1 /*
  2  * Copyright (c) 2017, 2025, 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 "gc/g1/g1CollectedHeap.hpp"
 27 #include "gc/g1/g1FullCollector.inline.hpp"
 28 #include "gc/g1/g1FullGCAdjustTask.hpp"
 29 #include "gc/g1/g1FullGCCompactTask.hpp"
 30 #include "gc/g1/g1FullGCMarker.inline.hpp"
 31 #include "gc/g1/g1FullGCMarkTask.hpp"
 32 #include "gc/g1/g1FullGCPrepareTask.inline.hpp"
 33 #include "gc/g1/g1FullGCResetMetadataTask.hpp"
 34 #include "gc/g1/g1FullGCScope.hpp"
 35 #include "gc/g1/g1OopClosures.hpp"
 36 #include "gc/g1/g1Policy.hpp"
 37 #include "gc/g1/g1RegionMarkStatsCache.inline.hpp"
 38 #include "gc/shared/gcTraceTime.inline.hpp"
 39 #include "gc/shared/preservedMarks.inline.hpp"
 40 #include "gc/shared/classUnloadingContext.hpp"
 41 #include "gc/shared/referenceProcessor.hpp"
 42 #include "gc/shared/verifyOption.hpp"
 43 #include "gc/shared/weakProcessor.inline.hpp"
 44 #include "gc/shared/workerPolicy.hpp"
 45 #include "logging/log.hpp"
 46 #include "runtime/handles.inline.hpp"
 47 #include "utilities/debug.hpp"
 48 
 49 static void clear_and_activate_derived_pointers() {
 50 #if COMPILER2_OR_JVMCI
 51   DerivedPointerTable::clear();
 52 #endif
 53 }
 54 
 55 static void deactivate_derived_pointers() {
 56 #if COMPILER2_OR_JVMCI
 57   DerivedPointerTable::set_active(false);
 58 #endif
 59 }
 60 
 61 static void update_derived_pointers() {
 62 #if COMPILER2_OR_JVMCI
 63   DerivedPointerTable::update_pointers();
 64 #endif
 65 }
 66 
 67 G1CMBitMap* G1FullCollector::mark_bitmap() {
 68   return _heap->concurrent_mark()->mark_bitmap();
 69 }
 70 
 71 ReferenceProcessor* G1FullCollector::reference_processor() {
 72   return _heap->ref_processor_stw();
 73 }
 74 
 75 uint G1FullCollector::calc_active_workers() {
 76   G1CollectedHeap* heap = G1CollectedHeap::heap();
 77   uint max_worker_count = heap->workers()->max_workers();
 78   // Only calculate number of workers if UseDynamicNumberOfGCThreads
 79   // is enabled, otherwise use max.
 80   if (!UseDynamicNumberOfGCThreads) {
 81     return max_worker_count;
 82   }
 83 
 84   // Consider G1HeapWastePercent to decide max number of workers. Each worker
 85   // will in average cause half a region waste.
 86   uint max_wasted_regions_allowed = ((heap->num_regions() * G1HeapWastePercent) / 100);
 87   uint waste_worker_count = MAX2((max_wasted_regions_allowed * 2) , 1u);
 88   uint heap_waste_worker_limit = MIN2(waste_worker_count, max_worker_count);
 89 
 90   // Also consider HeapSizePerGCThread by calling WorkerPolicy to calculate
 91   // the number of workers.
 92   uint current_active_workers = heap->workers()->active_workers();
 93   uint active_worker_limit = WorkerPolicy::calc_active_workers(max_worker_count, current_active_workers, 0);
 94 
 95   // Finally consider the amount of used regions.
 96   uint used_worker_limit = heap->num_used_regions();
 97   assert(used_worker_limit > 0, "Should never have zero used regions.");
 98 
 99   // Update active workers to the lower of the limits.
