1 /* 2 * Copyright (c) 2001, 2011, 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 DirtyCardToOopClosure* cl = 126 r->new_dcto_closure(_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 // We want the parallel threads to start their scanning at 213 // different collection set regions to avoid contention. 214 // If we have: 215 // n collection set regions 216 // p threads 217 // Then thread t will start at region t * floor (n/p) 218 219 HeapRegion* G1RemSet::calculateStartRegion(int worker_i) { 220 HeapRegion* result = _g1p->collection_set(); 221 if (ParallelGCThreads > 0) { 222 size_t cs_size = _g1p->collection_set_size(); 223 int n_workers = _g1->workers()->total_workers(); 224 size_t cs_spans = cs_size / n_workers; 225 size_t ind = cs_spans * worker_i; 226 for (size_t i = 0; i < ind; i++) 227 result = result->next_in_collection_set(); 228 } 229 return result; 230 } 231 232 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { 233 double rs_time_start = os::elapsedTime(); 234 HeapRegion *startRegion = calculateStartRegion(worker_i); 235 236 ScanRSClosure scanRScl(oc, worker_i); 237 238 _g1->collection_set_iterate_from(startRegion, &scanRScl); 239 scanRScl.set_try_claimed(); 240 _g1->collection_set_iterate_from(startRegion, &scanRScl); 241 242 double scan_rs_time_sec = os::elapsedTime() - rs_time_start; 243 244 assert( _cards_scanned != NULL, "invariant" ); 245 _cards_scanned[worker_i] = scanRScl.cards_done(); 246 247 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); 248 } 249 250 // Closure used for updating RSets and recording references that 251 // point into the collection set. Only called during an 252 // evacuation pause. 253 254 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure { 255 G1RemSet* _g1rs; 256 DirtyCardQueue* _into_cset_dcq; 257 public: 258 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h, 259 DirtyCardQueue* into_cset_dcq) : 260 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq) 261 {} 262 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 263 // The only time we care about recording cards that 264 // contain references that point into the collection set 265 // is during RSet updating within an evacuation pause. 266 // In this case worker_i should be the id of a GC worker thread. 267 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); 268 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker"); 269 270 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) { 271 // 'card_ptr' contains references that point into the collection 272 // set. We need to record the card in the DCQS 273 // (G1CollectedHeap::into_cset_dirty_card_queue_set()) 274 // that's used for that purpose. 275 // 276 // Enqueue the card 277 _into_cset_dcq->enqueue(card_ptr); 278 } 279 return true; 280 } 281 }; 282 283 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { 284 double start = os::elapsedTime(); 285 // Apply the given closure to all remaining log entries. 286 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq); 287 288 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i); 289 290 // Now there should be no dirty cards. 291 if (G1RSLogCheckCardTable) { 292 CountNonCleanMemRegionClosure cl(_g1); 293 _ct_bs->mod_card_iterate(&cl); 294 // XXX This isn't true any more: keeping cards of young regions 295 // marked dirty broke it. Need some reasonable fix. 296 guarantee(cl.n() == 0, "Card table should be clean."); 297 } 298 299 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0); 300 } 301 302 #ifndef PRODUCT 303 class PrintRSClosure : public HeapRegionClosure { 304 int _count; 305 public: 306 PrintRSClosure() : _count(0) {} 307 bool doHeapRegion(HeapRegion* r) { 308 HeapRegionRemSet* hrrs = r->rem_set(); 309 _count += (int) hrrs->occupied(); 310 if (hrrs->occupied() == 0) { 311 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") " 312 "has no remset entries\n", 313 r->bottom(), r->end()); 314 } else { 315 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n", 316 r->bottom(), r->end()); 317 r->print(); 318 hrrs->print(); 319 gclog_or_tty->print("\nDone printing rem set\n"); 320 } 321 return false; 322 } 323 int occupied() {return _count;} 324 }; 325 #endif 326 327 class CountRSSizeClosure: public HeapRegionClosure { 328 size_t _n; 329 size_t _tot; 330 size_t _max; 331 HeapRegion* _max_r; 332 enum { 333 N = 20, 334 MIN = 6 335 }; 336 int _histo[N]; 337 public: 338 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) { 339 for (int i = 0; i < N; i++) _histo[i] = 0; 340 } 341 bool doHeapRegion(HeapRegion* r) { 342 if (!r->continuesHumongous()) { 343 size_t occ = r->rem_set()->occupied(); 344 _n++; 345 _tot += occ; 346 if (occ > _max) { 347 _max = occ; 348 _max_r = r; 349 } 350 // Fit it into a histo bin. 351 int s = 1 << MIN; 352 int i = 0; 353 while (occ > (size_t) s && i < (N-1)) { 354 s = s << 1; 355 i++; 356 } 357 _histo[i]++; 358 } 359 return false; 360 } 361 size_t n() { return _n; } 362 size_t tot() { return _tot; } 363 size_t mx() { return _max; } 364 HeapRegion* mxr() { return _max_r; } 365 void print_histo() { 366 int mx = N; 367 while (mx >= 0) { 368 if (_histo[mx-1] > 0) break; 369 mx--; 370 } 371 gclog_or_tty->print_cr("Number of regions with given RS sizes:"); 372 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]); 373 for (int i = 1; i < mx-1; i++) { 374 gclog_or_tty->print_cr(" %8d - %8d %8d", 375 (1 << (MIN + i - 1)) + 1, 376 1 << (MIN + i), 377 _histo[i]); 378 } 379 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]); 380 } 381 }; 382 383 void G1RemSet::cleanupHRRS() { 384 HeapRegionRemSet::cleanup(); 385 } 386 387 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, 388 int worker_i) { 389 #if CARD_REPEAT_HISTO 390 ct_freq_update_histo_and_reset(); 391 #endif 392 if (worker_i == 0) { 393 _cg1r->clear_and_record_card_counts(); 394 } 395 396 // Make this into a command-line flag... 397 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) { 398 CountRSSizeClosure count_cl; 399 _g1->heap_region_iterate(&count_cl); 400 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, " 401 "max region is " PTR_FORMAT, 402 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(), 403 count_cl.mx(), count_cl.mxr()); 404 count_cl.print_histo(); 405 } 406 407 // We cache the value of 'oc' closure into the appropriate slot in the 408 // _cset_rs_update_cl for this worker 409 assert(worker_i < (int)n_workers(), "sanity"); 410 _cset_rs_update_cl[worker_i] = oc; 411 412 // A DirtyCardQueue that is used to hold cards containing references 413 // that point into the collection set. This DCQ is associated with a 414 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal 415 // circumstances (i.e. the pause successfully completes), these cards 416 // are just discarded (there's no need to update the RSets of regions 417 // that were in the collection set - after the pause these regions 418 // are wholly 'free' of live objects. In the event of an evacuation 419 // failure the cards/buffers in this queue set are: 420 // * passed to the DirtyCardQueueSet that is used to manage deferred 421 // RSet updates, or 422 // * scanned for references that point into the collection set 423 // and the RSet of the corresponding region in the collection set 424 // is updated immediately. 425 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 426 427 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant"); 428 429 // The two flags below were introduced temporarily to serialize 430 // the updating and scanning of remembered sets. There are some 431 // race conditions when these two operations are done in parallel 432 // and they are causing failures. When we resolve said race 433 // conditions, we'll revert back to parallel remembered set 434 // updating and scanning. See CRs 6677707 and 6677708. 435 if (G1UseParallelRSetUpdating || (worker_i == 0)) { 436 updateRS(&into_cset_dcq, worker_i); 437 } else { 438 _g1p->record_update_rs_processed_buffers(worker_i, 0.0); 439 _g1p->record_update_rs_time(worker_i, 0.0); 440 } 441 if (G1UseParallelRSetScanning || (worker_i == 0)) { 442 scanRS(oc, worker_i); 443 } else { 444 _g1p->record_scan_rs_time(worker_i, 0.0); 445 } 446 447 // We now clear the cached values of _cset_rs_update_cl for this worker 448 _cset_rs_update_cl[worker_i] = NULL; 449 } 450 451 void G1RemSet::prepare_for_oops_into_collection_set_do() { 452 #if G1_REM_SET_LOGGING 453 PrintRSClosure cl; 454 _g1->collection_set_iterate(&cl); 455 #endif 456 cleanupHRRS(); 457 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); 458 _g1->set_refine_cte_cl_concurrency(false); 459 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 460 dcqs.concatenate_logs(); 461 462 if (ParallelGCThreads > 0) { 463 _seq_task->set_n_threads((int)n_workers()); 464 } 465 guarantee( _cards_scanned == NULL, "invariant" ); 466 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); 467 for (uint i = 0; i < n_workers(); ++i) { 468 _cards_scanned[i] = 0; 469 } 470 _total_cards_scanned = 0; 471 } 472 473 474 class cleanUpIteratorsClosure : public HeapRegionClosure { 475 bool doHeapRegion(HeapRegion *r) { 476 HeapRegionRemSet* hrrs = r->rem_set(); 477 hrrs->init_for_par_iteration(); 478 return false; 479 } 480 }; 481 482 // This closure, applied to a DirtyCardQueueSet, is used to immediately 483 // update the RSets for the regions in the CSet. For each card it iterates 484 // through the oops which coincide with that card. It scans the reference 485 // fields in each oop; when it finds an oop that points into the collection 486 // set, the RSet for the region containing the referenced object is updated. 487 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure { 488 G1CollectedHeap* _g1; 489 CardTableModRefBS* _ct_bs; 490 public: 491 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1, 492 CardTableModRefBS* bs): 493 _g1(g1), _ct_bs(bs) 494 { } 495 496 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 497 // Construct the region representing the card. 498 HeapWord* start = _ct_bs->addr_for(card_ptr); 499 // And find the region containing it. 500 HeapRegion* r = _g1->heap_region_containing(start); 501 assert(r != NULL, "unexpected null"); 502 503 // Scan oops in the card looking for references into the collection set 504 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 505 MemRegion scanRegion(start, end); 506 507 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set()); 508 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl, NULL /* rp */); 509 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl); 510 511 // We can pass false as the "filter_young" parameter here as: 512 // * we should be in a STW pause, 513 // * the DCQS to which this closure is applied is used to hold 514 // references that point into the collection set from the prior 515 // RSet updating, 516 // * the post-write barrier shouldn't be logging updates to young 517 // regions (but there is a situation where this can happen - see 518 // the comment in G1RemSet::concurrentRefineOneCard below - 519 // that should not be applicable here), and 520 // * during actual RSet updating, the filtering of cards in young 521 // regions in HeapRegion::oops_on_card_seq_iterate_careful is 522 // employed. 523 // As a result, when this closure is applied to "refs into cset" 524 // DCQS, we shouldn't see any cards in young regions. 525 update_rs_cl.set_region(r); 526 HeapWord* stop_point = 527 r->oops_on_card_seq_iterate_careful(scanRegion, 528 &filter_then_update_rs_cset_oop_cl, 529 false /* filter_young */, 530 NULL /* card_ptr */); 531 532 // Since this is performed in the event of an evacuation failure, we 533 // we shouldn't see a non-null stop point 534 assert(stop_point == NULL, "saw an unallocated region"); 535 return true; 536 } 537 }; 538 539 void G1RemSet::cleanup_after_oops_into_collection_set_do() { 540 guarantee( _cards_scanned != NULL, "invariant" ); 541 _total_cards_scanned = 0; 542 for (uint i = 0; i < n_workers(); ++i) 543 _total_cards_scanned += _cards_scanned[i]; 544 FREE_C_HEAP_ARRAY(size_t, _cards_scanned); 545 _cards_scanned = NULL; 546 // Cleanup after copy 547 #if G1_REM_SET_LOGGING 548 PrintRSClosure cl; 549 _g1->heap_region_iterate(&cl); 550 #endif 551 _g1->set_refine_cte_cl_concurrency(true); 552 cleanUpIteratorsClosure iterClosure; 553 _g1->collection_set_iterate(&iterClosure); 554 // Set all cards back to clean. 