1 /*
2 * Copyright (c) 2001, 2015, 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/metadataOnStackMark.hpp"
27 #include "classfile/symbolTable.hpp"
28 #include "code/codeCache.hpp"
29 #include "gc/g1/concurrentMark.inline.hpp"
30 #include "gc/g1/concurrentMarkThread.inline.hpp"
31 #include "gc/g1/g1CollectedHeap.inline.hpp"
32 #include "gc/g1/g1CollectorPolicy.hpp"
33 #include "gc/g1/g1ErgoVerbose.hpp"
34 #include "gc/g1/g1Log.hpp"
35 #include "gc/g1/g1OopClosures.inline.hpp"
36 #include "gc/g1/g1RemSet.hpp"
37 #include "gc/g1/g1StringDedup.hpp"
38 #include "gc/g1/heapRegion.inline.hpp"
39 #include "gc/g1/heapRegionManager.inline.hpp"
40 #include "gc/g1/heapRegionRemSet.hpp"
41 #include "gc/g1/heapRegionSet.inline.hpp"
42 #include "gc/g1/suspendibleThreadSet.hpp"
43 #include "gc/shared/gcTimer.hpp"
44 #include "gc/shared/gcTrace.hpp"
45 #include "gc/shared/gcTraceTime.hpp"
46 #include "gc/shared/genOopClosures.inline.hpp"
47 #include "gc/shared/referencePolicy.hpp"
48 #include "gc/shared/strongRootsScope.hpp"
49 #include "gc/shared/taskqueue.inline.hpp"
50 #include "gc/shared/vmGCOperations.hpp"
51 #include "memory/allocation.hpp"
52 #include "memory/resourceArea.hpp"
53 #include "oops/oop.inline.hpp"
54 #include "runtime/atomic.inline.hpp"
55 #include "runtime/handles.inline.hpp"
56 #include "runtime/java.hpp"
57 #include "runtime/prefetch.inline.hpp"
58 #include "services/memTracker.hpp"
59
60 // Concurrent marking bit map wrapper
61
62 CMBitMapRO::CMBitMapRO(int shifter) :
63 _bm(),
64 _shifter(shifter) {
65 _bmStartWord = 0;
66 _bmWordSize = 0;
67 }
68
69 HeapWord* CMBitMapRO::getNextMarkedWordAddress(const HeapWord* addr,
70 const HeapWord* limit) const {
71 // First we must round addr *up* to a possible object boundary.
72 addr = (HeapWord*)align_size_up((intptr_t)addr,
73 HeapWordSize << _shifter);
74 size_t addrOffset = heapWordToOffset(addr);
75 if (limit == NULL) {
76 limit = _bmStartWord + _bmWordSize;
77 }
78 size_t limitOffset = heapWordToOffset(limit);
79 size_t nextOffset = _bm.get_next_one_offset(addrOffset, limitOffset);
80 HeapWord* nextAddr = offsetToHeapWord(nextOffset);
81 assert(nextAddr >= addr, "get_next_one postcondition");
82 assert(nextAddr == limit || isMarked(nextAddr),
83 "get_next_one postcondition");
84 return nextAddr;
85 }
86
87 HeapWord* CMBitMapRO::getNextUnmarkedWordAddress(const HeapWord* addr,
88 const HeapWord* limit) const {
89 size_t addrOffset = heapWordToOffset(addr);
90 if (limit == NULL) {
91 limit = _bmStartWord + _bmWordSize;
92 }
93 size_t limitOffset = heapWordToOffset(limit);
94 size_t nextOffset = _bm.get_next_zero_offset(addrOffset, limitOffset);
95 HeapWord* nextAddr = offsetToHeapWord(nextOffset);
96 assert(nextAddr >= addr, "get_next_one postcondition");
97 assert(nextAddr == limit || !isMarked(nextAddr),
98 "get_next_one postcondition");
99 return nextAddr;
100 }
101
102 int CMBitMapRO::heapWordDiffToOffsetDiff(size_t diff) const {
103 assert((diff & ((1 << _shifter) - 1)) == 0, "argument check");
104 return (int) (diff >> _shifter);
105 }
106
107 #ifndef PRODUCT
108 bool CMBitMapRO::covers(MemRegion heap_rs) const {
109 // assert(_bm.map() == _virtual_space.low(), "map inconsistency");
110 assert(((size_t)_bm.size() * ((size_t)1 << _shifter)) == _bmWordSize,
111 "size inconsistency");
112 return _bmStartWord == (HeapWord*)(heap_rs.start()) &&
113 _bmWordSize == heap_rs.word_size();
114 }
115 #endif
116
117 void CMBitMapRO::print_on_error(outputStream* st, const char* prefix) const {
118 _bm.print_on_error(st, prefix);
119 }
120
121 size_t CMBitMap::compute_size(size_t heap_size) {
122 return ReservedSpace::allocation_align_size_up(heap_size / mark_distance());
123 }
124
125 size_t CMBitMap::mark_distance() {
126 return MinObjAlignmentInBytes * BitsPerByte;
127 }
128
129 void CMBitMap::initialize(MemRegion heap, G1RegionToSpaceMapper* storage) {
130 _bmStartWord = heap.start();
131 _bmWordSize = heap.word_size();
132
133 _bm.set_map((BitMap::bm_word_t*) storage->reserved().start());
134 _bm.set_size(_bmWordSize >> _shifter);
135
136 storage->set_mapping_changed_listener(&_listener);
137 }
138
139 void CMBitMapMappingChangedListener::on_commit(uint start_region, size_t num_regions, bool zero_filled) {
140 if (zero_filled) {
141 return;
142 }
143 // We need to clear the bitmap on commit, removing any existing information.
144 MemRegion mr(G1CollectedHeap::heap()->bottom_addr_for_region(start_region), num_regions * HeapRegion::GrainWords);
145 _bm->clearRange(mr);
146 }
147
148 // Closure used for clearing the given mark bitmap.
149 class ClearBitmapHRClosure : public HeapRegionClosure {
150 private:
151 ConcurrentMark* _cm;
152 CMBitMap* _bitmap;
153 bool _may_yield; // The closure may yield during iteration. If yielded, abort the iteration.
154 public:
155 ClearBitmapHRClosure(ConcurrentMark* cm, CMBitMap* bitmap, bool may_yield) : HeapRegionClosure(), _cm(cm), _bitmap(bitmap), _may_yield(may_yield) {
156 assert(!may_yield || cm != NULL, "CM must be non-NULL if this closure is expected to yield.");
157 }
158
159 virtual bool doHeapRegion(HeapRegion* r) {
160 size_t const chunk_size_in_words = M / HeapWordSize;
161
162 HeapWord* cur = r->bottom();
163 HeapWord* const end = r->end();
164
165 while (cur < end) {
166 MemRegion mr(cur, MIN2(cur + chunk_size_in_words, end));
167 _bitmap->clearRange(mr);
168
169 cur += chunk_size_in_words;
170
171 // Abort iteration if after yielding the marking has been aborted.
172 if (_may_yield && _cm->do_yield_check() && _cm->has_aborted()) {
173 return true;
174 }
175 // Repeat the asserts from before the start of the closure. We will do them
176 // as asserts here to minimize their overhead on the product. However, we
177 // will have them as guarantees at the beginning / end of the bitmap
178 // clearing to get some checking in the product.
179 assert(!_may_yield || _cm->cmThread()->during_cycle(), "invariant");
180 assert(!_may_yield || !G1CollectedHeap::heap()->mark_in_progress(), "invariant");
181 }
182
183 return false;
184 }
185 };
186
187 class ParClearNextMarkBitmapTask : public AbstractGangTask {
188 ClearBitmapHRClosure* _cl;
189 HeapRegionClaimer _hrclaimer;
190 bool _suspendible; // If the task is suspendible, workers must join the STS.
191
192 public:
193 ParClearNextMarkBitmapTask(ClearBitmapHRClosure *cl, uint n_workers, bool suspendible) :
194 _cl(cl), _suspendible(suspendible), AbstractGangTask("Parallel Clear Bitmap Task"), _hrclaimer(n_workers) {}
195
196 void work(uint worker_id) {
197 SuspendibleThreadSetJoiner sts_join(_suspendible);
198 G1CollectedHeap::heap()->heap_region_par_iterate(_cl, worker_id, &_hrclaimer, true);
199 }
200 };
201
202 void CMBitMap::clearAll() {
203 G1CollectedHeap* g1h = G1CollectedHeap::heap();
204 ClearBitmapHRClosure cl(NULL, this, false /* may_yield */);
205 uint n_workers = g1h->workers()->active_workers();
206 ParClearNextMarkBitmapTask task(&cl, n_workers, false);
207 g1h->workers()->run_task(&task);
208 guarantee(cl.complete(), "Must have completed iteration.");
209 return;
210 }
211
212 void CMBitMap::markRange(MemRegion mr) {
213 mr.intersection(MemRegion(_bmStartWord, _bmWordSize));
214 assert(!mr.is_empty(), "unexpected empty region");
215 assert((offsetToHeapWord(heapWordToOffset(mr.end())) ==
216 ((HeapWord *) mr.end())),
217 "markRange memory region end is not card aligned");
218 // convert address range into offset range
219 _bm.at_put_range(heapWordToOffset(mr.start()),
220 heapWordToOffset(mr.end()), true);
221 }
222
223 void CMBitMap::clearRange(MemRegion mr) {
224 mr.intersection(MemRegion(_bmStartWord, _bmWordSize));
225 assert(!mr.is_empty(), "unexpected empty region");
226 // convert address range into offset range
227 _bm.at_put_range(heapWordToOffset(mr.start()),
228 heapWordToOffset(mr.end()), false);
229 }
230
231 MemRegion CMBitMap::getAndClearMarkedRegion(HeapWord* addr,
232 HeapWord* end_addr) {
233 HeapWord* start = getNextMarkedWordAddress(addr);
234 start = MIN2(start, end_addr);
235 HeapWord* end = getNextUnmarkedWordAddress(start);
236 end = MIN2(end, end_addr);
237 assert(start <= end, "Consistency check");
238 MemRegion mr(start, end);
239 if (!mr.is_empty()) {
240 clearRange(mr);
241 }
242 return mr;
243 }
244
245 CMMarkStack::CMMarkStack(ConcurrentMark* cm) :
246 _base(NULL), _cm(cm)
247 #ifdef ASSERT
248 , _drain_in_progress(false)
249 , _drain_in_progress_yields(false)
250 #endif
251 {}
252
253 bool CMMarkStack::allocate(size_t capacity) {
254 // allocate a stack of the requisite depth
255 ReservedSpace rs(ReservedSpace::allocation_align_size_up(capacity * sizeof(oop)));
256 if (!rs.is_reserved()) {
257 warning("ConcurrentMark MarkStack allocation failure");
258 return false;
259 }
260 MemTracker::record_virtual_memory_type((address)rs.base(), mtGC);
261 if (!_virtual_space.initialize(rs, rs.size())) {
262 warning("ConcurrentMark MarkStack backing store failure");
263 // Release the virtual memory reserved for the marking stack
264 rs.release();
265 return false;
266 }
267 assert(_virtual_space.committed_size() == rs.size(),
268 "Didn't reserve backing store for all of ConcurrentMark stack?");
269 _base = (oop*) _virtual_space.low();
270 setEmpty();
271 _capacity = (jint) capacity;
272 _saved_index = -1;
273 _should_expand = false;
274 return true;
275 }
276
277 void CMMarkStack::expand() {
278 // Called, during remark, if we've overflown the marking stack during marking.
279 assert(isEmpty(), "stack should been emptied while handling overflow");
280 assert(_capacity <= (jint) MarkStackSizeMax, "stack bigger than permitted");
281 // Clear expansion flag
282 _should_expand = false;
283 if (_capacity == (jint) MarkStackSizeMax) {
284 if (PrintGCDetails && Verbose) {
285 gclog_or_tty->print_cr(" (benign) Can't expand marking stack capacity, at max size limit");
286 }
287 return;
288 }
289 // Double capacity if possible
290 jint new_capacity = MIN2(_capacity*2, (jint) MarkStackSizeMax);
291 // Do not give up existing stack until we have managed to
292 // get the double capacity that we desired.
293 ReservedSpace rs(ReservedSpace::allocation_align_size_up(new_capacity *
294 sizeof(oop)));
295 if (rs.is_reserved()) {
296 // Release the backing store associated with old stack
297 _virtual_space.release();
298 // Reinitialize virtual space for new stack
299 if (!_virtual_space.initialize(rs, rs.size())) {
300 fatal("Not enough swap for expanded marking stack capacity");
301 }
302 _base = (oop*)(_virtual_space.low());
303 _index = 0;
304 _capacity = new_capacity;
305 } else {
306 if (PrintGCDetails && Verbose) {
307 // Failed to double capacity, continue;
308 gclog_or_tty->print(" (benign) Failed to expand marking stack capacity from "
309 SIZE_FORMAT "K to " SIZE_FORMAT "K",
310 _capacity / K, new_capacity / K);
311 }
312 }
313 }
314
315 void CMMarkStack::set_should_expand() {
316 // If we're resetting the marking state because of an
317 // marking stack overflow, record that we should, if
318 // possible, expand the stack.
319 _should_expand = _cm->has_overflown();
320 }
321
322 CMMarkStack::~CMMarkStack() {
323 if (_base != NULL) {
324 _base = NULL;
325 _virtual_space.release();
326 }
327 }
328
329 void CMMarkStack::par_push_arr(oop* ptr_arr, int n) {
330 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
331 jint start = _index;
332 jint next_index = start + n;
333 if (next_index > _capacity) {
334 _overflow = true;
335 return;
336 }
337 // Otherwise.
338 _index = next_index;
339 for (int i = 0; i < n; i++) {
340 int ind = start + i;
341 assert(ind < _capacity, "By overflow test above.");
342 _base[ind] = ptr_arr[i];
343 }
344 }
345
346 bool CMMarkStack::par_pop_arr(oop* ptr_arr, int max, int* n) {
347 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
348 jint index = _index;
349 if (index == 0) {
350 *n = 0;
351 return false;
352 } else {
353 int k = MIN2(max, index);
354 jint new_ind = index - k;
355 for (int j = 0; j < k; j++) {
356 ptr_arr[j] = _base[new_ind + j];
357 }
358 _index = new_ind;
359 *n = k;
360 return true;
361 }
362 }
363
364 template<class OopClosureClass>
365 bool CMMarkStack::drain(OopClosureClass* cl, CMBitMap* bm, bool yield_after) {
366 assert(!_drain_in_progress || !_drain_in_progress_yields || yield_after
367 || SafepointSynchronize::is_at_safepoint(),
368 "Drain recursion must be yield-safe.");
369 bool res = true;
370 debug_only(_drain_in_progress = true);
371 debug_only(_drain_in_progress_yields = yield_after);
372 while (!isEmpty()) {
373 oop newOop = pop();
374 assert(G1CollectedHeap::heap()->is_in_reserved(newOop), "Bad pop");
375 assert(newOop->is_oop(), "Expected an oop");
376 assert(bm == NULL || bm->isMarked((HeapWord*)newOop),
377 "only grey objects on this stack");
378 newOop->oop_iterate(cl);
379 if (yield_after && _cm->do_yield_check()) {
380 res = false;
381 break;
382 }
383 }
384 debug_only(_drain_in_progress = false);
385 return res;
386 }
387
388 void CMMarkStack::note_start_of_gc() {
389 assert(_saved_index == -1,
390 "note_start_of_gc()/end_of_gc() bracketed incorrectly");
391 _saved_index = _index;
392 }
393
394 void CMMarkStack::note_end_of_gc() {
395 // This is intentionally a guarantee, instead of an assert. If we
396 // accidentally add something to the mark stack during GC, it
397 // will be a correctness issue so it's better if we crash. we'll
398 // only check this once per GC anyway, so it won't be a performance
399 // issue in any way.
400 guarantee(_saved_index == _index,
401 err_msg("saved index: %d index: %d", _saved_index, _index));
402 _saved_index = -1;
403 }
404
405 void CMMarkStack::oops_do(OopClosure* f) {
406 assert(_saved_index == _index,
407 err_msg("saved index: %d index: %d", _saved_index, _index));
408 for (int i = 0; i < _index; i += 1) {
409 f->do_oop(&_base[i]);
410 }
411 }
412
413 CMRootRegions::CMRootRegions() :
414 _young_list(NULL), _cm(NULL), _scan_in_progress(false),
415 _should_abort(false), _next_survivor(NULL) { }
416
417 void CMRootRegions::init(G1CollectedHeap* g1h, ConcurrentMark* cm) {
418 _young_list = g1h->young_list();
419 _cm = cm;
420 }
421
422 void CMRootRegions::prepare_for_scan() {
423 assert(!scan_in_progress(), "pre-condition");
424
425 // Currently, only survivors can be root regions.
