1 /* 2 * Copyright (c) 2001, 2015, 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 "classfile/metadataOnStackMark.hpp" 27 #include "classfile/symbolTable.hpp" 28 #include "code/codeCache.hpp" 29 #include "gc/g1/concurrentMark.inline.hpp" 30 #include "gc/g1/concurrentMarkThread.inline.hpp" 31 #include "gc/g1/g1CollectedHeap.inline.hpp" 32 #include "gc/g1/g1CollectorPolicy.hpp" 33 #include "gc/g1/g1CollectorState.hpp" 34 #include "gc/g1/g1ErgoVerbose.hpp" 35 #include "gc/g1/g1Log.hpp" 36 #include "gc/g1/g1OopClosures.inline.hpp" 37 #include "gc/g1/g1RemSet.hpp" 38 #include "gc/g1/g1StringDedup.hpp" 39 #include "gc/g1/heapRegion.inline.hpp" 40 #include "gc/g1/heapRegionManager.inline.hpp" 41 #include "gc/g1/heapRegionRemSet.hpp" 42 #include "gc/g1/heapRegionSet.inline.hpp" 43 #include "gc/g1/suspendibleThreadSet.hpp" 44 #include "gc/shared/gcId.hpp" 45 #include "gc/shared/gcTimer.hpp" 46 #include "gc/shared/gcTrace.hpp" 47 #include "gc/shared/gcTraceTime.hpp" 48 #include "gc/shared/genOopClosures.inline.hpp" 49 #include "gc/shared/referencePolicy.hpp" 50 #include "gc/shared/strongRootsScope.hpp" 51 #include "gc/shared/taskqueue.inline.hpp" 52 #include "gc/shared/vmGCOperations.hpp" 53 #include "memory/allocation.hpp" 54 #include "memory/resourceArea.hpp" 55 #include "oops/oop.inline.hpp" 56 #include "runtime/atomic.inline.hpp" 57 #include "runtime/handles.inline.hpp" 58 #include "runtime/java.hpp" 59 #include "runtime/prefetch.inline.hpp" 60 #include "services/memTracker.hpp" 61 62 // Concurrent marking bit map wrapper 63 64 CMBitMapRO::CMBitMapRO(int shifter) : 65 _bm(), 66 _shifter(shifter) { 67 _bmStartWord = 0; 68 _bmWordSize = 0; 69 } 70 71 HeapWord* CMBitMapRO::getNextMarkedWordAddress(const HeapWord* addr, 72 const HeapWord* limit) const { 73 // First we must round addr *up* to a possible object boundary. 74 addr = (HeapWord*)align_size_up((intptr_t)addr, 75 HeapWordSize << _shifter); 76 size_t addrOffset = heapWordToOffset(addr); 77 if (limit == NULL) { 78 limit = _bmStartWord + _bmWordSize; 79 } 80 size_t limitOffset = heapWordToOffset(limit); 81 size_t nextOffset = _bm.get_next_one_offset(addrOffset, limitOffset); 82 HeapWord* nextAddr = offsetToHeapWord(nextOffset); 83 assert(nextAddr >= addr, "get_next_one postcondition"); 84 assert(nextAddr == limit || isMarked(nextAddr), 85 "get_next_one postcondition"); 86 return nextAddr; 87 } 88 89 HeapWord* CMBitMapRO::getNextUnmarkedWordAddress(const HeapWord* addr, 90 const HeapWord* limit) const { 91 size_t addrOffset = heapWordToOffset(addr); 92 if (limit == NULL) { 93 limit = _bmStartWord + _bmWordSize; 94 } 95 size_t limitOffset = heapWordToOffset(limit); 96 size_t nextOffset = _bm.get_next_zero_offset(addrOffset, limitOffset); 97 HeapWord* nextAddr = offsetToHeapWord(nextOffset); 98 assert(nextAddr >= addr, "get_next_one postcondition"); 99 assert(nextAddr == limit || !isMarked(nextAddr), 100 "get_next_one postcondition"); 101 return nextAddr; 102 } 103 104 int CMBitMapRO::heapWordDiffToOffsetDiff(size_t diff) const { 105 assert((diff & ((1 << _shifter) - 1)) == 0, "argument check"); 106 return (int) (diff >> _shifter); 107 } 108 109 #ifndef PRODUCT 110 bool CMBitMapRO::covers(MemRegion heap_rs) const { 111 // assert(_bm.map() == _virtual_space.low(), "map inconsistency"); 112 assert(((size_t)_bm.size() * ((size_t)1 << _shifter)) == _bmWordSize, 113 "size inconsistency"); 114 return _bmStartWord == (HeapWord*)(heap_rs.start()) && 115 _bmWordSize == heap_rs.word_size(); 116 } 117 #endif 118 119 void CMBitMapRO::print_on_error(outputStream* st, const char* prefix) const { 120 _bm.print_on_error(st, prefix); 121 } 122 123 size_t CMBitMap::compute_size(size_t heap_size) { 124 return ReservedSpace::allocation_align_size_up(heap_size / mark_distance()); 125 } 126 127 size_t CMBitMap::mark_distance() { 128 return MinObjAlignmentInBytes * BitsPerByte; 129 } 130 131 void CMBitMap::initialize(MemRegion heap, G1RegionToSpaceMapper* storage) { 132 _bmStartWord = heap.start(); 133 _bmWordSize = heap.word_size(); 134 135 _bm.set_map((BitMap::bm_word_t*) storage->reserved().start()); 136 _bm.set_size(_bmWordSize >> _shifter); 137 138 storage->set_mapping_changed_listener(&_listener); 139 } 140 141 void CMBitMapMappingChangedListener::on_commit(uint start_region, size_t num_regions, bool zero_filled) { 142 if (zero_filled) { 143 return; 144 } 145 // We need to clear the bitmap on commit, removing any existing information. 146 MemRegion mr(G1CollectedHeap::heap()->bottom_addr_for_region(start_region), num_regions * HeapRegion::GrainWords); 147 _bm->clearRange(mr); 148 } 149 150 // Closure used for clearing the given mark bitmap. 151 class ClearBitmapHRClosure : public HeapRegionClosure { 152 private: 153 ConcurrentMark* _cm; 154 CMBitMap* _bitmap; 155 bool _may_yield; // The closure may yield during iteration. If yielded, abort the iteration. 156 public: 157 ClearBitmapHRClosure(ConcurrentMark* cm, CMBitMap* bitmap, bool may_yield) : HeapRegionClosure(), _cm(cm), _bitmap(bitmap), _may_yield(may_yield) { 158 assert(!may_yield || cm != NULL, "CM must be non-NULL if this closure is expected to yield."); 159 } 160 161 virtual bool doHeapRegion(HeapRegion* r) { 162 size_t const chunk_size_in_words = M / HeapWordSize; 163 164 HeapWord* cur = r->bottom(); 165 HeapWord* const end = r->end(); 166 167 while (cur < end) { 168 MemRegion mr(cur, MIN2(cur + chunk_size_in_words, end)); 169 _bitmap->clearRange(mr); 170 171 cur += chunk_size_in_words; 172 173 // Abort iteration if after yielding the marking has been aborted. 174 if (_may_yield && _cm->do_yield_check() && _cm->has_aborted()) { 175 return true; 176 } 177 // Repeat the asserts from before the start of the closure. We will do them 178 // as asserts here to minimize their overhead on the product. However, we 179 // will have them as guarantees at the beginning / end of the bitmap 180 // clearing to get some checking in the product. 181 assert(!_may_yield || _cm->cmThread()->during_cycle(), "invariant"); 182 assert(!_may_yield || !G1CollectedHeap::heap()->collector_state()->mark_in_progress(), "invariant"); 183 } 184 185 return false; 186 } 187 }; 188 189 class ParClearNextMarkBitmapTask : public AbstractGangTask { 190 ClearBitmapHRClosure* _cl; 191 HeapRegionClaimer _hrclaimer; 192 bool _suspendible; // If the task is suspendible, workers must join the STS. 193 194 public: 195 ParClearNextMarkBitmapTask(ClearBitmapHRClosure *cl, uint n_workers, bool suspendible) : 196 _cl(cl), _suspendible(suspendible), AbstractGangTask("Parallel Clear Bitmap Task"), _hrclaimer(n_workers) {} 197 198 void work(uint worker_id) { 199 SuspendibleThreadSetJoiner sts_join(_suspendible); 200 G1CollectedHeap::heap()->heap_region_par_iterate(_cl, worker_id, &_hrclaimer, true); 201 } 202 }; 203 204 void CMBitMap::clearAll() { 205 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 206 ClearBitmapHRClosure cl(NULL, this, false /* may_yield */); 207 uint n_workers = g1h->workers()->active_workers(); 208 ParClearNextMarkBitmapTask task(&cl, n_workers, false); 209 g1h->workers()->run_task(&task); 210 guarantee(cl.complete(), "Must have completed iteration."); 211 return; 212 } 213 214 void CMBitMap::markRange(MemRegion mr) { 215 mr.intersection(MemRegion(_bmStartWord, _bmWordSize)); 216 assert(!mr.is_empty(), "unexpected empty region"); 217 assert((offsetToHeapWord(heapWordToOffset(mr.end())) == 218 ((HeapWord *) mr.end())), 219 "markRange memory region end is not card aligned"); 220 // convert address range into offset range 221 _bm.at_put_range(heapWordToOffset(mr.start()), 222 heapWordToOffset(mr.end()), true); 223 } 224 225 void CMBitMap::clearRange(MemRegion mr) { 226 mr.intersection(MemRegion(_bmStartWord, _bmWordSize)); 227 assert(!mr.is_empty(), "unexpected empty region"); 228 // convert address range into offset range 229 _bm.at_put_range(heapWordToOffset(mr.start()), 230 heapWordToOffset(mr.end()), false); 231 } 232 233 MemRegion CMBitMap::getAndClearMarkedRegion(HeapWord* addr, 234 HeapWord* end_addr) { 235 HeapWord* start = getNextMarkedWordAddress(addr); 236 start = MIN2(start, end_addr); 237 HeapWord* end = getNextUnmarkedWordAddress(start); 238 end = MIN2(end, end_addr); 239 assert(start <= end, "Consistency check"); 240 MemRegion mr(start, end); 241 if (!mr.is_empty()) { 242 clearRange(mr); 243 } 244 return mr; 245 } 246 247 CMMarkStack::CMMarkStack(ConcurrentMark* cm) : 248 _base(NULL), _cm(cm) 249 {} 250 251 bool CMMarkStack::allocate(size_t capacity) { 252 // allocate a stack of the requisite depth 253 ReservedSpace rs(ReservedSpace::allocation_align_size_up(capacity * sizeof(oop))); 254 if (!rs.is_reserved()) { 255 warning("ConcurrentMark MarkStack allocation failure"); 256 return false; 257 } 258 MemTracker::record_virtual_memory_type((address)rs.base(), mtGC); 259 if (!_virtual_space.initialize(rs, rs.size())) { 260 warning("ConcurrentMark MarkStack backing store failure"); 261 // Release the virtual memory reserved for the marking stack 262 rs.release(); 263 return false; 264 } 265 assert(_virtual_space.committed_size() == rs.size(), 266 "Didn't reserve backing store for all of ConcurrentMark stack?"); 267 _base = (oop*) _virtual_space.low(); 268 setEmpty(); 269 _capacity = (jint) capacity; 270 _saved_index = -1; 271 _should_expand = false; 272 return true; 273 } 274 275 void CMMarkStack::expand() { 276 // Called, during remark, if we've overflown the marking stack during marking. 277 assert(isEmpty(), "stack should been emptied while handling overflow"); 278 assert(_capacity <= (jint) MarkStackSizeMax, "stack bigger than permitted"); 279 // Clear expansion flag 280 _should_expand = false; 281 if (_capacity == (jint) MarkStackSizeMax) { 282 if (PrintGCDetails && Verbose) { 283 gclog_or_tty->print_cr(" (benign) Can't expand marking stack capacity, at max size limit"); 284 } 285 return; 286 } 287 // Double capacity if possible 288 jint new_capacity = MIN2(_capacity*2, (jint) MarkStackSizeMax); 289 // Do not give up existing stack until we have managed to 290 // get the double capacity that we desired. 291 ReservedSpace rs(ReservedSpace::allocation_align_size_up(new_capacity * 292 sizeof(oop))); 293 if (rs.is_reserved()) { 294 // Release the backing store associated with old stack 295 _virtual_space.release(); 296 // Reinitialize virtual space for new stack 297 if (!_virtual_space.initialize(rs, rs.size())) { 298 fatal("Not enough swap for expanded marking stack capacity"); 299 } 300 _base = (oop*)(_virtual_space.low()); 301 _index = 0; 302 _capacity = new_capacity; 303 } else { 304 if (PrintGCDetails && Verbose) { 305 // Failed to double capacity, continue; 306 gclog_or_tty->print(" (benign) Failed to expand marking stack capacity from " 307 SIZE_FORMAT "K to " SIZE_FORMAT "K", 308 _capacity / K, new_capacity / K); 309 } 310 } 311 } 312 313 void CMMarkStack::set_should_expand() { 314 // If we're resetting the marking state because of an 315 // marking stack overflow, record that we should, if 316 // possible, expand the stack. 317 _should_expand = _cm->has_overflown(); 318 } 319 320 CMMarkStack::~CMMarkStack() { 321 if (_base != NULL) { 322 _base = NULL; 323 _virtual_space.release(); 324 } 325 } 326 327 void CMMarkStack::par_push_arr(oop* ptr_arr, int n) { 328 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag); 329 jint start = _index; 330 jint next_index = start + n; 331 if (next_index > _capacity) { 332 _overflow = true; 333 return; 334 } 335 // Otherwise. 336 _index = next_index; 337 for (int i = 0; i < n; i++) { 338 int ind = start + i; 339 assert(ind < _capacity, "By overflow test above."); 340 _base[ind] = ptr_arr[i]; 341 } 342 } 343 344 bool CMMarkStack::par_pop_arr(oop* ptr_arr, int max, int* n) { 345 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag); 346 jint index = _index; 347 if (index == 0) { 348 *n = 0; 349 return false; 350 } else { 351 int k = MIN2(max, index); 352 jint new_ind = index - k; 353 for (int j = 0; j < k; j++) { 354 ptr_arr[j] = _base[new_ind + j]; 355 } 356 _index = new_ind; 357 *n = k; 358 return true; 359 } 360 } 361 362 void CMMarkStack::note_start_of_gc() { 363 assert(_saved_index == -1, 364 "note_start_of_gc()/end_of_gc() bracketed incorrectly"); 365 _saved_index = _index; 366 } 367 368 void CMMarkStack::note_end_of_gc() { 369 // This is intentionally a guarantee, instead of an assert. If we 370 // accidentally add something to the mark stack during GC, it 371 // will be a correctness issue so it's better if we crash. we'll 372 // only check this once per GC anyway, so it won't be a performance 373 // issue in any way. 374 guarantee(_saved_index == _index, 375 "saved index: %d index: %d", _saved_index, _index); 376 _saved_index = -1; 377 } 378 379 CMRootRegions::CMRootRegions() : 380 _young_list(NULL), _cm(NULL), _scan_in_progress(false), 381 _should_abort(false), _next_survivor(NULL) { } 382 383 void CMRootRegions::init(G1CollectedHeap* g1h, ConcurrentMark* cm) { 384 _young_list = g1h->young_list(); 385 _cm = cm; 386 } 387 388 void CMRootRegions::prepare_for_scan() { 389 assert(!scan_in_progress(), "pre-condition"); 390 391 // Currently, only survivors can be root regions. 392 assert(_next_survivor == NULL, "pre-condition"); 393 _next_survivor = _young_list->first_survivor_region(); 394 _scan_in_progress = (_next_survivor != NULL); 395 _should_abort = false; 396 } 397 398 HeapRegion* CMRootRegions::claim_next() { 399 if (_should_abort) { 400 // If someone has set the should_abort flag, we return NULL to 401 // force the caller to bail out of their loop. 402 return NULL; 403 } 404 405 // Currently, only survivors can be root regions. 406 HeapRegion* res = _next_survivor; 407 if (res != NULL) { 408 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag); 409 // Read it again in case it changed while we were waiting for the lock. 410 res = _next_survivor; 411 if (res != NULL) { 412 if (res == _young_list->last_survivor_region()) { 413 // We just claimed the last survivor so store NULL to indicate 414 // that we're done. 415 _next_survivor = NULL; 416 } else { 417 _next_survivor = res->get_next_young_region(); 418 } 419 } else { 420 // Someone else claimed the last survivor while we were trying 421 // to take the lock so nothing else to do. 422 } 423 } 424 assert(res == NULL || res->is_survivor(), "post-condition"); 425 426 return res; 427 } 428 429 void CMRootRegions::scan_finished() { 430 assert(scan_in_progress(), "pre-condition"); 431 432 // Currently, only survivors can be root regions. 433 if (!_should_abort) { 434 assert(_next_survivor == NULL, "we should have claimed all survivors"); 435 } 436 _next_survivor = NULL; 437 438 { 439 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag); 440 _scan_in_progress = false; 441 RootRegionScan_lock->notify_all(); 442 } 443 } 444 445 bool CMRootRegions::wait_until_scan_finished() { 446 if (!scan_in_progress()) return false; 447 448 { 449 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag); 450 while (scan_in_progress()) { 451 RootRegionScan_lock->wait(Mutex::_no_safepoint_check_flag); 452 } 453 } 454 return true; 455 } 456 457 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away 458 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list 459 #endif // _MSC_VER 460 461 uint ConcurrentMark::scale_parallel_threads(uint n_par_threads) { 462 return MAX2((n_par_threads + 2) / 4, 1U); 463 } 464 465 ConcurrentMark::ConcurrentMark(G1CollectedHeap* g1h, G1RegionToSpaceMapper* prev_bitmap_storage, G1RegionToSpaceMapper* next_bitmap_storage) : 466 _g1h(g1h), 467 _markBitMap1(), 468 _markBitMap2(), 469 _parallel_marking_threads(0), 470 _max_parallel_marking_threads(0), 471 _sleep_factor(0.0), 472 _marking_task_overhead(1.0), 473 _cleanup_sleep_factor(0.0), 474 _cleanup_task_overhead(1.0), 475 _cleanup_list("Cleanup List"), 476 _region_bm((BitMap::idx_t)(g1h->max_regions()), false /* in_resource_area*/), 477 _card_bm((g1h->reserved_region().byte_size() + CardTableModRefBS::card_size - 1) >> 478 CardTableModRefBS::card_shift, 479 false /* in_resource_area*/), 480 481 _prevMarkBitMap(&_markBitMap1), 482 _nextMarkBitMap(&_markBitMap2), 483 484 _markStack(this), 485 // _finger set in set_non_marking_state 486 487 _max_worker_id(ParallelGCThreads), 488 // _active_tasks set in set_non_marking_state 489 // _tasks set inside the constructor 490 _task_queues(new CMTaskQueueSet((int) _max_worker_id)), 491 _terminator(ParallelTaskTerminator((int) _max_worker_id, _task_queues)), 492 493 _has_overflown(false), 494 _concurrent(false), 495 _has_aborted(false), 496 _restart_for_overflow(false), 497 _concurrent_marking_in_progress(false), 498 499 // _verbose_level set below 500 501 _init_times(), 502 _remark_times(), _remark_mark_times(), _remark_weak_ref_times(), 503 _cleanup_times(), 504 _total_counting_time(0.0), 505 _total_rs_scrub_time(0.0), 506 507 _parallel_workers(NULL), 508 509 _count_card_bitmaps(NULL), 510 _count_marked_bytes(NULL), 511 _completed_initialization(false) { 512 513 _markBitMap1.initialize(g1h->reserved_region(), prev_bitmap_storage); 514 _markBitMap2.initialize(g1h->reserved_region(), next_bitmap_storage); 515 516 // Create & start a ConcurrentMark thread. 517 _cmThread = new ConcurrentMarkThread(this); 518 assert(cmThread() != NULL, "CM Thread should have been created"); 519 assert(cmThread()->cm() != NULL, "CM Thread should refer to this cm"); 520 if (_cmThread->osthread() == NULL) { 521 vm_shutdown_during_initialization("Could not create ConcurrentMarkThread"); 522 } 523 524 assert(CGC_lock != NULL, "Where's the CGC_lock?"); 525 assert(_markBitMap1.covers(g1h->reserved_region()), "_markBitMap1 inconsistency"); 526 assert(_markBitMap2.covers(g1h->reserved_region()), "_markBitMap2 inconsistency"); 527 528 SATBMarkQueueSet& satb_qs = JavaThread::satb_mark_queue_set(); 529 satb_qs.set_buffer_size(G1SATBBufferSize); 530 531 _root_regions.init(_g1h, this); 532 533 if (ConcGCThreads > ParallelGCThreads) { 534 warning("Can't have more ConcGCThreads (%u) " 535 "than ParallelGCThreads (%u).", 536 ConcGCThreads, ParallelGCThreads); 537 return; 538 } 539 if (!FLAG_IS_DEFAULT(ConcGCThreads) && ConcGCThreads > 0) { 540 // Note: ConcGCThreads has precedence over G1MarkingOverheadPercent 541 // if both are set 542 _sleep_factor = 0.0; 543 _marking_task_overhead = 1.0; 544 } else if (G1MarkingOverheadPercent > 0) { 545 // We will calculate the number of parallel marking threads based 546 // on a target overhead with respect to the soft real-time goal 547 double marking_overhead = (double) G1MarkingOverheadPercent / 100.0; 548 double overall_cm_overhead = 549 (double) MaxGCPauseMillis * marking_overhead / 550 (double) GCPauseIntervalMillis; 551 double cpu_ratio = 1.0 / (double) os::processor_count(); 552 double marking_thread_num = ceil(overall_cm_overhead / cpu_ratio); 553 double marking_task_overhead = 554 overall_cm_overhead / marking_thread_num * 555 (double) os::processor_count(); 556 double sleep_factor = 557 (1.0 - marking_task_overhead) / marking_task_overhead; 558 559 FLAG_SET_ERGO(uint, ConcGCThreads, (uint) marking_thread_num); 560 _sleep_factor = sleep_factor; 561 _marking_task_overhead = marking_task_overhead; 562 } else { 563 // Calculate the number of parallel marking threads by scaling 564 // the number of parallel GC threads. 