1 /* 2 * Copyright (c) 2005, 2020, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP 26 #define SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP 27 28 #include "gc/parallel/mutableSpace.hpp" 29 #include "gc/parallel/objectStartArray.hpp" 30 #include "gc/parallel/parMarkBitMap.hpp" 31 #include "gc/parallel/parallelScavengeHeap.hpp" 32 #include "gc/shared/collectedHeap.hpp" 33 #include "gc/shared/collectorCounters.hpp" 34 #include "oops/oop.hpp" 35 #include "runtime/atomic.hpp" 36 #include "runtime/orderAccess.hpp" 37 38 class ParallelScavengeHeap; 39 class PSAdaptiveSizePolicy; 40 class PSYoungGen; 41 class PSOldGen; 42 class ParCompactionManager; 43 class PSParallelCompact; 44 class MoveAndUpdateClosure; 45 class RefProcTaskExecutor; 46 class ParallelOldTracer; 47 class STWGCTimer; 48 49 // The SplitInfo class holds the information needed to 'split' a source region 50 // so that the live data can be copied to two destination *spaces*. Normally, 51 // all the live data in a region is copied to a single destination space (e.g., 52 // everything live in a region in eden is copied entirely into the old gen). 53 // However, when the heap is nearly full, all the live data in eden may not fit 54 // into the old gen. Copying only some of the regions from eden to old gen 55 // requires finding a region that does not contain a partial object (i.e., no 56 // live object crosses the region boundary) somewhere near the last object that 57 // does fit into the old gen. Since it's not always possible to find such a 58 // region, splitting is necessary for predictable behavior. 59 // 60 // A region is always split at the end of the partial object. This avoids 61 // additional tests when calculating the new location of a pointer, which is a 62 // very hot code path. The partial object and everything to its left will be 63 // copied to another space (call it dest_space_1). The live data to the right 64 // of the partial object will be copied either within the space itself, or to a 65 // different destination space (distinct from dest_space_1). 66 // 67 // Split points are identified during the summary phase, when region 68 // destinations are computed: data about the split, including the 69 // partial_object_size, is recorded in a SplitInfo record and the 70 // partial_object_size field in the summary data is set to zero. The zeroing is 71 // possible (and necessary) since the partial object will move to a different 72 // destination space than anything to its right, thus the partial object should 73 // not affect the locations of any objects to its right. 74 // 75 // The recorded data is used during the compaction phase, but only rarely: when 76 // the partial object on the split region will be copied across a destination 77 // region boundary. This test is made once each time a region is filled, and is 78 // a simple address comparison, so the overhead is negligible (see 79 // PSParallelCompact::first_src_addr()). 80 // 81 // Notes: 82 // 83 // Only regions with partial objects are split; a region without a partial 84 // object does not need any extra bookkeeping. 85 // 86 // At most one region is split per space, so the amount of data required is 87 // constant. 88 // 89 // A region is split only when the destination space would overflow. Once that 90 // happens, the destination space is abandoned and no other data (even from 91 // other source spaces) is targeted to that destination space. Abandoning the 92 // destination space may leave a somewhat large unused area at the end, if a 93 // large object caused the overflow. 94 // 95 // Future work: 96 // 97 // More bookkeeping would be required to continue to use the destination space. 98 // The most general solution would allow data from regions in two different 99 // source spaces to be "joined" in a single destination region. At the very 100 // least, additional code would be required in next_src_region() to detect the 101 // join and skip to an out-of-order source region. If the join region was also 102 // the last destination region to which a split region was copied (the most 103 // likely case), then additional work would be needed to get fill_region() to 104 // stop iteration and switch to a new source region at the right point. Basic 105 // idea would be to use a fake value for the top of the source space. It is 106 // doable, if a bit tricky. 107 // 108 // A simpler (but less general) solution would fill the remainder of the 109 // destination region with a dummy object and continue filling the next 110 // destination region. 111 112 class SplitInfo 113 { 114 public: 115 // Return true if this split info is valid (i.e., if a split has been 116 // recorded). The very first region cannot have a partial object and thus is 117 // never split, so 0 is the 'invalid' value. 118 bool is_valid() const { return _src_region_idx > 0; } 119 120 // Return true if this split holds data for the specified source region. 121 inline bool is_split(size_t source_region) const; 122 123 // The index of the split region, the size of the partial object on that 124 // region and the destination of the partial object. 125 size_t src_region_idx() const { return _src_region_idx; } 126 size_t partial_obj_size() const { return _partial_obj_size; } 127 HeapWord* destination() const { return _destination; } 128 129 // The destination count of the partial object referenced by this split 130 // (either 1 or 2). This must be added to the destination count of the 131 // remainder of the source region. 132 unsigned int destination_count() const { return _destination_count; } 133 134 // If a word within the partial object will be written to the first word of a 135 // destination region, this is the address of the destination region; 136 // otherwise this is NULL. 137 HeapWord* dest_region_addr() const { return _dest_region_addr; } 138 139 // If a word within the partial object will be written to the first word of a 140 // destination region, this is the address of that word within the partial 141 // object; otherwise this is NULL. 142 HeapWord* first_src_addr() const { return _first_src_addr; } 143 144 // Record the data necessary to split the region src_region_idx. 145 void record(size_t src_region_idx, size_t partial_obj_size, 146 HeapWord* destination); 147 148 void clear(); 149 150 DEBUG_ONLY(void verify_clear();) 151 152 private: 153 size_t _src_region_idx; 154 size_t _partial_obj_size; 155 HeapWord* _destination; 156 unsigned int _destination_count; 157 HeapWord* _dest_region_addr; 158 HeapWord* _first_src_addr; 159 }; 160 161 inline bool SplitInfo::is_split(size_t region_idx) const 162 { 163 return _src_region_idx == region_idx && is_valid(); 164 } 165 166 class SpaceInfo 167 { 168 public: 169 MutableSpace* space() const { return _space; } 170 171 // Where the free space will start after the collection. Valid only after the 172 // summary phase completes. 