectIterator it(page, NULL); 1189 Address end_of_previous_object = page->area_start(); 1190 Address top = page->area_end(); 1191 int black_size = 0; 1192 for (HeapObject* object = it.Next(); object != NULL; object = it.Next()) { 1193 CHECK(end_of_previous_object <= object->address()); 1194 1195 // The first word should be a map, and we expect all map pointers to 1196 // be in map space. 1197 Map* map = object->map(); 1198 CHECK(map->IsMap()); 1199 CHECK(heap()->map_space()->Contains(map)); 1200 1201 // Perform space-specific object verification. 1202 VerifyObject(object); 1203 1204 // The object itself should look OK. 1205 object->Verify(); 1206 1207 // All the interior pointers should be contained in the heap. 1208 int size = object->Size(); 1209 object->IterateBody(map->instance_type(), size, visitor); 1210 if (Marking::IsBlack(Marking::MarkBitFrom(object))) { 1211 black_size += size; 1212 } 1213 1214 CHECK(object->address() + size <= top); 1215 end_of_previous_object = object->address() + size; 1216 } 1217 CHECK_LE(black_size, page->LiveBytes()); 1218 } 1219 CHECK(allocation_pointer_found_in_space); 1220 } 1221 #endif // VERIFY_HEAP 1222 1223 // ----------------------------------------------------------------------------- 1224 // NewSpace implementation 1225 1226 1227 bool NewSpace::SetUp(int reserved_semispace_capacity, 1228 int maximum_semispace_capacity) { 1229 // Set up new space based on the preallocated memory block defined by 1230 // start and size. The provided space is divided into two semi-spaces. 1231 // To support fast containment testing in the new space, the size of 1232 // this chunk must be a power of two and it must be aligned to its size. 1233 int initial_semispace_capacity = heap()->InitialSemiSpaceSize(); 1234 1235 size_t size = 2 * reserved_semispace_capacity; 1236 Address base = 1237 heap()->isolate()->memory_allocator()->ReserveAlignedMemory( 1238 size, size, &reservation_); 1239 if (base == NULL) return false; 1240 1241 chunk_base_ = base; 1242 chunk_size_ = static_cast<uintptr_t>(size); 1243 LOG(heap()->isolate(), NewEvent("InitialChunk", chunk_base_, chunk_size_)); 1244 1245 ASSERT(initial_semispace_capacity <= maximum_semispace_capacity); 1246 ASSERT(IsPowerOf2(maximum_semispace_capacity)); 1247 1248 // Allocate and set up the histogram arrays if necessary. 1249 allocated_histogram_ = NewArray<HistogramInfo>(LAST_TYPE + 1); 1250 promoted_histogram_ = NewArray<HistogramInfo>(LAST_TYPE + 1); 1251 1252 #define SET_NAME(name) allocated_histogram_[name].set_name(#name); \ 1253 promoted_histogram_[name].set_name(#name); 1254 INSTANCE_TYPE_LIST(SET_NAME) 1255 #undef SET_NAME 1256 1257 ASSERT(reserved_semispace_capacity == heap()->ReservedSemiSpaceSize()); 1258 ASSERT(static_cast<intptr_t>(chunk_size_) >= 1259 2 * heap()->ReservedSemiSpaceSize()); 1260 ASSERT(IsAddressAligned(chunk_base_, 2 * reserved_semispace_capacity, 0)); 1261 1262 to_space_.SetUp(chunk_base_, 1263 initial_semispace_capacity, 1264 maximum_semispace_capacity); 1265 from_space_.SetUp(chunk_base_ + reserved_semispace_capacity, 1266 initial_semispace_capacity, 1267 maximum_semispace_capacity); 1268 if (!to_space_.Commit()) { 1269 return false; 1270 } 1271 ASSERT(!from_space_.is_committed()); // No need to use memory yet. 1272 1273 start_ = chunk_base_; 1274 address_mask_ = ~(2 * reserved_semispace_capacity - 1); 1275 object_mask_ = address_mask_ | kHeapObjectTagMask; 1276 object_expected_ = reinterpret_cast<uintptr_t>(start_) | kHeapObjectTag; 1277 1278 ResetAllocationInfo(); 1279 1280 return true; 1281 } 1282 1283 1284 void NewSpace::TearDown() { 1285 if (allocated_histogram_) { 1286 DeleteArray(allocated_histogram_); 1287 allocated_histogram_ = NULL; 1288 } 1289 if (promoted_histogram_) { 1290 DeleteArray(promoted_histogram_); 1291 promoted_histogram_ = NULL; 1292 } 1293 1294 start_ = NULL; 1295 allocation_info_.set_top(NULL); 1296 allocation_info_.set_limit(NULL); 1297 1298 to_space_.TearDown(); 1299 from_space_.TearDown(); 1300 1301 LOG(heap()->isolate(), DeleteEvent("InitialChunk", chunk_base_)); 1302 1303 ASSERT(reservation_.IsReserved()); 1304 heap()->isolate()->memory_allocator()->FreeMemory(&reservation_, 1305 NOT_EXECUTABLE); 1306 chunk_base_ = NULL; 1307 chunk_size_ = 0; 1308 } 1309 1310 1311 void NewSpace::Flip() { 1312 SemiSpace::Swap(&from_space_, &to_space_); 1313 } 1314 1315 1316 void NewSpace::Grow() { 1317 // Double the semispace size but only up to maximum capacity. 1318 ASSERT(Capacity() < MaximumCapacity()); 1319 int new_capacity = Min(MaximumCapacity(), 2 * static_cast<int>(Capacity())); 1320 if (to_space_.GrowTo(new_capacity)) { 1321 // Only grow from space if we managed to grow to-space. 1322 if (!from_space_.GrowTo(new_capacity)) { 1323 // If we managed to grow to-space but couldn't grow from-space, 1324 // attempt to shrink to-space. 1325 if (!to_space_.ShrinkTo(from_space_.Capacity())) { 1326 // We are in an inconsistent state because we could not 1327 // commit/uncommit memory from new space. 1328 V8::FatalProcessOutOfMemory("Failed to grow new space."); 1329 } 1330 } 1331 } 1332 ASSERT_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1333 } 1334 1335 1336 void NewSpace::Shrink() { 1337 int new_capacity = Max(InitialCapacity(), 2 * SizeAsInt()); 1338 int rounded_new_capacity = RoundUp(new_capacity, Page::kPageSize); 1339 if (rounded_new_capacity < Capacity() && 1340 to_space_.ShrinkTo(rounded_new_capacity)) { 1341 // Only shrink from-space if we managed to shrink to-space. 1342 from_space_.Reset(); 1343 if (!from_space_.ShrinkTo(rounded_new_capacity)) { 1344 // If we managed to shrink to-space but couldn't shrink from 1345 // space, attempt to grow to-space again. 1346 if (!to_space_.GrowTo(from_space_.Capacity())) { 1347 // We are in an inconsistent state because we could not 1348 // commit/uncommit memory from new space. 1349 V8::FatalProcessOutOfMemory("Failed to shrink new space."); 1350 } 1351 } 1352 } 1353 ASSERT_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1354 } 1355 1356 1357 void NewSpace::UpdateAllocationInfo() { 1358 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 1359 allocation_info_.set_top(to_space_.page_low()); 1360 allocation_info_.set_limit(to_space_.page_high()); 1361 UpdateInlineAllocationLimit(0); 1362 ASSERT_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1363 } 1364 1365 1366 void NewSpace::ResetAllocationInfo() { 1367 to_space_.Reset(); 1368 UpdateAllocationInfo(); 1369 pages_used_ = 0; 1370 // Clear all mark-bits in the to-space. 1371 NewSpacePageIterator it(&to_space_); 1372 while (it.has_next()) { 1373 Bitmap::Clear(it.next()); 1374 } 1375 } 1376 1377 1378 void NewSpace::UpdateInlineAllocationLimit(int size_in_bytes) { 1379 if (heap()->inline_allocation_disabled()) { 1380 // Lowest limit when linear allocation was disabled. 1381 Address high = to_space_.page_high(); 1382 Address new_top = allocation_info_.top() + size_in_bytes; 1383 allocation_info_.set_limit(Min(new_top, high)); 1384 } else if (inline_allocation_limit_step() == 0) { 1385 // Normal limit is the end of the current page. 1386 allocation_info_.set_limit(to_space_.page_high()); 1387 } else { 1388 // Lower limit during incremental marking. 1389 Address high = to_space_.page_high(); 1390 Address new_top = allocation_info_.top() + size_in_bytes; 1391 Address new_limit = new_top + inline_allocation_limit_step_; 1392 allocation_info_.set_limit(Min(new_limit, high)); 1393 } 1394 ASSERT_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1395 } 1396 1397 1398 bool NewSpace::AddFreshPage() { 1399 Address top = allocation_info_.top(); 1400 if (NewSpacePage::IsAtStart(top)) { 1401 // The current page is already empty. Don't try to make another. 1402 1403 // We should only get here if someone asks to allocate more 1404 // than what can be stored in a single page. 1405 // TODO(gc): Change the limit on new-space allocation to prevent this 1406 // from happening (all such allocations should go directly to LOSpace). 1407 return false; 1408 } 1409 if (!to_space_.AdvancePage()) { 1410 // Failed to get a new page in to-space. 