1 /* 2 * Copyright (c) 2017, 2023, 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/classLoaderDataGraph.hpp" 27 #include "classfile/systemDictionary.hpp" 28 #include "code/codeCache.hpp" 29 #include "compiler/oopMap.hpp" 30 #include "gc/g1/g1CollectedHeap.hpp" 31 #include "gc/g1/g1FullCollector.inline.hpp" 32 #include "gc/g1/g1FullGCAdjustTask.hpp" 33 #include "gc/g1/g1FullGCCompactTask.hpp" 34 #include "gc/g1/g1FullGCMarker.inline.hpp" 35 #include "gc/g1/g1FullGCMarkTask.hpp" 36 #include "gc/g1/g1FullGCPrepareTask.inline.hpp" 37 #include "gc/g1/g1FullGCResetMetadataTask.hpp" 38 #include "gc/g1/g1FullGCScope.hpp" 39 #include "gc/g1/g1OopClosures.hpp" 40 #include "gc/g1/g1Policy.hpp" 41 #include "gc/g1/g1RegionMarkStatsCache.inline.hpp" 42 #include "gc/shared/gcTraceTime.inline.hpp" 43 #include "gc/shared/preservedMarks.inline.hpp" 44 #include "gc/shared/referenceProcessor.hpp" 45 #include "gc/shared/verifyOption.hpp" 46 #include "gc/shared/weakProcessor.inline.hpp" 47 #include "gc/shared/workerPolicy.hpp" 48 #include "logging/log.hpp" 49 #include "runtime/handles.inline.hpp" 50 #include "utilities/debug.hpp" 51 52 static void clear_and_activate_derived_pointers() { 53 #if COMPILER2_OR_JVMCI 54 DerivedPointerTable::clear(); 55 #endif 56 } 57 58 static void deactivate_derived_pointers() { 59 #if COMPILER2_OR_JVMCI 60 DerivedPointerTable::set_active(false); 61 #endif 62 } 63 64 static void update_derived_pointers() { 65 #if COMPILER2_OR_JVMCI 66 DerivedPointerTable::update_pointers(); 67 #endif 68 } 69 70 G1CMBitMap* G1FullCollector::mark_bitmap() { 71 return _heap->concurrent_mark()->mark_bitmap(); 72 } 73 74 ReferenceProcessor* G1FullCollector::reference_processor() { 75 return _heap->ref_processor_stw(); 76 } 77 78 uint G1FullCollector::calc_active_workers() { 79 G1CollectedHeap* heap = G1CollectedHeap::heap(); 80 uint max_worker_count = heap->workers()->max_workers(); 81 // Only calculate number of workers if UseDynamicNumberOfGCThreads 82 // is enabled, otherwise use max. 83 if (!UseDynamicNumberOfGCThreads) { 84 return max_worker_count; 85 } 86 87 // Consider G1HeapWastePercent to decide max number of workers. Each worker 88 // will in average cause half a region waste. 89 uint max_wasted_regions_allowed = ((heap->num_regions() * G1HeapWastePercent) / 100); 90 uint waste_worker_count = MAX2((max_wasted_regions_allowed * 2) , 1u); 91 uint heap_waste_worker_limit = MIN2(waste_worker_count, max_worker_count); 92 93 // Also consider HeapSizePerGCThread by calling WorkerPolicy to calculate 94 // the number of workers. 95 uint current_active_workers = heap->workers()->active_workers(); 96 uint active_worker_limit = WorkerPolicy::calc_active_workers(max_worker_count, current_active_workers, 0); 97 98 // Finally consider the amount of used regions. 99 uint used_worker_limit = heap->num_used_regions(); 100 assert(used_worker_limit > 0, "Should never have zero used regions."); 101 102 // Update active workers to the lower of the limits. 103 uint worker_count = MIN3(heap_waste_worker_limit, active_worker_limit, used_worker_limit); 104 log_debug(gc, task)("Requesting %u active workers for full compaction (waste limited workers: %u, " 105 "adaptive workers: %u, used limited workers: %u)", 106 worker_count, heap_waste_worker_limit, active_worker_limit, used_worker_limit); 107 worker_count = heap->workers()->set_active_workers(worker_count); 108 log_info(gc, task)("Using %u workers of %u for full compaction", worker_count, max_worker_count); 109 110 return worker_count; 111 } 112 113 G1FullCollector::G1FullCollector(G1CollectedHeap* heap, 114 bool clear_soft_refs, 115 bool do_maximal_compaction, 116 G1FullGCTracer* tracer) : 117 _heap(heap), 118 _scope(heap->monitoring_support(), clear_soft_refs, do_maximal_compaction, tracer), 119 _num_workers(calc_active_workers()), 120 _has_compaction_targets(false), 121 _has_humongous(false), 122 _oop_queue_set(_num_workers), 123 _array_queue_set(_num_workers), 124 _preserved_marks_set(true), 125 _serial_compaction_point(this), 126 _humongous_compaction_point(this), 127 _is_alive(this, heap->concurrent_mark()->mark_bitmap()), 128 _is_alive_mutator(heap->ref_processor_stw(), &_is_alive), 129 _humongous_compaction_regions(8), 130 _always_subject_to_discovery(), 131 _is_subject_mutator(heap->ref_processor_stw(), &_always_subject_to_discovery), 132 _region_attr_table() { 133 assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint"); 134 135 _preserved_marks_set.init(_num_workers); 136 _markers = NEW_C_HEAP_ARRAY(G1FullGCMarker*, _num_workers, mtGC); 137 _compaction_points = NEW_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _num_workers, mtGC); 138 139 _live_stats = NEW_C_HEAP_ARRAY(G1RegionMarkStats, _heap->max_regions(), mtGC); 140 _compaction_tops = NEW_C_HEAP_ARRAY(HeapWord*, _heap->max_regions(), mtGC); 141 for (uint j = 0; j < heap->max_regions(); j++) { 142 _live_stats[j].clear(); 143 _compaction_tops[j] = nullptr; 144 } 145 146 for (uint i = 0; i < _num_workers; i++) { 147 _markers[i] = new G1FullGCMarker(this, i, _preserved_marks_set.get(i), _live_stats); 148 _compaction_points[i] = new G1FullGCCompactionPoint(this); 149 _oop_queue_set.register_queue(i, marker(i)->oop_stack()); 150 _array_queue_set.register_queue(i, marker(i)->objarray_stack()); 151 } 152 _region_attr_table.initialize(heap->reserved(), HeapRegion::GrainBytes); 153 } 154 155 G1FullCollector::~G1FullCollector() { 156 for (uint i = 0; i < _num_workers; i++) { 157 delete _markers[i]; 158 delete _compaction_points[i]; 159 } 160 161 FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers); 162 FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points); 163 FREE_C_HEAP_ARRAY(HeapWord*, _compaction_tops); 164 FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats); 165 } 166 167 class PrepareRegionsClosure : public HeapRegionClosure { 168 G1FullCollector* _collector; 169 170 public: 171 PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { } 172 173 bool do_heap_region(HeapRegion* hr) { 174 G1CollectedHeap::heap()->prepare_region_for_full_compaction(hr); 175 _collector->before_marking_update_attribute_table(hr); 176 return false; 177 } 178 }; 179 180 void G1FullCollector::prepare_collection() { 181 _heap->policy()->record_full_collection_start(); 182 183 // Verification needs the bitmap, so we should clear the bitmap only later. 184 bool in_concurrent_cycle = _heap->abort_concurrent_cycle(); 185 _heap->verify_before_full_collection(); 186 if (in_concurrent_cycle) { 187 GCTraceTime(Debug, gc) debug("Clear Bitmap"); 188 _heap->concurrent_mark()->clear_bitmap(_heap->workers()); 189 } 190 191 _heap->gc_prologue(true); 192 _heap->retire_tlabs(); 193 _heap->prepare_heap_for_full_collection(); 194 195 PrepareRegionsClosure cl(this); 196 _heap->heap_region_iterate(&cl); 197 198 reference_processor()->start_discovery(scope()->should_clear_soft_refs()); 199 200 // Clear and activate derived pointer collection. 201 clear_and_activate_derived_pointers(); 202 } 203 204 void G1FullCollector::collect() { 205 G1CollectedHeap::start_codecache_marking_cycle_if_inactive(false /* concurrent_mark_start */); 206 207 phase1_mark_live_objects(); 208 verify_after_marking(); 209 210 // Don't add any more derived pointers during later phases 211 deactivate_derived_pointers(); 212 213 phase2_prepare_compaction(); 214 215 if (has_compaction_targets()) { 216 phase3_adjust_pointers(); 217 218 phase4_do_compaction(); 219 } else { 220 // All regions have a high live ratio thus will not be compacted. 221 // The live ratio is only considered if do_maximal_compaction is false. 