565 uint marking_thread_num = scale_parallel_threads(ParallelGCThreads); 566 FLAG_SET_ERGO(uint, ConcGCThreads, marking_thread_num); 567 _sleep_factor = 0.0; 568 _marking_task_overhead = 1.0; 569 } 570 571 assert(ConcGCThreads > 0, "Should have been set"); 572 _parallel_marking_threads = ConcGCThreads; 573 _max_parallel_marking_threads = _parallel_marking_threads; 574 575 if (parallel_marking_threads() > 1) { 576 _cleanup_task_overhead = 1.0; 577 } else { 578 _cleanup_task_overhead = marking_task_overhead(); 579 } 580 _cleanup_sleep_factor = 581 (1.0 - cleanup_task_overhead()) / cleanup_task_overhead(); 582 583 #if 0 584 gclog_or_tty->print_cr("Marking Threads %d", parallel_marking_threads()); 585 gclog_or_tty->print_cr("CM Marking Task Overhead %1.4lf", marking_task_overhead()); 586 gclog_or_tty->print_cr("CM Sleep Factor %1.4lf", sleep_factor()); 587 gclog_or_tty->print_cr("CL Marking Task Overhead %1.4lf", cleanup_task_overhead()); 588 gclog_or_tty->print_cr("CL Sleep Factor %1.4lf", cleanup_sleep_factor()); 589 #endif 590 591 _parallel_workers = new WorkGang("G1 Marker", 592 _max_parallel_marking_threads, false, true); 593 if (_parallel_workers == NULL) { 594 vm_exit_during_initialization("Failed necessary allocation."); 595 } else { 596 _parallel_workers->initialize_workers(); 597 } 598 599 if (FLAG_IS_DEFAULT(MarkStackSize)) { 600 size_t mark_stack_size = 601 MIN2(MarkStackSizeMax, 602 MAX2(MarkStackSize, (size_t) (parallel_marking_threads() * TASKQUEUE_SIZE))); 603 // Verify that the calculated value for MarkStackSize is in range. 604 // It would be nice to use the private utility routine from Arguments. 605 if (!(mark_stack_size >= 1 && mark_stack_size <= MarkStackSizeMax)) { 606 warning("Invalid value calculated for MarkStackSize (" SIZE_FORMAT "): " 607 "must be between 1 and " SIZE_FORMAT, 608 mark_stack_size, MarkStackSizeMax); 609 return; 610 } 611 FLAG_SET_ERGO(size_t, MarkStackSize, mark_stack_size); 612 } else { 613 // Verify MarkStackSize is in range. 614 if (FLAG_IS_CMDLINE(MarkStackSize)) { 615 if (FLAG_IS_DEFAULT(MarkStackSizeMax)) { 616 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) { 617 warning("Invalid value specified for MarkStackSize (" SIZE_FORMAT "): " 618 "must be between 1 and " SIZE_FORMAT, 619 MarkStackSize, MarkStackSizeMax); 620 return; 621 } 622 } else if (FLAG_IS_CMDLINE(MarkStackSizeMax)) { 623 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) { 624 warning("Invalid value specified for MarkStackSize (" SIZE_FORMAT ")" 625 " or for MarkStackSizeMax (" SIZE_FORMAT ")", 626 MarkStackSize, MarkStackSizeMax); 627 return; 628 } 629 } 630 } 631 } 632 633 if (!_markStack.allocate(MarkStackSize)) { 634 warning("Failed to allocate CM marking stack"); 635 return; 636 } 637 638 _tasks = NEW_C_HEAP_ARRAY(CMTask*, _max_worker_id, mtGC); 639 _accum_task_vtime = NEW_C_HEAP_ARRAY(double, _max_worker_id, mtGC); 640 641 _count_card_bitmaps = NEW_C_HEAP_ARRAY(BitMap, _max_worker_id, mtGC); 642 _count_marked_bytes = NEW_C_HEAP_ARRAY(size_t*, _max_worker_id, mtGC); 643 644 BitMap::idx_t card_bm_size = _card_bm.size(); 645 646 // so that the assertion in MarkingTaskQueue::task_queue doesn't fail 647 _active_tasks = _max_worker_id; 648 649 uint max_regions = _g1h->max_regions(); 650 for (uint i = 0; i < _max_worker_id; ++i) { 651 CMTaskQueue* task_queue = new CMTaskQueue(); 652 task_queue->initialize(); 653 _task_queues->register_queue(i, task_queue); 654 655 _count_card_bitmaps[i] = BitMap(card_bm_size, false); 656 _count_marked_bytes[i] = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC); 657 658 _tasks[i] = new CMTask(i, this, 659 _count_marked_bytes[i], 660 &_count_card_bitmaps[i], 661 task_queue, _task_queues); 662 663 _accum_task_vtime[i] = 0.0; 664 } 665 666 // Calculate the card number for the bottom of the heap. Used 667 // in biasing indexes into the accounting card bitmaps. 668 _heap_bottom_card_num = 669 intptr_t(uintptr_t(_g1h->reserved_region().start()) >> 670 CardTableModRefBS::card_shift); 671 672 // Clear all the liveness counting data 673 clear_all_count_data(); 674 675 // so that the call below can read a sensible value 676 _heap_start = g1h->reserved_region().start(); 677 set_non_marking_state(); 678 _completed_initialization = true; 679 } 680 681 void ConcurrentMark::reset() { 682 // Starting values for these two. This should be called in a STW 683 // phase. 684 MemRegion reserved = _g1h->g1_reserved(); 685 _heap_start = reserved.start(); 686 _heap_end = reserved.end(); 687 688 // Separated the asserts so that we know which one fires. 689 assert(_heap_start != NULL, "heap bounds should look ok"); 690 assert(_heap_end != NULL, "heap bounds should look ok"); 691 assert(_heap_start < _heap_end, "heap bounds should look ok"); 692 693 // Reset all the marking data structures and any necessary flags 694 reset_marking_state(); 695 696 // We do reset all of them, since different phases will use 697 // different number of active threads. So, it's easiest to have all 698 // of them ready. 699 for (uint i = 0; i < _max_worker_id; ++i) { 700 _tasks[i]->reset(_nextMarkBitMap); 701 } 702 703 // we need this to make sure that the flag is on during the evac 704 // pause with initial mark piggy-backed 705 set_concurrent_marking_in_progress(); 706 } 707 708 709 void ConcurrentMark::reset_marking_state(bool clear_overflow) { 710 _markStack.set_should_expand(); 711 _markStack.setEmpty(); // Also clears the _markStack overflow flag 712 if (clear_overflow) { 713 clear_has_overflown(); 714 } else { 715 assert(has_overflown(), "pre-condition"); 716 } 717 _finger = _heap_start; 718 719 for (uint i = 0; i < _max_worker_id; ++i) { 720 CMTaskQueue* queue = _task_queues->queue(i); 721 queue->set_empty(); 722 } 723 } 724 725 void ConcurrentMark::set_concurrency(uint active_tasks) { 726 assert(active_tasks <= _max_worker_id, "we should not have more"); 727 728 _active_tasks = active_tasks; 729 // Need to update the three data structures below according to the 730 // number of active threads for this phase. 731 _terminator = ParallelTaskTerminator((int) active_tasks, _task_queues); 732 _first_overflow_barrier_sync.set_n_workers((int) active_tasks); 733 _second_overflow_barrier_sync.set_n_workers((int) active_tasks); 734 } 735 736 void ConcurrentMark::set_concurrency_and_phase(uint active_tasks, bool concurrent) { 737 set_concurrency(active_tasks); 738 739 _concurrent = concurrent; 740 // We propagate this to all tasks, not just the active ones. 741 for (uint i = 0; i < _max_worker_id; ++i) 742 _tasks[i]->set_concurrent(concurrent); 743 744 if (concurrent) { 745 set_concurrent_marking_in_progress(); 746 } else { 747 // We currently assume that the concurrent flag has been set to 748 // false before we start remark. At this point we should also be 749 // in a STW phase. 750 assert(!concurrent_marking_in_progress(), "invariant"); 751 assert(out_of_regions(), 752 "only way to get here: _finger: " PTR_FORMAT ", _heap_end: " PTR_FORMAT, 753 p2i(_finger), p2i(_heap_end)); 754 } 755 } 756 757 void ConcurrentMark::set_non_marking_state() { 758 // We set the global marking state to some default values when we're 759 // not doing marking. 760 reset_marking_state(); 761 _active_tasks = 0; 762 clear_concurrent_marking_in_progress(); 763 } 764 765 ConcurrentMark::~ConcurrentMark() { 766 // The ConcurrentMark instance is never freed. 767 ShouldNotReachHere(); 768 } 769 770 void ConcurrentMark::clearNextBitmap() { 771 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 772 773 // Make sure that the concurrent mark thread looks to still be in 774 // the current cycle. 775 guarantee(cmThread()->during_cycle(), "invariant"); 776 777 // We are finishing up the current cycle by clearing the next 778 // marking bitmap and getting it ready for the next cycle. During 779 // this time no other cycle can start. So, let's make sure that this 780 // is the case. 781 guarantee(!g1h->collector_state()->mark_in_progress(), "invariant"); 782 783 ClearBitmapHRClosure cl(this, _nextMarkBitMap, true /* may_yield */); 784 ParClearNextMarkBitmapTask task(&cl, parallel_marking_threads(), true); 785 _parallel_workers->run_task(&task); 786 787 // Clear the liveness counting data. If the marking has been aborted, the abort() 788 // call already did that. 789 if (cl.complete()) { 790 clear_all_count_data(); 791 } 792 793 // Repeat the asserts from above. 794 guarantee(cmThread()->during_cycle(), "invariant"); 795 guarantee(!g1h->collector_state()->mark_in_progress(), "invariant"); 796 } 797 798 class CheckBitmapClearHRClosure : public HeapRegionClosure { 799 CMBitMap* _bitmap; 800 bool _error; 801 public: 802 CheckBitmapClearHRClosure(CMBitMap* bitmap) : _bitmap(bitmap) { 803 } 804 805 virtual bool doHeapRegion(HeapRegion* r) { 806 // This closure can be called concurrently to the mutator, so we must make sure 807 // that the result of the getNextMarkedWordAddress() call is compared to the 808 // value passed to it as limit to detect any found bits. 809 // We can use the region's orig_end() for the limit and the comparison value 810 // as it always contains the "real" end of the region that never changes and 811 // has no side effects. 812 // Due to the latter, there can also be no problem with the compiler generating 813 // reloads of the orig_end() call. 814 HeapWord* end = r->orig_end(); 815 return _bitmap->getNextMarkedWordAddress(r->bottom(), end) != end; 816 } 817 }; 818 819 bool ConcurrentMark::nextMarkBitmapIsClear() { 820 CheckBitmapClearHRClosure cl(_nextMarkBitMap); 821 _g1h->heap_region_iterate(&cl); 822 return cl.complete(); 823 } 824 825 class NoteStartOfMarkHRClosure: public HeapRegionClosure { 826 public: 827 bool doHeapRegion(HeapRegion* r) { 828 if (!r->is_continues_humongous()) { 829 r->note_start_of_marking(); 830 } 831 return false; 832 } 833 }; 834 835 void ConcurrentMark::checkpointRootsInitialPre() { 836 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 837 G1CollectorPolicy* g1p = g1h->g1_policy(); 838 839 _has_aborted = false; 840 841 // Initialize marking structures. This has to be done in a STW phase. 842 reset(); 843 844 // For each region note start of marking. 845 NoteStartOfMarkHRClosure startcl; 846 g1h->heap_region_iterate(&startcl); 847 } 848 849 850 void ConcurrentMark::checkpointRootsInitialPost() { 851 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 852 853 // If we force an overflow during remark, the remark operation will 854 // actually abort and we'll restart concurrent marking. If we always 855 // force an overflow during remark we'll never actually complete the 856 // marking phase. So, we initialize this here, at the start of the 857 // cycle, so that at the remaining overflow number will decrease at 858 // every remark and we'll eventually not need to cause one. 859 force_overflow_stw()->init(); 860 861 // Start Concurrent Marking weak-reference discovery. 862 ReferenceProcessor* rp = g1h->ref_processor_cm(); 863 // enable ("weak") refs discovery 864 rp->enable_discovery(); 865 rp->setup_policy(false); // snapshot the soft ref policy to be used in this cycle 866 867 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 868 // This is the start of the marking cycle, we're expected all 869 // threads to have SATB queues with active set to false. 870 satb_mq_set.set_active_all_threads(true, /* new active value */ 871 false /* expected_active */); 872 873 _root_regions.prepare_for_scan(); 874 875 // update_g1_committed() will be called at the end of an evac pause 876 // when marking is on. So, it's also called at the end of the 877 // initial-mark pause to update the heap end, if the heap expands 878 // during it. No need to call it here. 879 } 880 881 /* 882 * Notice that in the next two methods, we actually leave the STS 883 * during the barrier sync and join it immediately afterwards. If we 884 * do not do this, the following deadlock can occur: one thread could 885 * be in the barrier sync code, waiting for the other thread to also 886 * sync up, whereas another one could be trying to yield, while also 887 * waiting for the other threads to sync up too. 888 * 889 * Note, however, that this code is also used during remark and in 890 * this case we should not attempt to leave / enter the STS, otherwise 891 * we'll either hit an assert (debug / fastdebug) or deadlock 892 * (product). So we should only leave / enter the STS if we are 893 * operating concurrently. 894 * 895 * Because the thread that does the sync barrier has left the STS, it 896 * is possible to be suspended for a Full GC or an evacuation pause 897 * could occur. This is actually safe, since the entering the sync 898 * barrier is one of the last things do_marking_step() does, and it 899 * doesn't manipulate any data structures afterwards. 900 */ 901 902 void ConcurrentMark::enter_first_sync_barrier(uint worker_id) { 903 bool barrier_aborted; 904 { 905 SuspendibleThreadSetLeaver sts_leave(concurrent()); 906 barrier_aborted = !_first_overflow_barrier_sync.enter(); 907 } 908 909 // at this point everyone should have synced up and not be doing any 910 // more work 911 912 if (barrier_aborted) { 913 // If the barrier aborted we ignore the overflow condition and 914 // just abort the whole marking phase as quickly as possible. 915 return; 916 } 917 918 // If we're executing the concurrent phase of marking, reset the marking 919 // state; otherwise the marking state is reset after reference processing, 920 // during the remark pause. 921 // If we reset here as a result of an overflow during the remark we will 922 // see assertion failures from any subsequent set_concurrency_and_phase() 923 // calls. 924 if (concurrent()) { 925 // let the task associated with with worker 0 do this 926 if (worker_id == 0) { 927 // task 0 is responsible for clearing the global data structures 928 // We should be here because of an overflow. During STW we should 929 // not clear the overflow flag since we rely on it being true when 930 // we exit this method to abort the pause and restart concurrent 931 // marking. 932 reset_marking_state(true /* clear_overflow */); 933 force_overflow()->update(); 934 935 if (G1Log::fine()) { 936 gclog_or_tty->gclog_stamp(); 937 gclog_or_tty->print_cr("[GC concurrent-mark-reset-for-overflow]"); 938 } 939 } 940 } 941 942 // after this, each task should reset its own data structures then 943 // then go into the second barrier 944 } 945 946 void ConcurrentMark::enter_second_sync_barrier(uint worker_id) { 947 SuspendibleThreadSetLeaver sts_leave(concurrent()); 948 _second_overflow_barrier_sync.enter(); 949 950 // at this point everything should be re-initialized and ready to go 951 } 952 953 #ifndef PRODUCT 954 void ForceOverflowSettings::init() { 955 _num_remaining = G1ConcMarkForceOverflow; 956 _force = false; 957 update(); 958 } 959 960 void ForceOverflowSettings::update() { 961 if (_num_remaining > 0) { 962 _num_remaining -= 1; 963 _force = true; 964 } else { 965 _force = false; 966 } 967 } 968 969 bool ForceOverflowSettings::should_force() { 970 if (_force) { 971 _force = false; 972 return true; 973 } else { 974 return false; 975 } 976 } 977 #endif // !PRODUCT 978 979 class CMConcurrentMarkingTask: public AbstractGangTask { 980 private: 981 ConcurrentMark* _cm; 982 ConcurrentMarkThread* _cmt; 983 984 public: 985 void work(uint worker_id) { 986 assert(Thread::current()->is_ConcurrentGC_thread(), 987 "this should only be done by a conc GC thread"); 988 ResourceMark rm; 989 990 double start_vtime = os::elapsedVTime(); 991 992 { 993 SuspendibleThreadSetJoiner sts_join; 994 995 assert(worker_id < _cm->active_tasks(), "invariant"); 996 CMTask* the_task = _cm->task(worker_id); 997 the_task->record_start_time(); 998 if (!_cm->has_aborted()) { 999 do { 1000 double start_vtime_sec = os::elapsedVTime(); 1001 double mark_step_duration_ms = G1ConcMarkStepDurationMillis; 1002 1003 the_task->do_marking_step(mark_step_duration_ms, 1004 true /* do_termination */, 1005 false /* is_serial*/); 1006 1007 double end_vtime_sec = os::elapsedVTime(); 1008 double elapsed_vtime_sec = end_vtime_sec - start_vtime_sec; 1009 _cm->clear_has_overflown(); 1010 1011 _cm->do_yield_check(worker_id); 1012 1013 jlong sleep_time_ms; 1014 if (!_cm->has_aborted() && the_task->has_aborted()) { 1015 sleep_time_ms = 1016 (jlong) (elapsed_vtime_sec * _cm->sleep_factor() * 1000.0); 1017 { 1018 SuspendibleThreadSetLeaver sts_leave; 1019 os::sleep(Thread::current(), sleep_time_ms, false); 1020 } 1021 } 1022 } while (!_cm->has_aborted() && the_task->has_aborted()); 1023 } 1024 the_task->record_end_time(); 1025 guarantee(!the_task->has_aborted() || _cm->has_aborted(), "invariant"); 1026 } 1027 1028 double end_vtime = os::elapsedVTime(); 1029 _cm->update_accum_task_vtime(worker_id, end_vtime - start_vtime); 1030 } 1031 1032 CMConcurrentMarkingTask(ConcurrentMark* cm, 1033 ConcurrentMarkThread* cmt) : 1034 AbstractGangTask("Concurrent Mark"), _cm(cm), _cmt(cmt) { } 1035 1036 ~CMConcurrentMarkingTask() { } 1037 }; 1038 1039 // Calculates the number of active workers for a concurrent 1040 // phase. 1041 uint ConcurrentMark::calc_parallel_marking_threads() { 1042 uint n_conc_workers = 0; 1043 if (!UseDynamicNumberOfGCThreads || 1044 (!FLAG_IS_DEFAULT(ConcGCThreads) && 1045 !ForceDynamicNumberOfGCThreads)) { 1046 n_conc_workers = max_parallel_marking_threads(); 1047 } else { 1048 n_conc_workers = 1049 AdaptiveSizePolicy::calc_default_active_workers( 1050 max_parallel_marking_threads(), 1051 1, /* Minimum workers */ 1052 parallel_marking_threads(), 1053 Threads::number_of_non_daemon_threads()); 1054 // Don't scale down "n_conc_workers" by scale_parallel_threads() because 1055 // that scaling has already gone into "_max_parallel_marking_threads". 