1 /*
  2  * Copyright (c) 2017, 2026, 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 "classfile/classLoaderDataGraph.hpp"
 26 #include "cppstdlib/new.hpp"
 27 #include "gc/g1/g1CollectedHeap.hpp"
 28 #include "gc/g1/g1FullCollector.inline.hpp"
 29 #include "gc/g1/g1FullGCAdjustTask.hpp"
 30 #include "gc/g1/g1FullGCCompactTask.hpp"
 31 #include "gc/g1/g1FullGCMarker.inline.hpp"
 32 #include "gc/g1/g1FullGCMarkTask.hpp"
 33 #include "gc/g1/g1FullGCPrepareTask.inline.hpp"
 34 #include "gc/g1/g1FullGCResetMetadataTask.hpp"
 35 #include "gc/g1/g1FullGCScope.hpp"
 36 #include "gc/g1/g1OopClosures.hpp"
 37 #include "gc/g1/g1Policy.hpp"
 38 #include "gc/g1/g1RegionMarkStatsCache.inline.hpp"
 39 #include "gc/shared/classUnloadingContext.hpp"
 40 #include "gc/shared/gcTraceTime.inline.hpp"
 41 #include "gc/shared/preservedMarks.inline.hpp"
 42 #include "gc/shared/referenceProcessor.hpp"
 43 #include "gc/shared/verifyOption.hpp"
 44 #include "gc/shared/weakProcessor.inline.hpp"
 45 #include "gc/shared/workerPolicy.hpp"
 46 #include "logging/log.hpp"
 47 #include "runtime/handles.inline.hpp"
 48 #include "utilities/debug.hpp"
 49 
 50 static void clear_and_activate_derived_pointers() {
 51 #if COMPILER2_OR_JVMCI
 52   DerivedPointerTable::clear();
 53 #endif
 54 }
 55 
 56 static void deactivate_derived_pointers() {
 57 #if COMPILER2_OR_JVMCI
 58   DerivedPointerTable::set_active(false);
 59 #endif
 60 }
 61 
 62 static void update_derived_pointers() {
 63 #if COMPILER2_OR_JVMCI
 64   DerivedPointerTable::update_pointers();
 65 #endif
 66 }
 67 
 68 G1CMBitMap* G1FullCollector::mark_bitmap() {
 69   return _heap->concurrent_mark()->mark_bitmap();
 70 }
 71 
 72 ReferenceProcessor* G1FullCollector::reference_processor() {
 73   return _heap->ref_processor_stw();
 74 }
 75 
 76 uint G1FullCollector::calc_active_workers() {
 77   G1CollectedHeap* heap = G1CollectedHeap::heap();
 78   uint max_worker_count = heap->workers()->max_workers();
 79   // Only calculate number of workers if UseDynamicNumberOfGCThreads
 80   // is enabled, otherwise use max.
 81   if (!UseDynamicNumberOfGCThreads) {
 82     return max_worker_count;
 83   }
 84 
 85   // Consider G1HeapWastePercent to decide max number of workers. Each worker
 86   // will in average cause half a region waste.
 87   uint max_wasted_regions_allowed = ((heap->num_committed_regions() * G1HeapWastePercent) / 100);
 88   uint waste_worker_count = MAX2((max_wasted_regions_allowed * 2) , 1u);
 89   uint heap_waste_worker_limit = MIN2(waste_worker_count, max_worker_count);
 90 
 91   // Also consider HeapSizePerGCThread by calling WorkerPolicy to calculate
 92   // the number of workers.
 93   uint current_active_workers = heap->workers()->active_workers();
 94   uint active_worker_limit = WorkerPolicy::calc_active_workers(max_worker_count, current_active_workers, 0);
 95 
 96   // Finally consider the amount of used regions.
 97   uint used_worker_limit = heap->num_used_regions();
 98   assert(used_worker_limit > 0, "Should never have zero used regions.");
 99 
100   // Update active workers to the lower of the limits.
