1 /*
2 * Copyright (c) 2001, 2019, 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 "code/codeCache.hpp"
28 #include "gc/g1/g1BarrierSet.hpp"
29 #include "gc/g1/g1CollectedHeap.inline.hpp"
30 #include "gc/g1/g1CollectorState.hpp"
31 #include "gc/g1/g1ConcurrentMark.inline.hpp"
32 #include "gc/g1/g1ConcurrentMarkThread.inline.hpp"
33 #include "gc/g1/g1DirtyCardQueue.hpp"
34 #include "gc/g1/g1HeapVerifier.hpp"
35 #include "gc/g1/g1OopClosures.inline.hpp"
36 #include "gc/g1/g1Policy.hpp"
37 #include "gc/g1/g1RegionMarkStatsCache.inline.hpp"
38 #include "gc/g1/g1StringDedup.hpp"
39 #include "gc/g1/g1ThreadLocalData.hpp"
40 #include "gc/g1/g1Trace.hpp"
41 #include "gc/g1/heapRegion.inline.hpp"
42 #include "gc/g1/heapRegionRemSet.hpp"
43 #include "gc/g1/heapRegionSet.inline.hpp"
44 #include "gc/shared/gcId.hpp"
45 #include "gc/shared/gcTimer.hpp"
46 #include "gc/shared/gcTraceTime.inline.hpp"
47 #include "gc/shared/gcVMOperations.hpp"
48 #include "gc/shared/genOopClosures.inline.hpp"
49 #include "gc/shared/referencePolicy.hpp"
50 #include "gc/shared/strongRootsScope.hpp"
51 #include "gc/shared/suspendibleThreadSet.hpp"
52 #include "gc/shared/taskqueue.inline.hpp"
53 #include "gc/shared/weakProcessor.inline.hpp"
54 #include "gc/shared/workerPolicy.hpp"
55 #include "include/jvm.h"
56 #include "logging/log.hpp"
57 #include "memory/allocation.hpp"
58 #include "memory/resourceArea.hpp"
59 #include "memory/universe.hpp"
60 #include "oops/access.inline.hpp"
61 #include "oops/oop.inline.hpp"
62 #include "runtime/atomic.hpp"
63 #include "runtime/handles.inline.hpp"
64 #include "runtime/mutexLocker.inline.hpp"
65 #include "runtime/java.hpp"
66 #include "runtime/prefetch.inline.hpp"
67 #include "services/memTracker.hpp"
68 #include "utilities/align.hpp"
69 #include "utilities/growableArray.hpp"
70
71 bool G1CMBitMapClosure::do_addr(HeapWord* const addr) {
72 assert(addr < _cm->finger(), "invariant");
73 assert(addr >= _task->finger(), "invariant");
74
75 // We move that task's local finger along.
76 _task->move_finger_to(addr);
77
78 _task->scan_task_entry(G1TaskQueueEntry::from_oop(oop(addr)));
79 // we only partially drain the local queue and global stack
80 _task->drain_local_queue(true);
81 _task->drain_global_stack(true);
82
83 // if the has_aborted flag has been raised, we need to bail out of
84 // the iteration
85 return !_task->has_aborted();
86 }
87
88 G1CMMarkStack::G1CMMarkStack() :
89 _max_chunk_capacity(0),
90 _base(NULL),
91 _chunk_capacity(0) {
92 set_empty();
93 }
94
95 bool G1CMMarkStack::resize(size_t new_capacity) {
96 assert(is_empty(), "Only resize when stack is empty.");
97 assert(new_capacity <= _max_chunk_capacity,
98 "Trying to resize stack to " SIZE_FORMAT " chunks when the maximum is " SIZE_FORMAT, new_capacity, _max_chunk_capacity);
99
100 TaskQueueEntryChunk* new_base = MmapArrayAllocator<TaskQueueEntryChunk>::allocate_or_null(new_capacity, mtGC);
101
102 if (new_base == NULL) {
103 log_warning(gc)("Failed to reserve memory for new overflow mark stack with " SIZE_FORMAT " chunks and size " SIZE_FORMAT "B.", new_capacity, new_capacity * sizeof(TaskQueueEntryChunk));
104 return false;
105 }
106 // Release old mapping.
107 if (_base != NULL) {
108 MmapArrayAllocator<TaskQueueEntryChunk>::free(_base, _chunk_capacity);
109 }
110
111 _base = new_base;
112 _chunk_capacity = new_capacity;
113 set_empty();
114
115 return true;
116 }
117
118 size_t G1CMMarkStack::capacity_alignment() {
119 return (size_t)lcm(os::vm_allocation_granularity(), sizeof(TaskQueueEntryChunk)) / sizeof(G1TaskQueueEntry);
120 }
121
122 bool G1CMMarkStack::initialize(size_t initial_capacity, size_t max_capacity) {
123 guarantee(_max_chunk_capacity == 0, "G1CMMarkStack already initialized.");
124
125 size_t const TaskEntryChunkSizeInVoidStar = sizeof(TaskQueueEntryChunk) / sizeof(G1TaskQueueEntry);
126
127 _max_chunk_capacity = align_up(max_capacity, capacity_alignment()) / TaskEntryChunkSizeInVoidStar;
128 size_t initial_chunk_capacity = align_up(initial_capacity, capacity_alignment()) / TaskEntryChunkSizeInVoidStar;
129
130 guarantee(initial_chunk_capacity <= _max_chunk_capacity,
131 "Maximum chunk capacity " SIZE_FORMAT " smaller than initial capacity " SIZE_FORMAT,
132 _max_chunk_capacity,
133 initial_chunk_capacity);
134
135 log_debug(gc)("Initialize mark stack with " SIZE_FORMAT " chunks, maximum " SIZE_FORMAT,
136 initial_chunk_capacity, _max_chunk_capacity);
137
138 return resize(initial_chunk_capacity);
139 }
140
141 void G1CMMarkStack::expand() {
142 if (_chunk_capacity == _max_chunk_capacity) {
143 log_debug(gc)("Can not expand overflow mark stack further, already at maximum capacity of " SIZE_FORMAT " chunks.", _chunk_capacity);
144 return;
145 }
146 size_t old_capacity = _chunk_capacity;
147 // Double capacity if possible
148 size_t new_capacity = MIN2(old_capacity * 2, _max_chunk_capacity);
149
150 if (resize(new_capacity)) {
151 log_debug(gc)("Expanded mark stack capacity from " SIZE_FORMAT " to " SIZE_FORMAT " chunks",
152 old_capacity, new_capacity);
153 } else {
154 log_warning(gc)("Failed to expand mark stack capacity from " SIZE_FORMAT " to " SIZE_FORMAT " chunks",
155 old_capacity, new_capacity);
156 }
157 }
158
159 G1CMMarkStack::~G1CMMarkStack() {
160 if (_base != NULL) {
161 MmapArrayAllocator<TaskQueueEntryChunk>::free(_base, _chunk_capacity);
162 }
163 }
164
165 void G1CMMarkStack::add_chunk_to_list(TaskQueueEntryChunk* volatile* list, TaskQueueEntryChunk* elem) {
166 elem->next = *list;
167 *list = elem;
168 }
169
170 void G1CMMarkStack::add_chunk_to_chunk_list(TaskQueueEntryChunk* elem) {
171 MutexLocker x(MarkStackChunkList_lock, Mutex::_no_safepoint_check_flag);
172 add_chunk_to_list(&_chunk_list, elem);
173 _chunks_in_chunk_list++;
174 }
175
176 void G1CMMarkStack::add_chunk_to_free_list(TaskQueueEntryChunk* elem) {
177 MutexLocker x(MarkStackFreeList_lock, Mutex::_no_safepoint_check_flag);
178 add_chunk_to_list(&_free_list, elem);
179 }
180
181 G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::remove_chunk_from_list(TaskQueueEntryChunk* volatile* list) {
182 TaskQueueEntryChunk* result = *list;
183 if (result != NULL) {
184 *list = (*list)->next;
185 }
186 return result;
187 }
188
189 G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::remove_chunk_from_chunk_list() {
190 MutexLocker x(MarkStackChunkList_lock, Mutex::_no_safepoint_check_flag);
191 TaskQueueEntryChunk* result = remove_chunk_from_list(&_chunk_list);
192 if (result != NULL) {
193 _chunks_in_chunk_list--;
194 }
195 return result;
196 }
197
198 G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::remove_chunk_from_free_list() {
199 MutexLocker x(MarkStackFreeList_lock, Mutex::_no_safepoint_check_flag);
200 return remove_chunk_from_list(&_free_list);
201 }
202
203 G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::allocate_new_chunk() {
204 // This dirty read of _hwm is okay because we only ever increase the _hwm in parallel code.
205 // Further this limits _hwm to a value of _chunk_capacity + #threads, avoiding
206 // wraparound of _hwm.
207 if (_hwm >= _chunk_capacity) {
208 return NULL;
209 }
210
211 size_t cur_idx = Atomic::add(1u, &_hwm) - 1;
212 if (cur_idx >= _chunk_capacity) {
213 return NULL;
214 }
215
216 TaskQueueEntryChunk* result = ::new (&_base[cur_idx]) TaskQueueEntryChunk;
217 result->next = NULL;
218 return result;
219 }
220
221 bool G1CMMarkStack::par_push_chunk(G1TaskQueueEntry* ptr_arr) {
222 // Get a new chunk.
223 TaskQueueEntryChunk* new_chunk = remove_chunk_from_free_list();
224
225 if (new_chunk == NULL) {
226 // Did not get a chunk from the free list. Allocate from backing memory.
227 new_chunk = allocate_new_chunk();
228
229 if (new_chunk == NULL) {
230 return false;
231 }
232 }
233
234 Copy::conjoint_memory_atomic(ptr_arr, new_chunk->data, EntriesPerChunk * sizeof(G1TaskQueueEntry));
235
236 add_chunk_to_chunk_list(new_chunk);
237
238 return true;
239 }
240
241 bool G1CMMarkStack::par_pop_chunk(G1TaskQueueEntry* ptr_arr) {
242 TaskQueueEntryChunk* cur = remove_chunk_from_chunk_list();
243
244 if (cur == NULL) {
245 return false;
246 }
247
248 Copy::conjoint_memory_atomic(cur->data, ptr_arr, EntriesPerChunk * sizeof(G1TaskQueueEntry));
249
250 add_chunk_to_free_list(cur);
251 return true;
252 }
253
254 void G1CMMarkStack::set_empty() {
255 _chunks_in_chunk_list = 0;
256 _hwm = 0;
257 _chunk_list = NULL;
258 _free_list = NULL;
259 }
260
261 G1CMRootMemRegions::G1CMRootMemRegions(uint const max_regions) :
262 _root_regions(NULL),
263 _max_regions(max_regions),
264 _num_root_regions(0),
265 _claimed_root_regions(0),
266 _scan_in_progress(false),
267 _should_abort(false) {
268 _root_regions = new MemRegion[_max_regions];
269 if (_root_regions == NULL) {
270 vm_exit_during_initialization("Could not allocate root MemRegion set.");
271 }
272 }
273
274 G1CMRootMemRegions::~G1CMRootMemRegions() {
275 delete[] _root_regions;
276 }
277
278 void G1CMRootMemRegions::reset() {
279 _num_root_regions = 0;
280 }
281
282 void G1CMRootMemRegions::add(HeapWord* start, HeapWord* end) {
283 assert_at_safepoint();
284 size_t idx = Atomic::add((size_t)1, &_num_root_regions) - 1;
285 assert(idx < _max_regions, "Trying to add more root MemRegions than there is space " SIZE_FORMAT, _max_regions);
286 assert(start != NULL && end != NULL && start <= end, "Start (" PTR_FORMAT ") should be less or equal to "
287 "end (" PTR_FORMAT ")", p2i(start), p2i(end));
288 _root_regions[idx].set_start(start);
289 _root_regions[idx].set_end(end);
290 }
291
292 void G1CMRootMemRegions::prepare_for_scan() {
293 assert(!scan_in_progress(), "pre-condition");
294
295 _scan_in_progress = _num_root_regions > 0;
296
297 _claimed_root_regions = 0;
298 _should_abort = false;
299 }
300
301 const MemRegion* G1CMRootMemRegions::claim_next() {
302 if (_should_abort) {
303 // If someone has set the should_abort flag, we return NULL to
304 // force the caller to bail out of their loop.
305 return NULL;
306 }
307
308 if (_claimed_root_regions >= _num_root_regions) {
309 return NULL;
310 }
311
312 size_t claimed_index = Atomic::add((size_t)1, &_claimed_root_regions) - 1;
313 if (claimed_index < _num_root_regions) {
314 return &_root_regions[claimed_index];
315 }
316 return NULL;
317 }
318
319 uint G1CMRootMemRegions::num_root_regions() const {
320 return (uint)_num_root_regions;
321 }
322
323 void G1CMRootMemRegions::notify_scan_done() {
324 MutexLocker x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
325 _scan_in_progress = false;
326 RootRegionScan_lock->notify_all();
327 }
328
329 void G1CMRootMemRegions::cancel_scan() {
330 notify_scan_done();
331 }
332
333 void G1CMRootMemRegions::scan_finished() {
334 assert(scan_in_progress(), "pre-condition");
335
336 if (!_should_abort) {
337 assert(_claimed_root_regions >= num_root_regions(),
338 "we should have claimed all root regions, claimed " SIZE_FORMAT ", length = %u",
339 _claimed_root_regions, num_root_regions());
340 }
341
342 notify_scan_done();
343 }
344
345 bool G1CMRootMemRegions::wait_until_scan_finished() {
346 if (!scan_in_progress()) {
347 return false;
348 }
349
350 {
351 MonitorLocker ml(RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
352 while (scan_in_progress()) {
353 ml.wait();
354 }
355 }
356 return true;
357 }
358
359 // Returns the maximum number of workers to be used in a concurrent
360 // phase based on the number of GC workers being used in a STW
361 // phase.
362 static uint scale_concurrent_worker_threads(uint num_gc_workers) {
363 return MAX2((num_gc_workers + 2) / 4, 1U);
364 }
365
366 G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
367 G1RegionToSpaceMapper* prev_bitmap_storage,
368 G1RegionToSpaceMapper* next_bitmap_storage) :
369 // _cm_thread set inside the constructor
370 _g1h(g1h),
371 _completed_initialization(false),
372
373 _mark_bitmap_1(),
374 _mark_bitmap_2(),
375 _prev_mark_bitmap(&_mark_bitmap_1),
376 _next_mark_bitmap(&_mark_bitmap_2),
377
378 _heap(_g1h->reserved_region()),
379
380 _root_regions(_g1h->max_regions()),
381
382 _global_mark_stack(),
383
384 // _finger set in set_non_marking_state
385
386 _worker_id_offset(G1DirtyCardQueueSet::num_par_ids() + G1ConcRefinementThreads),
387 _max_num_tasks(ParallelGCThreads),
388 // _num_active_tasks set in set_non_marking_state()
389 // _tasks set inside the constructor
390
391 _task_queues(new G1CMTaskQueueSet((int) _max_num_tasks)),
392 _terminator((int) _max_num_tasks, _task_queues),
393
394 _first_overflow_barrier_sync(),
395 _second_overflow_barrier_sync(),
396
397 _has_overflown(false),
398 _concurrent(false),
399 _has_aborted(false),
400 _restart_for_overflow(false),
401 _gc_timer_cm(new (ResourceObj::C_HEAP, mtGC) ConcurrentGCTimer()),
402 _gc_tracer_cm(new (ResourceObj::C_HEAP, mtGC) G1OldTracer()),
403
404 // _verbose_level set below
405
406 _init_times(),
407 _remark_times(),
408 _remark_mark_times(),
409 _remark_weak_ref_times(),
410 _cleanup_times(),
411 _total_cleanup_time(0.0),
412
413 _accum_task_vtime(NULL),
414
415 _concurrent_workers(NULL),
416 _num_concurrent_workers(0),
417 _max_concurrent_workers(0),
418
419 _region_mark_stats(NEW_C_HEAP_ARRAY(G1RegionMarkStats, _g1h->max_regions(), mtGC)),
420 _top_at_rebuild_starts(NEW_C_HEAP_ARRAY(HeapWord*, _g1h->max_regions(), mtGC))
421 {
422 _mark_bitmap_1.initialize(g1h->reserved_region(), prev_bitmap_storage);
423 _mark_bitmap_2.initialize(g1h->reserved_region(), next_bitmap_storage);
424
425 // Create & start ConcurrentMark thread.