1056 } 1057 assert(n_conc_workers > 0, "Always need at least 1"); 1058 return n_conc_workers; 1059 } 1060 1061 void ConcurrentMark::scanRootRegion(HeapRegion* hr, uint worker_id) { 1062 // Currently, only survivors can be root regions. 1063 assert(hr->next_top_at_mark_start() == hr->bottom(), "invariant"); 1064 G1RootRegionScanClosure cl(_g1h, this, worker_id); 1065 1066 const uintx interval = PrefetchScanIntervalInBytes; 1067 HeapWord* curr = hr->bottom(); 1068 const HeapWord* end = hr->top(); 1069 while (curr < end) { 1070 Prefetch::read(curr, interval); 1071 oop obj = oop(curr); 1072 int size = obj->oop_iterate_size(&cl); 1073 assert(size == obj->size(), "sanity"); 1074 curr += size; 1075 } 1076 } 1077 1078 class CMRootRegionScanTask : public AbstractGangTask { 1079 private: 1080 ConcurrentMark* _cm; 1081 1082 public: 1083 CMRootRegionScanTask(ConcurrentMark* cm) : 1084 AbstractGangTask("Root Region Scan"), _cm(cm) { } 1085 1086 void work(uint worker_id) { 1087 assert(Thread::current()->is_ConcurrentGC_thread(), 1088 "this should only be done by a conc GC thread"); 1089 1090 CMRootRegions* root_regions = _cm->root_regions(); 1091 HeapRegion* hr = root_regions->claim_next(); 1092 while (hr != NULL) { 1093 _cm->scanRootRegion(hr, worker_id); 1094 hr = root_regions->claim_next(); 1095 } 1096 } 1097 }; 1098 1099 void ConcurrentMark::scanRootRegions() { 1100 double scan_start = os::elapsedTime(); 1101 1102 // Start of concurrent marking. 1103 ClassLoaderDataGraph::clear_claimed_marks(); 1104 1105 // scan_in_progress() will have been set to true only if there was 1106 // at least one root region to scan. So, if it's false, we 1107 // should not attempt to do any further work. 1108 if (root_regions()->scan_in_progress()) { 1109 if (G1Log::fine()) { 1110 gclog_or_tty->gclog_stamp(); 1111 gclog_or_tty->print_cr("[GC concurrent-root-region-scan-start]"); 1112 } 1113 1114 _parallel_marking_threads = calc_parallel_marking_threads(); 1115 assert(parallel_marking_threads() <= max_parallel_marking_threads(), 1116 "Maximum number of marking threads exceeded"); 1117 uint active_workers = MAX2(1U, parallel_marking_threads()); 1118 1119 CMRootRegionScanTask task(this); 1120 _parallel_workers->set_active_workers(active_workers); 1121 _parallel_workers->run_task(&task); 1122 1123 if (G1Log::fine()) { 1124 gclog_or_tty->gclog_stamp(); 1125 gclog_or_tty->print_cr("[GC concurrent-root-region-scan-end, %1.7lf secs]", os::elapsedTime() - scan_start); 1126 } 1127 1128 // It's possible that has_aborted() is true here without actually 1129 // aborting the survivor scan earlier. This is OK as it's 1130 // mainly used for sanity checking. 1131 root_regions()->scan_finished(); 1132 } 1133 } 1134 1135 void ConcurrentMark::markFromRoots() { 1136 // we might be tempted to assert that: 1137 // assert(asynch == !SafepointSynchronize::is_at_safepoint(), 1138 // "inconsistent argument?"); 1139 // However that wouldn't be right, because it's possible that 1140 // a safepoint is indeed in progress as a younger generation 1141 // stop-the-world GC happens even as we mark in this generation. 1142 1143 _restart_for_overflow = false; 1144 force_overflow_conc()->init(); 1145 1146 // _g1h has _n_par_threads 1147 _parallel_marking_threads = calc_parallel_marking_threads(); 1148 assert(parallel_marking_threads() <= max_parallel_marking_threads(), 1149 "Maximum number of marking threads exceeded"); 1150 1151 uint active_workers = MAX2(1U, parallel_marking_threads()); 1152 assert(active_workers > 0, "Should have been set"); 1153 1154 // Parallel task terminator is set in "set_concurrency_and_phase()" 1155 set_concurrency_and_phase(active_workers, true /* concurrent */); 1156 1157 CMConcurrentMarkingTask markingTask(this, cmThread()); 1158 _parallel_workers->set_active_workers(active_workers); 1159 _parallel_workers->run_task(&markingTask); 1160 print_stats(); 1161 } 1162 1163 // Helper class to get rid of some boilerplate code. 1164 class G1CMTraceTime : public StackObj { 1165 GCTraceTimeImpl _gc_trace_time; 1166 static bool doit_and_prepend(bool doit) { 1167 if (doit) { 1168 gclog_or_tty->put(' '); 1169 } 1170 return doit; 1171 } 1172 1173 public: 1174 G1CMTraceTime(const char* title, bool doit) 1175 : _gc_trace_time(title, doit_and_prepend(doit), false, G1CollectedHeap::heap()->gc_timer_cm()) { 1176 } 1177 }; 1178 1179 void ConcurrentMark::checkpointRootsFinal(bool clear_all_soft_refs) { 1180 // world is stopped at this checkpoint 1181 assert(SafepointSynchronize::is_at_safepoint(), 1182 "world should be stopped"); 1183 1184 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 1185 1186 // If a full collection has happened, we shouldn't do this. 1187 if (has_aborted()) { 1188 g1h->collector_state()->set_mark_in_progress(false); // So bitmap clearing isn't confused 1189 return; 1190 } 1191 1192 SvcGCMarker sgcm(SvcGCMarker::OTHER); 1193 1194 if (VerifyDuringGC) { 1195 HandleMark hm; // handle scope 1196 g1h->prepare_for_verify(); 1197 Universe::verify(VerifyOption_G1UsePrevMarking, 1198 " VerifyDuringGC:(before)"); 1199 } 1200 g1h->check_bitmaps("Remark Start"); 1201 1202 G1CollectorPolicy* g1p = g1h->g1_policy(); 1203 g1p->record_concurrent_mark_remark_start(); 1204 1205 double start = os::elapsedTime(); 1206 1207 checkpointRootsFinalWork(); 1208 1209 double mark_work_end = os::elapsedTime(); 1210 1211 weakRefsWork(clear_all_soft_refs); 1212 1213 if (has_overflown()) { 1214 // Oops. We overflowed. Restart concurrent marking. 1215 _restart_for_overflow = true; 1216 if (G1TraceMarkStackOverflow) { 1217 gclog_or_tty->print_cr("\nRemark led to restart for overflow."); 1218 } 1219 1220 // Verify the heap w.r.t. the previous marking bitmap. 1221 if (VerifyDuringGC) { 1222 HandleMark hm; // handle scope 1223 g1h->prepare_for_verify(); 1224 Universe::verify(VerifyOption_G1UsePrevMarking, 1225 " VerifyDuringGC:(overflow)"); 1226 } 1227 1228 // Clear the marking state because we will be restarting 1229 // marking due to overflowing the global mark stack. 1230 reset_marking_state(); 1231 } else { 1232 { 1233 G1CMTraceTime trace("GC aggregate-data", G1Log::finer()); 1234 1235 // Aggregate the per-task counting data that we have accumulated 1236 // while marking. 1237 aggregate_count_data(); 1238 } 1239 1240 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 1241 // We're done with marking. 1242 // This is the end of the marking cycle, we're expected all 1243 // threads to have SATB queues with active set to true. 1244 satb_mq_set.set_active_all_threads(false, /* new active value */ 1245 true /* expected_active */); 1246 1247 if (VerifyDuringGC) { 1248 HandleMark hm; // handle scope 1249 g1h->prepare_for_verify(); 1250 Universe::verify(VerifyOption_G1UseNextMarking, 1251 " VerifyDuringGC:(after)"); 1252 } 1253 g1h->check_bitmaps("Remark End"); 1254 assert(!restart_for_overflow(), "sanity"); 1255 // Completely reset the marking state since marking completed 1256 set_non_marking_state(); 1257 } 1258 1259 // Expand the marking stack, if we have to and if we can. 1260 if (_markStack.should_expand()) { 1261 _markStack.expand(); 1262 } 1263 1264 // Statistics 1265 double now = os::elapsedTime(); 1266 _remark_mark_times.add((mark_work_end - start) * 1000.0); 1267 _remark_weak_ref_times.add((now - mark_work_end) * 1000.0); 1268 _remark_times.add((now - start) * 1000.0); 1269 1270 g1p->record_concurrent_mark_remark_end(); 1271 1272 G1CMIsAliveClosure is_alive(g1h); 1273 g1h->gc_tracer_cm()->report_object_count_after_gc(&is_alive); 1274 } 1275 1276 // Base class of the closures that finalize and verify the 1277 // liveness counting data. 1278 class CMCountDataClosureBase: public HeapRegionClosure { 1279 protected: 1280 G1CollectedHeap* _g1h; 1281 ConcurrentMark* _cm; 1282 CardTableModRefBS* _ct_bs; 1283 1284 BitMap* _region_bm; 1285 BitMap* _card_bm; 1286 1287 // Takes a region that's not empty (i.e., it has at least one 1288 // live object in it and sets its corresponding bit on the region 1289 // bitmap to 1. If the region is "starts humongous" it will also set 1290 // to 1 the bits on the region bitmap that correspond to its 1291 // associated "continues humongous" regions. 1292 void set_bit_for_region(HeapRegion* hr) { 1293 assert(!hr->is_continues_humongous(), "should have filtered those out"); 1294 1295 BitMap::idx_t index = (BitMap::idx_t) hr->hrm_index(); 1296 if (!hr->is_starts_humongous()) { 1297 // Normal (non-humongous) case: just set the bit. 1298 _region_bm->par_at_put(index, true); 1299 } else { 1300 // Starts humongous case: calculate how many regions are part of 1301 // this humongous region and then set the bit range. 1302 BitMap::idx_t end_index = (BitMap::idx_t) hr->last_hc_index(); 1303 _region_bm->par_at_put_range(index, end_index, true); 1304 } 1305 } 1306 1307 public: 1308 CMCountDataClosureBase(G1CollectedHeap* g1h, 1309 BitMap* region_bm, BitMap* card_bm): 1310 _g1h(g1h), _cm(g1h->concurrent_mark()), 1311 _ct_bs(barrier_set_cast<CardTableModRefBS>(g1h->barrier_set())), 1312 _region_bm(region_bm), _card_bm(card_bm) { } 1313 }; 1314 1315 // Closure that calculates the # live objects per region. Used 1316 // for verification purposes during the cleanup pause. 1317 class CalcLiveObjectsClosure: public CMCountDataClosureBase { 1318 CMBitMapRO* _bm; 1319 size_t _region_marked_bytes; 1320 1321 public: 1322 CalcLiveObjectsClosure(CMBitMapRO *bm, G1CollectedHeap* g1h, 1323 BitMap* region_bm, BitMap* card_bm) : 1324 CMCountDataClosureBase(g1h, region_bm, card_bm), 1325 _bm(bm), _region_marked_bytes(0) { } 1326 1327 bool doHeapRegion(HeapRegion* hr) { 1328 1329 if (hr->is_continues_humongous()) { 1330 // We will ignore these here and process them when their 1331 // associated "starts humongous" region is processed (see 1332 // set_bit_for_heap_region()). Note that we cannot rely on their 1333 // associated "starts humongous" region to have their bit set to 1334 // 1 since, due to the region chunking in the parallel region 1335 // iteration, a "continues humongous" region might be visited 1336 // before its associated "starts humongous". 1337 return false; 1338 } 1339 1340 HeapWord* ntams = hr->next_top_at_mark_start(); 1341 HeapWord* start = hr->bottom(); 1342 1343 assert(start <= hr->end() && start <= ntams && ntams <= hr->end(), 1344 "Preconditions not met - " 1345 "start: " PTR_FORMAT ", ntams: " PTR_FORMAT ", end: " PTR_FORMAT, 1346 p2i(start), p2i(ntams), p2i(hr->end())); 1347 1348 // Find the first marked object at or after "start". 1349 start = _bm->getNextMarkedWordAddress(start, ntams); 1350 1351 size_t marked_bytes = 0; 1352 1353 while (start < ntams) { 1354 oop obj = oop(start); 1355 int obj_sz = obj->size(); 1356 HeapWord* obj_end = start + obj_sz; 1357 1358 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(start); 1359 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(obj_end); 1360 1361 // Note: if we're looking at the last region in heap - obj_end 1362 // could be actually just beyond the end of the heap; end_idx 1363 // will then correspond to a (non-existent) card that is also 1364 // just beyond the heap. 1365 if (_g1h->is_in_g1_reserved(obj_end) && !_ct_bs->is_card_aligned(obj_end)) { 1366 // end of object is not card aligned - increment to cover 1367 // all the cards spanned by the object 1368 end_idx += 1; 1369 } 1370 1371 // Set the bits in the card BM for the cards spanned by this object. 1372 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */); 1373 1374 // Add the size of this object to the number of marked bytes. 1375 marked_bytes += (size_t)obj_sz * HeapWordSize; 1376 1377 // Find the next marked object after this one. 1378 start = _bm->getNextMarkedWordAddress(obj_end, ntams); 1379 } 1380 1381 // Mark the allocated-since-marking portion... 1382 HeapWord* top = hr->top(); 1383 if (ntams < top) { 1384 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(ntams); 1385 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(top); 1386 1387 // Note: if we're looking at the last region in heap - top 1388 // could be actually just beyond the end of the heap; end_idx 1389 // will then correspond to a (non-existent) card that is also 1390 // just beyond the heap. 1391 if (_g1h->is_in_g1_reserved(top) && !_ct_bs->is_card_aligned(top)) { 1392 // end of object is not card aligned - increment to cover 1393 // all the cards spanned by the object 1394 end_idx += 1; 1395 } 1396 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */); 1397 1398 // This definitely means the region has live objects. 1399 set_bit_for_region(hr); 1400 } 1401 1402 // Update the live region bitmap. 1403 if (marked_bytes > 0) { 1404 set_bit_for_region(hr); 1405 } 1406 1407 // Set the marked bytes for the current region so that 1408 // it can be queried by a calling verification routine 1409 _region_marked_bytes = marked_bytes; 1410 1411 return false; 1412 } 1413 1414 size_t region_marked_bytes() const { return _region_marked_bytes; } 1415 }; 1416 1417 // Heap region closure used for verifying the counting data 1418 // that was accumulated concurrently and aggregated during 1419 // the remark pause. This closure is applied to the heap 1420 // regions during the STW cleanup pause. 1421 1422 class VerifyLiveObjectDataHRClosure: public HeapRegionClosure { 1423 G1CollectedHeap* _g1h; 1424 ConcurrentMark* _cm; 1425 CalcLiveObjectsClosure _calc_cl; 1426 BitMap* _region_bm; // Region BM to be verified 1427 BitMap* _card_bm; // Card BM to be verified 1428 1429 BitMap* _exp_region_bm; // Expected Region BM values 1430 BitMap* _exp_card_bm; // Expected card BM values 1431 1432 int _failures; 1433 1434 public: 1435 VerifyLiveObjectDataHRClosure(G1CollectedHeap* g1h, 1436 BitMap* region_bm, 1437 BitMap* card_bm, 1438 BitMap* exp_region_bm, 1439 BitMap* exp_card_bm) : 1440 _g1h(g1h), _cm(g1h->concurrent_mark()), 1441 _calc_cl(_cm->nextMarkBitMap(), g1h, exp_region_bm, exp_card_bm), 1442 _region_bm(region_bm), _card_bm(card_bm), 1443 _exp_region_bm(exp_region_bm), _exp_card_bm(exp_card_bm), 1444 _failures(0) { } 1445 1446 int failures() const { return _failures; } 1447 1448 bool doHeapRegion(HeapRegion* hr) { 1449 if (hr->is_continues_humongous()) { 1450 // We will ignore these here and process them when their 1451 // associated "starts humongous" region is processed (see 1452 // set_bit_for_heap_region()). Note that we cannot rely on their 1453 // associated "starts humongous" region to have their bit set to 1454 // 1 since, due to the region chunking in the parallel region 1455 // iteration, a "continues humongous" region might be visited 1456 // before its associated "starts humongous". 1457 return false; 1458 } 1459 1460 int failures = 0; 1461 1462 // Call the CalcLiveObjectsClosure to walk the marking bitmap for 1463 // this region and set the corresponding bits in the expected region 1464 // and card bitmaps. 1465 bool res = _calc_cl.doHeapRegion(hr); 1466 assert(res == false, "should be continuing"); 1467 1468 // Verify the marked bytes for this region. 1469 size_t exp_marked_bytes = _calc_cl.region_marked_bytes(); 1470 size_t act_marked_bytes = hr->next_marked_bytes(); 1471 1472 // We're not OK if expected marked bytes > actual marked bytes. It means 1473 // we have missed accounting some objects during the actual marking. 1474 if (exp_marked_bytes > act_marked_bytes) { 1475 failures += 1; 1476 } 1477 1478 // Verify the bit, for this region, in the actual and expected 1479 // (which was just calculated) region bit maps. 1480 // We're not OK if the bit in the calculated expected region 1481 // bitmap is set and the bit in the actual region bitmap is not. 1482 BitMap::idx_t index = (BitMap::idx_t) hr->hrm_index(); 1483 1484 bool expected = _exp_region_bm->at(index); 1485 bool actual = _region_bm->at(index); 1486 if (expected && !actual) { 1487 failures += 1; 1488 } 1489 1490 // Verify that the card bit maps for the cards spanned by the current 1491 // region match. We have an error if we have a set bit in the expected 1492 // bit map and the corresponding bit in the actual bitmap is not set. 1493 1494 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(hr->bottom()); 1495 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(hr->top()); 1496 1497 for (BitMap::idx_t i = start_idx; i < end_idx; i+=1) { 1498 expected = _exp_card_bm->at(i); 1499 actual = _card_bm->at(i); 1500 1501 if (expected && !actual) { 1502 failures += 1; 1503 } 1504 } 1505 1506 _failures += failures; 1507 1508 // We could stop iteration over the heap when we 1509 // find the first violating region by returning true. 1510 return false; 1511 } 1512 }; 1513 1514 class G1ParVerifyFinalCountTask: public AbstractGangTask { 1515 protected: 1516 G1CollectedHeap* _g1h; 1517 ConcurrentMark* _cm; 1518 BitMap* _actual_region_bm; 1519 BitMap* _actual_card_bm; 1520 1521 uint _n_workers; 1522 1523 BitMap* _expected_region_bm; 1524 BitMap* _expected_card_bm; 1525 1526 int _failures; 1527 1528 HeapRegionClaimer _hrclaimer; 1529 1530 public: 1531 G1ParVerifyFinalCountTask(G1CollectedHeap* g1h, 1532 BitMap* region_bm, BitMap* card_bm, 1533 BitMap* expected_region_bm, BitMap* expected_card_bm) 1534 : AbstractGangTask("G1 verify final counting"), 1535 _g1h(g1h), _cm(_g1h->concurrent_mark()), 1536 _actual_region_bm(region_bm), _actual_card_bm(card_bm), 1537 _expected_region_bm(expected_region_bm), _expected_card_bm(expected_card_bm), 1538 _failures(0), 1539 _n_workers(_g1h->workers()->active_workers()), _hrclaimer(_n_workers) { 1540 assert(VerifyDuringGC, "don't call this otherwise"); 1541 assert(_expected_card_bm->size() == _actual_card_bm->size(), "sanity"); 1542 assert(_expected_region_bm->size() == _actual_region_bm->size(), "sanity"); 1543 } 1544 1545 void work(uint worker_id) { 1546 assert(worker_id < _n_workers, "invariant"); 1547 1548 VerifyLiveObjectDataHRClosure verify_cl(_g1h, 1549 _actual_region_bm, _actual_card_bm, 1550 _expected_region_bm, 1551 _expected_card_bm); 1552 1553 _g1h->heap_region_par_iterate(&verify_cl, worker_id, &_hrclaimer); 1554 1555 Atomic::add(verify_cl.failures(), &_failures); 1556 } 1557 1558 int failures() const { return _failures; } 1559 }; 1560 1561 // Closure that finalizes the liveness counting data. 1562 // Used during the cleanup pause. 1563 // Sets the bits corresponding to the interval [NTAMS, top] 1564 // (which contains the implicitly live objects) in the 1565 // card liveness bitmap. Also sets the bit for each region, 1566 // containing live data, in the region liveness bitmap. 