1 /* 2 * Copyright (c) 2005, 2019, 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 36 class ParallelScavengeHeap; 37 class PSAdaptiveSizePolicy; 38 class PSYoungGen; 39 class PSOldGen; 40 class ParCompactionManager; 41 class ParallelTaskTerminator; 42 class PSParallelCompact; 43 class PreGCValues; 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 // UNUSED -> try_push() -> FINISHED 336 // Steal Path: 337 // UNUSED -> try_steal() -> SHADOW -> mark_filled() -> FILLED -> try_copy() -> FINISHED 338 static const int UNUSED; // Original state 339 static const int SHADOW; // Stolen by an idle thread, and a shadow region is created for it 340 static const int FILLED; // Its shadow region has been filled and ready to be copied back 341 static const int FINISH; // Work has been done 342 343 // Preempt the region to avoid double processes 344 inline bool try_push(); 345 inline bool try_steal(); 346 // Mark the region as filled and ready to be copied back 347 inline void mark_filled(); 348 // Preempt the region to copy the shadow region content back 349 inline bool try_copy(); 350 // Special case: see the comment in PSParallelCompact::fill_shadow_region. 351 // Return to the normal path here 352 inline void mark_normal(); 353 354 355 int shadow_state() { return _shadow_state; } 356 357 private: 358 // The type used to represent object sizes within a region. 359 typedef uint region_sz_t; 360 361 // Constants for manipulating the _dc_and_los field, which holds both the 362 // destination count and live obj size. The live obj size lives at the 363 // least significant end so no masking is necessary when adding. 364 static const region_sz_t dc_shift; // Shift amount. 365 static const region_sz_t dc_mask; // Mask for destination count. 366 static const region_sz_t dc_one; // 1, shifted appropriately. 367 static const region_sz_t dc_claimed; // Region has been claimed. 368 static const region_sz_t dc_completed; // Region has been completed. 369 static const region_sz_t los_mask; // Mask for live obj size. 370 371 HeapWord* _destination; 372 size_t _source_region; 373 HeapWord* _partial_obj_addr; 374 region_sz_t _partial_obj_size; 375 region_sz_t volatile _dc_and_los; 376 bool volatile _blocks_filled; 377 int volatile _shadow_state; 378 379 #ifdef ASSERT 380 size_t _blocks_filled_count; // Number of block table fills. 381 382 // These enable optimizations that are only partially implemented. Use 383 // debug builds to prevent the code fragments from breaking. 384 HeapWord* _data_location; 385 HeapWord* _highest_ref; 386 #endif // #ifdef ASSERT 387 388 #ifdef ASSERT 389 public: 390 uint _pushed; // 0 until region is pushed onto a stack 391 private: 392 #endif 393 }; 394 395 // "Blocks" allow shorter sections of the bitmap to be searched. Each Block 396 // holds an offset, which is the amount of live data in the Region to the left 397 // of the first live object that starts in the Block. 398 class BlockData 399 { 400 public: 401 typedef unsigned short int blk_ofs_t; 402 403 blk_ofs_t offset() const { return _offset; } 404 void set_offset(size_t val) { _offset = (blk_ofs_t)val; } 405 406 private: 407 blk_ofs_t _offset; 408 }; 409 410 public: 411 ParallelCompactData(); 412 bool initialize(MemRegion covered_region); 413 414 size_t region_count() const { return _region_count; } 415 size_t reserved_byte_size() const { return _reserved_byte_size; } 416 417 // Convert region indices to/from RegionData pointers. 418 inline RegionData* region(size_t region_idx) const; 419 inline size_t region(const RegionData* const region_ptr) const; 420 421 size_t block_count() const { return _block_count; } 422 inline BlockData* block(size_t block_idx) const; 423 inline size_t block(const BlockData* block_ptr) const; 424 425 void add_obj(HeapWord* addr, size_t len); 426 void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); } 427 428 // Fill in the regions covering [beg, end) so that no data moves; i.e., the 429 // destination of region n is simply the start of region n. The argument beg 430 // must be region-aligned; end need not be. 431 void summarize_dense_prefix(HeapWord* beg, HeapWord* end); 432 433 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info, 434 HeapWord* destination, HeapWord* target_end, 435 HeapWord** target_next); 436 bool summarize(SplitInfo& split_info, 437 HeapWord* source_beg, HeapWord* source_end, 438 HeapWord** source_next, 439 HeapWord* target_beg, HeapWord* target_end, 440 HeapWord** target_next); 441 442 void clear(); 443 void clear_range(size_t beg_region, size_t end_region); 444 void clear_range(HeapWord* beg, HeapWord* end) { 445 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end)); 446 } 447 448 // Return the number of words between addr and the start of the region 449 // containing addr. 450 inline size_t region_offset(const HeapWord* addr) const; 451 452 // Convert addresses to/from a region index or region pointer. 