100   uint worker_count = MIN3(heap_waste_worker_limit, active_worker_limit, used_worker_limit);
101   log_debug(gc, task)("Requesting %u active workers for full compaction (waste limited workers: %u, "
102                       "adaptive workers: %u, used limited workers: %u)",
103                       worker_count, heap_waste_worker_limit, active_worker_limit, used_worker_limit);
104   worker_count = heap->workers()->set_active_workers(worker_count);
105   log_info(gc, task)("Using %u workers of %u for full compaction", worker_count, max_worker_count);
106 
107   return worker_count;
108 }
109 
110 G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
111                                  bool clear_soft_refs,
112                                  bool do_maximal_compaction,
113                                  G1FullGCTracer* tracer) :
114     _heap(heap),
115     _scope(heap->monitoring_support(), clear_soft_refs, do_maximal_compaction, tracer),
116     _num_workers(calc_active_workers()),
117     _has_compaction_targets(false),
118     _has_humongous(false),
119     _oop_queue_set(_num_workers),
120     _array_queue_set(_num_workers),
121     _preserved_marks_set(true),
122     _serial_compaction_point(this, nullptr),
123     _humongous_compaction_point(this, nullptr),
124     _is_alive(this, heap->concurrent_mark()->mark_bitmap()),
125     _is_alive_mutator(heap->ref_processor_stw(), &_is_alive),
126     _humongous_compaction_regions(8),
127     _always_subject_to_discovery(),
128     _is_subject_mutator(heap->ref_processor_stw(), &_always_subject_to_discovery),
129     _region_attr_table() {
130   assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint");
131 
132   _preserved_marks_set.init(_num_workers);
133   _markers = NEW_C_HEAP_ARRAY(G1FullGCMarker*, _num_workers, mtGC);
134   _compaction_points = NEW_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _num_workers, mtGC);
135 
136   _live_stats = NEW_C_HEAP_ARRAY(G1RegionMarkStats, _heap->max_regions(), mtGC);
137   _compaction_tops = NEW_C_HEAP_ARRAY(HeapWord*, _heap->max_regions(), mtGC);
138   for (uint j = 0; j < heap->max_regions(); j++) {
139     _live_stats[j].clear();
140     _compaction_tops[j] = nullptr;
141   }
142 
143   for (uint i = 0; i < _num_workers; i++) {
144     _markers[i] = new G1FullGCMarker(this, i, _live_stats);
145     _compaction_points[i] = new G1FullGCCompactionPoint(this, _preserved_marks_set.get(i));
146     _oop_queue_set.register_queue(i, marker(i)->oop_stack());
147     _array_queue_set.register_queue(i, marker(i)->objarray_stack());
148   }
149   _serial_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
150   _humongous_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
151   _region_attr_table.initialize(heap->reserved(), G1HeapRegion::GrainBytes);
152 }
153 
154 G1FullCollector::~G1FullCollector() {
155   for (uint i = 0; i < _num_workers; i++) {
156     delete _markers[i];
157     delete _compaction_points[i];
158   }
159 
160   FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers);
161   FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points);
162   FREE_C_HEAP_ARRAY(HeapWord*, _compaction_tops);
163   FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats);
164 }
165 
166 class PrepareRegionsClosure : public G1HeapRegionClosure {
167   G1FullCollector* _collector;
168 
169 public:
170   PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { }
171 
172   bool do_heap_region(G1HeapRegion* hr) {
173     hr->prepare_for_full_gc();
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->flush_region_pin_cache();
194   _heap->prepare_heap_for_full_collection();
195 
196   PrepareRegionsClosure cl(this);
197   _heap->heap_region_iterate(&cl);
198 
199   reference_processor()->start_discovery(scope()->should_clear_soft_refs());
200 
201   // Clear and activate derived pointer collection.
202   clear_and_activate_derived_pointers();
203 }
204 
205 void G1FullCollector::collect() {
206   G1CollectedHeap::start_codecache_marking_cycle_if_inactive(false /* concurrent_mark_start */);
207 
208   phase1_mark_live_objects();
209   verify_after_marking();
210 
211   // Don't add any more derived pointers during later phases
212   deactivate_derived_pointers();
213 
214   phase2_prepare_compaction();
215 
216   if (has_compaction_targets()) {
217     phase3_adjust_pointers();
218 
219     phase4_do_compaction();
220   } else {
221     // All regions have a high live ratio thus will not be compacted.
