1 /* 2 * Copyright (c) 2001, 2012, 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_implementation/g1/bufferingOopClosure.hpp" 27 #include "gc_implementation/g1/concurrentG1Refine.hpp" 28 #include "gc_implementation/g1/concurrentG1RefineThread.hpp" 29 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp" 30 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp" 31 #include "gc_implementation/g1/g1CollectorPolicy.hpp" 32 #include "gc_implementation/g1/g1OopClosures.inline.hpp" 33 #include "gc_implementation/g1/g1RemSet.inline.hpp" 34 #include "gc_implementation/g1/heapRegionSeq.inline.hpp" 35 #include "memory/iterator.hpp" 36 #include "oops/oop.inline.hpp" 37 #include "utilities/intHisto.hpp" 38 39 #define CARD_REPEAT_HISTO 0 40 41 #if CARD_REPEAT_HISTO 42 static size_t ct_freq_sz; 43 static jbyte* ct_freq = NULL; 44 45 void init_ct_freq_table(size_t heap_sz_bytes) { 46 if (ct_freq == NULL) { 47 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size; 48 ct_freq = new jbyte[ct_freq_sz]; 49 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0; 50 } 51 } 52 53 void ct_freq_note_card(size_t index) { 54 assert(0 <= index && index < ct_freq_sz, "Bounds error."); 55 if (ct_freq[index] < 100) { ct_freq[index]++; } 56 } 57 58 static IntHistogram card_repeat_count(10, 10); 59 60 void ct_freq_update_histo_and_reset() { 61 for (size_t j = 0; j < ct_freq_sz; j++) { 62 card_repeat_count.add_entry(ct_freq[j]); 63 ct_freq[j] = 0; 64 } 65 66 } 67 #endif 68 69 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) 70 : _g1(g1), _conc_refine_cards(0), 71 _ct_bs(ct_bs), _g1p(_g1->g1_policy()), 72 _cg1r(g1->concurrent_g1_refine()), 73 _cset_rs_update_cl(NULL), 74 _cards_scanned(NULL), _total_cards_scanned(0) 75 { 76 _seq_task = new SubTasksDone(NumSeqTasks); 77 guarantee(n_workers() > 0, "There should be some workers"); 78 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers()); 79 for (uint i = 0; i < n_workers(); i++) { 80 _cset_rs_update_cl[i] = NULL; 81 } 82 } 83 84 G1RemSet::~G1RemSet() { 85 delete _seq_task; 86 for (uint i = 0; i < n_workers(); i++) { 87 assert(_cset_rs_update_cl[i] == NULL, "it should be"); 88 } 89 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl); 90 } 91 92 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) { 93 if (_g1->is_in_g1_reserved(mr.start())) { 94 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size)); 95 if (_start_first == NULL) _start_first = mr.start(); 96 } 97 } 98 99 class ScanRSClosure : public HeapRegionClosure { 100 size_t _cards_done, _cards; 101 G1CollectedHeap* _g1h; 102 OopsInHeapRegionClosure* _oc; 103 G1BlockOffsetSharedArray* _bot_shared; 104 CardTableModRefBS *_ct_bs; 105 int _worker_i; 106 int _block_size; 107 bool _try_claimed; 108 public: 109 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) : 110 _oc(oc), 111 _cards(0), 112 _cards_done(0), 113 _worker_i(worker_i), 114 _try_claimed(false) 115 { 116 _g1h = G1CollectedHeap::heap(); 117 _bot_shared = _g1h->bot_shared(); 118 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set()); 119 _block_size = MAX2<int>(G1RSetScanBlockSize, 1); 120 } 121 122 void set_try_claimed() { _try_claimed = true; } 123 124 void scanCard(size_t index, HeapRegion *r) { 125 // Stack allocate the DirtyCardToOopClosure instance 126 HeapRegionDCTOC cl(_g1h, r, _oc, 127 CardTableModRefBS::Precise, 128 HeapRegionDCTOC::IntoCSFilterKind); 129 130 // Set the "from" region in the closure. 131 _oc->set_region(r); 132 HeapWord* card_start = _bot_shared->address_for_index(index); 133 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words; 134 Space *sp = SharedHeap::heap()->space_containing(card_start); 135 MemRegion sm_region = sp->used_region_at_save_marks(); 136 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end)); 137 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) { 138 // We make the card as "claimed" lazily (so races are possible 139 // but they're benign), which reduces the number of duplicate 140 // scans (the rsets of the regions in the cset can intersect). 141 _ct_bs->set_card_claimed(index); 142 _cards_done++; 143 cl.do_MemRegion(mr); 144 } 145 } 146 147 void printCard(HeapRegion* card_region, size_t card_index, 148 HeapWord* card_start) { 149 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") " 150 "RS names card %p: " 151 "[" PTR_FORMAT ", " PTR_FORMAT ")", 152 _worker_i, 153 card_region->bottom(), card_region->end(), 154 card_index, 155 card_start, card_start + G1BlockOffsetSharedArray::N_words); 156 } 157 158 bool doHeapRegion(HeapRegion* r) { 159 assert(r->in_collection_set(), "should only be called on elements of CS."); 160 HeapRegionRemSet* hrrs = r->rem_set(); 161 if (hrrs->iter_is_complete()) return false; // All done. 162 if (!_try_claimed && !hrrs->claim_iter()) return false; 163 // If we ever free the collection set concurrently, we should also 164 // clear the card table concurrently therefore we won't need to 165 // add regions of the collection set to the dirty cards region. 166 _g1h->push_dirty_cards_region(r); 167 // If we didn't return above, then 168 // _try_claimed || r->claim_iter() 169 // is true: either we're supposed to work on claimed-but-not-complete 170 // regions, or we successfully claimed the region. 171 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i); 172 hrrs->init_iterator(iter); 173 size_t card_index; 174 175 // We claim cards in block so as to recude the contention. The block size is determined by 176 // the G1RSetScanBlockSize parameter. 177 size_t jump_to_card = hrrs->iter_claimed_next(_block_size); 178 for (size_t current_card = 0; iter->has_next(card_index); current_card++) { 179 if (current_card >= jump_to_card + _block_size) { 180 jump_to_card = hrrs->iter_claimed_next(_block_size); 181 } 182 if (current_card < jump_to_card) continue; 183 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index); 184 #if 0 185 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n", 186 card_start, card_start + CardTableModRefBS::card_size_in_words); 187 #endif 188 189 HeapRegion* card_region = _g1h->heap_region_containing(card_start); 190 assert(card_region != NULL, "Yielding cards not in the heap?"); 191 _cards++; 192 193 if (!card_region->is_on_dirty_cards_region_list()) { 194 _g1h->push_dirty_cards_region(card_region); 195 } 196 197 // If the card is dirty, then we will scan it during updateRS. 198 if (!card_region->in_collection_set() && 199 !_ct_bs->is_card_dirty(card_index)) { 200 scanCard(card_index, card_region); 201 } 202 } 203 if (!_try_claimed) { 204 hrrs->set_iter_complete(); 205 } 206 return false; 207 } 208 size_t cards_done() { return _cards_done;} 209 size_t cards_looked_up() { return _cards;} 210 }; 211 212 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { 213 double rs_time_start = os::elapsedTime(); 214 HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i); 215 216 ScanRSClosure scanRScl(oc, worker_i); 217 218 _g1->collection_set_iterate_from(startRegion, &scanRScl); 219 scanRScl.set_try_claimed(); 220 _g1->collection_set_iterate_from(startRegion, &scanRScl); 221 222 double scan_rs_time_sec = os::elapsedTime() - rs_time_start; 223 224 assert( _cards_scanned != NULL, "invariant" ); 225 _cards_scanned[worker_i] = scanRScl.cards_done(); 226 227 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); 228 } 229 230 // Closure used for updating RSets and recording references that 231 // point into the collection set. Only called during an 232 // evacuation pause. 233 234 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure { 235 G1RemSet* _g1rs; 236 DirtyCardQueue* _into_cset_dcq; 237 public: 238 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h, 239 DirtyCardQueue* into_cset_dcq) : 240 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq) 241 {} 242 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 243 // The only time we care about recording cards that 244 // contain references that point into the collection set 245 // is during RSet updating within an evacuation pause. 246 // In this case worker_i should be the id of a GC worker thread. 247 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); 248 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker"); 249 250 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) { 251 // 'card_ptr' contains references that point into the collection 252 // set. We need to record the card in the DCQS 253 // (G1CollectedHeap::into_cset_dirty_card_queue_set()) 254 // that's used for that purpose. 255 // 256 // Enqueue the card 257 _into_cset_dcq->enqueue(card_ptr); 258 } 259 return true; 260 } 261 }; 262 263 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { 264 double start = os::elapsedTime(); 265 // Apply the given closure to all remaining log entries. 266 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq); 267 268 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i); 269 270 // Now there should be no dirty cards. 271 if (G1RSLogCheckCardTable) { 272 CountNonCleanMemRegionClosure cl(_g1); 273 _ct_bs->mod_card_iterate(&cl); 274 // XXX This isn't true any more: keeping cards of young regions 275 // marked dirty broke it. Need some reasonable fix. 276 guarantee(cl.n() == 0, "Card table should be clean."); 277 } 278 279 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0); 280 } 281 282 void G1RemSet::cleanupHRRS() { 283 HeapRegionRemSet::cleanup(); 284 } 285 286 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, 287 int worker_i) { 288 #if CARD_REPEAT_HISTO 289 ct_freq_update_histo_and_reset(); 290 #endif 291 if (worker_i == 0) { 292 _cg1r->clear_and_record_card_counts(); 293 } 294 295 // We cache the value of 'oc' closure into the appropriate slot in the 296 // _cset_rs_update_cl for this worker 297 assert(worker_i < (int)n_workers(), "sanity"); 298 _cset_rs_update_cl[worker_i] = oc; 299 300 // A DirtyCardQueue that is used to hold cards containing references 301 // that point into the collection set. This DCQ is associated with a 302 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal 303 // circumstances (i.e. the pause successfully completes), these cards 304 // are just discarded (there's no need to update the RSets of regions 305 // that were in the collection set - after the pause these regions 306 // are wholly 'free' of live objects. In the event of an evacuation 307 // failure the cards/buffers in this queue set are: 308 // * passed to the DirtyCardQueueSet that is used to manage deferred 309 // RSet updates, or 310 // * scanned for references that point into the collection set 311 // and the RSet of the corresponding region in the collection set 312 // is updated immediately. 313 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 314 315 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant"); 316 317 // The two flags below were introduced temporarily to serialize 318 // the updating and scanning of remembered sets. There are some 319 // race conditions when these two operations are done in parallel 320 // and they are causing failures. When we resolve said race 321 // conditions, we'll revert back to parallel remembered set 322 // updating and scanning. See CRs 6677707 and 6677708. 323 if (G1UseParallelRSetUpdating || (worker_i == 0)) { 324 updateRS(&into_cset_dcq, worker_i); 325 } else { 326 _g1p->record_update_rs_processed_buffers(worker_i, 0.0); 327 _g1p->record_update_rs_time(worker_i, 0.0); 328 } 329 if (G1UseParallelRSetScanning || (worker_i == 0)) { 330 scanRS(oc, worker_i); 331 } else { 332 _g1p->record_scan_rs_time(worker_i, 0.0); 333 } 334 335 // We now clear the cached values of _cset_rs_update_cl for this worker 336 _cset_rs_update_cl[worker_i] = NULL; 337 } 338 339 void G1RemSet::prepare_for_oops_into_collection_set_do() { 340 cleanupHRRS(); 341 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); 342 _g1->set_refine_cte_cl_concurrency(false); 343 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 344 dcqs.concatenate_logs(); 345 346 if (G1CollectedHeap::use_parallel_gc_threads()) { 347 // Don't set the number of workers here. It will be set 348 // when the task is run 349 // _seq_task->set_n_termination((int)n_workers()); 350 } 351 guarantee( _cards_scanned == NULL, "invariant" ); 352 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); 353 for (uint i = 0; i < n_workers(); ++i) { 354 _cards_scanned[i] = 0; 355 } 356 _total_cards_scanned = 0; 357 } 358 359 360 // This closure, applied to a DirtyCardQueueSet, is used to immediately 361 // update the RSets for the regions in the CSet. For each card it iterates 362 // through the oops which coincide with that card. It scans the reference 363 // fields in each oop; when it finds an oop that points into the collection 364 // set, the RSet for the region containing the referenced object is updated. 365 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure { 366 G1CollectedHeap* _g1; 367 CardTableModRefBS* _ct_bs; 368 public: 369 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1, 370 CardTableModRefBS* bs): 371 _g1(g1), _ct_bs(bs) 372 { } 373 374 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 375 // Construct the region representing the card. 376 HeapWord* start = _ct_bs->addr_for(card_ptr); 377 // And find the region containing it. 378 HeapRegion* r = _g1->heap_region_containing(start); 379 assert(r != NULL, "unexpected null"); 380 381 // Scan oops in the card looking for references into the collection set 382 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 383 MemRegion scanRegion(start, end); 384 385 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set()); 386 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl); 387 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl); 388 389 // We can pass false as the "filter_young" parameter here as: 390 // * we should be in a STW pause, 391 // * the DCQS to which this closure is applied is used to hold 392 // references that point into the collection set from the prior 393 // RSet updating, 394 // * the post-write barrier shouldn't be logging updates to young 395 // regions (but there is a situation where this can happen - see 396 // the comment in G1RemSet::concurrentRefineOneCard below - 397 // that should not be applicable here), and 398 // * during actual RSet updating, the filtering of cards in young 399 // regions in HeapRegion::oops_on_card_seq_iterate_careful is 400 // employed. 401 // As a result, when this closure is applied to "refs into cset" 402 // DCQS, we shouldn't see any cards in young regions. 403 update_rs_cl.set_region(r); 404 HeapWord* stop_point = 405 r->oops_on_card_seq_iterate_careful(scanRegion, 406 &filter_then_update_rs_cset_oop_cl, 407 false /* filter_young */, 408 NULL /* card_ptr */); 409 410 // Since this is performed in the event of an evacuation failure, we 411 // we shouldn't see a non-null stop point 412 assert(stop_point == NULL, "saw an unallocated region"); 413 return true; 414 } 415 }; 416 417 void G1RemSet::cleanup_after_oops_into_collection_set_do() { 418 guarantee( _cards_scanned != NULL, "invariant" ); 419 _total_cards_scanned = 0; 420 for (uint i = 0; i < n_workers(); ++i) { 421 _total_cards_scanned += _cards_scanned[i]; 422 } 423 FREE_C_HEAP_ARRAY(size_t, _cards_scanned); 424 _cards_scanned = NULL; 425 // Cleanup after copy 426 _g1->set_refine_cte_cl_concurrency(true); 427 // Set all cards back to clean. 428 _g1->cleanUpCardTable(); 429 430 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set(); 431 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num(); 432 433 if (_g1->evacuation_failed()) { 434 // Restore remembered sets for the regions pointing into the collection set. 435 436 if (G1DeferredRSUpdate) { 437 // If deferred RS updates are enabled then we just need to transfer 438 // the completed buffers from (a) the DirtyCardQueueSet used to hold 439 // cards that contain references that point into the collection set 440 // to (b) the DCQS used to hold the deferred RS updates 441 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs); 442 } else { 443 444 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set(); 445 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs); 446 447 int n_completed_buffers = 0; 448 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate, 449 0, 0, true)) { 450 n_completed_buffers++; 451 } 452 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers"); 453 } 454 } 455 456 // Free any completed buffers in the DirtyCardQueueSet used to hold cards 457 // which contain references that point into the collection. 458 _g1->into_cset_dirty_card_queue_set().clear(); 459 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0, 460 "all buffers should be freed"); 461 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers(); 462 } 463 464 class ScrubRSClosure: public HeapRegionClosure { 465 G1CollectedHeap* _g1h; 466 BitMap* _region_bm; 467 BitMap* _card_bm; 468 CardTableModRefBS* _ctbs; 469 public: 470 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) : 471 _g1h(G1CollectedHeap::heap()), 472 _region_bm(region_bm), _card_bm(card_bm), 473 _ctbs(NULL) 474 { 475 ModRefBarrierSet* bs = _g1h->mr_bs(); 476 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition"); 477 _ctbs = (CardTableModRefBS*)bs; 478 } 479 480 bool doHeapRegion(HeapRegion* r) { 481 if (!r->continuesHumongous()) { 482 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm); 483 } 484 return false; 485 } 486 }; 487 488 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { 489 ScrubRSClosure scrub_cl(region_bm, card_bm); 490 _g1->heap_region_iterate(&scrub_cl); 491 } 492 493 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, 494 uint worker_num, int claim_val) { 495 ScrubRSClosure scrub_cl(region_bm, card_bm); 496 _g1->heap_region_par_iterate_chunked(&scrub_cl, 497 worker_num, 498 n_workers(), 499 claim_val); 500 } 501 502 503 504 G1TriggerClosure::G1TriggerClosure() : 505 _triggered(false) { } 506 507 G1InvokeIfNotTriggeredClosure::G1InvokeIfNotTriggeredClosure(G1TriggerClosure* t_cl, 508 OopClosure* oop_cl) : 509 _trigger_cl(t_cl), _oop_cl(oop_cl) { } 510 511 G1Mux2Closure::G1Mux2Closure(OopClosure *c1, OopClosure *c2) : 512 _c1(c1), _c2(c2) { } 513 514 G1UpdateRSOrPushRefOopClosure:: 515 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h, 516 G1RemSet* rs, 517 OopsInHeapRegionClosure* push_ref_cl, 518 bool record_refs_into_cset, 519 int worker_i) : 520 _g1(g1h), _g1_rem_set(rs), _from(NULL), 521 _record_refs_into_cset(record_refs_into_cset), 522 _push_ref_cl(push_ref_cl), _worker_i(worker_i) { } 523 524 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, 525 bool check_for_refs_into_cset) { 526 // Construct the region representing the card. 527 HeapWord* start = _ct_bs->addr_for(card_ptr); 528 // And find the region containing it. 529 HeapRegion* r = _g1->heap_region_containing(start); 530 assert(r != NULL, "unexpected null"); 531 532 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 533 MemRegion dirtyRegion(start, end); 534 535 #if CARD_REPEAT_HISTO 536 init_ct_freq_table(_g1->max_capacity()); 537 ct_freq_note_card(_ct_bs->index_for(start)); 538 #endif 539 540 OopsInHeapRegionClosure* oops_in_heap_closure = NULL; 541 if (check_for_refs_into_cset) { 542 // ConcurrentG1RefineThreads have worker numbers larger than what 543 // _cset_rs_update_cl[] is set up to handle. But those threads should 544 // only be active outside of a collection which means that when they 545 // reach here they should have check_for_refs_into_cset == false. 546 assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length"); 547 oops_in_heap_closure = _cset_rs_update_cl[worker_i]; 548 } 549 G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1, 550 _g1->g1_rem_set(), 551 oops_in_heap_closure, 552 check_for_refs_into_cset, 553 worker_i); 554 update_rs_oop_cl.set_from(r); 555 556 G1TriggerClosure trigger_cl; 557 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl); 558 G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl); 559 G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl); 560 561 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, 562 (check_for_refs_into_cset ? 