1411 return false; 1412 } 1413 1414 // Clear remainder of current page. 1415 Address limit = NewSpacePage::FromLimit(top)->area_end(); 1416 if (heap()->gc_state() == Heap::SCAVENGE) { 1417 heap()->promotion_queue()->SetNewLimit(limit); 1418 heap()->promotion_queue()->ActivateGuardIfOnTheSamePage(); 1419 } 1420 1421 int remaining_in_page = static_cast<int>(limit - top); 1422 heap()->CreateFillerObjectAt(top, remaining_in_page); 1423 pages_used_++; 1424 UpdateAllocationInfo(); 1425 1426 return true; 1427 } 1428 1429 1430 MaybeObject* NewSpace::SlowAllocateRaw(int size_in_bytes) { 1431 Address old_top = allocation_info_.top(); 1432 Address high = to_space_.page_high(); 1433 if (allocation_info_.limit() < high) { 1434 // Either the limit has been lowered because linear allocation was disabled 1435 // or because incremental marking wants to get a chance to do a step. Set 1436 // the new limit accordingly. 1437 Address new_top = old_top + size_in_bytes; 1438 int bytes_allocated = static_cast<int>(new_top - top_on_previous_step_); 1439 heap()->incremental_marking()->Step( 1440 bytes_allocated, IncrementalMarking::GC_VIA_STACK_GUARD); 1441 UpdateInlineAllocationLimit(size_in_bytes); 1442 top_on_previous_step_ = new_top; 1443 return AllocateRaw(size_in_bytes); 1444 } else if (AddFreshPage()) { 1445 // Switched to new page. Try allocating again. 1446 int bytes_allocated = static_cast<int>(old_top - top_on_previous_step_); 1447 heap()->incremental_marking()->Step( 1448 bytes_allocated, IncrementalMarking::GC_VIA_STACK_GUARD); 1449 top_on_previous_step_ = to_space_.page_low(); 1450 return AllocateRaw(size_in_bytes); 1451 } else { 1452 return Failure::RetryAfterGC(); 1453 } 1454 } 1455 1456 1457 #ifdef VERIFY_HEAP 1458 // We do not use the SemiSpaceIterator because verification doesn't assume 1459 // that it works (it depends on the invariants we are checking). 1460 void NewSpace::Verify() { 1461 // The allocation pointer should be in the space or at the very end. 1462 ASSERT_SEMISPACE_ALLOCATION_INFO(allocation_info_, to_space_); 1463 1464 // There should be objects packed in from the low address up to the 1465 // allocation pointer. 1466 Address current = to_space_.first_page()->area_start(); 1467 CHECK_EQ(current, to_space_.space_start()); 1468 1469 while (current != top()) { 1470 if (!NewSpacePage::IsAtEnd(current)) { 1471 // The allocation pointer should not be in the middle of an object. 1472 CHECK(!NewSpacePage::FromLimit(current)->ContainsLimit(top()) || 1473 current < top()); 1474 1475 HeapObject* object = HeapObject::FromAddress(current); 1476 1477 // The first word should be a map, and we expect all map pointers to 1478 // be in map space. 1479 Map* map = object->map(); 1480 CHECK(map->IsMap()); 1481 CHECK(heap()->map_space()->Contains(map)); 1482 1483 // The object should not be code or a map. 1484 CHECK(!object->IsMap()); 1485 CHECK(!object->IsCode()); 1486 1487 // The object itself should look OK. 1488 object->Verify(); 1489 1490 // All the interior pointers should be contained in the heap. 1491 VerifyPointersVisitor visitor; 1492 int size = object->Size(); 1493 object->IterateBody(map->instance_type(), size, &visitor); 1494 1495 current += size; 1496 } else { 1497 // At end of page, switch to next page. 1498 NewSpacePage* page = NewSpacePage::FromLimit(current)->next_page(); 1499 // Next page should be valid. 1500 CHECK(!page->is_anchor()); 1501 current = page->area_start(); 1502 } 1503 } 1504 1505 // Check semi-spaces. 1506 CHECK_EQ(from_space_.id(), kFromSpace); 1507 CHECK_EQ(to_space_.id(), kToSpace); 1508 from_space_.Verify(); 1509 to_space_.Verify(); 1510 } 1511 #endif 1512 1513 // ----------------------------------------------------------------------------- 1514 // SemiSpace implementation 1515 1516 void SemiSpace::SetUp(Address start, 1517 int initial_capacity, 1518 int maximum_capacity) { 1519 // Creates a space in the young generation. The constructor does not 1520 // allocate memory from the OS. A SemiSpace is given a contiguous chunk of 1521 // memory of size 'capacity' when set up, and does not grow or shrink 1522 // otherwise. In the mark-compact collector, the memory region of the from 1523 // space is used as the marking stack. It requires contiguous memory 1524 // addresses. 1525 ASSERT(maximum_capacity >= Page::kPageSize); 1526 initial_capacity_ = RoundDown(initial_capacity, Page::kPageSize); 1527 capacity_ = initial_capacity; 1528 maximum_capacity_ = RoundDown(maximum_capacity, Page::kPageSize); 1529 maximum_committed_ = 0; 1530 committed_ = false; 1531 start_ = start; 1532 address_mask_ = ~(maximum_capacity - 1); 1533 object_mask_ = address_mask_ | kHeapObjectTagMask; 1534 object_expected_ = reinterpret_cast<uintptr_t>(start) | kHeapObjectTag; 1535 age_mark_ = start_; 1536 } 1537 1538 1539 void SemiSpace::TearDown() { 1540 start_ = NULL; 1541 capacity_ = 0; 1542 } 1543 1544 1545 bool SemiSpace::Commit() { 1546 ASSERT(!is_committed()); 1547 int pages = capacity_ / Page::kPageSize; 1548 if (!heap()->isolate()->memory_allocator()->CommitBlock(start_, 1549 capacity_, 1550 executable())) { 1551 return false; 1552 } 1553 1554 NewSpacePage* current = anchor(); 1555 for (int i = 0; i < pages; i++) { 1556 NewSpacePage* new_page = 1557 NewSpacePage::Initialize(heap(), start_ + i * Page::kPageSize, this); 1558 new_page->InsertAfter(current); 1559 current = new_page; 1560 } 1561 1562 SetCapacity(capacity_); 1563 committed_ = true; 1564 Reset(); 1565 return true; 1566 } 1567 1568 1569 bool SemiSpace::Uncommit() { 1570 ASSERT(is_committed()); 1571 Address start = start_ + maximum_capacity_ - capacity_; 1572 if (!heap()->isolate()->memory_allocator()->UncommitBlock(start, capacity_)) { 1573 return false; 1574 } 1575 anchor()->set_next_page(anchor()); 1576 anchor()->set_prev_page(anchor()); 1577 1578 committed_ = false; 1579 return true; 1580 } 1581 1582 1583 size_t SemiSpace::CommittedPhysicalMemory() { 1584 if (!is_committed()) return 0; 1585 size_t size = 0; 1586 NewSpacePageIterator it(this); 1587 while (it.has_next()) { 1588 size += it.next()->CommittedPhysicalMemory(); 1589 } 1590 return size; 1591 } 1592 1593 1594 bool SemiSpace::GrowTo(int new_capacity) { 1595 if (!is_committed()) { 1596 if (!Commit()) return false; 1597 } 1598 ASSERT((new_capacity & Page::kPageAlignmentMask) == 0); 1599 ASSERT(new_capacity <= maximum_capacity_); 1600 ASSERT(new_capacity > capacity_); 1601 int pages_before = capacity_ / Page::kPageSize; 1602 int pages_after = new_capacity / Page::kPageSize; 1603 1604 size_t delta = new_capacity - capacity_; 1605 1606 ASSERT(IsAligned(delta, OS::AllocateAlignment())); 1607 if (!heap()->isolate()->memory_allocator()->CommitBlock( 1608 start_ + capacity_, delta, executable())) { 1609 return false; 1610 } 1611 SetCapacity(new_capacity); 1612 NewSpacePage* last_page = anchor()->prev_page(); 1613 ASSERT(last_page != anchor()); 1614 for (int i = pages_before; i < pages_after; i++) { 1615 Address page_address = start_ + i * Page::kPageSize; 1616 NewSpacePage* new_page = NewSpacePage::Initialize(heap(), 1617 page_address, 1618 this); 1619 new_page->InsertAfter(last_page); 1620 Bitmap::Clear(new_page); 1621 // Duplicate the flags that was set on the old page. 1622 new_page->SetFlags(last_page->GetFlags(), 1623 NewSpacePage::kCopyOnFlipFlagsMask); 1624 last_page = new_page; 1625 } 1626 return true; 1627 } 1628 1629 1630 bool SemiSpace::ShrinkTo(int new_capacity) { 1631 ASSERT((new_capacity & Page::kPageAlignmentMask) == 0); 1632 ASSERT(new_capacity >= initial_capacity_); 1633 ASSERT(new_capacity < capacity_); 1634 if (is_committed()) { 1635 size_t delta = capacity_ - new_capacity; 1636 ASSERT(IsAligned(delta, OS::AllocateAlignment())); 1637 1638 MemoryAllocator* allocator = heap()->isolate()->memory_allocator(); 1639 if (!allocator->UncommitBlock(start_ + new_capacity, delta)) { 1640 return false; 1641 } 1642 1643 int pages_after = new_capacity / Page::kPageSize; 1644 NewSpacePage* new_last_page = 1645 NewSpacePage::FromAddress(start_ + (pages_after - 1) * Page::kPageSize); 1646 new_last_page->set_next_page(anchor()); 1647 anchor()->set_prev_page(new_last_page); 1648 ASSERT((current_page_ >= first_page()) && (current_page_ <= new_last_page)); 1649 } 1650 1651 SetCapacity(new_capacity); 1652 1653 return true; 1654 } 1655 1656 1657 void SemiSpace::FlipPages(intptr_t flags, intptr_t mask) { 1658 anchor_.set_owner(this); 1659 // Fixup back-pointers to anchor. Address of anchor changes 1660 // when we swap. 1661 anchor_.prev_page()->set_next_page(&anchor_); 1662 anchor_.next_page()->set_prev_page(&anchor_); 1663 1664 bool becomes_to_space = (id_ == kFromSpace); 1665 id_ = becomes_to_space ? kToSpace : kFromSpace; 1666 NewSpacePage* page = anchor_.next_page(); 1667 while (page != &anchor_) { 1668 page->set_owner(this); 1669 page->SetFlags(flags, mask); 1670 if (becomes_to_space) { 1671 page->ClearFlag(MemoryChunk::IN_FROM_SPACE); 1672 page->SetFlag(MemoryChunk::IN_TO_SPACE); 1673 page->ClearFlag(MemoryChunk::NEW_SPACE_BELOW_AGE_MARK); 1674 page->ResetLiveBytes(); 1675 } else { 1676 page->SetFlag(MemoryChunk::IN_FROM_SPACE); 1677 page->ClearFlag(MemoryChunk::IN_TO_SPACE); 1678 } 1679 ASSERT(page->IsFlagSet(MemoryChunk::SCAN_ON_SCAVENGE)); 1680 ASSERT(page->IsFlagSet(MemoryChunk::IN_TO_SPACE) || 1681 page->IsFlagSet(MemoryChunk::IN_FROM_SPACE)); 1682 page = page->next_page(); 1683 } 1684 } 1685 1686 1687 void SemiSpace::Reset() { 1688 ASSERT(anchor_.next_page() != &anchor_); 1689 current_page_ = anchor_.next_page(); 1690 } 1691 1692 1693 void SemiSpace::Swap(SemiSpace* from, SemiSpace* to) { 1694 // We won't be swapping semispaces without data in them. 1695 ASSERT(from->anchor_.next_page() != &from->anchor_); 1696 ASSERT(to->anchor_.next_page() != &to->anchor_); 1697 1698 // Swap bits. 1699 SemiSpace tmp = *from; 1700 *from = *to; 1701 *to = tmp; 1702 1703 // Fixup back-pointers to the page list anchor now that its address 1704 // has changed. 