426 assert(_next_survivor == NULL, "pre-condition");
427 _next_survivor = _young_list->first_survivor_region();
428 _scan_in_progress = (_next_survivor != NULL);
429 _should_abort = false;
430 }
431
432 HeapRegion* CMRootRegions::claim_next() {
433 if (_should_abort) {
434 // If someone has set the should_abort flag, we return NULL to
435 // force the caller to bail out of their loop.
436 return NULL;
437 }
438
439 // Currently, only survivors can be root regions.
440 HeapRegion* res = _next_survivor;
441 if (res != NULL) {
442 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
443 // Read it again in case it changed while we were waiting for the lock.
444 res = _next_survivor;
445 if (res != NULL) {
446 if (res == _young_list->last_survivor_region()) {
447 // We just claimed the last survivor so store NULL to indicate
448 // that we're done.
449 _next_survivor = NULL;
450 } else {
451 _next_survivor = res->get_next_young_region();
452 }
453 } else {
454 // Someone else claimed the last survivor while we were trying
455 // to take the lock so nothing else to do.
456 }
457 }
458 assert(res == NULL || res->is_survivor(), "post-condition");
459
460 return res;
461 }
462
463 void CMRootRegions::scan_finished() {
464 assert(scan_in_progress(), "pre-condition");
465
466 // Currently, only survivors can be root regions.
467 if (!_should_abort) {
468 assert(_next_survivor == NULL, "we should have claimed all survivors");
469 }
470 _next_survivor = NULL;
471
472 {
473 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
474 _scan_in_progress = false;
475 RootRegionScan_lock->notify_all();
476 }
477 }
478
479 bool CMRootRegions::wait_until_scan_finished() {
480 if (!scan_in_progress()) return false;
481
482 {
483 MutexLockerEx x(RootRegionScan_lock, Mutex::_no_safepoint_check_flag);
484 while (scan_in_progress()) {
485 RootRegionScan_lock->wait(Mutex::_no_safepoint_check_flag);
486 }
487 }
488 return true;
489 }
490
491 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
492 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list
493 #endif // _MSC_VER
494
495 uint ConcurrentMark::scale_parallel_threads(uint n_par_threads) {
496 return MAX2((n_par_threads + 2) / 4, 1U);
497 }
498
499 ConcurrentMark::ConcurrentMark(G1CollectedHeap* g1h, G1RegionToSpaceMapper* prev_bitmap_storage, G1RegionToSpaceMapper* next_bitmap_storage) :
500 _g1h(g1h),
501 _markBitMap1(),
502 _markBitMap2(),
503 _parallel_marking_threads(0),
504 _max_parallel_marking_threads(0),
505 _sleep_factor(0.0),
506 _marking_task_overhead(1.0),
507 _cleanup_sleep_factor(0.0),
508 _cleanup_task_overhead(1.0),
509 _cleanup_list("Cleanup List"),
510 _region_bm((BitMap::idx_t)(g1h->max_regions()), false /* in_resource_area*/),
511 _card_bm((g1h->reserved_region().byte_size() + CardTableModRefBS::card_size - 1) >>
512 CardTableModRefBS::card_shift,
513 false /* in_resource_area*/),
514
515 _prevMarkBitMap(&_markBitMap1),
516 _nextMarkBitMap(&_markBitMap2),
517
518 _markStack(this),
519 // _finger set in set_non_marking_state
520
521 _max_worker_id(MAX2((uint)ParallelGCThreads, 1U)),
522 // _active_tasks set in set_non_marking_state
523 // _tasks set inside the constructor
524 _task_queues(new CMTaskQueueSet((int) _max_worker_id)),
525 _terminator(ParallelTaskTerminator((int) _max_worker_id, _task_queues)),
526
527 _has_overflown(false),
528 _concurrent(false),
529 _has_aborted(false),
530 _aborted_gc_id(GCId::undefined()),
531 _restart_for_overflow(false),
532 _concurrent_marking_in_progress(false),
533
534 // _verbose_level set below
535
536 _init_times(),
537 _remark_times(), _remark_mark_times(), _remark_weak_ref_times(),
538 _cleanup_times(),
539 _total_counting_time(0.0),
540 _total_rs_scrub_time(0.0),
541
542 _parallel_workers(NULL),
543
544 _count_card_bitmaps(NULL),
545 _count_marked_bytes(NULL),
546 _completed_initialization(false) {
547 CMVerboseLevel verbose_level = (CMVerboseLevel) G1MarkingVerboseLevel;
548 if (verbose_level < no_verbose) {
549 verbose_level = no_verbose;
550 }
551 if (verbose_level > high_verbose) {
552 verbose_level = high_verbose;
553 }
554 _verbose_level = verbose_level;
555
556 if (verbose_low()) {
557 gclog_or_tty->print_cr("[global] init, heap start = " PTR_FORMAT ", "
558 "heap end = " PTR_FORMAT, p2i(_heap_start), p2i(_heap_end));
559 }
560
561 _markBitMap1.initialize(g1h->reserved_region(), prev_bitmap_storage);
562 _markBitMap2.initialize(g1h->reserved_region(), next_bitmap_storage);
563
564 // Create & start a ConcurrentMark thread.
565 _cmThread = new ConcurrentMarkThread(this);
566 assert(cmThread() != NULL, "CM Thread should have been created");
567 assert(cmThread()->cm() != NULL, "CM Thread should refer to this cm");
568 if (_cmThread->osthread() == NULL) {
569 vm_shutdown_during_initialization("Could not create ConcurrentMarkThread");
570 }
571
572 assert(CGC_lock != NULL, "Where's the CGC_lock?");
573 assert(_markBitMap1.covers(g1h->reserved_region()), "_markBitMap1 inconsistency");
574 assert(_markBitMap2.covers(g1h->reserved_region()), "_markBitMap2 inconsistency");
575
576 SATBMarkQueueSet& satb_qs = JavaThread::satb_mark_queue_set();
577 satb_qs.set_buffer_size(G1SATBBufferSize);
578
579 _root_regions.init(_g1h, this);
580
581 if (ConcGCThreads > ParallelGCThreads) {
582 warning("Can't have more ConcGCThreads (" UINTX_FORMAT ") "
583 "than ParallelGCThreads (" UINTX_FORMAT ").",
584 ConcGCThreads, ParallelGCThreads);
585 return;
586 }
587 if (!FLAG_IS_DEFAULT(ConcGCThreads) && ConcGCThreads > 0) {
588 // Note: ConcGCThreads has precedence over G1MarkingOverheadPercent
589 // if both are set
590 _sleep_factor = 0.0;
591 _marking_task_overhead = 1.0;
592 } else if (G1MarkingOverheadPercent > 0) {
593 // We will calculate the number of parallel marking threads based
594 // on a target overhead with respect to the soft real-time goal
595 double marking_overhead = (double) G1MarkingOverheadPercent / 100.0;
596 double overall_cm_overhead =
597 (double) MaxGCPauseMillis * marking_overhead /
598 (double) GCPauseIntervalMillis;
599 double cpu_ratio = 1.0 / (double) os::processor_count();
600 double marking_thread_num = ceil(overall_cm_overhead / cpu_ratio);
601 double marking_task_overhead =
602 overall_cm_overhead / marking_thread_num *
603 (double) os::processor_count();
604 double sleep_factor =
605 (1.0 - marking_task_overhead) / marking_task_overhead;
606
607 FLAG_SET_ERGO(uintx, ConcGCThreads, (uint) marking_thread_num);
608 _sleep_factor = sleep_factor;
609 _marking_task_overhead = marking_task_overhead;
610 } else {
611 // Calculate the number of parallel marking threads by scaling
612 // the number of parallel GC threads.
613 uint marking_thread_num = scale_parallel_threads((uint) ParallelGCThreads);
614 FLAG_SET_ERGO(uintx, ConcGCThreads, marking_thread_num);
615 _sleep_factor = 0.0;
616 _marking_task_overhead = 1.0;
617 }
618
619 assert(ConcGCThreads > 0, "Should have been set");
620 _parallel_marking_threads = (uint) ConcGCThreads;
621 _max_parallel_marking_threads = _parallel_marking_threads;
622
623 if (parallel_marking_threads() > 1) {
624 _cleanup_task_overhead = 1.0;
625 } else {
626 _cleanup_task_overhead = marking_task_overhead();
627 }
628 _cleanup_sleep_factor =
629 (1.0 - cleanup_task_overhead()) / cleanup_task_overhead();
630
631 #if 0
632 gclog_or_tty->print_cr("Marking Threads %d", parallel_marking_threads());
633 gclog_or_tty->print_cr("CM Marking Task Overhead %1.4lf", marking_task_overhead());
634 gclog_or_tty->print_cr("CM Sleep Factor %1.4lf", sleep_factor());
635 gclog_or_tty->print_cr("CL Marking Task Overhead %1.4lf", cleanup_task_overhead());
636 gclog_or_tty->print_cr("CL Sleep Factor %1.4lf", cleanup_sleep_factor());
637 #endif
638
639 _parallel_workers = new FlexibleWorkGang("G1 Marker",
640 _max_parallel_marking_threads, false, true);
641 if (_parallel_workers == NULL) {
642 vm_exit_during_initialization("Failed necessary allocation.");
643 } else {
644 _parallel_workers->initialize_workers();
645 }
646
647 if (FLAG_IS_DEFAULT(MarkStackSize)) {
648 size_t mark_stack_size =
649 MIN2(MarkStackSizeMax,
650 MAX2(MarkStackSize, (size_t) (parallel_marking_threads() * TASKQUEUE_SIZE)));
651 // Verify that the calculated value for MarkStackSize is in range.
652 // It would be nice to use the private utility routine from Arguments.
653 if (!(mark_stack_size >= 1 && mark_stack_size <= MarkStackSizeMax)) {
654 warning("Invalid value calculated for MarkStackSize (" SIZE_FORMAT "): "
655 "must be between 1 and " SIZE_FORMAT,
656 mark_stack_size, MarkStackSizeMax);
657 return;
658 }
659 FLAG_SET_ERGO(size_t, MarkStackSize, mark_stack_size);
660 } else {
661 // Verify MarkStackSize is in range.
662 if (FLAG_IS_CMDLINE(MarkStackSize)) {
663 if (FLAG_IS_DEFAULT(MarkStackSizeMax)) {
664 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) {
665 warning("Invalid value specified for MarkStackSize (" SIZE_FORMAT "): "
666 "must be between 1 and " SIZE_FORMAT,
667 MarkStackSize, MarkStackSizeMax);
668 return;
669 }
670 } else if (FLAG_IS_CMDLINE(MarkStackSizeMax)) {
671 if (!(MarkStackSize >= 1 && MarkStackSize <= MarkStackSizeMax)) {
672 warning("Invalid value specified for MarkStackSize (" SIZE_FORMAT ")"
673 " or for MarkStackSizeMax (" SIZE_FORMAT ")",
674 MarkStackSize, MarkStackSizeMax);
675 return;
676 }
677 }
678 }
679 }
680
681 if (!_markStack.allocate(MarkStackSize)) {
682 warning("Failed to allocate CM marking stack");
683 return;
684 }
685
686 _tasks = NEW_C_HEAP_ARRAY(CMTask*, _max_worker_id, mtGC);
687 _accum_task_vtime = NEW_C_HEAP_ARRAY(double, _max_worker_id, mtGC);
688
689 _count_card_bitmaps = NEW_C_HEAP_ARRAY(BitMap, _max_worker_id, mtGC);
690 _count_marked_bytes = NEW_C_HEAP_ARRAY(size_t*, _max_worker_id, mtGC);
691
692 BitMap::idx_t card_bm_size = _card_bm.size();
693
694 // so that the assertion in MarkingTaskQueue::task_queue doesn't fail
695 _active_tasks = _max_worker_id;
696
697 uint max_regions = _g1h->max_regions();
698 for (uint i = 0; i < _max_worker_id; ++i) {
699 CMTaskQueue* task_queue = new CMTaskQueue();
700 task_queue->initialize();
701 _task_queues->register_queue(i, task_queue);
702
703 _count_card_bitmaps[i] = BitMap(card_bm_size, false);
704 _count_marked_bytes[i] = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC);
705
706 _tasks[i] = new CMTask(i, this,
707 _count_marked_bytes[i],
708 &_count_card_bitmaps[i],
709 task_queue, _task_queues);
710
711 _accum_task_vtime[i] = 0.0;
712 }
713
714 // Calculate the card number for the bottom of the heap. Used
715 // in biasing indexes into the accounting card bitmaps.
716 _heap_bottom_card_num =
717 intptr_t(uintptr_t(_g1h->reserved_region().start()) >>
718 CardTableModRefBS::card_shift);
719
720 // Clear all the liveness counting data
721 clear_all_count_data();
722
723 // so that the call below can read a sensible value
724 _heap_start = g1h->reserved_region().start();
725 set_non_marking_state();
726 _completed_initialization = true;
727 }
728
729 void ConcurrentMark::reset() {
730 // Starting values for these two. This should be called in a STW
731 // phase.
732 MemRegion reserved = _g1h->g1_reserved();
733 _heap_start = reserved.start();
734 _heap_end = reserved.end();
735
736 // Separated the asserts so that we know which one fires.
737 assert(_heap_start != NULL, "heap bounds should look ok");
738 assert(_heap_end != NULL, "heap bounds should look ok");
739 assert(_heap_start < _heap_end, "heap bounds should look ok");
740
741 // Reset all the marking data structures and any necessary flags
742 reset_marking_state();
743
744 if (verbose_low()) {
745 gclog_or_tty->print_cr("[global] resetting");
746 }
747
748 // We do reset all of them, since different phases will use
749 // different number of active threads. So, it's easiest to have all
750 // of them ready.
751 for (uint i = 0; i < _max_worker_id; ++i) {
752 _tasks[i]->reset(_nextMarkBitMap);
753 }
754
755 // we need this to make sure that the flag is on during the evac
756 // pause with initial mark piggy-backed
757 set_concurrent_marking_in_progress();
758 }
759
760
761 void ConcurrentMark::reset_marking_state(bool clear_overflow) {
762 _markStack.set_should_expand();
763 _markStack.setEmpty(); // Also clears the _markStack overflow flag
764 if (clear_overflow) {
765 clear_has_overflown();
766 } else {
767 assert(has_overflown(), "pre-condition");
768 }
769 _finger = _heap_start;
770
771 for (uint i = 0; i < _max_worker_id; ++i) {
772 CMTaskQueue* queue = _task_queues->queue(i);
773 queue->set_empty();
774 }
775 }
776
777 void ConcurrentMark::set_concurrency(uint active_tasks) {
778 assert(active_tasks <= _max_worker_id, "we should not have more");
779
780 _active_tasks = active_tasks;
781 // Need to update the three data structures below according to the
782 // number of active threads for this phase.
783 _terminator = ParallelTaskTerminator((int) active_tasks, _task_queues);
784 _first_overflow_barrier_sync.set_n_workers((int) active_tasks);
785 _second_overflow_barrier_sync.set_n_workers((int) active_tasks);
786 }
787
788 void ConcurrentMark::set_concurrency_and_phase(uint active_tasks, bool concurrent) {
789 set_concurrency(active_tasks);
790
791 _concurrent = concurrent;
792 // We propagate this to all tasks, not just the active ones.
793 for (uint i = 0; i < _max_worker_id; ++i)
794 _tasks[i]->set_concurrent(concurrent);
795
796 if (concurrent) {
797 set_concurrent_marking_in_progress();
798 } else {
799 // We currently assume that the concurrent flag has been set to
800 // false before we start remark. At this point we should also be
801 // in a STW phase.
802 assert(!concurrent_marking_in_progress(), "invariant");
803 assert(out_of_regions(),
804 err_msg("only way to get here: _finger: " PTR_FORMAT ", _heap_end: " PTR_FORMAT,
805 p2i(_finger), p2i(_heap_end)));
806 }
807 }
808
809 void ConcurrentMark::set_non_marking_state() {
810 // We set the global marking state to some default values when we're
811 // not doing marking.
812 reset_marking_state();
813 _active_tasks = 0;
814 clear_concurrent_marking_in_progress();
815 }
816
817 ConcurrentMark::~ConcurrentMark() {
818 // The ConcurrentMark instance is never freed.
819 ShouldNotReachHere();
820 }
821
822 void ConcurrentMark::clearNextBitmap() {
823 G1CollectedHeap* g1h = G1CollectedHeap::heap();
824
825 // Make sure that the concurrent mark thread looks to still be in
826 // the current cycle.