555 _g1->cleanUpCardTable(); 556 557 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set(); 558 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num(); 559 560 if (_g1->evacuation_failed()) { 561 // Restore remembered sets for the regions pointing into the collection set. 562 563 if (G1DeferredRSUpdate) { 564 // If deferred RS updates are enabled then we just need to transfer 565 // the completed buffers from (a) the DirtyCardQueueSet used to hold 566 // cards that contain references that point into the collection set 567 // to (b) the DCQS used to hold the deferred RS updates 568 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs); 569 } else { 570 571 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set(); 572 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs); 573 574 int n_completed_buffers = 0; 575 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate, 576 0, 0, true)) { 577 n_completed_buffers++; 578 } 579 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers"); 580 } 581 } 582 583 // Free any completed buffers in the DirtyCardQueueSet used to hold cards 584 // which contain references that point into the collection. 585 _g1->into_cset_dirty_card_queue_set().clear(); 586 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0, 587 "all buffers should be freed"); 588 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers(); 589 } 590 591 class ScrubRSClosure: public HeapRegionClosure { 592 G1CollectedHeap* _g1h; 593 BitMap* _region_bm; 594 BitMap* _card_bm; 595 CardTableModRefBS* _ctbs; 596 public: 597 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) : 598 _g1h(G1CollectedHeap::heap()), 599 _region_bm(region_bm), _card_bm(card_bm), 600 _ctbs(NULL) 601 { 602 ModRefBarrierSet* bs = _g1h->mr_bs(); 603 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition"); 604 _ctbs = (CardTableModRefBS*)bs; 605 } 606 607 bool doHeapRegion(HeapRegion* r) { 608 if (!r->continuesHumongous()) { 609 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm); 610 } 611 return false; 612 } 613 }; 614 615 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { 616 ScrubRSClosure scrub_cl(region_bm, card_bm); 617 _g1->heap_region_iterate(&scrub_cl); 618 } 619 620 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, 621 int worker_num, int claim_val) { 622 ScrubRSClosure scrub_cl(region_bm, card_bm); 623 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val); 624 } 625 626 627 static IntHistogram out_of_histo(50, 50); 628 629 class TriggerClosure : public OopClosure { 630 bool _trigger; 631 public: 632 TriggerClosure() : _trigger(false) { } 633 bool value() const { return _trigger; } 634 template <class T> void do_oop_nv(T* p) { _trigger = true; } 635 virtual void do_oop(oop* p) { do_oop_nv(p); } 636 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 637 }; 638 639 class InvokeIfNotTriggeredClosure: public OopClosure { 640 TriggerClosure* _t; 641 OopClosure* _oc; 642 public: 643 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc): 644 _t(t), _oc(oc) { } 645 template <class T> void do_oop_nv(T* p) { 646 if (!_t->value()) _oc->do_oop(p); 647 } 648 virtual void do_oop(oop* p) { do_oop_nv(p); } 649 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 650 }; 651 652 class Mux2Closure : public OopClosure { 653 OopClosure* _c1; 654 OopClosure* _c2; 655 public: 656 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { } 657 template <class T> void do_oop_nv(T* p) { 658 _c1->do_oop(p); _c2->do_oop(p); 659 } 660 virtual void do_oop(oop* p) { do_oop_nv(p); } 661 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 662 }; 663 664 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, 665 bool check_for_refs_into_cset) { 666 // Construct the region representing the card. 667 HeapWord* start = _ct_bs->addr_for(card_ptr); 668 // And find the region containing it. 669 HeapRegion* r = _g1->heap_region_containing(start); 670 assert(r != NULL, "unexpected null"); 671 672 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 673 MemRegion dirtyRegion(start, end); 674 675 #if CARD_REPEAT_HISTO 676 init_ct_freq_table(_g1->max_capacity()); 677 ct_freq_note_card(_ct_bs->index_for(start)); 678 #endif 679 680 assert(!check_for_refs_into_cset || _cset_rs_update_cl[worker_i] != NULL, "sanity"); 681 UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1, 682 _g1->g1_rem_set(), 683 _cset_rs_update_cl[worker_i], 684 check_for_refs_into_cset, 685 worker_i); 686 update_rs_oop_cl.set_from(r); 687 688 TriggerClosure trigger_cl; 689 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl, NULL /* rp */); 690 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl); 691 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl); 692 693 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, 694 (check_for_refs_into_cset ? 695 (OopClosure*)&mux : 696 (OopClosure*)&update_rs_oop_cl)); 697 698 // The region for the current card may be a young region. The 699 // current card may have been a card that was evicted from the 700 // card cache. When the card was inserted into the cache, we had 701 // determined that its region was non-young. While in the cache, 702 // the region may have been freed during a cleanup pause, reallocated 703 // and tagged as young. 704 // 705 // We wish to filter out cards for such a region but the current 706 // thread, if we're running concurrently, may "see" the young type 707 // change at any time (so an earlier "is_young" check may pass or 708 // fail arbitrarily). We tell the iteration code to perform this 709 // filtering when it has been determined that there has been an actual 710 // allocation in this region and making it safe to check the young type. 711 bool filter_young = true; 712 713 HeapWord* stop_point = 714 r->oops_on_card_seq_iterate_careful(dirtyRegion, 715 &filter_then_update_rs_oop_cl, 716 filter_young, 717 card_ptr); 718 719 // If stop_point is non-null, then we encountered an unallocated region 720 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the 721 // card and re-enqueue: if we put off the card until a GC pause, then the 722 // unallocated portion will be filled in. Alternatively, we might try 723 // the full complexity of the technique used in "regular" precleaning. 724 if (stop_point != NULL) { 725 // The card might have gotten re-dirtied and re-enqueued while we 726 // worked. (In fact, it's pretty likely.) 727 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 728 *card_ptr = CardTableModRefBS::dirty_card_val(); 729 MutexLockerEx x(Shared_DirtyCardQ_lock, 730 Mutex::_no_safepoint_check_flag); 731 DirtyCardQueue* sdcq = 732 JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); 733 sdcq->enqueue(card_ptr); 734 } 735 } else { 736 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region()); 737 _conc_refine_cards++; 738 } 739 740 return trigger_cl.value(); 741 } 742 743 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, 744 bool check_for_refs_into_cset) { 745 // If the card is no longer dirty, nothing to do. 746 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 747 // No need to return that this card contains refs that point 748 // into the collection set. 749 return false; 750 } 751 752 // Construct the region representing the card. 753 HeapWord* start = _ct_bs->addr_for(card_ptr); 754 // And find the region containing it. 755 HeapRegion* r = _g1->heap_region_containing(start); 756 if (r == NULL) { 757 guarantee(_g1->is_in_permanent(start), "Or else where?"); 758 // Again no need to return that this card contains refs that 759 // point into the collection set. 760 return false; // Not in the G1 heap (might be in perm, for example.) 761 } 762 // Why do we have to check here whether a card is on a young region, 763 // given that we dirty young regions and, as a result, the 764 // post-barrier is supposed to filter them out and never to enqueue 765 // them? When we allocate a new region as the "allocation region" we 766 // actually dirty its cards after we release the lock, since card 767 // dirtying while holding the lock was a performance bottleneck. So, 768 // as a result, it is possible for other threads to actually 769 // allocate objects in the region (after the acquire the lock) 770 // before all the cards on the region are dirtied. This is unlikely, 771 // and it doesn't happen often, but it can happen. So, the extra 772 // check below filters out those cards. 