173 HeapWord* new_top() const { return _new_top; } 174 175 // Allows new_top to be set. 176 HeapWord** new_top_addr() { return &_new_top; } 177 178 // Where the smallest allowable dense prefix ends (used only for perm gen). 179 HeapWord* min_dense_prefix() const { return _min_dense_prefix; } 180 181 // Where the dense prefix ends, or the compacted region begins. 182 HeapWord* dense_prefix() const { return _dense_prefix; } 183 184 // The start array for the (generation containing the) space, or NULL if there 185 // is no start array. 186 ObjectStartArray* start_array() const { return _start_array; } 187 188 SplitInfo& split_info() { return _split_info; } 189 190 void set_space(MutableSpace* s) { _space = s; } 191 void set_new_top(HeapWord* addr) { _new_top = addr; } 192 void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; } 193 void set_dense_prefix(HeapWord* addr) { _dense_prefix = addr; } 194 void set_start_array(ObjectStartArray* s) { _start_array = s; } 195 196 void publish_new_top() const { _space->set_top(_new_top); } 197 198 private: 199 MutableSpace* _space; 200 HeapWord* _new_top; 201 HeapWord* _min_dense_prefix; 202 HeapWord* _dense_prefix; 203 ObjectStartArray* _start_array; 204 SplitInfo _split_info; 205 }; 206 207 class ParallelCompactData 208 { 209 public: 210 // Sizes are in HeapWords, unless indicated otherwise. 211 static const size_t Log2RegionSize; 212 static const size_t RegionSize; 213 static const size_t RegionSizeBytes; 214 215 // Mask for the bits in a size_t to get an offset within a region. 216 static const size_t RegionSizeOffsetMask; 217 // Mask for the bits in a pointer to get an offset within a region. 218 static const size_t RegionAddrOffsetMask; 219 // Mask for the bits in a pointer to get the address of the start of a region. 220 static const size_t RegionAddrMask; 221 222 static const size_t Log2BlockSize; 223 static const size_t BlockSize; 224 static const size_t BlockSizeBytes; 225 226 static const size_t BlockSizeOffsetMask; 227 static const size_t BlockAddrOffsetMask; 228 static const size_t BlockAddrMask; 229 230 static const size_t BlocksPerRegion; 231 static const size_t Log2BlocksPerRegion; 232 233 class RegionData 234 { 235 public: 236 // Destination address of the region. 237 HeapWord* destination() const { return _destination; } 238 239 // The first region containing data destined for this region. 240 size_t source_region() const { return _source_region; } 241 242 // Reuse _source_region to store the corresponding shadow region index 243 size_t shadow_region() const { return _source_region; } 244 245 // The object (if any) starting in this region and ending in a different 246 // region that could not be updated during the main (parallel) compaction 247 // phase. This is different from _partial_obj_addr, which is an object that 248 // extends onto a source region. However, the two uses do not overlap in 249 // time, so the same field is used to save space. 250 HeapWord* deferred_obj_addr() const { return _partial_obj_addr; } 251 252 // The starting address of the partial object extending onto the region. 253 HeapWord* partial_obj_addr() const { return _partial_obj_addr; } 254 255 // Size of the partial object extending onto the region (words). 256 size_t partial_obj_size() const { return _partial_obj_size; } 257 258 // Size of live data that lies within this region due to objects that start 259 // in this region (words). This does not include the partial object 260 // extending onto the region (if any), or the part of an object that extends 261 // onto the next region (if any). 262 size_t live_obj_size() const { return _dc_and_los & los_mask; } 263 264 // Total live data that lies within the region (words). 265 size_t data_size() const { return partial_obj_size() + live_obj_size(); } 266 267 // The destination_count is the number of other regions to which data from 268 // this region will be copied. At the end of the summary phase, the valid 269 // values of destination_count are 270 // 271 // 0 - data from the region will be compacted completely into itself, or the 272 // region is empty. The region can be claimed and then filled. 273 // 1 - data from the region will be compacted into 1 other region; some 274 // data from the region may also be compacted into the region itself. 275 // 2 - data from the region will be copied to 2 other regions. 276 // 277 // During compaction as regions are emptied, the destination_count is 278 // decremented (atomically) and when it reaches 0, it can be claimed and 279 // then filled. 280 // 281 // A region is claimed for processing by atomically changing the 282 // destination_count to the claimed value (dc_claimed). After a region has 283 // been filled, the destination_count should be set to the completed value 284 // (dc_completed). 285 inline uint destination_count() const; 286 inline uint destination_count_raw() const; 287 288 // Whether the block table for this region has been filled. 289 inline bool blocks_filled() const; 290 291 // Number of times the block table was filled. 292 DEBUG_ONLY(inline size_t blocks_filled_count() const;) 293 294 // The location of the java heap data that corresponds to this region. 295 inline HeapWord* data_location() const; 296 297 // The highest address referenced by objects in this region. 298 inline HeapWord* highest_ref() const; 299 300 // Whether this region is available to be claimed, has been claimed, or has 301 // been completed. 302 // 303 // Minor subtlety: claimed() returns true if the region is marked 304 // completed(), which is desirable since a region must be claimed before it 305 // can be completed. 306 bool available() const { return _dc_and_los < dc_one; } 307 bool claimed() const { return _dc_and_los >= dc_claimed; } 308 bool completed() const { return _dc_and_los >= dc_completed; } 309 310 // These are not atomic. 311 void set_destination(HeapWord* addr) { _destination = addr; } 312 void set_source_region(size_t region) { _source_region = region; } 313 void set_shadow_region(size_t region) { _source_region = region; } 314 void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } 315 void set_partial_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } 316 void set_partial_obj_size(size_t words) { 317 _partial_obj_size = (region_sz_t) words; 318 } 319 inline void set_blocks_filled(); 320 321 inline void set_destination_count(uint count); 322 inline void set_live_obj_size(size_t words); 323 inline void set_data_location(HeapWord* addr); 324 inline void set_completed(); 325 inline bool claim_unsafe(); 326 327 // These are atomic. 