1 #ifdef USE_PRAGMA_IDENT_HDR 2 #pragma ident "@(#)psParallelCompact.hpp 1.48 07/10/04 10:49:38 JVM" 3 #endif 4 /* 5 * Copyright 2005-2008 Sun Microsystems, Inc. All Rights Reserved. 6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 7 * 8 * This code is free software; you can redistribute it and/or modify it 9 * under the terms of the GNU General Public License version 2 only, as 10 * published by the Free Software Foundation. 11 * 12 * This code is distributed in the hope that it will be useful, but WITHOUT 13 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 14 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 15 * version 2 for more details (a copy is included in the LICENSE file that 16 * accompanied this code). 17 * 18 * You should have received a copy of the GNU General Public License version 19 * 2 along with this work; if not, write to the Free Software Foundation, 20 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 21 * 22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, 23 * CA 95054 USA or visit www.sun.com if you need additional information or 24 * have any questions. 25 * 26 */ 27 28 class ParallelScavengeHeap; 29 class PSAdaptiveSizePolicy; 30 class PSYoungGen; 31 class PSOldGen; 32 class PSPermGen; 33 class ParCompactionManager; 34 class ParallelTaskTerminator; 35 class PSParallelCompact; 36 class GCTaskManager; 37 class GCTaskQueue; 38 class PreGCValues; 39 class MoveAndUpdateClosure; 40 class RefProcTaskExecutor; 41 42 // The SplitInfo class holds the information needed to 'split' a source region 43 // so that the live data can be copied to two destination *spaces*. Normally, 44 // all the live data in a region is copied to a single destination space (e.g., 45 // everything live in a region in eden is copied entirely into the old gen). 46 // However, when the heap is nearly full, all the live data in eden may not fit 47 // into the old gen. Copying only some of the regions from eden to old gen 48 // requires finding a region that does not contain a partial object (i.e., no 49 // live object crosses the region boundary) somewhere near the last object that 50 // does fit into the old gen. Since it's not always possible to find such a 51 // region, splitting is necessary for predictable behavior. 52 // 53 // A region is always split at the end of the partial object. This avoids 54 // additional tests when calculating the new location of a pointer, which is a 55 // very hot code path. The partial object and everything to its left will be 56 // copied to another space (call it dest_space_1). The live data to the right 57 // of the partial object will be copied either within the space itself, or to a 58 // different destination space (distinct from dest_space_1). 59 // 60 // Split points are identified during the summary phase, when region 61 // destinations are computed: data about the split, including the 62 // partial_object_size, is recorded in a SplitInfo record and the 63 // partial_object_size field in the summary data is set to zero. The zeroing is 64 // possible (and necessary) since the partial object will move to a different 65 // destination space than anything to its right, thus the partial object should 66 // not affect the locations of any objects to its right. 67 // 68 // The recorded data is used during the compaction phase, but only rarely: when 69 // the partial object on the split region will be copied across a destination 70 // region boundary. This test is made once each time a region is filled, and is 71 // a simple address comparison, so the overhead is negligible (see 72 // PSParallelCompact::first_src_addr()). 73 // 74 // Notes: 75 // 76 // Only regions with partial objects are split; a region without a partial 77 // object does not need any extra bookkeeping. 78 // 79 // At most one region is split per space, so the amount of data required is 80 // constant. 81 // 82 // A region is split only when the destination space would overflow. Once that 83 // happens, the destination space is abandoned and no other data (even from 84 // other source spaces) is targeted to that destination space. Abandoning the 85 // destination space may leave a somewhat large unused area at the end, if a 86 // large object caused the overflow. 87 // 88 // Future work: 89 // 90 // More bookkeeping would be required to continue to use the destination space. 91 // The most general solution would allow data from regions in two different 92 // source spaces to be "joined" in a single destination region. At the very 93 // least, additional code would be required in next_src_region() to detect the 94 // join and skip to an out-of-order source region. If the join region was also 95 // the last destination region to which a split region was copied (the most 96 // likely case), then additional work would be needed to get fill_region() to 97 // stop iteration and switch to a new source region at the right point. Basic 98 // idea would be to use a fake value for the top of the source space. It is 99 // doable, if a bit tricky. 100 // 101 // A simpler (but less general) solution would fill the remainder of the 102 // destination region with a dummy object and continue filling the next 103 // destination region. 104 105 class SplitInfo 106 { 107 public: 108 // Return true if this split info is valid (i.e., if a split has been 109 // recorded). The very first region cannot have a partial object and thus is 110 // never split, so 0 is the 'invalid' value. 111 bool is_valid() const { return _src_region_idx > 0; } 112 113 // Return true if this split holds data for the specified source region. 114 inline bool is_split(size_t source_region) const; 115 116 // The index of the split region, the size of the partial object on that 117 // region and the destination of the partial object. 118 size_t src_region_idx() const { return _src_region_idx; } 119 size_t partial_obj_size() const { return _partial_obj_size; } 120 HeapWord* destination() const { return _destination; } 121 122 // The destination count of the partial object referenced by this split 123 // (either 1 or 2). This must be added to the destination count of the 124 // remainder of the source region. 125 unsigned int destination_count() const { return _destination_count; } 126 127 // If a word within the partial object will be written to the first word of a 128 // destination region, this is the address of the destination region; 129 // otherwise this is NULL. 130 HeapWord* dest_region_addr() const { return _dest_region_addr; } 131 132 // If a word within the partial object will be written to the first word of a 133 // destination region, this is the address of that word within the partial 134 // object; otherwise this is NULL. 135 HeapWord* first_src_addr() const { return _first_src_addr; } 136 137 // Record the data necessary to split the region src_region_idx. 138 void record(size_t src_region_idx, size_t partial_obj_size, 139 HeapWord* destination); 140 141 void clear(); 142 143 DEBUG_ONLY(void verify_clear();) 144 145 private: 146 size_t _src_region_idx; 147 size_t _partial_obj_size; 148 HeapWord* _destination; 149 unsigned int _destination_count; 150 HeapWord* _dest_region_addr; 151 HeapWord* _first_src_addr; 152 }; 153 154 inline bool SplitInfo::is_split(size_t region_idx) const 155 { 156 return _src_region_idx == region_idx && is_valid(); 157 } 158 159 class SpaceInfo 160 { 161 public: 162 MutableSpace* space() const { return _space; } 163 164 // Where the free space will start after the collection. Valid only after the 165 // summary phase completes. 