453 inline size_t addr_to_region_idx(const HeapWord* addr) const; 454 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const; 455 inline HeapWord* region_to_addr(size_t region) const; 456 inline HeapWord* region_to_addr(size_t region, size_t offset) const; 457 inline HeapWord* region_to_addr(const RegionData* region) const; 458 459 inline HeapWord* region_align_down(HeapWord* addr) const; 460 inline HeapWord* region_align_up(HeapWord* addr) const; 461 inline bool is_region_aligned(HeapWord* addr) const; 462 463 // Analogous to region_offset() for blocks. 464 size_t block_offset(const HeapWord* addr) const; 465 size_t addr_to_block_idx(const HeapWord* addr) const; 466 size_t addr_to_block_idx(const oop obj) const { 467 return addr_to_block_idx((HeapWord*) obj); 468 } 469 inline BlockData* addr_to_block_ptr(const HeapWord* addr) const; 470 inline HeapWord* block_to_addr(size_t block) const; 471 inline size_t region_to_block_idx(size_t region) const; 472 473 inline HeapWord* block_align_down(HeapWord* addr) const; 474 inline HeapWord* block_align_up(HeapWord* addr) const; 475 inline bool is_block_aligned(HeapWord* addr) const; 476 477 // Return the address one past the end of the partial object. 478 HeapWord* partial_obj_end(size_t region_idx) const; 479 480 // Return the location of the object after compaction. 481 HeapWord* calc_new_pointer(HeapWord* addr, ParCompactionManager* cm); 482 483 HeapWord* calc_new_pointer(oop p, ParCompactionManager* cm) { 484 return calc_new_pointer((HeapWord*) p, cm); 485 } 486 487 #ifdef ASSERT 488 void verify_clear(const PSVirtualSpace* vspace); 489 void verify_clear(); 490 #endif // #ifdef ASSERT 491 492 private: 493 bool initialize_block_data(); 494 bool initialize_region_data(size_t region_size); 495 PSVirtualSpace* create_vspace(size_t count, size_t element_size); 496 497 private: 498 HeapWord* _region_start; 499 #ifdef ASSERT 500 HeapWord* _region_end; 501 #endif // #ifdef ASSERT 502 503 PSVirtualSpace* _region_vspace; 504 size_t _reserved_byte_size; 505 RegionData* _region_data; 506 size_t _region_count; 507 508 PSVirtualSpace* _block_vspace; 509 BlockData* _block_data; 510 size_t _block_count; 511 }; 512 513 inline uint 514 ParallelCompactData::RegionData::destination_count_raw() const 515 { 516 return _dc_and_los & dc_mask; 517 } 518 519 inline uint 520 ParallelCompactData::RegionData::destination_count() const 521 { 522 return destination_count_raw() >> dc_shift; 523 } 524 525 inline bool 526 ParallelCompactData::RegionData::blocks_filled() const 527 { 528 bool result = _blocks_filled; 529 OrderAccess::acquire(); 530 return result; 531 } 532 533 #ifdef ASSERT 534 inline size_t 535 ParallelCompactData::RegionData::blocks_filled_count() const 536 { 537 return _blocks_filled_count; 538 } 539 #endif // #ifdef ASSERT 540 541 inline void 542 ParallelCompactData::RegionData::set_blocks_filled() 543 { 544 OrderAccess::release(); 545 _blocks_filled = true; 546 // Debug builds count the number of times the table was filled. 547 DEBUG_ONLY(Atomic::inc(&_blocks_filled_count)); 548 } 549 550 inline void 551 ParallelCompactData::RegionData::set_destination_count(uint count) 552 { 553 assert(count <= (dc_completed >> dc_shift), "count too large"); 554 const region_sz_t live_sz = (region_sz_t) live_obj_size(); 555 _dc_and_los = (count << dc_shift) | live_sz; 556 } 557 558 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words) 559 { 560 assert(words <= los_mask, "would overflow"); 561 _dc_and_los = destination_count_raw() | (region_sz_t)words; 562 } 563 564 inline void ParallelCompactData::RegionData::decrement_destination_count() 565 { 566 assert(_dc_and_los < dc_claimed, "already claimed"); 567 assert(_dc_and_los >= dc_one, "count would go negative"); 568 Atomic::add(dc_mask, &_dc_and_los); 569 } 570 571 inline HeapWord* ParallelCompactData::RegionData::data_location() const 572 { 573 DEBUG_ONLY(return _data_location;) 574 NOT_DEBUG(return NULL;) 575 } 576 577 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const 578 { 579 DEBUG_ONLY(return _highest_ref;) 580 NOT_DEBUG(return NULL;) 581 } 582 583 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr) 584 { 585 DEBUG_ONLY(_data_location = addr;) 586 } 587 588 inline void ParallelCompactData::RegionData::set_completed() 589 { 590 assert(claimed(), "must be claimed first"); 591 _dc_and_los = dc_completed | (region_sz_t) live_obj_size(); 592 } 593 594 // MT-unsafe claiming of a region. Should only be used during single threaded 595 // execution. 