1 /* 2 * Copyright (c) 2001, 2010, 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 "incls/_precompiled.incl" 26 #include "incls/_g1RemSet.cpp.incl" 27 28 #define CARD_REPEAT_HISTO 0 29 30 #if CARD_REPEAT_HISTO 31 static size_t ct_freq_sz; 32 static jbyte* ct_freq = NULL; 33 34 void init_ct_freq_table(size_t heap_sz_bytes) { 35 if (ct_freq == NULL) { 36 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size; 37 ct_freq = new jbyte[ct_freq_sz]; 38 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0; 39 } 40 } 41 42 void ct_freq_note_card(size_t index) { 43 assert(0 <= index && index < ct_freq_sz, "Bounds error."); 44 if (ct_freq[index] < 100) { ct_freq[index]++; } 45 } 46 47 static IntHistogram card_repeat_count(10, 10); 48 49 void ct_freq_update_histo_and_reset() { 50 for (size_t j = 0; j < ct_freq_sz; j++) { 51 card_repeat_count.add_entry(ct_freq[j]); 52 ct_freq[j] = 0; 53 } 54 55 } 56 #endif 57 58 59 class IntoCSOopClosure: public OopsInHeapRegionClosure { 60 OopsInHeapRegionClosure* _blk; 61 G1CollectedHeap* _g1; 62 public: 63 IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : 64 _g1(g1), _blk(blk) {} 65 void set_region(HeapRegion* from) { 66 _blk->set_region(from); 67 } 68 virtual void do_oop(narrowOop* p) { do_oop_work(p); } 69 virtual void do_oop( oop* p) { do_oop_work(p); } 70 template <class T> void do_oop_work(T* p) { 71 oop obj = oopDesc::load_decode_heap_oop(p); 72 if (_g1->obj_in_cs(obj)) _blk->do_oop(p); 73 } 74 bool apply_to_weak_ref_discovered_field() { return true; } 75 bool idempotent() { return true; } 76 }; 77 78 class IntoCSRegionClosure: public HeapRegionClosure { 79 IntoCSOopClosure _blk; 80 G1CollectedHeap* _g1; 81 public: 82 IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : 83 _g1(g1), _blk(g1, blk) {} 84 bool doHeapRegion(HeapRegion* r) { 85 if (!r->in_collection_set()) { 86 _blk.set_region(r); 87 if (r->isHumongous()) { 88 if (r->startsHumongous()) { 89 oop obj = oop(r->bottom()); 90 obj->oop_iterate(&_blk); 91 } 92 } else { 93 r->oop_before_save_marks_iterate(&_blk); 94 } 95 } 96 return false; 97 } 98 }; 99 100 class VerifyRSCleanCardOopClosure: public OopClosure { 101 G1CollectedHeap* _g1; 102 public: 103 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {} 104 105 virtual void do_oop(narrowOop* p) { do_oop_work(p); } 106 virtual void do_oop( oop* p) { do_oop_work(p); } 107 template <class T> void do_oop_work(T* p) { 108 oop obj = oopDesc::load_decode_heap_oop(p); 109 HeapRegion* to = _g1->heap_region_containing(obj); 110 guarantee(to == NULL || !to->in_collection_set(), 111 "Missed a rem set member."); 112 } 113 }; 114 115 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) 116 : _g1(g1), _conc_refine_cards(0), 117 _ct_bs(ct_bs), _g1p(_g1->g1_policy()), 118 _cg1r(g1->concurrent_g1_refine()), 119 _traversal_in_progress(false), 120 _cset_rs_update_cl(NULL), 121 _cards_scanned(NULL), _total_cards_scanned(0) 122 { 123 _seq_task = new SubTasksDone(NumSeqTasks); 124 guarantee(n_workers() > 0, "There should be some workers"); 125 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers()); 126 for (uint i = 0; i < n_workers(); i++) { 127 _cset_rs_update_cl[i] = NULL; 128 } 129 } 130 131 G1RemSet::~G1RemSet() { 132 delete _seq_task; 133 for (uint i = 0; i < n_workers(); i++) { 134 assert(_cset_rs_update_cl[i] == NULL, "it should be"); 135 } 136 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl); 137 } 138 139 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) { 140 if (_g1->is_in_g1_reserved(mr.start())) { 141 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size)); 142 if (_start_first == NULL) _start_first = mr.start(); 143 } 144 } 145 146 class ScanRSClosure : public HeapRegionClosure { 147 size_t _cards_done, _cards; 148 G1CollectedHeap* _g1h; 149 OopsInHeapRegionClosure* _oc; 150 G1BlockOffsetSharedArray* _bot_shared; 151 CardTableModRefBS *_ct_bs; 152 int _worker_i; 153 int _block_size; 154 bool _try_claimed; 155 public: 156 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) : 157 _oc(oc), 158 _cards(0), 159 _cards_done(0), 160 _worker_i(worker_i), 161 _try_claimed(false) 162 { 163 _g1h = G1CollectedHeap::heap(); 164 _bot_shared = _g1h->bot_shared(); 165 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set()); 166 _block_size = MAX2<int>(G1RSetScanBlockSize, 1); 167 } 168 169 void set_try_claimed() { _try_claimed = true; } 170 171 void scanCard(size_t index, HeapRegion *r) { 172 _cards_done++; 173 DirtyCardToOopClosure* cl = 174 r->new_dcto_closure(_oc, 175 CardTableModRefBS::Precise, 176 HeapRegionDCTOC::IntoCSFilterKind); 177 178 // Set the "from" region in the closure. 