166 HeapWord* new_top() const { return _new_top; } 167 168 // Allows new_top to be set. 169 HeapWord** new_top_addr() { return &_new_top; } 170 171 // Where the smallest allowable dense prefix ends (used only for perm gen). 172 HeapWord* min_dense_prefix() const { return _min_dense_prefix; } 173 174 // Where the dense prefix ends, or the compacted region begins. 175 HeapWord* dense_prefix() const { return _dense_prefix; } 176 177 // The start array for the (generation containing the) space, or NULL if there 178 // is no start array. 179 ObjectStartArray* start_array() const { return _start_array; } 180 181 SplitInfo& split_info() { return _split_info; } 182 183 void set_space(MutableSpace* s) { _space = s; } 184 void set_new_top(HeapWord* addr) { _new_top = addr; } 185 void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; } 186 void set_dense_prefix(HeapWord* addr) { _dense_prefix = addr; } 187 void set_start_array(ObjectStartArray* s) { _start_array = s; } 188 189 void publish_new_top() const { _space->set_top(_new_top); } 190 191 private: 192 MutableSpace* _space; 193 HeapWord* _new_top; 194 HeapWord* _min_dense_prefix; 195 HeapWord* _dense_prefix; 196 ObjectStartArray* _start_array; 197 SplitInfo _split_info; 198 }; 199 200 class ParallelCompactData 201 { 202 public: 203 // Sizes are in HeapWords, unless indicated otherwise. 204 static const size_t Log2RegionSize; 205 static const size_t RegionSize; 206 static const size_t RegionSizeBytes; 207 208 // Mask for the bits in a size_t to get an offset within a region. 209 static const size_t RegionSizeOffsetMask; 210 // Mask for the bits in a pointer to get an offset within a region. 211 static const size_t RegionAddrOffsetMask; 212 // Mask for the bits in a pointer to get the address of the start of a region. 213 static const size_t RegionAddrMask; 214 215 class RegionData 216 { 217 public: 218 // Destination address of the region. 219 HeapWord* destination() const { return _destination; } 220 221 // The first region containing data destined for this region. 222 size_t source_region() const { return _source_region; } 223 224 // The object (if any) starting in this region and ending in a different 225 // region that could not be updated during the main (parallel) compaction 226 // phase. This is different from _partial_obj_addr, which is an object that 227 // extends onto a source region. However, the two uses do not overlap in 228 // time, so the same field is used to save space. 229 HeapWord* deferred_obj_addr() const { return _partial_obj_addr; } 230 231 // The starting address of the partial object extending onto the region. 232 HeapWord* partial_obj_addr() const { return _partial_obj_addr; } 233 234 // Size of the partial object extending onto the region (words). 235 size_t partial_obj_size() const { return _partial_obj_size; } 236 237 // Size of live data that lies within this region due to objects that start 238 // in this region (words). This does not include the partial object 239 // extending onto the region (if any), or the part of an object that extends 240 // onto the next region (if any). 241 size_t live_obj_size() const { return _dc_and_los & los_mask; } 242 243 // Total live data that lies within the region (words). 244 size_t data_size() const { return partial_obj_size() + live_obj_size(); } 245 246 // The destination_count is the number of other regions to which data from 247 // this region will be copied. At the end of the summary phase, the valid 248 // values of destination_count are 249 // 250 // 0 - data from the region will be compacted completely into itself, or the 251 // region is empty. The region can be claimed and then filled. 252 // 1 - data from the region will be compacted into 1 other region; some 253 // data from the region may also be compacted into the region itself. 254 // 2 - data from the region will be copied to 2 other regions. 255 // 256 // During compaction as regions are emptied, the destination_count is 257 // decremented (atomically) and when it reaches 0, it can be claimed and 258 // then filled. 259 // 260 // A region is claimed for processing by atomically changing the 261 // destination_count to the claimed value (dc_claimed). After a region has 262 // been filled, the destination_count should be set to the completed value 263 // (dc_completed). 264 inline uint destination_count() const; 265 inline uint destination_count_raw() const; 266 267 // The location of the java heap data that corresponds to this region. 268 inline HeapWord* data_location() const; 269 270 // The highest address referenced by objects in this region. 271 inline HeapWord* highest_ref() const; 272 273 // Whether this region is available to be claimed, has been claimed, or has 274 // been completed. 275 // 276 // Minor subtlety: claimed() returns true if the region is marked 277 // completed(), which is desirable since a region must be claimed before it 278 // can be completed. 279 bool available() const { return _dc_and_los < dc_one; } 280 bool claimed() const { return _dc_and_los >= dc_claimed; } 281 bool completed() const { return _dc_and_los >= dc_completed; } 282 283 // These are not atomic. 284 void set_destination(HeapWord* addr) { _destination = addr; } 285 void set_source_region(size_t region) { _source_region = region; } 286 void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } 287 void set_partial_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } 288 void set_partial_obj_size(size_t words) { 289 _partial_obj_size = (region_sz_t) words; 290 } 291 292 inline void set_destination_count(uint count); 293 inline void set_live_obj_size(size_t words); 294 inline void set_data_location(HeapWord* addr); 295 inline void set_completed(); 296 inline bool claim_unsafe(); 297 298 // These are atomic. 299 inline void add_live_obj(size_t words); 300 inline void set_highest_ref(HeapWord* addr); 301 inline void decrement_destination_count(); 302 inline bool claim(); 303 304 private: 305 // The type used to represent object sizes within a region. 306 typedef uint region_sz_t; 307 308 // Constants for manipulating the _dc_and_los field, which holds both the 309 // destination count and live obj size. The live obj size lives at the 310 // least significant end so no masking is necessary when adding. 311 static const region_sz_t dc_shift; // Shift amount. 312 static const region_sz_t dc_mask; // Mask for destination count. 313 static const region_sz_t dc_one; // 1, shifted appropriately. 314 static const region_sz_t dc_claimed; // Region has been claimed. 315 static const region_sz_t dc_completed; // Region has been completed. 316 static const region_sz_t los_mask; // Mask for live obj size. 317 318 HeapWord* _destination; 319 size_t _source_region; 320 HeapWord* _partial_obj_addr; 321 region_sz_t _partial_obj_size; 322 region_sz_t volatile _dc_and_los; 323 #ifdef ASSERT 324 // These enable optimizations that are only partially implemented. Use 325 // debug builds to prevent the code fragments from breaking. 