222     // The live ratio is only considered if do_maximal_compaction is false.
223     log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap");
224   }
225 
226   phase5_reset_metadata();
227 
228   G1CollectedHeap::finish_codecache_marking_cycle();
229 }
230 
231 void G1FullCollector::complete_collection() {
232   // Restore all marks.
233   restore_marks();
234 
235   // When the pointers have been adjusted and moved, we can
236   // update the derived pointer table.
237   update_derived_pointers();
238 
239   // Need completely cleared claim bits for the next concurrent marking or full gc.
240   ClassLoaderDataGraph::clear_claimed_marks();
241 
242   // Prepare the bitmap for the next (potentially concurrent) marking.
243   _heap->concurrent_mark()->clear_bitmap(_heap->workers());
244 
245   _heap->prepare_for_mutator_after_full_collection();
246 
247   _heap->resize_all_tlabs();
248 
249   _heap->young_regions_cardset()->clear();
250 
251   _heap->policy()->record_full_collection_end();
252   _heap->gc_epilogue(true);
253 
254   _heap->verify_after_full_collection();
255 
256   _heap->print_heap_after_full_collection();
257 }
258 
259 void G1FullCollector::before_marking_update_attribute_table(G1HeapRegion* hr) {
260   if (hr->is_free()) {
261     _region_attr_table.set_free(hr->hrm_index());
262   } else if (hr->is_humongous() || hr->has_pinned_objects()) {
263     // Humongous objects or pinned regions will never be moved in the "main"
264     // compaction phase, but non-pinned regions might afterwards in a special phase.
265     _region_attr_table.set_skip_compacting(hr->hrm_index());
266   } else {
267     // Everything else should be compacted.
268     _region_attr_table.set_compacting(hr->hrm_index());
269   }
270 }
271 
272 class G1FullGCRefProcProxyTask : public RefProcProxyTask {
273   G1FullCollector& _collector;
274 
275 public:
276   G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
277     : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
278       _collector(collector) {}
279 
280   void work(uint worker_id) override {
281     assert(worker_id < _max_workers, "sanity");
282     G1IsAliveClosure is_alive(&_collector);
283     uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
284     G1FullKeepAliveClosure keep_alive(_collector.marker(index));
285     BarrierEnqueueDiscoveredFieldClosure enqueue;
286     G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure();
287     _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc);
288   }
289 };
290 
291 void G1FullCollector::phase1_mark_live_objects() {
292   // Recursively traverse all live objects and mark them.
293   GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer());
294 
295   {
296     // Do the actual marking.
297     G1FullGCMarkTask marking_task(this);
298     run_task(&marking_task);
299   }
300 
301   {
302     uint old_active_mt_degree = reference_processor()->num_queues();
303     reference_processor()->set_active_mt_degree(workers());
304     GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer());
305     // Process reference objects found during marking.
306     ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues());
307     G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues());
308     const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, pt);
309     scope()->tracer()->report_gc_reference_stats(stats);
310     pt.print_all_references();
311     assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack");
312 
313     reference_processor()->set_active_mt_degree(old_active_mt_degree);
314   }
315 
316   {
317     GCTraceTime(Debug, gc, phases) debug("Phase 1: Flush Mark Stats Cache", scope()->timer());
318     for (uint i = 0; i < workers(); i++) {
319       marker(i)->flush_mark_stats_cache();
320     }
321   }
322 
323   // Weak oops cleanup.
324   {
325     GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer());
326     WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1);
327   }
328 
329   // Class unloading and cleanup.