179 _oc->set_region(r); 180 HeapWord* card_start = _bot_shared->address_for_index(index); 181 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words; 182 Space *sp = SharedHeap::heap()->space_containing(card_start); 183 MemRegion sm_region; 184 if (ParallelGCThreads > 0) { 185 // first find the used area 186 sm_region = sp->used_region_at_save_marks(); 187 } else { 188 // The closure is not idempotent. We shouldn't look at objects 189 // allocated during the GC. 190 sm_region = sp->used_region_at_save_marks(); 191 } 192 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end)); 193 if (!mr.is_empty()) { 194 cl->do_MemRegion(mr); 195 } 196 } 197 198 void printCard(HeapRegion* card_region, size_t card_index, 199 HeapWord* card_start) { 200 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") " 201 "RS names card %p: " 202 "[" PTR_FORMAT ", " PTR_FORMAT ")", 203 _worker_i, 204 card_region->bottom(), card_region->end(), 205 card_index, 206 card_start, card_start + G1BlockOffsetSharedArray::N_words); 207 } 208 209 bool doHeapRegion(HeapRegion* r) { 210 assert(r->in_collection_set(), "should only be called on elements of CS."); 211 HeapRegionRemSet* hrrs = r->rem_set(); 212 if (hrrs->iter_is_complete()) return false; // All done. 213 if (!_try_claimed && !hrrs->claim_iter()) return false; 214 _g1h->push_dirty_cards_region(r); 215 // If we didn't return above, then 216 // _try_claimed || r->claim_iter() 217 // is true: either we're supposed to work on claimed-but-not-complete 218 // regions, or we successfully claimed the region. 219 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i); 220 hrrs->init_iterator(iter); 221 size_t card_index; 222 223 // We claim cards in block so as to recude the contention. The block size is determined by 224 // the G1RSetScanBlockSize parameter. 225 size_t jump_to_card = hrrs->iter_claimed_next(_block_size); 226 for (size_t current_card = 0; iter->has_next(card_index); current_card++) { 227 if (current_card >= jump_to_card + _block_size) { 228 jump_to_card = hrrs->iter_claimed_next(_block_size); 229 } 230 if (current_card < jump_to_card) continue; 231 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index); 232 #if 0 233 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n", 234 card_start, card_start + CardTableModRefBS::card_size_in_words); 235 #endif 236 237 HeapRegion* card_region = _g1h->heap_region_containing(card_start); 238 assert(card_region != NULL, "Yielding cards not in the heap?"); 239 _cards++; 240 241 if (!card_region->is_on_dirty_cards_region_list()) { 242 _g1h->push_dirty_cards_region(card_region); 243 } 244 245 // If the card is dirty, then we will scan it during updateRS. 246 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) { 247 // We make the card as "claimed" lazily (so races are possible but they're benign), 248 // which reduces the number of duplicate scans (the rsets of the regions in the cset 249 // can intersect). 250 if (!_ct_bs->is_card_claimed(card_index)) { 251 _ct_bs->set_card_claimed(card_index); 252 scanCard(card_index, card_region); 253 } 254 } 255 } 256 if (!_try_claimed) { 257 hrrs->set_iter_complete(); 258 } 259 return false; 260 } 261 // Set all cards back to clean. 262 void cleanup() {_g1h->cleanUpCardTable();} 263 size_t cards_done() { return _cards_done;} 264 size_t cards_looked_up() { return _cards;} 265 }; 266 267 // We want the parallel threads to start their scanning at 268 // different collection set regions to avoid contention. 269 // If we have: 270 // n collection set regions 271 // p threads 272 // Then thread t will start at region t * floor (n/p) 273 274 HeapRegion* G1RemSet::calculateStartRegion(int worker_i) { 275 HeapRegion* result = _g1p->collection_set(); 276 if (ParallelGCThreads > 0) { 277 size_t cs_size = _g1p->collection_set_size(); 278 int n_workers = _g1->workers()->total_workers(); 279 size_t cs_spans = cs_size / n_workers; 280 size_t ind = cs_spans * worker_i; 281 for (size_t i = 0; i < ind; i++) 282 result = result->next_in_collection_set(); 283 } 284 return result; 285 } 286 287 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { 288 double rs_time_start = os::elapsedTime(); 289 HeapRegion *startRegion = calculateStartRegion(worker_i); 290 291 ScanRSClosure scanRScl(oc, worker_i); 292 _g1->collection_set_iterate_from(startRegion, &scanRScl); 293 scanRScl.set_try_claimed(); 294 _g1->collection_set_iterate_from(startRegion, &scanRScl); 295 296 double scan_rs_time_sec = os::elapsedTime() - rs_time_start; 297 298 assert( _cards_scanned != NULL, "invariant" ); 299 _cards_scanned[worker_i] = scanRScl.cards_done(); 300 301 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); 302 } 303 304 // Closure used for updating RSets and recording references that 305 // point into the collection set. Only called during an 306 // evacuation pause. 