326 HeapWord* _data_location; 327 HeapWord* _highest_ref; 328 #endif // #ifdef ASSERT 329 330 #ifdef ASSERT 331 public: 332 uint _pushed; // 0 until region is pushed onto a worker's stack 333 private: 334 #endif 335 }; 336 337 public: 338 ParallelCompactData(); 339 bool initialize(MemRegion covered_region); 340 341 size_t region_count() const { return _region_count; } 342 343 // Convert region indices to/from RegionData pointers. 344 inline RegionData* region(size_t region_idx) const; 345 inline size_t region(const RegionData* const region_ptr) const; 346 347 // Returns true if the given address is contained within the region 348 bool region_contains(size_t region_index, HeapWord* addr); 349 350 void add_obj(HeapWord* addr, size_t len); 351 void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); } 352 353 // Fill in the regions covering [beg, end) so that no data moves; i.e., the 354 // destination of region n is simply the start of region n. The argument beg 355 // must be region-aligned; end need not be. 356 void summarize_dense_prefix(HeapWord* beg, HeapWord* end); 357 358 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info, 359 HeapWord* destination, HeapWord* target_end, 360 HeapWord** target_next); 361 bool summarize(SplitInfo& split_info, 362 HeapWord* source_beg, HeapWord* source_end, 363 HeapWord** source_next, 364 HeapWord* target_beg, HeapWord* target_end, 365 HeapWord** target_next); 366 367 void clear(); 368 void clear_range(size_t beg_region, size_t end_region); 369 void clear_range(HeapWord* beg, HeapWord* end) { 370 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end)); 371 } 372 373 // Return the number of words between addr and the start of the region 374 // containing addr. 375 inline size_t region_offset(const HeapWord* addr) const; 376 377 // Convert addresses to/from a region index or region pointer. 378 inline size_t addr_to_region_idx(const HeapWord* addr) const; 379 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const; 380 inline HeapWord* region_to_addr(size_t region) const; 381 inline HeapWord* region_to_addr(size_t region, size_t offset) const; 382 inline HeapWord* region_to_addr(const RegionData* region) const; 383 384 inline HeapWord* region_align_down(HeapWord* addr) const; 385 inline HeapWord* region_align_up(HeapWord* addr) const; 386 inline bool is_region_aligned(HeapWord* addr) const; 387 388 // Return the address one past the end of the partial object. 389 HeapWord* partial_obj_end(size_t region_idx) const; 390 391 // Return the new location of the object p after the 392 // the compaction. 393 HeapWord* calc_new_pointer(HeapWord* addr); 394 395 HeapWord* calc_new_pointer(oop p) { 396 return calc_new_pointer((HeapWord*) p); 397 } 398 399 // Return the updated address for the given klass 400 klassOop calc_new_klass(klassOop); 401 402 #ifdef ASSERT 403 void verify_clear(const PSVirtualSpace* vspace); 404 void verify_clear(); 405 #endif // #ifdef ASSERT 406 407 private: 408 bool initialize_region_data(size_t region_size); 409 PSVirtualSpace* create_vspace(size_t count, size_t element_size); 410 411 private: 412 HeapWord* _region_start; 413 #ifdef ASSERT 414 HeapWord* _region_end; 415 #endif // #ifdef ASSERT 416 417 PSVirtualSpace* _region_vspace; 418 RegionData* _region_data; 419 size_t _region_count; 420 }; 421 422 inline uint 423 ParallelCompactData::RegionData::destination_count_raw() const 424 { 425 return _dc_and_los & dc_mask; 426 } 427 428 inline uint 429 ParallelCompactData::RegionData::destination_count() const 430 { 431 return destination_count_raw() >> dc_shift; 432 } 433 434 inline void 435 ParallelCompactData::RegionData::set_destination_count(uint count) 436 { 437 assert(count <= (dc_completed >> dc_shift), "count too large"); 438 const region_sz_t live_sz = (region_sz_t) live_obj_size(); 439 _dc_and_los = (count << dc_shift) | live_sz; 440 } 441 442 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words) 443 { 444 assert(words <= los_mask, "would overflow"); 445 _dc_and_los = destination_count_raw() | (region_sz_t)words; 446 } 447 448 inline void ParallelCompactData::RegionData::decrement_destination_count() 449 { 450 assert(_dc_and_los < dc_claimed, "already claimed"); 451 assert(_dc_and_los >= dc_one, "count would go negative"); 452 Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los); 453 } 454 455 inline HeapWord* ParallelCompactData::RegionData::data_location() const 456 { 457 DEBUG_ONLY(return _data_location;) 458 NOT_DEBUG(return NULL;) 459 } 460 461 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const 462 { 463 DEBUG_ONLY(return _highest_ref;) 464 NOT_DEBUG(return NULL;) 465 } 466 467 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr) 468 { 469 DEBUG_ONLY(_data_location = addr;) 470 } 471 472 inline void ParallelCompactData::RegionData::set_completed() 473 { 474 assert(claimed(), "must be claimed first"); 475 _dc_and_los = dc_completed | (region_sz_t) live_obj_size(); 476 } 477 478 // MT-unsafe claiming of a region. Should only be used during single threaded 479 // execution. 480 inline bool ParallelCompactData::RegionData::claim_unsafe() 481 { 482 if (available()) { 483 _dc_and_los |= dc_claimed; 484 return true; 485 } 486 return false; 487 } 488 489 inline void ParallelCompactData::RegionData::add_live_obj(size_t words) 490 { 491 assert(words <= (size_t)los_mask - live_obj_size(), "overflow"); 492 Atomic::add((int) words, (volatile int*) &_dc_and_los); 493 } 494 495 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr) 496 { 497 #ifdef ASSERT 498 HeapWord* tmp = _highest_ref; 499 while (addr > tmp) { 500 tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp); 501 } 502 #endif // #ifdef ASSERT 503 } 504 505 inline bool ParallelCompactData::RegionData::claim() 506 { 507 const int los = (int) live_obj_size(); 508 const int old = Atomic::cmpxchg(dc_claimed | los, 509 (volatile int*) &_dc_and_los, los); 510 return old == los; 511 } 512 513 inline ParallelCompactData::RegionData* 514 ParallelCompactData::region(size_t region_idx) const 515 { 516 assert(region_idx <= region_count(), "bad arg"); 517 return _region_data + region_idx; 518 } 519 520 inline size_t 521 ParallelCompactData::region(const RegionData* const region_ptr) const 522 { 523 assert(region_ptr >= _region_data, "bad arg"); 524 assert(region_ptr <= _region_data + region_count(), "bad arg"); 525 return pointer_delta(region_ptr, _region_data, sizeof(RegionData)); 526 } 527 528 inline size_t 529 ParallelCompactData::region_offset(const HeapWord* addr) const 530 { 531 assert(addr >= _region_start, "bad addr"); 532 assert(addr <= _region_end, "bad addr"); 533 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize; 534 } 535 536 inline size_t 537 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const 538 { 539 assert(addr >= _region_start, "bad addr"); 540 assert(addr <= _region_end, "bad addr"); 541 return pointer_delta(addr, _region_start) >> Log2RegionSize; 542 } 543 544 inline ParallelCompactData::RegionData* 545 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const 546 { 547 return region(addr_to_region_idx(addr)); 548 } 549 550 inline HeapWord* 551 ParallelCompactData::region_to_addr(size_t region) const 552 { 553 assert(region <= _region_count, "region out of range"); 554 return _region_start + (region << Log2RegionSize); 555 } 556 557 inline HeapWord* 558 ParallelCompactData::region_to_addr(const RegionData* region) const 559 { 560 return region_to_addr(pointer_delta(region, _region_data, 561 sizeof(RegionData))); 562 } 563 564 inline HeapWord* 565 ParallelCompactData::region_to_addr(size_t region, size_t offset) const 566 { 567 assert(region <= _region_count, "region out of range"); 568 assert(offset < RegionSize, "offset too big"); // This may be too strict. 