101   uint worker_count = MIN3(heap_waste_worker_limit, active_worker_limit, used_worker_limit);
102   log_debug(gc, task)("Requesting %u active workers for full compaction (waste limited workers: %u, "
103                       "adaptive workers: %u, used limited workers: %u)",
104                       worker_count, heap_waste_worker_limit, active_worker_limit, used_worker_limit);
105   worker_count = heap->workers()->set_active_workers(worker_count);
106   log_info(gc, task)("Using %u workers of %u for full compaction", worker_count, max_worker_count);
107 
108   return worker_count;
109 }
110 
111 G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
112                                  bool clear_soft_refs,
113                                  bool do_maximal_compaction,
114                                  GCTracer* tracer) :
115     _heap(heap),
116     _scope(heap->monitoring_support(), clear_soft_refs, do_maximal_compaction, tracer),
117     _num_workers(calc_active_workers()),
118     _has_compaction_targets(false),
119     _has_humongous(false),
120     _marking_task_queues(_num_workers),
121     _partial_array_state_manager(nullptr),
122     _preserved_marks_set(true),
123     _serial_compaction_point(this, nullptr),
124     _humongous_compaction_point(this, nullptr),
125     _is_alive(this, heap->concurrent_mark()->mark_bitmap()),
126     _is_alive_mutator(heap->ref_processor_stw(), &_is_alive),
127     _humongous_compaction_regions(8),
128     _always_subject_to_discovery(),
129     _is_subject_mutator(heap->ref_processor_stw(), &_always_subject_to_discovery),
130     _region_attr_table() {
131   assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint");
132 
133   _preserved_marks_set.init(_num_workers);
134   _markers = NEW_C_HEAP_ARRAY(G1FullGCMarker*, _num_workers, mtGC);
135   _compaction_points = NEW_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _num_workers, mtGC);
136 
137   _live_stats = NEW_C_HEAP_ARRAY(G1RegionMarkStats, _heap->max_num_regions(), mtGC);
138   for (uint j = 0; j < heap->max_num_regions(); j++) {
139     _live_stats[j].clear();
140   }
141 
142   _compaction_tops = NEW_C_HEAP_ARRAY(Atomic<HeapWord*>, _heap->max_num_regions(), mtGC);
143   ::new (_compaction_tops) Atomic<HeapWord*>[heap->max_num_regions()]{};
144 
145   _partial_array_state_manager = new PartialArrayStateManager(_num_workers);
146 
147   for (uint i = 0; i < _num_workers; i++) {
148     _markers[i] = new G1FullGCMarker(this, i, _live_stats);
149     _compaction_points[i] = new G1FullGCCompactionPoint(this, _preserved_marks_set.get(i));
150     _marking_task_queues.register_queue(i, marker(i)->task_queue());
151   }
152 
153   _serial_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
154   _humongous_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
155   _region_attr_table.initialize(heap->reserved(), G1HeapRegion::GrainBytes);
156 }
157 
158 PartialArrayStateManager* G1FullCollector::partial_array_state_manager() const {
159   return _partial_array_state_manager;
160 }
161 
162 G1FullCollector::~G1FullCollector() {
163   for (uint i = 0; i < _num_workers; i++) {
164     delete _markers[i];
165     delete _compaction_points[i];
166   }
167 
168   delete _partial_array_state_manager;
169 
170   FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers);
171   FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points);
172   FREE_C_HEAP_ARRAY(Atomic<HeapWord*>, _compaction_tops);
173   FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats);
174 }
175 
176 class PrepareRegionsClosure : public G1HeapRegionClosure {
177   G1FullCollector* _collector;
178 
179 public:
180   PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { }
181 
182   bool do_heap_region(G1HeapRegion* hr) {
183     hr->prepare_for_full_gc();
184     G1CollectedHeap::heap()->prepare_region_for_full_compaction(hr);
185     _collector->before_marking_update_attribute_table(hr);
186     return false;
187   }
188 };
189 
190 void G1FullCollector::prepare_collection() {
191   _heap->policy()->record_full_collection_start();
192 
193   // Verification needs the bitmap, so we should clear the bitmap only later.
194   bool in_concurrent_cycle = _heap->abort_concurrent_cycle();
195   _heap->verify_before_full_collection();
196   if (in_concurrent_cycle) {
197     GCTraceTime(Debug, gc) debug("Clear Bitmap");
198     _heap->concurrent_mark()->clear_bitmap(_heap->workers());
199   }
200 
201   _heap->gc_prologue(true);
202   _heap->retire_tlabs();
203   _heap->flush_region_pin_cache();
204   _heap->prepare_heap_for_full_collection();
205 
206   PrepareRegionsClosure cl(this);
207   _heap->heap_region_iterate(&cl);
208 
209   reference_processor()->start_discovery(scope()->should_clear_soft_refs());
210 
211   // Clear and activate derived pointer collection.