307 308 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure { 309 G1RemSet* _g1rs; 310 DirtyCardQueue* _into_cset_dcq; 311 public: 312 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h, 313 DirtyCardQueue* into_cset_dcq) : 314 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq) 315 {} 316 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 317 // The only time we care about recording cards that 318 // contain references that point into the collection set 319 // is during RSet updating within an evacuation pause. 320 // In this case worker_i should be the id of a GC worker thread. 321 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause"); 322 assert(worker_i < (int) DirtyCardQueueSet::num_par_ids(), "should be a GC worker"); 323 324 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) { 325 // 'card_ptr' contains references that point into the collection 326 // set. We need to record the card in the DCQS 327 // (G1CollectedHeap::into_cset_dirty_card_queue_set()) 328 // that's used for that purpose. 329 // 330 // Enqueue the card 331 _into_cset_dcq->enqueue(card_ptr); 332 } 333 return true; 334 } 335 }; 336 337 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) { 338 double start = os::elapsedTime(); 339 // Apply the given closure to all remaining log entries. 340 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq); 341 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i); 342 343 // Now there should be no dirty cards. 344 if (G1RSLogCheckCardTable) { 345 CountNonCleanMemRegionClosure cl(_g1); 346 _ct_bs->mod_card_iterate(&cl); 347 // XXX This isn't true any more: keeping cards of young regions 348 // marked dirty broke it. Need some reasonable fix. 349 guarantee(cl.n() == 0, "Card table should be clean."); 350 } 351 352 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0); 353 } 354 355 #ifndef PRODUCT 356 class PrintRSClosure : public HeapRegionClosure { 357 int _count; 358 public: 359 PrintRSClosure() : _count(0) {} 360 bool doHeapRegion(HeapRegion* r) { 361 HeapRegionRemSet* hrrs = r->rem_set(); 362 _count += (int) hrrs->occupied(); 363 if (hrrs->occupied() == 0) { 364 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") " 365 "has no remset entries\n", 366 r->bottom(), r->end()); 367 } else { 368 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n", 369 r->bottom(), r->end()); 370 r->print(); 371 hrrs->print(); 372 gclog_or_tty->print("\nDone printing rem set\n"); 373 } 374 return false; 375 } 376 int occupied() {return _count;} 377 }; 378 #endif 379 380 class CountRSSizeClosure: public HeapRegionClosure { 381 size_t _n; 382 size_t _tot; 383 size_t _max; 384 HeapRegion* _max_r; 385 enum { 386 N = 20, 387 MIN = 6 388 }; 389 int _histo[N]; 390 public: 391 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) { 392 for (int i = 0; i < N; i++) _histo[i] = 0; 393 } 394 bool doHeapRegion(HeapRegion* r) { 395 if (!r->continuesHumongous()) { 396 size_t occ = r->rem_set()->occupied(); 397 _n++; 398 _tot += occ; 399 if (occ > _max) { 400 _max = occ; 401 _max_r = r; 402 } 403 // Fit it into a histo bin. 404 int s = 1 << MIN; 405 int i = 0; 406 while (occ > (size_t) s && i < (N-1)) { 407 s = s << 1; 408 i++; 409 } 410 _histo[i]++; 411 } 412 return false; 413 } 414 size_t n() { return _n; } 415 size_t tot() { return _tot; } 416 size_t mx() { return _max; } 417 HeapRegion* mxr() { return _max_r; } 418 void print_histo() { 419 int mx = N; 420 while (mx >= 0) { 421 if (_histo[mx-1] > 0) break; 422 mx--; 423 } 424 gclog_or_tty->print_cr("Number of regions with given RS sizes:"); 425 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]); 426 for (int i = 1; i < mx-1; i++) { 427 gclog_or_tty->print_cr(" %8d - %8d %8d", 428 (1 << (MIN + i - 1)) + 1, 429 1 << (MIN + i), 430 _histo[i]); 431 } 432 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]); 433 } 434 }; 435 436 void G1RemSet::cleanupHRRS() { 437 HeapRegionRemSet::cleanup(); 438 } 439 440 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, 441 int worker_i) { 442 #if CARD_REPEAT_HISTO 443 ct_freq_update_histo_and_reset(); 444 #endif 445 if (worker_i == 0) { 446 _cg1r->clear_and_record_card_counts(); 447 } 448 449 // Make this into a command-line flag... 450 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) { 451 CountRSSizeClosure count_cl; 452 _g1->heap_region_iterate(&count_cl); 453 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, " 454 "max region is " PTR_FORMAT, 455 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(), 456 count_cl.mx(), count_cl.mxr()); 457 count_cl.print_histo(); 458 } 459 460 // We cache the value of 'oc' closure into the appropriate slot in the 461 // _cset_rs_update_cl for this worker 462 assert(worker_i < (int)n_workers(), "sanity"); 463 _cset_rs_update_cl[worker_i] = oc; 464 465 // A DirtyCardQueue