563 (OopClosure*)&mux : 564 (OopClosure*)&update_rs_oop_cl)); 565 566 // The region for the current card may be a young region. The 567 // current card may have been a card that was evicted from the 568 // card cache. When the card was inserted into the cache, we had 569 // determined that its region was non-young. While in the cache, 570 // the region may have been freed during a cleanup pause, reallocated 571 // and tagged as young. 572 // 573 // We wish to filter out cards for such a region but the current 574 // thread, if we're running concurrently, may "see" the young type 575 // change at any time (so an earlier "is_young" check may pass or 576 // fail arbitrarily). We tell the iteration code to perform this 577 // filtering when it has been determined that there has been an actual 578 // allocation in this region and making it safe to check the young type. 579 bool filter_young = true; 580 581 HeapWord* stop_point = 582 r->oops_on_card_seq_iterate_careful(dirtyRegion, 583 &filter_then_update_rs_oop_cl, 584 filter_young, 585 card_ptr); 586 587 // If stop_point is non-null, then we encountered an unallocated region 588 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the 589 // card and re-enqueue: if we put off the card until a GC pause, then the 590 // unallocated portion will be filled in. Alternatively, we might try 591 // the full complexity of the technique used in "regular" precleaning. 592 if (stop_point != NULL) { 593 // The card might have gotten re-dirtied and re-enqueued while we 594 // worked. (In fact, it's pretty likely.) 595 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 596 *card_ptr = CardTableModRefBS::dirty_card_val(); 597 MutexLockerEx x(Shared_DirtyCardQ_lock, 598 Mutex::_no_safepoint_check_flag); 599 DirtyCardQueue* sdcq = 600 JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); 601 sdcq->enqueue(card_ptr); 602 } 603 } else { 604 _conc_refine_cards++; 605 } 606 607 return trigger_cl.triggered(); 608 } 609 610 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, 611 bool check_for_refs_into_cset) { 612 // If the card is no longer dirty, nothing to do. 613 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 614 // No need to return that this card contains refs that point 615 // into the collection set. 616 return false; 617 } 618 619 // Construct the region representing the card. 620 HeapWord* start = _ct_bs->addr_for(card_ptr); 621 // And find the region containing it. 622 HeapRegion* r = _g1->heap_region_containing(start); 623 if (r == NULL) { 624 guarantee(_g1->is_in_permanent(start), "Or else where?"); 625 // Again no need to return that this card contains refs that 626 // point into the collection set. 627 return false; // Not in the G1 heap (might be in perm, for example.) 628 } 629 // Why do we have to check here whether a card is on a young region, 630 // given that we dirty young regions and, as a result, the 631 // post-barrier is supposed to filter them out and never to enqueue 632 // them? When we allocate a new region as the "allocation region" we 633 // actually dirty its cards after we release the lock, since card 634 // dirtying while holding the lock was a performance bottleneck. So, 635 // as a result, it is possible for other threads to actually 636 // allocate objects in the region (after the acquire the lock) 637 // before all the cards on the region are dirtied. This is unlikely, 638 // and it doesn't happen often, but it can happen. So, the extra 639 // check below filters out those cards. 640 if (r->is_young()) { 641 return false; 642 } 643 // While we are processing RSet buffers during the collection, we 644 // actually don't want to scan any cards on the collection set, 645 // since we don't want to update remebered sets with entries that 646 // point into the collection set, given that live objects from the 647 // collection set are about to move and such entries will be stale 648 // very soon. This change also deals with a reliability issue which 649 // involves scanning a card in the collection set and coming across 650 // an array that was being chunked and looking malformed. Note, 651 // however, that if evacuation fails, we have to scan any objects 652 // that were not moved and create any missing entries. 