328 inline void add_live_obj(size_t words); 329 inline void set_highest_ref(HeapWord* addr); 330 inline void decrement_destination_count(); 331 inline bool claim(); 332 333 // Possible values of _shadow_state, and transition is as follows 334 // Normal Path: 335 // UnusedRegion -> mark_normal() -> NormalRegion 336 // Shadow Path: 337 // UnusedRegion -> mark_shadow() -> ShadowRegion -> 338 // mark_filled() -> FilledShadow -> mark_copied() -> CopiedShadow 339 static const int UnusedRegion = 0; // The region is not collected yet 340 static const int ShadowRegion = 1; // Stolen by an idle thread, and a shadow region is created for it 341 static const int FilledShadow = 2; // Its shadow region has been filled and ready to be copied back 342 static const int CopiedShadow = 3; // The data of the shadow region has been copied back 343 static const int NormalRegion = 4; // The region will be collected by the original parallel algorithm 344 345 // Mark the current region as normal or shadow to enter different processing paths 346 inline bool mark_normal(); 347 inline bool mark_shadow(); 348 // Mark the shadow region as filled and ready to be copied back 349 inline void mark_filled(); 350 // Mark the shadow region as copied back to avoid double copying. 351 inline bool mark_copied(); 352 // Special case: see the comment in PSParallelCompact::fill_and_update_shadow_region. 353 // Return to the normal path here 354 inline void shadow_to_normal(); 355 356 357 int shadow_state() { return _shadow_state; } 358 359 private: 360 // The type used to represent object sizes within a region. 361 typedef uint region_sz_t; 362 363 // Constants for manipulating the _dc_and_los field, which holds both the 364 // destination count and live obj size. The live obj size lives at the 365 // least significant end so no masking is necessary when adding. 366 static const region_sz_t dc_shift; // Shift amount. 367 static const region_sz_t dc_mask; // Mask for destination count. 368 static const region_sz_t dc_one; // 1, shifted appropriately. 369 static const region_sz_t dc_claimed; // Region has been claimed. 370 static const region_sz_t dc_completed; // Region has been completed. 371 static const region_sz_t los_mask; // Mask for live obj size. 372 373 HeapWord* _destination; 374 size_t _source_region; 375 HeapWord* _partial_obj_addr; 376 region_sz_t _partial_obj_size; 377 region_sz_t volatile _dc_and_los; 378 bool volatile _blocks_filled; 379 int volatile _shadow_state; 380 381 #ifdef ASSERT 382 size_t _blocks_filled_count; // Number of block table fills. 383 384 // These enable optimizations that are only partially implemented. Use 385 // debug builds to prevent the code fragments from breaking. 386 HeapWord* _data_location; 387 HeapWord* _highest_ref; 388 #endif // #ifdef ASSERT 389 390 #ifdef ASSERT 391 public: 392 uint _pushed; // 0 until region is pushed onto a stack 393 private: 394 #endif 395 }; 396 397 // "Blocks" allow shorter sections of the bitmap to be searched. Each Block 398 // holds an offset, which is the amount of live data in the Region to the left 399 // of the first live object that starts in the Block. 400 class BlockData 401 { 402 public: 403 typedef unsigned short int blk_ofs_t; 404 405 blk_ofs_t offset() const { return _offset; } 406 void set_offset(size_t val) { _offset = (blk_ofs_t)val; } 407 408 private: 409 blk_ofs_t _offset; 410 }; 411 412 public: 413 ParallelCompactData(); 414 bool initialize(MemRegion covered_region); 415 416 size_t region_count() const { return _region_count; } 417 size_t reserved_byte_size() const { return _reserved_byte_size; } 418 419 // Convert region indices to/from RegionData pointers. 420 inline RegionData* region(size_t region_idx) const; 421 inline size_t region(const RegionData* const region_ptr) const; 422 423 size_t block_count() const { return _block_count; } 424 inline BlockData* block(size_t block_idx) const; 425 inline size_t block(const BlockData* block_ptr) const; 426 427 void add_obj(HeapWord* addr, size_t len); 428 void add_obj(oop p, size_t len) { add_obj(cast_from_oop<HeapWord*>(p), len); } 429 430 // Fill in the regions covering [beg, end) so that no data moves; i.e., the 431 // destination of region n is simply the start of region n. The argument beg 432 // must be region-aligned; end need not be. 433 void summarize_dense_prefix(HeapWord* beg, HeapWord* end); 434 435 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info, 436 HeapWord* destination, HeapWord* target_end, 437 HeapWord** target_next); 438 bool summarize(SplitInfo& split_info, 439 HeapWord* source_beg, HeapWord* source_end, 440 HeapWord** source_next, 441 HeapWord* target_beg, HeapWord* target_end, 442 HeapWord** target_next); 443 444 void clear(); 445 void clear_range(size_t beg_region, size_t end_region); 446 void clear_range(HeapWord* beg, HeapWord* end) { 447 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end)); 448 } 449 450 // Return the number of words between addr and the start of the region 451 // containing addr. 452 inline size_t region_offset(const HeapWord* addr) const; 453 454 // Convert addresses to/from a region index or region pointer. 455 inline size_t addr_to_region_idx(const HeapWord* addr) const; 456 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const; 457 inline HeapWord* region_to_addr(size_t region) const; 458 inline HeapWord* region_to_addr(size_t region, size_t offset) const; 459 inline HeapWord* region_to_addr(const RegionData* region) const; 460 461 inline HeapWord* region_align_down(HeapWord* addr) const; 462 inline HeapWord* region_align_up(HeapWord* addr) const; 463 inline bool is_region_aligned(HeapWord* addr) const; 464 465 // Analogous to region_offset() for blocks. 466 size_t block_offset(const HeapWord* addr) const; 467 size_t addr_to_block_idx(const HeapWord* addr) const; 468 size_t addr_to_block_idx(const oop obj) const { 469 return addr_to_block_idx(cast_from_oop<HeapWord*>(obj)); 470 } 471 inline BlockData* addr_to_block_ptr(const HeapWord* addr) const; 472 inline HeapWord* block_to_addr(size_t block) const; 473 inline size_t region_to_block_idx(size_t region) const; 474 475 inline HeapWord* block_align_down(HeapWord* addr) const; 476 inline HeapWord* block_align_up(HeapWord* addr) const; 477 inline bool is_block_aligned(HeapWord* addr) const; 478 479 // Return the address one past the end of the partial object. 480 HeapWord* partial_obj_end(size_t region_idx) const; 481 482 // Return the location of the object after compaction. 