569 return region_to_addr(region) + offset; 570 } 571 572 inline HeapWord* 573 ParallelCompactData::region_align_down(HeapWord* addr) const 574 { 575 assert(addr >= _region_start, "bad addr"); 576 assert(addr < _region_end + RegionSize, "bad addr"); 577 return (HeapWord*)(size_t(addr) & RegionAddrMask); 578 } 579 580 inline HeapWord* 581 ParallelCompactData::region_align_up(HeapWord* addr) const 582 { 583 assert(addr >= _region_start, "bad addr"); 584 assert(addr <= _region_end, "bad addr"); 585 return region_align_down(addr + RegionSizeOffsetMask); 586 } 587 588 inline bool 589 ParallelCompactData::is_region_aligned(HeapWord* addr) const 590 { 591 return region_offset(addr) == 0; 592 } 593 594 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the 595 // do_addr() method. 596 // 597 // The closure is initialized with the number of heap words to process 598 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr() 599 // methods in subclasses should update the total as words are processed. Since 600 // only one subclass actually uses this mechanism to terminate iteration, the 601 // default initial value is > 0. The implementation is here and not in the 602 // single subclass that uses it to avoid making is_full() virtual, and thus 603 // adding a virtual call per live object. 604 605 class ParMarkBitMapClosure: public StackObj { 606 public: 607 typedef ParMarkBitMap::idx_t idx_t; 608 typedef ParMarkBitMap::IterationStatus IterationStatus; 609 610 public: 611 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm, 612 size_t words = max_uintx); 613 614 inline ParCompactionManager* compaction_manager() const; 615 inline ParMarkBitMap* bitmap() const; 616 inline size_t words_remaining() const; 617 inline bool is_full() const; 618 inline HeapWord* source() const; 619 620 inline void set_source(HeapWord* addr); 621 622 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0; 623 624 protected: 625 inline void decrement_words_remaining(size_t words); 626 627 private: 628 ParMarkBitMap* const _bitmap; 629 ParCompactionManager* const _compaction_manager; 630 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger. 631 size_t _words_remaining; // Words left to copy. 632 633 protected: 634 HeapWord* _source; // Next addr that would be read. 635 }; 636 637 inline 638 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap, 639 ParCompactionManager* cm, 640 size_t words): 641 _bitmap(bitmap), _compaction_manager(cm) 642 #ifdef ASSERT 643 , _initial_words_remaining(words) 644 #endif 645 { 646 _words_remaining = words; 647 _source = NULL; 648 } 649 650 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const { 651 return _compaction_manager; 652 } 653 654 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const { 655 return _bitmap; 656 } 657 658 inline size_t ParMarkBitMapClosure::words_remaining() const { 659 return _words_remaining; 660 } 661 662 inline bool ParMarkBitMapClosure::is_full() const { 663 return words_remaining() == 0; 664 } 665 666 inline HeapWord* ParMarkBitMapClosure::source() const { 667 return _source; 668 } 669 670 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) { 671 _source = addr; 672 } 673 674 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) { 675 assert(_words_remaining >= words, "processed too many words"); 676 _words_remaining -= words; 677 } 678 679 // The UseParallelOldGC collector is a stop-the-world garbage collector that 680 // does parts of the collection using parallel threads. The collection includes 681 // the tenured generation and the young generation. The permanent generation is 682 // collected at the same time as the other two generations but the permanent 683 // generation is collect by a single GC thread. The permanent generation is 684 // collected serially because of the requirement that during the processing of a 685 // klass AAA, any objects reference by AAA must already have been processed. 686 // This requirement is enforced by a left (lower address) to right (higher 687 // address) sliding compaction. 688 // 689 // There are four phases of the collection. 690 // 691 // - marking phase 692 // - summary phase 693 // - compacting phase 694 // - clean up phase 695 // 696 // Roughly speaking these phases correspond, respectively, to 697 // - mark all the live objects 698 // - calculate the destination of each object at the end of the collection 699 // - move the objects to their destination 700 // - update some references and reinitialize some variables 701 // 702 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The 703 // marking phase is implemented in PSParallelCompact::marking_phase() and does a 704 // complete marking of the heap. The summary phase is implemented in 705 // PSParallelCompact::summary_phase(). The move and update phase is implemented 706 // in PSParallelCompact::compact(). 707 // 708 // A space that is being collected is divided into regions and with each region 709 // is associated an object of type ParallelCompactData. Each region is of a 710 // fixed size and typically will contain more than 1 object and may have parts 711 // of objects at the front and back of the region. 712 // 713 // region -----+---------------------+---------- 714 // objects covered [ AAA )[ BBB )[ CCC )[ DDD ) 715 // 716 // The marking phase does a complete marking of all live objects in the heap. 717 // The marking also compiles the size of the data for all live objects covered 718 // by the region. This size includes the part of any live object spanning onto 719 // the region (part of AAA if it is live) from the front, all live objects 720 // contained in the region (BBB and/or CCC if they are live), and the part of 721 // any live objects covered by the region that extends off the region (part of 722 // DDD if it is live). The marking phase uses multiple GC threads and marking 723 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is 724 // done atomically as is the accumulation of the size of the live objects 725 // covered by a region. 726 // 727 // The summary phase calculates the total live data to the left of each region 728 // XXX. Based on that total and the bottom of the space, it can calculate the 729 // starting location of the live data in XXX. The summary phase calculates for 730 // each region XXX quantites such as 731 // 732 // - the amount of live data at the beginning of a region from an object 733 // entering the region. 734 // - the location of the first live data on the region 735 // - a count of the number of regions receiving live data from XXX. 736 // 737 // See ParallelCompactData for precise details. The summary phase also 738 // calculates the dense prefix for the compaction. The dense prefix is a 739 // portion at the beginning of the space that is not moved. The objects in the 740 // dense prefix do need to have their object references updated. See method 741 // summarize_dense_prefix(). 742 // 743 // The summary phase is done using 1 GC thread. 744 // 745 // The compaction phase moves objects to their new location and updates all 746 // references in the object. 