426 _cm_thread = new G1ConcurrentMarkThread(this);
427 if (_cm_thread->osthread() == NULL) {
428 vm_shutdown_during_initialization("Could not create ConcurrentMarkThread");
429 }
430
431 assert(CGC_lock != NULL, "CGC_lock must be initialized");
432
433 if (FLAG_IS_DEFAULT(ConcGCThreads) || ConcGCThreads == 0) {
434 // Calculate the number of concurrent worker threads by scaling
435 // the number of parallel GC threads.
436 uint marking_thread_num = scale_concurrent_worker_threads(ParallelGCThreads);
437 FLAG_SET_ERGO(ConcGCThreads, marking_thread_num);
438 }
439
440 assert(ConcGCThreads > 0, "ConcGCThreads have been set.");
441 if (ConcGCThreads > ParallelGCThreads) {
442 log_warning(gc)("More ConcGCThreads (%u) than ParallelGCThreads (%u).",
443 ConcGCThreads, ParallelGCThreads);
444 return;
445 }
446
447 log_debug(gc)("ConcGCThreads: %u offset %u", ConcGCThreads, _worker_id_offset);
448 log_debug(gc)("ParallelGCThreads: %u", ParallelGCThreads);
449
450 _num_concurrent_workers = ConcGCThreads;
451 _max_concurrent_workers = _num_concurrent_workers;
452
453 _concurrent_workers = new WorkGang("G1 Conc", _max_concurrent_workers, false, true);
454 _concurrent_workers->initialize_workers();
455
456 if (FLAG_IS_DEFAULT(MarkStackSize)) {
457 size_t mark_stack_size =
458 MIN2(MarkStackSizeMax,
459 MAX2(MarkStackSize, (size_t) (_max_concurrent_workers * TASKQUEUE_SIZE)));
460 // Verify that the calculated value for MarkStackSize is in range.
461 // It would be nice to use the private utility routine from Arguments.
462 if (!(mark_stack_size >= 1 && mark_stack_size <= MarkStackSizeMax)) {
463 log_warning(gc)("Invalid value calculated for MarkStackSize (" SIZE_FORMAT "): "
464 "must be between 1 and " SIZE_FORMAT,
465 mark_stack_size, MarkStackSizeMax);
466 return;
467 }
468 FLAG_SET_ERGO(MarkStackSize, mark_stack_size);
469 } else {
470 // Verify MarkStackSize is in range.
471 if (FLAG_IS_CMDLINE(MarkStackSize)) {
472 if (FLAG_IS_DEFAULT(MarkStackSizeMax)) {
473 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) {
474 log_warning(gc)("Invalid value specified for MarkStackSize (" SIZE_FORMAT "): "
475 "must be between 1 and " SIZE_FORMAT,
476 MarkStackSize, MarkStackSizeMax);
477 return;
478 }
479 } else if (FLAG_IS_CMDLINE(MarkStackSizeMax)) {
480 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) {
481 log_warning(gc)("Invalid value specified for MarkStackSize (" SIZE_FORMAT ")"
482 " or for MarkStackSizeMax (" SIZE_FORMAT ")",
483 MarkStackSize, MarkStackSizeMax);
484 return;
485 }
486 }
487 }
488 }
489
490 if (!_global_mark_stack.initialize(MarkStackSize, MarkStackSizeMax)) {
491 vm_exit_during_initialization("Failed to allocate initial concurrent mark overflow mark stack.");
492 }
493
494 _tasks = NEW_C_HEAP_ARRAY(G1CMTask*, _max_num_tasks, mtGC);
495 _accum_task_vtime = NEW_C_HEAP_ARRAY(double, _max_num_tasks, mtGC);
496
497 // so that the assertion in MarkingTaskQueue::task_queue doesn't fail
498 _num_active_tasks = _max_num_tasks;
499
500 for (uint i = 0; i < _max_num_tasks; ++i) {
501 G1CMTaskQueue* task_queue = new G1CMTaskQueue();
502 task_queue->initialize();
503 _task_queues->register_queue(i, task_queue);
504
505 _tasks[i] = new G1CMTask(i, this, task_queue, _region_mark_stats, _g1h->max_regions());
506
507 _accum_task_vtime[i] = 0.0;
508 }
509
510 reset_at_marking_complete();
511 _completed_initialization = true;
512 }
513
514 void G1ConcurrentMark::reset() {
515 _has_aborted = false;
516
517 reset_marking_for_restart();
518
519 // Reset all tasks, since different phases will use different number of active
520 // threads. So, it's easiest to have all of them ready.
521 for (uint i = 0; i < _max_num_tasks; ++i) {
522 _tasks[i]->reset(_next_mark_bitmap);
523 }
524
525 uint max_regions = _g1h->max_regions();
526 for (uint i = 0; i < max_regions; i++) {
527 _top_at_rebuild_starts[i] = NULL;
528 _region_mark_stats[i].clear();
529 }
530 }
531
532 void G1ConcurrentMark::clear_statistics_in_region(uint region_idx) {
533 for (uint j = 0; j < _max_num_tasks; ++j) {
534 _tasks[j]->clear_mark_stats_cache(region_idx);
535 }
536 _top_at_rebuild_starts[region_idx] = NULL;
537 _region_mark_stats[region_idx].clear();
538 }
539
540 void G1ConcurrentMark::clear_statistics(HeapRegion* r) {
541 uint const region_idx = r->hrm_index();
542 if (r->is_humongous()) {
543 assert(r->is_starts_humongous(), "Got humongous continues region here");
544 uint const size_in_regions = (uint)_g1h->humongous_obj_size_in_regions(oop(r->humongous_start_region()->bottom())->size());
545 for (uint j = region_idx; j < (region_idx + size_in_regions); j++) {
546 clear_statistics_in_region(j);
547 }
548 } else {
549 clear_statistics_in_region(region_idx);
550 }
551 }
552
553 static void clear_mark_if_set(G1CMBitMap* bitmap, HeapWord* addr) {
554 if (bitmap->is_marked(addr)) {
555 bitmap->clear(addr);
556 }
557 }
558
559 void G1ConcurrentMark::humongous_object_eagerly_reclaimed(HeapRegion* r) {
560 assert_at_safepoint_on_vm_thread();
561
562 // Need to clear all mark bits of the humongous object.
563 clear_mark_if_set(_prev_mark_bitmap, r->bottom());
564 clear_mark_if_set(_next_mark_bitmap, r->bottom());
565
566 if (!_g1h->collector_state()->mark_or_rebuild_in_progress()) {
567 return;
568 }
569
570 // Clear any statistics about the region gathered so far.
571 clear_statistics(r);
572 }
573
574 void G1ConcurrentMark::reset_marking_for_restart() {
575 _global_mark_stack.set_empty();
576
577 // Expand the marking stack, if we have to and if we can.
578 if (has_overflown()) {
579 _global_mark_stack.expand();
580
581 uint max_regions = _g1h->max_regions();
582 for (uint i = 0; i < max_regions; i++) {
583 _region_mark_stats[i].clear_during_overflow();
584 }
585 }
586
587 clear_has_overflown();
588 _finger = _heap.start();
589
590 for (uint i = 0; i < _max_num_tasks; ++i) {
591 G1CMTaskQueue* queue = _task_queues->queue(i);
592 queue->set_empty();
593 }
594 }
595
596 void G1ConcurrentMark::set_concurrency(uint active_tasks) {
597 assert(active_tasks <= _max_num_tasks, "we should not have more");
598
599 _num_active_tasks = active_tasks;
600 // Need to update the three data structures below according to the
601 // number of active threads for this phase.
602 _terminator.terminator()->reset_for_reuse((int) active_tasks);
603 _first_overflow_barrier_sync.set_n_workers((int) active_tasks);
604 _second_overflow_barrier_sync.set_n_workers((int) active_tasks);
605 }
606
607 void G1ConcurrentMark::set_concurrency_and_phase(uint active_tasks, bool concurrent) {
608 set_concurrency(active_tasks);
609
610 _concurrent = concurrent;
611
612 if (!concurrent) {
613 // At this point we should be in a STW phase, and completed marking.
614 assert_at_safepoint_on_vm_thread();
615 assert(out_of_regions(),
616 "only way to get here: _finger: " PTR_FORMAT ", _heap_end: " PTR_FORMAT,
617 p2i(_finger), p2i(_heap.end()));
618 }
619 }
620
621 void G1ConcurrentMark::reset_at_marking_complete() {
622 // We set the global marking state to some default values when we're
623 // not doing marking.
624 reset_marking_for_restart();
625 _num_active_tasks = 0;
626 }
627
628 G1ConcurrentMark::~G1ConcurrentMark() {
629 FREE_C_HEAP_ARRAY(HeapWord*, _top_at_rebuild_starts);
630 FREE_C_HEAP_ARRAY(G1RegionMarkStats, _region_mark_stats);
631 // The G1ConcurrentMark instance is never freed.
632 ShouldNotReachHere();
633 }
634
635 class G1ClearBitMapTask : public AbstractGangTask {
636 public:
637 static size_t chunk_size() { return M; }
638
639 private:
640 // Heap region closure used for clearing the given mark bitmap.
641 class G1ClearBitmapHRClosure : public HeapRegionClosure {
642 private:
643 G1CMBitMap* _bitmap;
644 G1ConcurrentMark* _cm;
645 public:
646 G1ClearBitmapHRClosure(G1CMBitMap* bitmap, G1ConcurrentMark* cm) : HeapRegionClosure(), _bitmap(bitmap), _cm(cm) {
647 }
648
649 virtual bool do_heap_region(HeapRegion* r) {
650 size_t const chunk_size_in_words = G1ClearBitMapTask::chunk_size() / HeapWordSize;
651
652 HeapWord* cur = r->bottom();
653 HeapWord* const end = r->end();
654
655 while (cur < end) {
656 MemRegion mr(cur, MIN2(cur + chunk_size_in_words, end));
657 _bitmap->clear_range(mr);
658
659 cur += chunk_size_in_words;
660
661 // Abort iteration if after yielding the marking has been aborted.
662 if (_cm != NULL && _cm->do_yield_check() && _cm->has_aborted()) {
663 return true;
664 }
665 // Repeat the asserts from before the start of the closure. We will do them
666 // as asserts here to minimize their overhead on the product. However, we
667 // will have them as guarantees at the beginning / end of the bitmap
668 // clearing to get some checking in the product.
669 assert(_cm == NULL || _cm->cm_thread()->during_cycle(), "invariant");
670 assert(_cm == NULL || !G1CollectedHeap::heap()->collector_state()->mark_or_rebuild_in_progress(), "invariant");
671 }
672 assert(cur == end, "Must have completed iteration over the bitmap for region %u.", r->hrm_index());
673
674 return false;
675 }
676 };
677
678 G1ClearBitmapHRClosure _cl;
679 HeapRegionClaimer _hr_claimer;
680 bool _suspendible; // If the task is suspendible, workers must join the STS.
681
682 public:
683 G1ClearBitMapTask(G1CMBitMap* bitmap, G1ConcurrentMark* cm, uint n_workers, bool suspendible) :
684 AbstractGangTask("G1 Clear Bitmap"),
685 _cl(bitmap, suspendible ? cm : NULL),
686 _hr_claimer(n_workers),
687 _suspendible(suspendible)
688 { }
689
690 void work(uint worker_id) {
691 SuspendibleThreadSetJoiner sts_join(_suspendible);
692 G1CollectedHeap::heap()->heap_region_par_iterate_from_worker_offset(&_cl, &_hr_claimer, worker_id);
693 }
694
695 bool is_complete() {
696 return _cl.is_complete();
697 }
698 };
699
700 void G1ConcurrentMark::clear_bitmap(G1CMBitMap* bitmap, WorkGang* workers, bool may_yield) {
701 assert(may_yield || SafepointSynchronize::is_at_safepoint(), "Non-yielding bitmap clear only allowed at safepoint.");
702
703 size_t const num_bytes_to_clear = (HeapRegion::GrainBytes * _g1h->num_regions()) / G1CMBitMap::heap_map_factor();
704 size_t const num_chunks = align_up(num_bytes_to_clear, G1ClearBitMapTask::chunk_size()) / G1ClearBitMapTask::chunk_size();
705
706 uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers());
707
708 G1ClearBitMapTask cl(bitmap, this, num_workers, may_yield);
709
710 log_debug(gc, ergo)("Running %s with %u workers for " SIZE_FORMAT " work units.", cl.name(), num_workers, num_chunks);
711 workers->run_task(&cl, num_workers);
712 guarantee(!may_yield || cl.is_complete(), "Must have completed iteration when not yielding.");
713 }
714
715 void G1ConcurrentMark::cleanup_for_next_mark() {
716 // Make sure that the concurrent mark thread looks to still be in
717 // the current cycle.
718 guarantee(cm_thread()->during_cycle(), "invariant");
719
720 // We are finishing up the current cycle by clearing the next
721 // marking bitmap and getting it ready for the next cycle. During
722 // this time no other cycle can start. So, let's make sure that this
723 // is the case.
724 guarantee(!_g1h->collector_state()->mark_or_rebuild_in_progress(), "invariant");
725
726 clear_bitmap(_next_mark_bitmap, _concurrent_workers, true);
727
728 // Repeat the asserts from above.
729 guarantee(cm_thread()->during_cycle(), "invariant");
730 guarantee(!_g1h->collector_state()->mark_or_rebuild_in_progress(), "invariant");
731 }
732
733 void G1ConcurrentMark::clear_prev_bitmap(WorkGang* workers) {
734 assert_at_safepoint_on_vm_thread();
735 clear_bitmap(_prev_mark_bitmap, workers, false);
736 }
737
738 class NoteStartOfMarkHRClosure : public HeapRegionClosure {
739 public:
740 bool do_heap_region(HeapRegion* r) {
741 r->note_start_of_marking();
742 return false;
743 }
744 };
745
746 void G1ConcurrentMark::pre_initial_mark() {
747 assert_at_safepoint_on_vm_thread();
748
749 // Reset marking state.
750 reset();
751
752 // For each region note start of marking.
753 NoteStartOfMarkHRClosure startcl;
754 _g1h->heap_region_iterate(&startcl);
755
756 _root_regions.reset();
757 }
758
759
760 void G1ConcurrentMark::post_initial_mark() {
761 // Start Concurrent Marking weak-reference discovery.
762 ReferenceProcessor* rp = _g1h->ref_processor_cm();
763 // enable ("weak") refs discovery
764 rp->enable_discovery();
765 rp->setup_policy(false); // snapshot the soft ref policy to be used in this cycle
766
767 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
768 // This is the start of the marking cycle, we're expected all
769 // threads to have SATB queues with active set to false.
770 satb_mq_set.set_active_all_threads(true, /* new active value */
771 false /* expected_active */);
772
773 _root_regions.prepare_for_scan();
774
775 // update_g1_committed() will be called at the end of an evac pause
776 // when marking is on. So, it's also called at the end of the
777 // initial-mark pause to update the heap end, if the heap expands
778 // during it. No need to call it here.
779 }
780
781 /*
782 * Notice that in the next two methods, we actually leave the STS
783 * during the barrier sync and join it immediately afterwards. If we
784 * do not do this, the following deadlock can occur: one thread could
785 * be in the barrier sync code, waiting for the other thread to also
786 * sync up, whereas another one could be trying to yield, while also
787 * waiting for the other threads to sync up too.
788 *
789 * Note, however, that this code is also used during remark and in
790 * this case we should not attempt to leave / enter the STS, otherwise
791 * we'll either hit an assert (debug / fastdebug) or deadlock
792 * (product). So we should only leave / enter the STS if we are
793 * operating concurrently.
794 *
795 * Because the thread that does the sync barrier has left the STS, it
796 * is possible to be suspended for a Full GC or an evacuation pause
797 * could occur. This is actually safe, since the entering the sync
798 * barrier is one of the last things do_marking_step() does, and it
799 * doesn't manipulate any data structures afterwards.
800 */
801
802 void G1ConcurrentMark::enter_first_sync_barrier(uint worker_id) {
803 bool barrier_aborted;
804 {
805 SuspendibleThreadSetLeaver sts_leave(concurrent());
806 barrier_aborted = !_first_overflow_barrier_sync.enter();
807 }
808
809 // at this point everyone should have synced up and not be doing any
810 // more work
811
812 if (barrier_aborted) {
813 // If the barrier aborted we ignore the overflow condition and
814 // just abort the whole marking phase as quickly as possible.
