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.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 explicit_gc,
115                                  bool clear_soft_refs,
116                                  bool do_maximal_compaction,
117                                  G1FullGCTracer* tracer) :
118     _heap(heap),
119     _scope(heap->monitoring_support(), explicit_gc, clear_soft_refs, do_maximal_compaction, tracer),
120     _num_workers(calc_active_workers()),
121     _has_compaction_targets(false),
122     _oop_queue_set(_num_workers),
123     _array_queue_set(_num_workers),
124     _preserved_marks_set(true),
125     _serial_compaction_point(this),
126     _is_alive(this, heap->concurrent_mark()->mark_bitmap()),
127     _is_alive_mutator(heap->ref_processor_stw(), &_is_alive),
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_regions(), mtGC);
138   _compaction_tops = NEW_C_HEAP_ARRAY(HeapWord*, _heap->max_regions(), mtGC);
139   for (uint j = 0; j < heap->max_regions(); j++) {
140     _live_stats[j].clear();
141     _compaction_tops[j] = nullptr;
142   }
143 
144   for (uint i = 0; i < _num_workers; i++) {
145     _markers[i] = new G1FullGCMarker(this, i, _preserved_marks_set.get(i), _live_stats);
146     _compaction_points[i] = new G1FullGCCompactionPoint(this);
147     _oop_queue_set.register_queue(i, marker(i)->oop_stack());
148     _array_queue_set.register_queue(i, marker(i)->objarray_stack());
149   }
150   _region_attr_table.initialize(heap->reserved(), HeapRegion::GrainBytes);
151 }
152 
153 G1FullCollector::~G1FullCollector() {
154   for (uint i = 0; i < _num_workers; i++) {
155     delete _markers[i];
156     delete _compaction_points[i];
157   }
158   FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers);
159   FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points);
160   FREE_C_HEAP_ARRAY(HeapWord*, _compaction_tops);
161   FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats);
162 }
163 
164 class PrepareRegionsClosure : public HeapRegionClosure {
165   G1FullCollector* _collector;
166 
167 public:
168   PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { }
169 
170   bool do_heap_region(HeapRegion* hr) {
171     G1CollectedHeap::heap()->prepare_region_for_full_compaction(hr);
172     _collector->before_marking_update_attribute_table(hr);
173     return false;
174   }
175 };
176 
177 void G1FullCollector::prepare_collection() {
178   _heap->policy()->record_full_collection_start();
179 
180   // Verification needs the bitmap, so we should clear the bitmap only later.
181   bool in_concurrent_cycle = _heap->abort_concurrent_cycle();
182   _heap->verify_before_full_collection(scope()->is_explicit_gc());
183   if (in_concurrent_cycle) {
184     GCTraceTime(Debug, gc) debug("Clear Bitmap");
185     _heap->concurrent_mark()->clear_bitmap(_heap->workers());
186   }
187 
188   _heap->gc_prologue(true);
189   _heap->retire_tlabs();
190   _heap->prepare_heap_for_full_collection();
191 
192   PrepareRegionsClosure cl(this);
193   _heap->heap_region_iterate(&cl);
194 
195   reference_processor()->start_discovery(scope()->should_clear_soft_refs());
196 
197   // Clear and activate derived pointer collection.
198   clear_and_activate_derived_pointers();
199 }
200 
201 void G1FullCollector::collect() {
202   G1CollectedHeap::start_codecache_marking_cycle_if_inactive(false /* concurrent_mark_start */);
203 
204   phase1_mark_live_objects();
205   verify_after_marking();
206 
207   // Don't add any more derived pointers during later phases
208   deactivate_derived_pointers();
209 
210   phase2_prepare_compaction();
211 
212   if (has_compaction_targets()) {
213     phase3_adjust_pointers();
214 
215     phase4_do_compaction();
216   } else {
217     // All regions have a high live ratio thus will not be compacted.
218     // The live ratio is only considered if do_maximal_compaction is false.
219     log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap");
220   }
221 
222   phase5_reset_metadata();
223 
224   G1CollectedHeap::finish_codecache_marking_cycle();
225 }
226 
227 void G1FullCollector::complete_collection() {
228   // Restore all marks.
229   restore_marks();
230 
231   // When the pointers have been adjusted and moved, we can
232   // update the derived pointer table.
233   update_derived_pointers();
234 
235   // Need completely cleared claim bits for the next concurrent marking or full gc.
236   ClassLoaderDataGraph::clear_claimed_marks();
237 
238   // Prepare the bitmap for the next (potentially concurrent) marking.