222 log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap"); 223 } 224 225 phase5_reset_metadata(); 226 227 G1CollectedHeap::finish_codecache_marking_cycle(); 228 } 229 230 void G1FullCollector::complete_collection() { 231 // Restore all marks. 232 restore_marks(); 233 234 // When the pointers have been adjusted and moved, we can 235 // update the derived pointer table. 236 update_derived_pointers(); 237 238 // Need completely cleared claim bits for the next concurrent marking or full gc. 239 ClassLoaderDataGraph::clear_claimed_marks(); 240 241 // Prepare the bitmap for the next (potentially concurrent) marking. 242 _heap->concurrent_mark()->clear_bitmap(_heap->workers()); 243 244 _heap->prepare_for_mutator_after_full_collection(); 245 246 _heap->resize_all_tlabs(); 247 248 _heap->policy()->record_full_collection_end(); 249 _heap->gc_epilogue(true); 250 251 _heap->verify_after_full_collection(); 252 253 _heap->print_heap_after_full_collection(); 254 } 255 256 void G1FullCollector::before_marking_update_attribute_table(HeapRegion* hr) { 257 if (hr->is_free()) { 258 _region_attr_table.set_free(hr->hrm_index()); 259 } else if (hr->is_humongous()) { 260 // Humongous objects will never be moved in the "main" compaction phase, but 261 // afterwards in a special phase if needed. 262 _region_attr_table.set_skip_compacting(hr->hrm_index()); 263 } else { 264 // Everything else should be compacted. 265 _region_attr_table.set_compacting(hr->hrm_index()); 266 } 267 } 268 269 class G1FullGCRefProcProxyTask : public RefProcProxyTask { 270 G1FullCollector& _collector; 271 272 public: 273 G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers) 274 : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers), 275 _collector(collector) {} 276 277 void work(uint worker_id) override { 278 assert(worker_id < _max_workers, "sanity"); 279 G1IsAliveClosure is_alive(&_collector); 280 uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id; 281 G1FullKeepAliveClosure keep_alive(_collector.marker(index)); 282 BarrierEnqueueDiscoveredFieldClosure enqueue; 283 G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure(); 284 _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc); 285 } 286 }; 287 288 void G1FullCollector::phase1_mark_live_objects() { 289 // Recursively traverse all live objects and mark them. 290 GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer()); 291 292 { 293 // Do the actual marking. 294 G1FullGCMarkTask marking_task(this); 295 run_task(&marking_task); 296 } 297 298 { 299 uint old_active_mt_degree = reference_processor()->num_queues(); 300 reference_processor()->set_active_mt_degree(workers()); 301 GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer()); 302 // Process reference objects found during marking. 303 ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues()); 304 G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues()); 305 const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, pt); 306 scope()->tracer()->report_gc_reference_stats(stats); 307 pt.print_all_references(); 308 assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack"); 309 310 reference_processor()->set_active_mt_degree(old_active_mt_degree); 311 } 312 313 // Weak oops cleanup. 314 { 315 GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer()); 316 WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1); 317 } 318 319 // Class unloading and cleanup. 320 if (ClassUnloading) { 321 GCTraceTime(Debug, gc, phases) debug("Phase 1: Class Unloading and Cleanup", scope()->timer()); 322 CodeCache::UnloadingScope unloading_scope(&_is_alive); 323 // Unload classes and purge the SystemDictionary. 