212   clear_and_activate_derived_pointers();
213 }
214 
215 void G1FullCollector::collect() {
216   G1CollectedHeap::start_codecache_marking_cycle_if_inactive(false /* concurrent_mark_start */);
217 
218   phase1_mark_live_objects();
219   verify_after_marking();
220 
221   // Don't add any more derived pointers during later phases
222   deactivate_derived_pointers();
223 
224   phase2_prepare_compaction();
225 
226   if (has_compaction_targets()) {
227     phase3_adjust_pointers();
228 
229     phase4_do_compaction();
230   } else {
231     // All regions have a high live ratio thus will not be compacted.
232     // The live ratio is only considered if do_maximal_compaction is false.
233     log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap");
234   }
235 
236   phase5_reset_metadata();
237 }
238 
239 void G1FullCollector::complete_collection(size_t allocation_word_size) {
240   // Restore all marks.
241   restore_marks();
242 
243   // When the pointers have been adjusted and moved, we can
244   // update the derived pointer table.
245   update_derived_pointers();
246 
247   // Need completely cleared claim bits for the next concurrent marking or full gc.
248   ClassLoaderDataGraph::clear_claimed_marks();
249 
250   // Prepare the bitmap for the next (potentially concurrent) marking.
251   _heap->concurrent_mark()->clear_bitmap(_heap->workers());
252 
253   _heap->prepare_for_mutator_after_full_collection(allocation_word_size);
254 
255   _heap->resize_all_tlabs();
256 
257   _heap->policy()->record_full_collection_end(allocation_word_size);
258   _heap->gc_epilogue(true);
259 
260   _heap->verify_after_full_collection();
261 
262   _heap->print_heap_after_full_collection();
263 }
264 
265 void G1FullCollector::before_marking_update_attribute_table(G1HeapRegion* hr) {
266   if (hr->is_free()) {
267     _region_attr_table.set_free(hr->hrm_index());
268   } else if (hr->is_humongous() || hr->has_pinned_objects()) {
269     // Humongous objects or pinned regions will never be moved in the "main"
270     // compaction phase, but non-pinned regions might afterwards in a special phase.
271     _region_attr_table.set_skip_compacting(hr->hrm_index());
272   } else {
273     // Everything else should be compacted.
274     _region_attr_table.set_compacting(hr->hrm_index());
275   }
276 }
277 
278 class G1FullGCRefProcProxyTask : public RefProcProxyTask {
279   G1FullCollector& _collector;
280 
281   // G1 Full GC specific closure for handling discovered fields. Do NOT need any
282   // barriers as Full GC discards all this information anyway.
283   class G1FullGCDiscoveredFieldClosure : public EnqueueDiscoveredFieldClosure {
284     G1CollectedHeap* _g1h;
285 
286   public:
287     G1FullGCDiscoveredFieldClosure() : _g1h(G1CollectedHeap::heap()) { }
288 
289     void enqueue(HeapWord* discovered_field_addr, oop value) override {
290       assert(_g1h->is_in(discovered_field_addr), PTR_FORMAT " is not in heap ", p2i(discovered_field_addr));
291       // Store the value and done.
292       RawAccess<>::oop_store(discovered_field_addr, value);
293     }
294   };
295 
296 public:
297   G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
298     : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
299       _collector(collector) {}
300 
301   void work(uint worker_id) override {
302     assert(worker_id < _max_workers, "sanity");
303     G1IsAliveClosure is_alive(&_collector);
304     uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
305     G1FullKeepAliveClosure keep_alive(_collector.marker(index));
306     G1FullGCDiscoveredFieldClosure enqueue;
307     G1MarkStackClosure* complete_marking = _collector.marker(index)->stack_closure();
308     _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_marking);
309   }
310 };
311 
312 void G1FullCollector::phase1_mark_live_objects() {
313   // Recursively traverse all live objects and mark them.
314   GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer());
315 
316   {
317     // Do the actual marking.
318     G1FullGCMarkTask marking_task(this);
319     run_task(&marking_task);
320   }
321 
322   {
323     GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer());
324     // Process reference objects found during marking.