1705 // Swap to/from-space bits on pages. 1706 // Copy GC flags from old active space (from-space) to new (to-space). 1707 intptr_t flags = from->current_page()->GetFlags(); 1708 to->FlipPages(flags, NewSpacePage::kCopyOnFlipFlagsMask); 1709 1710 from->FlipPages(0, 0); 1711 } 1712 1713 1714 void SemiSpace::SetCapacity(int new_capacity) { 1715 capacity_ = new_capacity; 1716 if (capacity_ > maximum_committed_) { 1717 maximum_committed_ = capacity_; 1718 } 1719 } 1720 1721 1722 void SemiSpace::set_age_mark(Address mark) { 1723 ASSERT(NewSpacePage::FromLimit(mark)->semi_space() == this); 1724 age_mark_ = mark; 1725 // Mark all pages up to the one containing mark. 1726 NewSpacePageIterator it(space_start(), mark); 1727 while (it.has_next()) { 1728 it.next()->SetFlag(MemoryChunk::NEW_SPACE_BELOW_AGE_MARK); 1729 } 1730 } 1731 1732 1733 #ifdef DEBUG 1734 void SemiSpace::Print() { } 1735 #endif 1736 1737 #ifdef VERIFY_HEAP 1738 void SemiSpace::Verify() { 1739 bool is_from_space = (id_ == kFromSpace); 1740 NewSpacePage* page = anchor_.next_page(); 1741 CHECK(anchor_.semi_space() == this); 1742 while (page != &anchor_) { 1743 CHECK(page->semi_space() == this); 1744 CHECK(page->InNewSpace()); 1745 CHECK(page->IsFlagSet(is_from_space ? MemoryChunk::IN_FROM_SPACE 1746 : MemoryChunk::IN_TO_SPACE)); 1747 CHECK(!page->IsFlagSet(is_from_space ? MemoryChunk::IN_TO_SPACE 1748 : MemoryChunk::IN_FROM_SPACE)); 1749 CHECK(page->IsFlagSet(MemoryChunk::POINTERS_TO_HERE_ARE_INTERESTING)); 1750 if (!is_from_space) { 1751 // The pointers-from-here-are-interesting flag isn't updated dynamically 1752 // on from-space pages, so it might be out of sync with the marking state. 1753 if (page->heap()->incremental_marking()->IsMarking()) { 1754 CHECK(page->IsFlagSet(MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING)); 1755 } else { 1756 CHECK(!page->IsFlagSet( 1757 MemoryChunk::POINTERS_FROM_HERE_ARE_INTERESTING)); 1758 } 1759 // TODO(gc): Check that the live_bytes_count_ field matches the 1760 // black marking on the page (if we make it match in new-space). 1761 } 1762 CHECK(page->IsFlagSet(MemoryChunk::SCAN_ON_SCAVENGE)); 1763 CHECK(page->prev_page()->next_page() == page); 1764 page = page->next_page(); 1765 } 1766 } 1767 #endif 1768 1769 #ifdef DEBUG 1770 void SemiSpace::AssertValidRange(Address start, Address end) { 1771 // Addresses belong to same semi-space 1772 NewSpacePage* page = NewSpacePage::FromLimit(start); 1773 NewSpacePage* end_page = NewSpacePage::FromLimit(end); 1774 SemiSpace* space = page->semi_space(); 1775 CHECK_EQ(space, end_page->semi_space()); 1776 // Start address is before end address, either on same page, 1777 // or end address is on a later page in the linked list of 1778 // semi-space pages. 1779 if (page == end_page) { 1780 CHECK(start <= end); 1781 } else { 1782 while (page != end_page) { 1783 page = page->next_page(); 1784 CHECK_NE(page, space->anchor()); 1785 } 1786 } 1787 } 1788 #endif 1789 1790 1791 // ----------------------------------------------------------------------------- 1792 // SemiSpaceIterator implementation. 1793 SemiSpaceIterator::SemiSpaceIterator(NewSpace* space) { 1794 Initialize(space->bottom(), space->top(), NULL); 1795 } 1796 1797 1798 SemiSpaceIterator::SemiSpaceIterator(NewSpace* space, 1799 HeapObjectCallback size_func) { 1800 Initialize(space->bottom(), space->top(), size_func); 1801 } 1802 1803 1804 SemiSpaceIterator::SemiSpaceIterator(NewSpace* space, Address start) { 1805 Initialize(start, space->top(), NULL); 1806 } 1807 1808 1809 SemiSpaceIterator::SemiSpaceIterator(Address from, Address to) { 1810 Initialize(from, to, NULL); 1811 } 1812 1813 1814 void SemiSpaceIterator::Initialize(Address start, 1815 Address end, 1816 HeapObjectCallback size_func) { 1817 SemiSpace::AssertValidRange(start, end); 1818 current_ = start; 1819 limit_ = end; 1820 size_func_ = size_func; 1821 } 1822 1823 1824 #ifdef DEBUG 1825 // heap_histograms is shared, always clear it before using it. 1826 static void ClearHistograms(Isolate* isolate) { 1827 // We reset the name each time, though it hasn't changed. 1828 #define DEF_TYPE_NAME(name) isolate->heap_histograms()[name].set_name(#name); 1829 INSTANCE_TYPE_LIST(DEF_TYPE_NAME) 1830 #undef DEF_TYPE_NAME 1831 1832 #define CLEAR_HISTOGRAM(name) isolate->heap_histograms()[name].clear(); 1833 INSTANCE_TYPE_LIST(CLEAR_HISTOGRAM) 1834 #undef CLEAR_HISTOGRAM 1835 1836 isolate->js_spill_information()->Clear(); 1837 } 1838 1839 1840 static void ClearCodeKindStatistics(int* code_kind_statistics) { 1841 for (int i = 0; i < Code::NUMBER_OF_KINDS; i++) { 1842 code_kind_statistics[i] = 0; 1843 } 1844 } 1845 1846 1847 static void ReportCodeKindStatistics(int* code_kind_statistics) { 1848 PrintF("\n Code kind histograms: \n"); 1849 for (int i = 0; i < Code::NUMBER_OF_KINDS; i++) { 1850 if (code_kind_statistics[i] > 0) { 1851 PrintF(" %-20s: %10d bytes\n", 1852 Code::Kind2String(static_cast<Code::Kind>(i)), 1853 code_kind_statistics[i]); 1854 } 1855 } 1856 PrintF("\n"); 1857 } 1858 1859 1860 static int CollectHistogramInfo(HeapObject* obj) { 1861 Isolate* isolate = obj->GetIsolate(); 1862 InstanceType type = obj->map()->instance_type(); 1863 ASSERT(0 <= type && type <= LAST_TYPE); 1864 ASSERT(isolate->heap_histograms()[type].name() != NULL); 1865 isolate->heap_histograms()[type].increment_number(1); 1866 isolate->heap_histograms()[type].increment_bytes(obj->Size()); 1867 1868 if (FLAG_collect_heap_spill_statistics && obj->IsJSObject()) { 1869 JSObject::cast(obj)->IncrementSpillStatistics( 1870 isolate->js_spill_information()); 1871 } 1872 1873 return obj->Size(); 1874 } 1875 1876 1877 static void ReportHistogram(Isolate* isolate, bool print_spill) { 1878 PrintF("\n Object Histogram:\n"); 1879 for (int i = 0; i <= LAST_TYPE; i++) { 1880 if (isolate->heap_histograms()[i].number() > 0) { 1881 PrintF(" %-34s%10d (%10d bytes)\n", 1882 isolate->heap_histograms()[i].name(), 1883 isolate->heap_histograms()[i].number(), 1884 isolate->heap_histograms()[i].bytes()); 1885 } 1886 } 1887 PrintF("\n"); 1888 1889 // Summarize string types. 1890 int string_number = 0; 1891 int string_bytes = 0; 1892 #define INCREMENT(type, size, name, camel_name) \ 1893 string_number += isolate->heap_histograms()[type].number(); \ 1894 string_bytes += isolate->heap_histograms()[type].bytes(); 1895 STRING_TYPE_LIST(INCREMENT) 1896 #undef INCREMENT 1897 if (string_number > 0) { 1898 PrintF(" %-34s%10d (%10d bytes)\n\n", "STRING_TYPE", string_number, 1899 string_bytes); 1900 } 1901 1902 if (FLAG_collect_heap_spill_statistics && print_spill) { 1903 isolate->js_spill_information()->Print(); 1904 } 1905 } 1906 #endif // DEBUG 1907 1908 1909 // Support for statistics gathering for --heap-stats and --log-gc. 1910 void NewSpace::ClearHistograms() { 1911 for (int i = 0; i <= LAST_TYPE; i++) { 1912 allocated_histogram_[i].clear(); 1913 promoted_histogram_[i].clear(); 1914 } 1915 } 1916 1917 1918 // Because the copying collector does not touch garbage objects, we iterate 1919 // the new space before a collection to get a histogram of allocated objects. 1920 // This only happens when --log-gc flag is set. 1921 void NewSpace::CollectStatistics() { 1922 ClearHistograms(); 1923 SemiSpaceIterator it(this); 1924 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) 1925 RecordAllocation(obj); 1926 } 1927 1928 1929 static void DoReportStatistics(Isolate* isolate, 1930 HistogramInfo* info, const char* description) { 1931 LOG(isolate, HeapSampleBeginEvent("NewSpace", description)); 1932 // Lump all the string types together. 