483 HeapWord* calc_new_pointer(HeapWord* addr, ParCompactionManager* cm); 484 485 HeapWord* calc_new_pointer(oop p, ParCompactionManager* cm) { 486 return calc_new_pointer(cast_from_oop<HeapWord*>(p), cm); 487 } 488 489 #ifdef ASSERT 490 void verify_clear(const PSVirtualSpace* vspace); 491 void verify_clear(); 492 #endif // #ifdef ASSERT 493 494 private: 495 bool initialize_block_data(); 496 bool initialize_region_data(size_t region_size); 497 PSVirtualSpace* create_vspace(size_t count, size_t element_size); 498 499 private: 500 HeapWord* _region_start; 501 #ifdef ASSERT 502 HeapWord* _region_end; 503 #endif // #ifdef ASSERT 504 505 PSVirtualSpace* _region_vspace; 506 size_t _reserved_byte_size; 507 RegionData* _region_data; 508 size_t _region_count; 509 510 PSVirtualSpace* _block_vspace; 511 BlockData* _block_data; 512 size_t _block_count; 513 }; 514 515 inline uint 516 ParallelCompactData::RegionData::destination_count_raw() const 517 { 518 return _dc_and_los & dc_mask; 519 } 520 521 inline uint 522 ParallelCompactData::RegionData::destination_count() const 523 { 524 return destination_count_raw() >> dc_shift; 525 } 526 527 inline bool 528 ParallelCompactData::RegionData::blocks_filled() const 529 { 530 bool result = _blocks_filled; 531 OrderAccess::acquire(); 532 return result; 533 } 534 535 #ifdef ASSERT 536 inline size_t 537 ParallelCompactData::RegionData::blocks_filled_count() const 538 { 539 return _blocks_filled_count; 540 } 541 #endif // #ifdef ASSERT 542 543 inline void 544 ParallelCompactData::RegionData::set_blocks_filled() 545 { 546 OrderAccess::release(); 547 _blocks_filled = true; 548 // Debug builds count the number of times the table was filled. 549 DEBUG_ONLY(Atomic::inc(&_blocks_filled_count)); 550 } 551 552 inline void 553 ParallelCompactData::RegionData::set_destination_count(uint count) 554 { 555 assert(count <= (dc_completed >> dc_shift), "count too large"); 556 const region_sz_t live_sz = (region_sz_t) live_obj_size(); 557 _dc_and_los = (count << dc_shift) | live_sz; 558 } 559 560 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words) 561 { 562 assert(words <= los_mask, "would overflow"); 563 _dc_and_los = destination_count_raw() | (region_sz_t)words; 564 } 565 566 inline void ParallelCompactData::RegionData::decrement_destination_count() 567 { 568 assert(_dc_and_los < dc_claimed, "already claimed"); 569 assert(_dc_and_los >= dc_one, "count would go negative"); 570 Atomic::add(&_dc_and_los, dc_mask); 571 } 572 573 inline HeapWord* ParallelCompactData::RegionData::data_location() const 574 { 575 DEBUG_ONLY(return _data_location;) 576 NOT_DEBUG(return NULL;) 577 } 578 579 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const 580 { 581 DEBUG_ONLY(return _highest_ref;) 582 NOT_DEBUG(return NULL;) 583 } 584 585 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr) 586 { 587 DEBUG_ONLY(_data_location = addr;) 588 } 589 590 inline void ParallelCompactData::RegionData::set_completed() 591 { 592 assert(claimed(), "must be claimed first"); 593 _dc_and_los = dc_completed | (region_sz_t) live_obj_size(); 594 } 595 596 // MT-unsafe claiming of a region. Should only be used during single threaded 597 // execution. 598 inline bool ParallelCompactData::RegionData::claim_unsafe() 599 { 600 if (available()) { 601 _dc_and_los |= dc_claimed; 602 return true; 603 } 604 return false; 605 } 606 607 inline void ParallelCompactData::RegionData::add_live_obj(size_t words) 608 { 609 assert(words <= (size_t)los_mask - live_obj_size(), "overflow"); 610 Atomic::add(&_dc_and_los, static_cast<region_sz_t>(words)); 611 } 612 613 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr) 614 { 615 #ifdef ASSERT 616 HeapWord* tmp = _highest_ref; 617 while (addr > tmp) { 618 tmp = Atomic::cmpxchg(&_highest_ref, tmp, addr); 619 } 620 #endif // #ifdef ASSERT 621 } 622 623 inline bool ParallelCompactData::RegionData::claim() 624 { 625 const region_sz_t los = static_cast<region_sz_t>(live_obj_size()); 626 const region_sz_t old = Atomic::cmpxchg(&_dc_and_los, los, dc_claimed | los); 627 return old == los; 628 } 629 630 inline bool ParallelCompactData::RegionData::mark_normal() { 631 return Atomic::cmpxchg(&_shadow_state, UnusedRegion, NormalRegion) == UnusedRegion; 632 } 633 634 inline bool ParallelCompactData::RegionData::mark_shadow() { 635 if (_shadow_state != UnusedRegion) return false; 636 return Atomic::cmpxchg(&_shadow_state, UnusedRegion, ShadowRegion) == UnusedRegion; 637 } 638 639 inline void ParallelCompactData::RegionData::mark_filled() { 640 int old = Atomic::cmpxchg(&_shadow_state, ShadowRegion, FilledShadow); 641 assert(old == ShadowRegion, "Fail to mark the region as filled"); 642 } 643 644 inline bool ParallelCompactData::RegionData::mark_copied() { 645 return Atomic::cmpxchg(&_shadow_state, FilledShadow, CopiedShadow) == FilledShadow; 646 } 647 648 void ParallelCompactData::RegionData::shadow_to_normal() { 649 int old = Atomic::cmpxchg(&_shadow_state, ShadowRegion, NormalRegion); 650 assert(old == ShadowRegion, "Fail to mark the region as finish"); 651 } 652 653 inline ParallelCompactData::RegionData* 654 ParallelCompactData::region(size_t region_idx) const 655 { 656 assert(region_idx <= region_count(), "bad arg"); 657 return _region_data + region_idx; 658 } 659 660 inline size_t 661 ParallelCompactData::region(const RegionData* const region_ptr) const 662 { 663 assert(region_ptr >= _region_data, "bad arg"); 664 assert(region_ptr <= _region_data + region_count(), "bad arg"); 665 return pointer_delta(region_ptr, _region_data, sizeof(RegionData)); 666 } 667 668 inline ParallelCompactData::BlockData* 669 ParallelCompactData::block(size_t n) const { 670 assert(n < block_count(), "bad arg"); 671 return _block_data + n; 672 } 673 674 inline size_t 675 ParallelCompactData::region_offset(const HeapWord* addr) const 676 { 677 assert(addr >= _region_start, "bad addr"); 678 assert(addr <= _region_end, "bad addr"); 679 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize; 680 } 681 682 inline size_t 683 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const 684 { 685 assert(addr >= _region_start, "bad addr " PTR_FORMAT " _region_start " PTR_FORMAT, p2i(addr), p2i(_region_start)); 686 assert(addr <= _region_end, "bad addr " PTR_FORMAT " _region_end " PTR_FORMAT, p2i(addr), p2i(_region_end)); 687 return pointer_delta(addr, _region_start) >> Log2RegionSize; 688 } 689 690 inline ParallelCompactData::RegionData* 691 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const 692 { 693 return region(addr_to_region_idx(addr)); 694 } 695 696 inline HeapWord* 697 ParallelCompactData::region_to_addr(size_t region) const 698 { 699 assert(region <= _region_count, "region out of range"); 700 return _region_start + (region << Log2RegionSize); 701 } 702 703 inline HeapWord* 704 ParallelCompactData::region_to_addr(const RegionData* region) const 705 { 706 return region_to_addr(pointer_delta(region, _region_data, 707 sizeof(RegionData))); 708 } 709 710 inline HeapWord* 711 ParallelCompactData::region_to_addr(size_t region, size_t offset) const 712 { 713 assert(region <= _region_count, "region out of range"); 714 assert(offset < RegionSize, "offset too big"); // This may be too strict. 