815 return;
816 }
817 }
818
819 void G1ConcurrentMark::enter_second_sync_barrier(uint worker_id) {
820 SuspendibleThreadSetLeaver sts_leave(concurrent());
821 _second_overflow_barrier_sync.enter();
822
823 // at this point everything should be re-initialized and ready to go
824 }
825
826 class G1CMConcurrentMarkingTask : public AbstractGangTask {
827 G1ConcurrentMark* _cm;
828
829 public:
830 void work(uint worker_id) {
831 assert(Thread::current()->is_ConcurrentGC_thread(), "Not a concurrent GC thread");
832 ResourceMark rm;
833
834 double start_vtime = os::elapsedVTime();
835
836 {
837 SuspendibleThreadSetJoiner sts_join;
838
839 assert(worker_id < _cm->active_tasks(), "invariant");
840
841 G1CMTask* task = _cm->task(worker_id);
842 task->record_start_time();
843 if (!_cm->has_aborted()) {
844 do {
845 task->do_marking_step(G1ConcMarkStepDurationMillis,
846 true /* do_termination */,
847 false /* is_serial*/);
848
849 _cm->do_yield_check();
850 } while (!_cm->has_aborted() && task->has_aborted());
851 }
852 task->record_end_time();
853 guarantee(!task->has_aborted() || _cm->has_aborted(), "invariant");
854 }
855
856 double end_vtime = os::elapsedVTime();
857 _cm->update_accum_task_vtime(worker_id, end_vtime - start_vtime);
858 }
859
860 G1CMConcurrentMarkingTask(G1ConcurrentMark* cm) :
861 AbstractGangTask("Concurrent Mark"), _cm(cm) { }
862
863 ~G1CMConcurrentMarkingTask() { }
864 };
865
866 uint G1ConcurrentMark::calc_active_marking_workers() {
867 uint result = 0;
868 if (!UseDynamicNumberOfGCThreads ||
869 (!FLAG_IS_DEFAULT(ConcGCThreads) &&
870 !ForceDynamicNumberOfGCThreads)) {
871 result = _max_concurrent_workers;
872 } else {
873 result =
874 WorkerPolicy::calc_default_active_workers(_max_concurrent_workers,
875 1, /* Minimum workers */
876 _num_concurrent_workers,
877 Threads::number_of_non_daemon_threads());
878 // Don't scale the result down by scale_concurrent_workers() because
879 // that scaling has already gone into "_max_concurrent_workers".
880 }
881 assert(result > 0 && result <= _max_concurrent_workers,
882 "Calculated number of marking workers must be larger than zero and at most the maximum %u, but is %u",
883 _max_concurrent_workers, result);
884 return result;
885 }
886
887 void G1ConcurrentMark::scan_root_region(const MemRegion* region, uint worker_id) {
888 #ifdef ASSERT
889 HeapWord* last = region->last();
890 HeapRegion* hr = _g1h->heap_region_containing(last);
891 assert(hr->is_old() || hr->next_top_at_mark_start() == hr->bottom(),
892 "Root regions must be old or survivor/eden but region %u is %s", hr->hrm_index(), hr->get_type_str());
893 assert(hr->next_top_at_mark_start() == region->start(),
894 "MemRegion start should be equal to nTAMS");
895 #endif
896
897 G1RootRegionScanClosure cl(_g1h, this, worker_id);
898
899 const uintx interval = PrefetchScanIntervalInBytes;
900 HeapWord* curr = region->start();
901 const HeapWord* end = region->end();
902 while (curr < end) {
903 Prefetch::read(curr, interval);
904 oop obj = oop(curr);
905 int size = obj->oop_iterate_size(&cl);
906 assert(size == obj->size(), "sanity");
907 curr += size;
908 }
909 }
910
911 class G1CMRootRegionScanTask : public AbstractGangTask {
912 G1ConcurrentMark* _cm;
913 public:
914 G1CMRootRegionScanTask(G1ConcurrentMark* cm) :
915 AbstractGangTask("G1 Root Region Scan"), _cm(cm) { }
916
917 void work(uint worker_id) {
918 assert(Thread::current()->is_ConcurrentGC_thread(),
919 "this should only be done by a conc GC thread");
920
921 G1CMRootMemRegions* root_regions = _cm->root_regions();
922 const MemRegion* region = root_regions->claim_next();
923 while (region != NULL) {
924 _cm->scan_root_region(region, worker_id);
925 region = root_regions->claim_next();
926 }
927 }
928 };
929
930 void G1ConcurrentMark::scan_root_regions() {
931 // scan_in_progress() will have been set to true only if there was
932 // at least one root region to scan. So, if it's false, we
933 // should not attempt to do any further work.
934 if (root_regions()->scan_in_progress()) {
935 assert(!has_aborted(), "Aborting before root region scanning is finished not supported.");
936
937 _num_concurrent_workers = MIN2(calc_active_marking_workers(),
938 // We distribute work on a per-region basis, so starting
939 // more threads than that is useless.
940 root_regions()->num_root_regions());
941 assert(_num_concurrent_workers <= _max_concurrent_workers,
942 "Maximum number of marking threads exceeded");
943
944 G1CMRootRegionScanTask task(this);
945 log_debug(gc, ergo)("Running %s using %u workers for %u work units.",
946 task.name(), _num_concurrent_workers, root_regions()->num_root_regions());
947 _concurrent_workers->run_task(&task, _num_concurrent_workers);
948
949 // It's possible that has_aborted() is true here without actually
950 // aborting the survivor scan earlier. This is OK as it's
951 // mainly used for sanity checking.
952 root_regions()->scan_finished();
953 }
954 }
955
956 void G1ConcurrentMark::concurrent_cycle_start() {
957 _gc_timer_cm->register_gc_start();
958
959 _gc_tracer_cm->report_gc_start(GCCause::_no_gc /* first parameter is not used */, _gc_timer_cm->gc_start());
960
961 _g1h->trace_heap_before_gc(_gc_tracer_cm);
962 }
963
964 void G1ConcurrentMark::concurrent_cycle_end() {
965 _g1h->collector_state()->set_clearing_next_bitmap(false);
966
967 _g1h->trace_heap_after_gc(_gc_tracer_cm);
968
969 if (has_aborted()) {
970 log_info(gc, marking)("Concurrent Mark Abort");
971 _gc_tracer_cm->report_concurrent_mode_failure();
972 }
973
974 _gc_timer_cm->register_gc_end();
975
976 _gc_tracer_cm->report_gc_end(_gc_timer_cm->gc_end(), _gc_timer_cm->time_partitions());
977 }
978
979 void G1ConcurrentMark::mark_from_roots() {
980 _restart_for_overflow = false;
981
982 _num_concurrent_workers = calc_active_marking_workers();
983
984 uint active_workers = MAX2(1U, _num_concurrent_workers);
985
986 // Setting active workers is not guaranteed since fewer
987 // worker threads may currently exist and more may not be
988 // available.
989 active_workers = _concurrent_workers->update_active_workers(active_workers);
990 log_info(gc, task)("Using %u workers of %u for marking", active_workers, _concurrent_workers->total_workers());
991
992 // Parallel task terminator is set in "set_concurrency_and_phase()"
993 set_concurrency_and_phase(active_workers, true /* concurrent */);
994
995 G1CMConcurrentMarkingTask marking_task(this);
996 _concurrent_workers->run_task(&marking_task);
997 print_stats();
998 }
999
1000 void G1ConcurrentMark::verify_during_pause(G1HeapVerifier::G1VerifyType type, VerifyOption vo, const char* caller) {
1001 G1HeapVerifier* verifier = _g1h->verifier();
1002
1003 verifier->verify_region_sets_optional();
1004
1005 if (VerifyDuringGC) {
1006 GCTraceTime(Debug, gc, phases) debug(caller, _gc_timer_cm);
1007
1008 size_t const BufLen = 512;
1009 char buffer[BufLen];
1010
1011 jio_snprintf(buffer, BufLen, "During GC (%s)", caller);
1012 verifier->verify(type, vo, buffer);
1013 }
1014
1015 verifier->check_bitmaps(caller);
1016 }
1017
1018 class G1UpdateRemSetTrackingBeforeRebuildTask : public AbstractGangTask {
1019 G1CollectedHeap* _g1h;
1020 G1ConcurrentMark* _cm;
1021 HeapRegionClaimer _hrclaimer;
1022 uint volatile _total_selected_for_rebuild;
1023
1024 G1PrintRegionLivenessInfoClosure _cl;
1025
1026 class G1UpdateRemSetTrackingBeforeRebuild : public HeapRegionClosure {
1027 G1CollectedHeap* _g1h;
1028 G1ConcurrentMark* _cm;
1029
1030 G1PrintRegionLivenessInfoClosure* _cl;
1031
1032 uint _num_regions_selected_for_rebuild; // The number of regions actually selected for rebuild.
1033
1034 void update_remset_before_rebuild(HeapRegion* hr) {
1035 G1RemSetTrackingPolicy* tracking_policy = _g1h->policy()->remset_tracker();
1036
1037 bool selected_for_rebuild;
1038 if (hr->is_humongous()) {
1039 bool const is_live = _cm->liveness(hr->humongous_start_region()->hrm_index()) > 0;
1040 selected_for_rebuild = tracking_policy->update_humongous_before_rebuild(hr, is_live);
1041 } else {
1042 size_t const live_bytes = _cm->liveness(hr->hrm_index());
1043 selected_for_rebuild = tracking_policy->update_before_rebuild(hr, live_bytes);
1044 }
1045 if (selected_for_rebuild) {
1046 _num_regions_selected_for_rebuild++;
1047 }
1048 _cm->update_top_at_rebuild_start(hr);
1049 }
1050
1051 // Distribute the given words across the humongous object starting with hr and
1052 // note end of marking.
1053 void distribute_marked_bytes(HeapRegion* hr, size_t marked_words) {
1054 uint const region_idx = hr->hrm_index();
1055 size_t const obj_size_in_words = (size_t)oop(hr->bottom())->size();
1056 uint const num_regions_in_humongous = (uint)G1CollectedHeap::humongous_obj_size_in_regions(obj_size_in_words);
1057
1058 // "Distributing" zero words means that we only note end of marking for these
1059 // regions.
1060 assert(marked_words == 0 || obj_size_in_words == marked_words,
1061 "Marked words should either be 0 or the same as humongous object (" SIZE_FORMAT ") but is " SIZE_FORMAT,
1062 obj_size_in_words, marked_words);
1063
1064 for (uint i = region_idx; i < (region_idx + num_regions_in_humongous); i++) {
1065 HeapRegion* const r = _g1h->region_at(i);
1066 size_t const words_to_add = MIN2(HeapRegion::GrainWords, marked_words);
1067
1068 log_trace(gc, marking)("Adding " SIZE_FORMAT " words to humongous region %u (%s)",
1069 words_to_add, i, r->get_type_str());
1070 add_marked_bytes_and_note_end(r, words_to_add * HeapWordSize);
1071 marked_words -= words_to_add;
1072 }
1073 assert(marked_words == 0,
1074 SIZE_FORMAT " words left after distributing space across %u regions",
1075 marked_words, num_regions_in_humongous);
1076 }
1077
1078 void update_marked_bytes(HeapRegion* hr) {
1079 uint const region_idx = hr->hrm_index();
1080 size_t const marked_words = _cm->liveness(region_idx);
1081 // The marking attributes the object's size completely to the humongous starts
1082 // region. We need to distribute this value across the entire set of regions a
1083 // humongous object spans.
1084 if (hr->is_humongous()) {
1085 assert(hr->is_starts_humongous() || marked_words == 0,
1086 "Should not have marked words " SIZE_FORMAT " in non-starts humongous region %u (%s)",
1087 marked_words, region_idx, hr->get_type_str());
1088 if (hr->is_starts_humongous()) {
1089 distribute_marked_bytes(hr, marked_words);
1090 }
1091 } else {
1092 log_trace(gc, marking)("Adding " SIZE_FORMAT " words to region %u (%s)", marked_words, region_idx, hr->get_type_str());
1093 add_marked_bytes_and_note_end(hr, marked_words * HeapWordSize);
1094 }
1095 }
1096
1097 void add_marked_bytes_and_note_end(HeapRegion* hr, size_t marked_bytes) {
1098 hr->add_to_marked_bytes(marked_bytes);
1099 _cl->do_heap_region(hr);
1100 hr->note_end_of_marking();
1101 }
1102
1103 public:
1104 G1UpdateRemSetTrackingBeforeRebuild(G1CollectedHeap* g1h, G1ConcurrentMark* cm, G1PrintRegionLivenessInfoClosure* cl) :
1105 _g1h(g1h), _cm(cm), _cl(cl), _num_regions_selected_for_rebuild(0) { }
1106
1107 virtual bool do_heap_region(HeapRegion* r) {
1108 update_remset_before_rebuild(r);
1109 update_marked_bytes(r);
1110
1111 return false;
1112 }
1113
1114 uint num_selected_for_rebuild() const { return _num_regions_selected_for_rebuild; }
1115 };
1116
1117 public:
1118 G1UpdateRemSetTrackingBeforeRebuildTask(G1CollectedHeap* g1h, G1ConcurrentMark* cm, uint num_workers) :
1119 AbstractGangTask("G1 Update RemSet Tracking Before Rebuild"),
1120 _g1h(g1h), _cm(cm), _hrclaimer(num_workers), _total_selected_for_rebuild(0), _cl("Post-Marking") { }
1121
1122 virtual void work(uint worker_id) {
1123 G1UpdateRemSetTrackingBeforeRebuild update_cl(_g1h, _cm, &_cl);
1124 _g1h->heap_region_par_iterate_from_worker_offset(&update_cl, &_hrclaimer, worker_id);
1125 Atomic::add(update_cl.num_selected_for_rebuild(), &_total_selected_for_rebuild);
1126 }
1127
1128 uint total_selected_for_rebuild() const { return _total_selected_for_rebuild; }
1129
1130 // Number of regions for which roughly one thread should be spawned for this work.
1131 static const uint RegionsPerThread = 384;
1132 };
1133
1134 class G1UpdateRemSetTrackingAfterRebuild : public HeapRegionClosure {
1135 G1CollectedHeap* _g1h;
1136 public:
1137 G1UpdateRemSetTrackingAfterRebuild(G1CollectedHeap* g1h) : _g1h(g1h) { }
1138
1139 virtual bool do_heap_region(HeapRegion* r) {
1140 _g1h->policy()->remset_tracker()->update_after_rebuild(r);
1141 return false;
1142 }
1143 };
1144
1145 void G1ConcurrentMark::remark() {
1146 assert_at_safepoint_on_vm_thread();
1147
1148 // If a full collection has happened, we should not continue. However we might
1149 // have ended up here as the Remark VM operation has been scheduled already.
1150 if (has_aborted()) {
1151 return;
1152 }
1153
1154 G1Policy* policy = _g1h->policy();
1155 policy->record_concurrent_mark_remark_start();
1156
1157 double start = os::elapsedTime();
1158
1159 verify_during_pause(G1HeapVerifier::G1VerifyRemark, VerifyOption_G1UsePrevMarking, "Remark before");
1160
1161 {
1162 GCTraceTime(Debug, gc, phases) debug("Finalize Marking", _gc_timer_cm);
1163 finalize_marking();
1164 }
1165
1166 double mark_work_end = os::elapsedTime();
1167
1168 bool const mark_finished = !has_overflown();
1169 if (mark_finished) {
1170 weak_refs_work(false /* clear_all_soft_refs */);
1171
1172 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
1173 // We're done with marking.
1174 // This is the end of the marking cycle, we're expected all
1175 // threads to have SATB queues with active set to true.
1176 satb_mq_set.set_active_all_threads(false, /* new active value */
1177 true /* expected_active */);
1178
1179 {
1180 GCTraceTime(Debug, gc, phases) debug("Flush Task Caches", _gc_timer_cm);
1181 flush_all_task_caches();
1182 }
1183
1184 // Install newly created mark bitmap as "prev".