330   if (ClassUnloading) {
331     _heap->unload_classes_and_code("Phase 1: Class Unloading and Cleanup", &_is_alive, scope()->timer());
332   }
333 
334   {
335     GCTraceTime(Debug, gc, phases) debug("Report Object Count", scope()->timer());
336     scope()->tracer()->report_object_count_after_gc(&_is_alive, _heap->workers());
337   }
338 #if TASKQUEUE_STATS
339   oop_queue_set()->print_and_reset_taskqueue_stats("Oop Queue");
340   array_queue_set()->print_and_reset_taskqueue_stats("ObjArrayOop Queue");
341 #endif
342 }
343 
344 void G1FullCollector::phase2_prepare_compaction() {
345   GCTraceTime(Info, gc, phases) info("Phase 2: Prepare compaction", scope()->timer());
346 
347   phase2a_determine_worklists();
348 
349   if (!has_compaction_targets()) {
350     return;
351   }
352 
353   bool has_free_compaction_targets = phase2b_forward_oops();
354 
355   // Try to avoid OOM immediately after Full GC in case there are no free regions
356   // left after determining the result locations (i.e. this phase). Prepare to
357   // maximally compact the tail regions of the compaction queues serially.
358   if (scope()->do_maximal_compaction() || !has_free_compaction_targets) {
359     phase2c_prepare_serial_compaction();
360 
361     if (scope()->do_maximal_compaction() &&
362         has_humongous() &&
363         serial_compaction_point()->has_regions()) {
364       phase2d_prepare_humongous_compaction();
365     }
366   }
367 }
368 
369 void G1FullCollector::phase2a_determine_worklists() {
370   GCTraceTime(Debug, gc, phases) debug("Phase 2: Determine work lists", scope()->timer());
371 
372   G1DetermineCompactionQueueClosure cl(this);
373   _heap->heap_region_iterate(&cl);
374 }
375 
376 bool G1FullCollector::phase2b_forward_oops() {
377   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare parallel compaction", scope()->timer());
378 
379   G1FullGCPrepareTask task(this);
380   run_task(&task);
381 
382   return task.has_free_compaction_targets();
383 }
384 
385 uint G1FullCollector::truncate_parallel_cps() {
386   uint lowest_current = UINT_MAX;
387   for (uint i = 0; i < workers(); i++) {
388     G1FullGCCompactionPoint* cp = compaction_point(i);
389     if (cp->has_regions()) {
390       lowest_current = MIN2(lowest_current, cp->current_region()->hrm_index());
391     }
392   }
393 
394   for (uint i = 0; i < workers(); i++) {
395     G1FullGCCompactionPoint* cp = compaction_point(i);
396     if (cp->has_regions()) {
397       cp->remove_at_or_above(lowest_current);
398     }
399   }
400   return lowest_current;
401 }
402 
403 void G1FullCollector::phase2c_prepare_serial_compaction() {
404   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer());
405   // At this point, we know that after parallel compaction there will be regions that
406   // are partially compacted into. Thus, the last compaction region of all
407   // compaction queues still have space in them. We try to re-compact these regions
408   // in serial to avoid a premature OOM when the mutator wants to allocate the first
409   // eden region after gc.
410 
411   // For maximum compaction, we need to re-prepare all objects above the lowest
412   // region among the current regions for all thread compaction points. It may
413   // happen that due to the uneven distribution of objects to parallel threads, holes
414   // have been created as threads compact to different target regions between the
415   // lowest and the highest region in the tails of the compaction points.