715 return region_to_addr(region) + offset; 716 } 717 718 inline HeapWord* 719 ParallelCompactData::region_align_down(HeapWord* addr) const 720 { 721 assert(addr >= _region_start, "bad addr"); 722 assert(addr < _region_end + RegionSize, "bad addr"); 723 return (HeapWord*)(size_t(addr) & RegionAddrMask); 724 } 725 726 inline HeapWord* 727 ParallelCompactData::region_align_up(HeapWord* addr) const 728 { 729 assert(addr >= _region_start, "bad addr"); 730 assert(addr <= _region_end, "bad addr"); 731 return region_align_down(addr + RegionSizeOffsetMask); 732 } 733 734 inline bool 735 ParallelCompactData::is_region_aligned(HeapWord* addr) const 736 { 737 return region_offset(addr) == 0; 738 } 739 740 inline size_t 741 ParallelCompactData::block_offset(const HeapWord* addr) const 742 { 743 assert(addr >= _region_start, "bad addr"); 744 assert(addr <= _region_end, "bad addr"); 745 return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize; 746 } 747 748 inline size_t 749 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const 750 { 751 assert(addr >= _region_start, "bad addr"); 752 assert(addr <= _region_end, "bad addr"); 753 return pointer_delta(addr, _region_start) >> Log2BlockSize; 754 } 755 756 inline ParallelCompactData::BlockData* 757 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const 758 { 759 return block(addr_to_block_idx(addr)); 760 } 761 762 inline HeapWord* 763 ParallelCompactData::block_to_addr(size_t block) const 764 { 765 assert(block < _block_count, "block out of range"); 766 return _region_start + (block << Log2BlockSize); 767 } 768 769 inline size_t 770 ParallelCompactData::region_to_block_idx(size_t region) const 771 { 772 return region << Log2BlocksPerRegion; 773 } 774 775 inline HeapWord* 776 ParallelCompactData::block_align_down(HeapWord* addr) const 777 { 778 assert(addr >= _region_start, "bad addr"); 779 assert(addr < _region_end + RegionSize, "bad addr"); 780 return (HeapWord*)(size_t(addr) & BlockAddrMask); 781 } 782 783 inline HeapWord* 784 ParallelCompactData::block_align_up(HeapWord* addr) const 785 { 786 assert(addr >= _region_start, "bad addr"); 787 assert(addr <= _region_end, "bad addr"); 788 return block_align_down(addr + BlockSizeOffsetMask); 789 } 790 791 inline bool 792 ParallelCompactData::is_block_aligned(HeapWord* addr) const 793 { 794 return block_offset(addr) == 0; 795 } 796 797 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the 798 // do_addr() method. 799 // 800 // The closure is initialized with the number of heap words to process 801 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr() 802 // methods in subclasses should update the total as words are processed. Since 803 // only one subclass actually uses this mechanism to terminate iteration, the 804 // default initial value is > 0. The implementation is here and not in the 805 // single subclass that uses it to avoid making is_full() virtual, and thus 806 // adding a virtual call per live object. 807 808 class ParMarkBitMapClosure: public StackObj { 809 public: 810 typedef ParMarkBitMap::idx_t idx_t; 811 typedef ParMarkBitMap::IterationStatus IterationStatus; 812 813 public: 814 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm, 815 size_t words = max_uintx); 816 817 inline ParCompactionManager* compaction_manager() const; 818 inline ParMarkBitMap* bitmap() const; 819 inline size_t words_remaining() const; 820 inline bool is_full() const; 821 inline HeapWord* source() const; 822 823 inline void set_source(HeapWord* addr); 824 825 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0; 826 827 protected: 828 inline void decrement_words_remaining(size_t words); 829 830 private: 831 ParMarkBitMap* const _bitmap; 832 ParCompactionManager* const _compaction_manager; 833 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger. 834 size_t _words_remaining; // Words left to copy. 835 836 protected: 837 HeapWord* _source; // Next addr that would be read. 838 }; 839 840 inline 841 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap, 842 ParCompactionManager* cm, 843 size_t words): 844 _bitmap(bitmap), _compaction_manager(cm) 845 #ifdef ASSERT 846 , _initial_words_remaining(words) 847 #endif 848 { 849 _words_remaining = words; 850 _source = NULL; 851 } 852 853 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const { 854 return _compaction_manager; 855 } 856 857 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const { 858 return _bitmap; 859 } 860 861 inline size_t ParMarkBitMapClosure::words_remaining() const { 862 return _words_remaining; 863 } 864 865 inline bool ParMarkBitMapClosure::is_full() const { 866 return words_remaining() == 0; 867 } 868 869 inline HeapWord* ParMarkBitMapClosure::source() const { 870 return _source; 871 } 872 873 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) { 874 _source = addr; 875 } 876 877 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) { 878 assert(_words_remaining >= words, "processed too many words"); 879 _words_remaining -= words; 880 } 881 882 // The Parallel collector is a stop-the-world garbage collector that 883 // does parts of the collection using parallel threads. The collection includes 884 // the tenured generation and the young generation. 885 // 886 // There are four phases of the collection. 887 // 888 // - marking phase 889 // - summary phase 890 // - compacting phase 891 // - clean up phase 892 // 893 // Roughly speaking these phases correspond, respectively, to 894 // - mark all the live objects 895 // - calculate the destination of each object at the end of the collection 896 // - move the objects to their destination 897 // - update some references and reinitialize some variables 898 // 899 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The 900 // marking phase is implemented in PSParallelCompact::marking_phase() and does a 901 // complete marking of the heap. The summary phase is implemented in 902 // PSParallelCompact::summary_phase(). The move and update phase is implemented 903 // in PSParallelCompact::compact(). 904 // 905 // A space that is being collected is divided into regions and with each region 906 // is associated an object of type ParallelCompactData. Each region is of a 907 // fixed size and typically will contain more than 1 object and may have parts 908 // of objects at the front and back of the region. 