1933 int string_number = 0; 1934 int string_bytes = 0; 1935 #define INCREMENT(type, size, name, camel_name) \ 1936 string_number += info[type].number(); \ 1937 string_bytes += info[type].bytes(); 1938 STRING_TYPE_LIST(INCREMENT) 1939 #undef INCREMENT 1940 if (string_number > 0) { 1941 LOG(isolate, 1942 HeapSampleItemEvent("STRING_TYPE", string_number, string_bytes)); 1943 } 1944 1945 // Then do the other types. 1946 for (int i = FIRST_NONSTRING_TYPE; i <= LAST_TYPE; ++i) { 1947 if (info[i].number() > 0) { 1948 LOG(isolate, 1949 HeapSampleItemEvent(info[i].name(), info[i].number(), 1950 info[i].bytes())); 1951 } 1952 } 1953 LOG(isolate, HeapSampleEndEvent("NewSpace", description)); 1954 } 1955 1956 1957 void NewSpace::ReportStatistics() { 1958 #ifdef DEBUG 1959 if (FLAG_heap_stats) { 1960 float pct = static_cast<float>(Available()) / Capacity(); 1961 PrintF(" capacity: %" V8_PTR_PREFIX "d" 1962 ", available: %" V8_PTR_PREFIX "d, %%%d\n", 1963 Capacity(), Available(), static_cast<int>(pct*100)); 1964 PrintF("\n Object Histogram:\n"); 1965 for (int i = 0; i <= LAST_TYPE; i++) { 1966 if (allocated_histogram_[i].number() > 0) { 1967 PrintF(" %-34s%10d (%10d bytes)\n", 1968 allocated_histogram_[i].name(), 1969 allocated_histogram_[i].number(), 1970 allocated_histogram_[i].bytes()); 1971 } 1972 } 1973 PrintF("\n"); 1974 } 1975 #endif // DEBUG 1976 1977 if (FLAG_log_gc) { 1978 Isolate* isolate = heap()->isolate(); 1979 DoReportStatistics(isolate, allocated_histogram_, "allocated"); 1980 DoReportStatistics(isolate, promoted_histogram_, "promoted"); 1981 } 1982 } 1983 1984 1985 void NewSpace::RecordAllocation(HeapObject* obj) { 1986 InstanceType type = obj->map()->instance_type(); 1987 ASSERT(0 <= type && type <= LAST_TYPE); 1988 allocated_histogram_[type].increment_number(1); 1989 allocated_histogram_[type].increment_bytes(obj->Size()); 1990 } 1991 1992 1993 void NewSpace::RecordPromotion(HeapObject* obj) { 1994 InstanceType type = obj->map()->instance_type(); 1995 ASSERT(0 <= type && type <= LAST_TYPE); 1996 promoted_histogram_[type].increment_number(1); 1997 promoted_histogram_[type].increment_bytes(obj->Size()); 1998 } 1999 2000 2001 size_t NewSpace::CommittedPhysicalMemory() { 2002 if (!VirtualMemory::HasLazyCommits()) return CommittedMemory(); 2003 MemoryChunk::UpdateHighWaterMark(allocation_info_.top()); 2004 size_t size = to_space_.CommittedPhysicalMemory(); 2005 if (from_space_.is_committed()) { 2006 size += from_space_.CommittedPhysicalMemory(); 2007 } 2008 return size; 2009 } 2010 2011 2012 // ----------------------------------------------------------------------------- 2013 // Free lists for old object spaces implementation 2014 2015 void FreeListNode::set_size(Heap* heap, int size_in_bytes) { 2016 ASSERT(size_in_bytes > 0); 2017 ASSERT(IsAligned(size_in_bytes, kPointerSize)); 2018 2019 // We write a map and possibly size information to the block. If the block 2020 // is big enough to be a FreeSpace with at least one extra word (the next 2021 // pointer), we set its map to be the free space map and its size to an 2022 // appropriate array length for the desired size from HeapObject::Size(). 2023 // If the block is too small (eg, one or two words), to hold both a size 2024 // field and a next pointer, we give it a filler map that gives it the 2025 // correct size. 2026 if (size_in_bytes > FreeSpace::kHeaderSize) { 2027 set_map_no_write_barrier(heap->raw_unchecked_free_space_map()); 2028 // Can't use FreeSpace::cast because it fails during deserialization. 2029 FreeSpace* this_as_free_space = reinterpret_cast<FreeSpace*>(this); 2030 this_as_free_space->set_size(size_in_bytes); 2031 } else if (size_in_bytes == kPointerSize) { 2032 set_map_no_write_barrier(heap->raw_unchecked_one_pointer_filler_map()); 2033 } else if (size_in_bytes == 2 * kPointerSize) { 2034 set_map_no_write_barrier(heap->raw_unchecked_two_pointer_filler_map()); 2035 } else { 2036 UNREACHABLE(); 2037 } 2038 // We would like to ASSERT(Size() == size_in_bytes) but this would fail during 2039 // deserialization because the free space map is not done yet. 2040 } 2041 2042 2043 FreeListNode* FreeListNode::next() { 2044 ASSERT(IsFreeListNode(this)); 2045 if (map() == GetHeap()->raw_unchecked_free_space_map()) { 2046 ASSERT(map() == NULL || Size() >= kNextOffset + kPointerSize); 2047 return reinterpret_cast<FreeListNode*>( 2048 Memory::Address_at(address() + kNextOffset)); 2049 } else { 2050 return reinterpret_cast<FreeListNode*>( 2051 Memory::Address_at(address() + kPointerSize)); 2052 } 2053 } 2054 2055 2056 FreeListNode** FreeListNode::next_address() { 2057 ASSERT(IsFreeListNode(this)); 2058 if (map() == GetHeap()->raw_unchecked_free_space_map()) { 2059 ASSERT(Size() >= kNextOffset + kPointerSize); 2060 return reinterpret_cast<FreeListNode**>(address() + kNextOffset); 2061 } else { 2062 return reinterpret_cast<FreeListNode**>(address() + kPointerSize); 2063 } 2064 } 2065 2066 2067 void FreeListNode::set_next(FreeListNode* next) { 2068 ASSERT(IsFreeListNode(this)); 2069 // While we are booting the VM the free space map will actually be null. So 2070 // we have to make sure that we don't try to use it for anything at that 2071 // stage. 2072 if (map() == GetHeap()->raw_unchecked_free_space_map()) { 2073 ASSERT(map() == NULL || Size() >= kNextOffset + kPointerSize); 2074 Memory::Address_at(address() + kNextOffset) = 2075 reinterpret_cast<Address>(next); 2076 } else { 2077 Memory::Address_at(address() + kPointerSize) = 2078 reinterpret_cast<Address>(next); 2079 } 2080 } 2081 2082 2083 intptr_t FreeListCategory::Concatenate(FreeListCategory* category) { 2084 intptr_t free_bytes = 0; 2085 if (category->top_ != NULL) { 2086 ASSERT(category->end_ != NULL); 2087 // This is safe (not going to deadlock) since Concatenate operations 2088 // are never performed on the same free lists at the same time in 2089 // reverse order. 2090 LockGuard<Mutex> target_lock_guard(mutex()); 2091 LockGuard<Mutex> source_lock_guard(category->mutex()); 2092 free_bytes = category->available(); 2093 if (end_ == NULL) { 2094 end_ = category->end(); 2095 } else { 2096 category->end()->set_next(top_); 2097 } 2098 top_ = category->top(); 2099 available_ += category->available(); 2100 category->Reset(); 2101 } 2102 return free_bytes; 2103 } 2104 2105 2106 void FreeListCategory::Reset() { 2107 top_ = NULL; 2108 end_ = NULL; 2109 available_ = 0; 2110 } 2111 2112 2113 intptr_t FreeListCategory::EvictFreeListItemsInList(Page* p) { 2114 int sum = 0; 2115 FreeListNode** n = &top_; 2116 while (*n != NULL) { 2117 if (Page::FromAddress((*n)->address()) == p) { 2118 FreeSpace* free_space = reinterpret_cast<FreeSpace*>(*n); 2119 sum += free_space->Size(); 2120 *n = (*n)->next(); 2121 } else { 2122 n = (*n)->next_address(); 2123 } 2124 } 2125 if (top_ == NULL) { 2126 end_ = NULL; 2127 } 2128 available_ -= sum; 2129 return sum; 2130 } 2131 2132 2133 FreeListNode* FreeListCategory::PickNodeFromList(int *node_size) { 2134 FreeListNode* node = top_; 2135 2136 if (node == NULL) return NULL; 2137 2138 while (node != NULL && 2139 Page::FromAddress(node->address())->IsEvacuationCandidate()) { 2140 available_ -= reinterpret_cast<FreeSpace*>(node)->Size(); 2141 node = node->next(); 2142 } 2143 2144 if (node != NULL) { 2145 set_top(node->next()); 2146 *node_size = reinterpret_cast<FreeSpace*>(node)->Size(); 2147 available_ -= *node_size; 2148 } else { 2149 set_top(NULL); 2150 } 2151 2152 if (top() == NULL) { 2153 set_end(NULL); 2154 } 2155 2156 return node; 2157 } 2158 2159 2160 FreeListNode* FreeListCategory::PickNodeFromList(int size_in_bytes, 2161 int *node_size) { 2162 FreeListNode* node = PickNodeFromList(node_size); 2163 if (node != NULL && *node_size < size_in_bytes) { 2164 Free(node, *node_size); 2165 *node_size = 0; 2166 return NULL; 2167 } 2168 return node; 2169 } 2170 2171 2172 void FreeListCategory::Free(FreeListNode* node, int size_in_bytes) { 2173 node->set_next(top_); 2174 top_ = node; 2175 if (end_ == NULL) { 2176 end_ = node; 2177 } 2178 available_ += size_in_bytes; 2179 } 2180 2181 2182 void FreeListCategory::RepairFreeList(Heap* heap) { 2183 FreeListNode* n = top_; 2184 while (n != NULL) { 2185 Map** map_location = reinterpret_cast<Map**>(n->address()); 2186 if (*map_location == NULL) { 2187 *map_location = heap->free_space_map(); 2188 } else { 2189 ASSERT(*map_location == heap->free_space_map()); 2190 } 2191 n = n->next(); 2192 } 2193 } 2194 2195 2196 FreeList::FreeList(PagedSpace* owner) 2197 : owner_(owner), heap_(owner->heap()) { 2198 Reset(); 2199 } 2200 2201 2202 intptr_t FreeList::Concatenate(FreeList* free_list) { 2203 intptr_t free_bytes = 0; 2204 free_bytes += small_list_.Concatenate(free_list->small_list()); 2205 free_bytes += medium_list_.Concatenate(free_list->medium_list()); 2206 free_bytes += large_list_.Concatenate(free_list->large_list()); 2207 free_bytes += huge_list_.Concatenate(free_list->huge_list()); 2208 return free_bytes; 2209 } 2210 2211 2212 void FreeList::Reset() { 2213 small_list_.Reset(); 2214 medium_list_.Reset(); 2215 large_list_.Reset(); 2216 huge_list_.Reset(); 2217 } 2218 2219 2220 int FreeList::Free(Address start, int size_in_bytes) { 2221 if (size_in_bytes == 0) return 0; 2222 2223 FreeListNode* node = FreeListNode::FromAddress(start); 2224 node->set_size(heap_, size_in_bytes); 2225 Page* page = Page::FromAddress(start); 2226 2227 // Early return to drop too-small blocks on the floor. 