that is used to hold cards containing references 466 // that point into the collection set. This DCQ is associated with a 467 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal 468 // circumstances (i.e. the pause successfully completes), these cards 469 // are just discarded (there's no need to update the RSets of regions 470 // that were in the collection set - after the pause these regions 471 // are wholly 'free' of live objects. In the event of an evacuation 472 // failure the cards/buffers in this queue set are: 473 // * passed to the DirtyCardQueueSet that is used to manage deferred 474 // RSet updates, or 475 // * scanned for references that point into the collection set 476 // and the RSet of the corresponding region in the collection set 477 // is updated immediately. 478 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 479 480 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant"); 481 482 // The two flags below were introduced temporarily to serialize 483 // the updating and scanning of remembered sets. There are some 484 // race conditions when these two operations are done in parallel 485 // and they are causing failures. When we resolve said race 486 // conditions, we'll revert back to parallel remembered set 487 // updating and scanning. See CRs 6677707 and 6677708. 488 if (G1UseParallelRSetUpdating || (worker_i == 0)) { 489 updateRS(&into_cset_dcq, worker_i); 490 } else { 491 _g1p->record_update_rs_processed_buffers(worker_i, 0.0); 492 _g1p->record_update_rs_time(worker_i, 0.0); 493 } 494 if (G1UseParallelRSetScanning || (worker_i == 0)) { 495 scanRS(oc, worker_i); 496 } else { 497 _g1p->record_scan_rs_time(worker_i, 0.0); 498 } 499 500 // We now clear the cached values of _cset_rs_update_cl for this worker 501 _cset_rs_update_cl[worker_i] = NULL; 502 } 503 504 void G1RemSet::prepare_for_oops_into_collection_set_do() { 505 #if G1_REM_SET_LOGGING 506 PrintRSClosure cl; 507 _g1->collection_set_iterate(&cl); 508 #endif 509 cleanupHRRS(); 510 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); 511 _g1->set_refine_cte_cl_concurrency(false); 512 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 513 dcqs.concatenate_logs(); 514 515 assert(!_traversal_in_progress, "Invariant between iterations."); 516 set_traversal(true); 517 if (ParallelGCThreads > 0) { 518 _seq_task->set_n_threads((int)n_workers()); 519 } 520 guarantee( _cards_scanned == NULL, "invariant" ); 521 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); 522 for (uint i = 0; i < n_workers(); ++i) { 523 _cards_scanned[i] = 0; 524 } 525 _total_cards_scanned = 0; 526 } 527 528 529 class cleanUpIteratorsClosure : public HeapRegionClosure { 530 bool doHeapRegion(HeapRegion *r) { 531 HeapRegionRemSet* hrrs = r->rem_set(); 532 hrrs->init_for_par_iteration(); 533 return false; 534 } 535 }; 536 537 // This closure, applied to a DirtyCardQueueSet, is used to immediately 538 // update the RSets for the regions in the CSet. For each card it iterates 539 // through the oops which coincide with that card. It scans the reference 540 // fields in each oop; when it finds an oop that points into the collection 541 // set, the RSet for the region containing the referenced object is updated. 542 // Note: _par_traversal_in_progress in the G1RemSet must be FALSE; otherwise 543 // the UpdateRSetImmediate closure will cause cards to be enqueued on to 544 // the DCQS that we're iterating over, causing an infinite loop. 545 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure { 546 G1CollectedHeap* _g1; 547 CardTableModRefBS* _ct_bs; 548 public: 549 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1, 550 CardTableModRefBS* bs): 551 _g1(g1), _ct_bs(bs) 552 { } 553 554 bool do_card_ptr(jbyte* card_ptr, int worker_i) { 555 // Construct the region representing the card. 556 HeapWord* start = _ct_bs->addr_for(card_ptr); 557 // And find the region containing it. 558 HeapRegion* r = _g1->heap_region_containing(start); 559 assert(r != NULL, "unexpected null"); 560 561 // Scan oops in the card looking for references into the collection set 562 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 563 MemRegion scanRegion(start, end); 564 565 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set()); 566 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl); 567 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl); 568 569 // We can pass false as the "filter_young" parameter here as: 570 // * we should be in a STW pause, 571 // * the DCQS to which this closure is applied is used to hold 572 // references that point into the collection set from the prior 573 // RSet updating, 574 // * the post-write barrier shouldn't be logging updates to young 575 // regions (but there is a situation where this can happen - see 576 // the comment in G1RemSet::concurrentRefineOneCard below - 577 // that should not be applicable here), and 578 // * during actual RSet updating, the filtering of cards in young 579 // regions in HeapRegion::oops_on_card_seq_iterate_careful is 580 // employed. 