909 // 910 // region -----+---------------------+---------- 911 // objects covered [ AAA )[ BBB )[ CCC )[ DDD ) 912 // 913 // The marking phase does a complete marking of all live objects in the heap. 914 // The marking also compiles the size of the data for all live objects covered 915 // by the region. This size includes the part of any live object spanning onto 916 // the region (part of AAA if it is live) from the front, all live objects 917 // contained in the region (BBB and/or CCC if they are live), and the part of 918 // any live objects covered by the region that extends off the region (part of 919 // DDD if it is live). The marking phase uses multiple GC threads and marking 920 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is 921 // done atomically as is the accumulation of the size of the live objects 922 // covered by a region. 923 // 924 // The summary phase calculates the total live data to the left of each region 925 // XXX. Based on that total and the bottom of the space, it can calculate the 926 // starting location of the live data in XXX. The summary phase calculates for 927 // each region XXX quantities such as 928 // 929 // - the amount of live data at the beginning of a region from an object 930 // entering the region. 931 // - the location of the first live data on the region 932 // - a count of the number of regions receiving live data from XXX. 933 // 934 // See ParallelCompactData for precise details. The summary phase also 935 // calculates the dense prefix for the compaction. The dense prefix is a 936 // portion at the beginning of the space that is not moved. The objects in the 937 // dense prefix do need to have their object references updated. See method 938 // summarize_dense_prefix(). 939 // 940 // The summary phase is done using 1 GC thread. 941 // 942 // The compaction phase moves objects to their new location and updates all 943 // references in the object. 944 // 945 // A current exception is that objects that cross a region boundary are moved 946 // but do not have their references updated. References are not updated because 947 // it cannot easily be determined if the klass pointer KKK for the object AAA 948 // has been updated. KKK likely resides in a region to the left of the region 949 // containing AAA. These AAA's have there references updated at the end in a 950 // clean up phase. See the method PSParallelCompact::update_deferred_objects(). 951 // An alternate strategy is being investigated for this deferral of updating. 952 // 953 // Compaction is done on a region basis. A region that is ready to be filled is 954 // put on a ready list and GC threads take region off the list and fill them. A 955 // region is ready to be filled if it empty of live objects. Such a region may 956 // have been initially empty (only contained dead objects) or may have had all 957 // its live objects copied out already. A region that compacts into itself is 958 // also ready for filling. The ready list is initially filled with empty 959 // regions and regions compacting into themselves. There is always at least 1 960 // region that can be put on the ready list. The regions are atomically added 961 // and removed from the ready list. 962 963 class TaskQueue; 964 965 class PSParallelCompact : AllStatic { 966 public: 967 // Convenient access to type names. 968 typedef ParMarkBitMap::idx_t idx_t; 969 typedef ParallelCompactData::RegionData RegionData; 970 typedef ParallelCompactData::BlockData BlockData; 971 972 typedef enum { 973 old_space_id, eden_space_id, 974 from_space_id, to_space_id, last_space_id 975 } SpaceId; 976 977 struct UpdateDensePrefixTask : public CHeapObj<mtGC> { 978 SpaceId _space_id; 979 size_t _region_index_start; 980 size_t _region_index_end; 981 982 UpdateDensePrefixTask() : 983 _space_id(SpaceId(0)), 984 _region_index_start(0), 985 _region_index_end(0) {} 986 987 UpdateDensePrefixTask(SpaceId space_id, 988 size_t region_index_start, 989 size_t region_index_end) : 990 _space_id(space_id), 991 _region_index_start(region_index_start), 992 _region_index_end(region_index_end) {} 993 }; 994 995 public: 996 // Inline closure decls 997 // 998 class IsAliveClosure: public BoolObjectClosure { 999 public: 1000 virtual bool do_object_b(oop p); 1001 }; 1002 1003 friend class RefProcTaskProxy; 1004 friend class PSParallelCompactTest; 1005 1006 private: 1007 static STWGCTimer _gc_timer; 1008 static ParallelOldTracer _gc_tracer; 1009 static elapsedTimer _accumulated_time; 1010 static unsigned int _total_invocations; 1011 static unsigned int _maximum_compaction_gc_num; 1012 static CollectorCounters* _counters; 1013 static ParMarkBitMap _mark_bitmap; 1014 static ParallelCompactData _summary_data; 1015 static IsAliveClosure _is_alive_closure; 1016 static SpaceInfo _space_info[last_space_id]; 1017 1018 // Reference processing (used in ...follow_contents) 1019 static SpanSubjectToDiscoveryClosure _span_based_discoverer; 1020 static ReferenceProcessor* _ref_processor; 1021 1022 // Values computed at initialization and used by dead_wood_limiter(). 1023 static double _dwl_mean; 1024 static double _dwl_std_dev; 1025 static double _dwl_first_term; 1026 static double _dwl_adjustment; 1027 #ifdef ASSERT 1028 static bool _dwl_initialized; 1029 #endif // #ifdef ASSERT 1030 1031 public: 1032 static ParallelOldTracer* gc_tracer() { return &_gc_tracer; } 1033 1034 private: 1035 1036 static void initialize_space_info(); 1037 1038 // Clear the marking bitmap and summary data that cover the specified space. 1039 static void clear_data_covering_space(SpaceId id); 1040 1041 static void pre_compact(); 1042 static void post_compact(); 1043 1044 // Mark live objects 1045 static void marking_phase(ParCompactionManager* cm, 1046 bool maximum_heap_compaction, 1047 ParallelOldTracer *gc_tracer); 1048 1049 // Compute the dense prefix for the designated space. This is an experimental 1050 // implementation currently not used in production. 1051 static HeapWord* compute_dense_prefix_via_density(const SpaceId id, 1052 bool maximum_compaction); 1053 1054 // Methods used to compute the dense prefix. 1055 1056 // Compute the value of the normal distribution at x = density. The mean and 1057 // standard deviation are values saved by initialize_dead_wood_limiter(). 1058 static inline double normal_distribution(double density); 1059 1060 // Initialize the static vars used by dead_wood_limiter(). 1061 static void initialize_dead_wood_limiter(); 1062 1063 // Return the percentage of space that can be treated as "dead wood" (i.e., 1064 // not reclaimed). 