2228 if (size_in_bytes < kSmallListMin) { 2229 page->add_non_available_small_blocks(size_in_bytes); 2230 return size_in_bytes; 2231 } 2232 2233 // Insert other blocks at the head of a free list of the appropriate 2234 // magnitude. 2235 if (size_in_bytes <= kSmallListMax) { 2236 small_list_.Free(node, size_in_bytes); 2237 page->add_available_in_small_free_list(size_in_bytes); 2238 } else if (size_in_bytes <= kMediumListMax) { 2239 medium_list_.Free(node, size_in_bytes); 2240 page->add_available_in_medium_free_list(size_in_bytes); 2241 } else if (size_in_bytes <= kLargeListMax) { 2242 large_list_.Free(node, size_in_bytes); 2243 page->add_available_in_large_free_list(size_in_bytes); 2244 } else { 2245 huge_list_.Free(node, size_in_bytes); 2246 page->add_available_in_huge_free_list(size_in_bytes); 2247 } 2248 2249 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2250 return 0; 2251 } 2252 2253 2254 FreeListNode* FreeList::FindNodeFor(int size_in_bytes, int* node_size) { 2255 FreeListNode* node = NULL; 2256 Page* page = NULL; 2257 2258 if (size_in_bytes <= kSmallAllocationMax) { 2259 node = small_list_.PickNodeFromList(node_size); 2260 if (node != NULL) { 2261 ASSERT(size_in_bytes <= *node_size); 2262 page = Page::FromAddress(node->address()); 2263 page->add_available_in_small_free_list(-(*node_size)); 2264 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2265 return node; 2266 } 2267 } 2268 2269 if (size_in_bytes <= kMediumAllocationMax) { 2270 node = medium_list_.PickNodeFromList(node_size); 2271 if (node != NULL) { 2272 ASSERT(size_in_bytes <= *node_size); 2273 page = Page::FromAddress(node->address()); 2274 page->add_available_in_medium_free_list(-(*node_size)); 2275 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2276 return node; 2277 } 2278 } 2279 2280 if (size_in_bytes <= kLargeAllocationMax) { 2281 node = large_list_.PickNodeFromList(node_size); 2282 if (node != NULL) { 2283 ASSERT(size_in_bytes <= *node_size); 2284 page = Page::FromAddress(node->address()); 2285 page->add_available_in_large_free_list(-(*node_size)); 2286 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2287 return node; 2288 } 2289 } 2290 2291 int huge_list_available = huge_list_.available(); 2292 for (FreeListNode** cur = huge_list_.GetTopAddress(); 2293 *cur != NULL; 2294 cur = (*cur)->next_address()) { 2295 FreeListNode* cur_node = *cur; 2296 while (cur_node != NULL && 2297 Page::FromAddress(cur_node->address())->IsEvacuationCandidate()) { 2298 int size = reinterpret_cast<FreeSpace*>(cur_node)->Size(); 2299 huge_list_available -= size; 2300 page = Page::FromAddress(cur_node->address()); 2301 page->add_available_in_huge_free_list(-size); 2302 cur_node = cur_node->next(); 2303 } 2304 2305 *cur = cur_node; 2306 if (cur_node == NULL) { 2307 huge_list_.set_end(NULL); 2308 break; 2309 } 2310 2311 ASSERT((*cur)->map() == heap_->raw_unchecked_free_space_map()); 2312 FreeSpace* cur_as_free_space = reinterpret_cast<FreeSpace*>(*cur); 2313 int size = cur_as_free_space->Size(); 2314 if (size >= size_in_bytes) { 2315 // Large enough node found. Unlink it from the list. 2316 node = *cur; 2317 *cur = node->next(); 2318 *node_size = size; 2319 huge_list_available -= size; 2320 page = Page::FromAddress(node->address()); 2321 page->add_available_in_huge_free_list(-size); 2322 break; 2323 } 2324 } 2325 2326 if (huge_list_.top() == NULL) { 2327 huge_list_.set_end(NULL); 2328 } 2329 huge_list_.set_available(huge_list_available); 2330 2331 if (node != NULL) { 2332 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2333 return node; 2334 } 2335 2336 if (size_in_bytes <= kSmallListMax) { 2337 node = small_list_.PickNodeFromList(size_in_bytes, node_size); 2338 if (node != NULL) { 2339 ASSERT(size_in_bytes <= *node_size); 2340 page = Page::FromAddress(node->address()); 2341 page->add_available_in_small_free_list(-(*node_size)); 2342 } 2343 } else if (size_in_bytes <= kMediumListMax) { 2344 node = medium_list_.PickNodeFromList(size_in_bytes, node_size); 2345 if (node != NULL) { 2346 ASSERT(size_in_bytes <= *node_size); 2347 page = Page::FromAddress(node->address()); 2348 page->add_available_in_medium_free_list(-(*node_size)); 2349 } 2350 } else if (size_in_bytes <= kLargeListMax) { 2351 node = large_list_.PickNodeFromList(size_in_bytes, node_size); 2352 if (node != NULL) { 2353 ASSERT(size_in_bytes <= *node_size); 2354 page = Page::FromAddress(node->address()); 2355 page->add_available_in_large_free_list(-(*node_size)); 2356 } 2357 } 2358 2359 ASSERT(IsVeryLong() || available() == SumFreeLists()); 2360 return node; 2361 } 2362 2363 2364 // Allocation on the old space free list. If it succeeds then a new linear 2365 // allocation space has been set up with the top and limit of the space. If 2366 // the allocation fails then NULL is returned, and the caller can perform a GC 2367 // or allocate a new page before retrying. 2368 HeapObject* FreeList::Allocate(int size_in_bytes) { 2369 ASSERT(0 < size_in_bytes); 2370 ASSERT(size_in_bytes <= kMaxBlockSize); 2371 ASSERT(IsAligned(size_in_bytes, kPointerSize)); 2372 // Don't free list allocate if there is linear space available. 2373 ASSERT(owner_->limit() - owner_->top() < size_in_bytes); 2374 2375 int old_linear_size = static_cast<int>(owner_->limit() - owner_->top()); 2376 // Mark the old linear allocation area with a free space map so it can be 2377 // skipped when scanning the heap. This also puts it back in the free list 2378 // if it is big enough. 2379 owner_->Free(owner_->top(), old_linear_size); 2380 2381 owner_->heap()->incremental_marking()->OldSpaceStep( 2382 size_in_bytes - old_linear_size); 2383 2384 int new_node_size = 0; 2385 FreeListNode* new_node = FindNodeFor(size_in_bytes, &new_node_size); 2386 if (new_node == NULL) { 2387 owner_->SetTopAndLimit(NULL, NULL); 2388 return NULL; 2389 } 2390 2391 int bytes_left = new_node_size - size_in_bytes; 2392 ASSERT(bytes_left >= 0); 2393 2394 #ifdef DEBUG 2395 for (int i = 0; i < size_in_bytes / kPointerSize; i++) { 2396 reinterpret_cast<Object**>(new_node->address())[i] = 2397 Smi::FromInt(kCodeZapValue); 2398 } 2399 #endif 2400 2401 // The old-space-step might have finished sweeping and restarted marking. 2402 // Verify that it did not turn the page of the new node into an evacuation 2403 // candidate. 2404 ASSERT(!MarkCompactCollector::IsOnEvacuationCandidate(new_node)); 2405 2406 const int kThreshold = IncrementalMarking::kAllocatedThreshold; 2407 2408 // Memory in the linear allocation area is counted as allocated. We may free 2409 // a little of this again immediately - see below. 2410 owner_->Allocate(new_node_size); 2411 2412 if (owner_->heap()->inline_allocation_disabled()) { 2413 // Keep the linear allocation area empty if requested to do so, just 2414 // return area back to the free list instead. 