596 inline bool ParallelCompactData::RegionData::claim_unsafe() 597 { 598 if (available()) { 599 _dc_and_los |= dc_claimed; 600 return true; 601 } 602 return false; 603 } 604 605 inline void ParallelCompactData::RegionData::add_live_obj(size_t words) 606 { 607 assert(words <= (size_t)los_mask - live_obj_size(), "overflow"); 608 Atomic::add(static_cast<region_sz_t>(words), &_dc_and_los); 609 } 610 611 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr) 612 { 613 #ifdef ASSERT 614 HeapWord* tmp = _highest_ref; 615 while (addr > tmp) { 616 tmp = Atomic::cmpxchg(addr, &_highest_ref, tmp); 617 } 618 #endif // #ifdef ASSERT 619 } 620 621 inline bool ParallelCompactData::RegionData::claim() 622 { 623 const region_sz_t los = static_cast<region_sz_t>(live_obj_size()); 624 const region_sz_t old = Atomic::cmpxchg(dc_claimed | los, &_dc_and_los, los); 625 return old == los; 626 } 627 628 inline bool ParallelCompactData::RegionData::try_push() { 629 return Atomic::cmpxchg(FINISH, &_shadow_state, UNUSED) == UNUSED; 630 } 631 632 inline bool ParallelCompactData::RegionData::try_steal() { 633 return Atomic::cmpxchg(SHADOW, &_shadow_state, UNUSED) == UNUSED; 634 } 635 636 inline void ParallelCompactData::RegionData::mark_filled() { 637 int old = Atomic::cmpxchg(FILLED, &_shadow_state, SHADOW); 638 assert(old == SHADOW, "Fail to mark the region as filled"); 639 } 640 641 inline bool ParallelCompactData::RegionData::try_copy() { 642 return Atomic::cmpxchg(FINISH, &_shadow_state, FILLED) == FILLED; 643 } 644 645 void ParallelCompactData::RegionData::mark_normal() { 646 _shadow_state = FINISH; 647 } 648 649 inline ParallelCompactData::RegionData* 650 ParallelCompactData::region(size_t region_idx) const 651 { 652 assert(region_idx <= region_count(), "bad arg"); 653 return _region_data + region_idx; 654 } 655 656 inline size_t 657 ParallelCompactData::region(const RegionData* const region_ptr) const 658 { 659 assert(region_ptr >= _region_data, "bad arg"); 660 assert(region_ptr <= _region_data + region_count(), "bad arg"); 661 return pointer_delta(region_ptr, _region_data, sizeof(RegionData)); 662 } 663 664 inline ParallelCompactData::BlockData* 665 ParallelCompactData::block(size_t n) const { 666 assert(n < block_count(), "bad arg"); 667 return _block_data + n; 668 } 669 670 inline size_t 671 ParallelCompactData::region_offset(const HeapWord* addr) const 672 { 673 assert(addr >= _region_start, "bad addr"); 674 assert(addr <= _region_end, "bad addr"); 675 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize; 676 } 677 678 inline size_t 679 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const 680 { 681 assert(addr >= _region_start, "bad addr " PTR_FORMAT " _region_start " PTR_FORMAT, p2i(addr), p2i(_region_start)); 682 assert(addr <= _region_end, "bad addr " PTR_FORMAT " _region_end " PTR_FORMAT, p2i(addr), p2i(_region_end)); 683 return pointer_delta(addr, _region_start) >> Log2RegionSize; 684 } 685 686 inline ParallelCompactData::RegionData* 687 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const 688 { 689 return region(addr_to_region_idx(addr)); 690 } 691 692 inline HeapWord* 693 ParallelCompactData::region_to_addr(size_t region) const 694 { 695 assert(region <= _region_count, "region out of range"); 696 return _region_start + (region << Log2RegionSize); 697 } 698 699 inline HeapWord* 700 ParallelCompactData::region_to_addr(const RegionData* region) const 701 { 702 return region_to_addr(pointer_delta(region, _region_data, 703 sizeof(RegionData))); 704 } 705 706 inline HeapWord* 707 ParallelCompactData::region_to_addr(size_t region, size_t offset) const 708 { 709 assert(region <= _region_count, "region out of range"); 710 assert(offset < RegionSize, "offset too big"); // This may be too strict. 711 return region_to_addr(region) + offset; 712 } 713 714 inline HeapWord* 715 ParallelCompactData::region_align_down(HeapWord* addr) const 716 { 717 assert(addr >= _region_start, "bad addr"); 718 assert(addr < _region_end + RegionSize, "bad addr"); 719 return (HeapWord*)(size_t(addr) & RegionAddrMask); 720 } 721 722 inline HeapWord* 723 ParallelCompactData::region_align_up(HeapWord* addr) const 724 { 725 assert(addr >= _region_start, "bad addr"); 726 assert(addr <= _region_end, "bad addr"); 727 return region_align_down(addr + RegionSizeOffsetMask); 728 } 729 730 inline bool 731 ParallelCompactData::is_region_aligned(HeapWord* addr) const 732 { 733 return region_offset(addr) == 0; 734 } 735 736 inline size_t 737 ParallelCompactData::block_offset(const HeapWord* addr) const 738 { 739 assert(addr >= _region_start, "bad addr"); 740 assert(addr <= _region_end, "bad addr"); 741 return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize; 742 } 743 744 inline size_t 745 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const 746 { 747 assert(addr >= _region_start, "bad addr"); 748 assert(addr <= _region_end, "bad addr"); 749 return pointer_delta(addr, _region_start) >> Log2BlockSize; 750 } 751 752 inline ParallelCompactData::BlockData* 753 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const 754 { 755 return block(addr_to_block_idx(addr)); 756 } 757 758 inline HeapWord* 759 ParallelCompactData::block_to_addr(size_t block) const 760 { 761 assert(block < _block_count, "block out of range"); 762 return _region_start + (block << Log2BlockSize); 763 } 764 765 inline size_t 766 ParallelCompactData::region_to_block_idx(size_t region) const 767 { 768 return region << Log2BlocksPerRegion; 769 } 770 771 inline HeapWord* 772 ParallelCompactData::block_align_down(HeapWord* addr) const 773 { 774 assert(addr >= _region_start, "bad addr"); 775 assert(addr < _region_end + RegionSize, "bad addr"); 776 return (HeapWord*)(size_t(addr) & BlockAddrMask); 777 } 778 779 inline HeapWord* 780 ParallelCompactData::block_align_up(HeapWord* addr) const 781 { 782 assert(addr >= _region_start, "bad addr"); 783 assert(addr <= _region_end, "bad addr"); 784 return block_align_down(addr + BlockSizeOffsetMask); 785 } 786 787 inline bool 788 ParallelCompactData::is_block_aligned(HeapWord* addr) const 789 { 790 return block_offset(addr) == 0; 791 } 792 793 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the 794 // do_addr() method. 