827 guarantee(cmThread()->during_cycle(), "invariant");
828
829 // We are finishing up the current cycle by clearing the next
830 // marking bitmap and getting it ready for the next cycle. During
831 // this time no other cycle can start. So, let's make sure that this
832 // is the case.
833 guarantee(!g1h->mark_in_progress(), "invariant");
834
835 ClearBitmapHRClosure cl(this, _nextMarkBitMap, true /* may_yield */);
836 ParClearNextMarkBitmapTask task(&cl, parallel_marking_threads(), true);
837 _parallel_workers->run_task(&task);
838
839 // Clear the liveness counting data. If the marking has been aborted, the abort()
840 // call already did that.
841 if (cl.complete()) {
842 clear_all_count_data();
843 }
844
845 // Repeat the asserts from above.
846 guarantee(cmThread()->during_cycle(), "invariant");
847 guarantee(!g1h->mark_in_progress(), "invariant");
848 }
849
850 class CheckBitmapClearHRClosure : public HeapRegionClosure {
851 CMBitMap* _bitmap;
852 bool _error;
853 public:
854 CheckBitmapClearHRClosure(CMBitMap* bitmap) : _bitmap(bitmap) {
855 }
856
857 virtual bool doHeapRegion(HeapRegion* r) {
858 // This closure can be called concurrently to the mutator, so we must make sure
859 // that the result of the getNextMarkedWordAddress() call is compared to the
860 // value passed to it as limit to detect any found bits.
861 // We can use the region's orig_end() for the limit and the comparison value
862 // as it always contains the "real" end of the region that never changes and
863 // has no side effects.
864 // Due to the latter, there can also be no problem with the compiler generating
865 // reloads of the orig_end() call.
866 HeapWord* end = r->orig_end();
867 return _bitmap->getNextMarkedWordAddress(r->bottom(), end) != end;
868 }
869 };
870
871 bool ConcurrentMark::nextMarkBitmapIsClear() {
872 CheckBitmapClearHRClosure cl(_nextMarkBitMap);
873 _g1h->heap_region_iterate(&cl);
874 return cl.complete();
875 }
876
877 class NoteStartOfMarkHRClosure: public HeapRegionClosure {
878 public:
879 bool doHeapRegion(HeapRegion* r) {
880 if (!r->is_continues_humongous()) {
881 r->note_start_of_marking();
882 }
883 return false;
884 }
885 };
886
887 void ConcurrentMark::checkpointRootsInitialPre() {
888 G1CollectedHeap* g1h = G1CollectedHeap::heap();
889 G1CollectorPolicy* g1p = g1h->g1_policy();
890
891 _has_aborted = false;
892
893 // Initialize marking structures. This has to be done in a STW phase.
894 reset();
895
896 // For each region note start of marking.
897 NoteStartOfMarkHRClosure startcl;
898 g1h->heap_region_iterate(&startcl);
899 }
900
901
902 void ConcurrentMark::checkpointRootsInitialPost() {
903 G1CollectedHeap* g1h = G1CollectedHeap::heap();
904
905 // If we force an overflow during remark, the remark operation will
906 // actually abort and we'll restart concurrent marking. If we always
907 // force an overflow during remark we'll never actually complete the
908 // marking phase. So, we initialize this here, at the start of the
909 // cycle, so that at the remaining overflow number will decrease at
910 // every remark and we'll eventually not need to cause one.
911 force_overflow_stw()->init();
912
913 // Start Concurrent Marking weak-reference discovery.
914 ReferenceProcessor* rp = g1h->ref_processor_cm();
915 // enable ("weak") refs discovery
916 rp->enable_discovery();
917 rp->setup_policy(false); // snapshot the soft ref policy to be used in this cycle
918
919 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
920 // This is the start of the marking cycle, we're expected all
921 // threads to have SATB queues with active set to false.
922 satb_mq_set.set_active_all_threads(true, /* new active value */
923 false /* expected_active */);
924
925 _root_regions.prepare_for_scan();
926
927 // update_g1_committed() will be called at the end of an evac pause
928 // when marking is on. So, it's also called at the end of the
929 // initial-mark pause to update the heap end, if the heap expands
930 // during it. No need to call it here.
931 }
932
933 /*
934 * Notice that in the next two methods, we actually leave the STS
935 * during the barrier sync and join it immediately afterwards. If we
936 * do not do this, the following deadlock can occur: one thread could
937 * be in the barrier sync code, waiting for the other thread to also
938 * sync up, whereas another one could be trying to yield, while also
939 * waiting for the other threads to sync up too.
940 *
941 * Note, however, that this code is also used during remark and in
942 * this case we should not attempt to leave / enter the STS, otherwise
943 * we'll either hit an assert (debug / fastdebug) or deadlock
944 * (product). So we should only leave / enter the STS if we are
945 * operating concurrently.
946 *
947 * Because the thread that does the sync barrier has left the STS, it
948 * is possible to be suspended for a Full GC or an evacuation pause
949 * could occur. This is actually safe, since the entering the sync
950 * barrier is one of the last things do_marking_step() does, and it
951 * doesn't manipulate any data structures afterwards.
952 */
953
954 void ConcurrentMark::enter_first_sync_barrier(uint worker_id) {
955 bool barrier_aborted;
956
957 if (verbose_low()) {
958 gclog_or_tty->print_cr("[%u] entering first barrier", worker_id);
959 }
960
961 {
962 SuspendibleThreadSetLeaver sts_leave(concurrent());
963 barrier_aborted = !_first_overflow_barrier_sync.enter();
964 }
965
966 // at this point everyone should have synced up and not be doing any
967 // more work
968
969 if (verbose_low()) {
970 if (barrier_aborted) {
971 gclog_or_tty->print_cr("[%u] aborted first barrier", worker_id);
972 } else {
973 gclog_or_tty->print_cr("[%u] leaving first barrier", worker_id);
974 }
975 }
976
977 if (barrier_aborted) {
978 // If the barrier aborted we ignore the overflow condition and
979 // just abort the whole marking phase as quickly as possible.
980 return;
981 }
982
983 // If we're executing the concurrent phase of marking, reset the marking
984 // state; otherwise the marking state is reset after reference processing,
985 // during the remark pause.
986 // If we reset here as a result of an overflow during the remark we will
987 // see assertion failures from any subsequent set_concurrency_and_phase()
988 // calls.
989 if (concurrent()) {
990 // let the task associated with with worker 0 do this
991 if (worker_id == 0) {
992 // task 0 is responsible for clearing the global data structures
993 // We should be here because of an overflow. During STW we should
994 // not clear the overflow flag since we rely on it being true when
995 // we exit this method to abort the pause and restart concurrent
996 // marking.
997 reset_marking_state(true /* clear_overflow */);
998 force_overflow()->update();
999
1000 if (G1Log::fine()) {
1001 gclog_or_tty->gclog_stamp(concurrent_gc_id());
1002 gclog_or_tty->print_cr("[GC concurrent-mark-reset-for-overflow]");
1003 }
1004 }
1005 }
1006
1007 // after this, each task should reset its own data structures then
1008 // then go into the second barrier
1009 }
1010
1011 void ConcurrentMark::enter_second_sync_barrier(uint worker_id) {
1012 bool barrier_aborted;
1013
1014 if (verbose_low()) {
1015 gclog_or_tty->print_cr("[%u] entering second barrier", worker_id);
1016 }
1017
1018 {
1019 SuspendibleThreadSetLeaver sts_leave(concurrent());
1020 barrier_aborted = !_second_overflow_barrier_sync.enter();
1021 }
1022
1023 // at this point everything should be re-initialized and ready to go
1024
1025 if (verbose_low()) {
1026 if (barrier_aborted) {
1027 gclog_or_tty->print_cr("[%u] aborted second barrier", worker_id);
1028 } else {
1029 gclog_or_tty->print_cr("[%u] leaving second barrier", worker_id);
1030 }
1031 }
1032 }
1033
1034 #ifndef PRODUCT
1035 void ForceOverflowSettings::init() {
1036 _num_remaining = G1ConcMarkForceOverflow;
1037 _force = false;
1038 update();
1039 }
1040
1041 void ForceOverflowSettings::update() {
1042 if (_num_remaining > 0) {
1043 _num_remaining -= 1;
1044 _force = true;
1045 } else {
1046 _force = false;
1047 }
1048 }
1049
1050 bool ForceOverflowSettings::should_force() {
1051 if (_force) {
1052 _force = false;
1053 return true;
1054 } else {
1055 return false;
1056 }
1057 }
1058 #endif // !PRODUCT
1059
1060 class CMConcurrentMarkingTask: public AbstractGangTask {
1061 private:
1062 ConcurrentMark* _cm;
1063 ConcurrentMarkThread* _cmt;
1064
1065 public:
1066 void work(uint worker_id) {
1067 assert(Thread::current()->is_ConcurrentGC_thread(),
1068 "this should only be done by a conc GC thread");
1069 ResourceMark rm;
1070
1071 double start_vtime = os::elapsedVTime();
1072
1073 {
1074 SuspendibleThreadSetJoiner sts_join;
1075
1076 assert(worker_id < _cm->active_tasks(), "invariant");
1077 CMTask* the_task = _cm->task(worker_id);
1078 the_task->record_start_time();
1079 if (!_cm->has_aborted()) {
1080 do {
1081 double start_vtime_sec = os::elapsedVTime();
1082 double mark_step_duration_ms = G1ConcMarkStepDurationMillis;
1083
1084 the_task->do_marking_step(mark_step_duration_ms,
1085 true /* do_termination */,
1086 false /* is_serial*/);
1087
1088 double end_vtime_sec = os::elapsedVTime();
1089 double elapsed_vtime_sec = end_vtime_sec - start_vtime_sec;
1090 _cm->clear_has_overflown();
1091
1092 _cm->do_yield_check(worker_id);
1093
1094 jlong sleep_time_ms;
1095 if (!_cm->has_aborted() && the_task->has_aborted()) {
1096 sleep_time_ms =
1097 (jlong) (elapsed_vtime_sec * _cm->sleep_factor() * 1000.0);
1098 {
1099 SuspendibleThreadSetLeaver sts_leave;
1100 os::sleep(Thread::current(), sleep_time_ms, false);
1101 }
1102 }
1103 } while (!_cm->has_aborted() && the_task->has_aborted());
1104 }
1105 the_task->record_end_time();
1106 guarantee(!the_task->has_aborted() || _cm->has_aborted(), "invariant");
1107 }
1108
1109 double end_vtime = os::elapsedVTime();
1110 _cm->update_accum_task_vtime(worker_id, end_vtime - start_vtime);
1111 }
1112
1113 CMConcurrentMarkingTask(ConcurrentMark* cm,
1114 ConcurrentMarkThread* cmt) :
1115 AbstractGangTask("Concurrent Mark"), _cm(cm), _cmt(cmt) { }
1116
1117 ~CMConcurrentMarkingTask() { }
1118 };
1119
1120 // Calculates the number of active workers for a concurrent
1121 // phase.
1122 uint ConcurrentMark::calc_parallel_marking_threads() {
1123 uint n_conc_workers = 0;
1124 if (!UseDynamicNumberOfGCThreads ||
1125 (!FLAG_IS_DEFAULT(ConcGCThreads) &&
1126 !ForceDynamicNumberOfGCThreads)) {
1127 n_conc_workers = max_parallel_marking_threads();
1128 } else {
1129 n_conc_workers =
1130 AdaptiveSizePolicy::calc_default_active_workers(
1131 max_parallel_marking_threads(),
1132 1, /* Minimum workers */
1133 parallel_marking_threads(),
1134 Threads::number_of_non_daemon_threads());
1135 // Don't scale down "n_conc_workers" by scale_parallel_threads() because
1136 // that scaling has already gone into "_max_parallel_marking_threads".
1137 }
1138 assert(n_conc_workers > 0, "Always need at least 1");
1139 return n_conc_workers;
1140 }
1141
1142 void ConcurrentMark::scanRootRegion(HeapRegion* hr, uint worker_id) {
1143 // Currently, only survivors can be root regions.
1144 assert(hr->next_top_at_mark_start() == hr->bottom(), "invariant");
1145 G1RootRegionScanClosure cl(_g1h, this, worker_id);
1146
1147 const uintx interval = PrefetchScanIntervalInBytes;
1148 HeapWord* curr = hr->bottom();
1149 const HeapWord* end = hr->top();
1150 while (curr < end) {
1151 Prefetch::read(curr, interval);
1152 oop obj = oop(curr);
1153 int size = obj->oop_iterate(&cl);
1154 assert(size == obj->size(), "sanity");
1155 curr += size;
1156 }
1157 }
1158
1159 class CMRootRegionScanTask : public AbstractGangTask {
1160 private:
1161 ConcurrentMark* _cm;
1162
1163 public:
1164 CMRootRegionScanTask(ConcurrentMark* cm) :
1165 AbstractGangTask("Root Region Scan"), _cm(cm) { }
1166
1167 void work(uint worker_id) {
1168 assert(Thread::current()->is_ConcurrentGC_thread(),
1169 "this should only be done by a conc GC thread");
1170
1171 CMRootRegions* root_regions = _cm->root_regions();
1172 HeapRegion* hr = root_regions->claim_next();
1173 while (hr != NULL) {
1174 _cm->scanRootRegion(hr, worker_id);
1175 hr = root_regions->claim_next();
1176 }
1177 }
1178 };
1179
1180 void ConcurrentMark::scanRootRegions() {
1181 // Start of concurrent marking.
1182 ClassLoaderDataGraph::clear_claimed_marks();
1183
1184 // scan_in_progress() will have been set to true only if there was
1185 // at least one root region to scan. So, if it's false, we
1186 // should not attempt to do any further work.
1187 if (root_regions()->scan_in_progress()) {
1188 _parallel_marking_threads = calc_parallel_marking_threads();
1189 assert(parallel_marking_threads() <= max_parallel_marking_threads(),
1190 "Maximum number of marking threads exceeded");
1191 uint active_workers = MAX2(1U, parallel_marking_threads());
1192
1193 CMRootRegionScanTask task(this);
1194 _parallel_workers->set_active_workers(active_workers);
1195 _parallel_workers->run_task(&task);
1196
1197 // It's possible that has_aborted() is true here without actually
1198 // aborting the survivor scan earlier. This is OK as it's
1199 // mainly used for sanity checking.
1200 root_regions()->scan_finished();
1201 }
1202 }
1203
1204 void ConcurrentMark::markFromRoots() {
1205 // we might be tempted to assert that:
1206 // assert(asynch == !SafepointSynchronize::is_at_safepoint(),
1207 // "inconsistent argument?");
1208 // However that wouldn't be right, because it's possible that
1209 // a safepoint is indeed in progress as a younger generation
1210 // stop-the-world GC happens even as we mark in this generation.
1211
1212 _restart_for_overflow = false;
1213 force_overflow_conc()->init();
1214
1215 // _g1h has _n_par_threads
1216 _parallel_marking_threads = calc_parallel_marking_threads();
1217 assert(parallel_marking_threads() <= max_parallel_marking_threads(),
1218 "Maximum number of marking threads exceeded");
1219
1220 uint active_workers = MAX2(1U, parallel_marking_threads());
1221
1222 // Parallel task terminator is set in "set_concurrency_and_phase()"
1223 set_concurrency_and_phase(active_workers, true /* concurrent */);
1224
1225 CMConcurrentMarkingTask markingTask(this, cmThread());
1226 _parallel_workers->set_active_workers(active_workers);
1227 // Don't set _n_par_threads because it affects MT in process_roots()
1228 // and the decisions on that MT processing is made elsewhere.
1229 assert(_parallel_workers->active_workers() > 0, "Should have been set");
1230 _parallel_workers->run_task(&markingTask);
1231 print_stats();
1232 }
1233
1234 // Helper class to get rid of some boilerplate code.
1235 class G1CMTraceTime : public GCTraceTime {
1236 static bool doit_and_prepend(bool doit) {
1237 if (doit) {
1238 gclog_or_tty->put(' ');
1239 }
1240 return doit;
1241 }
1242
1243 public:
1244 G1CMTraceTime(const char* title, bool doit)
1245 : GCTraceTime(title, doit_and_prepend(doit), false, G1CollectedHeap::heap()->gc_timer_cm(),
1246 G1CollectedHeap::heap()->concurrent_mark()->concurrent_gc_id()) {
1247 }
1248 };
1249
1250 void ConcurrentMark::checkpointRootsFinal(bool clear_all_soft_refs) {
1251 // world is stopped at this checkpoint
1252 assert(SafepointSynchronize::is_at_safepoint(),
1253 "world should be stopped");
1254
1255 G1CollectedHeap* g1h = G1CollectedHeap::heap();
1256
1257 // If a full collection has happened, we shouldn't do this.