2415 owner_->Free(new_node->address() + size_in_bytes, bytes_left); 2416 ASSERT(owner_->top() == NULL && owner_->limit() == NULL); 2417 } else if (bytes_left > kThreshold && 2418 owner_->heap()->incremental_marking()->IsMarkingIncomplete() && 2419 FLAG_incremental_marking_steps) { 2420 int linear_size = owner_->RoundSizeDownToObjectAlignment(kThreshold); 2421 // We don't want to give too large linear areas to the allocator while 2422 // incremental marking is going on, because we won't check again whether 2423 // we want to do another increment until the linear area is used up. 2424 owner_->Free(new_node->address() + size_in_bytes + linear_size, 2425 new_node_size - size_in_bytes - linear_size); 2426 owner_->SetTopAndLimit(new_node->address() + size_in_bytes, 2427 new_node->address() + size_in_bytes + linear_size); 2428 } else if (bytes_left > 0) { 2429 // Normally we give the rest of the node to the allocator as its new 2430 // linear allocation area. 2431 owner_->SetTopAndLimit(new_node->address() + size_in_bytes, 2432 new_node->address() + new_node_size); 2433 } else { 2434 // TODO(gc) Try not freeing linear allocation region when bytes_left 2435 // are zero. 2436 owner_->SetTopAndLimit(NULL, NULL); 2437 } 2438 2439 return new_node; 2440 } 2441 2442 2443 intptr_t FreeList::EvictFreeListItems(Page* p) { 2444 intptr_t sum = huge_list_.EvictFreeListItemsInList(p); 2445 p->set_available_in_huge_free_list(0); 2446 2447 if (sum < p->area_size()) { 2448 sum += small_list_.EvictFreeListItemsInList(p) + 2449 medium_list_.EvictFreeListItemsInList(p) + 2450 large_list_.EvictFreeListItemsInList(p); 2451 p->set_available_in_small_free_list(0); 2452 p->set_available_in_medium_free_list(0); 2453 p->set_available_in_large_free_list(0); 2454 } 2455 2456 return sum; 2457 } 2458 2459 2460 void FreeList::RepairLists(Heap* heap) { 2461 small_list_.RepairFreeList(heap); 2462 medium_list_.RepairFreeList(heap); 2463 large_list_.RepairFreeList(heap); 2464 huge_list_.RepairFreeList(heap); 2465 } 2466 2467 2468 #ifdef DEBUG 2469 intptr_t FreeListCategory::SumFreeList() { 2470 intptr_t sum = 0; 2471 FreeListNode* cur = top_; 2472 while (cur != NULL) { 2473 ASSERT(cur->map() == cur->GetHeap()->raw_unchecked_free_space_map()); 2474 FreeSpace* cur_as_free_space = reinterpret_cast<FreeSpace*>(cur); 2475 sum += cur_as_free_space->Size(); 2476 cur = cur->next(); 2477 } 2478 return sum; 2479 } 2480 2481 2482 static const int kVeryLongFreeList = 500; 2483 2484 2485 int FreeListCategory::FreeListLength() { 2486 int length = 0; 2487 FreeListNode* cur = top_; 2488 while (cur != NULL) { 2489 length++; 2490 cur = cur->next(); 2491 if (length == kVeryLongFreeList) return length; 2492 } 2493 return length; 2494 } 2495 2496 2497 bool FreeList::IsVeryLong() { 2498 if (small_list_.FreeListLength() == kVeryLongFreeList) return true; 2499 if (medium_list_.FreeListLength() == kVeryLongFreeList) return true; 2500 if (large_list_.FreeListLength() == kVeryLongFreeList) return true; 2501 if (huge_list_.FreeListLength() == kVeryLongFreeList) return true; 2502 return false; 2503 } 2504 2505 2506 // This can take a very long time because it is linear in the number of entries 2507 // on the free list, so it should not be called if FreeListLength returns 2508 // kVeryLongFreeList. 2509 intptr_t FreeList::SumFreeLists() { 2510 intptr_t sum = small_list_.SumFreeList(); 2511 sum += medium_list_.SumFreeList(); 2512 sum += large_list_.SumFreeList(); 2513 sum += huge_list_.SumFreeList(); 2514 return sum; 2515 } 2516 #endif 2517 2518 2519 // ----------------------------------------------------------------------------- 2520 // OldSpace implementation 2521 2522 void PagedSpace::PrepareForMarkCompact() { 2523 // We don't have a linear allocation area while sweeping. It will be restored 2524 // on the first allocation after the sweep. 2525 EmptyAllocationInfo(); 2526 2527 // Stop lazy sweeping and clear marking bits for unswept pages. 2528 if (first_unswept_page_ != NULL) { 2529 Page* p = first_unswept_page_; 2530 do { 2531 // Do not use ShouldBeSweptLazily predicate here. 2532 // New evacuation candidates were selected but they still have 2533 // to be swept before collection starts. 2534 if (!p->WasSwept()) { 2535 Bitmap::Clear(p); 2536 if (FLAG_gc_verbose) { 2537 PrintF("Sweeping 0x%" V8PRIxPTR " lazily abandoned.\n", 2538 reinterpret_cast<intptr_t>(p)); 2539 } 2540 } 2541 p = p->next_page(); 2542 } while (p != anchor()); 2543 } 2544 first_unswept_page_ = Page::FromAddress(NULL); 2545 unswept_free_bytes_ = 0; 2546 2547 // Clear the free list before a full GC---it will be rebuilt afterward. 2548 free_list_.Reset(); 2549 } 2550 2551 2552 intptr_t PagedSpace::SizeOfObjects() { 2553 ASSERT(!heap()->IsSweepingComplete() || (unswept_free_bytes_ == 0)); 2554 return Size() - unswept_free_bytes_ - (limit() - top()); 2555 } 2556 2557 2558 // After we have booted, we have created a map which represents free space 2559 // on the heap. If there was already a free list then the elements on it 2560 // were created with the wrong FreeSpaceMap (normally NULL), so we need to 2561 // fix them. 2562 void PagedSpace::RepairFreeListsAfterBoot() { 2563 free_list_.RepairLists(heap()); 2564 } 2565 2566 2567 bool PagedSpace::AdvanceSweeper(intptr_t bytes_to_sweep) { 2568 if (IsLazySweepingComplete()) return true; 2569 2570 intptr_t freed_bytes = 0; 2571 Page* p = first_unswept_page_; 2572 do { 2573 Page* next_page = p->next_page(); 2574 if (ShouldBeSweptLazily(p)) { 2575 if (FLAG_gc_verbose) { 2576 PrintF("Sweeping 0x%" V8PRIxPTR " lazily advanced.\n", 2577 reinterpret_cast<intptr_t>(p)); 2578 } 2579 DecreaseUnsweptFreeBytes(p); 2580 freed_bytes += 2581 MarkCompactCollector:: 2582 SweepConservatively<MarkCompactCollector::SWEEP_SEQUENTIALLY>( 2583 this, NULL, p); 2584 } 2585 p = next_page; 2586 } while (p != anchor() && freed_bytes < bytes_to_sweep); 2587 2588 if (p == anchor()) { 2589 first_unswept_page_ = Page::FromAddress(NULL); 2590 } else { 2591 first_unswept_page_ = p; 2592 } 2593 2594 heap()->FreeQueuedChunks(); 2595 2596 return IsLazySweepingComplete(); 2597 } 2598 2599 2600 void PagedSpace::EvictEvacuationCandidatesFromFreeLists() { 2601 if (allocation_info_.top() >= allocation_info_.limit()) return; 2602 2603 if (Page::FromAllocationTop(allocation_info_.top())-> 2604 IsEvacuationCandidate()) { 2605 // Create filler object to keep page iterable if it was iterable. 2606 int remaining = 2607 static_cast<int>(allocation_info_.limit() - allocation_info_.top()); 2608 heap()->CreateFillerObjectAt(allocation_info_.top(), remaining); 2609 2610 allocation_info_.set_top(NULL); 2611 allocation_info_.set_limit(NULL); 2612 } 2613 } 2614 2615 2616 bool PagedSpace::EnsureSweeperProgress(intptr_t size_in_bytes) { 2617 MarkCompactCollector* collector = heap()->mark_compact_collector(); 2618 if (collector->AreSweeperThreadsActivated()) { 2619 if (collector->IsConcurrentSweepingInProgress()) { 2620 if (collector->StealMemoryFromSweeperThreads(this) < size_in_bytes) { 2621 if (!collector->sequential_sweeping()) { 2622 collector->WaitUntilSweepingCompleted(); 2623 return true; 2624 } 2625 } 2626 return false; 2627 } 2628 return true; 2629 } else { 2630 return AdvanceSweeper(size_in_bytes); 2631 } 2632 } 2633 2634 2635 HeapObject* PagedSpace::SlowAllocateRaw(int size_in_bytes) { 2636 // Allocation in this space has failed. 2637 2638 // If there are unswept pages advance lazy sweeper a bounded number of times 2639 // until we find a size_in_bytes contiguous piece of memory 2640 const int kMaxSweepingTries = 5; 2641 bool sweeping_complete = false; 2642 2643 for (int i = 0; i < kMaxSweepingTries && !sweeping_complete; i++) { 2644 sweeping_complete = EnsureSweeperProgress(size_in_bytes); 2645 2646 // Retry the free list allocation. 2647 HeapObject* object = free_list_.Allocate(size_in_bytes); 2648 if (object != NULL) return object; 2649 } 2650 2651 // Free list allocation failed and there is no next page. Fail if we have 2652 // hit the old generation size limit that should cause a garbage 2653 // collection. 2654 if (!heap()->always_allocate() && 2655 heap()->OldGenerationAllocationLimitReached()) { 2656 return NULL; 2657 } 2658 2659 // Try to expand the space and allocate in the new next page. 2660 if (Expand()) { 2661 ASSERT(CountTotalPages() > 1 || size_in_bytes <= free_list_.available()); 2662 return free_list_.Allocate(size_in_bytes); 2663 } 2664 2665 // Last ditch, sweep all the remaining pages to try to find space. This may 2666 // cause a pause. 2667 if (!IsLazySweepingComplete()) { 2668 EnsureSweeperProgress(kMaxInt); 2669 2670 // Retry the free list allocation. 