1567 1568 class FinalCountDataUpdateClosure: public CMCountDataClosureBase { 1569 public: 1570 FinalCountDataUpdateClosure(G1CollectedHeap* g1h, 1571 BitMap* region_bm, 1572 BitMap* card_bm) : 1573 CMCountDataClosureBase(g1h, region_bm, card_bm) { } 1574 1575 bool doHeapRegion(HeapRegion* hr) { 1576 1577 if (hr->is_continues_humongous()) { 1578 // We will ignore these here and process them when their 1579 // associated "starts humongous" region is processed (see 1580 // set_bit_for_heap_region()). Note that we cannot rely on their 1581 // associated "starts humongous" region to have their bit set to 1582 // 1 since, due to the region chunking in the parallel region 1583 // iteration, a "continues humongous" region might be visited 1584 // before its associated "starts humongous". 1585 return false; 1586 } 1587 1588 HeapWord* ntams = hr->next_top_at_mark_start(); 1589 HeapWord* top = hr->top(); 1590 1591 assert(hr->bottom() <= ntams && ntams <= hr->end(), "Preconditions."); 1592 1593 // Mark the allocated-since-marking portion... 1594 if (ntams < top) { 1595 // This definitely means the region has live objects. 1596 set_bit_for_region(hr); 1597 1598 // Now set the bits in the card bitmap for [ntams, top) 1599 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(ntams); 1600 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(top); 1601 1602 // Note: if we're looking at the last region in heap - top 1603 // could be actually just beyond the end of the heap; end_idx 1604 // will then correspond to a (non-existent) card that is also 1605 // just beyond the heap. 1606 if (_g1h->is_in_g1_reserved(top) && !_ct_bs->is_card_aligned(top)) { 1607 // end of object is not card aligned - increment to cover 1608 // all the cards spanned by the object 1609 end_idx += 1; 1610 } 1611 1612 assert(end_idx <= _card_bm->size(), 1613 "oob: end_idx= " SIZE_FORMAT ", bitmap size= " SIZE_FORMAT, 1614 end_idx, _card_bm->size()); 1615 assert(start_idx < _card_bm->size(), 1616 "oob: start_idx= " SIZE_FORMAT ", bitmap size= " SIZE_FORMAT, 1617 start_idx, _card_bm->size()); 1618 1619 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */); 1620 } 1621 1622 // Set the bit for the region if it contains live data 1623 if (hr->next_marked_bytes() > 0) { 1624 set_bit_for_region(hr); 1625 } 1626 1627 return false; 1628 } 1629 }; 1630 1631 class G1ParFinalCountTask: public AbstractGangTask { 1632 protected: 1633 G1CollectedHeap* _g1h; 1634 ConcurrentMark* _cm; 1635 BitMap* _actual_region_bm; 1636 BitMap* _actual_card_bm; 1637 1638 uint _n_workers; 1639 HeapRegionClaimer _hrclaimer; 1640 1641 public: 1642 G1ParFinalCountTask(G1CollectedHeap* g1h, BitMap* region_bm, BitMap* card_bm) 1643 : AbstractGangTask("G1 final counting"), 1644 _g1h(g1h), _cm(_g1h->concurrent_mark()), 1645 _actual_region_bm(region_bm), _actual_card_bm(card_bm), 1646 _n_workers(_g1h->workers()->active_workers()), _hrclaimer(_n_workers) { 1647 } 1648 1649 void work(uint worker_id) { 1650 assert(worker_id < _n_workers, "invariant"); 1651 1652 FinalCountDataUpdateClosure final_update_cl(_g1h, 1653 _actual_region_bm, 1654 _actual_card_bm); 1655 1656 _g1h->heap_region_par_iterate(&final_update_cl, worker_id, &_hrclaimer); 1657 } 1658 }; 1659 1660 class G1NoteEndOfConcMarkClosure : public HeapRegionClosure { 1661 G1CollectedHeap* _g1; 1662 size_t _freed_bytes; 1663 FreeRegionList* _local_cleanup_list; 1664 HeapRegionSetCount _old_regions_removed; 1665 HeapRegionSetCount _humongous_regions_removed; 1666 HRRSCleanupTask* _hrrs_cleanup_task; 1667 1668 public: 1669 G1NoteEndOfConcMarkClosure(G1CollectedHeap* g1, 1670 FreeRegionList* local_cleanup_list, 1671 HRRSCleanupTask* hrrs_cleanup_task) : 1672 _g1(g1), 1673 _freed_bytes(0), 1674 _local_cleanup_list(local_cleanup_list), 1675 _old_regions_removed(), 1676 _humongous_regions_removed(), 1677 _hrrs_cleanup_task(hrrs_cleanup_task) { } 1678 1679 size_t freed_bytes() { return _freed_bytes; } 1680 const HeapRegionSetCount& old_regions_removed() { return _old_regions_removed; } 1681 const HeapRegionSetCount& humongous_regions_removed() { return _humongous_regions_removed; } 1682 1683 bool doHeapRegion(HeapRegion *hr) { 1684 if (hr->is_continues_humongous() || hr->is_archive()) { 1685 return false; 1686 } 1687 // We use a claim value of zero here because all regions 1688 // were claimed with value 1 in the FinalCount task. 1689 _g1->reset_gc_time_stamps(hr); 1690 hr->note_end_of_marking(); 1691 1692 if (hr->used() > 0 && hr->max_live_bytes() == 0 && !hr->is_young()) { 1693 _freed_bytes += hr->used(); 1694 hr->set_containing_set(NULL); 1695 if (hr->is_humongous()) { 1696 assert(hr->is_starts_humongous(), "we should only see starts humongous"); 1697 _humongous_regions_removed.increment(1u, hr->capacity()); 1698 _g1->free_humongous_region(hr, _local_cleanup_list, true); 1699 } else { 1700 _old_regions_removed.increment(1u, hr->capacity()); 1701 _g1->free_region(hr, _local_cleanup_list, true); 1702 } 1703 } else { 1704 hr->rem_set()->do_cleanup_work(_hrrs_cleanup_task); 1705 } 1706 1707 return false; 1708 } 1709 }; 1710 1711 class G1ParNoteEndTask: public AbstractGangTask { 1712 friend class G1NoteEndOfConcMarkClosure; 1713 1714 protected: 1715 G1CollectedHeap* _g1h; 1716 FreeRegionList* _cleanup_list; 1717 HeapRegionClaimer _hrclaimer; 1718 1719 public: 1720 G1ParNoteEndTask(G1CollectedHeap* g1h, FreeRegionList* cleanup_list, uint n_workers) : 1721 AbstractGangTask("G1 note end"), _g1h(g1h), _cleanup_list(cleanup_list), _hrclaimer(n_workers) { 1722 } 1723 1724 void work(uint worker_id) { 1725 FreeRegionList local_cleanup_list("Local Cleanup List"); 1726 HRRSCleanupTask hrrs_cleanup_task; 1727 G1NoteEndOfConcMarkClosure g1_note_end(_g1h, &local_cleanup_list, 1728 &hrrs_cleanup_task); 1729 _g1h->heap_region_par_iterate(&g1_note_end, worker_id, &_hrclaimer); 1730 assert(g1_note_end.complete(), "Shouldn't have yielded!"); 1731 1732 // Now update the lists 1733 _g1h->remove_from_old_sets(g1_note_end.old_regions_removed(), g1_note_end.humongous_regions_removed()); 1734 { 1735 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag); 1736 _g1h->decrement_summary_bytes(g1_note_end.freed_bytes()); 1737 1738 // If we iterate over the global cleanup list at the end of 1739 // cleanup to do this printing we will not guarantee to only 1740 // generate output for the newly-reclaimed regions (the list 1741 // might not be empty at the beginning of cleanup; we might 1742 // still be working on its previous contents). So we do the 1743 // printing here, before we append the new regions to the global 1744 // cleanup list. 1745 1746 G1HRPrinter* hr_printer = _g1h->hr_printer(); 1747 if (hr_printer->is_active()) { 1748 FreeRegionListIterator iter(&local_cleanup_list); 1749 while (iter.more_available()) { 1750 HeapRegion* hr = iter.get_next(); 1751 hr_printer->cleanup(hr); 1752 } 1753 } 1754 1755 _cleanup_list->add_ordered(&local_cleanup_list); 1756 assert(local_cleanup_list.is_empty(), "post-condition"); 1757 1758 HeapRegionRemSet::finish_cleanup_task(&hrrs_cleanup_task); 1759 } 1760 } 1761 }; 1762 1763 class G1ParScrubRemSetTask: public AbstractGangTask { 1764 protected: 1765 G1RemSet* _g1rs; 1766 BitMap* _region_bm; 1767 BitMap* _card_bm; 1768 HeapRegionClaimer _hrclaimer; 1769 1770 public: 1771 G1ParScrubRemSetTask(G1CollectedHeap* g1h, BitMap* region_bm, BitMap* card_bm, uint n_workers) : 1772 AbstractGangTask("G1 ScrubRS"), _g1rs(g1h->g1_rem_set()), _region_bm(region_bm), _card_bm(card_bm), _hrclaimer(n_workers) { 1773 } 1774 1775 void work(uint worker_id) { 1776 _g1rs->scrub(_region_bm, _card_bm, worker_id, &_hrclaimer); 1777 } 1778 1779 }; 1780 1781 void ConcurrentMark::cleanup() { 1782 // world is stopped at this checkpoint 1783 assert(SafepointSynchronize::is_at_safepoint(), 1784 "world should be stopped"); 1785 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 1786 1787 // If a full collection has happened, we shouldn't do this. 1788 if (has_aborted()) { 1789 g1h->collector_state()->set_mark_in_progress(false); // So bitmap clearing isn't confused 1790 return; 1791 } 1792 1793 g1h->verify_region_sets_optional(); 1794 1795 if (VerifyDuringGC) { 1796 HandleMark hm; // handle scope 1797 g1h->prepare_for_verify(); 1798 Universe::verify(VerifyOption_G1UsePrevMarking, 1799 " VerifyDuringGC:(before)"); 1800 } 1801 g1h->check_bitmaps("Cleanup Start"); 1802 1803 G1CollectorPolicy* g1p = g1h->g1_policy(); 1804 g1p->record_concurrent_mark_cleanup_start(); 1805 1806 double start = os::elapsedTime(); 1807 1808 HeapRegionRemSet::reset_for_cleanup_tasks(); 1809 1810 // Do counting once more with the world stopped for good measure. 1811 G1ParFinalCountTask g1_par_count_task(g1h, &_region_bm, &_card_bm); 1812 1813 g1h->workers()->run_task(&g1_par_count_task); 1814 1815 if (VerifyDuringGC) { 1816 // Verify that the counting data accumulated during marking matches 1817 // that calculated by walking the marking bitmap. 1818 1819 // Bitmaps to hold expected values 1820 BitMap expected_region_bm(_region_bm.size(), true); 1821 BitMap expected_card_bm(_card_bm.size(), true); 1822 1823 G1ParVerifyFinalCountTask g1_par_verify_task(g1h, 1824 &_region_bm, 1825 &_card_bm, 1826 &expected_region_bm, 1827 &expected_card_bm); 1828 1829 g1h->workers()->run_task(&g1_par_verify_task); 1830 1831 guarantee(g1_par_verify_task.failures() == 0, "Unexpected accounting failures"); 1832 } 1833 1834 size_t start_used_bytes = g1h->used(); 1835 g1h->collector_state()->set_mark_in_progress(false); 1836 1837 double count_end = os::elapsedTime(); 1838 double this_final_counting_time = (count_end - start); 1839 _total_counting_time += this_final_counting_time; 1840 1841 if (G1PrintRegionLivenessInfo) { 1842 G1PrintRegionLivenessInfoClosure cl(gclog_or_tty, "Post-Marking"); 1843 _g1h->heap_region_iterate(&cl); 1844 } 1845 1846 // Install newly created mark bitMap as "prev". 1847 swapMarkBitMaps(); 1848 1849 g1h->reset_gc_time_stamp(); 1850 1851 uint n_workers = _g1h->workers()->active_workers(); 1852 1853 // Note end of marking in all heap regions. 1854 G1ParNoteEndTask g1_par_note_end_task(g1h, &_cleanup_list, n_workers); 1855 g1h->workers()->run_task(&g1_par_note_end_task); 1856 g1h->check_gc_time_stamps(); 1857 1858 if (!cleanup_list_is_empty()) { 1859 // The cleanup list is not empty, so we'll have to process it 1860 // concurrently. Notify anyone else that might be wanting free 1861 // regions that there will be more free regions coming soon. 1862 g1h->set_free_regions_coming(); 1863 } 1864 1865 // call below, since it affects the metric by which we sort the heap 1866 // regions. 1867 if (G1ScrubRemSets) { 1868 double rs_scrub_start = os::elapsedTime(); 1869 G1ParScrubRemSetTask g1_par_scrub_rs_task(g1h, &_region_bm, &_card_bm, n_workers); 1870 g1h->workers()->run_task(&g1_par_scrub_rs_task); 1871 1872 double rs_scrub_end = os::elapsedTime(); 1873 double this_rs_scrub_time = (rs_scrub_end - rs_scrub_start); 1874 _total_rs_scrub_time += this_rs_scrub_time; 1875 } 1876 1877 // this will also free any regions totally full of garbage objects, 1878 // and sort the regions. 1879 g1h->g1_policy()->record_concurrent_mark_cleanup_end(); 1880 1881 // Statistics. 1882 double end = os::elapsedTime(); 1883 _cleanup_times.add((end - start) * 1000.0); 1884 1885 if (G1Log::fine()) { 1886 g1h->g1_policy()->print_heap_transition(start_used_bytes); 1887 } 1888 1889 // Clean up will have freed any regions completely full of garbage. 1890 // Update the soft reference policy with the new heap occupancy. 1891 Universe::update_heap_info_at_gc(); 1892 1893 if (VerifyDuringGC) { 1894 HandleMark hm; // handle scope 1895 g1h->prepare_for_verify(); 1896 Universe::verify(VerifyOption_G1UsePrevMarking, 1897 " VerifyDuringGC:(after)"); 1898 } 1899 1900 g1h->check_bitmaps("Cleanup End"); 1901 1902 g1h->verify_region_sets_optional(); 1903 1904 // We need to make this be a "collection" so any collection pause that 1905 // races with it goes around and waits for completeCleanup to finish. 1906 g1h->increment_total_collections(); 1907 1908 // Clean out dead classes and update Metaspace sizes. 1909 if (ClassUnloadingWithConcurrentMark) { 1910 ClassLoaderDataGraph::purge(); 1911 } 1912 MetaspaceGC::compute_new_size(); 1913 1914 // We reclaimed old regions so we should calculate the sizes to make 1915 // sure we update the old gen/space data. 1916 g1h->g1mm()->update_sizes(); 1917 g1h->allocation_context_stats().update_after_mark(); 1918 1919 g1h->trace_heap_after_concurrent_cycle(); 1920 } 1921 1922 void ConcurrentMark::completeCleanup() { 1923 if (has_aborted()) return; 1924 1925 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 1926 1927 _cleanup_list.verify_optional(); 1928 FreeRegionList tmp_free_list("Tmp Free List"); 1929 1930 if (G1ConcRegionFreeingVerbose) { 1931 gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : " 1932 "cleanup list has %u entries", 1933 _cleanup_list.length()); 1934 } 1935 1936 // No one else should be accessing the _cleanup_list at this point, 1937 // so it is not necessary to take any locks 1938 while (!_cleanup_list.is_empty()) { 1939 HeapRegion* hr = _cleanup_list.remove_region(true /* from_head */); 1940 assert(hr != NULL, "Got NULL from a non-empty list"); 1941 hr->par_clear(); 1942 tmp_free_list.add_ordered(hr); 1943 1944 // Instead of adding one region at a time to the secondary_free_list, 1945 // we accumulate them in the local list and move them a few at a 1946 // time. This also cuts down on the number of notify_all() calls 1947 // we do during this process. We'll also append the local list when 1948 // _cleanup_list is empty (which means we just removed the last 1949 // region from the _cleanup_list). 1950 if ((tmp_free_list.length() % G1SecondaryFreeListAppendLength == 0) || 1951 _cleanup_list.is_empty()) { 1952 if (G1ConcRegionFreeingVerbose) { 1953 gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : " 1954 "appending %u entries to the secondary_free_list, " 1955 "cleanup list still has %u entries", 1956 tmp_free_list.length(), 1957 _cleanup_list.length()); 1958 } 1959 1960 { 1961 MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag); 1962 g1h->secondary_free_list_add(&tmp_free_list); 1963 SecondaryFreeList_lock->notify_all(); 1964 } 1965 #ifndef PRODUCT 1966 if (G1StressConcRegionFreeing) { 1967 for (uintx i = 0; i < G1StressConcRegionFreeingDelayMillis; ++i) { 1968 os::sleep(Thread::current(), (jlong) 1, false); 1969 } 1970 } 1971 #endif 1972 } 1973 } 1974 assert(tmp_free_list.is_empty(), "post-condition"); 1975 } 1976 1977 // Supporting Object and Oop closures for reference discovery 1978 // and processing in during marking 1979 1980 bool G1CMIsAliveClosure::do_object_b(oop obj) { 1981 HeapWord* addr = (HeapWord*)obj; 1982 return addr != NULL && 1983 (!_g1->is_in_g1_reserved(addr) || !_g1->is_obj_ill(obj)); 1984 } 1985 1986 // 'Keep Alive' oop closure used by both serial parallel reference processing. 1987 // Uses the CMTask associated with a worker thread (for serial reference 1988 // processing the CMTask for worker 0 is used) to preserve (mark) and 1989 // trace referent objects. 1990 // 1991 // Using the CMTask and embedded local queues avoids having the worker 1992 // threads operating on the global mark stack. This reduces the risk 1993 // of overflowing the stack - which we would rather avoid at this late 1994 // state. Also using the tasks' local queues removes the potential 1995 // of the workers interfering with each other that could occur if 1996 // operating on the global stack. 1997 1998 class G1CMKeepAliveAndDrainClosure: public OopClosure { 1999 ConcurrentMark* _cm; 2000 CMTask* _task; 2001 int _ref_counter_limit; 2002 int _ref_counter; 2003 bool _is_serial; 2004 public: 2005 G1CMKeepAliveAndDrainClosure(ConcurrentMark* cm, CMTask* task, bool is_serial) : 2006 _cm(cm), _task(task), _is_serial(is_serial), 2007 _ref_counter_limit(G1RefProcDrainInterval) { 2008 assert(_ref_counter_limit > 0, "sanity"); 2009 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code"); 2010 _ref_counter = _ref_counter_limit; 2011 } 2012 2013 virtual void do_oop(narrowOop* p) { do_oop_work(p); } 2014 virtual void do_oop( oop* p) { do_oop_work(p); } 2015 2016 template <class T> void do_oop_work(T* p) { 2017 if (!_cm->has_overflown()) { 2018 oop obj = oopDesc::load_decode_heap_oop(p); 2019 _task->deal_with_reference(obj); 2020 _ref_counter--; 2021 2022 if (_ref_counter == 0) { 2023 // We have dealt with _ref_counter_limit references, pushing them 2024 // and objects reachable from them on to the local stack (and 2025 // possibly the global stack). Call CMTask::do_marking_step() to 2026 // process these entries. 2027 // 2028 // We call CMTask::do_marking_step() in a loop, which we'll exit if 2029 // there's nothing more to do (i.e. we're done with the entries that 2030 // were pushed as a result of the CMTask::deal_with_reference() calls 2031 // above) or we overflow. 2032 // 2033 // Note: CMTask::do_marking_step() can set the CMTask::has_aborted() 2034 // flag while there may still be some work to do. (See the comment at 2035 // the beginning of CMTask::do_marking_step() for those conditions - 2036 // one of which is reaching the specified time target.) It is only 2037 // when CMTask::do_marking_step() returns without setting the 2038 // has_aborted() flag that the marking step has completed. 2039 do { 2040 double mark_step_duration_ms = G1ConcMarkStepDurationMillis; 2041 _task->do_marking_step(mark_step_duration_ms, 2042 false /* do_termination */, 2043 _is_serial); 2044 } while (_task->has_aborted() && !_cm->has_overflown()); 2045 _ref_counter = _ref_counter_limit; 2046 } 2047 } 2048 } 2049 }; 2050 2051 // 'Drain' oop closure used by both serial and parallel reference processing. 2052 // Uses the CMTask associated with a given worker thread (for serial 2053 // reference processing the CMtask for worker 0 is used). Calls the 2054 // do_marking_step routine, with an unbelievably large timeout value, 2055 // to drain the marking data structures of the remaining entries 2056 // added by the 'keep alive' oop closure above. 2057 2058 class G1CMDrainMarkingStackClosure: public VoidClosure { 2059 ConcurrentMark* _cm; 2060 CMTask* _task; 2061 bool _is_serial; 2062 public: 2063 G1CMDrainMarkingStackClosure(ConcurrentMark* cm, CMTask* task, bool is_serial) : 2064 _cm(cm), _task(task), _is_serial(is_serial) { 2065 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code"); 2066 } 2067 2068 void do_void() { 2069 do { 2070 // We call CMTask::do_marking_step() to completely drain the local 2071 // and global marking stacks of entries pushed by the 'keep alive' 2072 // oop closure (an instance of G1CMKeepAliveAndDrainClosure above). 