795 // 796 // The closure is initialized with the number of heap words to process 797 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr() 798 // methods in subclasses should update the total as words are processed. Since 799 // only one subclass actually uses this mechanism to terminate iteration, the 800 // default initial value is > 0. The implementation is here and not in the 801 // single subclass that uses it to avoid making is_full() virtual, and thus 802 // adding a virtual call per live object. 803 804 class ParMarkBitMapClosure: public StackObj { 805 public: 806 typedef ParMarkBitMap::idx_t idx_t; 807 typedef ParMarkBitMap::IterationStatus IterationStatus; 808 809 public: 810 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm, 811 size_t words = max_uintx); 812 813 inline ParCompactionManager* compaction_manager() const; 814 inline ParMarkBitMap* bitmap() const; 815 inline size_t words_remaining() const; 816 inline bool is_full() const; 817 inline HeapWord* source() const; 818 819 inline void set_source(HeapWord* addr); 820 821 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0; 822 823 protected: 824 inline void decrement_words_remaining(size_t words); 825 826 private: 827 ParMarkBitMap* const _bitmap; 828 ParCompactionManager* const _compaction_manager; 829 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger. 830 size_t _words_remaining; // Words left to copy. 831 832 protected: 833 HeapWord* _source; // Next addr that would be read. 834 }; 835 836 inline 837 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap, 838 ParCompactionManager* cm, 839 size_t words): 840 _bitmap(bitmap), _compaction_manager(cm) 841 #ifdef ASSERT 842 , _initial_words_remaining(words) 843 #endif 844 { 845 _words_remaining = words; 846 _source = NULL; 847 } 848 849 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const { 850 return _compaction_manager; 851 } 852 853 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const { 854 return _bitmap; 855 } 856 857 inline size_t ParMarkBitMapClosure::words_remaining() const { 858 return _words_remaining; 859 } 860 861 inline bool ParMarkBitMapClosure::is_full() const { 862 return words_remaining() == 0; 863 } 864 865 inline HeapWord* ParMarkBitMapClosure::source() const { 866 return _source; 867 } 868 869 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) { 870 _source = addr; 871 } 872 873 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) { 874 assert(_words_remaining >= words, "processed too many words"); 875 _words_remaining -= words; 876 } 877 878 // The UseParallelOldGC collector is a stop-the-world garbage collector that 879 // does parts of the collection using parallel threads. The collection includes 880 // the tenured generation and the young generation. The permanent generation is 881 // collected at the same time as the other two generations but the permanent 882 // generation is collect by a single GC thread. The permanent generation is 883 // collected serially because of the requirement that during the processing of a 884 // klass AAA, any objects reference by AAA must already have been processed. 885 // This requirement is enforced by a left (lower address) to right (higher 886 // address) sliding compaction. 887 // 888 // There are four phases of the collection. 889 // 890 // - marking phase 891 // - summary phase 892 // - compacting phase 893 // - clean up phase 894 // 895 // Roughly speaking these phases correspond, respectively, to 896 // - mark all the live objects 897 // - calculate the destination of each object at the end of the collection 898 // - move the objects to their destination 899 // - update some references and reinitialize some variables 900 // 901 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The 902 // marking phase is implemented in PSParallelCompact::marking_phase() and does a 903 // complete marking of the heap. The summary phase is implemented in 904 // PSParallelCompact::summary_phase(). The move and update phase is implemented 905 // in PSParallelCompact::compact(). 906 // 907 // A space that is being collected is divided into regions and with each region 908 // is associated an object of type ParallelCompactData. Each region is of a 909 // fixed size and typically will contain more than 1 object and may have parts 910 // of objects at the front and back of the region. 