1258 if (has_aborted()) {
1259 g1h->set_marking_complete(); // So bitmap clearing isn't confused
1260 return;
1261 }
1262
1263 SvcGCMarker sgcm(SvcGCMarker::OTHER);
1264
1265 if (VerifyDuringGC) {
1266 HandleMark hm; // handle scope
1267 g1h->prepare_for_verify();
1268 Universe::verify(VerifyOption_G1UsePrevMarking,
1269 " VerifyDuringGC:(before)");
1270 }
1271 g1h->check_bitmaps("Remark Start");
1272
1273 G1CollectorPolicy* g1p = g1h->g1_policy();
1274 g1p->record_concurrent_mark_remark_start();
1275
1276 double start = os::elapsedTime();
1277
1278 checkpointRootsFinalWork();
1279
1280 double mark_work_end = os::elapsedTime();
1281
1282 weakRefsWork(clear_all_soft_refs);
1283
1284 if (has_overflown()) {
1285 // Oops. We overflowed. Restart concurrent marking.
1286 _restart_for_overflow = true;
1287 if (G1TraceMarkStackOverflow) {
1288 gclog_or_tty->print_cr("\nRemark led to restart for overflow.");
1289 }
1290
1291 // Verify the heap w.r.t. the previous marking bitmap.
1292 if (VerifyDuringGC) {
1293 HandleMark hm; // handle scope
1294 g1h->prepare_for_verify();
1295 Universe::verify(VerifyOption_G1UsePrevMarking,
1296 " VerifyDuringGC:(overflow)");
1297 }
1298
1299 // Clear the marking state because we will be restarting
1300 // marking due to overflowing the global mark stack.
1301 reset_marking_state();
1302 } else {
1303 {
1304 G1CMTraceTime trace("GC aggregate-data", G1Log::finer());
1305
1306 // Aggregate the per-task counting data that we have accumulated
1307 // while marking.
1308 aggregate_count_data();
1309 }
1310
1311 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
1312 // We're done with marking.
1313 // This is the end of the marking cycle, we're expected all
1314 // threads to have SATB queues with active set to true.
1315 satb_mq_set.set_active_all_threads(false, /* new active value */
1316 true /* expected_active */);
1317
1318 if (VerifyDuringGC) {
1319 HandleMark hm; // handle scope
1320 g1h->prepare_for_verify();
1321 Universe::verify(VerifyOption_G1UseNextMarking,
1322 " VerifyDuringGC:(after)");
1323 }
1324 g1h->check_bitmaps("Remark End");
1325 assert(!restart_for_overflow(), "sanity");
1326 // Completely reset the marking state since marking completed
1327 set_non_marking_state();
1328 }
1329
1330 // Expand the marking stack, if we have to and if we can.
1331 if (_markStack.should_expand()) {
1332 _markStack.expand();
1333 }
1334
1335 // Statistics
1336 double now = os::elapsedTime();
1337 _remark_mark_times.add((mark_work_end - start) * 1000.0);
1338 _remark_weak_ref_times.add((now - mark_work_end) * 1000.0);
1339 _remark_times.add((now - start) * 1000.0);
1340
1341 g1p->record_concurrent_mark_remark_end();
1342
1343 G1CMIsAliveClosure is_alive(g1h);
1344 g1h->gc_tracer_cm()->report_object_count_after_gc(&is_alive);
1345 }
1346
1347 // Base class of the closures that finalize and verify the
1348 // liveness counting data.
1349 class CMCountDataClosureBase: public HeapRegionClosure {
1350 protected:
1351 G1CollectedHeap* _g1h;
1352 ConcurrentMark* _cm;
1353 CardTableModRefBS* _ct_bs;
1354
1355 BitMap* _region_bm;
1356 BitMap* _card_bm;
1357
1358 // Takes a region that's not empty (i.e., it has at least one
1359 // live object in it and sets its corresponding bit on the region
1360 // bitmap to 1. If the region is "starts humongous" it will also set
1361 // to 1 the bits on the region bitmap that correspond to its
1362 // associated "continues humongous" regions.
1363 void set_bit_for_region(HeapRegion* hr) {
1364 assert(!hr->is_continues_humongous(), "should have filtered those out");
1365
1366 BitMap::idx_t index = (BitMap::idx_t) hr->hrm_index();
1367 if (!hr->is_starts_humongous()) {
1368 // Normal (non-humongous) case: just set the bit.
1369 _region_bm->par_at_put(index, true);
1370 } else {
1371 // Starts humongous case: calculate how many regions are part of
1372 // this humongous region and then set the bit range.
1373 BitMap::idx_t end_index = (BitMap::idx_t) hr->last_hc_index();
1374 _region_bm->par_at_put_range(index, end_index, true);
1375 }
1376 }
1377
1378 public:
1379 CMCountDataClosureBase(G1CollectedHeap* g1h,
1380 BitMap* region_bm, BitMap* card_bm):
1381 _g1h(g1h), _cm(g1h->concurrent_mark()),
1382 _ct_bs(barrier_set_cast<CardTableModRefBS>(g1h->barrier_set())),
1383 _region_bm(region_bm), _card_bm(card_bm) { }
1384 };
1385
1386 // Closure that calculates the # live objects per region. Used
1387 // for verification purposes during the cleanup pause.
1388 class CalcLiveObjectsClosure: public CMCountDataClosureBase {
1389 CMBitMapRO* _bm;
1390 size_t _region_marked_bytes;
1391
1392 public:
1393 CalcLiveObjectsClosure(CMBitMapRO *bm, G1CollectedHeap* g1h,
1394 BitMap* region_bm, BitMap* card_bm) :
1395 CMCountDataClosureBase(g1h, region_bm, card_bm),
1396 _bm(bm), _region_marked_bytes(0) { }
1397
1398 bool doHeapRegion(HeapRegion* hr) {
1399
1400 if (hr->is_continues_humongous()) {
1401 // We will ignore these here and process them when their
1402 // associated "starts humongous" region is processed (see
1403 // set_bit_for_heap_region()). Note that we cannot rely on their
1404 // associated "starts humongous" region to have their bit set to
1405 // 1 since, due to the region chunking in the parallel region
1406 // iteration, a "continues humongous" region might be visited
1407 // before its associated "starts humongous".
1408 return false;
1409 }
1410
1411 HeapWord* ntams = hr->next_top_at_mark_start();
1412 HeapWord* start = hr->bottom();
1413
1414 assert(start <= hr->end() && start <= ntams && ntams <= hr->end(),
1415 err_msg("Preconditions not met - "
1416 "start: " PTR_FORMAT ", ntams: " PTR_FORMAT ", end: " PTR_FORMAT,
1417 p2i(start), p2i(ntams), p2i(hr->end())));
1418
1419 // Find the first marked object at or after "start".
1420 start = _bm->getNextMarkedWordAddress(start, ntams);
1421
1422 size_t marked_bytes = 0;
1423
1424 while (start < ntams) {
1425 oop obj = oop(start);
1426 int obj_sz = obj->size();
1427 HeapWord* obj_end = start + obj_sz;
1428
1429 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(start);
1430 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(obj_end);
1431
1432 // Note: if we're looking at the last region in heap - obj_end
1433 // could be actually just beyond the end of the heap; end_idx
1434 // will then correspond to a (non-existent) card that is also
1435 // just beyond the heap.
1436 if (_g1h->is_in_g1_reserved(obj_end) && !_ct_bs->is_card_aligned(obj_end)) {
1437 // end of object is not card aligned - increment to cover
1438 // all the cards spanned by the object
1439 end_idx += 1;
1440 }
1441
1442 // Set the bits in the card BM for the cards spanned by this object.
1443 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */);
1444
1445 // Add the size of this object to the number of marked bytes.
1446 marked_bytes += (size_t)obj_sz * HeapWordSize;
1447
1448 // Find the next marked object after this one.
1449 start = _bm->getNextMarkedWordAddress(obj_end, ntams);
1450 }
1451
1452 // Mark the allocated-since-marking portion...
1453 HeapWord* top = hr->top();
1454 if (ntams < top) {
1455 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(ntams);
1456 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(top);
1457
1458 // Note: if we're looking at the last region in heap - top
1459 // could be actually just beyond the end of the heap; end_idx
1460 // will then correspond to a (non-existent) card that is also
1461 // just beyond the heap.
1462 if (_g1h->is_in_g1_reserved(top) && !_ct_bs->is_card_aligned(top)) {
1463 // end of object is not card aligned - increment to cover
1464 // all the cards spanned by the object
1465 end_idx += 1;
1466 }
1467 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */);
1468
1469 // This definitely means the region has live objects.
1470 set_bit_for_region(hr);
1471 }
1472
1473 // Update the live region bitmap.
1474 if (marked_bytes > 0) {
1475 set_bit_for_region(hr);
1476 }
1477
1478 // Set the marked bytes for the current region so that
1479 // it can be queried by a calling verification routine
1480 _region_marked_bytes = marked_bytes;
1481
1482 return false;
1483 }
1484
1485 size_t region_marked_bytes() const { return _region_marked_bytes; }
1486 };
1487
1488 // Heap region closure used for verifying the counting data
1489 // that was accumulated concurrently and aggregated during
1490 // the remark pause. This closure is applied to the heap
1491 // regions during the STW cleanup pause.
1492
1493 class VerifyLiveObjectDataHRClosure: public HeapRegionClosure {
1494 G1CollectedHeap* _g1h;
1495 ConcurrentMark* _cm;
1496 CalcLiveObjectsClosure _calc_cl;
1497 BitMap* _region_bm; // Region BM to be verified
1498 BitMap* _card_bm; // Card BM to be verified
1499 bool _verbose; // verbose output?
1500
1501 BitMap* _exp_region_bm; // Expected Region BM values
1502 BitMap* _exp_card_bm; // Expected card BM values
1503
1504 int _failures;
1505
1506 public:
1507 VerifyLiveObjectDataHRClosure(G1CollectedHeap* g1h,
1508 BitMap* region_bm,
1509 BitMap* card_bm,
1510 BitMap* exp_region_bm,
1511 BitMap* exp_card_bm,
1512 bool verbose) :
1513 _g1h(g1h), _cm(g1h->concurrent_mark()),
1514 _calc_cl(_cm->nextMarkBitMap(), g1h, exp_region_bm, exp_card_bm),
1515 _region_bm(region_bm), _card_bm(card_bm), _verbose(verbose),
1516 _exp_region_bm(exp_region_bm), _exp_card_bm(exp_card_bm),
1517 _failures(0) { }
1518
1519 int failures() const { return _failures; }
1520
1521 bool doHeapRegion(HeapRegion* hr) {
1522 if (hr->is_continues_humongous()) {
1523 // We will ignore these here and process them when their
1524 // associated "starts humongous" region is processed (see
1525 // set_bit_for_heap_region()). Note that we cannot rely on their
1526 // associated "starts humongous" region to have their bit set to
1527 // 1 since, due to the region chunking in the parallel region
1528 // iteration, a "continues humongous" region might be visited
1529 // before its associated "starts humongous".
1530 return false;
1531 }
1532
1533 int failures = 0;
1534
1535 // Call the CalcLiveObjectsClosure to walk the marking bitmap for
1536 // this region and set the corresponding bits in the expected region
1537 // and card bitmaps.
1538 bool res = _calc_cl.doHeapRegion(hr);
1539 assert(res == false, "should be continuing");
1540
1541 MutexLockerEx x((_verbose ? ParGCRareEvent_lock : NULL),
1542 Mutex::_no_safepoint_check_flag);
1543
1544 // Verify the marked bytes for this region.
1545 size_t exp_marked_bytes = _calc_cl.region_marked_bytes();
1546 size_t act_marked_bytes = hr->next_marked_bytes();
1547
1548 // We're not OK if expected marked bytes > actual marked bytes. It means
1549 // we have missed accounting some objects during the actual marking.
1550 if (exp_marked_bytes > act_marked_bytes) {
1551 if (_verbose) {
1552 gclog_or_tty->print_cr("Region %u: marked bytes mismatch: "
1553 "expected: " SIZE_FORMAT ", actual: " SIZE_FORMAT,
1554 hr->hrm_index(), exp_marked_bytes, act_marked_bytes);
1555 }
1556 failures += 1;
1557 }
1558
1559 // Verify the bit, for this region, in the actual and expected
1560 // (which was just calculated) region bit maps.
1561 // We're not OK if the bit in the calculated expected region
1562 // bitmap is set and the bit in the actual region bitmap is not.
1563 BitMap::idx_t index = (BitMap::idx_t) hr->hrm_index();
1564
1565 bool expected = _exp_region_bm->at(index);
1566 bool actual = _region_bm->at(index);
1567 if (expected && !actual) {
1568 if (_verbose) {
1569 gclog_or_tty->print_cr("Region %u: region bitmap mismatch: "
1570 "expected: %s, actual: %s",
1571 hr->hrm_index(),
1572 BOOL_TO_STR(expected), BOOL_TO_STR(actual));
1573 }
1574 failures += 1;
1575 }
1576
1577 // Verify that the card bit maps for the cards spanned by the current
1578 // region match. We have an error if we have a set bit in the expected
1579 // bit map and the corresponding bit in the actual bitmap is not set.
1580
1581 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(hr->bottom());
1582 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(hr->top());
1583
1584 for (BitMap::idx_t i = start_idx; i < end_idx; i+=1) {
1585 expected = _exp_card_bm->at(i);
1586 actual = _card_bm->at(i);
1587
1588 if (expected && !actual) {
1589 if (_verbose) {
1590 gclog_or_tty->print_cr("Region %u: card bitmap mismatch at " SIZE_FORMAT ": "
1591 "expected: %s, actual: %s",
1592 hr->hrm_index(), i,
1593 BOOL_TO_STR(expected), BOOL_TO_STR(actual));
1594 }
1595 failures += 1;
1596 }
1597 }
1598
1599 if (failures > 0 && _verbose) {
1600 gclog_or_tty->print_cr("Region " HR_FORMAT ", ntams: " PTR_FORMAT ", "
1601 "marked_bytes: calc/actual " SIZE_FORMAT "/" SIZE_FORMAT,
1602 HR_FORMAT_PARAMS(hr), p2i(hr->next_top_at_mark_start()),
1603 _calc_cl.region_marked_bytes(), hr->next_marked_bytes());
1604 }
1605
1606 _failures += failures;
1607
1608 // We could stop iteration over the heap when we
1609 // find the first violating region by returning true.
1610 return false;
1611 }
1612 };
1613
1614 class G1ParVerifyFinalCountTask: public AbstractGangTask {
1615 protected:
1616 G1CollectedHeap* _g1h;
1617 ConcurrentMark* _cm;
1618 BitMap* _actual_region_bm;
1619 BitMap* _actual_card_bm;
1620
1621 uint _n_workers;
1622
1623 BitMap* _expected_region_bm;
1624 BitMap* _expected_card_bm;
1625
1626 int _failures;
1627 bool _verbose;
1628
1629 HeapRegionClaimer _hrclaimer;
1630
1631 public:
1632 G1ParVerifyFinalCountTask(G1CollectedHeap* g1h,
1633 BitMap* region_bm, BitMap* card_bm,
1634 BitMap* expected_region_bm, BitMap* expected_card_bm)
1635 : AbstractGangTask("G1 verify final counting"),
1636 _g1h(g1h), _cm(_g1h->concurrent_mark()),
1637 _actual_region_bm(region_bm), _actual_card_bm(card_bm),
1638 _expected_region_bm(expected_region_bm), _expected_card_bm(expected_card_bm),
1639 _failures(0), _verbose(false),
1640 _n_workers(_g1h->workers()->active_workers()), _hrclaimer(_n_workers) {
1641 assert(VerifyDuringGC, "don't call this otherwise");
1642 assert(_expected_card_bm->size() == _actual_card_bm->size(), "sanity");
1643 assert(_expected_region_bm->size() == _actual_region_bm->size(), "sanity");
1644
1645 _verbose = _cm->verbose_medium();
1646 }
1647
1648 void work(uint worker_id) {
1649 assert(worker_id < _n_workers, "invariant");
1650
1651 VerifyLiveObjectDataHRClosure verify_cl(_g1h,
1652 _actual_region_bm, _actual_card_bm,
1653 _expected_region_bm,
1654 _expected_card_bm,
1655 _verbose);
1656
1657 _g1h->heap_region_par_iterate(&verify_cl, worker_id, &_hrclaimer);
1658
1659 Atomic::add(verify_cl.failures(), &_failures);
1660 }
1661
1662 int failures() const { return _failures; }
1663 };
1664
1665 // Closure that finalizes the liveness counting data.