2073 // 2074 // CMTask::do_marking_step() is called in a loop, which we'll exit 2075 // if there's nothing more to do (i.e. we've completely drained the 2076 // entries that were pushed as a a result of applying the 'keep alive' 2077 // closure to the entries on the discovered ref lists) or we overflow 2078 // the global marking stack. 2079 // 2080 // Note: CMTask::do_marking_step() can set the CMTask::has_aborted() 2081 // flag while there may still be some work to do. (See the comment at 2082 // the beginning of CMTask::do_marking_step() for those conditions - 2083 // one of which is reaching the specified time target.) It is only 2084 // when CMTask::do_marking_step() returns without setting the 2085 // has_aborted() flag that the marking step has completed. 2086 2087 _task->do_marking_step(1000000000.0 /* something very large */, 2088 true /* do_termination */, 2089 _is_serial); 2090 } while (_task->has_aborted() && !_cm->has_overflown()); 2091 } 2092 }; 2093 2094 // Implementation of AbstractRefProcTaskExecutor for parallel 2095 // reference processing at the end of G1 concurrent marking 2096 2097 class G1CMRefProcTaskExecutor: public AbstractRefProcTaskExecutor { 2098 private: 2099 G1CollectedHeap* _g1h; 2100 ConcurrentMark* _cm; 2101 WorkGang* _workers; 2102 uint _active_workers; 2103 2104 public: 2105 G1CMRefProcTaskExecutor(G1CollectedHeap* g1h, 2106 ConcurrentMark* cm, 2107 WorkGang* workers, 2108 uint n_workers) : 2109 _g1h(g1h), _cm(cm), 2110 _workers(workers), _active_workers(n_workers) { } 2111 2112 // Executes the given task using concurrent marking worker threads. 2113 virtual void execute(ProcessTask& task); 2114 virtual void execute(EnqueueTask& task); 2115 }; 2116 2117 class G1CMRefProcTaskProxy: public AbstractGangTask { 2118 typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask; 2119 ProcessTask& _proc_task; 2120 G1CollectedHeap* _g1h; 2121 ConcurrentMark* _cm; 2122 2123 public: 2124 G1CMRefProcTaskProxy(ProcessTask& proc_task, 2125 G1CollectedHeap* g1h, 2126 ConcurrentMark* cm) : 2127 AbstractGangTask("Process reference objects in parallel"), 2128 _proc_task(proc_task), _g1h(g1h), _cm(cm) { 2129 ReferenceProcessor* rp = _g1h->ref_processor_cm(); 2130 assert(rp->processing_is_mt(), "shouldn't be here otherwise"); 2131 } 2132 2133 virtual void work(uint worker_id) { 2134 ResourceMark rm; 2135 HandleMark hm; 2136 CMTask* task = _cm->task(worker_id); 2137 G1CMIsAliveClosure g1_is_alive(_g1h); 2138 G1CMKeepAliveAndDrainClosure g1_par_keep_alive(_cm, task, false /* is_serial */); 2139 G1CMDrainMarkingStackClosure g1_par_drain(_cm, task, false /* is_serial */); 2140 2141 _proc_task.work(worker_id, g1_is_alive, g1_par_keep_alive, g1_par_drain); 2142 } 2143 }; 2144 2145 void G1CMRefProcTaskExecutor::execute(ProcessTask& proc_task) { 2146 assert(_workers != NULL, "Need parallel worker threads."); 2147 assert(_g1h->ref_processor_cm()->processing_is_mt(), "processing is not MT"); 2148 2149 G1CMRefProcTaskProxy proc_task_proxy(proc_task, _g1h, _cm); 2150 2151 // We need to reset the concurrency level before each 2152 // proxy task execution, so that the termination protocol 2153 // and overflow handling in CMTask::do_marking_step() knows 2154 // how many workers to wait for. 2155 _cm->set_concurrency(_active_workers); 2156 _workers->run_task(&proc_task_proxy); 2157 } 2158 2159 class G1CMRefEnqueueTaskProxy: public AbstractGangTask { 2160 typedef AbstractRefProcTaskExecutor::EnqueueTask EnqueueTask; 2161 EnqueueTask& _enq_task; 2162 2163 public: 2164 G1CMRefEnqueueTaskProxy(EnqueueTask& enq_task) : 2165 AbstractGangTask("Enqueue reference objects in parallel"), 2166 _enq_task(enq_task) { } 2167 2168 virtual void work(uint worker_id) { 2169 _enq_task.work(worker_id); 2170 } 2171 }; 2172 2173 void G1CMRefProcTaskExecutor::execute(EnqueueTask& enq_task) { 2174 assert(_workers != NULL, "Need parallel worker threads."); 2175 assert(_g1h->ref_processor_cm()->processing_is_mt(), "processing is not MT"); 2176 2177 G1CMRefEnqueueTaskProxy enq_task_proxy(enq_task); 2178 2179 // Not strictly necessary but... 2180 // 2181 // We need to reset the concurrency level before each 2182 // proxy task execution, so that the termination protocol 2183 // and overflow handling in CMTask::do_marking_step() knows 2184 // how many workers to wait for. 2185 _cm->set_concurrency(_active_workers); 2186 _workers->run_task(&enq_task_proxy); 2187 } 2188 2189 void ConcurrentMark::weakRefsWorkParallelPart(BoolObjectClosure* is_alive, bool purged_classes) { 2190 G1CollectedHeap::heap()->parallel_cleaning(is_alive, true, true, purged_classes); 2191 } 2192 2193 void ConcurrentMark::weakRefsWork(bool clear_all_soft_refs) { 2194 if (has_overflown()) { 2195 // Skip processing the discovered references if we have 2196 // overflown the global marking stack. Reference objects 2197 // only get discovered once so it is OK to not 2198 // de-populate the discovered reference lists. We could have, 2199 // but the only benefit would be that, when marking restarts, 2200 // less reference objects are discovered. 2201 return; 2202 } 2203 2204 ResourceMark rm; 2205 HandleMark hm; 2206 2207 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 2208 2209 // Is alive closure. 2210 G1CMIsAliveClosure g1_is_alive(g1h); 2211 2212 // Inner scope to exclude the cleaning of the string and symbol 2213 // tables from the displayed time. 2214 { 2215 G1CMTraceTime t("GC ref-proc", G1Log::finer()); 2216 2217 ReferenceProcessor* rp = g1h->ref_processor_cm(); 2218 2219 // See the comment in G1CollectedHeap::ref_processing_init() 2220 // about how reference processing currently works in G1. 2221 2222 // Set the soft reference policy 2223 rp->setup_policy(clear_all_soft_refs); 2224 assert(_markStack.isEmpty(), "mark stack should be empty"); 2225 2226 // Instances of the 'Keep Alive' and 'Complete GC' closures used 2227 // in serial reference processing. Note these closures are also 2228 // used for serially processing (by the the current thread) the 2229 // JNI references during parallel reference processing. 2230 // 2231 // These closures do not need to synchronize with the worker 2232 // threads involved in parallel reference processing as these 2233 // instances are executed serially by the current thread (e.g. 2234 // reference processing is not multi-threaded and is thus 2235 // performed by the current thread instead of a gang worker). 2236 // 2237 // The gang tasks involved in parallel reference processing create 2238 // their own instances of these closures, which do their own 2239 // synchronization among themselves. 2240 G1CMKeepAliveAndDrainClosure g1_keep_alive(this, task(0), true /* is_serial */); 2241 G1CMDrainMarkingStackClosure g1_drain_mark_stack(this, task(0), true /* is_serial */); 2242 2243 // We need at least one active thread. If reference processing 2244 // is not multi-threaded we use the current (VMThread) thread, 2245 // otherwise we use the work gang from the G1CollectedHeap and 2246 // we utilize all the worker threads we can. 2247 bool processing_is_mt = rp->processing_is_mt(); 2248 uint active_workers = (processing_is_mt ? g1h->workers()->active_workers() : 1U); 2249 active_workers = MAX2(MIN2(active_workers, _max_worker_id), 1U); 2250 2251 // Parallel processing task executor. 2252 G1CMRefProcTaskExecutor par_task_executor(g1h, this, 2253 g1h->workers(), active_workers); 2254 AbstractRefProcTaskExecutor* executor = (processing_is_mt ? &par_task_executor : NULL); 2255 2256 // Set the concurrency level. The phase was already set prior to 2257 // executing the remark task. 2258 set_concurrency(active_workers); 2259 2260 // Set the degree of MT processing here. If the discovery was done MT, 2261 // the number of threads involved during discovery could differ from 2262 // the number of active workers. This is OK as long as the discovered 2263 // Reference lists are balanced (see balance_all_queues() and balance_queues()). 2264 rp->set_active_mt_degree(active_workers); 2265 2266 // Process the weak references. 2267 const ReferenceProcessorStats& stats = 2268 rp->process_discovered_references(&g1_is_alive, 2269 &g1_keep_alive, 2270 &g1_drain_mark_stack, 2271 executor, 2272 g1h->gc_timer_cm()); 2273 g1h->gc_tracer_cm()->report_gc_reference_stats(stats); 2274 2275 // The do_oop work routines of the keep_alive and drain_marking_stack 2276 // oop closures will set the has_overflown flag if we overflow the 2277 // global marking stack. 2278 2279 assert(_markStack.overflow() || _markStack.isEmpty(), 2280 "mark stack should be empty (unless it overflowed)"); 2281 2282 if (_markStack.overflow()) { 2283 // This should have been done already when we tried to push an 2284 // entry on to the global mark stack. But let's do it again. 2285 set_has_overflown(); 2286 } 2287 2288 assert(rp->num_q() == active_workers, "why not"); 2289 2290 rp->enqueue_discovered_references(executor); 2291 2292 rp->verify_no_references_recorded(); 2293 assert(!rp->discovery_enabled(), "Post condition"); 2294 } 2295 2296 if (has_overflown()) { 2297 // We can not trust g1_is_alive if the marking stack overflowed 2298 return; 2299 } 2300 2301 assert(_markStack.isEmpty(), "Marking should have completed"); 2302 2303 // Unload Klasses, String, Symbols, Code Cache, etc. 2304 { 2305 G1CMTraceTime trace("Unloading", G1Log::finer()); 2306 2307 if (ClassUnloadingWithConcurrentMark) { 2308 bool purged_classes; 2309 2310 { 2311 G1CMTraceTime trace("System Dictionary Unloading", G1Log::finest()); 2312 purged_classes = SystemDictionary::do_unloading(&g1_is_alive, false /* Defer klass cleaning */); 2313 } 2314 2315 { 2316 G1CMTraceTime trace("Parallel Unloading", G1Log::finest()); 2317 weakRefsWorkParallelPart(&g1_is_alive, purged_classes); 2318 } 2319 } 2320 2321 if (G1StringDedup::is_enabled()) { 2322 G1CMTraceTime trace("String Deduplication Unlink", G1Log::finest()); 2323 G1StringDedup::unlink(&g1_is_alive); 2324 } 2325 } 2326 } 2327 2328 void ConcurrentMark::swapMarkBitMaps() { 2329 CMBitMapRO* temp = _prevMarkBitMap; 2330 _prevMarkBitMap = (CMBitMapRO*)_nextMarkBitMap; 2331 _nextMarkBitMap = (CMBitMap*) temp; 2332 } 2333 2334 // Closure for marking entries in SATB buffers. 2335 class CMSATBBufferClosure : public SATBBufferClosure { 2336 private: 2337 CMTask* _task; 2338 G1CollectedHeap* _g1h; 2339 2340 // This is very similar to CMTask::deal_with_reference, but with 2341 // more relaxed requirements for the argument, so this must be more 2342 // circumspect about treating the argument as an object. 2343 void do_entry(void* entry) const { 2344 _task->increment_refs_reached(); 2345 HeapRegion* hr = _g1h->heap_region_containing_raw(entry); 2346 if (entry < hr->next_top_at_mark_start()) { 2347 // Until we get here, we don't know whether entry refers to a valid 2348 // object; it could instead have been a stale reference. 2349 oop obj = static_cast<oop>(entry); 2350 assert(obj->is_oop(true /* ignore mark word */), 2351 "Invalid oop in SATB buffer: " PTR_FORMAT, p2i(obj)); 2352 _task->make_reference_grey(obj, hr); 2353 } 2354 } 2355 2356 public: 2357 CMSATBBufferClosure(CMTask* task, G1CollectedHeap* g1h) 2358 : _task(task), _g1h(g1h) { } 2359 2360 virtual void do_buffer(void** buffer, size_t size) { 2361 for (size_t i = 0; i < size; ++i) { 2362 do_entry(buffer[i]); 2363 } 2364 } 2365 }; 2366 2367 class G1RemarkThreadsClosure : public ThreadClosure { 2368 CMSATBBufferClosure _cm_satb_cl; 2369 G1CMOopClosure _cm_cl; 2370 MarkingCodeBlobClosure _code_cl; 2371 int _thread_parity; 2372 2373 public: 2374 G1RemarkThreadsClosure(G1CollectedHeap* g1h, CMTask* task) : 2375 _cm_satb_cl(task, g1h), 2376 _cm_cl(g1h, g1h->concurrent_mark(), task), 2377 _code_cl(&_cm_cl, !CodeBlobToOopClosure::FixRelocations), 2378 _thread_parity(Threads::thread_claim_parity()) {} 2379 2380 void do_thread(Thread* thread) { 2381 if (thread->is_Java_thread()) { 2382 if (thread->claim_oops_do(true, _thread_parity)) { 2383 JavaThread* jt = (JavaThread*)thread; 2384 2385 // In theory it should not be neccessary to explicitly walk the nmethods to find roots for concurrent marking 2386 // however the liveness of oops reachable from nmethods have very complex lifecycles: 2387 // * Alive if on the stack of an executing method 2388 // * Weakly reachable otherwise 2389 // Some objects reachable from nmethods, such as the class loader (or klass_holder) of the receiver should be 2390 // live by the SATB invariant but other oops recorded in nmethods may behave differently. 2391 jt->nmethods_do(&_code_cl); 2392 2393 jt->satb_mark_queue().apply_closure_and_empty(&_cm_satb_cl); 2394 } 2395 } else if (thread->is_VM_thread()) { 2396 if (thread->claim_oops_do(true, _thread_parity)) { 2397 JavaThread::satb_mark_queue_set().shared_satb_queue()->apply_closure_and_empty(&_cm_satb_cl); 2398 } 2399 } 2400 } 2401 }; 2402 2403 class CMRemarkTask: public AbstractGangTask { 2404 private: 2405 ConcurrentMark* _cm; 2406 public: 2407 void work(uint worker_id) { 2408 // Since all available tasks are actually started, we should 2409 // only proceed if we're supposed to be active. 2410 if (worker_id < _cm->active_tasks()) { 2411 CMTask* task = _cm->task(worker_id); 2412 task->record_start_time(); 2413 { 2414 ResourceMark rm; 2415 HandleMark hm; 2416 2417 G1RemarkThreadsClosure threads_f(G1CollectedHeap::heap(), task); 2418 Threads::threads_do(&threads_f); 2419 } 2420 2421 do { 2422 task->do_marking_step(1000000000.0 /* something very large */, 2423 true /* do_termination */, 2424 false /* is_serial */); 2425 } while (task->has_aborted() && !_cm->has_overflown()); 2426 // If we overflow, then we do not want to restart. We instead 2427 // want to abort remark and do concurrent marking again. 2428 task->record_end_time(); 2429 } 2430 } 2431 2432 CMRemarkTask(ConcurrentMark* cm, uint active_workers) : 2433 AbstractGangTask("Par Remark"), _cm(cm) { 2434 _cm->terminator()->reset_for_reuse(active_workers); 2435 } 2436 }; 2437 2438 void ConcurrentMark::checkpointRootsFinalWork() { 2439 ResourceMark rm; 2440 HandleMark hm; 2441 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 2442 2443 G1CMTraceTime trace("Finalize Marking", G1Log::finer()); 2444 2445 g1h->ensure_parsability(false); 2446 2447 // this is remark, so we'll use up all active threads 2448 uint active_workers = g1h->workers()->active_workers(); 2449 set_concurrency_and_phase(active_workers, false /* concurrent */); 2450 // Leave _parallel_marking_threads at it's 2451 // value originally calculated in the ConcurrentMark 2452 // constructor and pass values of the active workers 2453 // through the gang in the task. 2454 2455 { 2456 StrongRootsScope srs(active_workers); 2457 2458 CMRemarkTask remarkTask(this, active_workers); 2459 // We will start all available threads, even if we decide that the 2460 // active_workers will be fewer. The extra ones will just bail out 2461 // immediately. 2462 g1h->workers()->run_task(&remarkTask); 2463 } 2464 2465 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 2466 guarantee(has_overflown() || 2467 satb_mq_set.completed_buffers_num() == 0, 2468 "Invariant: has_overflown = %s, num buffers = %d", 2469 BOOL_TO_STR(has_overflown()), 2470 satb_mq_set.completed_buffers_num()); 2471 2472 print_stats(); 2473 } 2474 2475 void ConcurrentMark::clearRangePrevBitmap(MemRegion mr) { 2476 // Note we are overriding the read-only view of the prev map here, via 2477 // the cast. 2478 ((CMBitMap*)_prevMarkBitMap)->clearRange(mr); 2479 } 2480 2481 void ConcurrentMark::clearRangeNextBitmap(MemRegion mr) { 2482 _nextMarkBitMap->clearRange(mr); 2483 } 2484 2485 HeapRegion* 2486 ConcurrentMark::claim_region(uint worker_id) { 2487 // "checkpoint" the finger 2488 HeapWord* finger = _finger; 2489 2490 // _heap_end will not change underneath our feet; it only changes at 2491 // yield points. 2492 while (finger < _heap_end) { 2493 assert(_g1h->is_in_g1_reserved(finger), "invariant"); 2494 2495 // Note on how this code handles humongous regions. In the 2496 // normal case the finger will reach the start of a "starts 2497 // humongous" (SH) region. Its end will either be the end of the 2498 // last "continues humongous" (CH) region in the sequence, or the 2499 // standard end of the SH region (if the SH is the only region in 2500 // the sequence). That way claim_region() will skip over the CH 2501 // regions. However, there is a subtle race between a CM thread 2502 // executing this method and a mutator thread doing a humongous 2503 // object allocation. The two are not mutually exclusive as the CM 2504 // thread does not need to hold the Heap_lock when it gets 2505 // here. So there is a chance that claim_region() will come across 2506 // a free region that's in the progress of becoming a SH or a CH 2507 // region. In the former case, it will either 2508 // a) Miss the update to the region's end, in which case it will 2509 // visit every subsequent CH region, will find their bitmaps 2510 // empty, and do nothing, or 2511 // b) Will observe the update of the region's end (in which case 2512 // it will skip the subsequent CH regions). 2513 // If it comes across a region that suddenly becomes CH, the 2514 // scenario will be similar to b). So, the race between 2515 // claim_region() and a humongous object allocation might force us 2516 // to do a bit of unnecessary work (due to some unnecessary bitmap 2517 // iterations) but it should not introduce and correctness issues. 2518 HeapRegion* curr_region = _g1h->heap_region_containing_raw(finger); 2519 2520 // Above heap_region_containing_raw may return NULL as we always scan claim 2521 // until the end of the heap. In this case, just jump to the next region. 2522 HeapWord* end = curr_region != NULL ? curr_region->end() : finger + HeapRegion::GrainWords; 2523 2524 // Is the gap between reading the finger and doing the CAS too long? 2525 HeapWord* res = (HeapWord*) Atomic::cmpxchg_ptr(end, &_finger, finger); 2526 if (res == finger && curr_region != NULL) { 2527 // we succeeded 2528 HeapWord* bottom = curr_region->bottom(); 2529 HeapWord* limit = curr_region->next_top_at_mark_start(); 2530 2531 // notice that _finger == end cannot be guaranteed here since, 2532 // someone else might have moved the finger even further 2533 assert(_finger >= end, "the finger should have moved forward"); 2534 2535 if (limit > bottom) { 2536 return curr_region; 2537 } else { 2538 assert(limit == bottom, 2539 "the region limit should be at bottom"); 2540 // we return NULL and the caller should try calling 2541 // claim_region() again. 