1185 swap_mark_bitmaps();
1186 {
1187 GCTraceTime(Debug, gc, phases) debug("Update Remembered Set Tracking Before Rebuild", _gc_timer_cm);
1188
1189 uint const workers_by_capacity = (_g1h->num_regions() + G1UpdateRemSetTrackingBeforeRebuildTask::RegionsPerThread - 1) /
1190 G1UpdateRemSetTrackingBeforeRebuildTask::RegionsPerThread;
1191 uint const num_workers = MIN2(_g1h->workers()->active_workers(), workers_by_capacity);
1192
1193 G1UpdateRemSetTrackingBeforeRebuildTask cl(_g1h, this, num_workers);
1194 log_debug(gc,ergo)("Running %s using %u workers for %u regions in heap", cl.name(), num_workers, _g1h->num_regions());
1195 _g1h->workers()->run_task(&cl, num_workers);
1196
1197 log_debug(gc, remset, tracking)("Remembered Set Tracking update regions total %u, selected %u",
1198 _g1h->num_regions(), cl.total_selected_for_rebuild());
1199 }
1200 {
1201 GCTraceTime(Debug, gc, phases) debug("Reclaim Empty Regions", _gc_timer_cm);
1202 reclaim_empty_regions();
1203 }
1204
1205 // Clean out dead classes
1206 if (ClassUnloadingWithConcurrentMark) {
1207 GCTraceTime(Debug, gc, phases) debug("Purge Metaspace", _gc_timer_cm);
1208 ClassLoaderDataGraph::purge();
1209 }
1210
1211 _g1h->resize_heap_if_necessary();
1212
1213 compute_new_sizes();
1214
1215 verify_during_pause(G1HeapVerifier::G1VerifyRemark, VerifyOption_G1UsePrevMarking, "Remark after");
1216
1217 assert(!restart_for_overflow(), "sanity");
1218 // Completely reset the marking state since marking completed
1219 reset_at_marking_complete();
1220 } else {
1221 // We overflowed. Restart concurrent marking.
1222 _restart_for_overflow = true;
1223
1224 verify_during_pause(G1HeapVerifier::G1VerifyRemark, VerifyOption_G1UsePrevMarking, "Remark overflow");
1225
1226 // Clear the marking state because we will be restarting
1227 // marking due to overflowing the global mark stack.
1228 reset_marking_for_restart();
1229 }
1230
1231 {
1232 GCTraceTime(Debug, gc, phases) debug("Report Object Count", _gc_timer_cm);
1233 report_object_count(mark_finished);
1234 }
1235
1236 // Statistics
1237 double now = os::elapsedTime();
1238 _remark_mark_times.add((mark_work_end - start) * 1000.0);
1239 _remark_weak_ref_times.add((now - mark_work_end) * 1000.0);
1240 _remark_times.add((now - start) * 1000.0);
1241
1242 policy->record_concurrent_mark_remark_end();
1243 }
1244
1245 class G1ReclaimEmptyRegionsTask : public AbstractGangTask {
1246 // Per-region work during the Cleanup pause.
1247 class G1ReclaimEmptyRegionsClosure : public HeapRegionClosure {
1248 G1CollectedHeap* _g1h;
1249 size_t _freed_bytes;
1250 FreeRegionList* _local_cleanup_list;
1251 uint _old_regions_removed;
1252 uint _humongous_regions_removed;
1253
1254 public:
1255 G1ReclaimEmptyRegionsClosure(G1CollectedHeap* g1h,
1256 FreeRegionList* local_cleanup_list) :
1257 _g1h(g1h),
1258 _freed_bytes(0),
1259 _local_cleanup_list(local_cleanup_list),
1260 _old_regions_removed(0),
1261 _humongous_regions_removed(0) { }
1262
1263 size_t freed_bytes() { return _freed_bytes; }
1264 const uint old_regions_removed() { return _old_regions_removed; }
1265 const uint humongous_regions_removed() { return _humongous_regions_removed; }
1266
1267 bool do_heap_region(HeapRegion *hr) {
1268 if (hr->used() > 0 && hr->max_live_bytes() == 0 && !hr->is_young() && !hr->is_archive()) {
1269 _freed_bytes += hr->used();
1270 hr->set_containing_set(NULL);
1271 if (hr->is_humongous()) {
1272 _humongous_regions_removed++;
1273 _g1h->free_humongous_region(hr, _local_cleanup_list);
1274 } else {
1275 _old_regions_removed++;
1276 _g1h->free_region(hr, _local_cleanup_list, false /* skip_remset */, false /* skip_hcc */, true /* locked */);
1277 }
1278 hr->clear_cardtable();
1279 _g1h->concurrent_mark()->clear_statistics_in_region(hr->hrm_index());
1280 log_trace(gc)("Reclaimed empty region %u (%s) bot " PTR_FORMAT, hr->hrm_index(), hr->get_short_type_str(), p2i(hr->bottom()));
1281 }
1282
1283 return false;
1284 }
1285 };
1286
1287 G1CollectedHeap* _g1h;
1288 FreeRegionList* _cleanup_list;
1289 HeapRegionClaimer _hrclaimer;
1290
1291 public:
1292 G1ReclaimEmptyRegionsTask(G1CollectedHeap* g1h, FreeRegionList* cleanup_list, uint n_workers) :
1293 AbstractGangTask("G1 Cleanup"),
1294 _g1h(g1h),
1295 _cleanup_list(cleanup_list),
1296 _hrclaimer(n_workers) {
1297 }
1298
1299 void work(uint worker_id) {
1300 FreeRegionList local_cleanup_list("Local Cleanup List");
1301 G1ReclaimEmptyRegionsClosure cl(_g1h, &local_cleanup_list);
1302 _g1h->heap_region_par_iterate_from_worker_offset(&cl, &_hrclaimer, worker_id);
1303 assert(cl.is_complete(), "Shouldn't have aborted!");
1304
1305 // Now update the old/humongous region sets
1306 _g1h->remove_from_old_sets(cl.old_regions_removed(), cl.humongous_regions_removed());
1307 {
1308 MutexLocker x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
1309 _g1h->decrement_summary_bytes(cl.freed_bytes());
1310
1311 _cleanup_list->add_ordered(&local_cleanup_list);
1312 assert(local_cleanup_list.is_empty(), "post-condition");
1313 }
1314 }
1315 };
1316
1317 void G1ConcurrentMark::reclaim_empty_regions() {
1318 WorkGang* workers = _g1h->workers();
1319 FreeRegionList empty_regions_list("Empty Regions After Mark List");
1320
1321 G1ReclaimEmptyRegionsTask cl(_g1h, &empty_regions_list, workers->active_workers());
1322 workers->run_task(&cl);
1323
1324 if (!empty_regions_list.is_empty()) {
1325 log_debug(gc)("Reclaimed %u empty regions", empty_regions_list.length());
1326 // Now print the empty regions list.
1327 G1HRPrinter* hrp = _g1h->hr_printer();
1328 if (hrp->is_active()) {
1329 FreeRegionListIterator iter(&empty_regions_list);
1330 while (iter.more_available()) {
1331 HeapRegion* hr = iter.get_next();
1332 hrp->cleanup(hr);
1333 }
1334 }
1335 // And actually make them available.
1336 _g1h->prepend_to_freelist(&empty_regions_list);
1337 }
1338 }
1339
1340 void G1ConcurrentMark::compute_new_sizes() {
1341 MetaspaceGC::compute_new_size();
1342
1343 // Cleanup will have freed any regions completely full of garbage.
1344 // Update the soft reference policy with the new heap occupancy.
1345 Universe::update_heap_info_at_gc();
1346
1347 // We reclaimed old regions so we should calculate the sizes to make
1348 // sure we update the old gen/space data.
1349 _g1h->g1mm()->update_sizes();
1350 }
1351
1352 void G1ConcurrentMark::cleanup() {
1353 assert_at_safepoint_on_vm_thread();
1354
1355 // If a full collection has happened, we shouldn't do this.
1356 if (has_aborted()) {
1357 return;
1358 }
1359
1360 G1Policy* policy = _g1h->policy();
1361 policy->record_concurrent_mark_cleanup_start();
1362
1363 double start = os::elapsedTime();
1364
1365 verify_during_pause(G1HeapVerifier::G1VerifyCleanup, VerifyOption_G1UsePrevMarking, "Cleanup before");
1366
1367 {
1368 GCTraceTime(Debug, gc, phases) debug("Update Remembered Set Tracking After Rebuild", _gc_timer_cm);
1369 G1UpdateRemSetTrackingAfterRebuild cl(_g1h);
1370 _g1h->heap_region_iterate(&cl);
1371 }
1372
1373 if (log_is_enabled(Trace, gc, liveness)) {
1374 G1PrintRegionLivenessInfoClosure cl("Post-Cleanup");
1375 _g1h->heap_region_iterate(&cl);
1376 }
1377
1378 verify_during_pause(G1HeapVerifier::G1VerifyCleanup, VerifyOption_G1UsePrevMarking, "Cleanup after");
1379
1380 // We need to make this be a "collection" so any collection pause that
1381 // races with it goes around and waits for Cleanup to finish.
1382 _g1h->increment_total_collections();
1383
1384 // Local statistics
1385 double recent_cleanup_time = (os::elapsedTime() - start);
1386 _total_cleanup_time += recent_cleanup_time;
1387 _cleanup_times.add(recent_cleanup_time);
1388
1389 {
1390 GCTraceTime(Debug, gc, phases) debug("Finalize Concurrent Mark Cleanup", _gc_timer_cm);
1391 policy->record_concurrent_mark_cleanup_end();
1392 }
1393 }
1394
1395 // 'Keep Alive' oop closure used by both serial parallel reference processing.
1396 // Uses the G1CMTask associated with a worker thread (for serial reference
1397 // processing the G1CMTask for worker 0 is used) to preserve (mark) and
1398 // trace referent objects.
1399 //
1400 // Using the G1CMTask and embedded local queues avoids having the worker
1401 // threads operating on the global mark stack. This reduces the risk
1402 // of overflowing the stack - which we would rather avoid at this late
1403 // state. Also using the tasks' local queues removes the potential
1404 // of the workers interfering with each other that could occur if
1405 // operating on the global stack.
1406
1407 class G1CMKeepAliveAndDrainClosure : public OopClosure {
1408 G1ConcurrentMark* _cm;
1409 G1CMTask* _task;
1410 uint _ref_counter_limit;
1411 uint _ref_counter;
1412 bool _is_serial;
1413 public:
1414 G1CMKeepAliveAndDrainClosure(G1ConcurrentMark* cm, G1CMTask* task, bool is_serial) :
1415 _cm(cm), _task(task), _ref_counter_limit(G1RefProcDrainInterval),
1416 _ref_counter(_ref_counter_limit), _is_serial(is_serial) {
1417 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code");
1418 }
1419
1420 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
1421 virtual void do_oop( oop* p) { do_oop_work(p); }
1422
1423 template <class T> void do_oop_work(T* p) {
1424 if (_cm->has_overflown()) {
1425 return;
1426 }
1427 if (!_task->deal_with_reference(p)) {
1428 // We did not add anything to the mark bitmap (or mark stack), so there is
1429 // no point trying to drain it.
1430 return;
1431 }
1432 _ref_counter--;
1433
1434 if (_ref_counter == 0) {
1435 // We have dealt with _ref_counter_limit references, pushing them
1436 // and objects reachable from them on to the local stack (and
1437 // possibly the global stack). Call G1CMTask::do_marking_step() to
1438 // process these entries.
1439 //
1440 // We call G1CMTask::do_marking_step() in a loop, which we'll exit if
1441 // there's nothing more to do (i.e. we're done with the entries that
1442 // were pushed as a result of the G1CMTask::deal_with_reference() calls
1443 // above) or we overflow.
1444 //
1445 // Note: G1CMTask::do_marking_step() can set the G1CMTask::has_aborted()
1446 // flag while there may still be some work to do. (See the comment at
1447 // the beginning of G1CMTask::do_marking_step() for those conditions -
1448 // one of which is reaching the specified time target.) It is only
1449 // when G1CMTask::do_marking_step() returns without setting the
1450 // has_aborted() flag that the marking step has completed.
1451 do {
1452 double mark_step_duration_ms = G1ConcMarkStepDurationMillis;
1453 _task->do_marking_step(mark_step_duration_ms,
1454 false /* do_termination */,
1455 _is_serial);
1456 } while (_task->has_aborted() && !_cm->has_overflown());
1457 _ref_counter = _ref_counter_limit;
1458 }
1459 }
1460 };
1461
1462 // 'Drain' oop closure used by both serial and parallel reference processing.
1463 // Uses the G1CMTask associated with a given worker thread (for serial
1464 // reference processing the G1CMtask for worker 0 is used). Calls the
1465 // do_marking_step routine, with an unbelievably large timeout value,
1466 // to drain the marking data structures of the remaining entries
1467 // added by the 'keep alive' oop closure above.
1468
1469 class G1CMDrainMarkingStackClosure : public VoidClosure {
1470 G1ConcurrentMark* _cm;
1471 G1CMTask* _task;
1472 bool _is_serial;
1473 public:
1474 G1CMDrainMarkingStackClosure(G1ConcurrentMark* cm, G1CMTask* task, bool is_serial) :
1475 _cm(cm), _task(task), _is_serial(is_serial) {
1476 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code");
1477 }
1478
1479 void do_void() {
1480 do {
1481 // We call G1CMTask::do_marking_step() to completely drain the local
1482 // and global marking stacks of entries pushed by the 'keep alive'
1483 // oop closure (an instance of G1CMKeepAliveAndDrainClosure above).
1484 //
1485 // G1CMTask::do_marking_step() is called in a loop, which we'll exit
1486 // if there's nothing more to do (i.e. we've completely drained the
1487 // entries that were pushed as a a result of applying the 'keep alive'
1488 // closure to the entries on the discovered ref lists) or we overflow
1489 // the global marking stack.
1490 //
1491 // Note: G1CMTask::do_marking_step() can set the G1CMTask::has_aborted()
1492 // flag while there may still be some work to do. (See the comment at
1493 // the beginning of G1CMTask::do_marking_step() for those conditions -
1494 // one of which is reaching the specified time target.) It is only
1495 // when G1CMTask::do_marking_step() returns without setting the
1496 // has_aborted() flag that the marking step has completed.
1497
1498 _task->do_marking_step(1000000000.0 /* something very large */,
1499 true /* do_termination */,
1500 _is_serial);
1501 } while (_task->has_aborted() && !_cm->has_overflown());
1502 }
1503 };
1504
1505 // Implementation of AbstractRefProcTaskExecutor for parallel
1506 // reference processing at the end of G1 concurrent marking
1507
1508 class G1CMRefProcTaskExecutor : public AbstractRefProcTaskExecutor {
1509 private:
1510 G1CollectedHeap* _g1h;
1511 G1ConcurrentMark* _cm;
1512 WorkGang* _workers;
1513 uint _active_workers;
1514
1515 public:
1516 G1CMRefProcTaskExecutor(G1CollectedHeap* g1h,
1517 G1ConcurrentMark* cm,
1518 WorkGang* workers,
1519 uint n_workers) :
1520 _g1h(g1h), _cm(cm),
1521 _workers(workers), _active_workers(n_workers) { }
1522
1523 virtual void execute(ProcessTask& task, uint ergo_workers);
1524 };
1525
1526 class G1CMRefProcTaskProxy : public AbstractGangTask {
1527 typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask;
1528 ProcessTask& _proc_task;
1529 G1CollectedHeap* _g1h;
1530 G1ConcurrentMark* _cm;
1531
1532 public:
1533 G1CMRefProcTaskProxy(ProcessTask& proc_task,
1534 G1CollectedHeap* g1h,
1535 G1ConcurrentMark* cm) :
1536 AbstractGangTask("Process reference objects in parallel"),
1537 _proc_task(proc_task), _g1h(g1h), _cm(cm) {
1538 ReferenceProcessor* rp = _g1h->ref_processor_cm();
1539 assert(rp->processing_is_mt(), "shouldn't be here otherwise");
1540 }
1541
1542 virtual void work(uint worker_id) {
1543 ResourceMark rm;
1544 HandleMark hm;
1545 G1CMTask* task = _cm->task(worker_id);
1546 G1CMIsAliveClosure g1_is_alive(_g1h);
1547 G1CMKeepAliveAndDrainClosure g1_par_keep_alive(_cm, task, false /* is_serial */);
1548 G1CMDrainMarkingStackClosure g1_par_drain(_cm, task, false /* is_serial */);
1549
1550 _proc_task.work(worker_id, g1_is_alive, g1_par_keep_alive, g1_par_drain);
1551 }
1552 };
1553
1554 void G1CMRefProcTaskExecutor::execute(ProcessTask& proc_task, uint ergo_workers) {
1555 assert(_workers != NULL, "Need parallel worker threads.");
1556 assert(_g1h->ref_processor_cm()->processing_is_mt(), "processing is not MT");
1557 assert(_workers->active_workers() >= ergo_workers,
1558 "Ergonomically chosen workers(%u) should be less than or equal to active workers(%u)",
1559 ergo_workers, _workers->active_workers());
1560
1561 G1CMRefProcTaskProxy proc_task_proxy(proc_task, _g1h, _cm);
1562
1563 // We need to reset the concurrency level before each
1564 // proxy task execution, so that the termination protocol
1565 // and overflow handling in G1CMTask::do_marking_step() knows
1566 // how many workers to wait for.
