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 "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/gcTraceTime.inline.hpp"
 40 #include "gc/shared/preservedMarks.inline.hpp"
 41 #include "gc/shared/classUnloadingContext.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_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                                  G1FullGCTracer* 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     _oop_queue_set(_num_workers),
121     _array_queue_set(_num_workers),
122     _preserved_marks_set(true),
123     _serial_compaction_point(this),
124     _humongous_compaction_point(this),
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_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 
159   FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers);
160   FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points);
161   FREE_C_HEAP_ARRAY(HeapWord*, _compaction_tops);
162   FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats);
163 }
164 
165 class PrepareRegionsClosure : public HeapRegionClosure {
166   G1FullCollector* _collector;
167 
168 public:
169   PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { }
170 
171   bool do_heap_region(HeapRegion* hr) {
172     hr->prepare_for_full_gc();
173     G1CollectedHeap::heap()->prepare_region_for_full_compaction(hr);
174     _collector->before_marking_update_attribute_table(hr);
175     return false;
176   }
177 };
178 
179 void G1FullCollector::prepare_collection() {
180   _heap->policy()->record_full_collection_start();
181 
182   // Verification needs the bitmap, so we should clear the bitmap only later.
183   bool in_concurrent_cycle = _heap->abort_concurrent_cycle();
184   _heap->verify_before_full_collection();
185   if (in_concurrent_cycle) {
186     GCTraceTime(Debug, gc) debug("Clear Bitmap");
187     _heap->concurrent_mark()->clear_bitmap(_heap->workers());
188   }
189 
190   _heap->gc_prologue(true);
191   _heap->retire_tlabs();
192   _heap->prepare_heap_for_full_collection();
193 
194   PrepareRegionsClosure cl(this);
195   _heap->heap_region_iterate(&cl);
196 
197   reference_processor()->start_discovery(scope()->should_clear_soft_refs());
198 
199   // Clear and activate derived pointer collection.
200   clear_and_activate_derived_pointers();
201 }
202 
203 void G1FullCollector::collect() {
204   G1CollectedHeap::start_codecache_marking_cycle_if_inactive(false /* concurrent_mark_start */);
205 
206   phase1_mark_live_objects();
207   verify_after_marking();
208 
209   // Don't add any more derived pointers during later phases
210   deactivate_derived_pointers();
211 
212   phase2_prepare_compaction();
213 
214   if (has_compaction_targets()) {
215     phase3_adjust_pointers();
216 
217     phase4_do_compaction();
218   } else {
219     // All regions have a high live ratio thus will not be compacted.
220     // The live ratio is only considered if do_maximal_compaction is false.
221     log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap");
222   }
223 
224   phase5_reset_metadata();
225 
226   G1CollectedHeap::finish_codecache_marking_cycle();
227 }
228 
229 void G1FullCollector::complete_collection() {
230   // Restore all marks.
231   restore_marks();
232 
233   // When the pointers have been adjusted and moved, we can
234   // update the derived pointer table.
235   update_derived_pointers();
236 
237   // Need completely cleared claim bits for the next concurrent marking or full gc.
238   ClassLoaderDataGraph::clear_claimed_marks();
239 
240   // Prepare the bitmap for the next (potentially concurrent) marking.
241   _heap->concurrent_mark()->clear_bitmap(_heap->workers());
242 
243   _heap->prepare_for_mutator_after_full_collection();
244 
245   _heap->resize_all_tlabs();
246 
247   _heap->policy()->record_full_collection_end();
248   _heap->gc_epilogue(true);
249 
250   _heap->verify_after_full_collection();
251 
252   _heap->print_heap_after_full_collection();
253 }
254 
255 void G1FullCollector::before_marking_update_attribute_table(HeapRegion* hr) {
256   if (hr->is_free()) {
257     _region_attr_table.set_free(hr->hrm_index());
258   } else if (hr->is_humongous()) {
259     // Humongous objects will never be moved in the "main" compaction phase, but
260     // afterwards in a special phase if needed.
261     _region_attr_table.set_skip_compacting(hr->hrm_index());
262   } else {
263     // Everything else should be compacted.
264     _region_attr_table.set_compacting(hr->hrm_index());
265   }
266 }
267 
268 class G1FullGCRefProcProxyTask : public RefProcProxyTask {
269   G1FullCollector& _collector;
270 
271 public:
272   G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
273     : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
274       _collector(collector) {}
275 
276   void work(uint worker_id) override {
277     assert(worker_id < _max_workers, "sanity");
278     G1IsAliveClosure is_alive(&_collector);
279     uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
280     G1FullKeepAliveClosure keep_alive(_collector.marker(index));
281     BarrierEnqueueDiscoveredFieldClosure enqueue;
282     G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure();
283     _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc);
284   }
285 };
286 
287 void G1FullCollector::phase1_mark_live_objects() {
288   // Recursively traverse all live objects and mark them.