653 if (r->in_collection_set()) { 654 return false; 655 } 656 657 // Should we defer processing the card? 658 // 659 // Previously the result from the insert_cache call would be 660 // either card_ptr (implying that card_ptr was currently "cold"), 661 // null (meaning we had inserted the card ptr into the "hot" 662 // cache, which had some headroom), or a "hot" card ptr 663 // extracted from the "hot" cache. 664 // 665 // Now that the _card_counts cache in the ConcurrentG1Refine 666 // instance is an evicting hash table, the result we get back 667 // could be from evicting the card ptr in an already occupied 668 // bucket (in which case we have replaced the card ptr in the 669 // bucket with card_ptr and "defer" is set to false). To avoid 670 // having a data structure (updates to which would need a lock) 671 // to hold these unprocessed dirty cards, we need to immediately 672 // process card_ptr. The actions needed to be taken on return 673 // from cache_insert are summarized in the following table: 674 // 675 // res defer action 676 // -------------------------------------------------------------- 677 // null false card evicted from _card_counts & replaced with 678 // card_ptr; evicted ptr added to hot cache. 679 // No need to process res; immediately process card_ptr 680 // 681 // null true card not evicted from _card_counts; card_ptr added 682 // to hot cache. 683 // Nothing to do. 684 // 685 // non-null false card evicted from _card_counts & replaced with 686 // card_ptr; evicted ptr is currently "cold" or 687 // caused an eviction from the hot cache. 688 // Immediately process res; process card_ptr. 689 // 690 // non-null true card not evicted from _card_counts; card_ptr is 691 // currently cold, or caused an eviction from hot 692 // cache. 693 // Immediately process res; no need to process card_ptr. 694 695 696 jbyte* res = card_ptr; 697 bool defer = false; 698 699 // This gets set to true if the card being refined has references 700 // that point into the collection set. 701 bool oops_into_cset = false; 702 703 if (_cg1r->use_cache()) { 704 jbyte* res = _cg1r->cache_insert(card_ptr, &defer); 705 if (res != NULL && (res != card_ptr || defer)) { 706 start = _ct_bs->addr_for(res); 707 r = _g1->heap_region_containing(start); 708 if (r == NULL) { 709 assert(_g1->is_in_permanent(start), "Or else where?"); 710 } else { 711 // Checking whether the region we got back from the cache 712 // is young here is inappropriate. The region could have been 713 // freed, reallocated and tagged as young while in the cache. 714 // Hence we could see its young type change at any time. 715 // 716 // Process card pointer we get back from the hot card cache. This 717 // will check whether the region containing the card is young 718 // _after_ checking that the region has been allocated from. 719 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i, 720 false /* check_for_refs_into_cset */); 721 // The above call to concurrentRefineOneCard_impl is only 722 // performed if the hot card cache is enabled. This cache is 723 // disabled during an evacuation pause - which is the only 724 // time when we need know if the card contains references 725 // that point into the collection set. Also when the hot card 726 // cache is enabled, this code is executed by the concurrent 727 // refine threads - rather than the GC worker threads - and 728 // concurrentRefineOneCard_impl will return false. 729 assert(!oops_into_cset, "should not see true here"); 730 } 731 } 732 } 733 734 if (!defer) { 735 oops_into_cset = 736 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset); 737 // We should only be detecting that the card contains references 738 // that point into the collection set if the current thread is 739 // a GC worker thread. 