1567 _cm->set_concurrency(ergo_workers);
1568 _workers->run_task(&proc_task_proxy, ergo_workers);
1569 }
1570
1571 void G1ConcurrentMark::weak_refs_work(bool clear_all_soft_refs) {
1572 ResourceMark rm;
1573 HandleMark hm;
1574
1575 // Is alive closure.
1576 G1CMIsAliveClosure g1_is_alive(_g1h);
1577
1578 // Inner scope to exclude the cleaning of the string table
1579 // from the displayed time.
1580 {
1581 GCTraceTime(Debug, gc, phases) debug("Reference Processing", _gc_timer_cm);
1582
1583 ReferenceProcessor* rp = _g1h->ref_processor_cm();
1584
1585 // See the comment in G1CollectedHeap::ref_processing_init()
1586 // about how reference processing currently works in G1.
1587
1588 // Set the soft reference policy
1589 rp->setup_policy(clear_all_soft_refs);
1590 assert(_global_mark_stack.is_empty(), "mark stack should be empty");
1591
1592 // Instances of the 'Keep Alive' and 'Complete GC' closures used
1593 // in serial reference processing. Note these closures are also
1594 // used for serially processing (by the the current thread) the
1595 // JNI references during parallel reference processing.
1596 //
1597 // These closures do not need to synchronize with the worker
1598 // threads involved in parallel reference processing as these
1599 // instances are executed serially by the current thread (e.g.
1600 // reference processing is not multi-threaded and is thus
1601 // performed by the current thread instead of a gang worker).
1602 //
1603 // The gang tasks involved in parallel reference processing create
1604 // their own instances of these closures, which do their own
1605 // synchronization among themselves.
1606 G1CMKeepAliveAndDrainClosure g1_keep_alive(this, task(0), true /* is_serial */);
1607 G1CMDrainMarkingStackClosure g1_drain_mark_stack(this, task(0), true /* is_serial */);
1608
1609 // We need at least one active thread. If reference processing
1610 // is not multi-threaded we use the current (VMThread) thread,
1611 // otherwise we use the work gang from the G1CollectedHeap and
1612 // we utilize all the worker threads we can.
1613 bool processing_is_mt = rp->processing_is_mt();
1614 uint active_workers = (processing_is_mt ? _g1h->workers()->active_workers() : 1U);
1615 active_workers = MAX2(MIN2(active_workers, _max_num_tasks), 1U);
1616
1617 // Parallel processing task executor.
1618 G1CMRefProcTaskExecutor par_task_executor(_g1h, this,
1619 _g1h->workers(), active_workers);
1620 AbstractRefProcTaskExecutor* executor = (processing_is_mt ? &par_task_executor : NULL);
1621
1622 // Set the concurrency level. The phase was already set prior to
1623 // executing the remark task.
1624 set_concurrency(active_workers);
1625
1626 // Set the degree of MT processing here. If the discovery was done MT,
1627 // the number of threads involved during discovery could differ from
1628 // the number of active workers. This is OK as long as the discovered
1629 // Reference lists are balanced (see balance_all_queues() and balance_queues()).
1630 rp->set_active_mt_degree(active_workers);
1631
1632 ReferenceProcessorPhaseTimes pt(_gc_timer_cm, rp->max_num_queues());
1633
1634 // Process the weak references.
1635 const ReferenceProcessorStats& stats =
1636 rp->process_discovered_references(&g1_is_alive,
1637 &g1_keep_alive,
1638 &g1_drain_mark_stack,
1639 executor,
1640 &pt);
1641 _gc_tracer_cm->report_gc_reference_stats(stats);
1642 pt.print_all_references();
1643
1644 // The do_oop work routines of the keep_alive and drain_marking_stack
1645 // oop closures will set the has_overflown flag if we overflow the
1646 // global marking stack.
1647
1648 assert(has_overflown() || _global_mark_stack.is_empty(),
1649 "Mark stack should be empty (unless it has overflown)");
1650
1651 assert(rp->num_queues() == active_workers, "why not");
1652
1653 rp->verify_no_references_recorded();
1654 assert(!rp->discovery_enabled(), "Post condition");
1655 }
1656
1657 if (has_overflown()) {
1658 // We can not trust g1_is_alive and the contents of the heap if the marking stack
1659 // overflowed while processing references. Exit the VM.
1660 fatal("Overflow during reference processing, can not continue. Please "
1661 "increase MarkStackSizeMax (current value: " SIZE_FORMAT ") and "
1662 "restart.", MarkStackSizeMax);
1663 return;
1664 }
1665
1666 assert(_global_mark_stack.is_empty(), "Marking should have completed");
1667
1668 {
1669 GCTraceTime(Debug, gc, phases) debug("Weak Processing", _gc_timer_cm);
1670 WeakProcessor::weak_oops_do(_g1h->workers(), &g1_is_alive, &do_nothing_cl, 1);
1671 }
1672
1673 // Unload Klasses, String, Code Cache, etc.
1674 if (ClassUnloadingWithConcurrentMark) {
1675 GCTraceTime(Debug, gc, phases) debug("Class Unloading", _gc_timer_cm);
1676 bool purged_classes = SystemDictionary::do_unloading(_gc_timer_cm);
1677 _g1h->complete_cleaning(&g1_is_alive, purged_classes);
1678 } else if (StringDedup::is_enabled()) {
1679 GCTraceTime(Debug, gc, phases) debug("String Deduplication", _gc_timer_cm);
1680 _g1h->string_dedup_cleaning(&g1_is_alive, NULL);
1681 }
1682 }
1683
1684 class G1PrecleanYieldClosure : public YieldClosure {
1685 G1ConcurrentMark* _cm;
1686
1687 public:
1688 G1PrecleanYieldClosure(G1ConcurrentMark* cm) : _cm(cm) { }
1689
1690 virtual bool should_return() {
1691 return _cm->has_aborted();
1692 }
1693
1694 virtual bool should_return_fine_grain() {
1695 _cm->do_yield_check();
1696 return _cm->has_aborted();
1697 }
1698 };
1699
1700 void G1ConcurrentMark::preclean() {
1701 assert(G1UseReferencePrecleaning, "Precleaning must be enabled.");
1702
1703 SuspendibleThreadSetJoiner joiner;
1704
1705 G1CMKeepAliveAndDrainClosure keep_alive(this, task(0), true /* is_serial */);
1706 G1CMDrainMarkingStackClosure drain_mark_stack(this, task(0), true /* is_serial */);
1707
1708 set_concurrency_and_phase(1, true);
1709
1710 G1PrecleanYieldClosure yield_cl(this);
1711
1712 ReferenceProcessor* rp = _g1h->ref_processor_cm();
1713 // Precleaning is single threaded. Temporarily disable MT discovery.
1714 ReferenceProcessorMTDiscoveryMutator rp_mut_discovery(rp, false);
1715 rp->preclean_discovered_references(rp->is_alive_non_header(),
1716 &keep_alive,
1717 &drain_mark_stack,
1718 &yield_cl,
1719 _gc_timer_cm);
1720 }
1721
1722 // When sampling object counts, we already swapped the mark bitmaps, so we need to use
1723 // the prev bitmap determining liveness.
1724 class G1ObjectCountIsAliveClosure: public BoolObjectClosure {
1725 G1CollectedHeap* _g1h;
1726 public:
1727 G1ObjectCountIsAliveClosure(G1CollectedHeap* g1h) : _g1h(g1h) { }
1728
1729 bool do_object_b(oop obj) {
1730 HeapWord* addr = (HeapWord*)obj;
1731 return addr != NULL &&
1732 (!_g1h->is_in_g1_reserved(addr) || !_g1h->is_obj_dead(obj));
1733 }
1734 };
1735
1736 void G1ConcurrentMark::report_object_count(bool mark_completed) {
1737 // Depending on the completion of the marking liveness needs to be determined
1738 // using either the next or prev bitmap.
1739 if (mark_completed) {
1740 G1ObjectCountIsAliveClosure is_alive(_g1h);
1741 _gc_tracer_cm->report_object_count_after_gc(&is_alive);
1742 } else {
1743 G1CMIsAliveClosure is_alive(_g1h);
1744 _gc_tracer_cm->report_object_count_after_gc(&is_alive);
1745 }
1746 }
1747
1748
1749 void G1ConcurrentMark::swap_mark_bitmaps() {
1750 G1CMBitMap* temp = _prev_mark_bitmap;
1751 _prev_mark_bitmap = _next_mark_bitmap;
1752 _next_mark_bitmap = temp;
1753 _g1h->collector_state()->set_clearing_next_bitmap(true);
1754 }
1755
1756 // Closure for marking entries in SATB buffers.
1757 class G1CMSATBBufferClosure : public SATBBufferClosure {
1758 private:
1759 G1CMTask* _task;
1760 G1CollectedHeap* _g1h;
1761
1762 // This is very similar to G1CMTask::deal_with_reference, but with
1763 // more relaxed requirements for the argument, so this must be more
1764 // circumspect about treating the argument as an object.
1765 void do_entry(void* entry) const {
1766 _task->increment_refs_reached();
1767 oop const obj = static_cast<oop>(entry);
1768 _task->make_reference_grey(obj);
1769 }
1770
1771 public:
1772 G1CMSATBBufferClosure(G1CMTask* task, G1CollectedHeap* g1h)
1773 : _task(task), _g1h(g1h) { }
1774
1775 virtual void do_buffer(void** buffer, size_t size) {
1776 for (size_t i = 0; i < size; ++i) {
1777 do_entry(buffer[i]);
1778 }
1779 }
1780 };
1781
1782 class G1RemarkThreadsClosure : public ThreadClosure {
1783 G1CMSATBBufferClosure _cm_satb_cl;
1784 G1CMOopClosure _cm_cl;
1785 MarkingCodeBlobClosure _code_cl;
1786 uintx _claim_token;
1787
1788 public:
1789 G1RemarkThreadsClosure(G1CollectedHeap* g1h, G1CMTask* task) :
1790 _cm_satb_cl(task, g1h),
1791 _cm_cl(g1h, task),
1792 _code_cl(&_cm_cl, !CodeBlobToOopClosure::FixRelocations),
1793 _claim_token(Threads::thread_claim_token()) {}
1794
1795 void do_thread(Thread* thread) {
1796 if (thread->claim_threads_do(true, _claim_token)) {
1797 SATBMarkQueue& queue = G1ThreadLocalData::satb_mark_queue(thread);
1798 queue.apply_closure_and_empty(&_cm_satb_cl);
1799 if (thread->is_Java_thread()) {
1800 // In theory it should not be neccessary to explicitly walk the nmethods to find roots for concurrent marking
1801 // however the liveness of oops reachable from nmethods have very complex lifecycles:
1802 // * Alive if on the stack of an executing method
1803 // * Weakly reachable otherwise
1804 // Some objects reachable from nmethods, such as the class loader (or klass_holder) of the receiver should be
1805 // live by the SATB invariant but other oops recorded in nmethods may behave differently.
1806 JavaThread* jt = (JavaThread*)thread;
1807 jt->nmethods_do(&_code_cl);
1808 }
1809 }
1810 }
1811 };
1812
1813 class G1CMRemarkTask : public AbstractGangTask {
1814 G1ConcurrentMark* _cm;
1815 public:
1816 void work(uint worker_id) {
1817 G1CMTask* task = _cm->task(worker_id);
1818 task->record_start_time();
1819 {
1820 ResourceMark rm;
1821 HandleMark hm;
1822
1823 G1RemarkThreadsClosure threads_f(G1CollectedHeap::heap(), task);
1824 Threads::threads_do(&threads_f);
1825 }
1826
1827 do {
1828 task->do_marking_step(1000000000.0 /* something very large */,
1829 true /* do_termination */,
1830 false /* is_serial */);
1831 } while (task->has_aborted() && !_cm->has_overflown());
1832 // If we overflow, then we do not want to restart. We instead
1833 // want to abort remark and do concurrent marking again.
1834 task->record_end_time();
1835 }
1836
1837 G1CMRemarkTask(G1ConcurrentMark* cm, uint active_workers) :
1838 AbstractGangTask("Par Remark"), _cm(cm) {
1839 _cm->terminator()->reset_for_reuse(active_workers);
1840 }
1841 };
1842
1843 void G1ConcurrentMark::finalize_marking() {
1844 ResourceMark rm;
1845 HandleMark hm;
1846
1847 _g1h->ensure_parsability(false);
1848
1849 // this is remark, so we'll use up all active threads
1850 uint active_workers = _g1h->workers()->active_workers();
1851 set_concurrency_and_phase(active_workers, false /* concurrent */);
1852 // Leave _parallel_marking_threads at it's
1853 // value originally calculated in the G1ConcurrentMark
1854 // constructor and pass values of the active workers
1855 // through the gang in the task.
1856
1857 {
1858 StrongRootsScope srs(active_workers);
1859
1860 G1CMRemarkTask remarkTask(this, active_workers);
1861 // We will start all available threads, even if we decide that the
1862 // active_workers will be fewer. The extra ones will just bail out
1863 // immediately.
1864 _g1h->workers()->run_task(&remarkTask);
1865 }
1866
1867 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
1868 guarantee(has_overflown() ||
1869 satb_mq_set.completed_buffers_num() == 0,
1870 "Invariant: has_overflown = %s, num buffers = " SIZE_FORMAT,
1871 BOOL_TO_STR(has_overflown()),
1872 satb_mq_set.completed_buffers_num());
1873
1874 print_stats();
1875 }
1876
1877 void G1ConcurrentMark::flush_all_task_caches() {
1878 size_t hits = 0;
1879 size_t misses = 0;
1880 for (uint i = 0; i < _max_num_tasks; i++) {
1881 Pair<size_t, size_t> stats = _tasks[i]->flush_mark_stats_cache();
1882 hits += stats.first;
1883 misses += stats.second;
1884 }
1885 size_t sum = hits + misses;
1886 log_debug(gc, stats)("Mark stats cache hits " SIZE_FORMAT " misses " SIZE_FORMAT " ratio %1.3lf",
1887 hits, misses, percent_of(hits, sum));
1888 }
1889
1890 void G1ConcurrentMark::clear_range_in_prev_bitmap(MemRegion mr) {
1891 _prev_mark_bitmap->clear_range(mr);
1892 }
1893
1894 HeapRegion*
1895 G1ConcurrentMark::claim_region(uint worker_id) {
1896 // "checkpoint" the finger
1897 HeapWord* finger = _finger;
1898
1899 while (finger < _heap.end()) {
1900 assert(_g1h->is_in_g1_reserved(finger), "invariant");
1901
1902 HeapRegion* curr_region = _g1h->heap_region_containing(finger);
1903 // Make sure that the reads below do not float before loading curr_region.
1904 OrderAccess::loadload();
1905 // Above heap_region_containing may return NULL as we always scan claim
1906 // until the end of the heap. In this case, just jump to the next region.
1907 HeapWord* end = curr_region != NULL ? curr_region->end() : finger + HeapRegion::GrainWords;
1908
1909 // Is the gap between reading the finger and doing the CAS too long?
1910 HeapWord* res = Atomic::cmpxchg(end, &_finger, finger);
1911 if (res == finger && curr_region != NULL) {
1912 // we succeeded
1913 HeapWord* bottom = curr_region->bottom();
1914 HeapWord* limit = curr_region->next_top_at_mark_start();
1915
1916 // notice that _finger == end cannot be guaranteed here since,
1917 // someone else might have moved the finger even further
1918 assert(_finger >= end, "the finger should have moved forward");
1919
1920 if (limit > bottom) {
1921 return curr_region;
1922 } else {
1923 assert(limit == bottom,
1924 "the region limit should be at bottom");
1925 // we return NULL and the caller should try calling
1926 // claim_region() again.
1927 return NULL;
1928 }
1929 } else {
1930 assert(_finger > finger, "the finger should have moved forward");
1931 // read it again
1932 finger = _finger;
1933 }
1934 }
1935
1936 return NULL;
1937 }
1938
1939 #ifndef PRODUCT
1940 class VerifyNoCSetOops {
1941 G1CollectedHeap* _g1h;
1942 const char* _phase;
1943 int _info;
1944
1945 public:
1946 VerifyNoCSetOops(const char* phase, int info = -1) :
1947 _g1h(G1CollectedHeap::heap()),
1948 _phase(phase),
1949 _info(info)
1950 { }
1951
1952 void operator()(G1TaskQueueEntry task_entry) const {
1953 if (task_entry.is_array_slice()) {
1954 guarantee(_g1h->is_in_reserved(task_entry.slice()), "Slice " PTR_FORMAT " must be in heap.", p2i(task_entry.slice()));
1955 return;
1956 }
1957 guarantee(oopDesc::is_oop(task_entry.obj()),
1958 "Non-oop " PTR_FORMAT ", phase: %s, info: %d",
1959 p2i(task_entry.obj()), _phase, _info);
1960 HeapRegion* r = _g1h->heap_region_containing(task_entry.obj());
1961 guarantee(!(r->in_collection_set() || r->has_index_in_opt_cset()),
1962 "obj " PTR_FORMAT " from %s (%d) in region %u in (optional) collection set",
1963 p2i(task_entry.obj()), _phase, _info, r->hrm_index());
1964 }
1965 };
1966
1967 void G1ConcurrentMark::verify_no_collection_set_oops() {
1968 assert(SafepointSynchronize::is_at_safepoint(), "should be at a safepoint");
1969 if (!_g1h->collector_state()->mark_or_rebuild_in_progress()) {
1970 return;
1971 }
1972
1973 // Verify entries on the global mark stack
1974 _global_mark_stack.iterate(VerifyNoCSetOops("Stack"));
1975
1976 // Verify entries on the task queues
1977 for (uint i = 0; i < _max_num_tasks; ++i) {
1978 G1CMTaskQueue* queue = _task_queues->queue(i);
1979 queue->iterate(VerifyNoCSetOops("Queue", i));
1980 }
1981
1982 // Verify the global finger
1983 HeapWord* global_finger = finger();
1984 if (global_finger != NULL && global_finger < _heap.end()) {
1985 // Since we always iterate over all regions, we might get a NULL HeapRegion
1986 // here.
1987 HeapRegion* global_hr = _g1h->heap_region_containing(global_finger);
1988 guarantee(global_hr == NULL || global_finger == global_hr->bottom(),
1989 "global finger: " PTR_FORMAT " region: " HR_FORMAT,
1990 p2i(global_finger), HR_FORMAT_PARAMS(global_hr));
1991 }
1992
1993 // Verify the task fingers
1994 assert(_num_concurrent_workers <= _max_num_tasks, "sanity");
1995 for (uint i = 0; i < _num_concurrent_workers; ++i) {
1996 G1CMTask* task = _tasks[i];
1997 HeapWord* task_finger = task->finger();
1998 if (task_finger != NULL && task_finger < _heap.end()) {
1999 // See above note on the global finger verification.
2000 HeapRegion* r = _g1h->heap_region_containing(task_finger);
2001 guarantee(r == NULL || task_finger == r->bottom() ||
2002 !r->in_collection_set() || !r->has_index_in_opt_cset(),
2003 "task finger: " PTR_FORMAT " region: " HR_FORMAT,
2004 p2i(task_finger), HR_FORMAT_PARAMS(r));
2005 }
2006 }
2007 }
2008 #endif // PRODUCT
2009
2010 void G1ConcurrentMark::rebuild_rem_set_concurrently() {
2011 _g1h->rem_set()->rebuild_rem_set(this, _concurrent_workers, _worker_id_offset);
2012 }
2013
2014 void G1ConcurrentMark::print_stats() {
2015 if (!log_is_enabled(Debug, gc, stats)) {
2016 return;
2017 }
2018 log_debug(gc, stats)("---------------------------------------------------------------------");
2019 for (size_t i = 0; i < _num_active_tasks; ++i) {
2020 _tasks[i]->print_stats();
2021 log_debug(gc, stats)("---------------------------------------------------------------------");
2022 }
2023 }
2024
2025 void G1ConcurrentMark::concurrent_cycle_abort() {
2026 if (!cm_thread()->during_cycle() || _has_aborted) {
2027 // We haven't started a concurrent cycle or we have already aborted it. No need to do anything.
2028 return;
2029 }
2030
2031 // Clear all marks in the next bitmap for the next marking cycle. This will allow us to skip the next
2032 // concurrent bitmap clearing.
2033 {
2034 GCTraceTime(Debug, gc) debug("Clear Next Bitmap");
2035 clear_bitmap(_next_mark_bitmap, _g1h->workers(), false);
2036 }
2037 // Note we cannot clear the previous marking bitmap here
2038 // since VerifyDuringGC verifies the objects marked during
2039 // a full GC against the previous bitmap.
2040
2041 // Empty mark stack
2042 reset_marking_for_restart();
2043 for (uint i = 0; i < _max_num_tasks; ++i) {
2044 _tasks[i]->clear_region_fields();
2045 }
2046 _first_overflow_barrier_sync.abort();
2047 _second_overflow_barrier_sync.abort();
2048 _has_aborted = true;
2049
2050 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
2051 satb_mq_set.abandon_partial_marking();
2052 // This can be called either during or outside marking, we'll read
2053 // the expected_active value from the SATB queue set.
2054 satb_mq_set.set_active_all_threads(
2055 false, /* new active value */
2056 satb_mq_set.is_active() /* expected_active */);
2057 }
2058
2059 static void print_ms_time_info(const char* prefix, const char* name,
2060 NumberSeq& ns) {
2061 log_trace(gc, marking)("%s%5d %12s: total time = %8.2f s (avg = %8.2f ms).",
2062 prefix, ns.num(), name, ns.sum()/1000.0, ns.avg());
2063 if (ns.num() > 0) {
2064 log_trace(gc, marking)("%s [std. dev = %8.2f ms, max = %8.2f ms]",
2065 prefix, ns.sd(), ns.maximum());
2066 }
2067 }
2068
2069 void G1ConcurrentMark::print_summary_info() {
2070 Log(gc, marking) log;
2071 if (!log.is_trace()) {
2072 return;
2073 }
2074
2075 log.trace(" Concurrent marking:");
2076 print_ms_time_info(" ", "init marks", _init_times);
2077 print_ms_time_info(" ", "remarks", _remark_times);
2078 {
2079 print_ms_time_info(" ", "final marks", _remark_mark_times);
2080 print_ms_time_info(" ", "weak refs", _remark_weak_ref_times);
2081
2082 }
2083 print_ms_time_info(" ", "cleanups", _cleanup_times);
2084 log.trace(" Finalize live data total time = %8.2f s (avg = %8.2f ms).",
2085 _total_cleanup_time, (_cleanup_times.num() > 0 ? _total_cleanup_time * 1000.0 / (double)_cleanup_times.num() : 0.0));
2086 log.trace(" Total stop_world time = %8.2f s.",
2087 (_init_times.sum() + _remark_times.sum() + _cleanup_times.sum())/1000.0);
2088 log.trace(" Total concurrent time = %8.2f s (%8.2f s marking).",
2089 cm_thread()->vtime_accum(), cm_thread()->vtime_mark_accum());
2090 }
2091
2092 void G1ConcurrentMark::print_worker_threads_on(outputStream* st) const {
2093 _concurrent_workers->print_worker_threads_on(st);
2094 }
2095
2096 void G1ConcurrentMark::threads_do(ThreadClosure* tc) const {
2097 _concurrent_workers->threads_do(tc);
2098 }
2099
2100 void G1ConcurrentMark::print_on_error(outputStream* st) const {
2101 st->print_cr("Marking Bits (Prev, Next): (CMBitMap*) " PTR_FORMAT ", (CMBitMap*) " PTR_FORMAT,
2102 p2i(_prev_mark_bitmap), p2i(_next_mark_bitmap));
2103 _prev_mark_bitmap->print_on_error(st, " Prev Bits: ");
2104 _next_mark_bitmap->print_on_error(st, " Next Bits: ");
2105 }
2106
2107 static ReferenceProcessor* get_cm_oop_closure_ref_processor(G1CollectedHeap* g1h) {
2108 ReferenceProcessor* result = g1h->ref_processor_cm();
2109 assert(result != NULL, "CM reference processor should not be NULL");
2110 return result;
2111 }
2112
2113 G1CMOopClosure::G1CMOopClosure(G1CollectedHeap* g1h,
2114 G1CMTask* task)
2115 : MetadataVisitingOopIterateClosure(get_cm_oop_closure_ref_processor(g1h)),
2116 _g1h(g1h), _task(task)
2117 { }
2118
2119 void G1CMTask::setup_for_region(HeapRegion* hr) {
2120 assert(hr != NULL,
2121 "claim_region() should have filtered out NULL regions");
2122 _curr_region = hr;
2123 _finger = hr->bottom();
2124 update_region_limit();
2125 }
2126
2127 void G1CMTask::update_region_limit() {
2128 HeapRegion* hr = _curr_region;
2129 HeapWord* bottom = hr->bottom();
2130 HeapWord* limit = hr->next_top_at_mark_start();
2131
2132 if (limit == bottom) {
2133 // The region was collected underneath our feet.
2134 // We set the finger to bottom to ensure that the bitmap
2135 // iteration that will follow this will not do anything.
2136 // (this is not a condition that holds when we set the region up,
2137 // as the region is not supposed to be empty in the first place)
2138 _finger = bottom;
2139 } else if (limit >= _region_limit) {
2140 assert(limit >= _finger, "peace of mind");
2141 } else {
2142 assert(limit < _region_limit, "only way to get here");
2143 // This can happen under some pretty unusual circumstances. An
2144 // evacuation pause empties the region underneath our feet (NTAMS
2145 // at bottom). We then do some allocation in the region (NTAMS
2146 // stays at bottom), followed by the region being used as a GC
2147 // alloc region (NTAMS will move to top() and the objects
2148 // originally below it will be grayed). All objects now marked in
2149 // the region are explicitly grayed, if below the global finger,
2150 // and we do not need in fact to scan anything else. So, we simply
2151 // set _finger to be limit to ensure that the bitmap iteration
2152 // doesn't do anything.
2153 _finger = limit;
2154 }
2155
2156 _region_limit = limit;
2157 }
2158
2159 void G1CMTask::giveup_current_region() {
2160 assert(_curr_region != NULL, "invariant");
2161 clear_region_fields();
2162 }
2163
2164 void G1CMTask::clear_region_fields() {
2165 // Values for these three fields that indicate that we're not
2166 // holding on to a region.
2167 _curr_region = NULL;
2168 _finger = NULL;
2169 _region_limit = NULL;
2170 }
2171
2172 void G1CMTask::set_cm_oop_closure(G1CMOopClosure* cm_oop_closure) {
2173 if (cm_oop_closure == NULL) {
2174 assert(_cm_oop_closure != NULL, "invariant");
2175 } else {
2176 assert(_cm_oop_closure == NULL, "invariant");
2177 }
2178 _cm_oop_closure = cm_oop_closure;
2179 }
2180
2181 void G1CMTask::reset(G1CMBitMap* next_mark_bitmap) {
2182 guarantee(next_mark_bitmap != NULL, "invariant");
2183 _next_mark_bitmap = next_mark_bitmap;
2184 clear_region_fields();
2185
2186 _calls = 0;
2187 _elapsed_time_ms = 0.0;
2188 _termination_time_ms = 0.0;
2189 _termination_start_time_ms = 0.0;
2190
2191 _mark_stats_cache.reset();
2192 }
2193
2194 bool G1CMTask::should_exit_termination() {
2195 if (!regular_clock_call()) {
2196 return true;
2197 }
2198
2199 // This is called when we are in the termination protocol. We should
2200 // quit if, for some reason, this task wants to abort or the global
2201 // stack is not empty (this means that we can get work from it).
2202 return !_cm->mark_stack_empty() || has_aborted();
2203 }
2204
2205 void G1CMTask::reached_limit() {
2206 assert(_words_scanned >= _words_scanned_limit ||
2207 _refs_reached >= _refs_reached_limit ,
2208 "shouldn't have been called otherwise");
2209 abort_marking_if_regular_check_fail();
2210 }
2211
2212 bool G1CMTask::regular_clock_call() {
2213 if (has_aborted()) {
2214 return false;
2215 }
2216
2217 // First, we need to recalculate the words scanned and refs reached
2218 // limits for the next clock call.
2219 recalculate_limits();
2220
2221 // During the regular clock call we do the following
2222
2223 // (1) If an overflow has been flagged, then we abort.
2224 if (_cm->has_overflown()) {
2225 return false;
2226 }
2227
2228 // If we are not concurrent (i.e. we're doing remark) we don't need
2229 // to check anything else. The other steps are only needed during
2230 // the concurrent marking phase.
2231 if (!_cm->concurrent()) {
2232 return true;
2233 }
2234
2235 // (2) If marking has been aborted for Full GC, then we also abort.
2236 if (_cm->has_aborted()) {
2237 return false;
2238 }
2239
2240 double curr_time_ms = os::elapsedVTime() * 1000.0;
2241
2242 // (4) We check whether we should yield. If we have to, then we abort.
2243 if (SuspendibleThreadSet::should_yield()) {
2244 // We should yield. To do this we abort the task. The caller is
2245 // responsible for yielding.
2246 return false;
2247 }
2248
2249 // (5) We check whether we've reached our time quota. If we have,
2250 // then we abort.
2251 double elapsed_time_ms = curr_time_ms - _start_time_ms;
2252 if (elapsed_time_ms > _time_target_ms) {
2253 _has_timed_out = true;
2254 return false;
2255 }
2256
2257 // (6) Finally, we check whether there are enough completed STAB
2258 // buffers available for processing. If there are, we abort.
2259 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
2260 if (!_draining_satb_buffers && satb_mq_set.process_completed_buffers()) {
2261 // we do need to process SATB buffers, we'll abort and restart
2262 // the marking task to do so
2263 return false;
2264 }
2265 return true;
2266 }
2267
2268 void G1CMTask::recalculate_limits() {
2269 _real_words_scanned_limit = _words_scanned + words_scanned_period;
2270 _words_scanned_limit = _real_words_scanned_limit;
2271
2272 _real_refs_reached_limit = _refs_reached + refs_reached_period;
2273 _refs_reached_limit = _real_refs_reached_limit;
2274 }
2275
2276 void G1CMTask::decrease_limits() {
2277 // This is called when we believe that we're going to do an infrequent
2278 // operation which will increase the per byte scanned cost (i.e. move
2279 // entries to/from the global stack). It basically tries to decrease the
2280 // scanning limit so that the clock is called earlier.
2281
2282 _words_scanned_limit = _real_words_scanned_limit - 3 * words_scanned_period / 4;
2283 _refs_reached_limit = _real_refs_reached_limit - 3 * refs_reached_period / 4;
2284 }
2285
2286 void G1CMTask::move_entries_to_global_stack() {
2287 // Local array where we'll store the entries that will be popped
2288 // from the local queue.
2289 G1TaskQueueEntry buffer[G1CMMarkStack::EntriesPerChunk];
2290
2291 size_t n = 0;
2292 G1TaskQueueEntry task_entry;
2293 while (n < G1CMMarkStack::EntriesPerChunk && _task_queue->pop_local(task_entry)) {
2294 buffer[n] = task_entry;
2295 ++n;
2296 }
2297 if (n < G1CMMarkStack::EntriesPerChunk) {
2298 buffer[n] = G1TaskQueueEntry();
2299 }
2300
2301 if (n > 0) {
2302 if (!_cm->mark_stack_push(buffer)) {
2303 set_has_aborted();
2304 }
2305 }
2306
2307 // This operation was quite expensive, so decrease the limits.
2308 decrease_limits();
2309 }
2310
2311 bool G1CMTask::get_entries_from_global_stack() {
2312 // Local array where we'll store the entries that will be popped
2313 // from the global stack.
2314 G1TaskQueueEntry buffer[G1CMMarkStack::EntriesPerChunk];
2315
2316 if (!_cm->mark_stack_pop(buffer)) {
2317 return false;
2318 }
2319
2320 // We did actually pop at least one entry.
2321 for (size_t i = 0; i < G1CMMarkStack::EntriesPerChunk; ++i) {
2322 G1TaskQueueEntry task_entry = buffer[i];
2323 if (task_entry.is_null()) {
2324 break;
2325 }
2326 assert(task_entry.is_array_slice() || oopDesc::is_oop(task_entry.obj()), "Element " PTR_FORMAT " must be an array slice or oop", p2i(task_entry.obj()));
2327 bool success = _task_queue->push(task_entry);
2328 // We only call this when the local queue is empty or under a
2329 // given target limit. So, we do not expect this push to fail.
2330 assert(success, "invariant");
2331 }
2332
2333 // This operation was quite expensive, so decrease the limits
2334 decrease_limits();
2335 return true;
2336 }
2337
2338 void G1CMTask::drain_local_queue(bool partially) {
2339 if (has_aborted()) {
2340 return;
2341 }
2342
2343 // Decide what the target size is, depending whether we're going to
2344 // drain it partially (so that other tasks can steal if they run out
2345 // of things to do) or totally (at the very end).
2346 size_t target_size;
2347 if (partially) {
2348 target_size = MIN2((size_t)_task_queue->max_elems()/3, (size_t)GCDrainStackTargetSize);
2349 } else {
2350 target_size = 0;
2351 }
2352
2353 if (_task_queue->size() > target_size) {
2354 G1TaskQueueEntry entry;
2355 bool ret = _task_queue->pop_local(entry);
2356 while (ret) {
2357 scan_task_entry(entry);
2358 if (_task_queue->size() <= target_size || has_aborted()) {
2359 ret = false;
2360 } else {
2361 ret = _task_queue->pop_local(entry);
2362 }
2363 }
2364 }
2365 }
2366
2367 void G1CMTask::drain_global_stack(bool partially) {
2368 if (has_aborted()) {
2369 return;
2370 }
2371
2372 // We have a policy to drain the local queue before we attempt to
2373 // drain the global stack.
2374 assert(partially || _task_queue->size() == 0, "invariant");
2375
2376 // Decide what the target size is, depending whether we're going to
2377 // drain it partially (so that other tasks can steal if they run out
2378 // of things to do) or totally (at the very end).
2379 // Notice that when draining the global mark stack partially, due to the racyness
2380 // of the mark stack size update we might in fact drop below the target. But,
2381 // this is not a problem.
2382 // In case of total draining, we simply process until the global mark stack is
2383 // totally empty, disregarding the size counter.
2384 if (partially) {
2385 size_t const target_size = _cm->partial_mark_stack_size_target();
2386 while (!has_aborted() && _cm->mark_stack_size() > target_size) {
2387 if (get_entries_from_global_stack()) {
2388 drain_local_queue(partially);
2389 }
2390 }
2391 } else {
2392 while (!has_aborted() && get_entries_from_global_stack()) {
2393 drain_local_queue(partially);
2394 }
2395 }
2396 }
2397
2398 // SATB Queue has several assumptions on whether to call the par or
2399 // non-par versions of the methods. this is why some of the code is
2400 // replicated. We should really get rid of the single-threaded version
2401 // of the code to simplify things.
2402 void G1CMTask::drain_satb_buffers() {
2403 if (has_aborted()) {
2404 return;
2405 }
2406
2407 // We set this so that the regular clock knows that we're in the
2408 // middle of draining buffers and doesn't set the abort flag when it
2409 // notices that SATB buffers are available for draining. It'd be
2410 // very counter productive if it did that. :-)
2411 _draining_satb_buffers = true;
2412
2413 G1CMSATBBufferClosure satb_cl(this, _g1h);
2414 SATBMarkQueueSet& satb_mq_set = G1BarrierSet::satb_mark_queue_set();
2415
2416 // This keeps claiming and applying the closure to completed buffers
2417 // until we run out of buffers or we need to abort.
2418 while (!has_aborted() &&
2419 satb_mq_set.apply_closure_to_completed_buffer(&satb_cl)) {
2420 abort_marking_if_regular_check_fail();
2421 }
2422
2423 // Can't assert qset is empty here, even if not aborted. If concurrent,
2424 // some other thread might be adding to the queue. If not concurrent,
2425 // some other thread might have won the race for the last buffer, but
2426 // has not yet decremented the count.
2427
2428 _draining_satb_buffers = false;
2429
2430 // again, this was a potentially expensive operation, decrease the
2431 // limits to get the regular clock call early
2432 decrease_limits();
2433 }
2434
2435 void G1CMTask::clear_mark_stats_cache(uint region_idx) {
2436 _mark_stats_cache.reset(region_idx);
2437 }
2438
2439 Pair<size_t, size_t> G1CMTask::flush_mark_stats_cache() {
2440 return _mark_stats_cache.evict_all();
2441 }
2442
2443 void G1CMTask::print_stats() {
2444 log_debug(gc, stats)("Marking Stats, task = %u, calls = %u", _worker_id, _calls);
2445 log_debug(gc, stats)(" Elapsed time = %1.2lfms, Termination time = %1.2lfms",
2446 _elapsed_time_ms, _termination_time_ms);
2447 log_debug(gc, stats)(" Step Times (cum): num = %d, avg = %1.2lfms, sd = %1.2lfms max = %1.2lfms, total = %1.2lfms",
2448 _step_times_ms.num(),
2449 _step_times_ms.avg(),
2450 _step_times_ms.sd(),
2451 _step_times_ms.maximum(),
2452 _step_times_ms.sum());
2453 size_t const hits = _mark_stats_cache.hits();
2454 size_t const misses = _mark_stats_cache.misses();
2455 log_debug(gc, stats)(" Mark Stats Cache: hits " SIZE_FORMAT " misses " SIZE_FORMAT " ratio %.3f",
2456 hits, misses, percent_of(hits, hits + misses));
2457 }
2458
2459 bool G1ConcurrentMark::try_stealing(uint worker_id, G1TaskQueueEntry& task_entry) {
2460 return _task_queues->steal(worker_id, task_entry);
2461 }
2462
2463 /*****************************************************************************
2464
2465 The do_marking_step(time_target_ms, ...) method is the building
2466 block of the parallel marking framework. It can be called in parallel
2467 with other invocations of do_marking_step() on different tasks
2468 (but only one per task, obviously) and concurrently with the
2469 mutator threads, or during remark, hence it eliminates the need
2470 for two versions of the code. When called during remark, it will
2471 pick up from where the task left off during the concurrent marking
2472 phase. Interestingly, tasks are also claimable during evacuation
2473 pauses too, since do_marking_step() ensures that it aborts before
2474 it needs to yield.
2475
2476 The data structures that it uses to do marking work are the
2477 following:
2478
2479 (1) Marking Bitmap. If there are gray objects that appear only
2480 on the bitmap (this happens either when dealing with an overflow
2481 or when the initial marking phase has simply marked the roots
2482 and didn't push them on the stack), then tasks claim heap
2483 regions whose bitmap they then scan to find gray objects. A
2484 global finger indicates where the end of the last claimed region
2485 is. A local finger indicates how far into the region a task has
2486 scanned. The two fingers are used to determine how to gray an
2487 object (i.e. whether simply marking it is OK, as it will be
2488 visited by a task in the future, or whether it needs to be also
2489 pushed on a stack).
2490
2491 (2) Local Queue. The local queue of the task which is accessed
2492 reasonably efficiently by the task. Other tasks can steal from
2493 it when they run out of work. Throughout the marking phase, a
2494 task attempts to keep its local queue short but not totally
2495 empty, so that entries are available for stealing by other
2496 tasks. Only when there is no more work, a task will totally
2497 drain its local queue.
2498
2499 (3) Global Mark Stack. This handles local queue overflow. During
2500 marking only sets of entries are moved between it and the local
2501 queues, as access to it requires a mutex and more fine-grain
2502 interaction with it which might cause contention. If it
2503 overflows, then the marking phase should restart and iterate
2504 over the bitmap to identify gray objects. Throughout the marking
2505 phase, tasks attempt to keep the global mark stack at a small
2506 length but not totally empty, so that entries are available for
2507 popping by other tasks. Only when there is no more work, tasks
2508 will totally drain the global mark stack.
2509
2510 (4) SATB Buffer Queue. This is where completed SATB buffers are
2511 made available. Buffers are regularly removed from this queue
2512 and scanned for roots, so that the queue doesn't get too
2513 long. During remark, all completed buffers are processed, as
2514 well as the filled in parts of any uncompleted buffers.
2515
2516 The do_marking_step() method tries to abort when the time target
2517 has been reached. There are a few other cases when the
2518 do_marking_step() method also aborts:
2519
2520 (1) When the marking phase has been aborted (after a Full GC).
2521
2522 (2) When a global overflow (on the global stack) has been
2523 triggered. Before the task aborts, it will actually sync up with
2524 the other tasks to ensure that all the marking data structures
2525 (local queues, stacks, fingers etc.) are re-initialized so that
2526 when do_marking_step() completes, the marking phase can
2527 immediately restart.
2528
2529 (3) When enough completed SATB buffers are available. The
2530 do_marking_step() method only tries to drain SATB buffers right
2531 at the beginning. So, if enough buffers are available, the
2532 marking step aborts and the SATB buffers are processed at
2533 the beginning of the next invocation.
2534
2535 (4) To yield. when we have to yield then we abort and yield
2536 right at the end of do_marking_step(). This saves us from a lot
2537 of hassle as, by yielding we might allow a Full GC. If this
2538 happens then objects will be compacted underneath our feet, the
2539 heap might shrink, etc. We save checking for this by just
2540 aborting and doing the yield right at the end.
2541
2542 From the above it follows that the do_marking_step() method should
2543 be called in a loop (or, otherwise, regularly) until it completes.
2544
2545 If a marking step completes without its has_aborted() flag being
2546 true, it means it has completed the current marking phase (and
2547 also all other marking tasks have done so and have all synced up).
2548
2549 A method called regular_clock_call() is invoked "regularly" (in
2550 sub ms intervals) throughout marking. It is this clock method that
2551 checks all the abort conditions which were mentioned above and
2552 decides when the task should abort. A work-based scheme is used to
2553 trigger this clock method: when the number of object words the
2554 marking phase has scanned or the number of references the marking
2555 phase has visited reach a given limit. Additional invocations to
2556 the method clock have been planted in a few other strategic places
2557 too. The initial reason for the clock method was to avoid calling
2558 vtime too regularly, as it is quite expensive. So, once it was in
2559 place, it was natural to piggy-back all the other conditions on it
2560 too and not constantly check them throughout the code.
2561
2562 If do_termination is true then do_marking_step will enter its
2563 termination protocol.
2564
2565 The value of is_serial must be true when do_marking_step is being
2566 called serially (i.e. by the VMThread) and do_marking_step should
2567 skip any synchronization in the termination and overflow code.
2568 Examples include the serial remark code and the serial reference
2569 processing closures.
2570
2571 The value of is_serial must be false when do_marking_step is
2572 being called by any of the worker threads in a work gang.
2573 Examples include the concurrent marking code (CMMarkingTask),
2574 the MT remark code, and the MT reference processing closures.
2575
2576 *****************************************************************************/
2577
2578 void G1CMTask::do_marking_step(double time_target_ms,
2579 bool do_termination,
2580 bool is_serial) {
2581 assert(time_target_ms >= 1.0, "minimum granularity is 1ms");
2582
2583 _start_time_ms = os::elapsedVTime() * 1000.0;
2584
2585 // If do_stealing is true then do_marking_step will attempt to
2586 // steal work from the other G1CMTasks. It only makes sense to
2587 // enable stealing when the termination protocol is enabled
2588 // and do_marking_step() is not being called serially.
2589 bool do_stealing = do_termination && !is_serial;
2590
2591 double diff_prediction_ms = _g1h->policy()->predictor().get_new_prediction(&_marking_step_diff_ms);
2592 _time_target_ms = time_target_ms - diff_prediction_ms;
2593
2594 // set up the variables that are used in the work-based scheme to
2595 // call the regular clock method
2596 _words_scanned = 0;
2597 _refs_reached = 0;
2598 recalculate_limits();
2599
2600 // clear all flags
2601 clear_has_aborted();
2602 _has_timed_out = false;
2603 _draining_satb_buffers = false;
2604
2605 ++_calls;
2606
2607 // Set up the bitmap and oop closures. Anything that uses them is
2608 // eventually called from this method, so it is OK to allocate these
2609 // statically.
2610 G1CMBitMapClosure bitmap_closure(this, _cm);
2611 G1CMOopClosure cm_oop_closure(_g1h, this);
2612 set_cm_oop_closure(&cm_oop_closure);
2613
2614 if (_cm->has_overflown()) {
2615 // This can happen if the mark stack overflows during a GC pause
2616 // and this task, after a yield point, restarts. We have to abort
2617 // as we need to get into the overflow protocol which happens
2618 // right at the end of this task.
2619 set_has_aborted();
2620 }
2621
2622 // First drain any available SATB buffers. After this, we will not
2623 // look at SATB buffers before the next invocation of this method.
2624 // If enough completed SATB buffers are queued up, the regular clock
2625 // will abort this task so that it restarts.
2626 drain_satb_buffers();
2627 // ...then partially drain the local queue and the global stack
2628 drain_local_queue(true);
2629 drain_global_stack(true);
2630
2631 do {
2632 if (!has_aborted() && _curr_region != NULL) {
2633 // This means that we're already holding on to a region.
2634 assert(_finger != NULL, "if region is not NULL, then the finger "
2635 "should not be NULL either");
2636
2637 // We might have restarted this task after an evacuation pause
2638 // which might have evacuated the region we're holding on to
2639 // underneath our feet. Let's read its limit again to make sure
2640 // that we do not iterate over a region of the heap that
2641 // contains garbage (update_region_limit() will also move
2642 // _finger to the start of the region if it is found empty).
2643 update_region_limit();
2644 // We will start from _finger not from the start of the region,
2645 // as we might be restarting this task after aborting half-way
2646 // through scanning this region. In this case, _finger points to
2647 // the address where we last found a marked object. If this is a
2648 // fresh region, _finger points to start().
2649 MemRegion mr = MemRegion(_finger, _region_limit);
2650
2651 assert(!_curr_region->is_humongous() || mr.start() == _curr_region->bottom(),
2652 "humongous regions should go around loop once only");
2653
2654 // Some special cases:
2655 // If the memory region is empty, we can just give up the region.
2656 // If the current region is humongous then we only need to check
2657 // the bitmap for the bit associated with the start of the object,
2658 // scan the object if it's live, and give up the region.
2659 // Otherwise, let's iterate over the bitmap of the part of the region
2660 // that is left.
2661 // If the iteration is successful, give up the region.
2662 if (mr.is_empty()) {
2663 giveup_current_region();
2664 abort_marking_if_regular_check_fail();
2665 } else if (_curr_region->is_humongous() && mr.start() == _curr_region->bottom()) {
2666 if (_next_mark_bitmap->is_marked(mr.start())) {
2667 // The object is marked - apply the closure
2668 bitmap_closure.do_addr(mr.start());
2669 }
2670 // Even if this task aborted while scanning the humongous object
2671 // we can (and should) give up the current region.
2672 giveup_current_region();
2673 abort_marking_if_regular_check_fail();
2674 } else if (_next_mark_bitmap->iterate(&bitmap_closure, mr)) {
2675 giveup_current_region();
2676 abort_marking_if_regular_check_fail();
2677 } else {
2678 assert(has_aborted(), "currently the only way to do so");
2679 // The only way to abort the bitmap iteration is to return
2680 // false from the do_bit() method. However, inside the
2681 // do_bit() method we move the _finger to point to the
2682 // object currently being looked at. So, if we bail out, we
2683 // have definitely set _finger to something non-null.
2684 assert(_finger != NULL, "invariant");
2685
2686 // Region iteration was actually aborted. So now _finger
2687 // points to the address of the object we last scanned. If we
2688 // leave it there, when we restart this task, we will rescan
2689 // the object. It is easy to avoid this. We move the finger by
2690 // enough to point to the next possible object header.
2691 assert(_finger < _region_limit, "invariant");
2692 HeapWord* const new_finger = _finger + ((oop)_finger)->size();
2693 // Check if bitmap iteration was aborted while scanning the last object
2694 if (new_finger >= _region_limit) {
2695 giveup_current_region();
2696 } else {
2697 move_finger_to(new_finger);
2698 }
2699 }
2700 }
2701 // At this point we have either completed iterating over the
2702 // region we were holding on to, or we have aborted.
2703
2704 // We then partially drain the local queue and the global stack.
2705 // (Do we really need this?)
2706 drain_local_queue(true);
2707 drain_global_stack(true);
2708
2709 // Read the note on the claim_region() method on why it might
2710 // return NULL with potentially more regions available for
2711 // claiming and why we have to check out_of_regions() to determine
2712 // whether we're done or not.
2713 while (!has_aborted() && _curr_region == NULL && !_cm->out_of_regions()) {
2714 // We are going to try to claim a new region. We should have
2715 // given up on the previous one.