289   GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer());
290 
291   {
292     // Do the actual marking.
293     G1FullGCMarkTask marking_task(this);
294     run_task(&marking_task);
295   }
296 
297   {
298     uint old_active_mt_degree = reference_processor()->num_queues();
299     reference_processor()->set_active_mt_degree(workers());
300     GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer());
301     // Process reference objects found during marking.
302     ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues());
303     G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues());
304     const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, pt);
305     scope()->tracer()->report_gc_reference_stats(stats);
306     pt.print_all_references();
307     assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack");
308 
309     reference_processor()->set_active_mt_degree(old_active_mt_degree);
310   }
311 
312   // Weak oops cleanup.
313   {
314     GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer());
315     WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1);
316   }
317 
318   // Class unloading and cleanup.
319   if (ClassUnloading) {
320     _heap->unload_classes_and_code("Phase 1: Class Unloading and Cleanup", &_is_alive, scope()->timer());
321   }
322 
323   {
324     GCTraceTime(Debug, gc, phases) debug("Report Object Count", scope()->timer());
325     scope()->tracer()->report_object_count_after_gc(&_is_alive, _heap->workers());
326   }
327 #if TASKQUEUE_STATS
328   oop_queue_set()->print_and_reset_taskqueue_stats("Oop Queue");
329   array_queue_set()->print_and_reset_taskqueue_stats("ObjArrayOop Queue");
330 #endif
331 }
332 
333 void G1FullCollector::phase2_prepare_compaction() {
334   GCTraceTime(Info, gc, phases) info("Phase 2: Prepare compaction", scope()->timer());
335 
336   phase2a_determine_worklists();
337 
338   if (!has_compaction_targets()) {
339     return;
340   }
341 
342   bool has_free_compaction_targets = phase2b_forward_oops();
343 
344   // Try to avoid OOM immediately after Full GC in case there are no free regions
345   // left after determining the result locations (i.e. this phase). Prepare to
346   // maximally compact the tail regions of the compaction queues serially.
347   if (scope()->do_maximal_compaction() || !has_free_compaction_targets) {
348     phase2c_prepare_serial_compaction();
349 
350     if (scope()->do_maximal_compaction() &&
351         has_humongous() &&
352         serial_compaction_point()->has_regions()) {
353       phase2d_prepare_humongous_compaction();
354     }
355   }
356 }
357 
358 void G1FullCollector::phase2a_determine_worklists() {
359   GCTraceTime(Debug, gc, phases) debug("Phase 2: Determine work lists", scope()->timer());
360 
361   G1DetermineCompactionQueueClosure cl(this);
362   _heap->heap_region_iterate(&cl);
363 }
364 
365 bool G1FullCollector::phase2b_forward_oops() {
366   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare parallel compaction", scope()->timer());
367 
368   G1FullGCPrepareTask task(this);
369   run_task(&task);
370 
371   return task.has_free_compaction_targets();
372 }
373 
374 uint G1FullCollector::truncate_parallel_cps() {
375   uint lowest_current = UINT_MAX;
376   for (uint i = 0; i < workers(); i++) {
377     G1FullGCCompactionPoint* cp = compaction_point(i);
378     if (cp->has_regions()) {
379       lowest_current = MIN2(lowest_current, cp->current_region()->hrm_index());
380     }
381   }
382 
383   for (uint i = 0; i < workers(); i++) {
384     G1FullGCCompactionPoint* cp = compaction_point(i);
385     if (cp->has_regions()) {
386       cp->remove_at_or_above(lowest_current);
387     }
388   }
389   return lowest_current;
390 }
391 
392 void G1FullCollector::phase2c_prepare_serial_compaction() {
393   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer());
394   // At this point, we know that after parallel compaction there will be regions that
395   // are partially compacted into. Thus, the last compaction region of all
396   // compaction queues still have space in them. We try to re-compact these regions
397   // in serial to avoid a premature OOM when the mutator wants to allocate the first
398   // eden region after gc.
399 
400   // For maximum compaction, we need to re-prepare all objects above the lowest
401   // region among the current regions for all thread compaction points. It may
402   // happen that due to the uneven distribution of objects to parallel threads, holes
403   // have been created as threads compact to different target regions between the
404   // lowest and the highest region in the tails of the compaction points.