2542 return NULL; 2543 } 2544 } else { 2545 assert(_finger > finger, "the finger should have moved forward"); 2546 // read it again 2547 finger = _finger; 2548 } 2549 } 2550 2551 return NULL; 2552 } 2553 2554 #ifndef PRODUCT 2555 class VerifyNoCSetOops VALUE_OBJ_CLASS_SPEC { 2556 private: 2557 G1CollectedHeap* _g1h; 2558 const char* _phase; 2559 int _info; 2560 2561 public: 2562 VerifyNoCSetOops(const char* phase, int info = -1) : 2563 _g1h(G1CollectedHeap::heap()), 2564 _phase(phase), 2565 _info(info) 2566 { } 2567 2568 void operator()(oop obj) const { 2569 guarantee(obj->is_oop(), 2570 "Non-oop " PTR_FORMAT ", phase: %s, info: %d", 2571 p2i(obj), _phase, _info); 2572 guarantee(!_g1h->obj_in_cs(obj), 2573 "obj: " PTR_FORMAT " in CSet, phase: %s, info: %d", 2574 p2i(obj), _phase, _info); 2575 } 2576 }; 2577 2578 void ConcurrentMark::verify_no_cset_oops() { 2579 assert(SafepointSynchronize::is_at_safepoint(), "should be at a safepoint"); 2580 if (!G1CollectedHeap::heap()->collector_state()->mark_in_progress()) { 2581 return; 2582 } 2583 2584 // Verify entries on the global mark stack 2585 _markStack.iterate(VerifyNoCSetOops("Stack")); 2586 2587 // Verify entries on the task queues 2588 for (uint i = 0; i < _max_worker_id; ++i) { 2589 CMTaskQueue* queue = _task_queues->queue(i); 2590 queue->iterate(VerifyNoCSetOops("Queue", i)); 2591 } 2592 2593 // Verify the global finger 2594 HeapWord* global_finger = finger(); 2595 if (global_finger != NULL && global_finger < _heap_end) { 2596 // The global finger always points to a heap region boundary. We 2597 // use heap_region_containing_raw() to get the containing region 2598 // given that the global finger could be pointing to a free region 2599 // which subsequently becomes continues humongous. If that 2600 // happens, heap_region_containing() will return the bottom of the 2601 // corresponding starts humongous region and the check below will 2602 // not hold any more. 2603 // Since we always iterate over all regions, we might get a NULL HeapRegion 2604 // here. 2605 HeapRegion* global_hr = _g1h->heap_region_containing_raw(global_finger); 2606 guarantee(global_hr == NULL || global_finger == global_hr->bottom(), 2607 "global finger: " PTR_FORMAT " region: " HR_FORMAT, 2608 p2i(global_finger), HR_FORMAT_PARAMS(global_hr)); 2609 } 2610 2611 // Verify the task fingers 2612 assert(parallel_marking_threads() <= _max_worker_id, "sanity"); 2613 for (uint i = 0; i < parallel_marking_threads(); ++i) { 2614 CMTask* task = _tasks[i]; 2615 HeapWord* task_finger = task->finger(); 2616 if (task_finger != NULL && task_finger < _heap_end) { 2617 // See above note on the global finger verification. 2618 HeapRegion* task_hr = _g1h->heap_region_containing_raw(task_finger); 2619 guarantee(task_hr == NULL || task_finger == task_hr->bottom() || 2620 !task_hr->in_collection_set(), 2621 "task finger: " PTR_FORMAT " region: " HR_FORMAT, 2622 p2i(task_finger), HR_FORMAT_PARAMS(task_hr)); 2623 } 2624 } 2625 } 2626 #endif // PRODUCT 2627 2628 // Aggregate the counting data that was constructed concurrently 2629 // with marking. 2630 class AggregateCountDataHRClosure: public HeapRegionClosure { 2631 G1CollectedHeap* _g1h; 2632 ConcurrentMark* _cm; 2633 CardTableModRefBS* _ct_bs; 2634 BitMap* _cm_card_bm; 2635 uint _max_worker_id; 2636 2637 public: 2638 AggregateCountDataHRClosure(G1CollectedHeap* g1h, 2639 BitMap* cm_card_bm, 2640 uint max_worker_id) : 2641 _g1h(g1h), _cm(g1h->concurrent_mark()), 2642 _ct_bs(barrier_set_cast<CardTableModRefBS>(g1h->barrier_set())), 2643 _cm_card_bm(cm_card_bm), _max_worker_id(max_worker_id) { } 2644 2645 bool doHeapRegion(HeapRegion* hr) { 2646 if (hr->is_continues_humongous()) { 2647 // We will ignore these here and process them when their 2648 // associated "starts humongous" region is processed. 2649 // Note that we cannot rely on their associated 2650 // "starts humongous" region to have their bit set to 1 2651 // since, due to the region chunking in the parallel region 2652 // iteration, a "continues humongous" region might be visited 2653 // before its associated "starts humongous". 2654 return false; 2655 } 2656 2657 HeapWord* start = hr->bottom(); 2658 HeapWord* limit = hr->next_top_at_mark_start(); 2659 HeapWord* end = hr->end(); 2660 2661 assert(start <= limit && limit <= hr->top() && hr->top() <= hr->end(), 2662 "Preconditions not met - " 2663 "start: " PTR_FORMAT ", limit: " PTR_FORMAT ", " 2664 "top: " PTR_FORMAT ", end: " PTR_FORMAT, 2665 p2i(start), p2i(limit), p2i(hr->top()), p2i(hr->end())); 2666 2667 assert(hr->next_marked_bytes() == 0, "Precondition"); 2668 2669 if (start == limit) { 2670 // NTAMS of this region has not been set so nothing to do. 2671 return false; 2672 } 2673 2674 // 'start' should be in the heap. 2675 assert(_g1h->is_in_g1_reserved(start) && _ct_bs->is_card_aligned(start), "sanity"); 2676 // 'end' *may* be just beyond the end of the heap (if hr is the last region) 2677 assert(!_g1h->is_in_g1_reserved(end) || _ct_bs->is_card_aligned(end), "sanity"); 2678 2679 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(start); 2680 BitMap::idx_t limit_idx = _cm->card_bitmap_index_for(limit); 2681 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(end); 2682 2683 // If ntams is not card aligned then we bump card bitmap index 2684 // for limit so that we get the all the cards spanned by 2685 // the object ending at ntams. 2686 // Note: if this is the last region in the heap then ntams 2687 // could be actually just beyond the end of the the heap; 2688 // limit_idx will then correspond to a (non-existent) card 2689 // that is also outside the heap. 2690 if (_g1h->is_in_g1_reserved(limit) && !_ct_bs->is_card_aligned(limit)) { 2691 limit_idx += 1; 2692 } 2693 2694 assert(limit_idx <= end_idx, "or else use atomics"); 2695 2696 // Aggregate the "stripe" in the count data associated with hr. 2697 uint hrm_index = hr->hrm_index(); 2698 size_t marked_bytes = 0; 2699 2700 for (uint i = 0; i < _max_worker_id; i += 1) { 2701 size_t* marked_bytes_array = _cm->count_marked_bytes_array_for(i); 2702 BitMap* task_card_bm = _cm->count_card_bitmap_for(i); 2703 2704 // Fetch the marked_bytes in this region for task i and 2705 // add it to the running total for this region. 2706 marked_bytes += marked_bytes_array[hrm_index]; 2707 2708 // Now union the bitmaps[0,max_worker_id)[start_idx..limit_idx) 2709 // into the global card bitmap. 2710 BitMap::idx_t scan_idx = task_card_bm->get_next_one_offset(start_idx, limit_idx); 2711 2712 while (scan_idx < limit_idx) { 2713 assert(task_card_bm->at(scan_idx) == true, "should be"); 2714 _cm_card_bm->set_bit(scan_idx); 2715 assert(_cm_card_bm->at(scan_idx) == true, "should be"); 2716 2717 // BitMap::get_next_one_offset() can handle the case when 2718 // its left_offset parameter is greater than its right_offset 2719 // parameter. It does, however, have an early exit if 2720 // left_offset == right_offset. So let's limit the value 2721 // passed in for left offset here. 2722 BitMap::idx_t next_idx = MIN2(scan_idx + 1, limit_idx); 2723 scan_idx = task_card_bm->get_next_one_offset(next_idx, limit_idx); 2724 } 2725 } 2726 2727 // Update the marked bytes for this region. 2728 hr->add_to_marked_bytes(marked_bytes); 2729 2730 // Next heap region 2731 return false; 2732 } 2733 }; 2734 2735 class G1AggregateCountDataTask: public AbstractGangTask { 2736 protected: 2737 G1CollectedHeap* _g1h; 2738 ConcurrentMark* _cm; 2739 BitMap* _cm_card_bm; 2740 uint _max_worker_id; 2741 uint _active_workers; 2742 HeapRegionClaimer _hrclaimer; 2743 2744 public: 2745 G1AggregateCountDataTask(G1CollectedHeap* g1h, 2746 ConcurrentMark* cm, 2747 BitMap* cm_card_bm, 2748 uint max_worker_id, 2749 uint n_workers) : 2750 AbstractGangTask("Count Aggregation"), 2751 _g1h(g1h), _cm(cm), _cm_card_bm(cm_card_bm), 2752 _max_worker_id(max_worker_id), 2753 _active_workers(n_workers), 2754 _hrclaimer(_active_workers) { 2755 } 2756 2757 void work(uint worker_id) { 2758 AggregateCountDataHRClosure cl(_g1h, _cm_card_bm, _max_worker_id); 2759 2760 _g1h->heap_region_par_iterate(&cl, worker_id, &_hrclaimer); 2761 } 2762 }; 2763 2764 2765 void ConcurrentMark::aggregate_count_data() { 2766 uint n_workers = _g1h->workers()->active_workers(); 2767 2768 G1AggregateCountDataTask g1_par_agg_task(_g1h, this, &_card_bm, 2769 _max_worker_id, n_workers); 2770 2771 _g1h->workers()->run_task(&g1_par_agg_task); 2772 } 2773 2774 // Clear the per-worker arrays used to store the per-region counting data 2775 void ConcurrentMark::clear_all_count_data() { 2776 // Clear the global card bitmap - it will be filled during 2777 // liveness count aggregation (during remark) and the 2778 // final counting task. 2779 _card_bm.clear(); 2780 2781 // Clear the global region bitmap - it will be filled as part 2782 // of the final counting task. 2783 _region_bm.clear(); 2784 2785 uint max_regions = _g1h->max_regions(); 2786 assert(_max_worker_id > 0, "uninitialized"); 2787 2788 for (uint i = 0; i < _max_worker_id; i += 1) { 2789 BitMap* task_card_bm = count_card_bitmap_for(i); 2790 size_t* marked_bytes_array = count_marked_bytes_array_for(i); 2791 2792 assert(task_card_bm->size() == _card_bm.size(), "size mismatch"); 2793 assert(marked_bytes_array != NULL, "uninitialized"); 2794 2795 memset(marked_bytes_array, 0, (size_t) max_regions * sizeof(size_t)); 2796 task_card_bm->clear(); 2797 } 2798 } 2799 2800 void ConcurrentMark::print_stats() { 2801 if (G1MarkingVerboseLevel > 0) { 2802 gclog_or_tty->print_cr("---------------------------------------------------------------------"); 2803 for (size_t i = 0; i < _active_tasks; ++i) { 2804 _tasks[i]->print_stats(); 2805 gclog_or_tty->print_cr("---------------------------------------------------------------------"); 2806 } 2807 } 2808 } 2809 2810 // abandon current marking iteration due to a Full GC 2811 void ConcurrentMark::abort() { 2812 if (!cmThread()->during_cycle() || _has_aborted) { 2813 // We haven't started a concurrent cycle or we have already aborted it. No need to do anything. 2814 return; 2815 } 2816 2817 // Clear all marks in the next bitmap for the next marking cycle. This will allow us to skip the next 2818 // concurrent bitmap clearing. 2819 _nextMarkBitMap->clearAll(); 2820 2821 // Note we cannot clear the previous marking bitmap here 2822 // since VerifyDuringGC verifies the objects marked during 2823 // a full GC against the previous bitmap. 2824 2825 // Clear the liveness counting data 2826 clear_all_count_data(); 2827 // Empty mark stack 2828 reset_marking_state(); 2829 for (uint i = 0; i < _max_worker_id; ++i) { 2830 _tasks[i]->clear_region_fields(); 2831 } 2832 _first_overflow_barrier_sync.abort(); 2833 _second_overflow_barrier_sync.abort(); 2834 _has_aborted = true; 2835 2836 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 2837 satb_mq_set.abandon_partial_marking(); 2838 // This can be called either during or outside marking, we'll read 2839 // the expected_active value from the SATB queue set. 2840 satb_mq_set.set_active_all_threads( 2841 false, /* new active value */ 2842 satb_mq_set.is_active() /* expected_active */); 2843 2844 _g1h->trace_heap_after_concurrent_cycle(); 2845 _g1h->register_concurrent_cycle_end(); 2846 } 2847 2848 static void print_ms_time_info(const char* prefix, const char* name, 2849 NumberSeq& ns) { 2850 gclog_or_tty->print_cr("%s%5d %12s: total time = %8.2f s (avg = %8.2f ms).", 2851 prefix, ns.num(), name, ns.sum()/1000.0, ns.avg()); 2852 if (ns.num() > 0) { 2853 gclog_or_tty->print_cr("%s [std. dev = %8.2f ms, max = %8.2f ms]", 2854 prefix, ns.sd(), ns.maximum()); 2855 } 2856 } 2857 2858 void ConcurrentMark::print_summary_info() { 2859 gclog_or_tty->print_cr(" Concurrent marking:"); 2860 print_ms_time_info(" ", "init marks", _init_times); 2861 print_ms_time_info(" ", "remarks", _remark_times); 2862 { 2863 print_ms_time_info(" ", "final marks", _remark_mark_times); 2864 print_ms_time_info(" ", "weak refs", _remark_weak_ref_times); 2865 2866 } 2867 print_ms_time_info(" ", "cleanups", _cleanup_times); 2868 gclog_or_tty->print_cr(" Final counting total time = %8.2f s (avg = %8.2f ms).", 2869 _total_counting_time, 2870 (_cleanup_times.num() > 0 ? _total_counting_time * 1000.0 / 2871 (double)_cleanup_times.num() 2872 : 0.0)); 2873 if (G1ScrubRemSets) { 2874 gclog_or_tty->print_cr(" RS scrub total time = %8.2f s (avg = %8.2f ms).", 2875 _total_rs_scrub_time, 2876 (_cleanup_times.num() > 0 ? _total_rs_scrub_time * 1000.0 / 2877 (double)_cleanup_times.num() 2878 : 0.0)); 2879 } 2880 gclog_or_tty->print_cr(" Total stop_world time = %8.2f s.", 2881 (_init_times.sum() + _remark_times.sum() + 2882 _cleanup_times.sum())/1000.0); 2883 gclog_or_tty->print_cr(" Total concurrent time = %8.2f s " 2884 "(%8.2f s marking).", 2885 cmThread()->vtime_accum(), 2886 cmThread()->vtime_mark_accum()); 2887 } 2888 2889 void ConcurrentMark::print_worker_threads_on(outputStream* st) const { 2890 _parallel_workers->print_worker_threads_on(st); 2891 } 2892 2893 void ConcurrentMark::print_on_error(outputStream* st) const { 2894 st->print_cr("Marking Bits (Prev, Next): (CMBitMap*) " PTR_FORMAT ", (CMBitMap*) " PTR_FORMAT, 2895 p2i(_prevMarkBitMap), p2i(_nextMarkBitMap)); 2896 _prevMarkBitMap->print_on_error(st, " Prev Bits: "); 2897 _nextMarkBitMap->print_on_error(st, " Next Bits: "); 2898 } 2899 2900 // We take a break if someone is trying to stop the world. 2901 bool ConcurrentMark::do_yield_check(uint worker_id) { 2902 if (SuspendibleThreadSet::should_yield()) { 2903 if (worker_id == 0) { 2904 _g1h->g1_policy()->record_concurrent_pause(); 2905 } 2906 SuspendibleThreadSet::yield(); 2907 return true; 2908 } else { 2909 return false; 2910 } 2911 } 2912 2913 // Closure for iteration over bitmaps 2914 class CMBitMapClosure : public BitMapClosure { 2915 private: 2916 // the bitmap that is being iterated over 2917 CMBitMap* _nextMarkBitMap; 2918 ConcurrentMark* _cm; 2919 CMTask* _task; 2920 2921 public: 2922 CMBitMapClosure(CMTask *task, ConcurrentMark* cm, CMBitMap* nextMarkBitMap) : 2923 _task(task), _cm(cm), _nextMarkBitMap(nextMarkBitMap) { } 2924 2925 bool do_bit(size_t offset) { 2926 HeapWord* addr = _nextMarkBitMap->offsetToHeapWord(offset); 2927 assert(_nextMarkBitMap->isMarked(addr), "invariant"); 2928 assert( addr < _cm->finger(), "invariant"); 2929 assert(addr >= _task->finger(), "invariant"); 2930 2931 // We move that task's local finger along. 2932 _task->move_finger_to(addr); 2933 2934 _task->scan_object(oop(addr)); 2935 // we only partially drain the local queue and global stack 2936 _task->drain_local_queue(true); 2937 _task->drain_global_stack(true); 2938 2939 // if the has_aborted flag has been raised, we need to bail out of 2940 // the iteration 2941 return !_task->has_aborted(); 2942 } 2943 }; 2944 2945 static ReferenceProcessor* get_cm_oop_closure_ref_processor(G1CollectedHeap* g1h) { 2946 ReferenceProcessor* result = NULL; 2947 if (G1UseConcMarkReferenceProcessing) { 2948 result = g1h->ref_processor_cm(); 2949 assert(result != NULL, "should not be NULL"); 2950 } 2951 return result; 2952 } 2953 2954 G1CMOopClosure::G1CMOopClosure(G1CollectedHeap* g1h, 2955 ConcurrentMark* cm, 2956 CMTask* task) 2957 : MetadataAwareOopClosure(get_cm_oop_closure_ref_processor(g1h)), 2958 _g1h(g1h), _cm(cm), _task(task) 2959 { } 2960 2961 void CMTask::setup_for_region(HeapRegion* hr) { 2962 assert(hr != NULL, 2963 "claim_region() should have filtered out NULL regions"); 2964 assert(!hr->is_continues_humongous(), 2965 "claim_region() should have filtered out continues humongous regions"); 2966 _curr_region = hr; 2967 _finger = hr->bottom(); 2968 update_region_limit(); 2969 } 2970 2971 void CMTask::update_region_limit() { 2972 HeapRegion* hr = _curr_region; 2973 HeapWord* bottom = hr->bottom(); 2974 HeapWord* limit = hr->next_top_at_mark_start(); 2975 2976 if (limit == bottom) { 2977 // The region was collected underneath our feet. 2978 // We set the finger to bottom to ensure that the bitmap 2979 // iteration that will follow this will not do anything. 2980 // (this is not a condition that holds when we set the region up, 2981 // as the region is not supposed to be empty in the first place) 2982 _finger = bottom; 2983 } else if (limit >= _region_limit) { 2984 assert(limit >= _finger, "peace of mind"); 2985 } else { 2986 assert(limit < _region_limit, "only way to get here"); 2987 // This can happen under some pretty unusual circumstances. An 2988 // evacuation pause empties the region underneath our feet (NTAMS 2989 // at bottom). We then do some allocation in the region (NTAMS 2990 // stays at bottom), followed by the region being used as a GC 2991 // alloc region (NTAMS will move to top() and the objects 2992 // originally below it will be grayed). All objects now marked in 2993 // the region are explicitly grayed, if below the global finger, 2994 // and we do not need in fact to scan anything else. So, we simply 2995 // set _finger to be limit to ensure that the bitmap iteration 2996 // doesn't do anything. 2997 _finger = limit; 2998 } 2999 3000 _region_limit = limit; 3001 } 3002 3003 void CMTask::giveup_current_region() { 3004 assert(_curr_region != NULL, "invariant"); 3005 clear_region_fields(); 3006 } 3007 3008 void CMTask::clear_region_fields() { 3009 // Values for these three fields that indicate that we're not 3010 // holding on to a region. 3011 _curr_region = NULL; 3012 _finger = NULL; 3013 _region_limit = NULL; 3014 } 3015 3016 void CMTask::set_cm_oop_closure(G1CMOopClosure* cm_oop_closure) { 3017 if (cm_oop_closure == NULL) { 3018 assert(_cm_oop_closure != NULL, "invariant"); 3019 } else { 3020 assert(_cm_oop_closure == NULL, "invariant"); 3021 } 3022 _cm_oop_closure = cm_oop_closure; 3023 } 3024 3025 void CMTask::reset(CMBitMap* nextMarkBitMap) { 3026 guarantee(nextMarkBitMap != NULL, "invariant"); 3027 _nextMarkBitMap = nextMarkBitMap; 3028 clear_region_fields(); 3029 3030 _calls = 0; 3031 _elapsed_time_ms = 0.0; 3032 _termination_time_ms = 0.0; 3033 _termination_start_time_ms = 0.0; 3034 } 3035 3036 bool CMTask::should_exit_termination() { 3037 regular_clock_call(); 3038 // This is called when we are in the termination protocol. We should 3039 // quit if, for some reason, this task wants to abort or the global 3040 // stack is not empty (this means that we can get work from it). 