1065 static double dead_wood_limiter(double density, size_t min_percent); 1066 1067 // Find the first (left-most) region in the range [beg, end) that has at least 1068 // dead_words of dead space to the left. The argument beg must be the first 1069 // region in the space that is not completely live. 1070 static RegionData* dead_wood_limit_region(const RegionData* beg, 1071 const RegionData* end, 1072 size_t dead_words); 1073 1074 // Return a pointer to the first region in the range [beg, end) that is not 1075 // completely full. 1076 static RegionData* first_dead_space_region(const RegionData* beg, 1077 const RegionData* end); 1078 1079 // Return a value indicating the benefit or 'yield' if the compacted region 1080 // were to start (or equivalently if the dense prefix were to end) at the 1081 // candidate region. Higher values are better. 1082 // 1083 // The value is based on the amount of space reclaimed vs. the costs of (a) 1084 // updating references in the dense prefix plus (b) copying objects and 1085 // updating references in the compacted region. 1086 static inline double reclaimed_ratio(const RegionData* const candidate, 1087 HeapWord* const bottom, 1088 HeapWord* const top, 1089 HeapWord* const new_top); 1090 1091 // Compute the dense prefix for the designated space. 1092 static HeapWord* compute_dense_prefix(const SpaceId id, 1093 bool maximum_compaction); 1094 1095 // Return true if dead space crosses onto the specified Region; bit must be 1096 // the bit index corresponding to the first word of the Region. 1097 static inline bool dead_space_crosses_boundary(const RegionData* region, 1098 idx_t bit); 1099 1100 // Summary phase utility routine to fill dead space (if any) at the dense 1101 // prefix boundary. Should only be called if the the dense prefix is 1102 // non-empty. 1103 static void fill_dense_prefix_end(SpaceId id); 1104 1105 static void summarize_spaces_quick(); 1106 static void summarize_space(SpaceId id, bool maximum_compaction); 1107 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction); 1108 1109 // Adjust addresses in roots. Does not adjust addresses in heap. 1110 static void adjust_roots(ParCompactionManager* cm); 1111 1112 DEBUG_ONLY(static void write_block_fill_histogram();) 1113 1114 // Move objects to new locations. 1115 static void compact_perm(ParCompactionManager* cm); 1116 static void compact(); 1117 1118 // Add available regions to the stack and draining tasks to the task queue. 1119 static void prepare_region_draining_tasks(uint parallel_gc_threads); 1120 1121 // Add dense prefix update tasks to the task queue. 1122 static void enqueue_dense_prefix_tasks(TaskQueue& task_queue, 1123 uint parallel_gc_threads); 1124 1125 #ifndef PRODUCT 1126 // Print generic summary data 1127 static void print_generic_summary_data(ParallelCompactData& summary_data, 1128 HeapWord* const beg_addr, 1129 HeapWord* const end_addr); 1130 #endif // #ifndef PRODUCT 1131 1132 public: 1133 1134 PSParallelCompact(); 1135 1136 static void invoke(bool maximum_heap_compaction); 1137 static bool invoke_no_policy(bool maximum_heap_compaction); 1138 1139 static void post_initialize(); 1140 // Perform initialization for PSParallelCompact that requires 1141 // allocations. This should be called during the VM initialization 1142 // at a pointer where it would be appropriate to return a JNI_ENOMEM 1143 // in the event of a failure. 1144 static bool initialize(); 1145 1146 // Closure accessors 1147 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; } 1148 1149 // Public accessors 1150 static elapsedTimer* accumulated_time() { return &_accumulated_time; } 1151 static unsigned int total_invocations() { return _total_invocations; } 1152 static CollectorCounters* counters() { return _counters; } 1153 1154 // Marking support 1155 static inline bool mark_obj(oop obj); 1156 static inline bool is_marked(oop obj); 1157 1158 template <class T> static inline void adjust_pointer(T* p, ParCompactionManager* cm); 1159 1160 // Compaction support. 1161 // Return true if p is in the range [beg_addr, end_addr). 1162 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr); 1163 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr); 1164 1165 // Convenience wrappers for per-space data kept in _space_info. 1166 static inline MutableSpace* space(SpaceId space_id); 1167 static inline HeapWord* new_top(SpaceId space_id); 1168 static inline HeapWord* dense_prefix(SpaceId space_id); 1169 static inline ObjectStartArray* start_array(SpaceId space_id); 1170 1171 // Process the end of the given region range in the dense prefix. 1172 // This includes saving any object not updated. 1173 static void dense_prefix_regions_epilogue(ParCompactionManager* cm, 1174 size_t region_start_index, 1175 size_t region_end_index, 1176 idx_t exiting_object_offset, 1177 idx_t region_offset_start, 1178 idx_t region_offset_end); 1179 1180 // Update a region in the dense prefix. For each live object 1181 // in the region, update it's interior references. For each 1182 // dead object, fill it with deadwood. Dead space at the end 1183 // of a region range will be filled to the start of the next 1184 // live object regardless of the region_index_end. None of the 1185 // objects in the dense prefix move and dead space is dead 1186 // (holds only dead objects that don't need any processing), so 1187 // dead space can be filled in any order. 1188 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm, 1189 SpaceId space_id, 1190 size_t region_index_start, 1191 size_t region_index_end); 1192 1193 // Return the address of the count + 1st live word in the range [beg, end). 1194 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count); 1195 1196 // Return the address of the word to be copied to dest_addr, which must be 1197 // aligned to a region boundary. 1198 static HeapWord* first_src_addr(HeapWord* const dest_addr, 1199 SpaceId src_space_id, 1200 size_t src_region_idx); 1201 1202 // Determine the next source region, set closure.source() to the start of the 1203 // new region return the region index. Parameter end_addr is the address one 1204 // beyond the end of source range just processed. If necessary, switch to a 1205 // new source space and set src_space_id (in-out parameter) and src_space_top 1206 // (out parameter) accordingly. 