911 // 912 // region -----+---------------------+---------- 913 // objects covered [ AAA )[ BBB )[ CCC )[ DDD ) 914 // 915 // The marking phase does a complete marking of all live objects in the heap. 916 // The marking also compiles the size of the data for all live objects covered 917 // by the region. This size includes the part of any live object spanning onto 918 // the region (part of AAA if it is live) from the front, all live objects 919 // contained in the region (BBB and/or CCC if they are live), and the part of 920 // any live objects covered by the region that extends off the region (part of 921 // DDD if it is live). The marking phase uses multiple GC threads and marking 922 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is 923 // done atomically as is the accumulation of the size of the live objects 924 // covered by a region. 925 // 926 // The summary phase calculates the total live data to the left of each region 927 // XXX. Based on that total and the bottom of the space, it can calculate the 928 // starting location of the live data in XXX. The summary phase calculates for 929 // each region XXX quantities such as 930 // 931 // - the amount of live data at the beginning of a region from an object 932 // entering the region. 933 // - the location of the first live data on the region 934 // - a count of the number of regions receiving live data from XXX. 935 // 936 // See ParallelCompactData for precise details. The summary phase also 937 // calculates the dense prefix for the compaction. The dense prefix is a 938 // portion at the beginning of the space that is not moved. The objects in the 939 // dense prefix do need to have their object references updated. See method 940 // summarize_dense_prefix(). 941 // 942 // The summary phase is done using 1 GC thread. 943 // 944 // The compaction phase moves objects to their new location and updates all 945 // references in the object. 946 // 947 // A current exception is that objects that cross a region boundary are moved 948 // but do not have their references updated. References are not updated because 949 // it cannot easily be determined if the klass pointer KKK for the object AAA 950 // has been updated. KKK likely resides in a region to the left of the region 951 // containing AAA. These AAA's have there references updated at the end in a 952 // clean up phase. See the method PSParallelCompact::update_deferred_objects(). 953 // An alternate strategy is being investigated for this deferral of updating. 954 // 955 // Compaction is done on a region basis. A region that is ready to be filled is 956 // put on a ready list and GC threads take region off the list and fill them. A 957 // region is ready to be filled if it empty of live objects. Such a region may 958 // have been initially empty (only contained dead objects) or may have had all 959 // its live objects copied out already. A region that compacts into itself is 960 // also ready for filling. The ready list is initially filled with empty 961 // regions and regions compacting into themselves. There is always at least 1 962 // region that can be put on the ready list. The regions are atomically added 963 // and removed from the ready list. 964 965 class TaskQueue; 966 967 class PSParallelCompact : AllStatic { 968 public: 969 // Convenient access to type names. 970 typedef ParMarkBitMap::idx_t idx_t; 971 typedef ParallelCompactData::RegionData RegionData; 972 typedef ParallelCompactData::BlockData BlockData; 973 974 typedef enum { 975 old_space_id, eden_space_id, 976 from_space_id, to_space_id, last_space_id 977 } SpaceId; 978 979 struct UpdateDensePrefixTask : public CHeapObj<mtGC> { 980 SpaceId _space_id; 981 size_t _region_index_start; 982 size_t _region_index_end; 983 984 UpdateDensePrefixTask() : 985 _space_id(SpaceId(0)), 986 _region_index_start(0), 987 _region_index_end(0) {} 988 989 UpdateDensePrefixTask(SpaceId space_id, 990 size_t region_index_start, 991 size_t region_index_end) : 992 _space_id(space_id), 993 _region_index_start(region_index_start), 994 _region_index_end(region_index_end) {} 995 }; 996 997 public: 998 // Inline closure decls 999 // 1000 class IsAliveClosure: public BoolObjectClosure { 1001 public: 1002 virtual bool do_object_b(oop p); 1003 }; 1004 1005 friend class RefProcTaskProxy; 1006 friend class PSParallelCompactTest; 1007 1008 private: 1009 static STWGCTimer _gc_timer; 1010 static ParallelOldTracer _gc_tracer; 1011 static elapsedTimer _accumulated_time; 1012 static unsigned int _total_invocations; 1013 static unsigned int _maximum_compaction_gc_num; 1014 static jlong _time_of_last_gc; // ms 1015 static CollectorCounters* _counters; 1016 static ParMarkBitMap _mark_bitmap; 1017 static ParallelCompactData _summary_data; 1018 static IsAliveClosure _is_alive_closure; 1019 static SpaceInfo _space_info[last_space_id]; 1020 1021 // Reference processing (used in ...follow_contents) 1022 static SpanSubjectToDiscoveryClosure _span_based_discoverer; 1023 static ReferenceProcessor* _ref_processor; 1024 1025 // Values computed at initialization and used by dead_wood_limiter(). 1026 static double _dwl_mean; 1027 static double _dwl_std_dev; 1028 static double _dwl_first_term; 1029 static double _dwl_adjustment; 1030 #ifdef ASSERT 1031 static bool _dwl_initialized; 1032 #endif // #ifdef ASSERT 1033 1034 public: 1035 static ParallelOldTracer* gc_tracer() { return &_gc_tracer; } 1036 1037 private: 1038 1039 static void initialize_space_info(); 1040 1041 // Clear the marking bitmap and summary data that cover the specified space. 