325     ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues());
326     G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues());
327     const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, _heap->workers(), pt);
328     scope()->tracer()->report_gc_reference_stats(stats);
329     pt.print_all_references();
330     assert(marker(0)->task_queue()->is_empty(), "Should be no oops on the stack");
331   }
332 
333   {
334     GCTraceTime(Debug, gc, phases) debug("Phase 1: Flush Mark Stats Cache", scope()->timer());
335     for (uint i = 0; i < workers(); i++) {
336       marker(i)->flush_mark_stats_cache();
337     }
338   }
339 
340   // Weak oops cleanup.
341   {
342     GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer());
343     WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1);
344   }
345 
346   // Class unloading and cleanup.
347   if (ClassUnloading) {
348     _heap->unload_classes_and_code("Phase 1: Class Unloading and Cleanup", &_is_alive, scope()->timer());
349   }
350 
351   {
352     GCTraceTime(Debug, gc, phases) debug("Report Object Count", scope()->timer());
353     scope()->tracer()->report_object_count_after_gc(&_is_alive, _heap->workers());
354   }
355 #if TASKQUEUE_STATS
356   marking_task_queues()->print_and_reset_taskqueue_stats("Full GC");
357 
358   auto get_stats = [&](uint i) {
359     return marker(i)->partial_array_splitter().stats();
360   };
361   PartialArrayTaskStats::log_set(_num_workers, get_stats,
362                                  "Full GC Partial Array");
363 #endif
364 }
365 
366 void G1FullCollector::phase2_prepare_compaction() {
367   GCTraceTime(Info, gc, phases) info("Phase 2: Prepare compaction", scope()->timer());
368 
369   phase2a_determine_worklists();
370 
371   if (!has_compaction_targets()) {
372     return;
373   }
374 
375   bool has_free_compaction_targets = phase2b_forward_oops();
376 
377   // Try to avoid OOM immediately after Full GC in case there are no free regions
378   // left after determining the result locations (i.e. this phase). Prepare to
379   // maximally compact the tail regions of the compaction queues serially.
380   if (scope()->do_maximal_compaction() || !has_free_compaction_targets) {
381     phase2c_prepare_serial_compaction();
382 
383     if (scope()->do_maximal_compaction() &&
384         has_humongous() &&
385         serial_compaction_point()->has_regions()) {
386       phase2d_prepare_humongous_compaction();
387     }
388   }
389 }
390 
391 void G1FullCollector::phase2a_determine_worklists() {
392   GCTraceTime(Debug, gc, phases) debug("Phase 2: Determine work lists", scope()->timer());
393 
394   G1DetermineCompactionQueueClosure cl(this);
395   _heap->heap_region_iterate(&cl);
396 }
397 
398 bool G1FullCollector::phase2b_forward_oops() {
399   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare parallel compaction", scope()->timer());
400 
401   G1FullGCPrepareTask task(this);
402   run_task(&task);
403 
404   return task.has_free_compaction_targets();
405 }
406 
407 uint G1FullCollector::truncate_parallel_cps() {
408   uint lowest_current = UINT_MAX;
409   for (uint i = 0; i < workers(); i++) {
410     G1FullGCCompactionPoint* cp = compaction_point(i);
411     if (cp->has_regions()) {
412       lowest_current = MIN2(lowest_current, cp->current_region()->hrm_index());
413     }
414   }
415 
416   for (uint i = 0; i < workers(); i++) {
417     G1FullGCCompactionPoint* cp = compaction_point(i);
418     if (cp->has_regions()) {
419       cp->remove_at_or_above(lowest_current);
420     }
421   }
422   return lowest_current;
423 }
424 
425 void G1FullCollector::phase2c_prepare_serial_compaction() {
426   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer());
427   // At this point, we know that after parallel compaction there will be regions that
428   // are partially compacted into. Thus, the last compaction region of all
429   // compaction queues still have space in them. We try to re-compact these regions
430   // in serial to avoid a premature OOM when the mutator wants to allocate the first
431   // eden region after gc.
432 
433   // For maximum compaction, we need to re-prepare all objects above the lowest
434   // region among the current regions for all thread compaction points. It may
435   // happen that due to the uneven distribution of objects to parallel threads, holes
436   // have been created as threads compact to different target regions between the
437   // lowest and the highest region in the tails of the compaction points.