324 bool purged_class = SystemDictionary::do_unloading(scope()->timer()); 325 _heap->complete_cleaning(purged_class); 326 } 327 328 { 329 GCTraceTime(Debug, gc, phases) debug("Report Object Count", scope()->timer()); 330 scope()->tracer()->report_object_count_after_gc(&_is_alive, _heap->workers()); 331 } 332 #if TASKQUEUE_STATS 333 oop_queue_set()->print_and_reset_taskqueue_stats("Oop Queue"); 334 array_queue_set()->print_and_reset_taskqueue_stats("ObjArrayOop Queue"); 335 #endif 336 } 337 338 void G1FullCollector::phase2_prepare_compaction() { 339 GCTraceTime(Info, gc, phases) info("Phase 2: Prepare compaction", scope()->timer()); 340 341 phase2a_determine_worklists(); 342 343 if (!has_compaction_targets()) { 344 return; 345 } 346 347 bool has_free_compaction_targets = phase2b_forward_oops(); 348 349 // Try to avoid OOM immediately after Full GC in case there are no free regions 350 // left after determining the result locations (i.e. this phase). Prepare to 351 // maximally compact the tail regions of the compaction queues serially. 352 if (scope()->do_maximal_compaction() || !has_free_compaction_targets) { 353 phase2c_prepare_serial_compaction(); 354 355 if (scope()->do_maximal_compaction() && 356 has_humongous() && 357 serial_compaction_point()->has_regions()) { 358 phase2d_prepare_humongous_compaction(); 359 } 360 } 361 } 362 363 void G1FullCollector::phase2a_determine_worklists() { 364 GCTraceTime(Debug, gc, phases) debug("Phase 2: Determine work lists", scope()->timer()); 365 366 G1DetermineCompactionQueueClosure cl(this); 367 _heap->heap_region_iterate(&cl); 368 } 369 370 bool G1FullCollector::phase2b_forward_oops() { 371 GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare parallel compaction", scope()->timer()); 372 373 G1FullGCPrepareTask task(this); 374 run_task(&task); 375 376 return task.has_free_compaction_targets(); 377 } 378 379 uint G1FullCollector::truncate_parallel_cps() { 380 uint lowest_current = UINT_MAX; 381 for (uint i = 0; i < workers(); i++) { 382 G1FullGCCompactionPoint* cp = compaction_point(i); 383 if (cp->has_regions()) { 384 lowest_current = MIN2(lowest_current, cp->current_region()->hrm_index()); 385 } 386 } 387 388 for (uint i = 0; i < workers(); i++) { 389 G1FullGCCompactionPoint* cp = compaction_point(i); 390 if (cp->has_regions()) { 391 cp->remove_at_or_above(lowest_current); 392 } 393 } 394 return lowest_current; 395 } 396 397 void G1FullCollector::phase2c_prepare_serial_compaction() { 398 GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer()); 399 // At this point, we know that after parallel compaction there will be regions that 400 // are partially compacted into. Thus, the last compaction region of all 401 // compaction queues still have space in them. We try to re-compact these regions 402 // in serial to avoid a premature OOM when the mutator wants to allocate the first 403 // eden region after gc. 404 405 // For maximum compaction, we need to re-prepare all objects above the lowest 406 // region among the current regions for all thread compaction points. It may 407 // happen that due to the uneven distribution of objects to parallel threads, holes 408 // have been created as threads compact to different target regions between the 409 // lowest and the highest region in the tails of the compaction points. 410 411 uint start_serial = truncate_parallel_cps(); 412 assert(start_serial < _heap->max_reserved_regions(), "Called on empty parallel compaction queues"); 413 414 G1FullGCCompactionPoint* serial_cp = serial_compaction_point(); 415 assert(!serial_cp->is_initialized(), "sanity!"); 416 417 HeapRegion* start_hr = _heap->region_at(start_serial); 418 serial_cp->add(start_hr); 419 serial_cp->initialize(start_hr); 420 421 HeapWord* dense_prefix_top = compaction_top(start_hr); 422 G1SerialRePrepareClosure re_prepare(serial_cp, dense_prefix_top); 423 424 for (uint i = start_serial + 1; i < _heap->max_reserved_regions(); i++) { 