2671 HeapObject* object = free_list_.Allocate(size_in_bytes); 2672 if (object != NULL) return object; 2673 } 2674 2675 // Finally, fail. 2676 return NULL; 2677 } 2678 2679 2680 #ifdef DEBUG 2681 void PagedSpace::ReportCodeStatistics(Isolate* isolate) { 2682 CommentStatistic* comments_statistics = 2683 isolate->paged_space_comments_statistics(); 2684 ReportCodeKindStatistics(isolate->code_kind_statistics()); 2685 PrintF("Code comment statistics (\" [ comment-txt : size/ " 2686 "count (average)\"):\n"); 2687 for (int i = 0; i <= CommentStatistic::kMaxComments; i++) { 2688 const CommentStatistic& cs = comments_statistics[i]; 2689 if (cs.size > 0) { 2690 PrintF(" %-30s: %10d/%6d (%d)\n", cs.comment, cs.size, cs.count, 2691 cs.size/cs.count); 2692 } 2693 } 2694 PrintF("\n"); 2695 } 2696 2697 2698 void PagedSpace::ResetCodeStatistics(Isolate* isolate) { 2699 CommentStatistic* comments_statistics = 2700 isolate->paged_space_comments_statistics(); 2701 ClearCodeKindStatistics(isolate->code_kind_statistics()); 2702 for (int i = 0; i < CommentStatistic::kMaxComments; i++) { 2703 comments_statistics[i].Clear(); 2704 } 2705 comments_statistics[CommentStatistic::kMaxComments].comment = "Unknown"; 2706 comments_statistics[CommentStatistic::kMaxComments].size = 0; 2707 comments_statistics[CommentStatistic::kMaxComments].count = 0; 2708 } 2709 2710 2711 // Adds comment to 'comment_statistics' table. Performance OK as long as 2712 // 'kMaxComments' is small 2713 static void EnterComment(Isolate* isolate, const char* comment, int delta) { 2714 CommentStatistic* comments_statistics = 2715 isolate->paged_space_comments_statistics(); 2716 // Do not count empty comments 2717 if (delta <= 0) return; 2718 CommentStatistic* cs = &comments_statistics[CommentStatistic::kMaxComments]; 2719 // Search for a free or matching entry in 'comments_statistics': 'cs' 2720 // points to result. 2721 for (int i = 0; i < CommentStatistic::kMaxComments; i++) { 2722 if (comments_statistics[i].comment == NULL) { 2723 cs = &comments_statistics[i]; 2724 cs->comment = comment; 2725 break; 2726 } else if (strcmp(comments_statistics[i].comment, comment) == 0) { 2727 cs = &comments_statistics[i]; 2728 break; 2729 } 2730 } 2731 // Update entry for 'comment' 2732 cs->size += delta; 2733 cs->count += 1; 2734 } 2735 2736 2737 // Call for each nested comment start (start marked with '[ xxx', end marked 2738 // with ']'. RelocIterator 'it' must point to a comment reloc info. 2739 static void CollectCommentStatistics(Isolate* isolate, RelocIterator* it) { 2740 ASSERT(!it->done()); 2741 ASSERT(it->rinfo()->rmode() == RelocInfo::COMMENT); 2742 const char* tmp = reinterpret_cast<const char*>(it->rinfo()->data()); 2743 if (tmp[0] != '[') { 2744 // Not a nested comment; skip 2745 return; 2746 } 2747 2748 // Search for end of nested comment or a new nested comment 2749 const char* const comment_txt = 2750 reinterpret_cast<const char*>(it->rinfo()->data()); 2751 const byte* prev_pc = it->rinfo()->pc(); 2752 int flat_delta = 0; 2753 it->next(); 2754 while (true) { 2755 // All nested comments must be terminated properly, and therefore exit 2756 // from loop. 2757 ASSERT(!it->done()); 2758 if (it->rinfo()->rmode() == RelocInfo::COMMENT) { 2759 const char* const txt = 2760 reinterpret_cast<const char*>(it->rinfo()->data()); 2761 flat_delta += static_cast<int>(it->rinfo()->pc() - prev_pc); 2762 if (txt[0] == ']') break; // End of nested comment 2763 // A new comment 2764 CollectCommentStatistics(isolate, it); 2765 // Skip code that was covered with previous comment 2766 prev_pc = it->rinfo()->pc(); 2767 } 2768 it->next(); 2769 } 2770 EnterComment(isolate, comment_txt, flat_delta); 2771 } 2772 2773 2774 // Collects code size statistics: 2775 // - by code kind 2776 // - by code comment 2777 void PagedSpace::CollectCodeStatistics() { 2778 Isolate* isolate = heap()->isolate(); 2779 HeapObjectIterator obj_it(this); 2780 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) { 2781 if (obj->IsCode()) { 2782 Code* code = Code::cast(obj); 2783 isolate->code_kind_statistics()[code->kind()] += code->Size(); 2784 RelocIterator it(code); 2785 int delta = 0; 2786 const byte* prev_pc = code->instruction_start(); 2787 while (!it.done()) { 2788 if (it.rinfo()->rmode() == RelocInfo::COMMENT) { 2789 delta += static_cast<int>(it.rinfo()->pc() - prev_pc); 2790 CollectCommentStatistics(isolate, &it); 2791 prev_pc = it.rinfo()->pc(); 2792 } 2793 it.next(); 2794 } 2795 2796 ASSERT(code->instruction_start() <= prev_pc && 2797 prev_pc <= code->instruction_end()); 2798 delta += static_cast<int>(code->instruction_end() - prev_pc); 2799 EnterComment(isolate, "NoComment", delta); 2800 } 2801 } 2802 } 2803 2804 2805 void PagedSpace::ReportStatistics() { 2806 int pct = static_cast<int>(Available() * 100 / Capacity()); 2807 PrintF(" capacity: %" V8_PTR_PREFIX "d" 2808 ", waste: %" V8_PTR_PREFIX "d" 2809 ", available: %" V8_PTR_PREFIX "d, %%%d\n", 2810 Capacity(), Waste(), Available(), pct); 2811 2812 if (was_swept_conservatively_) return; 2813 ClearHistograms(heap()->isolate()); 2814 HeapObjectIterator obj_it(this); 2815 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) 2816 CollectHistogramInfo(obj); 2817 ReportHistogram(heap()->isolate(), true); 2818 } 2819 #endif 2820 2821 2822 // ----------------------------------------------------------------------------- 2823 // MapSpace implementation 2824 // TODO(mvstanton): this is weird...the compiler can't make a vtable unless 2825 // there is at least one non-inlined virtual function. I would prefer to hide 2826 // the VerifyObject definition behind VERIFY_HEAP. 2827 2828 void MapSpace::VerifyObject(HeapObject* object) { 2829 CHECK(object->IsMap()); 2830 } 2831 2832 2833 // ----------------------------------------------------------------------------- 2834 // CellSpace and PropertyCellSpace implementation 2835 // TODO(mvstanton): this is weird...the compiler can't make a vtable unless 2836 // there is at least one non-inlined virtual function. I would prefer to hide 2837 // the VerifyObject definition behind VERIFY_HEAP. 2838 2839 void CellSpace::VerifyObject(HeapObject* object) { 2840 CHECK(object->IsCell()); 2841 } 2842 2843 2844 void PropertyCellSpace::VerifyObject(HeapObject* object) { 2845 CHECK(object->IsPropertyCell()); 2846 } 2847 2848 2849 // ----------------------------------------------------------------------------- 2850 // LargeObjectIterator 2851 2852 LargeObjectIterator::LargeObjectIterator(LargeObjectSpace* space) { 2853 current_ = space->first_page_; 2854 size_func_ = NULL; 2855 } 2856 2857 2858 LargeObjectIterator::LargeObjectIterator(LargeObjectSpace* space, 2859 HeapObjectCallback size_func) { 2860 current_ = space->first_page_; 2861 size_func_ = size_func; 2862 } 2863 2864 2865 HeapObject* LargeObjectIterator::Next() { 2866 if (current_ == NULL) return NULL; 2867 2868 HeapObject* object = current_->GetObject(); 2869 current_ = current_->next_page(); 2870 return object; 2871 } 2872 2873 2874 // ----------------------------------------------------------------------------- 2875 // LargeObjectSpace 2876 static bool ComparePointers(void* key1, void* key2) { 2877 return key1 == key2; 2878 } 2879 2880 2881 LargeObjectSpace::LargeObjectSpace(Heap* heap, 2882 intptr_t max_capacity, 2883 AllocationSpace id) 2884 : Space(heap, id, NOT_EXECUTABLE), // Managed on a per-allocation basis 2885 max_capacity_(max_capacity), 2886 first_page_(NULL), 2887 size_(0), 2888 page_count_(0), 2889 objects_size_(0), 2890 chunk_map_(ComparePointers, 1024) {} 2891 2892 2893 bool LargeObjectSpace::SetUp() { 2894 first_page_ = NULL; 2895 size_ = 0; 2896 maximum_committed_ = 0; 2897 page_count_ = 0; 2898 objects_size_ = 0; 2899 chunk_map_.Clear(); 2900 return true; 2901 } 2902 2903 2904 void LargeObjectSpace::TearDown() { 2905 while (first_page_ != NULL) { 2906 LargePage* page = first_page_; 2907 first_page_ = first_page_->next_page(); 2908 LOG(heap()->isolate(), DeleteEvent("LargeObjectChunk", page->address())); 2909 2910 ObjectSpace space = static_cast<ObjectSpace>(1 << identity()); 2911 heap()->isolate()->memory_allocator()->PerformAllocationCallback( 2912 space, kAllocationActionFree, page->size()); 2913 heap()->isolate()->memory_allocator()->Free(page); 2914 } 2915 SetUp(); 2916 } 2917 2918 2919 MaybeObject* LargeObjectSpace::AllocateRaw(int object_size, 2920 Executability executable) { 2921 // Check if we want to force a GC before growing the old space further. 2922 // If so, fail the allocation. 