239   _heap->concurrent_mark()->clear_bitmap(_heap->workers());
240 
241   _heap->prepare_for_mutator_after_full_collection();
242 
243   _heap->resize_all_tlabs();
244 
245   _heap->policy()->record_full_collection_end();
246   _heap->gc_epilogue(true);
247 
248   _heap->verify_after_full_collection();
249 }
250 
251 void G1FullCollector::before_marking_update_attribute_table(HeapRegion* hr) {
252   if (hr->is_free()) {
253     _region_attr_table.set_free(hr->hrm_index());
254   } else if (hr->is_closed_archive()) {
255     _region_attr_table.set_skip_marking(hr->hrm_index());
256   } else if (hr->is_pinned()) {
257     _region_attr_table.set_skip_compacting(hr->hrm_index());
258   } else {
259     // Everything else should be compacted.
260     _region_attr_table.set_compacting(hr->hrm_index());
261   }
262 }
263 
264 class G1FullGCRefProcProxyTask : public RefProcProxyTask {
265   G1FullCollector& _collector;
266 
267 public:
268   G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
269     : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
270       _collector(collector) {}
271 
272   void work(uint worker_id) override {
273     assert(worker_id < _max_workers, "sanity");
274     G1IsAliveClosure is_alive(&_collector);
275     uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
276     G1FullKeepAliveClosure keep_alive(_collector.marker(index));
277     BarrierEnqueueDiscoveredFieldClosure enqueue;
278     G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure();
279     _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc);
280   }
281 };
282 
283 void G1FullCollector::phase1_mark_live_objects() {
284   // Recursively traverse all live objects and mark them.
285   GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer());
286 
287   {
288     // Do the actual marking.
289     G1FullGCMarkTask marking_task(this);
290     run_task(&marking_task);
291   }
292 
293   {
294     uint old_active_mt_degree = reference_processor()->num_queues();
295     reference_processor()->set_active_mt_degree(workers());
296     GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer());
297     // Process reference objects found during marking.
298     ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues());
299     G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues());
300     const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, pt);
301     scope()->tracer()->report_gc_reference_stats(stats);
302     pt.print_all_references();
303     assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack");
304 
305     reference_processor()->set_active_mt_degree(old_active_mt_degree);
306   }
307 
308   // Weak oops cleanup.
309   {
310     GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer());
311     WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1);
312   }
313 
314   // Class unloading and cleanup.
315   if (ClassUnloading) {
316     GCTraceTime(Debug, gc, phases) debug("Phase 1: Class Unloading and Cleanup", scope()->timer());
317     CodeCache::UnloadingScope unloading_scope(&_is_alive);
318     // Unload classes and purge the SystemDictionary.
319     bool purged_class = SystemDictionary::do_unloading(scope()->timer());
320     _heap->complete_cleaning(purged_class);
321   }
322 
323   scope()->tracer()->report_object_count_after_gc(&_is_alive);
324 #if TASKQUEUE_STATS
325   oop_queue_set()->print_and_reset_taskqueue_stats("Oop Queue");
326   array_queue_set()->print_and_reset_taskqueue_stats("ObjArrayOop Queue");
327 #endif
328 }
329 
330 void G1FullCollector::phase2_prepare_compaction() {
331   GCTraceTime(Info, gc, phases) info("Phase 2: Prepare compaction", scope()->timer());
332 
333   phase2a_determine_worklists();
334 
335   if (!has_compaction_targets()) {
336     return;
337   }
338 
339   bool has_free_compaction_targets = phase2b_forward_oops();
340 
341   // Try to avoid OOM immediately after Full GC in case there are no free regions
342   // left after determining the result locations (i.e. this phase). Prepare to
343   // maximally compact the tail regions of the compaction queues serially.