747 // 748 // A current exception is that objects that cross a region boundary are moved 749 // but do not have their references updated. References are not updated because 750 // it cannot easily be determined if the klass pointer KKK for the object AAA 751 // has been updated. KKK likely resides in a region to the left of the region 752 // containing AAA. These AAA's have there references updated at the end in a 753 // clean up phase. See the method PSParallelCompact::update_deferred_objects(). 754 // An alternate strategy is being investigated for this deferral of updating. 755 // 756 // Compaction is done on a region basis. A region that is ready to be filled is 757 // put on a ready list and GC threads take region off the list and fill them. A 758 // region is ready to be filled if it empty of live objects. Such a region may 759 // have been initially empty (only contained dead objects) or may have had all 760 // its live objects copied out already. A region that compacts into itself is 761 // also ready for filling. The ready list is initially filled with empty 762 // regions and regions compacting into themselves. There is always at least 1 763 // region that can be put on the ready list. The regions are atomically added 764 // and removed from the ready list. 765 766 class PSParallelCompact : AllStatic { 767 public: 768 // Convenient access to type names. 769 typedef ParMarkBitMap::idx_t idx_t; 770 typedef ParallelCompactData::RegionData RegionData; 771 772 typedef enum { 773 perm_space_id, old_space_id, eden_space_id, 774 from_space_id, to_space_id, last_space_id 775 } SpaceId; 776 777 public: 778 // Inline closure decls 779 // 780 class IsAliveClosure: public BoolObjectClosure { 781 public: 782 virtual void do_object(oop p); 783 virtual bool do_object_b(oop p); 784 }; 785 786 class KeepAliveClosure: public OopClosure { 787 private: 788 ParCompactionManager* _compaction_manager; 789 protected: 790 template <class T> inline void do_oop_work(T* p); 791 public: 792 KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } 793 virtual void do_oop(oop* p); 794 virtual void do_oop(narrowOop* p); 795 }; 796 797 // Current unused 798 class FollowRootClosure: public OopsInGenClosure { 799 private: 800 ParCompactionManager* _compaction_manager; 801 public: 802 FollowRootClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } 803 virtual void do_oop(oop* p); 804 virtual void do_oop(narrowOop* p); 805 virtual const bool do_nmethods() const { return true; } 806 }; 807 808 class FollowStackClosure: public VoidClosure { 809 private: 810 ParCompactionManager* _compaction_manager; 811 public: 812 FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } 813 virtual void do_void(); 814 }; 815 816 class AdjustPointerClosure: public OopsInGenClosure { 817 private: 818 bool _is_root; 819 public: 820 AdjustPointerClosure(bool is_root) : _is_root(is_root) { } 821 virtual void do_oop(oop* p); 822 virtual void do_oop(narrowOop* p); 823 }; 824 825 // Closure for verifying update of pointers. Does not 826 // have any side effects. 827 class VerifyUpdateClosure: public ParMarkBitMapClosure { 828 const MutableSpace* _space; // Is this ever used? 829 830 public: 831 VerifyUpdateClosure(ParCompactionManager* cm, const MutableSpace* sp) : 832 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm), _space(sp) 833 { } 834 835 virtual IterationStatus do_addr(HeapWord* addr, size_t words); 836 837 const MutableSpace* space() { return _space; } 838 }; 839 840 // Closure for updating objects altered for debug checking 841 class ResetObjectsClosure: public ParMarkBitMapClosure { 842 public: 843 ResetObjectsClosure(ParCompactionManager* cm): 844 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm) 845 { } 846 847 virtual IterationStatus do_addr(HeapWord* addr, size_t words); 848 }; 849 850 friend class KeepAliveClosure; 851 friend class FollowStackClosure; 852 friend class AdjustPointerClosure; 853 friend class FollowRootClosure; 854 friend class instanceKlassKlass; 855 friend class RefProcTaskProxy; 856 857 private: 858 static elapsedTimer _accumulated_time; 859 static unsigned int _total_invocations; 860 static unsigned int _maximum_compaction_gc_num; 861 static jlong _time_of_last_gc; // ms 862 static CollectorCounters* _counters; 863 static ParMarkBitMap _mark_bitmap; 864 static ParallelCompactData _summary_data; 865 static IsAliveClosure _is_alive_closure; 866 static SpaceInfo _space_info[last_space_id]; 867 static bool _print_phases; 868 static AdjustPointerClosure _adjust_root_pointer_closure; 869 static AdjustPointerClosure _adjust_pointer_closure; 870 871 // Reference processing (used in ...follow_contents) 872 static ReferenceProcessor* _ref_processor; 873 874 // Updated location of intArrayKlassObj. 875 static klassOop _updated_int_array_klass_obj; 876 877 // Values computed at initialization and used by dead_wood_limiter(). 878 static double _dwl_mean; 879 static double _dwl_std_dev; 880 static double _dwl_first_term; 881 static double _dwl_adjustment; 882 #ifdef ASSERT 883 static bool _dwl_initialized; 884 #endif // #ifdef ASSERT 885 886 private: 887 // Closure accessors 888 static OopClosure* adjust_pointer_closure() { return (OopClosure*)&_adjust_pointer_closure; } 889 static OopClosure* adjust_root_pointer_closure() { return (OopClosure*)&_adjust_root_pointer_closure; } 890 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; } 891 892 static void initialize_space_info(); 893 894 // Return true if details about individual phases should be printed. 895 static inline bool print_phases(); 896 897 // Clear the marking bitmap and summary data that cover the specified space. 898 static void clear_data_covering_space(SpaceId id); 899 900 static void pre_compact(PreGCValues* pre_gc_values); 901 static void post_compact(); 902 903 // Mark live objects 904 static void marking_phase(ParCompactionManager* cm, 905 bool maximum_heap_compaction); 906 static void follow_stack(ParCompactionManager* cm); 907 static void follow_weak_klass_links(ParCompactionManager* cm); 908 909 template <class T> static inline void adjust_pointer(T* p, bool is_root); 910 static void adjust_root_pointer(oop* p) { adjust_pointer(p, true); } 911 912 template <class T> 913 static inline void follow_root(ParCompactionManager* cm, T* p); 914 915 // Compute the dense prefix for the designated space. This is an experimental 916 // implementation currently not used in production. 917 static HeapWord* compute_dense_prefix_via_density(const SpaceId id, 918 bool maximum_compaction); 919 920 // Methods used to compute the dense prefix. 921 922 // Compute the value of the normal distribution at x = density. The mean and 923 // standard deviation are values saved by initialize_dead_wood_limiter(). 924 static inline double normal_distribution(double density); 925 926 // Initialize the static vars used by dead_wood_limiter(). 927 static void initialize_dead_wood_limiter(); 928 929 // Return the percentage of space that can be treated as "dead wood" (i.e., 930 // not reclaimed). 