1042 static void clear_data_covering_space(SpaceId id); 1043 1044 static void pre_compact(); 1045 static void post_compact(); 1046 1047 // Mark live objects 1048 static void marking_phase(ParCompactionManager* cm, 1049 bool maximum_heap_compaction, 1050 ParallelOldTracer *gc_tracer); 1051 1052 // Compute the dense prefix for the designated space. This is an experimental 1053 // implementation currently not used in production. 1054 static HeapWord* compute_dense_prefix_via_density(const SpaceId id, 1055 bool maximum_compaction); 1056 1057 // Methods used to compute the dense prefix. 1058 1059 // Compute the value of the normal distribution at x = density. The mean and 1060 // standard deviation are values saved by initialize_dead_wood_limiter(). 1061 static inline double normal_distribution(double density); 1062 1063 // Initialize the static vars used by dead_wood_limiter(). 1064 static void initialize_dead_wood_limiter(); 1065 1066 // Return the percentage of space that can be treated as "dead wood" (i.e., 1067 // not reclaimed). 1068 static double dead_wood_limiter(double density, size_t min_percent); 1069 1070 // Find the first (left-most) region in the range [beg, end) that has at least 1071 // dead_words of dead space to the left. The argument beg must be the first 1072 // region in the space that is not completely live. 1073 static RegionData* dead_wood_limit_region(const RegionData* beg, 1074 const RegionData* end, 1075 size_t dead_words); 1076 1077 // Return a pointer to the first region in the range [beg, end) that is not 1078 // completely full. 1079 static RegionData* first_dead_space_region(const RegionData* beg, 1080 const RegionData* end); 1081 1082 // Return a value indicating the benefit or 'yield' if the compacted region 1083 // were to start (or equivalently if the dense prefix were to end) at the 1084 // candidate region. Higher values are better. 1085 // 1086 // The value is based on the amount of space reclaimed vs. the costs of (a) 1087 // updating references in the dense prefix plus (b) copying objects and 1088 // updating references in the compacted region. 1089 static inline double reclaimed_ratio(const RegionData* const candidate, 1090 HeapWord* const bottom, 1091 HeapWord* const top, 1092 HeapWord* const new_top); 1093 1094 // Compute the dense prefix for the designated space. 1095 static HeapWord* compute_dense_prefix(const SpaceId id, 1096 bool maximum_compaction); 1097 1098 // Return true if dead space crosses onto the specified Region; bit must be 1099 // the bit index corresponding to the first word of the Region. 1100 static inline bool dead_space_crosses_boundary(const RegionData* region, 1101 idx_t bit); 1102 1103 // Summary phase utility routine to fill dead space (if any) at the dense 1104 // prefix boundary. Should only be called if the the dense prefix is 1105 // non-empty. 1106 static void fill_dense_prefix_end(SpaceId id); 1107 1108 static void summarize_spaces_quick(); 1109 static void summarize_space(SpaceId id, bool maximum_compaction); 1110 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction); 1111 1112 // Adjust addresses in roots. Does not adjust addresses in heap. 1113 static void adjust_roots(ParCompactionManager* cm); 1114 1115 DEBUG_ONLY(static void write_block_fill_histogram();) 1116 1117 // Move objects to new locations. 1118 static void compact_perm(ParCompactionManager* cm); 1119 static void compact(); 1120 1121 // Add available regions to the stack and draining tasks to the task queue. 1122 static void prepare_region_draining_tasks(uint parallel_gc_threads); 1123 1124 // Add dense prefix update tasks to the task queue. 1125 static void enqueue_dense_prefix_tasks(TaskQueue& task_queue, 1126 uint parallel_gc_threads); 1127 1128 // If objects are left in eden after a collection, try to move the boundary 1129 // and absorb them into the old gen. Returns true if eden was emptied. 1130 static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy, 1131 PSYoungGen* young_gen, 1132 PSOldGen* old_gen); 1133 1134 // Reset time since last full gc 1135 static void reset_millis_since_last_gc(); 1136 1137 #ifndef PRODUCT 1138 // Print generic summary data 1139 static void print_generic_summary_data(ParallelCompactData& summary_data, 1140 HeapWord* const beg_addr, 1141 HeapWord* const end_addr); 1142 #endif // #ifndef PRODUCT 1143 1144 public: 1145 1146 PSParallelCompact(); 1147 1148 static void invoke(bool maximum_heap_compaction); 1149 static bool invoke_no_policy(bool maximum_heap_compaction); 1150 1151 static void post_initialize(); 1152 // Perform initialization for PSParallelCompact that requires 1153 // allocations. This should be called during the VM initialization 1154 // at a pointer where it would be appropriate to return a JNI_ENOMEM 1155 // in the event of a failure. 