2923 if (!heap()->always_allocate() && 2924 heap()->OldGenerationAllocationLimitReached()) { 2925 return Failure::RetryAfterGC(identity()); 2926 } 2927 2928 if (Size() + object_size > max_capacity_) { 2929 return Failure::RetryAfterGC(identity()); 2930 } 2931 2932 LargePage* page = heap()->isolate()->memory_allocator()-> 2933 AllocateLargePage(object_size, this, executable); 2934 if (page == NULL) return Failure::RetryAfterGC(identity()); 2935 ASSERT(page->area_size() >= object_size); 2936 2937 size_ += static_cast<int>(page->size()); 2938 objects_size_ += object_size; 2939 page_count_++; 2940 page->set_next_page(first_page_); 2941 first_page_ = page; 2942 2943 if (size_ > maximum_committed_) { 2944 maximum_committed_ = size_; 2945 } 2946 2947 // Register all MemoryChunk::kAlignment-aligned chunks covered by 2948 // this large page in the chunk map. 2949 uintptr_t base = reinterpret_cast<uintptr_t>(page) / MemoryChunk::kAlignment; 2950 uintptr_t limit = base + (page->size() - 1) / MemoryChunk::kAlignment; 2951 for (uintptr_t key = base; key <= limit; key++) { 2952 HashMap::Entry* entry = chunk_map_.Lookup(reinterpret_cast<void*>(key), 2953 static_cast<uint32_t>(key), 2954 true); 2955 ASSERT(entry != NULL); 2956 entry->value = page; 2957 } 2958 2959 HeapObject* object = page->GetObject(); 2960 2961 if (Heap::ShouldZapGarbage()) { 2962 // Make the object consistent so the heap can be verified in OldSpaceStep. 2963 // We only need to do this in debug builds or if verify_heap is on. 2964 reinterpret_cast<Object**>(object->address())[0] = 2965 heap()->fixed_array_map(); 2966 reinterpret_cast<Object**>(object->address())[1] = Smi::FromInt(0); 2967 } 2968 2969 heap()->incremental_marking()->OldSpaceStep(object_size); 2970 return object; 2971 } 2972 2973 2974 size_t LargeObjectSpace::CommittedPhysicalMemory() { 2975 if (!VirtualMemory::HasLazyCommits()) return CommittedMemory(); 2976 size_t size = 0; 2977 LargePage* current = first_page_; 2978 while (current != NULL) { 2979 size += current->CommittedPhysicalMemory(); 2980 current = current->next_page(); 2981 } 2982 return size; 2983 } 2984 2985 2986 // GC support 2987 MaybeObject* LargeObjectSpace::FindObject(Address a) { 2988 LargePage* page = FindPage(a); 2989 if (page != NULL) { 2990 return page->GetObject(); 2991 } 2992 return Failure::Exception(); 2993 } 2994 2995 2996 LargePage* LargeObjectSpace::FindPage(Address a) { 2997 uintptr_t key = reinterpret_cast<uintptr_t>(a) / MemoryChunk::kAlignment; 2998 HashMap::Entry* e = chunk_map_.Lookup(reinterpret_cast<void*>(key), 2999 static_cast<uint32_t>(key), 3000 false); 3001 if (e != NULL) { 3002 ASSERT(e->value != NULL); 3003 LargePage* page = reinterpret_cast<LargePage*>(e->value); 3004 ASSERT(page->is_valid()); 3005 if (page->Contains(a)) { 3006 return page; 3007 } 3008 } 3009 return NULL; 3010 } 3011 3012 3013 void LargeObjectSpace::FreeUnmarkedObjects() { 3014 LargePage* previous = NULL; 3015 LargePage* current = first_page_; 3016 while (current != NULL) { 3017 HeapObject* object = current->GetObject(); 3018 // Can this large page contain pointers to non-trivial objects. No other 3019 // pointer object is this big. 3020 bool is_pointer_object = object->IsFixedArray(); 3021 MarkBit mark_bit = Marking::MarkBitFrom(object); 3022 if (mark_bit.Get()) { 3023 mark_bit.Clear(); 3024 Page::FromAddress(object->address())->ResetProgressBar(); 3025 Page::FromAddress(object->address())->ResetLiveBytes(); 3026 previous = current; 3027 current = current->next_page(); 3028 } else { 3029 LargePage* page = current; 3030 // Cut the chunk out from the chunk list. 3031 current = current->next_page(); 3032 if (previous == NULL) { 3033 first_page_ = current; 3034 } else { 3035 previous->set_next_page(current); 3036 } 3037 3038 // Free the chunk. 3039 heap()->mark_compact_collector()->ReportDeleteIfNeeded( 3040 object, heap()->isolate()); 3041 size_ -= static_cast<int>(page->size()); 3042 objects_size_ -= object->Size(); 3043 page_count_--; 3044 3045 // Remove entries belonging to this page. 3046 // Use variable alignment to help pass length check (<= 80 characters) 3047 // of single line in tools/presubmit.py. 3048 const intptr_t alignment = MemoryChunk::kAlignment; 3049 uintptr_t base = reinterpret_cast<uintptr_t>(page)/alignment; 3050 uintptr_t limit = base + (page->size()-1)/alignment; 3051 for (uintptr_t key = base; key <= limit; key++) { 3052 chunk_map_.Remove(reinterpret_cast<void*>(key), 3053 static_cast<uint32_t>(key)); 3054 } 3055 3056 if (is_pointer_object) { 3057 heap()->QueueMemoryChunkForFree(page); 3058 } else { 3059 heap()->isolate()->memory_allocator()->Free(page); 3060 } 3061 } 3062 } 3063 heap()->FreeQueuedChunks(); 3064 } 3065 3066 3067 bool LargeObjectSpace::Contains(HeapObject* object) { 3068 Address address = object->address(); 3069 MemoryChunk* chunk = MemoryChunk::FromAddress(address); 3070 3071 bool owned = (chunk->owner() == this); 3072 3073 SLOW_ASSERT(!owned || !FindObject(address)->IsFailure()); 3074 3075 return owned; 3076 } 3077 3078 3079 #ifdef VERIFY_HEAP 3080 // We do not assume that the large object iterator works, because it depends 3081 // on the invariants we are checking during verification. 3082 void LargeObjectSpace::Verify() { 3083 for (LargePage* chunk = first_page_; 3084 chunk != NULL; 3085 chunk = chunk->next_page()) { 3086 // Each chunk contains an object that starts at the large object page's 3087 // object area start. 3088 HeapObject* object = chunk->GetObject(); 3089 Page* page = Page::FromAddress(object->address()); 3090 CHECK(object->address() == page->area_start()); 3091 3092 // The first word should be a map, and we expect all map pointers to be 3093 // in map space. 3094 Map* map = object->map(); 3095 CHECK(map->IsMap()); 3096 CHECK(heap()->map_space()->Contains(map)); 3097 3098 // We have only code, sequential strings, external strings 3099 // (sequential strings that have been morphed into external 3100 // strings), fixed arrays, and byte arrays in large object space. 3101 CHECK(object->IsCode() || object->IsSeqString() || 3102 object->IsExternalString() || object->IsFixedArray() || 3103 object->IsFixedDoubleArray() || object->IsByteArray()); 3104 3105 // The object itself should look OK. 3106 object->Verify(); 3107 3108 // Byte arrays and strings don't have interior pointers. 3109 if (object->IsCode()) { 3110 VerifyPointersVisitor code_visitor; 3111 object->IterateBody(map->instance_type(), 3112 object->Size(), 3113 &code_visitor); 3114 } else if (object->IsFixedArray()) { 3115 FixedArray* array = FixedArray::cast(object); 3116 for (int j = 0; j < array->length(); j++) { 3117 Object* element = array->get(j); 3118 if (element->IsHeapObject()) { 3119 HeapObject* element_object = HeapObject::cast(element); 3120 CHECK(heap()->Contains(element_object)); 3121 CHECK(element_object->map()->IsMap()); 3122 } 3123 } 3124 } 3125 } 3126 } 3127 #endif 3128 3129 3130 #ifdef DEBUG 3131 void LargeObjectSpace::Print() { 3132 LargeObjectIterator it(this); 3133 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 3134 obj->Print(); 3135 } 3136 } 3137 3138 3139 void LargeObjectSpace::ReportStatistics() { 3140 PrintF(" size: %" V8_PTR_PREFIX "d\n", size_); 3141 int num_objects = 0; 3142 ClearHistograms(heap()->isolate()); 3143 LargeObjectIterator it(this); 3144 for (HeapObject* obj = it.Next(); obj != NULL; obj = it.Next()) { 3145 num_objects++; 3146 CollectHistogramInfo(obj); 3147 } 3148 3149 PrintF(" number of objects %d, " 3150 "size of objects %" V8_PTR_PREFIX "d\n", num_objects, objects_size_); 3151 if (num_objects > 0) ReportHistogram(heap()->isolate(), false); 3152 } 3153 3154 3155 void LargeObjectSpace::CollectCodeStatistics() { 3156 Isolate* isolate = heap()->isolate(); 3157 LargeObjectIterator obj_it(this); 3158 for (HeapObject* obj = obj_it.Next(); obj != NULL; obj = obj_it.Next()) { 3159 if (obj->IsCode()) { 3160 Code* code = Code::cast(obj); 3161 isolate->code_kind_statistics()[code->kind()] += code->Size(); 3162 } 3163 } 3164 } 3165 3166 3167 void Page::Print() { 3168 // Make a best-effort to print the objects in the page. 3169 PrintF("Page@%p in %s\n", 3170 this->address(), 3171 AllocationSpaceName(this->owner()->identity())); 3172 printf(" --------------------------------------\n"); 3173 HeapObjectIterator objects(this, heap()->GcSafeSizeOfOldObjectFunction()); 3174 unsigned mark_size = 0; 3175 for (HeapObject* object = objects.Next(); 3176 object != NULL; 3177 object = objects.Next()) { 3178 bool is_marked = Marking::MarkBitFrom(object).Get(); 3179 PrintF(" %c ", (is_marked ? '!' : ' ')); // Indent a little. 3180 if (is_marked) { 3181 mark_size += heap()->GcSafeSizeOfOldObjectFunction()(object); 3182 } 3183 object->ShortPrint(); 3184 PrintF("\n"); 3185 } 3186 printf(" --------------------------------------\n"); 3187 printf(" Marked: %x, LiveCount: %x\n", mark_size, LiveBytes()); 3188 } 3189 3190 #endif // DEBUG 3191 3192 } } // namespace v8::internal 3193