416 
417   uint start_serial = truncate_parallel_cps();
418   assert(start_serial < _heap->max_reserved_regions(), "Called on empty parallel compaction queues");
419 
420   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
421   assert(!serial_cp->is_initialized(), "sanity!");
422 
423   G1HeapRegion* start_hr = _heap->region_at(start_serial);
424   serial_cp->add(start_hr);
425   serial_cp->initialize(start_hr);
426 
427   HeapWord* dense_prefix_top = compaction_top(start_hr);
428   G1SerialRePrepareClosure re_prepare(serial_cp, dense_prefix_top);
429 
430   for (uint i = start_serial + 1; i < _heap->max_reserved_regions(); i++) {
431     if (is_compaction_target(i)) {
432       G1HeapRegion* current = _heap->region_at(i);
433       set_compaction_top(current, current->bottom());
434       serial_cp->add(current);
435       current->apply_to_marked_objects(mark_bitmap(), &re_prepare);
436     }
437   }
438   serial_cp->update();
439 }
440 
441 void G1FullCollector::phase2d_prepare_humongous_compaction() {
442   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare humongous compaction", scope()->timer());
443   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
444   assert(serial_cp->has_regions(), "Sanity!" );
445 
446   uint last_serial_target = serial_cp->current_region()->hrm_index();
447   uint region_index = last_serial_target + 1;
448   uint max_reserved_regions = _heap->max_reserved_regions();
449 
450   G1FullGCCompactionPoint* humongous_cp = humongous_compaction_point();
451 
452   while (region_index < max_reserved_regions) {
453     G1HeapRegion* hr = _heap->region_at_or_null(region_index);
454 
455     if (hr == nullptr) {
456       region_index++;
457       continue;
458     } else if (hr->is_starts_humongous()) {
459       size_t obj_size = cast_to_oop(hr->bottom())->size();
460       uint num_regions = (uint)G1CollectedHeap::humongous_obj_size_in_regions(obj_size);
461       // Even during last-ditch compaction we should not move pinned humongous objects.
462       if (!hr->has_pinned_objects()) {
463         humongous_cp->forward_humongous(hr);
464       }
465       region_index += num_regions; // Advance over all humongous regions.
466       continue;
467     } else if (is_compaction_target(region_index)) {
468       assert(!hr->has_pinned_objects(), "pinned regions should not be compaction targets");
469       // Add the region to the humongous compaction point.
470       humongous_cp->add(hr);
471     }
472     region_index++;
473   }
474 }
475 
476 void G1FullCollector::phase3_adjust_pointers() {
477   // Adjust the pointers to reflect the new locations
478   GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer());
479 
480   G1FullGCAdjustTask task(this);
481   run_task(&task);
482 }
483 
484 void G1FullCollector::phase4_do_compaction() {
485   // Compact the heap using the compaction queues created in phase 2.
486   GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer());
487   G1FullGCCompactTask task(this);
488   run_task(&task);
489 
490   // Serial compact to avoid OOM when very few free regions.
491   if (serial_compaction_point()->has_regions()) {
492     task.serial_compaction();
493   }
494 
495   if (!_humongous_compaction_regions.is_empty()) {
496     assert(scope()->do_maximal_compaction(), "Only compact humongous during maximal compaction");
497     task.humongous_compaction();
498   }
499 }
500 
501 void G1FullCollector::phase5_reset_metadata() {
502   // Clear region metadata that is invalid after GC for all regions.
503   GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer());
504   G1FullGCResetMetadataTask task(this);
505   run_task(&task);
506 }
507 
508 void G1FullCollector::restore_marks() {
509   _preserved_marks_set.restore(_heap->workers());
510   _preserved_marks_set.reclaim();
511 }
512 
513 void G1FullCollector::run_task(WorkerTask* task) {
514   _heap->workers()->run_task(task, _num_workers);
515 }
516 
517 void G1FullCollector::verify_after_marking() {
518   if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) {
519     // Only do verification if VerifyDuringGC and G1VerifyFull is set.
520     return;
521   }
522 
523 #if COMPILER2_OR_JVMCI
524   DerivedPointerTableDeactivate dpt_deact;
525 #endif
526   _heap->prepare_for_verify();
527   // Note: we can verify only the heap here. When an object is
528   // marked, the previous value of the mark word (including
529   // identity hash values, ages, etc) is preserved, and the mark
530   // word is set to markWord::marked_value - effectively removing
531   // any hash values from the mark word. These hash values are
532   // used when verifying the dictionaries and so removing them
533   // from the mark word can make verification of the dictionaries
534   // fail. At the end of the GC, the original mark word values
535   // (including hash values) are restored to the appropriate
536   // objects.
537   GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)");
538   _heap->verify(VerifyOption::G1UseFullMarking);
539 }