1156 static bool initialize(); 1157 1158 // Closure accessors 1159 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; } 1160 1161 // Public accessors 1162 static elapsedTimer* accumulated_time() { return &_accumulated_time; } 1163 static unsigned int total_invocations() { return _total_invocations; } 1164 static CollectorCounters* counters() { return _counters; } 1165 1166 // Marking support 1167 static inline bool mark_obj(oop obj); 1168 static inline bool is_marked(oop obj); 1169 1170 template <class T> static inline void adjust_pointer(T* p, ParCompactionManager* cm); 1171 1172 // Compaction support. 1173 // Return true if p is in the range [beg_addr, end_addr). 1174 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr); 1175 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr); 1176 1177 // Convenience wrappers for per-space data kept in _space_info. 1178 static inline MutableSpace* space(SpaceId space_id); 1179 static inline HeapWord* new_top(SpaceId space_id); 1180 static inline HeapWord* dense_prefix(SpaceId space_id); 1181 static inline ObjectStartArray* start_array(SpaceId space_id); 1182 1183 // Process the end of the given region range in the dense prefix. 1184 // This includes saving any object not updated. 1185 static void dense_prefix_regions_epilogue(ParCompactionManager* cm, 1186 size_t region_start_index, 1187 size_t region_end_index, 1188 idx_t exiting_object_offset, 1189 idx_t region_offset_start, 1190 idx_t region_offset_end); 1191 1192 // Update a region in the dense prefix. For each live object 1193 // in the region, update it's interior references. For each 1194 // dead object, fill it with deadwood. Dead space at the end 1195 // of a region range will be filled to the start of the next 1196 // live object regardless of the region_index_end. None of the 1197 // objects in the dense prefix move and dead space is dead 1198 // (holds only dead objects that don't need any processing), so 1199 // dead space can be filled in any order. 1200 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm, 1201 SpaceId space_id, 1202 size_t region_index_start, 1203 size_t region_index_end); 1204 1205 // Return the address of the count + 1st live word in the range [beg, end). 1206 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count); 1207 1208 // Return the address of the word to be copied to dest_addr, which must be 1209 // aligned to a region boundary. 1210 static HeapWord* first_src_addr(HeapWord* const dest_addr, 1211 SpaceId src_space_id, 1212 size_t src_region_idx); 1213 1214 // Determine the next source region, set closure.source() to the start of the 1215 // new region return the region index. Parameter end_addr is the address one 1216 // beyond the end of source range just processed. If necessary, switch to a 1217 // new source space and set src_space_id (in-out parameter) and src_space_top 1218 // (out parameter) accordingly. 1219 static size_t next_src_region(MoveAndUpdateClosure& closure, 1220 SpaceId& src_space_id, 1221 HeapWord*& src_space_top, 1222 HeapWord* end_addr); 1223 1224 // Decrement the destination count for each non-empty source region in the 1225 // range [beg_region, region(region_align_up(end_addr))). If the destination 1226 // count for a region goes to 0 and it needs to be filled, enqueue it. 1227 static void decrement_destination_counts(ParCompactionManager* cm, 1228 SpaceId src_space_id, 1229 size_t beg_region, 1230 HeapWord* end_addr); 1231 1232 static void fill_region(ParCompactionManager* cm, MoveAndUpdateClosure& closure, size_t region); 1233 static void fill_and_update_region(ParCompactionManager* cm, size_t region); 1234 1235 static bool steal_shadow_region(ParCompactionManager* cm, size_t& region_idx); 1236 static void fill_shadow_region(ParCompactionManager* cm, size_t region_idx); 1237 static void fill_and_update_shadow_region(ParCompactionManager* cm, size_t region) { 1238 fill_shadow_region(cm, region); 1239 } 1240 // Copy the content of a shadow region back to its corresponding heap region 1241 static void copy_back(HeapWord* shadow_addr, HeapWord* region_addr); 1242 // Initialize the steal record of a GC thread 1243 static void initialize_steal_record(uint which); 1244 // Reuse the empty heap regions as shadow regions, like to-space regions 1245 static void enqueue_shadow_region(); 1246 1247 // Fill in the block table for the specified region. 1248 static void fill_blocks(size_t region_idx); 1249 1250 // Update the deferred objects in the space. 1251 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id); 1252 1253 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; } 1254 static ParallelCompactData& summary_data() { return _summary_data; } 1255 1256 // Reference Processing 1257 static ReferenceProcessor* const ref_processor() { return _ref_processor; } 1258 1259 static STWGCTimer* gc_timer() { return &_gc_timer; } 1260 1261 // Return the SpaceId for the given address. 1262 static SpaceId space_id(HeapWord* addr); 1263 1264 // Time since last full gc (in milliseconds). 1265 static jlong millis_since_last_gc(); 1266 1267 static void print_on_error(outputStream* st); 1268 1269 #ifndef PRODUCT 1270 // Debugging support. 