2716 // Separated the asserts so that we know which one fires.
2717 assert(_curr_region == NULL, "invariant");
2718 assert(_finger == NULL, "invariant");
2719 assert(_region_limit == NULL, "invariant");
2720 HeapRegion* claimed_region = _cm->claim_region(_worker_id);
2721 if (claimed_region != NULL) {
2722 // Yes, we managed to claim one
2723 setup_for_region(claimed_region);
2724 assert(_curr_region == claimed_region, "invariant");
2725 }
2726 // It is important to call the regular clock here. It might take
2727 // a while to claim a region if, for example, we hit a large
2728 // block of empty regions. So we need to call the regular clock
2729 // method once round the loop to make sure it's called
2730 // frequently enough.
2731 abort_marking_if_regular_check_fail();
2732 }
2733
2734 if (!has_aborted() && _curr_region == NULL) {
2735 assert(_cm->out_of_regions(),
2736 "at this point we should be out of regions");
2737 }
2738 } while ( _curr_region != NULL && !has_aborted());
2739
2740 if (!has_aborted()) {
2741 // We cannot check whether the global stack is empty, since other
2742 // tasks might be pushing objects to it concurrently.
2743 assert(_cm->out_of_regions(),
2744 "at this point we should be out of regions");
2745 // Try to reduce the number of available SATB buffers so that
2746 // remark has less work to do.
2747 drain_satb_buffers();
2748 }
2749
2750 // Since we've done everything else, we can now totally drain the
2751 // local queue and global stack.
2752 drain_local_queue(false);
2753 drain_global_stack(false);
2754
2755 // Attempt at work stealing from other task's queues.
2756 if (do_stealing && !has_aborted()) {
2757 // We have not aborted. This means that we have finished all that
2758 // we could. Let's try to do some stealing...
2759
2760 // We cannot check whether the global stack is empty, since other
2761 // tasks might be pushing objects to it concurrently.
2762 assert(_cm->out_of_regions() && _task_queue->size() == 0,
2763 "only way to reach here");
2764 while (!has_aborted()) {
2765 G1TaskQueueEntry entry;
2766 if (_cm->try_stealing(_worker_id, entry)) {
2767 scan_task_entry(entry);
2768
2769 // And since we're towards the end, let's totally drain the
2770 // local queue and global stack.
2771 drain_local_queue(false);
2772 drain_global_stack(false);
2773 } else {
2774 break;
2775 }
2776 }
2777 }
2778
2779 // We still haven't aborted. Now, let's try to get into the
2780 // termination protocol.
2781 if (do_termination && !has_aborted()) {
2782 // We cannot check whether the global stack is empty, since other
2783 // tasks might be concurrently pushing objects on it.
2784 // Separated the asserts so that we know which one fires.
2785 assert(_cm->out_of_regions(), "only way to reach here");
2786 assert(_task_queue->size() == 0, "only way to reach here");
2787 _termination_start_time_ms = os::elapsedVTime() * 1000.0;
2788
2789 // The G1CMTask class also extends the TerminatorTerminator class,
2790 // hence its should_exit_termination() method will also decide
2791 // whether to exit the termination protocol or not.
2792 bool finished = (is_serial ||
2793 _cm->terminator()->offer_termination(this));
2794 double termination_end_time_ms = os::elapsedVTime() * 1000.0;
2795 _termination_time_ms +=
2796 termination_end_time_ms - _termination_start_time_ms;
2797
2798 if (finished) {
2799 // We're all done.
2800
2801 // We can now guarantee that the global stack is empty, since
2802 // all other tasks have finished. We separated the guarantees so
2803 // that, if a condition is false, we can immediately find out
2804 // which one.
2805 guarantee(_cm->out_of_regions(), "only way to reach here");
2806 guarantee(_cm->mark_stack_empty(), "only way to reach here");
2807 guarantee(_task_queue->size() == 0, "only way to reach here");
2808 guarantee(!_cm->has_overflown(), "only way to reach here");
2809 guarantee(!has_aborted(), "should never happen if termination has completed");
2810 } else {
2811 // Apparently there's more work to do. Let's abort this task. It
2812 // will restart it and we can hopefully find more things to do.
2813 set_has_aborted();
2814 }
2815 }
2816
2817 // Mainly for debugging purposes to make sure that a pointer to the
2818 // closure which was statically allocated in this frame doesn't
2819 // escape it by accident.
2820 set_cm_oop_closure(NULL);
2821 double end_time_ms = os::elapsedVTime() * 1000.0;
2822 double elapsed_time_ms = end_time_ms - _start_time_ms;
2823 // Update the step history.
2824 _step_times_ms.add(elapsed_time_ms);
2825
2826 if (has_aborted()) {
2827 // The task was aborted for some reason.
2828 if (_has_timed_out) {
2829 double diff_ms = elapsed_time_ms - _time_target_ms;
2830 // Keep statistics of how well we did with respect to hitting
2831 // our target only if we actually timed out (if we aborted for
2832 // other reasons, then the results might get skewed).
2833 _marking_step_diff_ms.add(diff_ms);
2834 }
2835
2836 if (_cm->has_overflown()) {
2837 // This is the interesting one. We aborted because a global
2838 // overflow was raised. This means we have to restart the
2839 // marking phase and start iterating over regions. However, in
2840 // order to do this we have to make sure that all tasks stop
2841 // what they are doing and re-initialize in a safe manner. We
2842 // will achieve this with the use of two barrier sync points.
2843
2844 if (!is_serial) {
2845 // We only need to enter the sync barrier if being called
2846 // from a parallel context
2847 _cm->enter_first_sync_barrier(_worker_id);
2848
2849 // When we exit this sync barrier we know that all tasks have
2850 // stopped doing marking work. So, it's now safe to
2851 // re-initialize our data structures.
2852 }
2853
2854 clear_region_fields();
2855 flush_mark_stats_cache();
2856
2857 if (!is_serial) {
2858 // If we're executing the concurrent phase of marking, reset the marking
2859 // state; otherwise the marking state is reset after reference processing,
2860 // during the remark pause.
2861 // If we reset here as a result of an overflow during the remark we will
2862 // see assertion failures from any subsequent set_concurrency_and_phase()
2863 // calls.
2864 if (_cm->concurrent() && _worker_id == 0) {
2865 // Worker 0 is responsible for clearing the global data structures because
2866 // of an overflow. During STW we should not clear the overflow flag (in
2867 // G1ConcurrentMark::reset_marking_state()) since we rely on it being true when we exit
2868 // method to abort the pause and restart concurrent marking.
2869 _cm->reset_marking_for_restart();
2870
2871 log_info(gc, marking)("Concurrent Mark reset for overflow");
2872 }
2873
2874 // ...and enter the second barrier.
2875 _cm->enter_second_sync_barrier(_worker_id);
2876 }
2877 // At this point, if we're during the concurrent phase of
2878 // marking, everything has been re-initialized and we're
2879 // ready to restart.
2880 }
2881 }
2882 }
2883
2884 G1CMTask::G1CMTask(uint worker_id,
2885 G1ConcurrentMark* cm,
2886 G1CMTaskQueue* task_queue,
2887 G1RegionMarkStats* mark_stats,
2888 uint max_regions) :
2889 _objArray_processor(this),
2890 _worker_id(worker_id),
2891 _g1h(G1CollectedHeap::heap()),
2892 _cm(cm),
2893 _next_mark_bitmap(NULL),
2894 _task_queue(task_queue),
2895 _mark_stats_cache(mark_stats, max_regions, RegionMarkStatsCacheSize),
2896 _calls(0),
2897 _time_target_ms(0.0),
2898 _start_time_ms(0.0),
2899 _cm_oop_closure(NULL),
2900 _curr_region(NULL),
2901 _finger(NULL),
2902 _region_limit(NULL),
2903 _words_scanned(0),
2904 _words_scanned_limit(0),
2905 _real_words_scanned_limit(0),
2906 _refs_reached(0),
2907 _refs_reached_limit(0),
2908 _real_refs_reached_limit(0),
2909 _has_aborted(false),
2910 _has_timed_out(false),
2911 _draining_satb_buffers(false),
2912 _step_times_ms(),
2913 _elapsed_time_ms(0.0),
2914 _termination_time_ms(0.0),
2915 _termination_start_time_ms(0.0),
2916 _marking_step_diff_ms()
2917 {
2918 guarantee(task_queue != NULL, "invariant");
2919
2920 _marking_step_diff_ms.add(0.5);
2921 }
2922
2923 // These are formatting macros that are used below to ensure
2924 // consistent formatting. The *_H_* versions are used to format the
2925 // header for a particular value and they should be kept consistent
2926 // with the corresponding macro. Also note that most of the macros add
2927 // the necessary white space (as a prefix) which makes them a bit
2928 // easier to compose.
2929
2930 // All the output lines are prefixed with this string to be able to
2931 // identify them easily in a large log file.
2932 #define G1PPRL_LINE_PREFIX "###"
2933
2934 #define G1PPRL_ADDR_BASE_FORMAT " " PTR_FORMAT "-" PTR_FORMAT
2935 #ifdef _LP64
2936 #define G1PPRL_ADDR_BASE_H_FORMAT " %37s"
2937 #else // _LP64
2938 #define G1PPRL_ADDR_BASE_H_FORMAT " %21s"
2939 #endif // _LP64
2940
2941 // For per-region info
2942 #define G1PPRL_TYPE_FORMAT " %-4s"
2943 #define G1PPRL_TYPE_H_FORMAT " %4s"
2944 #define G1PPRL_STATE_FORMAT " %-5s"
2945 #define G1PPRL_STATE_H_FORMAT " %5s"
2946 #define G1PPRL_BYTE_FORMAT " " SIZE_FORMAT_W(9)
2947 #define G1PPRL_BYTE_H_FORMAT " %9s"
2948 #define G1PPRL_DOUBLE_FORMAT " %14.1f"
2949 #define G1PPRL_DOUBLE_H_FORMAT " %14s"
2950
2951 // For summary info
2952 #define G1PPRL_SUM_ADDR_FORMAT(tag) " " tag ":" G1PPRL_ADDR_BASE_FORMAT
2953 #define G1PPRL_SUM_BYTE_FORMAT(tag) " " tag ": " SIZE_FORMAT
2954 #define G1PPRL_SUM_MB_FORMAT(tag) " " tag ": %1.2f MB"
2955 #define G1PPRL_SUM_MB_PERC_FORMAT(tag) G1PPRL_SUM_MB_FORMAT(tag) " / %1.2f %%"
2956
2957 G1PrintRegionLivenessInfoClosure::G1PrintRegionLivenessInfoClosure(const char* phase_name) :
2958 _total_used_bytes(0), _total_capacity_bytes(0),
2959 _total_prev_live_bytes(0), _total_next_live_bytes(0),
2960 _total_remset_bytes(0), _total_strong_code_roots_bytes(0)
2961 {
2962 if (!log_is_enabled(Trace, gc, liveness)) {
2963 return;
2964 }
2965
2966 G1CollectedHeap* g1h = G1CollectedHeap::heap();
2967 MemRegion g1_reserved = g1h->g1_reserved();
2968 double now = os::elapsedTime();
2969
2970 // Print the header of the output.
2971 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX" PHASE %s @ %1.3f", phase_name, now);
2972 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX" HEAP"
2973 G1PPRL_SUM_ADDR_FORMAT("reserved")
2974 G1PPRL_SUM_BYTE_FORMAT("region-size"),
2975 p2i(g1_reserved.start()), p2i(g1_reserved.end()),
2976 HeapRegion::GrainBytes);
2977 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX);
2978 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX
2979 G1PPRL_TYPE_H_FORMAT
2980 G1PPRL_ADDR_BASE_H_FORMAT
2981 G1PPRL_BYTE_H_FORMAT
2982 G1PPRL_BYTE_H_FORMAT
2983 G1PPRL_BYTE_H_FORMAT
2984 G1PPRL_DOUBLE_H_FORMAT
2985 G1PPRL_BYTE_H_FORMAT
2986 G1PPRL_STATE_H_FORMAT
2987 G1PPRL_BYTE_H_FORMAT,
2988 "type", "address-range",
2989 "used", "prev-live", "next-live", "gc-eff",
2990 "remset", "state", "code-roots");
2991 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX
2992 G1PPRL_TYPE_H_FORMAT
2993 G1PPRL_ADDR_BASE_H_FORMAT
2994 G1PPRL_BYTE_H_FORMAT
2995 G1PPRL_BYTE_H_FORMAT
2996 G1PPRL_BYTE_H_FORMAT
2997 G1PPRL_DOUBLE_H_FORMAT
2998 G1PPRL_BYTE_H_FORMAT
2999 G1PPRL_STATE_H_FORMAT
3000 G1PPRL_BYTE_H_FORMAT,
3001 "", "",
3002 "(bytes)", "(bytes)", "(bytes)", "(bytes/ms)",
3003 "(bytes)", "", "(bytes)");
3004 }
3005
3006 bool G1PrintRegionLivenessInfoClosure::do_heap_region(HeapRegion* r) {
3007 if (!log_is_enabled(Trace, gc, liveness)) {
3008 return false;
3009 }
3010
3011 const char* type = r->get_type_str();
3012 HeapWord* bottom = r->bottom();
3013 HeapWord* end = r->end();
3014 size_t capacity_bytes = r->capacity();
3015 size_t used_bytes = r->used();
3016 size_t prev_live_bytes = r->live_bytes();
3017 size_t next_live_bytes = r->next_live_bytes();
3018 double gc_eff = r->gc_efficiency();
3019 size_t remset_bytes = r->rem_set()->mem_size();
3020 size_t strong_code_roots_bytes = r->rem_set()->strong_code_roots_mem_size();
3021 const char* remset_type = r->rem_set()->get_short_state_str();
3022
3023 _total_used_bytes += used_bytes;
3024 _total_capacity_bytes += capacity_bytes;
3025 _total_prev_live_bytes += prev_live_bytes;
3026 _total_next_live_bytes += next_live_bytes;
3027 _total_remset_bytes += remset_bytes;
3028 _total_strong_code_roots_bytes += strong_code_roots_bytes;
3029
3030 // Print a line for this particular region.
3031 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX
3032 G1PPRL_TYPE_FORMAT
3033 G1PPRL_ADDR_BASE_FORMAT
3034 G1PPRL_BYTE_FORMAT
3035 G1PPRL_BYTE_FORMAT
3036 G1PPRL_BYTE_FORMAT
3037 G1PPRL_DOUBLE_FORMAT
3038 G1PPRL_BYTE_FORMAT
3039 G1PPRL_STATE_FORMAT
3040 G1PPRL_BYTE_FORMAT,
3041 type, p2i(bottom), p2i(end),
3042 used_bytes, prev_live_bytes, next_live_bytes, gc_eff,
3043 remset_bytes, remset_type, strong_code_roots_bytes);
3044
3045 return false;
3046 }
3047
3048 G1PrintRegionLivenessInfoClosure::~G1PrintRegionLivenessInfoClosure() {
3049 if (!log_is_enabled(Trace, gc, liveness)) {
3050 return;
3051 }
3052
3053 // add static memory usages to remembered set sizes
3054 _total_remset_bytes += HeapRegionRemSet::fl_mem_size() + HeapRegionRemSet::static_mem_size();
3055 // Print the footer of the output.
3056 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX);
3057 log_trace(gc, liveness)(G1PPRL_LINE_PREFIX
3058 " SUMMARY"
3059 G1PPRL_SUM_MB_FORMAT("capacity")
3060 G1PPRL_SUM_MB_PERC_FORMAT("used")
3061 G1PPRL_SUM_MB_PERC_FORMAT("prev-live")
3062 G1PPRL_SUM_MB_PERC_FORMAT("next-live")
3063 G1PPRL_SUM_MB_FORMAT("remset")
3064 G1PPRL_SUM_MB_FORMAT("code-roots"),
3065 bytes_to_mb(_total_capacity_bytes),
3066 bytes_to_mb(_total_used_bytes),
3067 percent_of(_total_used_bytes, _total_capacity_bytes),
3068 bytes_to_mb(_total_prev_live_bytes),
3069 percent_of(_total_prev_live_bytes, _total_capacity_bytes),
3070 bytes_to_mb(_total_next_live_bytes),
3071 percent_of(_total_next_live_bytes, _total_capacity_bytes),
3072 bytes_to_mb(_total_remset_bytes),
3073 bytes_to_mb(_total_strong_code_roots_bytes));
3074 }