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