1271 static const char* space_names[last_space_id]; 1272 static void print_region_ranges(); 1273 static void print_dense_prefix_stats(const char* const algorithm, 1274 const SpaceId id, 1275 const bool maximum_compaction, 1276 HeapWord* const addr); 1277 static void summary_phase_msg(SpaceId dst_space_id, 1278 HeapWord* dst_beg, HeapWord* dst_end, 1279 SpaceId src_space_id, 1280 HeapWord* src_beg, HeapWord* src_end); 1281 #endif // #ifndef PRODUCT 1282 1283 #ifdef ASSERT 1284 // Sanity check the new location of a word in the heap. 1285 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr); 1286 // Verify that all the regions have been emptied. 1287 static void verify_complete(SpaceId space_id); 1288 #endif // #ifdef ASSERT 1289 }; 1290 1291 class MoveAndUpdateClosure: public ParMarkBitMapClosure { 1292 static inline size_t calculate_words_remaining(size_t region); 1293 public: 1294 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, 1295 size_t region); 1296 1297 // Accessors. 1298 HeapWord* destination() const { return _destination; } 1299 HeapWord* copy_destination() const { return _destination + _offset; } 1300 1301 // If the object will fit (size <= words_remaining()), copy it to the current 1302 // destination, update the interior oops and the start array and return either 1303 // full (if the closure is full) or incomplete. If the object will not fit, 1304 // return would_overflow. 1305 IterationStatus do_addr(HeapWord* addr, size_t size); 1306 1307 // Copy enough words to fill this closure, starting at source(). Interior 1308 // oops and the start array are not updated. Return full. 1309 IterationStatus copy_until_full(); 1310 1311 // Copy enough words to fill this closure or to the end of an object, 1312 // whichever is smaller, starting at source(). Interior oops and the start 1313 // array are not updated. 1314 void copy_partial_obj(); 1315 1316 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr, 1317 PSParallelCompact::RegionData* region_ptr); 1318 1319 protected: 1320 // Update variables to indicate that word_count words were processed. 1321 inline void update_state(size_t word_count); 1322 1323 protected: 1324 HeapWord* _destination; // Next addr to be written. 1325 ObjectStartArray* const _start_array; 1326 size_t _offset; 1327 }; 1328 1329 inline size_t MoveAndUpdateClosure::calculate_words_remaining(size_t region) { 1330 HeapWord* dest_addr = PSParallelCompact::summary_data().region_to_addr(region); 1331 PSParallelCompact::SpaceId dest_space_id = PSParallelCompact::space_id(dest_addr); 1332 HeapWord* new_top = PSParallelCompact::new_top(dest_space_id); 1333 assert(dest_addr < new_top, "sanity"); 1334 1335 return MIN2(pointer_delta(new_top, dest_addr), ParallelCompactData::RegionSize); 1336 } 1337 1338 inline 1339 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap, 1340 ParCompactionManager* cm, 1341 size_t region_idx) : 1342 ParMarkBitMapClosure(bitmap, cm, calculate_words_remaining(region_idx)), 1343 _destination(PSParallelCompact::summary_data().region_to_addr(region_idx)), 1344 _start_array(PSParallelCompact::start_array(PSParallelCompact::space_id(_destination))), 1345 _offset(0) { } 1346 1347 1348 inline void MoveAndUpdateClosure::update_state(size_t words) 1349 { 1350 decrement_words_remaining(words); 1351 _source += words; 1352 _destination += words; 1353 } 1354 1355 class ShadowClosure: public MoveAndUpdateClosure { 1356 inline size_t calculate_shadow_offset(size_t region_idx, size_t shadow_idx); 1357 public: 1358 inline ShadowClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, 1359 size_t region, size_t shadow); 1360 1361 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr, 1362 PSParallelCompact::RegionData* region_ptr); 1363 1364 private: 1365 size_t _shadow; 1366 }; 1367 1368 inline size_t ShadowClosure::calculate_shadow_offset(size_t region_idx, size_t shadow_idx) { 1369 ParallelCompactData& sd = PSParallelCompact::summary_data(); 1370 HeapWord* dest_addr = sd.region_to_addr(region_idx); 1371 HeapWord* shadow_addr = sd.region_to_addr(shadow_idx); 1372 return pointer_delta(shadow_addr, dest_addr); 1373 } 1374 1375 inline 1376 ShadowClosure::ShadowClosure(ParMarkBitMap *bitmap, 1377 ParCompactionManager *cm, 1378 size_t region, 1379 size_t shadow) : 1380 MoveAndUpdateClosure(bitmap, cm, region), 1381 _shadow(shadow) { 1382 _offset = calculate_shadow_offset(region, shadow); 1383 } 1384 1385 class UpdateOnlyClosure: public ParMarkBitMapClosure { 1386 private: 1387 const PSParallelCompact::SpaceId _space_id; 1388 ObjectStartArray* const _start_array; 1389 1390 public: 1391 UpdateOnlyClosure(ParMarkBitMap* mbm, 1392 ParCompactionManager* cm, 1393 PSParallelCompact::SpaceId space_id); 1394 1395 // Update the object. 1396 virtual IterationStatus do_addr(HeapWord* addr, size_t words); 1397 1398 inline void do_addr(HeapWord* addr); 1399 }; 1400 1401 class FillClosure: public ParMarkBitMapClosure { 1402 public: 1403 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id); 1404 1405 virtual IterationStatus do_addr(HeapWord* addr, size_t size); 1406 1407 private: 1408 ObjectStartArray* const _start_array; 1409 }; 1410 1411 #endif // SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP