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/slidingForwarding.hpp"
44 #include "gc/shared/verifyOption.hpp"
45 #include "gc/shared/weakProcessor.inline.hpp"
46 #include "gc/shared/workerPolicy.hpp"
47 #include "logging/log.hpp"
48 #include "runtime/handles.inline.hpp"
49 #include "utilities/debug.hpp"
50
51 static void clear_and_activate_derived_pointers() {
52 #if COMPILER2_OR_JVMCI
53 DerivedPointerTable::clear();
54 #endif
55 }
56
57 static void deactivate_derived_pointers() {
58 #if COMPILER2_OR_JVMCI
59 DerivedPointerTable::set_active(false);
60 #endif
61 }
62
63 static void update_derived_pointers() {
64 #if COMPILER2_OR_JVMCI
65 DerivedPointerTable::update_pointers();
66 #endif
67 }
68
69 G1CMBitMap* G1FullCollector::mark_bitmap() {
70 return _heap->concurrent_mark()->mark_bitmap();
71 }
72
73 ReferenceProcessor* G1FullCollector::reference_processor() {
74 return _heap->ref_processor_stw();
75 }
76
77 uint G1FullCollector::calc_active_workers() {
78 G1CollectedHeap* heap = G1CollectedHeap::heap();
79 uint max_worker_count = heap->workers()->max_workers();
80 // Only calculate number of workers if UseDynamicNumberOfGCThreads
81 // is enabled, otherwise use max.
82 if (!UseDynamicNumberOfGCThreads) {
83 return max_worker_count;
84 }
85
86 // Consider G1HeapWastePercent to decide max number of workers. Each worker
87 // will in average cause half a region waste.
88 uint max_wasted_regions_allowed = ((heap->num_regions() * G1HeapWastePercent) / 100);
89 uint waste_worker_count = MAX2((max_wasted_regions_allowed * 2) , 1u);
90 uint heap_waste_worker_limit = MIN2(waste_worker_count, max_worker_count);
91
92 // Also consider HeapSizePerGCThread by calling WorkerPolicy to calculate
93 // the number of workers.
94 uint current_active_workers = heap->workers()->active_workers();
95 uint active_worker_limit = WorkerPolicy::calc_active_workers(max_worker_count, current_active_workers, 0);
96
97 // Finally consider the amount of used regions.
98 uint used_worker_limit = heap->num_used_regions();
99 assert(used_worker_limit > 0, "Should never have zero used regions.");
100
101 // Update active workers to the lower of the limits.
102 uint worker_count = MIN3(heap_waste_worker_limit, active_worker_limit, used_worker_limit);
103 log_debug(gc, task)("Requesting %u active workers for full compaction (waste limited workers: %u, "
104 "adaptive workers: %u, used limited workers: %u)",
105 worker_count, heap_waste_worker_limit, active_worker_limit, used_worker_limit);
106 worker_count = heap->workers()->set_active_workers(worker_count);
107 log_info(gc, task)("Using %u workers of %u for full compaction", worker_count, max_worker_count);
108
109 return worker_count;
110 }
111
112 G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
113 bool clear_soft_refs,
114 bool do_maximal_compaction,
115 G1FullGCTracer* tracer) :
116 _heap(heap),
117 _scope(heap->monitoring_support(), clear_soft_refs, do_maximal_compaction, tracer),
118 _num_workers(calc_active_workers()),
119 _has_compaction_targets(false),
120 _has_humongous(false),
121 _oop_queue_set(_num_workers),
122 _array_queue_set(_num_workers),
123 _preserved_marks_set(true),
124 _serial_compaction_point(this),
125 _humongous_compaction_point(this),
126 _is_alive(this, heap->concurrent_mark()->mark_bitmap()),
127 _is_alive_mutator(heap->ref_processor_stw(), &_is_alive),
128 _humongous_compaction_regions(8),
129 _always_subject_to_discovery(),
130 _is_subject_mutator(heap->ref_processor_stw(), &_always_subject_to_discovery),
131 _region_attr_table() {
132 assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint");
133
134 _preserved_marks_set.init(_num_workers);
135 _markers = NEW_C_HEAP_ARRAY(G1FullGCMarker*, _num_workers, mtGC);
136 _compaction_points = NEW_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _num_workers, mtGC);
137
138 _live_stats = NEW_C_HEAP_ARRAY(G1RegionMarkStats, _heap->max_regions(), mtGC);
139 _compaction_tops = NEW_C_HEAP_ARRAY(HeapWord*, _heap->max_regions(), mtGC);
140 for (uint j = 0; j < heap->max_regions(); j++) {
141 _live_stats[j].clear();
142 _compaction_tops[j] = nullptr;
143 }
144
145 for (uint i = 0; i < _num_workers; i++) {
146 _markers[i] = new G1FullGCMarker(this, i, _preserved_marks_set.get(i), _live_stats);
147 _compaction_points[i] = new G1FullGCCompactionPoint(this);
148 _oop_queue_set.register_queue(i, marker(i)->oop_stack());
149 _array_queue_set.register_queue(i, marker(i)->objarray_stack());
150 }
151 _region_attr_table.initialize(heap->reserved(), HeapRegion::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 HeapRegionClosure {
167 G1FullCollector* _collector;
168
169 public:
170 PrepareRegionsClosure(G1FullCollector* collector) : _collector(collector) { }
171
172 bool do_heap_region(HeapRegion* 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->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 SlidingForwarding::begin();
214
215 phase2_prepare_compaction();
216
217 if (has_compaction_targets()) {
218 phase3_adjust_pointers();
219
220 phase4_do_compaction();
221 } else {
222 // All regions have a high live ratio thus will not be compacted.
223 // The live ratio is only considered if do_maximal_compaction is false.
224 log_info(gc, phases) ("No Regions selected for compaction. Skipping Phase 3: Adjust pointers and Phase 4: Compact heap");
225 }
226
227 SlidingForwarding::end();
228
229 phase5_reset_metadata();
230
231 G1CollectedHeap::finish_codecache_marking_cycle();
232 }
233
234 void G1FullCollector::complete_collection() {
235 // Restore all marks.
236 restore_marks();
237
238 // When the pointers have been adjusted and moved, we can
239 // update the derived pointer table.
240 update_derived_pointers();
241
242 // Need completely cleared claim bits for the next concurrent marking or full gc.
243 ClassLoaderDataGraph::clear_claimed_marks();
244
245 // Prepare the bitmap for the next (potentially concurrent) marking.
246 _heap->concurrent_mark()->clear_bitmap(_heap->workers());
247
248 _heap->prepare_for_mutator_after_full_collection();
249
250 _heap->resize_all_tlabs();
251
252 _heap->policy()->record_full_collection_end();
253 _heap->gc_epilogue(true);
254
255 _heap->verify_after_full_collection();
256
257 _heap->print_heap_after_full_collection();
258 }
259
260 void G1FullCollector::before_marking_update_attribute_table(HeapRegion* hr) {
261 if (hr->is_free()) {
262 _region_attr_table.set_free(hr->hrm_index());
263 } else if (hr->is_humongous()) {
264 // Humongous objects will never be moved in the "main" compaction phase, but
265 // afterwards in a special phase if needed.
266 _region_attr_table.set_skip_compacting(hr->hrm_index());
267 } else {
268 // Everything else should be compacted.
269 _region_attr_table.set_compacting(hr->hrm_index());
270 }
271 }
272
273 class G1FullGCRefProcProxyTask : public RefProcProxyTask {
274 G1FullCollector& _collector;
275
276 public:
277 G1FullGCRefProcProxyTask(G1FullCollector &collector, uint max_workers)
278 : RefProcProxyTask("G1FullGCRefProcProxyTask", max_workers),
279 _collector(collector) {}
280
281 void work(uint worker_id) override {
282 assert(worker_id < _max_workers, "sanity");
283 G1IsAliveClosure is_alive(&_collector);
284 uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
285 G1FullKeepAliveClosure keep_alive(_collector.marker(index));
286 BarrierEnqueueDiscoveredFieldClosure enqueue;
287 G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure();
288 _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc);
289 }
290 };
291
292 void G1FullCollector::phase1_mark_live_objects() {
293 // Recursively traverse all live objects and mark them.
294 GCTraceTime(Info, gc, phases) info("Phase 1: Mark live objects", scope()->timer());
295
296 {
297 // Do the actual marking.
298 G1FullGCMarkTask marking_task(this);
299 run_task(&marking_task);
300 }
301
302 {
303 uint old_active_mt_degree = reference_processor()->num_queues();
304 reference_processor()->set_active_mt_degree(workers());
305 GCTraceTime(Debug, gc, phases) debug("Phase 1: Reference Processing", scope()->timer());
306 // Process reference objects found during marking.
307 ReferenceProcessorPhaseTimes pt(scope()->timer(), reference_processor()->max_num_queues());
308 G1FullGCRefProcProxyTask task(*this, reference_processor()->max_num_queues());
309 const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, pt);
310 scope()->tracer()->report_gc_reference_stats(stats);
311 pt.print_all_references();
312 assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack");
313
314 reference_processor()->set_active_mt_degree(old_active_mt_degree);
315 }
316
317 // Weak oops cleanup.
318 {
319 GCTraceTime(Debug, gc, phases) debug("Phase 1: Weak Processing", scope()->timer());
320 WeakProcessor::weak_oops_do(_heap->workers(), &_is_alive, &do_nothing_cl, 1);
321 }
322
323 // Class unloading and cleanup.
324 if (ClassUnloading) {
325 _heap->unload_classes_and_code("Phase 1: Class Unloading and Cleanup", &_is_alive, scope()->timer());
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 template <bool ALT_FWD>
398 void G1FullCollector::phase2c_prepare_serial_compaction_impl() {
399 GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare serial compaction", scope()->timer());
400 // At this point, we know that after parallel compaction there will be regions that
401 // are partially compacted into. Thus, the last compaction region of all
402 // compaction queues still have space in them. We try to re-compact these regions
403 // in serial to avoid a premature OOM when the mutator wants to allocate the first
404 // eden region after gc.
405
406 // For maximum compaction, we need to re-prepare all objects above the lowest
407 // region among the current regions for all thread compaction points. It may
408 // happen that due to the uneven distribution of objects to parallel threads, holes
409 // have been created as threads compact to different target regions between the
410 // lowest and the highest region in the tails of the compaction points.
411
412 uint start_serial = truncate_parallel_cps();
413 assert(start_serial < _heap->max_reserved_regions(), "Called on empty parallel compaction queues");
414
415 G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
416 assert(!serial_cp->is_initialized(), "sanity!");
417
418 HeapRegion* start_hr = _heap->region_at(start_serial);
419 serial_cp->add(start_hr);
420 serial_cp->initialize(start_hr);
421
422 HeapWord* dense_prefix_top = compaction_top(start_hr);
423 G1SerialRePrepareClosure<ALT_FWD> re_prepare(serial_cp, dense_prefix_top);
424
425 for (uint i = start_serial + 1; i < _heap->max_reserved_regions(); i++) {
426 if (is_compaction_target(i)) {
427 HeapRegion* current = _heap->region_at(i);
428 set_compaction_top(current, current->bottom());
429 serial_cp->add(current);
430 current->apply_to_marked_objects(mark_bitmap(), &re_prepare);
431 }
432 }
433 serial_cp->update();
434 }
435
436 void G1FullCollector::phase2c_prepare_serial_compaction() {
437 if (UseAltGCForwarding) {
438 phase2c_prepare_serial_compaction_impl<true>();
439 } else {
440 phase2c_prepare_serial_compaction_impl<false>();
441 }
442 }
443
444 template <bool ALT_FWD>
445 void G1FullCollector::phase2d_prepare_humongous_compaction_impl() {
446 GCTraceTime(Debug, gc, phases) debug("Phase 2: Prepare humongous compaction", scope()->timer());
447 G1FullGCCompactionPoint* serial_cp = serial_compaction_point();
448 assert(serial_cp->has_regions(), "Sanity!" );
449
450 uint last_serial_target = serial_cp->current_region()->hrm_index();
451 uint region_index = last_serial_target + 1;
452 uint max_reserved_regions = _heap->max_reserved_regions();
453
454 G1FullGCCompactionPoint* humongous_cp = humongous_compaction_point();
455
456 while (region_index < max_reserved_regions) {
457 HeapRegion* hr = _heap->region_at_or_null(region_index);
458
459 if (hr == nullptr) {
460 region_index++;
461 continue;
462 } else if (hr->is_starts_humongous()) {
463 uint num_regions = humongous_cp->forward_humongous<ALT_FWD>(hr);
464 region_index += num_regions; // Skip over the continues humongous regions.
465 continue;
466 } else if (is_compaction_target(region_index)) {
467 // Add the region to the humongous compaction point.
468 humongous_cp->add(hr);
469 }
470 region_index++;
471 }
472 }
473
474 void G1FullCollector::phase2d_prepare_humongous_compaction() {
475 if (UseAltGCForwarding) {
476 phase2d_prepare_humongous_compaction_impl<true>();
477 } else {
478 phase2d_prepare_humongous_compaction_impl<false>();
479 }
480 }
481
482 void G1FullCollector::phase3_adjust_pointers() {
483 // Adjust the pointers to reflect the new locations
484 GCTraceTime(Info, gc, phases) info("Phase 3: Adjust pointers", scope()->timer());
485
486 G1FullGCAdjustTask task(this);
487 run_task(&task);
488 }
489
490 void G1FullCollector::phase4_do_compaction() {
491 // Compact the heap using the compaction queues created in phase 2.
492 GCTraceTime(Info, gc, phases) info("Phase 4: Compact heap", scope()->timer());
493 G1FullGCCompactTask task(this);
494 run_task(&task);
495
496 // Serial compact to avoid OOM when very few free regions.
497 if (serial_compaction_point()->has_regions()) {
498 task.serial_compaction();
499 }
500
501 if (!_humongous_compaction_regions.is_empty()) {
502 assert(scope()->do_maximal_compaction(), "Only compact humongous during maximal compaction");
503 task.humongous_compaction();
504 }
505 }
506
507 void G1FullCollector::phase5_reset_metadata() {
508 // Clear region metadata that is invalid after GC for all regions.
509 GCTraceTime(Info, gc, phases) info("Phase 5: Reset Metadata", scope()->timer());
510 G1FullGCResetMetadataTask task(this);
511 run_task(&task);
512 }
513
514 void G1FullCollector::restore_marks() {
515 _preserved_marks_set.restore(_heap->workers());
516 _preserved_marks_set.reclaim();
517 }
518
519 void G1FullCollector::run_task(WorkerTask* task) {
520 _heap->workers()->run_task(task, _num_workers);
521 }
522
523 void G1FullCollector::verify_after_marking() {
524 if (!VerifyDuringGC || !_heap->verifier()->should_verify(G1HeapVerifier::G1VerifyFull)) {
525 // Only do verification if VerifyDuringGC and G1VerifyFull is set.
526 return;
527 }
528
529 #if COMPILER2_OR_JVMCI
530 DerivedPointerTableDeactivate dpt_deact;
531 #endif
532 _heap->prepare_for_verify();
533 // Note: we can verify only the heap here. When an object is
534 // marked, the previous value of the mark word (including
535 // identity hash values, ages, etc) is preserved, and the mark
536 // word is set to markWord::marked_value - effectively removing
537 // any hash values from the mark word. These hash values are
538 // used when verifying the dictionaries and so removing them
539 // from the mark word can make verification of the dictionaries
540 // fail. At the end of the GC, the original mark word values
541 // (including hash values) are restored to the appropriate
542 // objects.
543 GCTraceTime(Info, gc, verify) tm("Verifying During GC (full)");
544 _heap->verify(VerifyOption::G1UseFullMarking);
545 }