1666 // Used during the cleanup pause.
1667 // Sets the bits corresponding to the interval [NTAMS, top]
1668 // (which contains the implicitly live objects) in the
1669 // card liveness bitmap. Also sets the bit for each region,
1670 // containing live data, in the region liveness bitmap.
1671
1672 class FinalCountDataUpdateClosure: public CMCountDataClosureBase {
1673 public:
1674 FinalCountDataUpdateClosure(G1CollectedHeap* g1h,
1675 BitMap* region_bm,
1676 BitMap* card_bm) :
1677 CMCountDataClosureBase(g1h, region_bm, card_bm) { }
1678
1679 bool doHeapRegion(HeapRegion* hr) {
1680
1681 if (hr->is_continues_humongous()) {
1682 // We will ignore these here and process them when their
1683 // associated "starts humongous" region is processed (see
1684 // set_bit_for_heap_region()). Note that we cannot rely on their
1685 // associated "starts humongous" region to have their bit set to
1686 // 1 since, due to the region chunking in the parallel region
1687 // iteration, a "continues humongous" region might be visited
1688 // before its associated "starts humongous".
1689 return false;
1690 }
1691
1692 HeapWord* ntams = hr->next_top_at_mark_start();
1693 HeapWord* top = hr->top();
1694
1695 assert(hr->bottom() <= ntams && ntams <= hr->end(), "Preconditions.");
1696
1697 // Mark the allocated-since-marking portion...
1698 if (ntams < top) {
1699 // This definitely means the region has live objects.
1700 set_bit_for_region(hr);
1701
1702 // Now set the bits in the card bitmap for [ntams, top)
1703 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(ntams);
1704 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(top);
1705
1706 // Note: if we're looking at the last region in heap - top
1707 // could be actually just beyond the end of the heap; end_idx
1708 // will then correspond to a (non-existent) card that is also
1709 // just beyond the heap.
1710 if (_g1h->is_in_g1_reserved(top) && !_ct_bs->is_card_aligned(top)) {
1711 // end of object is not card aligned - increment to cover
1712 // all the cards spanned by the object
1713 end_idx += 1;
1714 }
1715
1716 assert(end_idx <= _card_bm->size(),
1717 err_msg("oob: end_idx= " SIZE_FORMAT ", bitmap size= " SIZE_FORMAT,
1718 end_idx, _card_bm->size()));
1719 assert(start_idx < _card_bm->size(),
1720 err_msg("oob: start_idx= " SIZE_FORMAT ", bitmap size= " SIZE_FORMAT,
1721 start_idx, _card_bm->size()));
1722
1723 _cm->set_card_bitmap_range(_card_bm, start_idx, end_idx, true /* is_par */);
1724 }
1725
1726 // Set the bit for the region if it contains live data
1727 if (hr->next_marked_bytes() > 0) {
1728 set_bit_for_region(hr);
1729 }
1730
1731 return false;
1732 }
1733 };
1734
1735 class G1ParFinalCountTask: public AbstractGangTask {
1736 protected:
1737 G1CollectedHeap* _g1h;
1738 ConcurrentMark* _cm;
1739 BitMap* _actual_region_bm;
1740 BitMap* _actual_card_bm;
1741
1742 uint _n_workers;
1743 HeapRegionClaimer _hrclaimer;
1744
1745 public:
1746 G1ParFinalCountTask(G1CollectedHeap* g1h, BitMap* region_bm, BitMap* card_bm)
1747 : AbstractGangTask("G1 final counting"),
1748 _g1h(g1h), _cm(_g1h->concurrent_mark()),
1749 _actual_region_bm(region_bm), _actual_card_bm(card_bm),
1750 _n_workers(_g1h->workers()->active_workers()), _hrclaimer(_n_workers) {
1751 }
1752
1753 void work(uint worker_id) {
1754 assert(worker_id < _n_workers, "invariant");
1755
1756 FinalCountDataUpdateClosure final_update_cl(_g1h,
1757 _actual_region_bm,
1758 _actual_card_bm);
1759
1760 _g1h->heap_region_par_iterate(&final_update_cl, worker_id, &_hrclaimer);
1761 }
1762 };
1763
1764 class G1ParNoteEndTask;
1765
1766 class G1NoteEndOfConcMarkClosure : public HeapRegionClosure {
1767 G1CollectedHeap* _g1;
1768 size_t _max_live_bytes;
1769 uint _regions_claimed;
1770 size_t _freed_bytes;
1771 FreeRegionList* _local_cleanup_list;
1772 HeapRegionSetCount _old_regions_removed;
1773 HeapRegionSetCount _humongous_regions_removed;
1774 HRRSCleanupTask* _hrrs_cleanup_task;
1775 double _claimed_region_time;
1776 double _max_region_time;
1777
1778 public:
1779 G1NoteEndOfConcMarkClosure(G1CollectedHeap* g1,
1780 FreeRegionList* local_cleanup_list,
1781 HRRSCleanupTask* hrrs_cleanup_task) :
1782 _g1(g1),
1783 _max_live_bytes(0), _regions_claimed(0),
1784 _freed_bytes(0),
1785 _claimed_region_time(0.0), _max_region_time(0.0),
1786 _local_cleanup_list(local_cleanup_list),
1787 _old_regions_removed(),
1788 _humongous_regions_removed(),
1789 _hrrs_cleanup_task(hrrs_cleanup_task) { }
1790
1791 size_t freed_bytes() { return _freed_bytes; }
1792 const HeapRegionSetCount& old_regions_removed() { return _old_regions_removed; }
1793 const HeapRegionSetCount& humongous_regions_removed() { return _humongous_regions_removed; }
1794
1795 bool doHeapRegion(HeapRegion *hr) {
1796 if (hr->is_continues_humongous()) {
1797 return false;
1798 }
1799 // We use a claim value of zero here because all regions
1800 // were claimed with value 1 in the FinalCount task.
1801 _g1->reset_gc_time_stamps(hr);
1802 double start = os::elapsedTime();
1803 _regions_claimed++;
1804 hr->note_end_of_marking();
1805 _max_live_bytes += hr->max_live_bytes();
1806
1807 if (hr->used() > 0 && hr->max_live_bytes() == 0 && !hr->is_young()) {
1808 _freed_bytes += hr->used();
1809 hr->set_containing_set(NULL);
1810 if (hr->is_humongous()) {
1811 assert(hr->is_starts_humongous(), "we should only see starts humongous");
1812 _humongous_regions_removed.increment(1u, hr->capacity());
1813 _g1->free_humongous_region(hr, _local_cleanup_list, true);
1814 } else {
1815 _old_regions_removed.increment(1u, hr->capacity());
1816 _g1->free_region(hr, _local_cleanup_list, true);
1817 }
1818 } else {
1819 hr->rem_set()->do_cleanup_work(_hrrs_cleanup_task);
1820 }
1821
1822 double region_time = (os::elapsedTime() - start);
1823 _claimed_region_time += region_time;
1824 if (region_time > _max_region_time) {
1825 _max_region_time = region_time;
1826 }
1827 return false;
1828 }
1829
1830 size_t max_live_bytes() { return _max_live_bytes; }
1831 uint regions_claimed() { return _regions_claimed; }
1832 double claimed_region_time_sec() { return _claimed_region_time; }
1833 double max_region_time_sec() { return _max_region_time; }
1834 };
1835
1836 class G1ParNoteEndTask: public AbstractGangTask {
1837 friend class G1NoteEndOfConcMarkClosure;
1838
1839 protected:
1840 G1CollectedHeap* _g1h;
1841 size_t _max_live_bytes;
1842 size_t _freed_bytes;
1843 FreeRegionList* _cleanup_list;
1844 HeapRegionClaimer _hrclaimer;
1845
1846 public:
1847 G1ParNoteEndTask(G1CollectedHeap* g1h, FreeRegionList* cleanup_list, uint n_workers) :
1848 AbstractGangTask("G1 note end"), _g1h(g1h), _max_live_bytes(0), _freed_bytes(0), _cleanup_list(cleanup_list), _hrclaimer(n_workers) {
1849 }
1850
1851 void work(uint worker_id) {
1852 FreeRegionList local_cleanup_list("Local Cleanup List");
1853 HRRSCleanupTask hrrs_cleanup_task;
1854 G1NoteEndOfConcMarkClosure g1_note_end(_g1h, &local_cleanup_list,
1855 &hrrs_cleanup_task);
1856 _g1h->heap_region_par_iterate(&g1_note_end, worker_id, &_hrclaimer);
1857 assert(g1_note_end.complete(), "Shouldn't have yielded!");
1858
1859 // Now update the lists
1860 _g1h->remove_from_old_sets(g1_note_end.old_regions_removed(), g1_note_end.humongous_regions_removed());
1861 {
1862 MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
1863 _g1h->decrement_summary_bytes(g1_note_end.freed_bytes());
1864 _max_live_bytes += g1_note_end.max_live_bytes();
1865 _freed_bytes += g1_note_end.freed_bytes();
1866
1867 // If we iterate over the global cleanup list at the end of
1868 // cleanup to do this printing we will not guarantee to only
1869 // generate output for the newly-reclaimed regions (the list
1870 // might not be empty at the beginning of cleanup; we might
1871 // still be working on its previous contents). So we do the
1872 // printing here, before we append the new regions to the global
1873 // cleanup list.
1874
1875 G1HRPrinter* hr_printer = _g1h->hr_printer();
1876 if (hr_printer->is_active()) {
1877 FreeRegionListIterator iter(&local_cleanup_list);
1878 while (iter.more_available()) {
1879 HeapRegion* hr = iter.get_next();
1880 hr_printer->cleanup(hr);
1881 }
1882 }
1883
1884 _cleanup_list->add_ordered(&local_cleanup_list);
1885 assert(local_cleanup_list.is_empty(), "post-condition");
1886
1887 HeapRegionRemSet::finish_cleanup_task(&hrrs_cleanup_task);
1888 }
1889 }
1890 size_t max_live_bytes() { return _max_live_bytes; }
1891 size_t freed_bytes() { return _freed_bytes; }
1892 };
1893
1894 class G1ParScrubRemSetTask: public AbstractGangTask {
1895 protected:
1896 G1RemSet* _g1rs;
1897 BitMap* _region_bm;
1898 BitMap* _card_bm;
1899 HeapRegionClaimer _hrclaimer;
1900
1901 public:
1902 G1ParScrubRemSetTask(G1CollectedHeap* g1h, BitMap* region_bm, BitMap* card_bm, uint n_workers) :
1903 AbstractGangTask("G1 ScrubRS"), _g1rs(g1h->g1_rem_set()), _region_bm(region_bm), _card_bm(card_bm), _hrclaimer(n_workers) {
1904 }
1905
1906 void work(uint worker_id) {
1907 _g1rs->scrub(_region_bm, _card_bm, worker_id, &_hrclaimer);
1908 }
1909
1910 };
1911
1912 void ConcurrentMark::cleanup() {
1913 // world is stopped at this checkpoint
1914 assert(SafepointSynchronize::is_at_safepoint(),
1915 "world should be stopped");
1916 G1CollectedHeap* g1h = G1CollectedHeap::heap();
1917
1918 // If a full collection has happened, we shouldn't do this.
1919 if (has_aborted()) {
1920 g1h->set_marking_complete(); // So bitmap clearing isn't confused
1921 return;
1922 }
1923
1924 g1h->verify_region_sets_optional();
1925
1926 if (VerifyDuringGC) {
1927 HandleMark hm; // handle scope
1928 g1h->prepare_for_verify();
1929 Universe::verify(VerifyOption_G1UsePrevMarking,
1930 " VerifyDuringGC:(before)");
1931 }
1932 g1h->check_bitmaps("Cleanup Start");
1933
1934 G1CollectorPolicy* g1p = g1h->g1_policy();
1935 g1p->record_concurrent_mark_cleanup_start();
1936
1937 double start = os::elapsedTime();
1938
1939 HeapRegionRemSet::reset_for_cleanup_tasks();
1940
1941 uint n_workers;
1942
1943 // Do counting once more with the world stopped for good measure.
1944 G1ParFinalCountTask g1_par_count_task(g1h, &_region_bm, &_card_bm);
1945
1946 g1h->set_par_threads();
1947 n_workers = g1h->n_par_threads();
1948 assert(g1h->n_par_threads() == n_workers,
1949 "Should not have been reset");
1950 g1h->workers()->run_task(&g1_par_count_task);
1951 // Done with the parallel phase so reset to 0.
1952 g1h->set_par_threads(0);
1953
1954 if (VerifyDuringGC) {
1955 // Verify that the counting data accumulated during marking matches
1956 // that calculated by walking the marking bitmap.
1957
1958 // Bitmaps to hold expected values
1959 BitMap expected_region_bm(_region_bm.size(), true);
1960 BitMap expected_card_bm(_card_bm.size(), true);
1961
1962 G1ParVerifyFinalCountTask g1_par_verify_task(g1h,
1963 &_region_bm,
1964 &_card_bm,
1965 &expected_region_bm,
1966 &expected_card_bm);
1967
1968 g1h->set_par_threads((int)n_workers);
1969 g1h->workers()->run_task(&g1_par_verify_task);
1970 // Done with the parallel phase so reset to 0.
1971 g1h->set_par_threads(0);
1972
1973 guarantee(g1_par_verify_task.failures() == 0, "Unexpected accounting failures");
1974 }
1975
1976 size_t start_used_bytes = g1h->used();
1977 g1h->set_marking_complete();
1978
1979 double count_end = os::elapsedTime();
1980 double this_final_counting_time = (count_end - start);
1981 _total_counting_time += this_final_counting_time;
1982
1983 if (G1PrintRegionLivenessInfo) {
1984 G1PrintRegionLivenessInfoClosure cl(gclog_or_tty, "Post-Marking");
1985 _g1h->heap_region_iterate(&cl);
1986 }
1987
1988 // Install newly created mark bitMap as "prev".
1989 swapMarkBitMaps();
1990
1991 g1h->reset_gc_time_stamp();
1992
1993 // Note end of marking in all heap regions.
1994 G1ParNoteEndTask g1_par_note_end_task(g1h, &_cleanup_list, n_workers);
1995 g1h->set_par_threads((int)n_workers);
1996 g1h->workers()->run_task(&g1_par_note_end_task);
1997 g1h->set_par_threads(0);
1998 g1h->check_gc_time_stamps();
1999
2000 if (!cleanup_list_is_empty()) {
2001 // The cleanup list is not empty, so we'll have to process it
2002 // concurrently. Notify anyone else that might be wanting free
2003 // regions that there will be more free regions coming soon.
2004 g1h->set_free_regions_coming();
2005 }
2006
2007 // call below, since it affects the metric by which we sort the heap
2008 // regions.
2009 if (G1ScrubRemSets) {
2010 double rs_scrub_start = os::elapsedTime();
2011 G1ParScrubRemSetTask g1_par_scrub_rs_task(g1h, &_region_bm, &_card_bm, n_workers);
2012 g1h->set_par_threads((int)n_workers);
2013 g1h->workers()->run_task(&g1_par_scrub_rs_task);
2014 g1h->set_par_threads(0);
2015
2016 double rs_scrub_end = os::elapsedTime();
2017 double this_rs_scrub_time = (rs_scrub_end - rs_scrub_start);
2018 _total_rs_scrub_time += this_rs_scrub_time;
2019 }
2020
2021 // this will also free any regions totally full of garbage objects,
2022 // and sort the regions.
2023 g1h->g1_policy()->record_concurrent_mark_cleanup_end((int)n_workers);
2024
2025 // Statistics.
2026 double end = os::elapsedTime();
2027 _cleanup_times.add((end - start) * 1000.0);
2028
2029 if (G1Log::fine()) {
2030 g1h->g1_policy()->print_heap_transition(start_used_bytes);
2031 }
2032
2033 // Clean up will have freed any regions completely full of garbage.
2034 // Update the soft reference policy with the new heap occupancy.
2035 Universe::update_heap_info_at_gc();
2036
2037 if (VerifyDuringGC) {
2038 HandleMark hm; // handle scope
2039 g1h->prepare_for_verify();
2040 Universe::verify(VerifyOption_G1UsePrevMarking,
2041 " VerifyDuringGC:(after)");
2042 }
2043
2044 g1h->check_bitmaps("Cleanup End");
2045
2046 g1h->verify_region_sets_optional();
2047
2048 // We need to make this be a "collection" so any collection pause that
2049 // races with it goes around and waits for completeCleanup to finish.
2050 g1h->increment_total_collections();
2051
2052 // Clean out dead classes and update Metaspace sizes.
2053 if (ClassUnloadingWithConcurrentMark) {
2054 ClassLoaderDataGraph::purge();
2055 }
2056 MetaspaceGC::compute_new_size();
2057
2058 // We reclaimed old regions so we should calculate the sizes to make
2059 // sure we update the old gen/space data.
2060 g1h->g1mm()->update_sizes();
2061 g1h->allocation_context_stats().update_after_mark();
2062
2063 g1h->trace_heap_after_concurrent_cycle();
2064 }
2065
2066 void ConcurrentMark::completeCleanup() {
2067 if (has_aborted()) return;
2068
2069 G1CollectedHeap* g1h = G1CollectedHeap::heap();
2070
2071 _cleanup_list.verify_optional();
2072 FreeRegionList tmp_free_list("Tmp Free List");
2073
2074 if (G1ConcRegionFreeingVerbose) {
2075 gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : "
2076 "cleanup list has %u entries",
2077 _cleanup_list.length());
2078 }
2079
2080 // No one else should be accessing the _cleanup_list at this point,
2081 // so it is not necessary to take any locks
2082 while (!_cleanup_list.is_empty()) {
2083 HeapRegion* hr = _cleanup_list.remove_region(true /* from_head */);
2084 assert(hr != NULL, "Got NULL from a non-empty list");
2085 hr->par_clear();
2086 tmp_free_list.add_ordered(hr);
2087
2088 // Instead of adding one region at a time to the secondary_free_list,
2089 // we accumulate them in the local list and move them a few at a
2090 // time. This also cuts down on the number of notify_all() calls
2091 // we do during this process. We'll also append the local list when
2092 // _cleanup_list is empty (which means we just removed the last
2093 // region from the _cleanup_list).
2094 if ((tmp_free_list.length() % G1SecondaryFreeListAppendLength == 0) ||
2095 _cleanup_list.is_empty()) {
2096 if (G1ConcRegionFreeingVerbose) {
2097 gclog_or_tty->print_cr("G1ConcRegionFreeing [complete cleanup] : "
2098 "appending %u entries to the secondary_free_list, "
2099 "cleanup list still has %u entries",
2100 tmp_free_list.length(),
2101 _cleanup_list.length());
2102 }
2103
2104 {
2105 MutexLockerEx x(SecondaryFreeList_lock, Mutex::_no_safepoint_check_flag);
2106 g1h->secondary_free_list_add(&tmp_free_list);
2107 SecondaryFreeList_lock->notify_all();
2108 }
2109 #ifndef PRODUCT
2110 if (G1StressConcRegionFreeing) {
2111 for (uintx i = 0; i < G1StressConcRegionFreeingDelayMillis; ++i) {
2112 os::sleep(Thread::current(), (jlong) 1, false);
2113 }
2114 }
2115 #endif
2116 }
2117 }
2118 assert(tmp_free_list.is_empty(), "post-condition");
2119 }
2120
2121 // Supporting Object and Oop closures for reference discovery
2122 // and processing in during marking
2123
2124 bool G1CMIsAliveClosure::do_object_b(oop obj) {
2125 HeapWord* addr = (HeapWord*)obj;
2126 return addr != NULL &&
2127 (!_g1->is_in_g1_reserved(addr) || !_g1->is_obj_ill(obj));
2128 }
2129
2130 // 'Keep Alive' oop closure used by both serial parallel reference processing.
2131 // Uses the CMTask associated with a worker thread (for serial reference
2132 // processing the CMTask for worker 0 is used) to preserve (mark) and
2133 // trace referent objects.
2134 //
2135 // Using the CMTask and embedded local queues avoids having the worker
2136 // threads operating on the global mark stack. This reduces the risk
2137 // of overflowing the stack - which we would rather avoid at this late
2138 // state. Also using the tasks' local queues removes the potential
2139 // of the workers interfering with each other that could occur if
2140 // operating on the global stack.
2141
2142 class G1CMKeepAliveAndDrainClosure: public OopClosure {
2143 ConcurrentMark* _cm;
2144 CMTask* _task;
2145 int _ref_counter_limit;
2146 int _ref_counter;
2147 bool _is_serial;
2148 public:
2149 G1CMKeepAliveAndDrainClosure(ConcurrentMark* cm, CMTask* task, bool is_serial) :
2150 _cm(cm), _task(task), _is_serial(is_serial),
2151 _ref_counter_limit(G1RefProcDrainInterval) {
2152 assert(_ref_counter_limit > 0, "sanity");
2153 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code");
2154 _ref_counter = _ref_counter_limit;
2155 }
2156
2157 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
2158 virtual void do_oop( oop* p) { do_oop_work(p); }
2159
2160 template <class T> void do_oop_work(T* p) {
2161 if (!_cm->has_overflown()) {
2162 oop obj = oopDesc::load_decode_heap_oop(p);
2163 if (_cm->verbose_high()) {
2164 gclog_or_tty->print_cr("\t[%u] we're looking at location "
2165 "*" PTR_FORMAT " = " PTR_FORMAT,
2166 _task->worker_id(), p2i(p), p2i((void*) obj));
2167 }
2168
2169 _task->deal_with_reference(obj);
2170 _ref_counter--;
2171
2172 if (_ref_counter == 0) {
2173 // We have dealt with _ref_counter_limit references, pushing them
2174 // and objects reachable from them on to the local stack (and
2175 // possibly the global stack). Call CMTask::do_marking_step() to
2176 // process these entries.
2177 //
2178 // We call CMTask::do_marking_step() in a loop, which we'll exit if
2179 // there's nothing more to do (i.e. we're done with the entries that
2180 // were pushed as a result of the CMTask::deal_with_reference() calls
2181 // above) or we overflow.
2182 //
2183 // Note: CMTask::do_marking_step() can set the CMTask::has_aborted()
2184 // flag while there may still be some work to do. (See the comment at
2185 // the beginning of CMTask::do_marking_step() for those conditions -
2186 // one of which is reaching the specified time target.) It is only
2187 // when CMTask::do_marking_step() returns without setting the
2188 // has_aborted() flag that the marking step has completed.
2189 do {
2190 double mark_step_duration_ms = G1ConcMarkStepDurationMillis;
2191 _task->do_marking_step(mark_step_duration_ms,
2192 false /* do_termination */,
2193 _is_serial);
2194 } while (_task->has_aborted() && !_cm->has_overflown());
2195 _ref_counter = _ref_counter_limit;
2196 }
2197 } else {
2198 if (_cm->verbose_high()) {
2199 gclog_or_tty->print_cr("\t[%u] CM Overflow", _task->worker_id());
2200 }
2201 }
2202 }
2203 };
2204
2205 // 'Drain' oop closure used by both serial and parallel reference processing.
2206 // Uses the CMTask associated with a given worker thread (for serial
2207 // reference processing the CMtask for worker 0 is used). Calls the
2208 // do_marking_step routine, with an unbelievably large timeout value,
2209 // to drain the marking data structures of the remaining entries
2210 // added by the 'keep alive' oop closure above.
2211
2212 class G1CMDrainMarkingStackClosure: public VoidClosure {
2213 ConcurrentMark* _cm;
2214 CMTask* _task;
2215 bool _is_serial;
2216 public:
2217 G1CMDrainMarkingStackClosure(ConcurrentMark* cm, CMTask* task, bool is_serial) :
2218 _cm(cm), _task(task), _is_serial(is_serial) {
2219 assert(!_is_serial || _task->worker_id() == 0, "only task 0 for serial code");
2220 }
2221
2222 void do_void() {
2223 do {
2224 if (_cm->verbose_high()) {
2225 gclog_or_tty->print_cr("\t[%u] Drain: Calling do_marking_step - serial: %s",
2226 _task->worker_id(), BOOL_TO_STR(_is_serial));
2227 }
2228
2229 // We call CMTask::do_marking_step() to completely drain the local
2230 // and global marking stacks of entries pushed by the 'keep alive'
2231 // oop closure (an instance of G1CMKeepAliveAndDrainClosure above).
2232 //
2233 // CMTask::do_marking_step() is called in a loop, which we'll exit
2234 // if there's nothing more to do (i.e. we've completely drained the
2235 // entries that were pushed as a a result of applying the 'keep alive'
2236 // closure to the entries on the discovered ref lists) or we overflow
2237 // the global marking stack.
2238 //
2239 // Note: CMTask::do_marking_step() can set the CMTask::has_aborted()
2240 // flag while there may still be some work to do. (See the comment at
2241 // the beginning of CMTask::do_marking_step() for those conditions -
2242 // one of which is reaching the specified time target.) It is only
2243 // when CMTask::do_marking_step() returns without setting the
2244 // has_aborted() flag that the marking step has completed.
2245
2246 _task->do_marking_step(1000000000.0 /* something very large */,
2247 true /* do_termination */,
2248 _is_serial);
2249 } while (_task->has_aborted() && !_cm->has_overflown());
2250 }
2251 };
2252
2253 // Implementation of AbstractRefProcTaskExecutor for parallel
2254 // reference processing at the end of G1 concurrent marking
2255
2256 class G1CMRefProcTaskExecutor: public AbstractRefProcTaskExecutor {
2257 private:
2258 G1CollectedHeap* _g1h;
2259 ConcurrentMark* _cm;
2260 WorkGang* _workers;
2261 uint _active_workers;
2262
2263 public:
2264 G1CMRefProcTaskExecutor(G1CollectedHeap* g1h,
2265 ConcurrentMark* cm,
2266 WorkGang* workers,
2267 uint n_workers) :
2268 _g1h(g1h), _cm(cm),
2269 _workers(workers), _active_workers(n_workers) { }
2270
2271 // Executes the given task using concurrent marking worker threads.
2272 virtual void execute(ProcessTask& task);
2273 virtual void execute(EnqueueTask& task);
2274 };
2275
2276 class G1CMRefProcTaskProxy: public AbstractGangTask {
2277 typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask;
2278 ProcessTask& _proc_task;
2279 G1CollectedHeap* _g1h;
2280 ConcurrentMark* _cm;
2281
2282 public:
2283 G1CMRefProcTaskProxy(ProcessTask& proc_task,
2284 G1CollectedHeap* g1h,
2285 ConcurrentMark* cm) :
2286 AbstractGangTask("Process reference objects in parallel"),
2287 _proc_task(proc_task), _g1h(g1h), _cm(cm) {
2288 ReferenceProcessor* rp = _g1h->ref_processor_cm();
2289 assert(rp->processing_is_mt(), "shouldn't be here otherwise");
2290 }
2291
2292 virtual void work(uint worker_id) {
2293 ResourceMark rm;
2294 HandleMark hm;
2295 CMTask* task = _cm->task(worker_id);
2296 G1CMIsAliveClosure g1_is_alive(_g1h);
2297 G1CMKeepAliveAndDrainClosure g1_par_keep_alive(_cm, task, false /* is_serial */);
2298 G1CMDrainMarkingStackClosure g1_par_drain(_cm, task, false /* is_serial */);
2299
2300 _proc_task.work(worker_id, g1_is_alive, g1_par_keep_alive, g1_par_drain);
2301 }
2302 };
2303
2304 void G1CMRefProcTaskExecutor::execute(ProcessTask& proc_task) {
2305 assert(_workers != NULL, "Need parallel worker threads.");
2306 assert(_g1h->ref_processor_cm()->processing_is_mt(), "processing is not MT");
2307
2308 G1CMRefProcTaskProxy proc_task_proxy(proc_task, _g1h, _cm);
2309
2310 // We need to reset the concurrency level before each
2311 // proxy task execution, so that the termination protocol
2312 // and overflow handling in CMTask::do_marking_step() knows
2313 // how many workers to wait for.
2314 _cm->set_concurrency(_active_workers);
2315 _g1h->set_par_threads(_active_workers);
2316 _workers->run_task(&proc_task_proxy);
2317 _g1h->set_par_threads(0);
2318 }
2319
2320 class G1CMRefEnqueueTaskProxy: public AbstractGangTask {
2321 typedef AbstractRefProcTaskExecutor::EnqueueTask EnqueueTask;
2322 EnqueueTask& _enq_task;
2323
2324 public:
2325 G1CMRefEnqueueTaskProxy(EnqueueTask& enq_task) :
2326 AbstractGangTask("Enqueue reference objects in parallel"),
2327 _enq_task(enq_task) { }
2328
2329 virtual void work(uint worker_id) {
2330 _enq_task.work(worker_id);
2331 }
2332 };
2333
2334 void G1CMRefProcTaskExecutor::execute(EnqueueTask& enq_task) {
2335 assert(_workers != NULL, "Need parallel worker threads.");
2336 assert(_g1h->ref_processor_cm()->processing_is_mt(), "processing is not MT");
2337
2338 G1CMRefEnqueueTaskProxy enq_task_proxy(enq_task);
2339
2340 // Not strictly necessary but...
2341 //
2342 // We need to reset the concurrency level before each
2343 // proxy task execution, so that the termination protocol
2344 // and overflow handling in CMTask::do_marking_step() knows
2345 // how many workers to wait for.
2346 _cm->set_concurrency(_active_workers);
2347 _g1h->set_par_threads(_active_workers);
2348 _workers->run_task(&enq_task_proxy);
2349 _g1h->set_par_threads(0);
2350 }
2351
2352 void ConcurrentMark::weakRefsWorkParallelPart(BoolObjectClosure* is_alive, bool purged_classes) {
2353 G1CollectedHeap::heap()->parallel_cleaning(is_alive, true, true, purged_classes);
2354 }
2355
2356 void ConcurrentMark::weakRefsWork(bool clear_all_soft_refs) {
2357 if (has_overflown()) {
2358 // Skip processing the discovered references if we have
2359 // overflown the global marking stack. Reference objects
2360 // only get discovered once so it is OK to not
2361 // de-populate the discovered reference lists. We could have,
2362 // but the only benefit would be that, when marking restarts,
2363 // less reference objects are discovered.
2364 return;
2365 }
2366
2367 ResourceMark rm;
2368 HandleMark hm;
2369
2370 G1CollectedHeap* g1h = G1CollectedHeap::heap();
2371
2372 // Is alive closure.
2373 G1CMIsAliveClosure g1_is_alive(g1h);
2374
2375 // Inner scope to exclude the cleaning of the string and symbol
2376 // tables from the displayed time.
2377 {
2378 G1CMTraceTime t("GC ref-proc", G1Log::finer());
2379
2380 ReferenceProcessor* rp = g1h->ref_processor_cm();
2381
2382 // See the comment in G1CollectedHeap::ref_processing_init()
2383 // about how reference processing currently works in G1.
2384
2385 // Set the soft reference policy
2386 rp->setup_policy(clear_all_soft_refs);
2387 assert(_markStack.isEmpty(), "mark stack should be empty");
2388
2389 // Instances of the 'Keep Alive' and 'Complete GC' closures used
2390 // in serial reference processing. Note these closures are also
2391 // used for serially processing (by the the current thread) the
2392 // JNI references during parallel reference processing.
2393 //
2394 // These closures do not need to synchronize with the worker
2395 // threads involved in parallel reference processing as these
2396 // instances are executed serially by the current thread (e.g.
2397 // reference processing is not multi-threaded and is thus
2398 // performed by the current thread instead of a gang worker).
2399 //
2400 // The gang tasks involved in parallel reference processing create
2401 // their own instances of these closures, which do their own
2402 // synchronization among themselves.
2403 G1CMKeepAliveAndDrainClosure g1_keep_alive(this, task(0), true /* is_serial */);
2404 G1CMDrainMarkingStackClosure g1_drain_mark_stack(this, task(0), true /* is_serial */);
2405
2406 // We need at least one active thread. If reference processing
2407 // is not multi-threaded we use the current (VMThread) thread,
2408 // otherwise we use the work gang from the G1CollectedHeap and
2409 // we utilize all the worker threads we can.
2410 bool processing_is_mt = rp->processing_is_mt();
2411 uint active_workers = (processing_is_mt ? g1h->workers()->active_workers() : 1U);
2412 active_workers = MAX2(MIN2(active_workers, _max_worker_id), 1U);
2413
2414 // Parallel processing task executor.
2415 G1CMRefProcTaskExecutor par_task_executor(g1h, this,
2416 g1h->workers(), active_workers);
2417 AbstractRefProcTaskExecutor* executor = (processing_is_mt ? &par_task_executor : NULL);
2418
2419 // Set the concurrency level. The phase was already set prior to
2420 // executing the remark task.
2421 set_concurrency(active_workers);
2422
2423 // Set the degree of MT processing here. If the discovery was done MT,
2424 // the number of threads involved during discovery could differ from
2425 // the number of active workers. This is OK as long as the discovered
2426 // Reference lists are balanced (see balance_all_queues() and balance_queues()).