581 // As a result, when this closure is applied to "refs into cset" 582 // DCQS, we shouldn't see any cards in young regions. 583 update_rs_cl.set_region(r); 584 HeapWord* stop_point = 585 r->oops_on_card_seq_iterate_careful(scanRegion, 586 &filter_then_update_rs_cset_oop_cl, 587 false /* filter_young */); 588 589 // Since this is performed in the event of an evacuation failure, we 590 // we shouldn't see a non-null stop point 591 assert(stop_point == NULL, "saw an unallocated region"); 592 return true; 593 } 594 }; 595 596 void G1RemSet::cleanup_after_oops_into_collection_set_do() { 597 guarantee( _cards_scanned != NULL, "invariant" ); 598 _total_cards_scanned = 0; 599 for (uint i = 0; i < n_workers(); ++i) 600 _total_cards_scanned += _cards_scanned[i]; 601 FREE_C_HEAP_ARRAY(size_t, _cards_scanned); 602 _cards_scanned = NULL; 603 // Cleanup after copy 604 #if G1_REM_SET_LOGGING 605 PrintRSClosure cl; 606 _g1->heap_region_iterate(&cl); 607 #endif 608 _g1->set_refine_cte_cl_concurrency(true); 609 cleanUpIteratorsClosure iterClosure; 610 _g1->collection_set_iterate(&iterClosure); 611 // Set all cards back to clean. 612 _g1->cleanUpCardTable(); 613 614 set_traversal(false); 615 616 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set(); 617 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num(); 618 619 if (_g1->evacuation_failed()) { 620 // Restore remembered sets for the regions pointing into the collection set. 621 622 if (G1DeferredRSUpdate) { 623 // If deferred RS updates are enabled then we just need to transfer 624 // the completed buffers from (a) the DirtyCardQueueSet used to hold 625 // cards that contain references that point into the collection set 626 // to (b) the DCQS used to hold the deferred RS updates 627 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs); 628 } else { 629 630 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set(); 631 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs); 632 633 int n_completed_buffers = 0; 634 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate, 635 0, 0, true)) { 636 n_completed_buffers++; 637 } 638 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers"); 639 } 640 } 641 642 // Free any completed buffers in the DirtyCardQueueSet used to hold cards 643 // which contain references that point into the collection. 644 _g1->into_cset_dirty_card_queue_set().clear(); 645 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0, 646 "all buffers should be freed"); 647 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers(); 648 649 assert(!_traversal_in_progress, "Invariant between iterations."); 650 } 651 652 class UpdateRSObjectClosure: public ObjectClosure { 653 UpdateRSOopClosure* _update_rs_oop_cl; 654 public: 655 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) : 656 _update_rs_oop_cl(update_rs_oop_cl) {} 657 void do_object(oop obj) { 658 obj->oop_iterate(_update_rs_oop_cl); 659 } 660 661 }; 662 663 class ScrubRSClosure: public HeapRegionClosure { 664 G1CollectedHeap* _g1h; 665 BitMap* _region_bm; 666 BitMap* _card_bm; 667 CardTableModRefBS* _ctbs; 668 public: 669 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) : 670 _g1h(G1CollectedHeap::heap()), 671 _region_bm(region_bm), _card_bm(card_bm), 672 _ctbs(NULL) 673 { 674 ModRefBarrierSet* bs = _g1h->mr_bs(); 675 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition"); 676 _ctbs = (CardTableModRefBS*)bs; 677 } 678 679 bool doHeapRegion(HeapRegion* r) { 680 if (!r->continuesHumongous()) { 681 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm); 682 } 683 return false; 684 } 685 }; 686 687 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { 688 ScrubRSClosure scrub_cl(region_bm, card_bm); 689 _g1->heap_region_iterate(&scrub_cl); 690 } 691 692 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, 693 int worker_num, int claim_val) { 694 ScrubRSClosure scrub_cl(region_bm, card_bm); 695 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val); 696 } 697 698 699 static IntHistogram out_of_histo(50, 50); 700 701 class TriggerClosure : public OopClosure { 702 bool _trigger; 703 public: 704 TriggerClosure() : _trigger(false) { } 705 bool value() const { return _trigger; } 706 template <class T> void do_oop_nv(T* p) { _trigger = true; } 707 virtual void do_oop(oop* p) { do_oop_nv(p); } 708 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 709 }; 710 711 class InvokeIfNotTriggeredClosure: public OopClosure { 712 TriggerClosure* _t; 713 OopClosure* _oc; 714 public: 715 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc): 716 _t(t), _oc(oc) { } 717 template <class T> void do_oop_nv(T* p) { 718 if (!