3041 return !_cm->mark_stack_empty() || has_aborted(); 3042 } 3043 3044 void CMTask::reached_limit() { 3045 assert(_words_scanned >= _words_scanned_limit || 3046 _refs_reached >= _refs_reached_limit , 3047 "shouldn't have been called otherwise"); 3048 regular_clock_call(); 3049 } 3050 3051 void CMTask::regular_clock_call() { 3052 if (has_aborted()) return; 3053 3054 // First, we need to recalculate the words scanned and refs reached 3055 // limits for the next clock call. 3056 recalculate_limits(); 3057 3058 // During the regular clock call we do the following 3059 3060 // (1) If an overflow has been flagged, then we abort. 3061 if (_cm->has_overflown()) { 3062 set_has_aborted(); 3063 return; 3064 } 3065 3066 // If we are not concurrent (i.e. we're doing remark) we don't need 3067 // to check anything else. The other steps are only needed during 3068 // the concurrent marking phase. 3069 if (!concurrent()) return; 3070 3071 // (2) If marking has been aborted for Full GC, then we also abort. 3072 if (_cm->has_aborted()) { 3073 set_has_aborted(); 3074 return; 3075 } 3076 3077 double curr_time_ms = os::elapsedVTime() * 1000.0; 3078 3079 // (4) We check whether we should yield. If we have to, then we abort. 3080 if (SuspendibleThreadSet::should_yield()) { 3081 // We should yield. To do this we abort the task. The caller is 3082 // responsible for yielding. 3083 set_has_aborted(); 3084 return; 3085 } 3086 3087 // (5) We check whether we've reached our time quota. If we have, 3088 // then we abort. 3089 double elapsed_time_ms = curr_time_ms - _start_time_ms; 3090 if (elapsed_time_ms > _time_target_ms) { 3091 set_has_aborted(); 3092 _has_timed_out = true; 3093 return; 3094 } 3095 3096 // (6) Finally, we check whether there are enough completed STAB 3097 // buffers available for processing. If there are, we abort. 3098 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 3099 if (!_draining_satb_buffers && satb_mq_set.process_completed_buffers()) { 3100 // we do need to process SATB buffers, we'll abort and restart 3101 // the marking task to do so 3102 set_has_aborted(); 3103 return; 3104 } 3105 } 3106 3107 void CMTask::recalculate_limits() { 3108 _real_words_scanned_limit = _words_scanned + words_scanned_period; 3109 _words_scanned_limit = _real_words_scanned_limit; 3110 3111 _real_refs_reached_limit = _refs_reached + refs_reached_period; 3112 _refs_reached_limit = _real_refs_reached_limit; 3113 } 3114 3115 void CMTask::decrease_limits() { 3116 // This is called when we believe that we're going to do an infrequent 3117 // operation which will increase the per byte scanned cost (i.e. move 3118 // entries to/from the global stack). It basically tries to decrease the 3119 // scanning limit so that the clock is called earlier. 3120 3121 _words_scanned_limit = _real_words_scanned_limit - 3122 3 * words_scanned_period / 4; 3123 _refs_reached_limit = _real_refs_reached_limit - 3124 3 * refs_reached_period / 4; 3125 } 3126 3127 void CMTask::move_entries_to_global_stack() { 3128 // local array where we'll store the entries that will be popped 3129 // from the local queue 3130 oop buffer[global_stack_transfer_size]; 3131 3132 int n = 0; 3133 oop obj; 3134 while (n < global_stack_transfer_size && _task_queue->pop_local(obj)) { 3135 buffer[n] = obj; 3136 ++n; 3137 } 3138 3139 if (n > 0) { 3140 // we popped at least one entry from the local queue 3141 3142 if (!_cm->mark_stack_push(buffer, n)) { 3143 set_has_aborted(); 3144 } 3145 } 3146 3147 // this operation was quite expensive, so decrease the limits 3148 decrease_limits(); 3149 } 3150 3151 void CMTask::get_entries_from_global_stack() { 3152 // local array where we'll store the entries that will be popped 3153 // from the global stack. 3154 oop buffer[global_stack_transfer_size]; 3155 int n; 3156 _cm->mark_stack_pop(buffer, global_stack_transfer_size, &n); 3157 assert(n <= global_stack_transfer_size, 3158 "we should not pop more than the given limit"); 3159 if (n > 0) { 3160 // yes, we did actually pop at least one entry 3161 for (int i = 0; i < n; ++i) { 3162 bool success = _task_queue->push(buffer[i]); 3163 // We only call this when the local queue is empty or under a 3164 // given target limit. So, we do not expect this push to fail. 3165 assert(success, "invariant"); 3166 } 3167 } 3168 3169 // this operation was quite expensive, so decrease the limits 3170 decrease_limits(); 3171 } 3172 3173 void CMTask::drain_local_queue(bool partially) { 3174 if (has_aborted()) return; 3175 3176 // Decide what the target size is, depending whether we're going to 3177 // drain it partially (so that other tasks can steal if they run out 3178 // of things to do) or totally (at the very end). 3179 size_t target_size; 3180 if (partially) { 3181 target_size = MIN2((size_t)_task_queue->max_elems()/3, GCDrainStackTargetSize); 3182 } else { 3183 target_size = 0; 3184 } 3185 3186 if (_task_queue->size() > target_size) { 3187 oop obj; 3188 bool ret = _task_queue->pop_local(obj); 3189 while (ret) { 3190 assert(_g1h->is_in_g1_reserved((HeapWord*) obj), "invariant" ); 3191 assert(!_g1h->is_on_master_free_list( 3192 _g1h->heap_region_containing((HeapWord*) obj)), "invariant"); 3193 3194 scan_object(obj); 3195 3196 if (_task_queue->size() <= target_size || has_aborted()) { 3197 ret = false; 3198 } else { 3199 ret = _task_queue->pop_local(obj); 3200 } 3201 } 3202 } 3203 } 3204 3205 void CMTask::drain_global_stack(bool partially) { 3206 if (has_aborted()) return; 3207 3208 // We have a policy to drain the local queue before we attempt to 3209 // drain the global stack. 3210 assert(partially || _task_queue->size() == 0, "invariant"); 3211 3212 // Decide what the target size is, depending whether we're going to 3213 // drain it partially (so that other tasks can steal if they run out 3214 // of things to do) or totally (at the very end). Notice that, 3215 // because we move entries from the global stack in chunks or 3216 // because another task might be doing the same, we might in fact 3217 // drop below the target. But, this is not a problem. 3218 size_t target_size; 3219 if (partially) { 3220 target_size = _cm->partial_mark_stack_size_target(); 3221 } else { 3222 target_size = 0; 3223 } 3224 3225 if (_cm->mark_stack_size() > target_size) { 3226 while (!has_aborted() && _cm->mark_stack_size() > target_size) { 3227 get_entries_from_global_stack(); 3228 drain_local_queue(partially); 3229 } 3230 } 3231 } 3232 3233 // SATB Queue has several assumptions on whether to call the par or 3234 // non-par versions of the methods. this is why some of the code is 3235 // replicated. We should really get rid of the single-threaded version 3236 // of the code to simplify things. 3237 void CMTask::drain_satb_buffers() { 3238 if (has_aborted()) return; 3239 3240 // We set this so that the regular clock knows that we're in the 3241 // middle of draining buffers and doesn't set the abort flag when it 3242 // notices that SATB buffers are available for draining. It'd be 3243 // very counter productive if it did that. :-) 3244 _draining_satb_buffers = true; 3245 3246 CMSATBBufferClosure satb_cl(this, _g1h); 3247 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set(); 3248 3249 // This keeps claiming and applying the closure to completed buffers 3250 // until we run out of buffers or we need to abort. 3251 while (!has_aborted() && 3252 satb_mq_set.apply_closure_to_completed_buffer(&satb_cl)) { 3253 regular_clock_call(); 3254 } 3255 3256 _draining_satb_buffers = false; 3257 3258 assert(has_aborted() || 3259 concurrent() || 3260 satb_mq_set.completed_buffers_num() == 0, "invariant"); 3261 3262 // again, this was a potentially expensive operation, decrease the 3263 // limits to get the regular clock call early 3264 decrease_limits(); 3265 } 3266 3267 void CMTask::print_stats() { 3268 gclog_or_tty->print_cr("Marking Stats, task = %u, calls = %d", 3269 _worker_id, _calls); 3270 gclog_or_tty->print_cr(" Elapsed time = %1.2lfms, Termination time = %1.2lfms", 3271 _elapsed_time_ms, _termination_time_ms); 3272 gclog_or_tty->print_cr(" Step Times (cum): num = %d, avg = %1.2lfms, sd = %1.2lfms", 3273 _step_times_ms.num(), _step_times_ms.avg(), 3274 _step_times_ms.sd()); 3275 gclog_or_tty->print_cr(" max = %1.2lfms, total = %1.2lfms", 3276 _step_times_ms.maximum(), _step_times_ms.sum()); 3277 } 3278 3279 bool ConcurrentMark::try_stealing(uint worker_id, int* hash_seed, oop& obj) { 3280 return _task_queues->steal(worker_id, hash_seed, obj); 3281 } 3282 3283 /***************************************************************************** 3284 3285 The do_marking_step(time_target_ms, ...) method is the building 3286 block of the parallel marking framework. It can be called in parallel 3287 with other invocations of do_marking_step() on different tasks 3288 (but only one per task, obviously) and concurrently with the 3289 mutator threads, or during remark, hence it eliminates the need 3290 for two versions of the code. When called during remark, it will 3291 pick up from where the task left off during the concurrent marking 3292 phase. Interestingly, tasks are also claimable during evacuation 3293 pauses too, since do_marking_step() ensures that it aborts before 3294 it needs to yield. 3295 3296 The data structures that it uses to do marking work are the 3297 following: 3298 3299 (1) Marking Bitmap. If there are gray objects that appear only 3300 on the bitmap (this happens either when dealing with an overflow 3301 or when the initial marking phase has simply marked the roots 3302 and didn't push them on the stack), then tasks claim heap 3303 regions whose bitmap they then scan to find gray objects. A 3304 global finger indicates where the end of the last claimed region 3305 is. A local finger indicates how far into the region a task has 3306 scanned. The two fingers are used to determine how to gray an 3307 object (i.e. whether simply marking it is OK, as it will be 3308 visited by a task in the future, or whether it needs to be also 3309 pushed on a stack). 3310 3311 (2) Local Queue. The local queue of the task which is accessed 3312 reasonably efficiently by the task. Other tasks can steal from 3313 it when they run out of work. Throughout the marking phase, a 3314 task attempts to keep its local queue short but not totally 3315 empty, so that entries are available for stealing by other 3316 tasks. Only when there is no more work, a task will totally 3317 drain its local queue. 3318 3319 (3) Global Mark Stack. This handles local queue overflow. During 3320 marking only sets of entries are moved between it and the local 3321 queues, as access to it requires a mutex and more fine-grain 3322 interaction with it which might cause contention. If it 3323 overflows, then the marking phase should restart and iterate 3324 over the bitmap to identify gray objects. Throughout the marking 3325 phase, tasks attempt to keep the global mark stack at a small 3326 length but not totally empty, so that entries are available for 3327 popping by other tasks. Only when there is no more work, tasks 3328 will totally drain the global mark stack. 3329 3330 (4) SATB Buffer Queue. This is where completed SATB buffers are 3331 made available. Buffers are regularly removed from this queue 3332 and scanned for roots, so that the queue doesn't get too 3333 long. During remark, all completed buffers are processed, as 3334 well as the filled in parts of any uncompleted buffers. 3335 3336 The do_marking_step() method tries to abort when the time target 3337 has been reached. There are a few other cases when the 3338 do_marking_step() method also aborts: 3339 3340 (1) When the marking phase has been aborted (after a Full GC). 3341 3342 (2) When a global overflow (on the global stack) has been 3343 triggered. Before the task aborts, it will actually sync up with 3344 the other tasks to ensure that all the marking data structures 3345 (local queues, stacks, fingers etc.) are re-initialized so that 3346 when do_marking_step() completes, the marking phase can 3347 immediately restart. 3348 3349 (3) When enough completed SATB buffers are available. The 3350 do_marking_step() method only tries to drain SATB buffers right 3351 at the beginning. So, if enough buffers are available, the 3352 marking step aborts and the SATB buffers are processed at 3353 the beginning of the next invocation. 3354 3355 (4) To yield. when we have to yield then we abort and yield 3356 right at the end of do_marking_step(). This saves us from a lot 3357 of hassle as, by yielding we might allow a Full GC. If this 3358 happens then objects will be compacted underneath our feet, the 3359 heap might shrink, etc. We save checking for this by just 3360 aborting and doing the yield right at the end. 3361 3362 From the above it follows that the do_marking_step() method should 3363 be called in a loop (or, otherwise, regularly) until it completes. 3364 3365 If a marking step completes without its has_aborted() flag being 3366 true, it means it has completed the current marking phase (and 3367 also all other marking tasks have done so and have all synced up). 3368 3369 A method called regular_clock_call() is invoked "regularly" (in 3370 sub ms intervals) throughout marking. It is this clock method that 3371 checks all the abort conditions which were mentioned above and 3372 decides when the task should abort. A work-based scheme is used to 3373 trigger this clock method: when the number of object words the 3374 marking phase has scanned or the number of references the marking 3375 phase has visited reach a given limit. Additional invocations to 3376 the method clock have been planted in a few other strategic places 3377 too. The initial reason for the clock method was to avoid calling 3378 vtime too regularly, as it is quite expensive. So, once it was in 3379 place, it was natural to piggy-back all the other conditions on it 3380 too and not constantly check them throughout the code. 3381 3382 If do_termination is true then do_marking_step will enter its 3383 termination protocol. 3384 3385 The value of is_serial must be true when do_marking_step is being 3386 called serially (i.e. by the VMThread) and do_marking_step should 3387 skip any synchronization in the termination and overflow code. 3388 Examples include the serial remark code and the serial reference 3389 processing closures. 3390 3391 The value of is_serial must be false when do_marking_step is 3392 being called by any of the worker threads in a work gang. 3393 Examples include the concurrent marking code (CMMarkingTask), 3394 the MT remark code, and the MT reference processing closures. 3395 3396 *****************************************************************************/ 3397 3398 void CMTask::do_marking_step(double time_target_ms, 3399 bool do_termination, 3400 bool is_serial) { 3401 assert(time_target_ms >= 1.0, "minimum granularity is 1ms"); 3402 assert(concurrent() == _cm->concurrent(), "they should be the same"); 3403 3404 G1CollectorPolicy* g1_policy = _g1h->g1_policy(); 3405 assert(_task_queues != NULL, "invariant"); 3406 assert(_task_queue != NULL, "invariant"); 3407 assert(_task_queues->queue(_worker_id) == _task_queue, "invariant"); 3408 3409 assert(!_claimed, 3410 "only one thread should claim this task at any one time"); 3411 3412 // OK, this doesn't safeguard again all possible scenarios, as it is 3413 // possible for two threads to set the _claimed flag at the same 3414 // time. But it is only for debugging purposes anyway and it will 3415 // catch most problems. 3416 _claimed = true; 3417 3418 _start_time_ms = os::elapsedVTime() * 1000.0; 3419 3420 // If do_stealing is true then do_marking_step will attempt to 3421 // steal work from the other CMTasks. It only makes sense to 3422 // enable stealing when the termination protocol is enabled 3423 // and do_marking_step() is not being called serially. 3424 bool do_stealing = do_termination && !is_serial; 3425 3426 double diff_prediction_ms = _g1h->g1_policy()->predictor().get_new_prediction(&_marking_step_diffs_ms); 3427 _time_target_ms = time_target_ms - diff_prediction_ms; 3428 3429 // set up the variables that are used in the work-based scheme to 3430 // call the regular clock method 3431 _words_scanned = 0; 3432 _refs_reached = 0; 3433 recalculate_limits(); 3434 3435 // clear all flags 3436 clear_has_aborted(); 3437 _has_timed_out = false; 3438 _draining_satb_buffers = false; 3439 3440 ++_calls; 3441 3442 // Set up the bitmap and oop closures. Anything that uses them is 3443 // eventually called from this method, so it is OK to allocate these 3444 // statically. 3445 CMBitMapClosure bitmap_closure(this, _cm, _nextMarkBitMap); 3446 G1CMOopClosure cm_oop_closure(_g1h, _cm, this); 3447 set_cm_oop_closure(&cm_oop_closure); 3448 3449 if (_cm->has_overflown()) { 3450 // This can happen if the mark stack overflows during a GC pause 3451 // and this task, after a yield point, restarts. We have to abort 3452 // as we need to get into the overflow protocol which happens 3453 // right at the end of this task. 3454 set_has_aborted(); 3455 } 3456 3457 // First drain any available SATB buffers. After this, we will not 3458 // look at SATB buffers before the next invocation of this method. 3459 // If enough completed SATB buffers are queued up, the regular clock 3460 // will abort this task so that it restarts. 3461 drain_satb_buffers(); 3462 // ...then partially drain the local queue and the global stack 3463 drain_local_queue(true); 3464 drain_global_stack(true); 3465 3466 do { 3467 if (!has_aborted() && _curr_region != NULL) { 3468 // This means that we're already holding on to a region. 3469 assert(_finger != NULL, "if region is not NULL, then the finger " 3470 "should not be NULL either"); 3471 3472 // We might have restarted this task after an evacuation pause 3473 // which might have evacuated the region we're holding on to 3474 // underneath our feet. Let's read its limit again to make sure 3475 // that we do not iterate over a region of the heap that 3476 // contains garbage (update_region_limit() will also move 3477 // _finger to the start of the region if it is found empty). 3478 update_region_limit(); 3479 // We will start from _finger not from the start of the region, 3480 // as we might be restarting this task after aborting half-way 3481 // through scanning this region. In this case, _finger points to 3482 // the address where we last found a marked object. If this is a 3483 // fresh region, _finger points to start(). 