344   if (scope()->do_maximal_compaction() || !has_free_compaction_targets) {
345     phase2c_prepare_serial_compaction();
346   }
347 }
348 
349 void G1FullCollector::phase2a_determine_worklists() {
350   GCTraceTime(Debug, gc, phases) debug("Phase 2: Determine work lists", scope()->timer());
351 
352   G1DetermineCompactionQueueClosure cl(this);
353   _heap->heap_region_iterate(&cl);
354 }
355 
356 bool G1FullCollector::phase2b_forward_oops() {
357   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare parallel compaction", scope()->timer());
358 
359   G1FullGCPrepareTask task(this);
360   run_task(&task);
361 
362   return task.has_free_compaction_targets();
363 }
364 
365 uint G1FullCollector::truncate_parallel_cps() {
366   uint lowest_current = (uint)-1;
367   for (uint i = 0; i < workers(); i++) {
368     G1FullGCCompactionPoint* cp = compaction_point(i);
369     if (cp->has_regions()) {
370       lowest_current = MIN2(lowest_current, cp->current_region()->hrm_index());
371     }
372   }
373 
374   for (uint i = 0; i < workers(); i++) {
375     G1FullGCCompactionPoint* cp = compaction_point(i);
376     if (cp->has_regions()) {
377       cp->remove_at_or_above(lowest_current);
378     }
379   }
380   return lowest_current;
381 }
382 
383 void G1FullCollector::phase2c_prepare_serial_compaction() {
384   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer());
385   // At this point, we know that after parallel compaction there will be regions that
386   // are partially compacted into. Thus, the last compaction region of all
387   // compaction queues still have space in them. We try to re-compact these regions
388   // in serial to avoid a premature OOM when the mutator wants to allocate the first
389   // eden region after gc.
390 
391   // For maximum compaction, we need to re-prepare all objects above the lowest
392   // region among the current regions for all thread compaction points. It may
393   // happen that due to the uneven distribution of objects to parallel threads, holes
394   // have been created as threads compact to different target regions between the
395   // lowest and the highest region in the tails of the compaction points.
396 
397   uint start_serial = truncate_parallel_cps();
398   assert(start_serial < _heap->max_reserved_regions(), "Called on empty parallel compaction queues");
399 
400   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
401   assert(!serial_cp->is_initialized(), "sanity!");
402 
403   HeapRegion* start_hr = _heap->region_at(start_serial);
404   serial_cp->add(start_hr);
405   serial_cp->initialize(start_hr);
406 
407   HeapWord* dense_prefix_top = compaction_top(start_hr);
408   G1SerialRePrepareClosure re_prepare(serial_cp, dense_prefix_top);
409 
410   for (uint i = start_serial + 1; i < _heap->max_reserved_regions(); i++) {
411     if (is_compaction_target(i)) {
412       HeapRegion* current = _heap->region_at(i);
413       set_compaction_top(current, current->bottom());
414       serial_cp->add(current);
415       current->apply_to_marked_objects(mark_bitmap(), &re_prepare);
416     }
417   }
418   serial_cp->update();
419 }
420 
421 void G1FullCollector::phase3_adjust_pointers() {
422   // Adjust the pointers to reflect the new locations
423   GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer());
424 
425   G1FullGCAdjustTask task(this);
426   run_task(&task);
427 }
428 
429 void G1FullCollector::phase4_do_compaction() {
430   // Compact the heap using the compaction queues created in phase 2.
431   GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer());
432   G1FullGCCompactTask task(this);
433   run_task(&task);
434 
435   // Serial compact to avoid OOM when very few free regions.
436   if (serial_compaction_point()->has_regions()) {
437     task.serial_compaction();
438   }
439 }
440 
441 void G1FullCollector::phase5_reset_metadata() {
442   // Clear region metadata that is invalid after GC for all regions.
443   GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer());
444   G1FullGCResetMetadataTask task(this);
445   run_task(&task);
446 }
447 
448 void G1FullCollector::restore_marks() {
449   _preserved_marks_set.restore(_heap->workers());
450   _preserved_marks_set.reclaim();
451 }
452 
453 void G1FullCollector::run_task(WorkerTask* task) {
454   _heap->workers()->run_task(task, _num_workers);
455 }
456 
457 void G1FullCollector::verify_after_marking() {
458   if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) {
459     // Only do verification if VerifyDuringGC and G1VerifyFull is set.
460     return;
461   }
462 
463 #if COMPILER2_OR_JVMCI
464   DerivedPointerTableDeactivate dpt_deact;
465 #endif
466   _heap->prepare_for_verify();
467   // Note: we can verify only the heap here. When an object is
468   // marked, the previous value of the mark word (including
469   // identity hash values, ages, etc) is preserved, and the mark
470   // word is set to markWord::marked_value - effectively removing
471   // any hash values from the mark word. These hash values are
472   // used when verifying the dictionaries and so removing them
473   // from the mark word can make verification of the dictionaries
474   // fail. At the end of the GC, the original mark word values
475   // (including hash values) are restored to the appropriate
476   // objects.
477   GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)");
478   _heap->verify(VerifyOption::G1UseFullMarking);
479 }