405 
406   uint start_serial = truncate_parallel_cps();
407   assert(start_serial < _heap->max_reserved_regions(), "Called on empty parallel compaction queues");
408 
409   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
410   assert(!serial_cp->is_initialized(), "sanity!");
411 
412   HeapRegion* start_hr = _heap->region_at(start_serial);
413   serial_cp->add(start_hr);
414   serial_cp->initialize(start_hr);
415 
416   HeapWord* dense_prefix_top = compaction_top(start_hr);
417   G1SerialRePrepareClosure re_prepare(serial_cp, dense_prefix_top);
418 
419   for (uint i = start_serial + 1; i < _heap->max_reserved_regions(); i++) {
420     if (is_compaction_target(i)) {
421       HeapRegion* current = _heap->region_at(i);
422       set_compaction_top(current, current->bottom());
423       serial_cp->add(current);
424       current->apply_to_marked_objects(mark_bitmap(), &re_prepare);
425     }
426   }
427   serial_cp->update();
428 }
429 
430 void G1FullCollector::phase2d_prepare_humongous_compaction() {
431   GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare humongous compaction", scope()->timer());
432   G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
433   assert(serial_cp->has_regions(), "Sanity!" );
434 
435   uint last_serial_target = serial_cp->current_region()->hrm_index();
436   uint region_index = last_serial_target + 1;
437   uint max_reserved_regions = _heap->max_reserved_regions();
438 
439   G1FullGCCompactionPoint* humongous_cp = humongous_compaction_point();
440 
441   while (region_index < max_reserved_regions) {
442     HeapRegion* hr = _heap->region_at_or_null(region_index);
443 
444     if (hr == nullptr) {
445       region_index++;
446       continue;
447     } else if (hr->is_starts_humongous()) {
448       uint num_regions = humongous_cp->forward_humongous(hr);
449       region_index += num_regions; // Skip over the continues humongous regions.
450       continue;
451     } else if (is_compaction_target(region_index)) {
452       // Add the region to the humongous compaction point.
453       humongous_cp->add(hr);
454     }
455     region_index++;
456   }
457 }
458 
459 void G1FullCollector::phase3_adjust_pointers() {
460   // Adjust the pointers to reflect the new locations
461   GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer());
462 
463   G1FullGCAdjustTask task(this);
464   run_task(&task);
465 }
466 
467 void G1FullCollector::phase4_do_compaction() {
468   // Compact the heap using the compaction queues created in phase 2.
469   GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer());
470   G1FullGCCompactTask task(this);
471   run_task(&task);
472 
473   // Serial compact to avoid OOM when very few free regions.
474   if (serial_compaction_point()->has_regions()) {
475     task.serial_compaction();
476   }
477 
478   if (!_humongous_compaction_regions.is_empty()) {
479     assert(scope()->do_maximal_compaction(), "Only compact humongous during maximal compaction");
480     task.humongous_compaction();
481   }
482 }
483 
484 void G1FullCollector::phase5_reset_metadata() {
485   // Clear region metadata that is invalid after GC for all regions.
486   GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer());
487   G1FullGCResetMetadataTask task(this);
488   run_task(&task);
489 }
490 
491 void G1FullCollector::restore_marks() {
492   _preserved_marks_set.restore(_heap->workers());
493   _preserved_marks_set.reclaim();
494 }
495 
496 void G1FullCollector::run_task(WorkerTask* task) {
497   _heap->workers()->run_task(task, _num_workers);
498 }
499 
500 void G1FullCollector::verify_after_marking() {
501   if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) {
502     // Only do verification if VerifyDuringGC and G1VerifyFull is set.
503     return;
504   }
505 
506 #if COMPILER2_OR_JVMCI
507   DerivedPointerTableDeactivate dpt_deact;
508 #endif
509   _heap->prepare_for_verify();
510   // Note: we can verify only the heap here. When an object is
511   // marked, the previous value of the mark word (including
512   // identity hash values, ages, etc) is preserved, and the mark
513   // word is set to markWord::marked_value - effectively removing
514   // any hash values from the mark word. These hash values are
515   // used when verifying the dictionaries and so removing them
516   // from the mark word can make verification of the dictionaries
517   // fail. At the end of the GC, the original mark word values
518   // (including hash values) are restored to the appropriate
519   // objects.
520   GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)");
521   _heap->verify(VerifyOption::G1UseFullMarking);
522 }