_t->value()) _oc->do_oop(p); 719 } 720 virtual void do_oop(oop* p) { do_oop_nv(p); } 721 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 722 }; 723 724 class Mux2Closure : public OopClosure { 725 OopClosure* _c1; 726 OopClosure* _c2; 727 public: 728 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { } 729 template <class T> void do_oop_nv(T* p) { 730 _c1->do_oop(p); _c2->do_oop(p); 731 } 732 virtual void do_oop(oop* p) { do_oop_nv(p); } 733 virtual void do_oop(narrowOop* p) { do_oop_nv(p); } 734 }; 735 736 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i, 737 bool check_for_refs_into_cset) { 738 // Construct the region representing the card. 739 HeapWord* start = _ct_bs->addr_for(card_ptr); 740 // And find the region containing it. 741 HeapRegion* r = _g1->heap_region_containing(start); 742 assert(r != NULL, "unexpected null"); 743 744 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); 745 MemRegion dirtyRegion(start, end); 746 747 #if CARD_REPEAT_HISTO 748 init_ct_freq_table(_g1->g1_reserved_obj_bytes()); 749 ct_freq_note_card(_ct_bs->index_for(start)); 750 #endif 751 752 UpdateRSOopClosure update_rs_oop_cl(this, worker_i); 753 update_rs_oop_cl.set_from(r); 754 755 TriggerClosure trigger_cl; 756 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl); 757 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl); 758 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl); 759 760 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, 761 (check_for_refs_into_cset ? 762 (OopClosure*)&mux : 763 (OopClosure*)&update_rs_oop_cl)); 764 765 // Undirty the card. 766 *card_ptr = CardTableModRefBS::clean_card_val(); 767 // We must complete this write before we do any of the reads below. 768 OrderAccess::storeload(); 769 // And process it, being careful of unallocated portions of TLAB's. 770 771 // The region for the current card may be a young region. The 772 // current card may have been a card that was evicted from the 773 // card cache. When the card was inserted into the cache, we had 774 // determined that its region was non-young. While in the cache, 775 // the region may have been freed during a cleanup pause, reallocated 776 // and tagged as young. 777 // 778 // We wish to filter out cards for such a region but the current 779 // thread, if we're running conucrrently, may "see" the young type 780 // change at any time (so an earlier "is_young" check may pass or 781 // fail arbitrarily). We tell the iteration code to perform this 782 // filtering when it has been determined that there has been an actual 783 // allocation in this region and making it safe to check the young type. 784 bool filter_young = true; 785 786 HeapWord* stop_point = 787 r->oops_on_card_seq_iterate_careful(dirtyRegion, 788 &filter_then_update_rs_oop_cl, 789 filter_young); 790 791 // If stop_point is non-null, then we encountered an unallocated region 792 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the 793 // card and re-enqueue: if we put off the card until a GC pause, then the 794 // unallocated portion will be filled in. Alternatively, we might try 795 // the full complexity of the technique used in "regular" precleaning. 796 if (stop_point != NULL) { 797 // The card might have gotten re-dirtied and re-enqueued while we 798 // worked. (In fact, it's pretty likely.) 799 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 800 *card_ptr = CardTableModRefBS::dirty_card_val(); 801 MutexLockerEx x(Shared_DirtyCardQ_lock, 802 Mutex::_no_safepoint_check_flag); 803 DirtyCardQueue* sdcq = 804 JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); 805 sdcq->enqueue(card_ptr); 806 } 807 } else { 808 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region()); 809 _conc_refine_cards++; 810 } 811 812 return trigger_cl.value(); 813 } 814 815 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i, 816 bool check_for_refs_into_cset) { 817 // If the card is no longer dirty, nothing to do. 818 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { 819 // No need to return that this card contains refs that point 820 // into the collection set. 821 return false; 822 } 823 824 // Construct the region representing the card. 825 HeapWord* start = _ct_bs->addr_for(card_ptr); 826 // And find the region containing it. 827 HeapRegion* r = _g1->heap_region_containing(start); 828 if (r == NULL) { 829 guarantee(_g1->is_in_permanent(start), "Or else where?"); 830 // Again no need to return that this card contains refs that 831 // point into the collection set. 832 return false; // Not in the G1 heap (might be in perm, for example.) 