773 if (r->is_young()) { 774 return false; 775 } 776 // While we are processing RSet buffers during the collection, we 777 // actually don't want to scan any cards on the collection set, 778 // since we don't want to update remebered sets with entries that 779 // point into the collection set, given that live objects from the 780 // collection set are about to move and such entries will be stale 781 // very soon. This change also deals with a reliability issue which 782 // involves scanning a card in the collection set and coming across 783 // an array that was being chunked and looking malformed. Note, 784 // however, that if evacuation fails, we have to scan any objects 785 // that were not moved and create any missing entries. 786 if (r->in_collection_set()) { 787 return false; 788 } 789 790 // Should we defer processing the card? 791 // 792 // Previously the result from the insert_cache call would be 793 // either card_ptr (implying that card_ptr was currently "cold"), 794 // null (meaning we had inserted the card ptr into the "hot" 795 // cache, which had some headroom), or a "hot" card ptr 796 // extracted from the "hot" cache. 797 // 798 // Now that the _card_counts cache in the ConcurrentG1Refine 799 // instance is an evicting hash table, the result we get back 800 // could be from evicting the card ptr in an already occupied 801 // bucket (in which case we have replaced the card ptr in the 802 // bucket with card_ptr and "defer" is set to false). To avoid 803 // having a data structure (updates to which would need a lock) 804 // to hold these unprocessed dirty cards, we need to immediately 805 // process card_ptr. The actions needed to be taken on return 806 // from cache_insert are summarized in the following table: 807 // 808 // res defer action 809 // -------------------------------------------------------------- 810 // null false card evicted from _card_counts & replaced with 811 // card_ptr; evicted ptr added to hot cache. 812 // No need to process res; immediately process card_ptr 813 // 814 // null true card not evicted from _card_counts; card_ptr added 815 // to hot cache. 816 // Nothing to do. 817 // 818 // non-null false card evicted from _card_counts & replaced with 819 // card_ptr; evicted ptr is currently "cold" or 820 // caused an eviction from the hot cache. 821 // Immediately process res; process card_ptr. 822 // 823 // non-null true card not evicted from _card_counts; card_ptr is 824 // currently cold, or caused an eviction from hot 825 // cache. 826 // Immediately process res; no need to process card_ptr. 827 828 829 jbyte* res = card_ptr; 830 bool defer = false; 831 832 // This gets set to true if the card being refined has references 833 // that point into the collection set. 834 bool oops_into_cset = false; 835 836 if (_cg1r->use_cache()) { 837 jbyte* res = _cg1r->cache_insert(card_ptr, &defer); 838 if (res != NULL && (res != card_ptr || defer)) { 839 start = _ct_bs->addr_for(res); 840 r = _g1->heap_region_containing(start); 841 if (r == NULL) { 842 assert(_g1->is_in_permanent(start), "Or else where?"); 843 } else { 844 // Checking whether the region we got back from the cache 845 // is young here is inappropriate. The region could have been 846 // freed, reallocated and tagged as young while in the cache. 847 // Hence we could see its young type change at any time. 848 // 849 // Process card pointer we get back from the hot card cache. This 850 // will check whether the region containing the card is young 851 // _after_ checking that the region has been allocated from. 852 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i, 853 false /* check_for_refs_into_cset */); 854 // The above call to concurrentRefineOneCard_impl is only 855 // performed if the hot card cache is enabled. This cache is 856 // disabled during an evacuation pause - which is the only 857 // time when we need know if the card contains references 858 // that point into the collection set. Also when the hot card 859 // cache is enabled, this code is executed by the concurrent 860 // refine threads - rather than the GC worker threads - and 861 // concurrentRefineOneCard_impl will return false. 862 assert(!oops_into_cset, "should not see true here"); 863 } 864 } 865 } 866 867 if (!defer) { 868 oops_into_cset = 869 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset); 870 // We should only be detecting that the card contains references 871 // that point into the collection set if the current thread is 872 // a GC worker thread. 