425 if (is_compaction_target(i)) { 426 HeapRegion* current = _heap->region_at(i); 427 set_compaction_top(current, current->bottom()); 428 serial_cp->add(current); 429 current->apply_to_marked_objects(mark_bitmap(), &re_prepare); 430 } 431 } 432 serial_cp->update(); 433 } 434 435 void G1FullCollector::phase2d_prepare_humongous_compaction() { 436 GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare humongous compaction", scope()->timer()); 437 G1FullGCCompactionPoint* serial_cp = serial_compaction_point(); 438 assert(serial_cp->has_regions(), "Sanity!" ); 439 440 uint last_serial_target = serial_cp->current_region()->hrm_index(); 441 uint region_index = last_serial_target + 1; 442 uint max_reserved_regions = _heap->max_reserved_regions(); 443 444 G1FullGCCompactionPoint* humongous_cp = humongous_compaction_point(); 445 446 while (region_index < max_reserved_regions) { 447 HeapRegion* hr = _heap->region_at_or_null(region_index); 448 449 if (hr == nullptr) { 450 region_index++; 451 continue; 452 } else if (hr->is_starts_humongous()) { 453 uint num_regions = humongous_cp->forward_humongous(hr); 454 region_index += num_regions; // Skip over the continues humongous regions. 455 continue; 456 } else if (is_compaction_target(region_index)) { 457 // Add the region to the humongous compaction point. 458 humongous_cp->add(hr); 459 } 460 region_index++; 461 } 462 } 463 464 void G1FullCollector::phase3_adjust_pointers() { 465 // Adjust the pointers to reflect the new locations 466 GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer()); 467 468 G1FullGCAdjustTask task(this); 469 run_task(&task); 470 } 471 472 void G1FullCollector::phase4_do_compaction() { 473 // Compact the heap using the compaction queues created in phase 2. 474 GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer()); 475 G1FullGCCompactTask task(this); 476 run_task(&task); 477 478 // Serial compact to avoid OOM when very few free regions. 479 if (serial_compaction_point()->has_regions()) { 480 task.serial_compaction(); 481 } 482 483 if (!_humongous_compaction_regions.is_empty()) { 484 assert(scope()->do_maximal_compaction(), "Only compact humongous during maximal compaction"); 485 task.humongous_compaction(); 486 } 487 } 488 489 void G1FullCollector::phase5_reset_metadata() { 490 // Clear region metadata that is invalid after GC for all regions. 491 GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer()); 492 G1FullGCResetMetadataTask task(this); 493 run_task(&task); 494 } 495 496 void G1FullCollector::restore_marks() { 497 _preserved_marks_set.restore(_heap->workers()); 498 _preserved_marks_set.reclaim(); 499 } 500 501 void G1FullCollector::run_task(WorkerTask* task) { 502 _heap->workers()->run_task(task, _num_workers); 503 } 504 505 void G1FullCollector::verify_after_marking() { 506 if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) { 507 // Only do verification if VerifyDuringGC and G1VerifyFull is set. 508 return; 509 } 510 511 #if COMPILER2_OR_JVMCI 512 DerivedPointerTableDeactivate dpt_deact; 513 #endif 514 _heap->prepare_for_verify(); 515 // Note: we can verify only the heap here. When an object is 516 // marked, the previous value of the mark word (including 517 // identity hash values, ages, etc) is preserved, and the mark 518 // word is set to markWord::marked_value - effectively removing 519 // any hash values from the mark word. These hash values are 520 // used when verifying the dictionaries and so removing them 521 // from the mark word can make verification of the dictionaries 522 // fail. At the end of the GC, the original mark word values 523 // (including hash values) are restored to the appropriate 524 // objects. 525 GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)"); 526 _heap->verify(VerifyOption::G1UseFullMarking); 527 }