833 } 834 // Why do we have to check here whether a card is on a young region, 835 // given that we dirty young regions and, as a result, the 836 // post-barrier is supposed to filter them out and never to enqueue 837 // them? When we allocate a new region as the "allocation region" we 838 // actually dirty its cards after we release the lock, since card 839 // dirtying while holding the lock was a performance bottleneck. So, 840 // as a result, it is possible for other threads to actually 841 // allocate objects in the region (after the acquire the lock) 842 // before all the cards on the region are dirtied. This is unlikely, 843 // and it doesn't happen often, but it can happen. So, the extra 844 // check below filters out those cards. 845 if (r->is_young()) { 846 return false; 847 } 848 // While we are processing RSet buffers during the collection, we 849 // actually don't want to scan any cards on the collection set, 850 // since we don't want to update remebered sets with entries that 851 // point into the collection set, given that live objects from the 852 // collection set are about to move and such entries will be stale 853 // very soon. This change also deals with a reliability issue which 854 // involves scanning a card in the collection set and coming across 855 // an array that was being chunked and looking malformed. Note, 856 // however, that if evacuation fails, we have to scan any objects 857 // that were not moved and create any missing entries. 858 if (r->in_collection_set()) { 859 return false; 860 } 861 862 // Should we defer processing the card? 863 // 864 // Previously the result from the insert_cache call would be 865 // either card_ptr (implying that card_ptr was currently "cold"), 866 // null (meaning we had inserted the card ptr into the "hot" 867 // cache, which had some headroom), or a "hot" card ptr 868 // extracted from the "hot" cache. 869 // 870 // Now that the _card_counts cache in the ConcurrentG1Refine 871 // instance is an evicting hash table, the result we get back 872 // could be from evicting the card ptr in an already occupied 873 // bucket (in which case we have replaced the card ptr in the 874 // bucket with card_ptr and "defer" is set to false). To avoid 875 // having a data structure (updates to which would need a lock) 876 // to hold these unprocessed dirty cards, we need to immediately 877 // process card_ptr. The actions needed to be taken on return 878 // from cache_insert are summarized in the following table: 879 // 880 // res defer action 881 // -------------------------------------------------------------- 882 // null false card evicted from _card_counts & replaced with 883 // card_ptr; evicted ptr added to hot cache. 884 // No need to process res; immediately process card_ptr 885 // 886 // null true card not evicted from _card_counts; card_ptr added 887 // to hot cache. 888 // Nothing to do. 889 // 890 // non-null false card evicted from _card_counts & replaced with 891 // card_ptr; evicted ptr is currently "cold" or 892 // caused an eviction from the hot cache. 893 // Immediately process res; process card_ptr. 894 // 895 // non-null true card not evicted from _card_counts; card_ptr is 896 // currently cold, or caused an eviction from hot 897 // cache. 898 // Immediately process res; no need to process card_ptr. 899 900 901 jbyte* res = card_ptr; 902 bool defer = false; 903 904 // This gets set to true if the card being refined has references 905 // that point into the collection set. 906 bool oops_into_cset = false; 907 908 if (_cg1r->use_cache()) { 909 jbyte* res = _cg1r->cache_insert(card_ptr, &defer); 910 if (res != NULL && (res != card_ptr || defer)) { 911 start = _ct_bs->addr_for(res); 912 r = _g1->heap_region_containing(start); 913 if (r == NULL) { 914 assert(_g1->is_in_permanent(start), "Or else where?"); 915 } else { 916 // Checking whether the region we got back from the cache 917 // is young here is inappropriate. The region could have been 918 // freed, reallocated and tagged as young while in the cache. 919 // Hence we could see its young type change at any time. 920 // 921 // Process card pointer we get back from the hot card cache. This 922 // will check whether the region containing the card is young 923 // _after_ checking that the region has been allocated from. 924 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i, 925 false /* check_for_refs_into_cset */); 926 // The above call to concurrentRefineOneCard_impl is only 927 // performed if the hot card cache is enabled. This cache is 928 // disabled during an evacuation pause - which is the only 929 // time when we need know if the card contains references 930 // that point into the collection set. Also when the hot card 931 // cache is enabled, this code is executed by the concurrent 932 // refine threads - rather than the GC worker threads - and 933 // concurrentRefineOneCard_impl will return false. 