740 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(), 741 "invalid result at non safepoint"); 742 } 743 return oops_into_cset; 744 } 745 746 class HRRSStatsIter: public HeapRegionClosure { 747 size_t _occupied; 748 size_t _total_mem_sz; 749 size_t _max_mem_sz; 750 HeapRegion* _max_mem_sz_region; 751 public: 752 HRRSStatsIter() : 753 _occupied(0), 754 _total_mem_sz(0), 755 _max_mem_sz(0), 756 _max_mem_sz_region(NULL) 757 {} 758 759 bool doHeapRegion(HeapRegion* r) { 760 if (r->continuesHumongous()) return false; 761 size_t mem_sz = r->rem_set()->mem_size(); 762 if (mem_sz > _max_mem_sz) { 763 _max_mem_sz = mem_sz; 764 _max_mem_sz_region = r; 765 } 766 _total_mem_sz += mem_sz; 767 size_t occ = r->rem_set()->occupied(); 768 _occupied += occ; 769 return false; 770 } 771 size_t total_mem_sz() { return _total_mem_sz; } 772 size_t max_mem_sz() { return _max_mem_sz; } 773 size_t occupied() { return _occupied; } 774 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; } 775 }; 776 777 class PrintRSThreadVTimeClosure : public ThreadClosure { 778 public: 779 virtual void do_thread(Thread *t) { 780 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t; 781 gclog_or_tty->print(" %5.2f", crt->vtime_accum()); 782 } 783 }; 784 785 void G1RemSet::print_summary_info() { 786 G1CollectedHeap* g1 = G1CollectedHeap::heap(); 787 788 #if CARD_REPEAT_HISTO 789 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: "); 790 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number."); 791 card_repeat_count.print_on(gclog_or_tty); 792 #endif 793 794 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards", 795 _conc_refine_cards); 796 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 797 jint tot_processed_buffers = 798 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread(); 799 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers); 800 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.", 801 dcqs.processed_buffers_rs_thread(), 802 100.0*(float)dcqs.processed_buffers_rs_thread()/ 803 (float)tot_processed_buffers); 804 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.", 805 dcqs.processed_buffers_mut(), 806 100.0*(float)dcqs.processed_buffers_mut()/ 807 (float)tot_processed_buffers); 808 gclog_or_tty->print_cr(" Conc RS threads times(s)"); 809 PrintRSThreadVTimeClosure p; 810 gclog_or_tty->print(" "); 811 g1->concurrent_g1_refine()->threads_do(&p); 812 gclog_or_tty->print_cr(""); 813 814 HRRSStatsIter blk; 815 g1->heap_region_iterate(&blk); 816 gclog_or_tty->print_cr(" Total heap region rem set sizes = "SIZE_FORMAT"K." 817 " Max = "SIZE_FORMAT"K.", 818 blk.total_mem_sz()/K, blk.max_mem_sz()/K); 819 gclog_or_tty->print_cr(" Static structures = "SIZE_FORMAT"K," 820 " free_lists = "SIZE_FORMAT"K.", 821 HeapRegionRemSet::static_mem_size() / K, 822 HeapRegionRemSet::fl_mem_size() / K); 823 gclog_or_tty->print_cr(" "SIZE_FORMAT" occupied cards represented.", 824 blk.occupied()); 825 HeapRegion* max_mem_sz_region = blk.max_mem_sz_region(); 826 HeapRegionRemSet* rem_set = max_mem_sz_region->rem_set(); 827 gclog_or_tty->print_cr(" Max size region = "HR_FORMAT", " 828 "size = "SIZE_FORMAT "K, occupied = "SIZE_FORMAT"K.", 829 HR_FORMAT_PARAMS(max_mem_sz_region), 830 (rem_set->mem_size() + K - 1)/K, 831 (rem_set->occupied() + K - 1)/K); 832 gclog_or_tty->print_cr(" Did %d coarsenings.", 833 HeapRegionRemSet::n_coarsenings()); 834 } 835 836 void G1RemSet::prepare_for_verify() { 837 if (G1HRRSFlushLogBuffersOnVerify && 838 (VerifyBeforeGC || VerifyAfterGC) 839 && !_g1->full_collection()) { 840 cleanupHRRS(); 841 _g1->set_refine_cte_cl_concurrency(false); 842 if (SafepointSynchronize::is_at_safepoint()) { 843 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 844 dcqs.concatenate_logs(); 845 } 846 bool cg1r_use_cache = _cg1r->use_cache(); 847 _cg1r->set_use_cache(false); 848 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 849 updateRS(&into_cset_dcq, 0); 850 _g1->into_cset_dirty_card_queue_set().clear(); 851 _cg1r->set_use_cache(cg1r_use_cache); 852 853 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); 854 } 855 }