2427 rp->set_active_mt_degree(active_workers);
2428
2429 // Process the weak references.
2430 const ReferenceProcessorStats& stats =
2431 rp->process_discovered_references(&g1_is_alive,
2432 &g1_keep_alive,
2433 &g1_drain_mark_stack,
2434 executor,
2435 g1h->gc_timer_cm(),
2436 concurrent_gc_id());
2437 g1h->gc_tracer_cm()->report_gc_reference_stats(stats);
2438
2439 // The do_oop work routines of the keep_alive and drain_marking_stack
2440 // oop closures will set the has_overflown flag if we overflow the
2441 // global marking stack.
2442
2443 assert(_markStack.overflow() || _markStack.isEmpty(),
2444 "mark stack should be empty (unless it overflowed)");
2445
2446 if (_markStack.overflow()) {
2447 // This should have been done already when we tried to push an
2448 // entry on to the global mark stack. But let's do it again.
2449 set_has_overflown();
2450 }
2451
2452 assert(rp->num_q() == active_workers, "why not");
2453
2454 rp->enqueue_discovered_references(executor);
2455
2456 rp->verify_no_references_recorded();
2457 assert(!rp->discovery_enabled(), "Post condition");
2458 }
2459
2460 if (has_overflown()) {
2461 // We can not trust g1_is_alive if the marking stack overflowed
2462 return;
2463 }
2464
2465 assert(_markStack.isEmpty(), "Marking should have completed");
2466
2467 // Unload Klasses, String, Symbols, Code Cache, etc.
2468 {
2469 G1CMTraceTime trace("Unloading", G1Log::finer());
2470
2471 if (ClassUnloadingWithConcurrentMark) {
2472 bool purged_classes;
2473
2474 {
2475 G1CMTraceTime trace("System Dictionary Unloading", G1Log::finest());
2476 purged_classes = SystemDictionary::do_unloading(&g1_is_alive, false /* Defer klass cleaning */);
2477 }
2478
2479 {
2480 G1CMTraceTime trace("Parallel Unloading", G1Log::finest());
2481 weakRefsWorkParallelPart(&g1_is_alive, purged_classes);
2482 }
2483 }
2484
2485 if (G1StringDedup::is_enabled()) {
2486 G1CMTraceTime trace("String Deduplication Unlink", G1Log::finest());
2487 G1StringDedup::unlink(&g1_is_alive);
2488 }
2489 }
2490 }
2491
2492 void ConcurrentMark::swapMarkBitMaps() {
2493 CMBitMapRO* temp = _prevMarkBitMap;
2494 _prevMarkBitMap = (CMBitMapRO*)_nextMarkBitMap;
2495 _nextMarkBitMap = (CMBitMap*) temp;
2496 }
2497
2498 // Closure for marking entries in SATB buffers.
2499 class CMSATBBufferClosure : public SATBBufferClosure {
2500 private:
2501 CMTask* _task;
2502 G1CollectedHeap* _g1h;
2503
2504 // This is very similar to CMTask::deal_with_reference, but with
2505 // more relaxed requirements for the argument, so this must be more
2506 // circumspect about treating the argument as an object.
2507 void do_entry(void* entry) const {
2508 _task->increment_refs_reached();
2509 HeapRegion* hr = _g1h->heap_region_containing_raw(entry);
2510 if (entry < hr->next_top_at_mark_start()) {
2511 // Until we get here, we don't know whether entry refers to a valid
2512 // object; it could instead have been a stale reference.
2513 oop obj = static_cast<oop>(entry);
2514 assert(obj->is_oop(true /* ignore mark word */),
2515 err_msg("Invalid oop in SATB buffer: " PTR_FORMAT, p2i(obj)));
2516 _task->make_reference_grey(obj, hr);
2517 }
2518 }
2519
2520 public:
2521 CMSATBBufferClosure(CMTask* task, G1CollectedHeap* g1h)
2522 : _task(task), _g1h(g1h) { }
2523
2524 virtual void do_buffer(void** buffer, size_t size) {
2525 for (size_t i = 0; i < size; ++i) {
2526 do_entry(buffer[i]);
2527 }
2528 }
2529 };
2530
2531 class G1RemarkThreadsClosure : public ThreadClosure {
2532 CMSATBBufferClosure _cm_satb_cl;
2533 G1CMOopClosure _cm_cl;
2534 MarkingCodeBlobClosure _code_cl;
2535 int _thread_parity;
2536
2537 public:
2538 G1RemarkThreadsClosure(G1CollectedHeap* g1h, CMTask* task) :
2539 _cm_satb_cl(task, g1h),
2540 _cm_cl(g1h, g1h->concurrent_mark(), task),
2541 _code_cl(&_cm_cl, !CodeBlobToOopClosure::FixRelocations),
2542 _thread_parity(Threads::thread_claim_parity()) {}
2543
2544 void do_thread(Thread* thread) {
2545 if (thread->is_Java_thread()) {
2546 if (thread->claim_oops_do(true, _thread_parity)) {
2547 JavaThread* jt = (JavaThread*)thread;
2548
2549 // In theory it should not be neccessary to explicitly walk the nmethods to find roots for concurrent marking
2550 // however the liveness of oops reachable from nmethods have very complex lifecycles:
2551 // * Alive if on the stack of an executing method
2552 // * Weakly reachable otherwise
2553 // Some objects reachable from nmethods, such as the class loader (or klass_holder) of the receiver should be
2554 // live by the SATB invariant but other oops recorded in nmethods may behave differently.
2555 jt->nmethods_do(&_code_cl);
2556
2557 jt->satb_mark_queue().apply_closure_and_empty(&_cm_satb_cl);
2558 }
2559 } else if (thread->is_VM_thread()) {
2560 if (thread->claim_oops_do(true, _thread_parity)) {
2561 JavaThread::satb_mark_queue_set().shared_satb_queue()->apply_closure_and_empty(&_cm_satb_cl);
2562 }
2563 }
2564 }
2565 };
2566
2567 class CMRemarkTask: public AbstractGangTask {
2568 private:
2569 ConcurrentMark* _cm;
2570 public:
2571 void work(uint worker_id) {
2572 // Since all available tasks are actually started, we should
2573 // only proceed if we're supposed to be active.
2574 if (worker_id < _cm->active_tasks()) {
2575 CMTask* task = _cm->task(worker_id);
2576 task->record_start_time();
2577 {
2578 ResourceMark rm;
2579 HandleMark hm;
2580
2581 G1RemarkThreadsClosure threads_f(G1CollectedHeap::heap(), task);
2582 Threads::threads_do(&threads_f);
2583 }
2584
2585 do {
2586 task->do_marking_step(1000000000.0 /* something very large */,
2587 true /* do_termination */,
2588 false /* is_serial */);
2589 } while (task->has_aborted() && !_cm->has_overflown());
2590 // If we overflow, then we do not want to restart. We instead
2591 // want to abort remark and do concurrent marking again.
2592 task->record_end_time();
2593 }
2594 }
2595
2596 CMRemarkTask(ConcurrentMark* cm, uint active_workers) :
2597 AbstractGangTask("Par Remark"), _cm(cm) {
2598 _cm->terminator()->reset_for_reuse(active_workers);
2599 }
2600 };
2601
2602 void ConcurrentMark::checkpointRootsFinalWork() {
2603 ResourceMark rm;
2604 HandleMark hm;
2605 G1CollectedHeap* g1h = G1CollectedHeap::heap();
2606
2607 G1CMTraceTime trace("Finalize Marking", G1Log::finer());
2608
2609 g1h->ensure_parsability(false);
2610
2611 StrongRootsScope srs;
2612 // this is remark, so we'll use up all active threads
2613 uint active_workers = g1h->workers()->active_workers();
2614 if (active_workers == 0) {
2615 assert(active_workers > 0, "Should have been set earlier");
2616 active_workers = (uint) ParallelGCThreads;
2617 g1h->workers()->set_active_workers(active_workers);
2618 }
2619 set_concurrency_and_phase(active_workers, false /* concurrent */);
2620 // Leave _parallel_marking_threads at it's
2621 // value originally calculated in the ConcurrentMark
2622 // constructor and pass values of the active workers
2623 // through the gang in the task.
2624
2625 CMRemarkTask remarkTask(this, active_workers);
2626 // We will start all available threads, even if we decide that the
2627 // active_workers will be fewer. The extra ones will just bail out
2628 // immediately.
2629 g1h->set_par_threads(active_workers);
2630 g1h->workers()->run_task(&remarkTask);
2631 g1h->set_par_threads(0);
2632
2633 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
2634 guarantee(has_overflown() ||
2635 satb_mq_set.completed_buffers_num() == 0,
2636 err_msg("Invariant: has_overflown = %s, num buffers = %d",
2637 BOOL_TO_STR(has_overflown()),
2638 satb_mq_set.completed_buffers_num()));
2639
2640 print_stats();
2641 }
2642
2643 void ConcurrentMark::clearRangePrevBitmap(MemRegion mr) {
2644 // Note we are overriding the read-only view of the prev map here, via
2645 // the cast.
2646 ((CMBitMap*)_prevMarkBitMap)->clearRange(mr);
2647 }
2648
2649 void ConcurrentMark::clearRangeNextBitmap(MemRegion mr) {
2650 _nextMarkBitMap->clearRange(mr);
2651 }
2652
2653 HeapRegion*
2654 ConcurrentMark::claim_region(uint worker_id) {
2655 // "checkpoint" the finger
2656 HeapWord* finger = _finger;
2657
2658 // _heap_end will not change underneath our feet; it only changes at
2659 // yield points.
2660 while (finger < _heap_end) {
2661 assert(_g1h->is_in_g1_reserved(finger), "invariant");
2662
2663 // Note on how this code handles humongous regions. In the
2664 // normal case the finger will reach the start of a "starts
2665 // humongous" (SH) region. Its end will either be the end of the
2666 // last "continues humongous" (CH) region in the sequence, or the
2667 // standard end of the SH region (if the SH is the only region in
2668 // the sequence). That way claim_region() will skip over the CH
2669 // regions. However, there is a subtle race between a CM thread
2670 // executing this method and a mutator thread doing a humongous
2671 // object allocation. The two are not mutually exclusive as the CM
2672 // thread does not need to hold the Heap_lock when it gets
2673 // here. So there is a chance that claim_region() will come across
2674 // a free region that's in the progress of becoming a SH or a CH
2675 // region. In the former case, it will either
2676 // a) Miss the update to the region's end, in which case it will
2677 // visit every subsequent CH region, will find their bitmaps
2678 // empty, and do nothing, or
2679 // b) Will observe the update of the region's end (in which case
2680 // it will skip the subsequent CH regions).
2681 // If it comes across a region that suddenly becomes CH, the
2682 // scenario will be similar to b). So, the race between
2683 // claim_region() and a humongous object allocation might force us
2684 // to do a bit of unnecessary work (due to some unnecessary bitmap
2685 // iterations) but it should not introduce and correctness issues.
2686 HeapRegion* curr_region = _g1h->heap_region_containing_raw(finger);
2687
2688 // Above heap_region_containing_raw may return NULL as we always scan claim
2689 // until the end of the heap. In this case, just jump to the next region.
2690 HeapWord* end = curr_region != NULL ? curr_region->end() : finger + HeapRegion::GrainWords;
2691
2692 // Is the gap between reading the finger and doing the CAS too long?
2693 HeapWord* res = (HeapWord*) Atomic::cmpxchg_ptr(end, &_finger, finger);
2694 if (res == finger && curr_region != NULL) {
2695 // we succeeded
2696 HeapWord* bottom = curr_region->bottom();
2697 HeapWord* limit = curr_region->next_top_at_mark_start();
2698
2699 if (verbose_low()) {
2700 gclog_or_tty->print_cr("[%u] curr_region = " PTR_FORMAT " "
2701 "[" PTR_FORMAT ", " PTR_FORMAT "), "
2702 "limit = " PTR_FORMAT,
2703 worker_id, p2i(curr_region), p2i(bottom), p2i(end), p2i(limit));
2704 }
2705
2706 // notice that _finger == end cannot be guaranteed here since,
2707 // someone else might have moved the finger even further
2708 assert(_finger >= end, "the finger should have moved forward");
2709
2710 if (verbose_low()) {
2711 gclog_or_tty->print_cr("[%u] we were successful with region = "
2712 PTR_FORMAT, worker_id, p2i(curr_region));
2713 }
2714
2715 if (limit > bottom) {
2716 if (verbose_low()) {
2717 gclog_or_tty->print_cr("[%u] region " PTR_FORMAT " is not empty, "
2718 "returning it ", worker_id, p2i(curr_region));
2719 }
2720 return curr_region;
2721 } else {
2722 assert(limit == bottom,
2723 "the region limit should be at bottom");
2724 if (verbose_low()) {
2725 gclog_or_tty->print_cr("[%u] region " PTR_FORMAT " is empty, "
2726 "returning NULL", worker_id, p2i(curr_region));
2727 }
2728 // we return NULL and the caller should try calling
2729 // claim_region() again.
2730 return NULL;
2731 }
2732 } else {
2733 assert(_finger > finger, "the finger should have moved forward");
2734 if (verbose_low()) {
2735 if (curr_region == NULL) {
2736 gclog_or_tty->print_cr("[%u] found uncommitted region, moving finger, "
2737 "global finger = " PTR_FORMAT ", "
2738 "our finger = " PTR_FORMAT,
2739 worker_id, p2i(_finger), p2i(finger));
2740 } else {
2741 gclog_or_tty->print_cr("[%u] somebody else moved the finger, "
2742 "global finger = " PTR_FORMAT ", "
2743 "our finger = " PTR_FORMAT,
2744 worker_id, p2i(_finger), p2i(finger));
2745 }
2746 }
2747
2748 // read it again
2749 finger = _finger;
2750 }
2751 }
2752
2753 return NULL;
2754 }
2755
2756 #ifndef PRODUCT
2757 enum VerifyNoCSetOopsPhase {
2758 VerifyNoCSetOopsStack,
2759 VerifyNoCSetOopsQueues
2760 };
2761
2762 class VerifyNoCSetOopsClosure : public OopClosure, public ObjectClosure {
2763 private:
2764 G1CollectedHeap* _g1h;
2765 VerifyNoCSetOopsPhase _phase;
2766 int _info;
2767
2768 const char* phase_str() {
2769 switch (_phase) {
2770 case VerifyNoCSetOopsStack: return "Stack";
2771 case VerifyNoCSetOopsQueues: return "Queue";
2772 default: ShouldNotReachHere();
2773 }
2774 return NULL;
2775 }
2776
2777 void do_object_work(oop obj) {
2778 guarantee(!_g1h->obj_in_cs(obj),
2779 err_msg("obj: " PTR_FORMAT " in CSet, phase: %s, info: %d",
2780 p2i((void*) obj), phase_str(), _info));
2781 }
2782
2783 public:
2784 VerifyNoCSetOopsClosure() : _g1h(G1CollectedHeap::heap()) { }
2785
2786 void set_phase(VerifyNoCSetOopsPhase phase, int info = -1) {
2787 _phase = phase;
2788 _info = info;
2789 }
2790
2791 virtual void do_oop(oop* p) {
2792 oop obj = oopDesc::load_decode_heap_oop(p);
2793 do_object_work(obj);
2794 }
2795
2796 virtual void do_oop(narrowOop* p) {
2797 // We should not come across narrow oops while scanning marking
2798 // stacks
2799 ShouldNotReachHere();
2800 }
2801
2802 virtual void do_object(oop obj) {
2803 do_object_work(obj);
2804 }
2805 };
2806
2807 void ConcurrentMark::verify_no_cset_oops() {
2808 assert(SafepointSynchronize::is_at_safepoint(), "should be at a safepoint");
2809 if (!G1CollectedHeap::heap()->mark_in_progress()) {
2810 return;
2811 }
2812
2813 VerifyNoCSetOopsClosure cl;
2814
2815 // Verify entries on the global mark stack
2816 cl.set_phase(VerifyNoCSetOopsStack);
2817 _markStack.oops_do(&cl);
2818
2819 // Verify entries on the task queues
2820 for (uint i = 0; i < _max_worker_id; i += 1) {
2821 cl.set_phase(VerifyNoCSetOopsQueues, i);
2822 CMTaskQueue* queue = _task_queues->queue(i);
2823 queue->oops_do(&cl);
2824 }
2825
2826 // Verify the global finger
2827 HeapWord* global_finger = finger();
2828 if (global_finger != NULL && global_finger < _heap_end) {
2829 // The global finger always points to a heap region boundary. We
2830 // use heap_region_containing_raw() to get the containing region
2831 // given that the global finger could be pointing to a free region
2832 // which subsequently becomes continues humongous. If that
2833 // happens, heap_region_containing() will return the bottom of the
2834 // corresponding starts humongous region and the check below will
2835 // not hold any more.