438 
439   uint start_serial = truncate_parallel_cps();
440   assert(start_serial < _heap->max_num_regions(), "Called on empty parallel compaction queues");
441 
442   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
443   assert(!serial_cp->is_initialized(), "sanity!");
444 
445   G1HeapRegion* start_hr = _heap->region_at(start_serial);
446   serial_cp->add(start_hr);
447   serial_cp->initialize(start_hr);
448 
449   HeapWord* dense_prefix_top = compaction_top(start_hr);
450   G1SerialRePrepareClosure re_prepare(serial_cp, dense_prefix_top);
451 
452   for (uint i = start_serial + 1; i < _heap->max_num_regions(); i++) {
453     if (is_compaction_target(i)) {
454       G1HeapRegion* current = _heap->region_at(i);
455       set_compaction_top(current, current->bottom());
456       serial_cp->add(current);
457       current->apply_to_marked_objects(mark_bitmap(), &re_prepare);
458     }
459   }
460   serial_cp->update();
461 }
462 
463 void G1FullCollector::phase2d_prepare_humongous_compaction() {
464   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare humongous compaction", scope()->timer());
465   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
466   assert(serial_cp->has_regions(), "Sanity!" );
467 
468   uint last_serial_target = serial_cp->current_region()->hrm_index();
469   uint region_index = last_serial_target + 1;
470   uint max_num_regions = _heap->max_num_regions();
471 
472   G1FullGCCompactionPoint* humongous_cp = humongous_compaction_point();
473 
474   while (region_index < max_num_regions) {
475     G1HeapRegion* hr = _heap->region_at_or_null(region_index);
476 
477     if (hr == nullptr) {
478       region_index++;
479       continue;
480     } else if (hr->is_starts_humongous()) {
481       size_t obj_size = cast_to_oop(hr->bottom())->size();
482       uint num_regions = (uint)G1CollectedHeap::humongous_obj_size_in_regions(obj_size);
483       // Even during last-ditch compaction we should not move pinned humongous objects.
484       if (!hr->has_pinned_objects()) {
485         humongous_cp->forward_humongous(hr);
486       }
487       region_index += num_regions; // Advance over all humongous regions.
488       continue;
489     } else if (is_compaction_target(region_index)) {
490       assert(!hr->has_pinned_objects(), "pinned regions should not be compaction targets");
491       // Add the region to the humongous compaction point.
492       humongous_cp->add(hr);
493     }
494     region_index++;
495   }
496 }
497 
498 void G1FullCollector::phase3_adjust_pointers() {
499   // Adjust the pointers to reflect the new locations
500   GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer());
501 
502   G1FullGCAdjustTask task(this);
503   run_task(&task);
504 }
505 
506 void G1FullCollector::phase4_do_compaction() {
507   // Compact the heap using the compaction queues created in phase 2.
508   GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer());
509   G1FullGCCompactTask task(this);
510   run_task(&task);
511 
512   // Serial compact to avoid OOM when very few free regions.
513   if (serial_compaction_point()->has_regions()) {
514     task.serial_compaction();
515   }
516 
517   if (!_humongous_compaction_regions.is_empty()) {
518     assert(scope()->do_maximal_compaction(), "Only compact humongous during maximal compaction");
519     task.humongous_compaction();
520   }
521 }
522 
523 void G1FullCollector::phase5_reset_metadata() {
524   // Clear region metadata that is invalid after GC for all regions.
525   GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer());
526   G1FullGCResetMetadataTask task(this);
527   run_task(&task);
528 }
529 
530 void G1FullCollector::restore_marks() {
531   _preserved_marks_set.restore(_heap->workers());
532   _preserved_marks_set.reclaim();
533 }
534 
535 void G1FullCollector::run_task(WorkerTask* task) {
536   _heap->workers()->run_task(task, _num_workers);
537 }
538 
539 void G1FullCollector::verify_after_marking() {
540   if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) {
541     // Only do verification if VerifyDuringGC and G1VerifyFull is set.
542     return;
543   }
544 
545 #if COMPILER2_OR_JVMCI
546   DerivedPointerTableDeactivate dpt_deact;
547 #endif
548   _heap->prepare_for_verify();
549   // Note: we can verify only the heap here. When an object is
550   // marked, the previous value of the mark word (including
551   // identity hash values, ages, etc) is preserved, and the mark
552   // word is set to markWord::marked_value - effectively removing
553   // any hash values from the mark word. These hash values are
554   // used when verifying the dictionaries and so removing them
555   // from the mark word can make verification of the dictionaries
556   // fail. At the end of the GC, the original mark word values
557   // (including hash values) are restored to the appropriate
558   // objects.
559   GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)");
560   _heap->verify(VerifyOption::G1UseFullMarking);
561 }