3484 MemRegion mr = MemRegion(_finger, _region_limit); 3485 3486 assert(!_curr_region->is_humongous() || mr.start() == _curr_region->bottom(), 3487 "humongous regions should go around loop once only"); 3488 3489 // Some special cases: 3490 // If the memory region is empty, we can just give up the region. 3491 // If the current region is humongous then we only need to check 3492 // the bitmap for the bit associated with the start of the object, 3493 // scan the object if it's live, and give up the region. 3494 // Otherwise, let's iterate over the bitmap of the part of the region 3495 // that is left. 3496 // If the iteration is successful, give up the region. 3497 if (mr.is_empty()) { 3498 giveup_current_region(); 3499 regular_clock_call(); 3500 } else if (_curr_region->is_humongous() && mr.start() == _curr_region->bottom()) { 3501 if (_nextMarkBitMap->isMarked(mr.start())) { 3502 // The object is marked - apply the closure 3503 BitMap::idx_t offset = _nextMarkBitMap->heapWordToOffset(mr.start()); 3504 bitmap_closure.do_bit(offset); 3505 } 3506 // Even if this task aborted while scanning the humongous object 3507 // we can (and should) give up the current region. 3508 giveup_current_region(); 3509 regular_clock_call(); 3510 } else if (_nextMarkBitMap->iterate(&bitmap_closure, mr)) { 3511 giveup_current_region(); 3512 regular_clock_call(); 3513 } else { 3514 assert(has_aborted(), "currently the only way to do so"); 3515 // The only way to abort the bitmap iteration is to return 3516 // false from the do_bit() method. However, inside the 3517 // do_bit() method we move the _finger to point to the 3518 // object currently being looked at. So, if we bail out, we 3519 // have definitely set _finger to something non-null. 3520 assert(_finger != NULL, "invariant"); 3521 3522 // Region iteration was actually aborted. So now _finger 3523 // points to the address of the object we last scanned. If we 3524 // leave it there, when we restart this task, we will rescan 3525 // the object. It is easy to avoid this. We move the finger by 3526 // enough to point to the next possible object header (the 3527 // bitmap knows by how much we need to move it as it knows its 3528 // granularity). 3529 assert(_finger < _region_limit, "invariant"); 3530 HeapWord* new_finger = _nextMarkBitMap->nextObject(_finger); 3531 // Check if bitmap iteration was aborted while scanning the last object 3532 if (new_finger >= _region_limit) { 3533 giveup_current_region(); 3534 } else { 3535 move_finger_to(new_finger); 3536 } 3537 } 3538 } 3539 // At this point we have either completed iterating over the 3540 // region we were holding on to, or we have aborted. 3541 3542 // We then partially drain the local queue and the global stack. 3543 // (Do we really need this?) 3544 drain_local_queue(true); 3545 drain_global_stack(true); 3546 3547 // Read the note on the claim_region() method on why it might 3548 // return NULL with potentially more regions available for 3549 // claiming and why we have to check out_of_regions() to determine 3550 // whether we're done or not. 3551 while (!has_aborted() && _curr_region == NULL && !_cm->out_of_regions()) { 3552 // We are going to try to claim a new region. We should have 3553 // given up on the previous one. 3554 // Separated the asserts so that we know which one fires. 3555 assert(_curr_region == NULL, "invariant"); 3556 assert(_finger == NULL, "invariant"); 3557 assert(_region_limit == NULL, "invariant"); 3558 HeapRegion* claimed_region = _cm->claim_region(_worker_id); 3559 if (claimed_region != NULL) { 3560 // Yes, we managed to claim one 3561 setup_for_region(claimed_region); 3562 assert(_curr_region == claimed_region, "invariant"); 3563 } 3564 // It is important to call the regular clock here. It might take 3565 // a while to claim a region if, for example, we hit a large 3566 // block of empty regions. So we need to call the regular clock 3567 // method once round the loop to make sure it's called 3568 // frequently enough. 3569 regular_clock_call(); 3570 } 3571 3572 if (!has_aborted() && _curr_region == NULL) { 3573 assert(_cm->out_of_regions(), 3574 "at this point we should be out of regions"); 3575 } 3576 } while ( _curr_region != NULL && !has_aborted()); 3577 3578 if (!has_aborted()) { 3579 // We cannot check whether the global stack is empty, since other 3580 // tasks might be pushing objects to it concurrently. 3581 assert(_cm->out_of_regions(), 3582 "at this point we should be out of regions"); 3583 // Try to reduce the number of available SATB buffers so that 3584 // remark has less work to do. 3585 drain_satb_buffers(); 3586 } 3587 3588 // Since we've done everything else, we can now totally drain the 3589 // local queue and global stack. 3590 drain_local_queue(false); 3591 drain_global_stack(false); 3592 3593 // Attempt at work stealing from other task's queues. 3594 if (do_stealing && !has_aborted()) { 3595 // We have not aborted. This means that we have finished all that 3596 // we could. Let's try to do some stealing... 3597 3598 // We cannot check whether the global stack is empty, since other 3599 // tasks might be pushing objects to it concurrently. 3600 assert(_cm->out_of_regions() && _task_queue->size() == 0, 3601 "only way to reach here"); 3602 while (!has_aborted()) { 3603 oop obj; 3604 if (_cm->try_stealing(_worker_id, &_hash_seed, obj)) { 3605 assert(_nextMarkBitMap->isMarked((HeapWord*) obj), 3606 "any stolen object should be marked"); 3607 scan_object(obj); 3608 3609 // And since we're towards the end, let's totally drain the 3610 // local queue and global stack. 3611 drain_local_queue(false); 3612 drain_global_stack(false); 3613 } else { 3614 break; 3615 } 3616 } 3617 } 3618 3619 // If we are about to wrap up and go into termination, check if we 3620 // should raise the overflow flag. 3621 if (do_termination && !has_aborted()) { 3622 if (_cm->force_overflow()->should_force()) { 3623 _cm->set_has_overflown(); 3624 regular_clock_call(); 3625 } 3626 } 3627 3628 // We still haven't aborted. Now, let's try to get into the 3629 // termination protocol. 3630 if (do_termination && !has_aborted()) { 3631 // We cannot check whether the global stack is empty, since other 3632 // tasks might be concurrently pushing objects on it. 3633 // Separated the asserts so that we know which one fires. 3634 assert(_cm->out_of_regions(), "only way to reach here"); 3635 assert(_task_queue->size() == 0, "only way to reach here"); 3636 _termination_start_time_ms = os::elapsedVTime() * 1000.0; 3637 3638 // The CMTask class also extends the TerminatorTerminator class, 3639 // hence its should_exit_termination() method will also decide 3640 // whether to exit the termination protocol or not. 3641 bool finished = (is_serial || 3642 _cm->terminator()->offer_termination(this)); 3643 double termination_end_time_ms = os::elapsedVTime() * 1000.0; 3644 _termination_time_ms += 3645 termination_end_time_ms - _termination_start_time_ms; 3646 3647 if (finished) { 3648 // We're all done. 3649 3650 if (_worker_id == 0) { 3651 // let's allow task 0 to do this 3652 if (concurrent()) { 3653 assert(_cm->concurrent_marking_in_progress(), "invariant"); 3654 // we need to set this to false before the next 3655 // safepoint. This way we ensure that the marking phase 3656 // doesn't observe any more heap expansions. 3657 _cm->clear_concurrent_marking_in_progress(); 3658 } 3659 } 3660 3661 // We can now guarantee that the global stack is empty, since 3662 // all other tasks have finished. We separated the guarantees so 3663 // that, if a condition is false, we can immediately find out 3664 // which one. 3665 guarantee(_cm->out_of_regions(), "only way to reach here"); 3666 guarantee(_cm->mark_stack_empty(), "only way to reach here"); 3667 guarantee(_task_queue->size() == 0, "only way to reach here"); 3668 guarantee(!_cm->has_overflown(), "only way to reach here"); 3669 guarantee(!_cm->mark_stack_overflow(), "only way to reach here"); 3670 } else { 3671 // Apparently there's more work to do. Let's abort this task. It 3672 // will restart it and we can hopefully find more things to do. 3673 set_has_aborted(); 3674 } 3675 } 3676 3677 // Mainly for debugging purposes to make sure that a pointer to the 3678 // closure which was statically allocated in this frame doesn't 3679 // escape it by accident. 3680 set_cm_oop_closure(NULL); 3681 double end_time_ms = os::elapsedVTime() * 1000.0; 3682 double elapsed_time_ms = end_time_ms - _start_time_ms; 3683 // Update the step history. 3684 _step_times_ms.add(elapsed_time_ms); 3685 3686 if (has_aborted()) { 3687 // The task was aborted for some reason. 3688 if (_has_timed_out) { 3689 double diff_ms = elapsed_time_ms - _time_target_ms; 3690 // Keep statistics of how well we did with respect to hitting 3691 // our target only if we actually timed out (if we aborted for 3692 // other reasons, then the results might get skewed). 3693 _marking_step_diffs_ms.add(diff_ms); 3694 } 3695 3696 if (_cm->has_overflown()) { 3697 // This is the interesting one. We aborted because a global 3698 // overflow was raised. This means we have to restart the 3699 // marking phase and start iterating over regions. However, in 3700 // order to do this we have to make sure that all tasks stop 3701 // what they are doing and re-initialize in a safe manner. We 3702 // will achieve this with the use of two barrier sync points. 3703 3704 if (!is_serial) { 3705 // We only need to enter the sync barrier if being called 3706 // from a parallel context 3707 _cm->enter_first_sync_barrier(_worker_id); 3708 3709 // When we exit this sync barrier we know that all tasks have 3710 // stopped doing marking work. So, it's now safe to 3711 // re-initialize our data structures. At the end of this method, 3712 // task 0 will clear the global data structures. 3713 } 3714 3715 // We clear the local state of this task... 3716 clear_region_fields(); 3717 3718 if (!is_serial) { 3719 // ...and enter the second barrier. 3720 _cm->enter_second_sync_barrier(_worker_id); 3721 } 3722 // At this point, if we're during the concurrent phase of 3723 // marking, everything has been re-initialized and we're 3724 // ready to restart. 3725 } 3726 } 3727 3728 _claimed = false; 3729 } 3730 3731 CMTask::CMTask(uint worker_id, 3732 ConcurrentMark* cm, 3733 size_t* marked_bytes, 3734 BitMap* card_bm, 3735 CMTaskQueue* task_queue, 3736 CMTaskQueueSet* task_queues) 3737 : _g1h(G1CollectedHeap::heap()), 3738 _worker_id(worker_id), _cm(cm), 3739 _claimed(false), 3740 _nextMarkBitMap(NULL), _hash_seed(17), 3741 _task_queue(task_queue), 3742 _task_queues(task_queues), 3743 _cm_oop_closure(NULL), 3744 _marked_bytes_array(marked_bytes), 3745 _card_bm(card_bm) { 3746 guarantee(task_queue != NULL, "invariant"); 3747 guarantee(task_queues != NULL, "invariant"); 3748 3749 _marking_step_diffs_ms.add(0.5); 3750 } 3751 3752 // These are formatting macros that are used below to ensure 3753 // consistent formatting. The *_H_* versions are used to format the 3754 // header for a particular value and they should be kept consistent 3755 // with the corresponding macro. Also note that most of the macros add 3756 // the necessary white space (as a prefix) which makes them a bit 3757 // easier to compose. 3758 3759 // All the output lines are prefixed with this string to be able to 3760 // identify them easily in a large log file. 3761 #define G1PPRL_LINE_PREFIX "###" 3762 3763 #define G1PPRL_ADDR_BASE_FORMAT " " PTR_FORMAT "-" PTR_FORMAT 3764 #ifdef _LP64 3765 #define G1PPRL_ADDR_BASE_H_FORMAT " %37s" 3766 #else // _LP64 3767 #define G1PPRL_ADDR_BASE_H_FORMAT " %21s" 3768 #endif // _LP64 3769 3770 // For per-region info 3771 #define G1PPRL_TYPE_FORMAT " %-4s" 3772 #define G1PPRL_TYPE_H_FORMAT " %4s" 3773 #define G1PPRL_BYTE_FORMAT " " SIZE_FORMAT_W(9) 3774 #define G1PPRL_BYTE_H_FORMAT " %9s" 3775 #define G1PPRL_DOUBLE_FORMAT " %14.1f" 3776 #define G1PPRL_DOUBLE_H_FORMAT " %14s" 3777 3778 // For summary info 3779 #define G1PPRL_SUM_ADDR_FORMAT(tag) " " tag ":" G1PPRL_ADDR_BASE_FORMAT 3780 #define G1PPRL_SUM_BYTE_FORMAT(tag) " " tag ": " SIZE_FORMAT 3781 #define G1PPRL_SUM_MB_FORMAT(tag) " " tag ": %1.2f MB" 3782 #define G1PPRL_SUM_MB_PERC_FORMAT(tag) G1PPRL_SUM_MB_FORMAT(tag) " / %1.2f %%" 3783 3784 G1PrintRegionLivenessInfoClosure:: 3785 G1PrintRegionLivenessInfoClosure(outputStream* out, const char* phase_name) 3786 : _out(out), 3787 _total_used_bytes(0), _total_capacity_bytes(0), 3788 _total_prev_live_bytes(0), _total_next_live_bytes(0), 3789 _hum_used_bytes(0), _hum_capacity_bytes(0), 3790 _hum_prev_live_bytes(0), _hum_next_live_bytes(0), 3791 _total_remset_bytes(0), _total_strong_code_roots_bytes(0) { 3792 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 3793 MemRegion g1_reserved = g1h->g1_reserved(); 3794 double now = os::elapsedTime(); 3795 3796 // Print the header of the output. 3797 _out->cr(); 3798 _out->print_cr(G1PPRL_LINE_PREFIX" PHASE %s @ %1.3f", phase_name, now); 3799 _out->print_cr(G1PPRL_LINE_PREFIX" HEAP" 3800 G1PPRL_SUM_ADDR_FORMAT("reserved") 3801 G1PPRL_SUM_BYTE_FORMAT("region-size"), 3802 p2i(g1_reserved.start()), p2i(g1_reserved.end()), 3803 HeapRegion::GrainBytes); 3804 _out->print_cr(G1PPRL_LINE_PREFIX); 3805 _out->print_cr(G1PPRL_LINE_PREFIX 3806 G1PPRL_TYPE_H_FORMAT 3807 G1PPRL_ADDR_BASE_H_FORMAT 3808 G1PPRL_BYTE_H_FORMAT 3809 G1PPRL_BYTE_H_FORMAT 3810 G1PPRL_BYTE_H_FORMAT 3811 G1PPRL_DOUBLE_H_FORMAT 3812 G1PPRL_BYTE_H_FORMAT 3813 G1PPRL_BYTE_H_FORMAT, 3814 "type", "address-range", 3815 "used", "prev-live", "next-live", "gc-eff", 3816 "remset", "code-roots"); 3817 _out->print_cr(G1PPRL_LINE_PREFIX 3818 G1PPRL_TYPE_H_FORMAT 3819 G1PPRL_ADDR_BASE_H_FORMAT 3820 G1PPRL_BYTE_H_FORMAT 3821 G1PPRL_BYTE_H_FORMAT 3822 G1PPRL_BYTE_H_FORMAT 3823 G1PPRL_DOUBLE_H_FORMAT 3824 G1PPRL_BYTE_H_FORMAT 3825 G1PPRL_BYTE_H_FORMAT, 3826 "", "", 3827 "(bytes)", "(bytes)", "(bytes)", "(bytes/ms)", 3828 "(bytes)", "(bytes)"); 3829 } 3830 3831 // It takes as a parameter a reference to one of the _hum_* fields, it 3832 // deduces the corresponding value for a region in a humongous region 3833 // series (either the region size, or what's left if the _hum_* field 3834 // is < the region size), and updates the _hum_* field accordingly. 3835 size_t G1PrintRegionLivenessInfoClosure::get_hum_bytes(size_t* hum_bytes) { 3836 size_t bytes = 0; 3837 // The > 0 check is to deal with the prev and next live bytes which 3838 // could be 0. 3839 if (*hum_bytes > 0) { 3840 bytes = MIN2(HeapRegion::GrainBytes, *hum_bytes); 3841 *hum_bytes -= bytes; 3842 } 3843 return bytes; 3844 } 3845 3846 // It deduces the values for a region in a humongous region series 3847 // from the _hum_* fields and updates those accordingly. It assumes 3848 // that that _hum_* fields have already been set up from the "starts 3849 // humongous" region and we visit the regions in address order. 3850 void G1PrintRegionLivenessInfoClosure::get_hum_bytes(size_t* used_bytes, 3851 size_t* capacity_bytes, 3852 size_t* prev_live_bytes, 3853 size_t* next_live_bytes) { 3854 assert(_hum_used_bytes > 0 && _hum_capacity_bytes > 0, "pre-condition"); 3855 *used_bytes = get_hum_bytes(&_hum_used_bytes); 3856 *capacity_bytes = get_hum_bytes(&_hum_capacity_bytes); 3857 *prev_live_bytes = get_hum_bytes(&_hum_prev_live_bytes); 3858 *next_live_bytes = get_hum_bytes(&_hum_next_live_bytes); 3859 } 3860 3861 bool G1PrintRegionLivenessInfoClosure::doHeapRegion(HeapRegion* r) { 3862 const char* type = r->get_type_str(); 3863 HeapWord* bottom = r->bottom(); 3864 HeapWord* end = r->end(); 3865 size_t capacity_bytes = r->capacity(); 3866 size_t used_bytes = r->used(); 3867 size_t prev_live_bytes = r->live_bytes(); 3868 size_t next_live_bytes = r->next_live_bytes(); 3869 double gc_eff = r->gc_efficiency(); 3870 size_t remset_bytes = r->rem_set()->mem_size(); 3871 size_t strong_code_roots_bytes = r->rem_set()->strong_code_roots_mem_size(); 3872 3873 if (r->is_starts_humongous()) { 3874 assert(_hum_used_bytes == 0 && _hum_capacity_bytes == 0 && 3875 _hum_prev_live_bytes == 0 && _hum_next_live_bytes == 0, 3876 "they should have been zeroed after the last time we used them"); 3877 // Set up the _hum_* fields. 3878 _hum_capacity_bytes = capacity_bytes; 3879 _hum_used_bytes = used_bytes; 3880 _hum_prev_live_bytes = prev_live_bytes; 3881 _hum_next_live_bytes = next_live_bytes; 3882 get_hum_bytes(&used_bytes, &capacity_bytes, 3883 &prev_live_bytes, &next_live_bytes); 3884 end = bottom + HeapRegion::GrainWords; 3885 } else if (r->is_continues_humongous()) { 3886 get_hum_bytes(&used_bytes, &capacity_bytes, 3887 &prev_live_bytes, &next_live_bytes); 3888 assert(end == bottom + HeapRegion::GrainWords, "invariant"); 3889 } 3890 3891 _total_used_bytes += used_bytes; 3892 _total_capacity_bytes += capacity_bytes; 3893 _total_prev_live_bytes += prev_live_bytes; 3894 _total_next_live_bytes += next_live_bytes; 3895 _total_remset_bytes += remset_bytes; 3896 _total_strong_code_roots_bytes += strong_code_roots_bytes; 3897 3898 // Print a line for this particular region. 3899 _out->print_cr(G1PPRL_LINE_PREFIX 3900 G1PPRL_TYPE_FORMAT 3901 G1PPRL_ADDR_BASE_FORMAT 3902 G1PPRL_BYTE_FORMAT 3903 G1PPRL_BYTE_FORMAT 3904 G1PPRL_BYTE_FORMAT 3905 G1PPRL_DOUBLE_FORMAT 3906 G1PPRL_BYTE_FORMAT 3907 G1PPRL_BYTE_FORMAT, 3908 type, p2i(bottom), p2i(end), 3909 used_bytes, prev_live_bytes, next_live_bytes, gc_eff, 3910 remset_bytes, strong_code_roots_bytes); 3911 3912 return false; 3913 } 3914 3915 G1PrintRegionLivenessInfoClosure::~G1PrintRegionLivenessInfoClosure() { 3916 // add static memory usages to remembered set sizes 3917 _total_remset_bytes += HeapRegionRemSet::fl_mem_size() + HeapRegionRemSet::static_mem_size(); 3918 // Print the footer of the output. 3919 _out->print_cr(G1PPRL_LINE_PREFIX); 3920 _out->print_cr(G1PPRL_LINE_PREFIX 3921 " SUMMARY" 3922 G1PPRL_SUM_MB_FORMAT("capacity") 3923 G1PPRL_SUM_MB_PERC_FORMAT("used") 3924 G1PPRL_SUM_MB_PERC_FORMAT("prev-live") 3925 G1PPRL_SUM_MB_PERC_FORMAT("next-live") 3926 G1PPRL_SUM_MB_FORMAT("remset") 3927 G1PPRL_SUM_MB_FORMAT("code-roots"), 3928 bytes_to_mb(_total_capacity_bytes), 3929 bytes_to_mb(_total_used_bytes), 3930 perc(_total_used_bytes, _total_capacity_bytes), 3931 bytes_to_mb(_total_prev_live_bytes), 3932 perc(_total_prev_live_bytes, _total_capacity_bytes), 3933 bytes_to_mb(_total_next_live_bytes), 3934 perc(_total_next_live_bytes, _total_capacity_bytes), 3935 bytes_to_mb(_total_remset_bytes), 3936 bytes_to_mb(_total_strong_code_roots_bytes)); 3937 _out->cr(); 3938 }