934 assert(!oops_into_cset, "should not see true here"); 935 } 936 } 937 } 938 939 if (!defer) { 940 oops_into_cset = 941 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset); 942 // We should only be detecting that the card contains references 943 // that point into the collection set if the current thread is 944 // a GC worker thread. 945 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(), 946 "invalid result at non safepoint"); 947 } 948 return oops_into_cset; 949 } 950 951 class HRRSStatsIter: public HeapRegionClosure { 952 size_t _occupied; 953 size_t _total_mem_sz; 954 size_t _max_mem_sz; 955 HeapRegion* _max_mem_sz_region; 956 public: 957 HRRSStatsIter() : 958 _occupied(0), 959 _total_mem_sz(0), 960 _max_mem_sz(0), 961 _max_mem_sz_region(NULL) 962 {} 963 964 bool doHeapRegion(HeapRegion* r) { 965 if (r->continuesHumongous()) return false; 966 size_t mem_sz = r->rem_set()->mem_size(); 967 if (mem_sz > _max_mem_sz) { 968 _max_mem_sz = mem_sz; 969 _max_mem_sz_region = r; 970 } 971 _total_mem_sz += mem_sz; 972 size_t occ = r->rem_set()->occupied(); 973 _occupied += occ; 974 return false; 975 } 976 size_t total_mem_sz() { return _total_mem_sz; } 977 size_t max_mem_sz() { return _max_mem_sz; } 978 size_t occupied() { return _occupied; } 979 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; } 980 }; 981 982 class PrintRSThreadVTimeClosure : public ThreadClosure { 983 public: 984 virtual void do_thread(Thread *t) { 985 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t; 986 gclog_or_tty->print(" %5.2f", crt->vtime_accum()); 987 } 988 }; 989 990 void G1RemSet::print_summary_info() { 991 G1CollectedHeap* g1 = G1CollectedHeap::heap(); 992 993 #if CARD_REPEAT_HISTO 994 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: "); 995 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number."); 996 card_repeat_count.print_on(gclog_or_tty); 997 #endif 998 999 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) { 1000 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: "); 1001 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number."); 1002 out_of_histo.print_on(gclog_or_tty); 1003 } 1004 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards", 1005 _conc_refine_cards); 1006 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 1007 jint tot_processed_buffers = 1008 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread(); 1009 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers); 1010 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.", 1011 dcqs.processed_buffers_rs_thread(), 1012 100.0*(float)dcqs.processed_buffers_rs_thread()/ 1013 (float)tot_processed_buffers); 1014 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.", 1015 dcqs.processed_buffers_mut(), 1016 100.0*(float)dcqs.processed_buffers_mut()/ 1017 (float)tot_processed_buffers); 1018 gclog_or_tty->print_cr(" Conc RS threads times(s)"); 1019 PrintRSThreadVTimeClosure p; 1020 gclog_or_tty->print(" "); 1021 g1->concurrent_g1_refine()->threads_do(&p); 1022 gclog_or_tty->print_cr(""); 1023 1024 HRRSStatsIter blk; 1025 g1->heap_region_iterate(&blk); 1026 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." 1027 " Max = " SIZE_FORMAT "K.", 1028 blk.total_mem_sz()/K, blk.max_mem_sz()/K); 1029 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," 1030 " free_lists = " SIZE_FORMAT "K.", 1031 HeapRegionRemSet::static_mem_size()/K, 1032 HeapRegionRemSet::fl_mem_size()/K); 1033 gclog_or_tty->print_cr(" %d occupied cards represented.", 1034 blk.occupied()); 1035 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" 1036 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", 1037 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), 1038 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, 1039 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); 1040 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings()); 1041 } 1042 1043 void G1RemSet::prepare_for_verify() { 1044 if (G1HRRSFlushLogBuffersOnVerify && 1045 (VerifyBeforeGC || VerifyAfterGC) 1046 && !_g1->full_collection()) { 1047 cleanupHRRS(); 1048 _g1->set_refine_cte_cl_concurrency(false); 1049 if (SafepointSynchronize::is_at_safepoint()) { 1050 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 1051 dcqs.concatenate_logs(); 1052 } 1053 bool cg1r_use_cache = _cg1r->use_cache(); 1054 _cg1r->set_use_cache(false); 1055 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set()); 1056 updateRS(&into_cset_dcq, 0); 1057 _g1->into_cset_dirty_card_queue_set().clear(); 1058 _cg1r->set_use_cache(cg1r_use_cache); 1059 1060 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); 1061 } 1062 }