931 static double dead_wood_limiter(double density, size_t min_percent); 932 933 // Find the first (left-most) region in the range [beg, end) that has at least 934 // dead_words of dead space to the left. The argument beg must be the first 935 // region in the space that is not completely live. 936 static RegionData* dead_wood_limit_region(const RegionData* beg, 937 const RegionData* end, 938 size_t dead_words); 939 940 // Return a pointer to the first region in the range [beg, end) that is not 941 // completely full. 942 static RegionData* first_dead_space_region(const RegionData* beg, 943 const RegionData* end); 944 945 // Return a value indicating the benefit or 'yield' if the compacted region 946 // were to start (or equivalently if the dense prefix were to end) at the 947 // candidate region. Higher values are better. 948 // 949 // The value is based on the amount of space reclaimed vs. the costs of (a) 950 // updating references in the dense prefix plus (b) copying objects and 951 // updating references in the compacted region. 952 static inline double reclaimed_ratio(const RegionData* const candidate, 953 HeapWord* const bottom, 954 HeapWord* const top, 955 HeapWord* const new_top); 956 957 // Compute the dense prefix for the designated space. 958 static HeapWord* compute_dense_prefix(const SpaceId id, 959 bool maximum_compaction); 960 961 // Return true if dead space crosses onto the specified Region; bit must be 962 // the bit index corresponding to the first word of the Region. 963 static inline bool dead_space_crosses_boundary(const RegionData* region, 964 idx_t bit); 965 966 // Summary phase utility routine to fill dead space (if any) at the dense 967 // prefix boundary. Should only be called if the the dense prefix is 968 // non-empty. 969 static void fill_dense_prefix_end(SpaceId id); 970 971 // Clear the summary data source_region field for the specified addresses. 972 static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr); 973 974 #ifndef PRODUCT 975 // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot). 976 977 // Fill the region [start, start + words) with live object(s). Only usable 978 // for the old and permanent generations. 979 static void fill_with_live_objects(SpaceId id, HeapWord* const start, 980 size_t words); 981 // Include the new objects in the summary data. 982 static void summarize_new_objects(SpaceId id, HeapWord* start); 983 984 // Add live objects to a survivor space since it's rare that both survivors 985 // are non-empty. 986 static void provoke_split_fill_survivor(SpaceId id); 987 988 // Add live objects and/or choose the dense prefix to provoke splitting. 989 static void provoke_split(bool & maximum_compaction); 990 #endif 991 992 static void summarize_spaces_quick(); 993 static void summarize_space(SpaceId id, bool maximum_compaction); 994 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction); 995 996 // Adjust addresses in roots. Does not adjust addresses in heap. 997 static void adjust_roots(); 998 999 // Serial code executed in preparation for the compaction phase. 1000 static void compact_prologue(); 1001 1002 // Move objects to new locations. 1003 static void compact_perm(ParCompactionManager* cm); 1004 static void compact(); 1005 1006 // Add available regions to the stack and draining tasks to the task queue. 1007 static void enqueue_region_draining_tasks(GCTaskQueue* q, 1008 uint parallel_gc_threads); 1009 1010 // Add dense prefix update tasks to the task queue. 1011 static void enqueue_dense_prefix_tasks(GCTaskQueue* q, 1012 uint parallel_gc_threads); 1013 1014 // Add region stealing tasks to the task queue. 1015 static void enqueue_region_stealing_tasks( 1016 GCTaskQueue* q, 1017 ParallelTaskTerminator* terminator_ptr, 1018 uint parallel_gc_threads); 1019 1020 // For debugging only - compacts the old gen serially 1021 static void compact_serial(ParCompactionManager* cm); 1022 1023 // If objects are left in eden after a collection, try to move the boundary 1024 // and absorb them into the old gen. Returns true if eden was emptied. 1025 static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy, 1026 PSYoungGen* young_gen, 1027 PSOldGen* old_gen); 1028 1029 // Reset time since last full gc 1030 static void reset_millis_since_last_gc(); 1031 1032 protected: 1033 #ifdef VALIDATE_MARK_SWEEP 1034 static GrowableArray<void*>* _root_refs_stack; 1035 static GrowableArray<oop> * _live_oops; 1036 static GrowableArray<oop> * _live_oops_moved_to; 1037 static GrowableArray<size_t>* _live_oops_size; 1038 static size_t _live_oops_index; 1039 static size_t _live_oops_index_at_perm; 1040 static GrowableArray<void*>* _other_refs_stack; 1041 static GrowableArray<void*>* _adjusted_pointers; 1042 static bool _pointer_tracking; 1043 static bool _root_tracking; 1044 1045 // The following arrays are saved since the time of the last GC and 1046 // assist in tracking down problems where someone has done an errant 1047 // store into the heap, usually to an oop that wasn't properly 1048 // handleized across a GC. If we crash or otherwise fail before the 1049 // next GC, we can query these arrays to find out the object we had 1050 // intended to do the store to (assuming it is still alive) and the 1051 // offset within that object. Covered under RecordMarkSweepCompaction. 1052 static GrowableArray<HeapWord*> * _cur_gc_live_oops; 1053 static GrowableArray<HeapWord*> * _cur_gc_live_oops_moved_to; 1054 static GrowableArray<size_t>* _cur_gc_live_oops_size; 1055 static GrowableArray<HeapWord*> * _last_gc_live_oops; 1056 static GrowableArray<HeapWord*> * _last_gc_live_oops_moved_to; 1057 static GrowableArray<size_t>* _last_gc_live_oops_size; 1058 #endif 1059 1060 public: 1061 class MarkAndPushClosure: public OopClosure { 1062 private: 1063 ParCompactionManager* _compaction_manager; 1064 public: 1065 MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } 1066 virtual void do_oop(oop* p); 1067 virtual void do_oop(narrowOop* p); 1068 virtual const bool do_nmethods() const { return true; } 1069 }; 1070 1071 PSParallelCompact(); 1072 1073 // Convenient accessor for Universe::heap(). 1074 static ParallelScavengeHeap* gc_heap() { 1075 return (ParallelScavengeHeap*)Universe::heap(); 1076 } 1077 1078 static void invoke(bool maximum_heap_compaction); 1079 static void invoke_no_policy(bool maximum_heap_compaction); 1080 1081 static void post_initialize(); 1082 // Perform initialization for PSParallelCompact that requires 1083 // allocations. This should be called during the VM initialization 1084 // at a pointer where it would be appropriate to return a JNI_ENOMEM 1085 // in the event of a failure. 1086 static bool initialize(); 1087 1088 // Public accessors 1089 static elapsedTimer* accumulated_time() { return &_accumulated_time; } 1090 static unsigned int total_invocations() { return _total_invocations; } 1091 static CollectorCounters* counters() { return _counters; } 1092 1093 // Used to add tasks 1094 static GCTaskManager* const gc_task_manager(); 1095 static klassOop updated_int_array_klass_obj() { 1096 return _updated_int_array_klass_obj; 1097 } 1098 1099 // Marking support 1100 static inline bool mark_obj(oop obj); 1101 // Check mark and maybe push on marking stack 1102 template <class T> static inline void mark_and_push(ParCompactionManager* cm, 1103 T* p); 1104 1105 // Compaction support. 1106 // Return true if p is in the range [beg_addr, end_addr). 1107 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr); 1108 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr); 1109 1110 // Convenience wrappers for per-space data kept in _space_info. 1111 static inline MutableSpace* space(SpaceId space_id); 1112 static inline HeapWord* new_top(SpaceId space_id); 1113 static inline HeapWord* dense_prefix(SpaceId space_id); 1114 static inline ObjectStartArray* start_array(SpaceId space_id); 1115 1116 // Return true if the klass should be updated. 1117 static inline bool should_update_klass(klassOop k); 1118 1119 // Move and update the live objects in the specified space. 1120 static void move_and_update(ParCompactionManager* cm, SpaceId space_id); 1121 1122 // Process the end of the given region range in the dense prefix. 1123 // This includes saving any object not updated. 1124 static void dense_prefix_regions_epilogue(ParCompactionManager* cm, 1125 size_t region_start_index, 1126 size_t region_end_index, 1127 idx_t exiting_object_offset, 1128 idx_t region_offset_start, 1129 idx_t region_offset_end); 1130 1131 // Update a region in the dense prefix. For each live object 1132 // in the region, update it's interior references. For each 1133 // dead object, fill it with deadwood. Dead space at the end 1134 // of a region range will be filled to the start of the next 1135 // live object regardless of the region_index_end. None of the 1136 // objects in the dense prefix move and dead space is dead 1137 // (holds only dead objects that don't need any processing), so 1138 // dead space can be filled in any order. 1139 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm, 1140 SpaceId space_id, 1141 size_t region_index_start, 1142 size_t region_index_end); 1143 1144 // Return the address of the count + 1st live word in the range [beg, end). 1145 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count); 1146 1147 // Return the address of the word to be copied to dest_addr, which must be 1148 // aligned to a region boundary. 1149 static HeapWord* first_src_addr(HeapWord* const dest_addr, 1150 SpaceId src_space_id, 1151 size_t src_region_idx); 1152 1153 // Determine the next source region, set closure.source() to the start of the 1154 // new region return the region index. Parameter end_addr is the address one 1155 // beyond the end of source range just processed. If necessary, switch to a 1156 // new source space and set src_space_id (in-out parameter) and src_space_top 1157 // (out parameter) accordingly. 1158 static size_t next_src_region(MoveAndUpdateClosure& closure, 1159 SpaceId& src_space_id, 1160 HeapWord*& src_space_top, 1161 HeapWord* end_addr); 1162 1163 // Decrement the destination count for each non-empty source region in the 1164 // range [beg_region, region(region_align_up(end_addr))). If the destination 1165 // count for a region goes to 0 and it needs to be filled, enqueue it. 1166 static void decrement_destination_counts(ParCompactionManager* cm, 1167 SpaceId src_space_id, 1168 size_t beg_region, 1169 HeapWord* end_addr); 1170 1171 // Fill a region, copying objects from one or more source regions. 1172 static void fill_region(ParCompactionManager* cm, size_t region_idx); 1173 static void fill_and_update_region(ParCompactionManager* cm, size_t region) { 1174 fill_region(cm, region); 1175 } 1176 1177 // Update the deferred objects in the space. 1178 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id); 1179 1180 // Mark pointer and follow contents. 1181 template <class T> 1182 static inline void mark_and_follow(ParCompactionManager* cm, T* p); 1183 1184 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; } 1185 static ParallelCompactData& summary_data() { return _summary_data; } 1186 1187 static inline void adjust_pointer(oop* p) { adjust_pointer(p, false); } 1188 static inline void adjust_pointer(narrowOop* p) { adjust_pointer(p, false); } 1189 1190 template <class T> 1191 static inline void adjust_pointer(T* p, 1192 HeapWord* beg_addr, 1193 HeapWord* end_addr); 1194 1195 // Reference Processing 1196 static ReferenceProcessor* const ref_processor() { return _ref_processor; } 1197 1198 // Return the SpaceId for the given address. 1199 static SpaceId space_id(HeapWord* addr); 1200 1201 // Time since last full gc (in milliseconds). 1202 static jlong millis_since_last_gc(); 1203 1204 #ifdef VALIDATE_MARK_SWEEP 1205 static void track_adjusted_pointer(void* p, bool isroot); 1206 static void check_adjust_pointer(void* p); 1207 static void track_interior_pointers(oop obj); 1208 static void check_interior_pointers(); 1209 1210 static void reset_live_oop_tracking(bool at_perm); 1211 static void register_live_oop(oop p, size_t size); 1212 static void validate_live_oop(oop p, size_t size); 1213 static void live_oop_moved_to(HeapWord* q, size_t size, HeapWord* compaction_top); 1214 static void compaction_complete(); 1215 1216 // Querying operation of RecordMarkSweepCompaction results. 1217 // Finds and prints the current base oop and offset for a word 1218 // within an oop that was live during the last GC. Helpful for 1219 // tracking down heap stomps. 1220 static void print_new_location_of_heap_address(HeapWord* q); 1221 #endif // #ifdef VALIDATE_MARK_SWEEP 1222 1223 // Call backs for class unloading 1224 // Update subklass/sibling/implementor links at end of marking. 1225 static void revisit_weak_klass_link(ParCompactionManager* cm, Klass* k); 1226 1227 #ifndef PRODUCT 1228 // Debugging support. 1229 static const char* space_names[last_space_id]; 1230 static void print_region_ranges(); 1231 static void print_dense_prefix_stats(const char* const algorithm, 1232 const SpaceId id, 1233 const bool maximum_compaction, 1234 HeapWord* const addr); 1235 static void summary_phase_msg(SpaceId dst_space_id, 1236 HeapWord* dst_beg, HeapWord* dst_end, 1237 SpaceId src_space_id, 1238 HeapWord* src_beg, HeapWord* src_end); 1239 #endif // #ifndef PRODUCT 1240 1241 #ifdef ASSERT 1242 // Sanity check the new location of a word in the heap. 1243 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr); 1244 // Verify that all the regions have been emptied. 1245 static void verify_complete(SpaceId space_id); 1246 #endif // #ifdef ASSERT 1247 }; 1248 1249 inline bool PSParallelCompact::mark_obj(oop obj) { 1250 const int obj_size = obj->size(); 1251 if (mark_bitmap()->mark_obj(obj, obj_size)) { 1252 _summary_data.add_obj(obj, obj_size); 1253 return true; 1254 } else { 1255 return false; 1256 } 1257 } 1258 1259 template <class T> 1260 inline void PSParallelCompact::follow_root(ParCompactionManager* cm, T* p) { 1261 assert(!Universe::heap()->is_in_reserved(p), 1262 "roots shouldn't be things within the heap"); 1263 #ifdef VALIDATE_MARK_SWEEP 1264 if (ValidateMarkSweep) { 1265 guarantee(!_root_refs_stack->contains(p), "should only be in here once"); 1266 _root_refs_stack->push(p); 1267 } 1268 #endif 1269 T heap_oop = oopDesc::load_heap_oop(p); 1270 if (!oopDesc::is_null(heap_oop)) { 1271 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); 1272 if (mark_bitmap()->is_unmarked(obj)) { 1273 if (mark_obj(obj)) { 1274 obj->follow_contents(cm); 1275 } 1276 } 1277 } 1278 follow_stack(cm); 1279 } 1280 1281 template <class T> 1282 inline void PSParallelCompact::mark_and_follow(ParCompactionManager* cm, 1283 T* p) { 1284 T heap_oop = oopDesc::load_heap_oop(p); 1285 if (!oopDesc::is_null(heap_oop)) { 1286 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); 1287 if (mark_bitmap()->is_unmarked(obj)) { 1288 if (mark_obj(obj)) { 1289 obj->follow_contents(cm); 1290 } 1291 } 1292 } 1293 } 1294 1295 template <class T> 1296 inline void PSParallelCompact::mark_and_push(ParCompactionManager* cm, T* p) { 1297 T heap_oop = oopDesc::load_heap_oop(p); 1298 if (!oopDesc::is_null(heap_oop)) { 1299 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); 1300 if (mark_bitmap()->is_unmarked(obj)) { 1301 if (mark_obj(obj)) { 1302 // This thread marked the object and owns the subsequent processing of it. 1303 cm->save_for_scanning(obj); 1304 } 1305 } 1306 } 1307 } 1308 1309 template <class T> 1310 inline void PSParallelCompact::adjust_pointer(T* p, bool isroot) { 1311 T heap_oop = oopDesc::load_heap_oop(p); 1312 if (!oopDesc::is_null(heap_oop)) { 1313 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); 1314 oop new_obj = (oop)summary_data().calc_new_pointer(obj); 1315 assert(new_obj != NULL || // is forwarding ptr? 1316 obj->is_shared(), // never forwarded? 1317 "should be forwarded"); 1318 // Just always do the update unconditionally? 1319 if (new_obj != NULL) { 1320 assert(Universe::heap()->is_in_reserved(new_obj), 1321 "should be in object space"); 1322 oopDesc::encode_store_heap_oop_not_null(p, new_obj); 1323 } 1324 } 1325 VALIDATE_MARK_SWEEP_ONLY(track_adjusted_pointer(p, isroot)); 1326 } 1327 1328 template <class T> 1329 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) { 1330 #ifdef VALIDATE_MARK_SWEEP 1331 if (ValidateMarkSweep) { 1332 if (!Universe::heap()->is_in_reserved(p)) { 1333 _root_refs_stack->push(p); 1334 } else { 1335 _other_refs_stack->push(p); 1336 } 1337 } 1338 #endif 1339 mark_and_push(_compaction_manager, p); 1340 } 1341 1342 inline bool PSParallelCompact::print_phases() { 1343 return _print_phases; 1344 } 1345 1346 inline double PSParallelCompact::normal_distribution(double density) { 1347 assert(_dwl_initialized, "uninitialized"); 1348 const double squared_term = (density - _dwl_mean) / _dwl_std_dev; 1349 return _dwl_first_term * exp(-0.5 * squared_term * squared_term); 1350 } 1351 1352 inline bool 1353 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region, 1354 idx_t bit) 1355 { 1356 assert(bit > 0, "cannot call this for the first bit/region"); 1357 assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit), 1358 "sanity check"); 1359 1360 // Dead space crosses the boundary if (1) a partial object does not extend 1361 // onto the region, (2) an object does not start at the beginning of the 1362 // region, and (3) an object does not end at the end of the prior region. 1363 return region->partial_obj_size() == 0 && 1364 !_mark_bitmap.is_obj_beg(bit) && 1365 !_mark_bitmap.is_obj_end(bit - 1); 1366 } 1367 1368 inline bool 1369 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) { 1370 return p >= beg_addr && p < end_addr; 1371 } 1372 1373 inline bool 1374 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) { 1375 return is_in((HeapWord*)p, beg_addr, end_addr); 1376 } 1377 1378 inline MutableSpace* PSParallelCompact::space(SpaceId id) { 1379 assert(id < last_space_id, "id out of range"); 1380 return _space_info[id].space(); 1381 } 1382 1383 inline HeapWord* PSParallelCompact::new_top(SpaceId id) { 1384 assert(id < last_space_id, "id out of range"); 1385 return _space_info[id].new_top(); 1386 } 1387 1388 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) { 1389 assert(id < last_space_id, "id out of range"); 1390 return _space_info[id].dense_prefix(); 1391 } 1392 1393 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) { 1394 assert(id < last_space_id, "id out of range"); 1395 return _space_info[id].start_array(); 1396 } 1397 1398 inline bool PSParallelCompact::should_update_klass(klassOop k) { 1399 return ((HeapWord*) k) >= dense_prefix(perm_space_id); 1400 } 1401 1402 template <class T> 1403 inline void PSParallelCompact::adjust_pointer(T* p, 1404 HeapWord* beg_addr, 1405 HeapWord* end_addr) { 1406 if (is_in((HeapWord*)p, beg_addr, end_addr)) { 1407 adjust_pointer(p); 1408 } 1409 } 1410 1411 #ifdef ASSERT 1412 inline void 1413 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr) 1414 { 1415 assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr), 1416 "must move left or to a different space"); 1417 } 1418 #endif // ASSERT 1419 1420 class MoveAndUpdateClosure: public ParMarkBitMapClosure { 1421 public: 1422 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, 1423 ObjectStartArray* start_array, 1424 HeapWord* destination, size_t words); 1425 1426 // Accessors. 1427 HeapWord* destination() const { return _destination; } 1428 1429 // If the object will fit (size <= words_remaining()), copy it to the current 1430 // destination, update the interior oops and the start array and return either 1431 // full (if the closure is full) or incomplete. If the object will not fit, 1432 // return would_overflow. 1433 virtual IterationStatus do_addr(HeapWord* addr, size_t size); 1434 1435 // Copy enough words to fill this closure, starting at source(). Interior 1436 // oops and the start array are not updated. Return full. 1437 IterationStatus copy_until_full(); 1438 1439 // Copy enough words to fill this closure or to the end of an object, 1440 // whichever is smaller, starting at source(). Interior oops and the start 1441 // array are not updated. 1442 void copy_partial_obj(); 1443 1444 protected: 1445 // Update variables to indicate that word_count words were processed. 1446 inline void update_state(size_t word_count); 1447 1448 protected: 1449 ObjectStartArray* const _start_array; 1450 HeapWord* _destination; // Next addr to be written. 1451 }; 1452 1453 inline 1454 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap, 1455 ParCompactionManager* cm, 1456 ObjectStartArray* start_array, 1457 HeapWord* destination, 1458 size_t words) : 1459 ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array) 1460 { 1461 _destination = destination; 1462 } 1463 1464 inline void MoveAndUpdateClosure::update_state(size_t words) 1465 { 1466 decrement_words_remaining(words); 1467 _source += words; 1468 _destination += words; 1469 } 1470 1471 class UpdateOnlyClosure: public ParMarkBitMapClosure { 1472 private: 1473 const PSParallelCompact::SpaceId _space_id; 1474 ObjectStartArray* const _start_array; 1475 1476 public: 1477 UpdateOnlyClosure(ParMarkBitMap* mbm, 1478 ParCompactionManager* cm, 1479 PSParallelCompact::SpaceId space_id); 1480 1481 // Update the object. 1482 virtual IterationStatus do_addr(HeapWord* addr, size_t words); 1483 1484 inline void do_addr(HeapWord* addr); 1485 }; 1486 1487 inline void UpdateOnlyClosure::do_addr(HeapWord* addr) 1488 { 1489 _start_array->allocate_block(addr); 1490 oop(addr)->update_contents(compaction_manager()); 1491 } 1492 1493 class FillClosure: public ParMarkBitMapClosure 1494 { 1495 public: 1496 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) : 1497 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm), 1498 _start_array(PSParallelCompact::start_array(space_id)) 1499 { 1500 assert(space_id == PSParallelCompact::perm_space_id || 1501 space_id == PSParallelCompact::old_space_id, 1502 "cannot use FillClosure in the young gen"); 1503 } 1504 1505 virtual IterationStatus do_addr(HeapWord* addr, size_t size) { 1506 CollectedHeap::fill_with_objects(addr, size); 1507 HeapWord* const end = addr + size; 1508 do { 1509 _start_array->allocate_block(addr); 1510 addr += oop(addr)->size(); 1511 } while (addr < end); 1512 return ParMarkBitMap::incomplete; 1513 } 1514 1515 private: 1516 ObjectStartArray* const _start_array; 1517 };