873 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(), 874 "invalid result at non safepoint"); 875 } 876 return oops_into_cset; 877 } 878 879 class HRRSStatsIter: public HeapRegionClosure { 880 size_t _occupied; 881 size_t _total_mem_sz; 882 size_t _max_mem_sz; 883 HeapRegion* _max_mem_sz_region; 884 public: 885 HRRSStatsIter() : 886 _occupied(0), 887 _total_mem_sz(0), 888 _max_mem_sz(0), 889 _max_mem_sz_region(NULL) 890 {} 891 892 bool doHeapRegion(HeapRegion* r) { 893 if (r->continuesHumongous()) return false; 894 size_t mem_sz = r->rem_set()->mem_size(); 895 if (mem_sz > _max_mem_sz) { 896 _max_mem_sz = mem_sz; 897 _max_mem_sz_region = r; 898 } 899 _total_mem_sz += mem_sz; 900 size_t occ = r->rem_set()->occupied(); 901 _occupied += occ; 902 return false; 903 } 904 size_t total_mem_sz() { return _total_mem_sz; } 905 size_t max_mem_sz() { return _max_mem_sz; } 906 size_t occupied() { return _occupied; } 907 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; } 908 }; 909 910 class PrintRSThreadVTimeClosure : public ThreadClosure { 911 public: 912 virtual void do_thread(Thread *t) { 913 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t; 914 gclog_or_tty->print(" %5.2f", crt->vtime_accum()); 915 } 916 }; 917 918 void G1RemSet::print_summary_info() { 919 G1CollectedHeap* g1 = G1CollectedHeap::heap(); 920 921 #if CARD_REPEAT_HISTO 922 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: "); 923 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number."); 924 card_repeat_count.print_on(gclog_or_tty); 925 #endif 926 927 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) { 928 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: "); 929 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number."); 930 out_of_histo.print_on(gclog_or_tty); 931 } 932 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards", 933 _conc_refine_cards); 934 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 935 jint tot_processed_buffers = 936 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread(); 937 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers); 938 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.", 939 dcqs.processed_buffers_rs_thread(), 940 100.0*(float)dcqs.processed_buffers_rs_thread()/ 941 (float)tot_processed_buffers); 942 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.", 943 dcqs.processed_buffers_mut(), 944 100.0*(float)dcqs.processed_buffers_mut()/ 945 (float)tot_processed_buffers); 946 gclog_or_tty->print_cr(" Conc RS threads times(s)"); 947 PrintRSThreadVTimeClosure p; 948 gclog_or_tty->print(" "); 949 g1->concurrent_g1_refine()->threads_do(&p); 950 gclog_or_tty->print_cr(""); 951 952 HRRSStatsIter blk; 953 g1->heap_region_iterate(&blk); 954 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." 955 " Max = " SIZE_FORMAT "K.", 956 blk.total_mem_sz()/K, blk.max_mem_sz()/K); 957 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," 958 " free_lists = " SIZE_FORMAT "K.", 959 HeapRegionRemSet::static_mem_size()/K, 960 HeapRegionRemSet::fl_mem_size()/K); 961 gclog_or_tty->print_cr(" %d occupied cards represented.", 962 blk.occupied()); 963 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" 964 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", 965 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), 966 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, 967 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); 968 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings()); 969 } 970 971 void G1RemSet::prepare_for_verify() { 972 if (G1HRRSFlushLogBuffersOnVerify && 973 (VerifyBeforeGC || VerifyAfterGC) 974 && !_g1->full_collection()) { 975 cleanupHRRS(); 976 _g1->set_refine_cte_cl_concurrency(false); 977 if (SafepointSynchronize::is_at_safepoint()) { 978 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 979 dcqs.concatenate_logs(); 980 } 981 bool cg1r_use_cache = _cg1r->use_cache(); 982 _cg1r->set_use_cache(false); 983 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 984 updateRS(&into_cset_dcq, 0); 985 _g1->into_cset_dirty_card_queue_set().clear(); 986 _cg1r->set_use_cache(cg1r_use_cache); 987 988 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); 989 } 990 }