1207 static size_t next_src_region(MoveAndUpdateClosure& closure, 1208 SpaceId& src_space_id, 1209 HeapWord*& src_space_top, 1210 HeapWord* end_addr); 1211 1212 // Decrement the destination count for each non-empty source region in the 1213 // range [beg_region, region(region_align_up(end_addr))). If the destination 1214 // count for a region goes to 0 and it needs to be filled, enqueue it. 1215 static void decrement_destination_counts(ParCompactionManager* cm, 1216 SpaceId src_space_id, 1217 size_t beg_region, 1218 HeapWord* end_addr); 1219 1220 static void fill_region(ParCompactionManager* cm, MoveAndUpdateClosure& closure, size_t region); 1221 static void fill_and_update_region(ParCompactionManager* cm, size_t region); 1222 1223 static bool steal_unavailable_region(ParCompactionManager* cm, size_t& region_idx); 1224 static void fill_and_update_shadow_region(ParCompactionManager* cm, size_t region); 1225 // Copy the content of a shadow region back to its corresponding heap region 1226 static void copy_back(HeapWord* shadow_addr, HeapWord* region_addr); 1227 // Collect empty regions as shadow regions and initialize the 1228 // _next_shadow_region filed for each compact manager 1229 static void initialize_shadow_regions(uint parallel_gc_threads); 1230 1231 // Fill in the block table for the specified region. 1232 static void fill_blocks(size_t region_idx); 1233 1234 // Update the deferred objects in the space. 1235 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id); 1236 1237 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; } 1238 static ParallelCompactData& summary_data() { return _summary_data; } 1239 1240 // Reference Processing 1241 static ReferenceProcessor* const ref_processor() { return _ref_processor; } 1242 1243 static STWGCTimer* gc_timer() { return &_gc_timer; } 1244 1245 // Return the SpaceId for the given address. 1246 static SpaceId space_id(HeapWord* addr); 1247 1248 static void print_on_error(outputStream* st); 1249 1250 #ifndef PRODUCT 1251 // Debugging support. 1252 static const char* space_names[last_space_id]; 1253 static void print_region_ranges(); 1254 static void print_dense_prefix_stats(const char* const algorithm, 1255 const SpaceId id, 1256 const bool maximum_compaction, 1257 HeapWord* const addr); 1258 static void summary_phase_msg(SpaceId dst_space_id, 1259 HeapWord* dst_beg, HeapWord* dst_end, 1260 SpaceId src_space_id, 1261 HeapWord* src_beg, HeapWord* src_end); 1262 #endif // #ifndef PRODUCT 1263 1264 #ifdef ASSERT 1265 // Sanity check the new location of a word in the heap. 1266 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr); 1267 // Verify that all the regions have been emptied. 1268 static void verify_complete(SpaceId space_id); 1269 #endif // #ifdef ASSERT 1270 }; 1271 1272 class MoveAndUpdateClosure: public ParMarkBitMapClosure { 1273 static inline size_t calculate_words_remaining(size_t region); 1274 public: 1275 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, 1276 size_t region); 1277 1278 // Accessors. 1279 HeapWord* destination() const { return _destination; } 1280 HeapWord* copy_destination() const { return _destination + _offset; } 1281 1282 // If the object will fit (size <= words_remaining()), copy it to the current 1283 // destination, update the interior oops and the start array and return either 1284 // full (if the closure is full) or incomplete. If the object will not fit, 1285 // return would_overflow. 1286 IterationStatus do_addr(HeapWord* addr, size_t size); 1287 1288 // Copy enough words to fill this closure, starting at source(). Interior 1289 // oops and the start array are not updated. Return full. 1290 IterationStatus copy_until_full(); 1291 1292 // Copy enough words to fill this closure or to the end of an object, 1293 // whichever is smaller, starting at source(). Interior oops and the start 1294 // array are not updated. 1295 void copy_partial_obj(); 1296 1297 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr, 1298 PSParallelCompact::RegionData* region_ptr); 1299 1300 protected: 1301 // Update variables to indicate that word_count words were processed. 1302 inline void update_state(size_t word_count); 1303 1304 protected: 1305 HeapWord* _destination; // Next addr to be written. 1306 ObjectStartArray* const _start_array; 1307 size_t _offset; 1308 }; 1309 1310 inline size_t MoveAndUpdateClosure::calculate_words_remaining(size_t region) { 1311 HeapWord* dest_addr = PSParallelCompact::summary_data().region_to_addr(region); 1312 PSParallelCompact::SpaceId dest_space_id = PSParallelCompact::space_id(dest_addr); 1313 HeapWord* new_top = PSParallelCompact::new_top(dest_space_id); 1314 assert(dest_addr < new_top, "sanity"); 1315 1316 return MIN2(pointer_delta(new_top, dest_addr), ParallelCompactData::RegionSize); 1317 } 1318 1319 inline 1320 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap, 1321 ParCompactionManager* cm, 1322 size_t region_idx) : 1323 ParMarkBitMapClosure(bitmap, cm, calculate_words_remaining(region_idx)), 1324 _destination(PSParallelCompact::summary_data().region_to_addr(region_idx)), 1325 _start_array(PSParallelCompact::start_array(PSParallelCompact::space_id(_destination))), 1326 _offset(0) { } 1327 1328 1329 inline void MoveAndUpdateClosure::update_state(size_t words) 1330 { 1331 decrement_words_remaining(words); 1332 _source += words; 1333 _destination += words; 1334 } 1335 1336 class MoveAndUpdateShadowClosure: public MoveAndUpdateClosure { 1337 inline size_t calculate_shadow_offset(size_t region_idx, size_t shadow_idx); 1338 public: 1339 inline MoveAndUpdateShadowClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, 1340 size_t region, size_t shadow); 1341 1342 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr, 1343 PSParallelCompact::RegionData* region_ptr); 1344 1345 private: 1346 size_t _shadow; 1347 }; 1348 1349 inline size_t MoveAndUpdateShadowClosure::calculate_shadow_offset(size_t region_idx, size_t shadow_idx) { 1350 ParallelCompactData& sd = PSParallelCompact::summary_data(); 1351 HeapWord* dest_addr = sd.region_to_addr(region_idx); 1352 HeapWord* shadow_addr = sd.region_to_addr(shadow_idx); 1353 return pointer_delta(shadow_addr, dest_addr); 1354 } 1355 1356 inline 1357 MoveAndUpdateShadowClosure::MoveAndUpdateShadowClosure(ParMarkBitMap *bitmap, 1358 ParCompactionManager *cm, 1359 size_t region, 1360 size_t shadow) : 1361 MoveAndUpdateClosure(bitmap, cm, region), 1362 _shadow(shadow) { 1363 _offset = calculate_shadow_offset(region, shadow); 1364 } 1365 1366 class UpdateOnlyClosure: public ParMarkBitMapClosure { 1367 private: 1368 const PSParallelCompact::SpaceId _space_id; 1369 ObjectStartArray* const _start_array; 1370 1371 public: 1372 UpdateOnlyClosure(ParMarkBitMap* mbm, 1373 ParCompactionManager* cm, 1374 PSParallelCompact::SpaceId space_id); 1375 1376 // Update the object. 1377 virtual IterationStatus do_addr(HeapWord* addr, size_t words); 1378 1379 inline void do_addr(HeapWord* addr); 1380 }; 1381 1382 class FillClosure: public ParMarkBitMapClosure { 1383 public: 1384 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id); 1385 1386 virtual IterationStatus do_addr(HeapWord* addr, size_t size); 1387 1388 private: 1389 ObjectStartArray* const _start_array; 1390 }; 1391 1392 #endif // SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP