1 /* 2 * Copyright (c) 2001, 2017, 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 #include "precompiled.hpp" 26 #include "gc/g1/concurrentG1Refine.hpp" 27 #include "gc/g1/dirtyCardQueue.hpp" 28 #include "gc/g1/g1BlockOffsetTable.inline.hpp" 29 #include "gc/g1/g1CollectedHeap.inline.hpp" 30 #include "gc/g1/g1FromCardCache.hpp" 31 #include "gc/g1/g1GCPhaseTimes.hpp" 32 #include "gc/g1/g1HotCardCache.hpp" 33 #include "gc/g1/g1OopClosures.inline.hpp" 34 #include "gc/g1/g1RemSet.inline.hpp" 35 #include "gc/g1/g1SATBCardTableModRefBS.inline.hpp" 36 #include "gc/g1/heapRegion.inline.hpp" 37 #include "gc/g1/heapRegionManager.inline.hpp" 38 #include "gc/g1/heapRegionRemSet.hpp" 39 #include "gc/shared/gcTraceTime.inline.hpp" 40 #include "memory/iterator.hpp" 41 #include "memory/resourceArea.hpp" 42 #include "oops/oop.inline.hpp" 43 #include "utilities/globalDefinitions.hpp" 44 #include "utilities/intHisto.hpp" 45 #include "utilities/stack.inline.hpp" 46 47 // Collects information about the overall remembered set scan progress during an evacuation. 48 class G1RemSetScanState : public CHeapObj<mtGC> { 49 private: 50 class G1ClearCardTableTask : public AbstractGangTask { 51 G1CollectedHeap* _g1h; 52 uint* _dirty_region_list; 53 size_t _num_dirty_regions; 54 size_t _chunk_length; 55 56 size_t volatile _cur_dirty_regions; 57 public: 58 G1ClearCardTableTask(G1CollectedHeap* g1h, 59 uint* dirty_region_list, 60 size_t num_dirty_regions, 61 size_t chunk_length) : 62 AbstractGangTask("G1 Clear Card Table Task"), 63 _g1h(g1h), 64 _dirty_region_list(dirty_region_list), 65 _num_dirty_regions(num_dirty_regions), 66 _chunk_length(chunk_length), 67 _cur_dirty_regions(0) { 68 69 assert(chunk_length > 0, "must be"); 70 } 71 72 static size_t chunk_size() { return M; } 73 74 void work(uint worker_id) { 75 G1SATBCardTableModRefBS* ct_bs = _g1h->g1_barrier_set(); 76 77 while (_cur_dirty_regions < _num_dirty_regions) { 78 size_t next = Atomic::add(_chunk_length, &_cur_dirty_regions) - _chunk_length; 79 size_t max = MIN2(next + _chunk_length, _num_dirty_regions); 80 81 for (size_t i = next; i < max; i++) { 82 HeapRegion* r = _g1h->region_at(_dirty_region_list[i]); 83 if (!r->is_survivor()) { 84 ct_bs->clear(MemRegion(r->bottom(), r->end())); 85 } 86 } 87 } 88 } 89 }; 90 91 size_t _max_regions; 92 93 // Scan progress for the remembered set of a single region. Transitions from 94 // Unclaimed -> Claimed -> Complete. 95 // At each of the transitions the thread that does the transition needs to perform 96 // some special action once. This is the reason for the extra "Claimed" state. 97 typedef jint G1RemsetIterState; 98 99 static const G1RemsetIterState Unclaimed = 0; // The remembered set has not been scanned yet. 100 static const G1RemsetIterState Claimed = 1; // The remembered set is currently being scanned. 101 static const G1RemsetIterState Complete = 2; // The remembered set has been completely scanned. 102 103 G1RemsetIterState volatile* _iter_states; 104 // The current location where the next thread should continue scanning in a region's 105 // remembered set. 106 size_t volatile* _iter_claims; 107 108 // Temporary buffer holding the regions we used to store remembered set scan duplicate 109 // information. These are also called "dirty". Valid entries are from [0.._cur_dirty_region) 110 uint* _dirty_region_buffer; 111 112 typedef jbyte IsDirtyRegionState; 113 static const IsDirtyRegionState Clean = 0; 114 static const IsDirtyRegionState Dirty = 1; 115 // Holds a flag for every region whether it is in the _dirty_region_buffer already 116 // to avoid duplicates. Uses jbyte since there are no atomic instructions for bools. 117 IsDirtyRegionState* _in_dirty_region_buffer; 118 size_t _cur_dirty_region; 119 120 // Creates a snapshot of the current _top values at the start of collection to 121 // filter out card marks that we do not want to scan. 122 class G1ResetScanTopClosure : public HeapRegionClosure { 123 private: 124 HeapWord** _scan_top; 125 public: 126 G1ResetScanTopClosure(HeapWord** scan_top) : _scan_top(scan_top) { } 127 128 virtual bool doHeapRegion(HeapRegion* r) { 129 uint hrm_index = r->hrm_index(); 130 if (!r->in_collection_set() && r->is_old_or_humongous()) { 131 _scan_top[hrm_index] = r->top(); 132 } else { 133 _scan_top[hrm_index] = r->bottom(); 134 } 135 return false; 136 } 137 }; 138 139 // For each region, contains the maximum top() value to be used during this garbage 140 // collection. Subsumes common checks like filtering out everything but old and 141 // humongous regions outside the collection set. 142 // This is valid because we are not interested in scanning stray remembered set 143 // entries from free or archive regions. 144 HeapWord** _scan_top; 145 public: 146 G1RemSetScanState() : 147 _max_regions(0), 148 _iter_states(NULL), 149 _iter_claims(NULL), 150 _dirty_region_buffer(NULL), 151 _in_dirty_region_buffer(NULL), 152 _cur_dirty_region(0), 153 _scan_top(NULL) { 154 } 155 156 ~G1RemSetScanState() { 157 if (_iter_states != NULL) { 158 FREE_C_HEAP_ARRAY(G1RemsetIterState, _iter_states); 159 } 160 if (_iter_claims != NULL) { 161 FREE_C_HEAP_ARRAY(size_t, _iter_claims); 162 } 163 if (_dirty_region_buffer != NULL) { 164 FREE_C_HEAP_ARRAY(uint, _dirty_region_buffer); 165 } 166 if (_in_dirty_region_buffer != NULL) { 167 FREE_C_HEAP_ARRAY(IsDirtyRegionState, _in_dirty_region_buffer); 168 } 169 if (_scan_top != NULL) { 170 FREE_C_HEAP_ARRAY(HeapWord*, _scan_top); 171 } 172 } 173 174 void initialize(uint max_regions) { 175 assert(_iter_states == NULL, "Must not be initialized twice"); 176 assert(_iter_claims == NULL, "Must not be initialized twice"); 177 _max_regions = max_regions; 178 _iter_states = NEW_C_HEAP_ARRAY(G1RemsetIterState, max_regions, mtGC); 179 _iter_claims = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC); 180 _dirty_region_buffer = NEW_C_HEAP_ARRAY(uint, max_regions, mtGC); 181 _in_dirty_region_buffer = NEW_C_HEAP_ARRAY(IsDirtyRegionState, max_regions, mtGC); 182 _scan_top = NEW_C_HEAP_ARRAY(HeapWord*, max_regions, mtGC); 183 } 184 185 void reset() { 186 for (uint i = 0; i < _max_regions; i++) { 187 _iter_states[i] = Unclaimed; 188 } 189 190 G1ResetScanTopClosure cl(_scan_top); 191 G1CollectedHeap::heap()->heap_region_iterate(&cl); 192 193 memset((void*)_iter_claims, 0, _max_regions * sizeof(size_t)); 194 memset(_in_dirty_region_buffer, Clean, _max_regions * sizeof(IsDirtyRegionState)); 195 _cur_dirty_region = 0; 196 } 197 198 // Attempt to claim the remembered set of the region for iteration. Returns true 199 // if this call caused the transition from Unclaimed to Claimed. 200 inline bool claim_iter(uint region) { 201 assert(region < _max_regions, "Tried to access invalid region %u", region); 202 if (_iter_states[region] != Unclaimed) { 203 return false; 204 } 205 jint res = Atomic::cmpxchg(Claimed, (jint*)(&_iter_states[region]), Unclaimed); 206 return (res == Unclaimed); 207 } 208 209 // Try to atomically sets the iteration state to "complete". Returns true for the 210 // thread that caused the transition. 211 inline bool set_iter_complete(uint region) { 212 if (iter_is_complete(region)) { 213 return false; 214 } 215 jint res = Atomic::cmpxchg(Complete, (jint*)(&_iter_states[region]), Claimed); 216 return (res == Claimed); 217 } 218 219 // Returns true if the region's iteration is complete. 220 inline bool iter_is_complete(uint region) const { 221 assert(region < _max_regions, "Tried to access invalid region %u", region); 222 return _iter_states[region] == Complete; 223 } 224 225 // The current position within the remembered set of the given region. 226 inline size_t iter_claimed(uint region) const { 227 assert(region < _max_regions, "Tried to access invalid region %u", region); 228 return _iter_claims[region]; 229 } 230 231 // Claim the next block of cards within the remembered set of the region with 232 // step size. 233 inline size_t iter_claimed_next(uint region, size_t step) { 234 return Atomic::add(step, &_iter_claims[region]) - step; 235 } 236 237 void add_dirty_region(uint region) { 238 if (_in_dirty_region_buffer[region] == Dirty) { 239 return; 240 } 241 242 bool marked_as_dirty = Atomic::cmpxchg(Dirty, &_in_dirty_region_buffer[region], Clean) == Clean; 243 if (marked_as_dirty) { 244 size_t allocated = Atomic::add(1, &_cur_dirty_region) - 1; 245 _dirty_region_buffer[allocated] = region; 246 } 247 } 248 249 HeapWord* scan_top(uint region_idx) const { 250 return _scan_top[region_idx]; 251 } 252 253 // Clear the card table of "dirty" regions. 254 void clear_card_table(WorkGang* workers) { 255 if (_cur_dirty_region == 0) { 256 return; 257 } 258 259 size_t const num_chunks = align_size_up(_cur_dirty_region * HeapRegion::CardsPerRegion, G1ClearCardTableTask::chunk_size()) / G1ClearCardTableTask::chunk_size(); 260 uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers()); 261 size_t const chunk_length = G1ClearCardTableTask::chunk_size() / HeapRegion::CardsPerRegion; 262 263 // Iterate over the dirty cards region list. 264 G1ClearCardTableTask cl(G1CollectedHeap::heap(), _dirty_region_buffer, _cur_dirty_region, chunk_length); 265 266 log_debug(gc, ergo)("Running %s using %u workers for " SIZE_FORMAT " " 267 "units of work for " SIZE_FORMAT " regions.", 268 cl.name(), num_workers, num_chunks, _cur_dirty_region); 269 workers->run_task(&cl, num_workers); 270 271 #ifndef PRODUCT 272 // Need to synchronize with concurrent cleanup since it needs to 273 // finish its card table clearing before we can verify. 274 G1CollectedHeap::heap()->wait_while_free_regions_coming(); 275 G1CollectedHeap::heap()->verifier()->verify_card_table_cleanup(); 276 #endif 277 } 278 }; 279 280 G1RemSet::G1RemSet(G1CollectedHeap* g1, 281 CardTableModRefBS* ct_bs, 282 G1HotCardCache* hot_card_cache) : 283 _g1(g1), 284 _scan_state(new G1RemSetScanState()), 285 _conc_refine_cards(0), 286 _ct_bs(ct_bs), 287 _g1p(_g1->g1_policy()), 288 _hot_card_cache(hot_card_cache), 289 _prev_period_summary(), 290 _into_cset_dirty_card_queue_set(false) 291 { 292 if (log_is_enabled(Trace, gc, remset)) { 293 _prev_period_summary.initialize(this); 294 } 295 // Initialize the card queue set used to hold cards containing 296 // references into the collection set. 297 _into_cset_dirty_card_queue_set.initialize(NULL, // Should never be called by the Java code 298 DirtyCardQ_CBL_mon, 299 DirtyCardQ_FL_lock, 300 -1, // never trigger processing 301 -1, // no limit on length 302 Shared_DirtyCardQ_lock, 303 &JavaThread::dirty_card_queue_set()); 304 } 305 306 G1RemSet::~G1RemSet() { 307 if (_scan_state != NULL) { 308 delete _scan_state; 309 } 310 } 311 312 uint G1RemSet::num_par_rem_sets() { 313 return MAX2(DirtyCardQueueSet::num_par_ids() + ConcurrentG1Refine::thread_num(), ParallelGCThreads); 314 } 315 316 void G1RemSet::initialize(size_t capacity, uint max_regions) { 317 G1FromCardCache::initialize(num_par_rem_sets(), max_regions); 318 _scan_state->initialize(max_regions); 319 { 320 GCTraceTime(Debug, gc, marking)("Initialize Card Live Data"); 321 _card_live_data.initialize(capacity, max_regions); 322 } 323 if (G1PretouchAuxiliaryMemory) { 324 GCTraceTime(Debug, gc, marking)("Pre-Touch Card Live Data"); 325 _card_live_data.pretouch(); 326 } 327 } 328 329 G1ScanRSClosure::G1ScanRSClosure(G1RemSetScanState* scan_state, 330 G1ParPushHeapRSClosure* push_heap_cl, 331 CodeBlobClosure* code_root_cl, 332 uint worker_i) : 333 _scan_state(scan_state), 334 _push_heap_cl(push_heap_cl), 335 _code_root_cl(code_root_cl), 336 _strong_code_root_scan_time_sec(0.0), 337 _cards_claimed(0), 338 _cards_scanned(0), 339 _cards_skipped(0), 340 _worker_i(worker_i) { 341 _g1h = G1CollectedHeap::heap(); 342 _bot = _g1h->bot(); 343 _ct_bs = _g1h->g1_barrier_set(); 344 _block_size = MAX2<size_t>(G1RSetScanBlockSize, 1); 345 } 346 347 void G1ScanRSClosure::scan_card(size_t index, HeapWord* card_start, HeapRegion *r) { 348 MemRegion card_region(card_start, BOTConstants::N_words); 349 MemRegion pre_gc_allocated(r->bottom(), _scan_state->scan_top(r->hrm_index())); 350 MemRegion mr = pre_gc_allocated.intersection(card_region); 351 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) { 352 // We make the card as "claimed" lazily (so races are possible 353 // but they're benign), which reduces the number of duplicate 354 // scans (the rsets of the regions in the cset can intersect). 355 _ct_bs->set_card_claimed(index); 356 _push_heap_cl->set_region(r); 357 r->oops_on_card_seq_iterate_careful<true>(mr, _push_heap_cl); 358 _cards_scanned++; 359 } 360 } 361 362 void G1ScanRSClosure::scan_strong_code_roots(HeapRegion* r) { 363 double scan_start = os::elapsedTime(); 364 r->strong_code_roots_do(_code_root_cl); 365 _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start); 366 } 367 368 bool G1ScanRSClosure::doHeapRegion(HeapRegion* r) { 369 assert(r->in_collection_set(), "should only be called on elements of CS."); 370 uint region_idx = r->hrm_index(); 371 372 if (_scan_state->iter_is_complete(region_idx)) { 373 return false; 374 } 375 if (_scan_state->claim_iter(region_idx)) { 376 // If we ever free the collection set concurrently, we should also 377 // clear the card table concurrently therefore we won't need to 378 // add regions of the collection set to the dirty cards region. 379 _scan_state->add_dirty_region(region_idx); 380 } 381 382 HeapRegionRemSetIterator iter(r->rem_set()); 383 size_t card_index; 384 385 // We claim cards in block so as to reduce the contention. The block size is determined by 386 // the G1RSetScanBlockSize parameter. 387 size_t claimed_card_block = _scan_state->iter_claimed_next(region_idx, _block_size); 388 for (size_t current_card = 0; iter.has_next(card_index); current_card++) { 389 if (current_card >= claimed_card_block + _block_size) { 390 claimed_card_block = _scan_state->iter_claimed_next(region_idx, _block_size); 391 } 392 if (current_card < claimed_card_block) { 393 _cards_skipped++; 394 continue; 395 } 396 HeapWord* card_start = _g1h->bot()->address_for_index(card_index); 397 398 HeapRegion* card_region = _g1h->heap_region_containing(card_start); 399 _cards_claimed++; 400 401 _scan_state->add_dirty_region(card_region->hrm_index()); 402 403 // If the card is dirty, then we will scan it during updateRS. 404 if (!card_region->in_collection_set() && 405 !_ct_bs->is_card_dirty(card_index)) { 406 scan_card(card_index, card_start, card_region); 407 } 408 } 409 if (_scan_state->set_iter_complete(region_idx)) { 410 // Scan the strong code root list attached to the current region 411 scan_strong_code_roots(r); 412 } 413 return false; 414 } 415 416 void G1RemSet::scan_rem_set(G1ParPushHeapRSClosure* oops_in_heap_closure, 417 CodeBlobClosure* heap_region_codeblobs, 418 uint worker_i) { 419 double rs_time_start = os::elapsedTime(); 420 421 G1ScanRSClosure cl(_scan_state, oops_in_heap_closure, heap_region_codeblobs, worker_i); 422 _g1->collection_set_iterate_from(&cl, worker_i); 423 424 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) - 425 cl.strong_code_root_scan_time_sec(); 426 427 G1GCPhaseTimes* p = _g1p->phase_times(); 428 429 p->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec); 430 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_scanned(), G1GCPhaseTimes::ScannedCards); 431 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_claimed(), G1GCPhaseTimes::ClaimedCards); 432 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_skipped(), G1GCPhaseTimes::SkippedCards); 433 434 p->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, cl.strong_code_root_scan_time_sec()); 435 } 436 437 // Closure used for updating RSets and recording references that 438 // point into the collection set. Only called during an 439 // evacuation pause. 440 441 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure { 442 G1RemSet* _g1rs; 443 DirtyCardQueue* _into_cset_dcq; 444 G1ParPushHeapRSClosure* _cl; 445 public: 446 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h, 447 DirtyCardQueue* into_cset_dcq, 448 G1ParPushHeapRSClosure* cl) : 449 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq), _cl(cl) 450 {} 451 452 bool do_card_ptr(jbyte* card_ptr, uint worker_i) { 453 // The only time we care about recording cards that 454 // contain references that point into the collection set 455 // is during RSet updating within an evacuation pause. 456 // In this case worker_i should be the id of a GC worker thread. 457 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); 458 assert(worker_i < ParallelGCThreads, "should be a GC worker"); 459 460 if (_g1rs->refine_card_during_gc(card_ptr, worker_i, _cl)) { 461 // 'card_ptr' contains references that point into the collection 462 // set. We need to record the card in the DCQS 463 // (_into_cset_dirty_card_queue_set) 464 // that's used for that purpose. 465 // 466 // Enqueue the card 467 _into_cset_dcq->enqueue(card_ptr); 468 } 469 return true; 470 } 471 }; 472 473 void G1RemSet::update_rem_set(DirtyCardQueue* into_cset_dcq, 474 G1ParPushHeapRSClosure* oops_in_heap_closure, 475 uint worker_i) { 476 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq, oops_in_heap_closure); 477 478 G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i); 479 if (G1HotCardCache::default_use_cache()) { 480 // Apply the closure to the entries of the hot card cache. 481 G1GCParPhaseTimesTracker y(_g1p->phase_times(), G1GCPhaseTimes::ScanHCC, worker_i); 482 _g1->iterate_hcc_closure(&into_cset_update_rs_cl, worker_i); 483 } 484 // Apply the closure to all remaining log entries. 485 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, worker_i); 486 } 487 488 void G1RemSet::cleanupHRRS() { 489 HeapRegionRemSet::cleanup(); 490 } 491 492 void G1RemSet::oops_into_collection_set_do(G1ParPushHeapRSClosure* cl, 493 CodeBlobClosure* heap_region_codeblobs, 494 uint worker_i) { 495 // A DirtyCardQueue that is used to hold cards containing references 496 // that point into the collection set. This DCQ is associated with a 497 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal 498 // circumstances (i.e. the pause successfully completes), these cards 499 // are just discarded (there's no need to update the RSets of regions 500 // that were in the collection set - after the pause these regions 501 // are wholly 'free' of live objects. In the event of an evacuation 502 // failure the cards/buffers in this queue set are passed to the 503 // DirtyCardQueueSet that is used to manage RSet updates 504 DirtyCardQueue into_cset_dcq(&_into_cset_dirty_card_queue_set); 505 506 update_rem_set(&into_cset_dcq, cl, worker_i); 507 scan_rem_set(cl, heap_region_codeblobs, worker_i);; 508 } 509 510 void G1RemSet::prepare_for_oops_into_collection_set_do() { 511 _g1->set_refine_cte_cl_concurrency(false); 512 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 513 dcqs.concatenate_logs(); 514 515 _scan_state->reset(); 516 } 517 518 void G1RemSet::cleanup_after_oops_into_collection_set_do() { 519 G1GCPhaseTimes* phase_times = _g1->g1_policy()->phase_times(); 520 // Cleanup after copy 521 _g1->set_refine_cte_cl_concurrency(true); 522 523 // Set all cards back to clean. 524 double start = os::elapsedTime(); 525 _scan_state->clear_card_table(_g1->workers()); 526 phase_times->record_clear_ct_time((os::elapsedTime() - start) * 1000.0); 527 528 DirtyCardQueueSet& into_cset_dcqs = _into_cset_dirty_card_queue_set; 529 530 if (_g1->evacuation_failed()) { 531 double restore_remembered_set_start = os::elapsedTime(); 532 533 // Restore remembered sets for the regions pointing into the collection set. 534 // We just need to transfer the completed buffers from the DirtyCardQueueSet 535 // used to hold cards that contain references that point into the collection set 536 // to the DCQS used to hold the deferred RS updates. 537 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs); 538 phase_times->record_evac_fail_restore_remsets((os::elapsedTime() - restore_remembered_set_start) * 1000.0); 539 } 540 541 // Free any completed buffers in the DirtyCardQueueSet used to hold cards 542 // which contain references that point into the collection. 543 _into_cset_dirty_card_queue_set.clear(); 544 assert(_into_cset_dirty_card_queue_set.completed_buffers_num() == 0, 545 "all buffers should be freed"); 546 _into_cset_dirty_card_queue_set.clear_n_completed_buffers(); 547 } 548 549 class G1ScrubRSClosure: public HeapRegionClosure { 550 G1CollectedHeap* _g1h; 551 G1CardLiveData* _live_data; 552 public: 553 G1ScrubRSClosure(G1CardLiveData* live_data) : 554 _g1h(G1CollectedHeap::heap()), 555 _live_data(live_data) { } 556 557 bool doHeapRegion(HeapRegion* r) { 558 if (!r->is_continues_humongous()) { 559 r->rem_set()->scrub(_live_data); 560 } 561 return false; 562 } 563 }; 564 565 void G1RemSet::scrub(uint worker_num, HeapRegionClaimer *hrclaimer) { 566 G1ScrubRSClosure scrub_cl(&_card_live_data); 567 _g1->heap_region_par_iterate(&scrub_cl, worker_num, hrclaimer); 568 } 569 570 inline void check_card_ptr(jbyte* card_ptr, CardTableModRefBS* ct_bs) { 571 #ifdef ASSERT 572 G1CollectedHeap* g1 = G1CollectedHeap::heap(); 573 assert(g1->is_in_exact(ct_bs->addr_for(card_ptr)), 574 "Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap", 575 p2i(card_ptr), 576 ct_bs->index_for(ct_bs->addr_for(card_ptr)), 577 p2i(ct_bs->addr_for(card_ptr)), 578 g1->addr_to_region(ct_bs->addr_for(card_ptr))); 579 #endif 580 } 581 582 G1UpdateRSOrPushRefOopClosure::G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h, 583 G1ParPushHeapRSClosure* push_ref_cl, 584 bool record_refs_into_cset, 585 uint worker_i) : 586 _g1(g1h), 587 _from(NULL), 588 _record_refs_into_cset(record_refs_into_cset), 589 _has_refs_into_cset(false), 590 _push_ref_cl(push_ref_cl), 591 _worker_i(worker_i) { } 592 593 void G1RemSet::refine_card_concurrently(jbyte* card_ptr, 594 uint worker_i) { 595 assert(!_g1->is_gc_active(), "Only call concurrently"); 596 597 check_card_ptr(card_ptr, _ct_bs); 598 599 // If the card is no longer dirty, nothing to do. 600 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 601 return; 602 } 603 604 // Construct the region representing the card. 605 HeapWord* start = _ct_bs->addr_for(card_ptr); 606 // And find the region containing it. 607 HeapRegion* r = _g1->heap_region_containing(start); 608 609 // This check is needed for some uncommon cases where we should 610 // ignore the card. 611 // 612 // The region could be young. Cards for young regions are 613 // distinctly marked (set to g1_young_gen), so the post-barrier will 614 // filter them out. However, that marking is performed 615 // concurrently. A write to a young object could occur before the 616 // card has been marked young, slipping past the filter. 617 // 618 // The card could be stale, because the region has been freed since 619 // the card was recorded. In this case the region type could be 620 // anything. If (still) free or (reallocated) young, just ignore 621 // it. If (reallocated) old or humongous, the later card trimming 622 // and additional checks in iteration may detect staleness. At 623 // worst, we end up processing a stale card unnecessarily. 624 // 625 // In the normal (non-stale) case, the synchronization between the 626 // enqueueing of the card and processing it here will have ensured 627 // we see the up-to-date region type here. 628 if (!r->is_old_or_humongous()) { 629 return; 630 } 631 632 // While we are processing RSet buffers during the collection, we 633 // actually don't want to scan any cards on the collection set, 634 // since we don't want to update remembered sets with entries that 635 // point into the collection set, given that live objects from the 636 // collection set are about to move and such entries will be stale 637 // very soon. This change also deals with a reliability issue which 638 // involves scanning a card in the collection set and coming across 639 // an array that was being chunked and looking malformed. Note, 640 // however, that if evacuation fails, we have to scan any objects 641 // that were not moved and create any missing entries. 642 if (r->in_collection_set()) { 643 return; 644 } 645 646 // The result from the hot card cache insert call is either: 647 // * pointer to the current card 648 // (implying that the current card is not 'hot'), 649 // * null 650 // (meaning we had inserted the card ptr into the "hot" card cache, 651 // which had some headroom), 652 // * a pointer to a "hot" card that was evicted from the "hot" cache. 653 // 654 655 if (_hot_card_cache->use_cache()) { 656 assert(!SafepointSynchronize::is_at_safepoint(), "sanity"); 657 658 const jbyte* orig_card_ptr = card_ptr; 659 card_ptr = _hot_card_cache->insert(card_ptr); 660 if (card_ptr == NULL) { 661 // There was no eviction. Nothing to do. 662 return; 663 } else if (card_ptr != orig_card_ptr) { 664 // Original card was inserted and an old card was evicted. 665 start = _ct_bs->addr_for(card_ptr); 666 r = _g1->heap_region_containing(start); 667 668 // Check whether the region formerly in the cache should be 669 // ignored, as discussed earlier for the original card. The 670 // region could have been freed while in the cache. The cset is 671 // not relevant here, since we're in concurrent phase. 672 if (!r->is_old_or_humongous()) { 673 return; 674 } 675 } // Else we still have the original card. 676 } 677 678 // Trim the region designated by the card to what's been allocated 679 // in the region. The card could be stale, or the card could cover 680 // (part of) an object at the end of the allocated space and extend 681 // beyond the end of allocation. 682 683 // Non-humongous objects are only allocated in the old-gen during 684 // GC, so if region is old then top is stable. Humongous object 685 // allocation sets top last; if top has not yet been set, this is 686 // a stale card and we'll end up with an empty intersection. If 687 // this is not a stale card, the synchronization between the 688 // enqueuing of the card and processing it here will have ensured 689 // we see the up-to-date top here. 690 HeapWord* scan_limit = r->top(); 691 692 if (scan_limit <= start) { 693 // If the trimmed region is empty, the card must be stale. 694 return; 695 } 696 697 // Okay to clean and process the card now. There are still some 698 // stale card cases that may be detected by iteration and dealt with 699 // as iteration failure. 700 *const_cast<volatile jbyte*>(card_ptr) = CardTableModRefBS::clean_card_val(); 701 702 // This fence serves two purposes. First, the card must be cleaned 703 // before processing the contents. Second, we can't proceed with 704 // processing until after the read of top, for synchronization with 705 // possibly concurrent humongous object allocation. It's okay that 706 // reading top and reading type were racy wrto each other. We need 707 // both set, in any order, to proceed. 708 OrderAccess::fence(); 709 710 // Don't use addr_for(card_ptr + 1) which can ask for 711 // a card beyond the heap. 712 HeapWord* end = start + CardTableModRefBS::card_size_in_words; 713 MemRegion dirty_region(start, MIN2(scan_limit, end)); 714 assert(!dirty_region.is_empty(), "sanity"); 715 716 G1ConcurrentRefineOopClosure conc_refine_cl(_g1, worker_i); 717 718 bool card_processed = 719 r->oops_on_card_seq_iterate_careful<false>(dirty_region, &conc_refine_cl); 720 721 // If unable to process the card then we encountered an unparsable 722 // part of the heap (e.g. a partially allocated object) while 723 // processing a stale card. Despite the card being stale, redirty 724 // and re-enqueue, because we've already cleaned the card. Without 725 // this we could incorrectly discard a non-stale card. 726 if (!card_processed) { 727 // The card might have gotten re-dirtied and re-enqueued while we 728 // worked. (In fact, it's pretty likely.) 729 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 730 *card_ptr = CardTableModRefBS::dirty_card_val(); 731 MutexLockerEx x(Shared_DirtyCardQ_lock, 732 Mutex::_no_safepoint_check_flag); 733 DirtyCardQueue* sdcq = 734 JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); 735 sdcq->enqueue(card_ptr); 736 } 737 } else { 738 _conc_refine_cards++; 739 } 740 } 741 742 bool G1RemSet::refine_card_during_gc(jbyte* card_ptr, 743 uint worker_i, 744 G1ParPushHeapRSClosure* oops_in_heap_closure) { 745 assert(_g1->is_gc_active(), "Only call during GC"); 746 747 check_card_ptr(card_ptr, _ct_bs); 748 749 // If the card is no longer dirty, nothing to do. This covers cards that were already 750 // scanned as parts of the remembered sets. 751 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 752 // No need to return that this card contains refs that point 753 // into the collection set. 754 return false; 755 } 756 757 // During GC we can immediately clean the card since we will not re-enqueue stale 758 // cards as we know they can be disregarded. 759 *card_ptr = CardTableModRefBS::clean_card_val(); 760 761 // Construct the region representing the card. 762 HeapWord* card_start = _ct_bs->addr_for(card_ptr); 763 // And find the region containing it. 764 HeapRegion* r = _g1->heap_region_containing(card_start); 765 766 HeapWord* scan_limit = _scan_state->scan_top(r->hrm_index()); 767 if (scan_limit <= card_start) { 768 // If the card starts above the area in the region containing objects to scan, skip it. 769 return false; 770 } 771 772 // Don't use addr_for(card_ptr + 1) which can ask for 773 // a card beyond the heap. 774 HeapWord* card_end = card_start + CardTableModRefBS::card_size_in_words; 775 MemRegion dirty_region(card_start, MIN2(scan_limit, card_end)); 776 assert(!dirty_region.is_empty(), "sanity"); 777 778 G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1, 779 oops_in_heap_closure, 780 true, 781 worker_i); 782 update_rs_oop_cl.set_from(r); 783 784 bool card_processed = 785 r->oops_on_card_seq_iterate_careful<true>(dirty_region, 786 &update_rs_oop_cl); 787 assert(card_processed, "must be"); 788 _conc_refine_cards++; 789 790 return update_rs_oop_cl.has_refs_into_cset(); 791 } 792 793 void G1RemSet::print_periodic_summary_info(const char* header, uint period_count) { 794 if ((G1SummarizeRSetStatsPeriod > 0) && log_is_enabled(Trace, gc, remset) && 795 (period_count % G1SummarizeRSetStatsPeriod == 0)) { 796 797 if (!_prev_period_summary.initialized()) { 798 _prev_period_summary.initialize(this); 799 } 800 801 G1RemSetSummary current; 802 current.initialize(this); 803 _prev_period_summary.subtract_from(¤t); 804 805 Log(gc, remset) log; 806 log.trace("%s", header); 807 ResourceMark rm; 808 _prev_period_summary.print_on(log.trace_stream()); 809 810 _prev_period_summary.set(¤t); 811 } 812 } 813 814 void G1RemSet::print_summary_info() { 815 Log(gc, remset, exit) log; 816 if (log.is_trace()) { 817 log.trace(" Cumulative RS summary"); 818 G1RemSetSummary current; 819 current.initialize(this); 820 ResourceMark rm; 821 current.print_on(log.trace_stream()); 822 } 823 } 824 825 void G1RemSet::prepare_for_verify() { 826 if (G1HRRSFlushLogBuffersOnVerify && 827 (VerifyBeforeGC || VerifyAfterGC) 828 && (!_g1->collector_state()->full_collection() || G1VerifyRSetsDuringFullGC)) { 829 cleanupHRRS(); 830 _g1->set_refine_cte_cl_concurrency(false); 831 if (SafepointSynchronize::is_at_safepoint()) { 832 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 833 dcqs.concatenate_logs(); 834 } 835 836 bool use_hot_card_cache = _hot_card_cache->use_cache(); 837 _hot_card_cache->set_use_cache(false); 838 839 DirtyCardQueue into_cset_dcq(&_into_cset_dirty_card_queue_set); 840 update_rem_set(&into_cset_dcq, NULL, 0); 841 _into_cset_dirty_card_queue_set.clear(); 842 843 _hot_card_cache->set_use_cache(use_hot_card_cache); 844 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); 845 } 846 } 847 848 void G1RemSet::create_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) { 849 _card_live_data.create(workers, mark_bitmap); 850 } 851 852 void G1RemSet::finalize_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) { 853 _card_live_data.finalize(workers, mark_bitmap); 854 } 855 856 void G1RemSet::verify_card_live_data(WorkGang* workers, G1CMBitMap* bitmap) { 857 _card_live_data.verify(workers, bitmap); 858 } 859 860 void G1RemSet::clear_card_live_data(WorkGang* workers) { 861 _card_live_data.clear(workers); 862 } 863 864 #ifdef ASSERT 865 void G1RemSet::verify_card_live_data_is_clear() { 866 _card_live_data.verify_is_clear(); 867 } 868 #endif