2836 // Since we always iterate over all regions, we might get a NULL HeapRegion
2837 // here.
2838 HeapRegion* global_hr = _g1h->heap_region_containing_raw(global_finger);
2839 guarantee(global_hr == NULL || global_finger == global_hr->bottom(),
2840 err_msg("global finger: " PTR_FORMAT " region: " HR_FORMAT,
2841 p2i(global_finger), HR_FORMAT_PARAMS(global_hr)));
2842 }
2843
2844 // Verify the task fingers
2845 assert(parallel_marking_threads() <= _max_worker_id, "sanity");
2846 for (int i = 0; i < (int) parallel_marking_threads(); i += 1) {
2847 CMTask* task = _tasks[i];
2848 HeapWord* task_finger = task->finger();
2849 if (task_finger != NULL && task_finger < _heap_end) {
2850 // See above note on the global finger verification.
2851 HeapRegion* task_hr = _g1h->heap_region_containing_raw(task_finger);
2852 guarantee(task_hr == NULL || task_finger == task_hr->bottom() ||
2853 !task_hr->in_collection_set(),
2854 err_msg("task finger: " PTR_FORMAT " region: " HR_FORMAT,
2855 p2i(task_finger), HR_FORMAT_PARAMS(task_hr)));
2856 }
2857 }
2858 }
2859 #endif // PRODUCT
2860
2861 // Aggregate the counting data that was constructed concurrently
2862 // with marking.
2863 class AggregateCountDataHRClosure: public HeapRegionClosure {
2864 G1CollectedHeap* _g1h;
2865 ConcurrentMark* _cm;
2866 CardTableModRefBS* _ct_bs;
2867 BitMap* _cm_card_bm;
2868 uint _max_worker_id;
2869
2870 public:
2871 AggregateCountDataHRClosure(G1CollectedHeap* g1h,
2872 BitMap* cm_card_bm,
2873 uint max_worker_id) :
2874 _g1h(g1h), _cm(g1h->concurrent_mark()),
2875 _ct_bs(barrier_set_cast<CardTableModRefBS>(g1h->barrier_set())),
2876 _cm_card_bm(cm_card_bm), _max_worker_id(max_worker_id) { }
2877
2878 bool doHeapRegion(HeapRegion* hr) {
2879 if (hr->is_continues_humongous()) {
2880 // We will ignore these here and process them when their
2881 // associated "starts humongous" region is processed.
2882 // Note that we cannot rely on their associated
2883 // "starts humongous" region to have their bit set to 1
2884 // since, due to the region chunking in the parallel region
2885 // iteration, a "continues humongous" region might be visited
2886 // before its associated "starts humongous".
2887 return false;
2888 }
2889
2890 HeapWord* start = hr->bottom();
2891 HeapWord* limit = hr->next_top_at_mark_start();
2892 HeapWord* end = hr->end();
2893
2894 assert(start <= limit && limit <= hr->top() && hr->top() <= hr->end(),
2895 err_msg("Preconditions not met - "
2896 "start: " PTR_FORMAT ", limit: " PTR_FORMAT ", "
2897 "top: " PTR_FORMAT ", end: " PTR_FORMAT,
2898 p2i(start), p2i(limit), p2i(hr->top()), p2i(hr->end())));
2899
2900 assert(hr->next_marked_bytes() == 0, "Precondition");
2901
2902 if (start == limit) {
2903 // NTAMS of this region has not been set so nothing to do.
2904 return false;
2905 }
2906
2907 // 'start' should be in the heap.
2908 assert(_g1h->is_in_g1_reserved(start) && _ct_bs->is_card_aligned(start), "sanity");
2909 // 'end' *may* be just beyond the end of the heap (if hr is the last region)
2910 assert(!_g1h->is_in_g1_reserved(end) || _ct_bs->is_card_aligned(end), "sanity");
2911
2912 BitMap::idx_t start_idx = _cm->card_bitmap_index_for(start);
2913 BitMap::idx_t limit_idx = _cm->card_bitmap_index_for(limit);
2914 BitMap::idx_t end_idx = _cm->card_bitmap_index_for(end);
2915
2916 // If ntams is not card aligned then we bump card bitmap index
2917 // for limit so that we get the all the cards spanned by
2918 // the object ending at ntams.
2919 // Note: if this is the last region in the heap then ntams
2920 // could be actually just beyond the end of the the heap;
2921 // limit_idx will then correspond to a (non-existent) card
2922 // that is also outside the heap.
2923 if (_g1h->is_in_g1_reserved(limit) && !_ct_bs->is_card_aligned(limit)) {
2924 limit_idx += 1;
2925 }
2926
2927 assert(limit_idx <= end_idx, "or else use atomics");
2928
2929 // Aggregate the "stripe" in the count data associated with hr.
2930 uint hrm_index = hr->hrm_index();
2931 size_t marked_bytes = 0;
2932
2933 for (uint i = 0; i < _max_worker_id; i += 1) {
2934 size_t* marked_bytes_array = _cm->count_marked_bytes_array_for(i);
2935 BitMap* task_card_bm = _cm->count_card_bitmap_for(i);
2936
2937 // Fetch the marked_bytes in this region for task i and
2938 // add it to the running total for this region.
2939 marked_bytes += marked_bytes_array[hrm_index];
2940
2941 // Now union the bitmaps[0,max_worker_id)[start_idx..limit_idx)
2942 // into the global card bitmap.
2943 BitMap::idx_t scan_idx = task_card_bm->get_next_one_offset(start_idx, limit_idx);
2944
2945 while (scan_idx < limit_idx) {
2946 assert(task_card_bm->at(scan_idx) == true, "should be");
2947 _cm_card_bm->set_bit(scan_idx);
2948 assert(_cm_card_bm->at(scan_idx) == true, "should be");
2949
2950 // BitMap::get_next_one_offset() can handle the case when
2951 // its left_offset parameter is greater than its right_offset
2952 // parameter. It does, however, have an early exit if
2953 // left_offset == right_offset. So let's limit the value
2954 // passed in for left offset here.
2955 BitMap::idx_t next_idx = MIN2(scan_idx + 1, limit_idx);
2956 scan_idx = task_card_bm->get_next_one_offset(next_idx, limit_idx);
2957 }
2958 }
2959
2960 // Update the marked bytes for this region.
2961 hr->add_to_marked_bytes(marked_bytes);
2962
2963 // Next heap region
2964 return false;
2965 }
2966 };
2967
2968 class G1AggregateCountDataTask: public AbstractGangTask {
2969 protected:
2970 G1CollectedHeap* _g1h;
2971 ConcurrentMark* _cm;
2972 BitMap* _cm_card_bm;
2973 uint _max_worker_id;
2974 uint _active_workers;
2975 HeapRegionClaimer _hrclaimer;
2976
2977 public:
2978 G1AggregateCountDataTask(G1CollectedHeap* g1h,
2979 ConcurrentMark* cm,
2980 BitMap* cm_card_bm,
2981 uint max_worker_id,
2982 uint n_workers) :
2983 AbstractGangTask("Count Aggregation"),
2984 _g1h(g1h), _cm(cm), _cm_card_bm(cm_card_bm),
2985 _max_worker_id(max_worker_id),
2986 _active_workers(n_workers),
2987 _hrclaimer(_active_workers) {
2988 }
2989
2990 void work(uint worker_id) {
2991 AggregateCountDataHRClosure cl(_g1h, _cm_card_bm, _max_worker_id);
2992
2993 _g1h->heap_region_par_iterate(&cl, worker_id, &_hrclaimer);
2994 }
2995 };
2996
2997
2998 void ConcurrentMark::aggregate_count_data() {
2999 uint n_workers = _g1h->workers()->active_workers();
3000
3001 G1AggregateCountDataTask g1_par_agg_task(_g1h, this, &_card_bm,
3002 _max_worker_id, n_workers);
3003
3004 _g1h->set_par_threads(n_workers);
3005 _g1h->workers()->run_task(&g1_par_agg_task);
3006 _g1h->set_par_threads(0);
3007 }
3008
3009 // Clear the per-worker arrays used to store the per-region counting data
3010 void ConcurrentMark::clear_all_count_data() {
3011 // Clear the global card bitmap - it will be filled during
3012 // liveness count aggregation (during remark) and the
3013 // final counting task.
3014 _card_bm.clear();
3015
3016 // Clear the global region bitmap - it will be filled as part
3017 // of the final counting task.
3018 _region_bm.clear();
3019
3020 uint max_regions = _g1h->max_regions();
3021 assert(_max_worker_id > 0, "uninitialized");
3022
3023 for (uint i = 0; i < _max_worker_id; i += 1) {
3024 BitMap* task_card_bm = count_card_bitmap_for(i);
3025 size_t* marked_bytes_array = count_marked_bytes_array_for(i);
3026
3027 assert(task_card_bm->size() == _card_bm.size(), "size mismatch");
3028 assert(marked_bytes_array != NULL, "uninitialized");
3029
3030 memset(marked_bytes_array, 0, (size_t) max_regions * sizeof(size_t));
3031 task_card_bm->clear();
3032 }
3033 }
3034
3035 void ConcurrentMark::print_stats() {
3036 if (verbose_stats()) {
3037 gclog_or_tty->print_cr("---------------------------------------------------------------------");
3038 for (size_t i = 0; i < _active_tasks; ++i) {
3039 _tasks[i]->print_stats();
3040 gclog_or_tty->print_cr("---------------------------------------------------------------------");
3041 }
3042 }
3043 }
3044
3045 // abandon current marking iteration due to a Full GC
3046 void ConcurrentMark::abort() {
3047 // Clear all marks in the next bitmap for the next marking cycle. This will allow us to skip the next
3048 // concurrent bitmap clearing.
3049 _nextMarkBitMap->clearAll();
3050
3051 // Note we cannot clear the previous marking bitmap here
3052 // since VerifyDuringGC verifies the objects marked during
3053 // a full GC against the previous bitmap.
3054
3055 // Clear the liveness counting data
3056 clear_all_count_data();
3057 // Empty mark stack
3058 reset_marking_state();
3059 for (uint i = 0; i < _max_worker_id; ++i) {
3060 _tasks[i]->clear_region_fields();
3061 }
3062 _first_overflow_barrier_sync.abort();
3063 _second_overflow_barrier_sync.abort();
3064 const GCId& gc_id = _g1h->gc_tracer_cm()->gc_id();
3065 if (!gc_id.is_undefined()) {
3066 // We can do multiple full GCs before ConcurrentMarkThread::run() gets a chance
3067 // to detect that it was aborted. Only keep track of the first GC id that we aborted.
3068 _aborted_gc_id = gc_id;
3069 }
3070 _has_aborted = true;
3071
3072 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
3073 satb_mq_set.abandon_partial_marking();
3074 // This can be called either during or outside marking, we'll read
3075 // the expected_active value from the SATB queue set.
3076 satb_mq_set.set_active_all_threads(
3077 false, /* new active value */
3078 satb_mq_set.is_active() /* expected_active */);
3079
3080 _g1h->trace_heap_after_concurrent_cycle();
3081 _g1h->register_concurrent_cycle_end();
3082 }
3083
3084 const GCId& ConcurrentMark::concurrent_gc_id() {
3085 if (has_aborted()) {
3086 return _aborted_gc_id;
3087 }
3088 return _g1h->gc_tracer_cm()->gc_id();
3089 }
3090
3091 static void print_ms_time_info(const char* prefix, const char* name,
3092 NumberSeq& ns) {
3093 gclog_or_tty->print_cr("%s%5d %12s: total time = %8.2f s (avg = %8.2f ms).",
3094 prefix, ns.num(), name, ns.sum()/1000.0, ns.avg());
3095 if (ns.num() > 0) {
3096 gclog_or_tty->print_cr("%s [std. dev = %8.2f ms, max = %8.2f ms]",
3097 prefix, ns.sd(), ns.maximum());
3098 }
3099 }
3100
3101 void ConcurrentMark::print_summary_info() {
3102 gclog_or_tty->print_cr(" Concurrent marking:");
3103 print_ms_time_info(" ", "init marks", _init_times);
3104 print_ms_time_info(" ", "remarks", _remark_times);
3105 {
3106 print_ms_time_info(" ", "final marks", _remark_mark_times);
3107 print_ms_time_info(" ", "weak refs", _remark_weak_ref_times);
3108
3109 }
3110 print_ms_time_info(" ", "cleanups", _cleanup_times);
3111 gclog_or_tty->print_cr(" Final counting total time = %8.2f s (avg = %8.2f ms).",
3112 _total_counting_time,
3113 (_cleanup_times.num() > 0 ? _total_counting_time * 1000.0 /
3114 (double)_cleanup_times.num()
3115 : 0.0));
3116 if (G1ScrubRemSets) {
3117 gclog_or_tty->print_cr(" RS scrub total time = %8.2f s (avg = %8.2f ms).",
3118 _total_rs_scrub_time,
3119 (_cleanup_times.num() > 0 ? _total_rs_scrub_time * 1000.0 /
3120 (double)_cleanup_times.num()
3121 : 0.0));
3122 }
3123 gclog_or_tty->print_cr(" Total stop_world time = %8.2f s.",
3124 (_init_times.sum() + _remark_times.sum() +
3125 _cleanup_times.sum())/1000.0);
3126 gclog_or_tty->print_cr(" Total concurrent time = %8.2f s "
3127 "(%8.2f s marking).",
3128 cmThread()->vtime_accum(),
3129 cmThread()->vtime_mark_accum());
3130 }
3131
3132 void ConcurrentMark::print_worker_threads_on(outputStream* st) const {
3133 _parallel_workers->print_worker_threads_on(st);
3134 }
3135
3136 void ConcurrentMark::print_on_error(outputStream* st) const {
3137 st->print_cr("Marking Bits (Prev, Next): (CMBitMap*) " PTR_FORMAT ", (CMBitMap*) " PTR_FORMAT,
3138 p2i(_prevMarkBitMap), p2i(_nextMarkBitMap));
3139 _prevMarkBitMap->print_on_error(st, " Prev Bits: ");
3140 _nextMarkBitMap->print_on_error(st, " Next Bits: ");
3141 }
3142
3143 // We take a break if someone is trying to stop the world.
3144 bool ConcurrentMark::do_yield_check(uint worker_id) {
3145 if (SuspendibleThreadSet::should_yield()) {
3146 if (worker_id == 0) {
3147 _g1h->g1_policy()->record_concurrent_pause();
3148 }
3149 SuspendibleThreadSet::yield();
3150 return true;
3151 } else {
3152 return false;
3153 }
3154 }
3155
3156 #ifndef PRODUCT
3157 // for debugging purposes
3158 void ConcurrentMark::print_finger() {
3159 gclog_or_tty->print_cr("heap [" PTR_FORMAT ", " PTR_FORMAT "), global finger = " PTR_FORMAT,
3160 p2i(_heap_start), p2i(_heap_end), p2i(_finger));
3161 for (uint i = 0; i < _max_worker_id; ++i) {
3162 gclog_or_tty->print(" %u: " PTR_FORMAT, i, p2i(_tasks[i]->finger()));
3163 }
3164 gclog_or_tty->cr();
3165 }
3166 #endif
3167
3168 template<bool scan>
3169 inline void CMTask::process_grey_object(oop obj) {
3170 assert(scan || obj->is_typeArray(), "Skipping scan of grey non-typeArray");
3171 assert(_nextMarkBitMap->isMarked((HeapWord*) obj), "invariant");
3172
3173 if (_cm->verbose_high()) {
3174 gclog_or_tty->print_cr("[%u] processing grey object " PTR_FORMAT,
3175 _worker_id, p2i((void*) obj));
3176 }
3177
3178 size_t obj_size = obj->size();
3179 _words_scanned += obj_size;
3180
3181 if (scan) {
3182 obj->oop_iterate(_cm_oop_closure);
3183 }
3184 statsOnly( ++_objs_scanned );
3185 check_limits();
3186 }
3187
3188 template void CMTask::process_grey_object<true>(oop);
3189 template void CMTask::process_grey_object<false>(oop);
3190
3191 // Closure for iteration over bitmaps
3192 class CMBitMapClosure : public BitMapClosure {
3193 private:
3194 // the bitmap that is being iterated over
3195 CMBitMap* _nextMarkBitMap;
3196 ConcurrentMark* _cm;
3197 CMTask* _task;
3198
3199 public:
3200 CMBitMapClosure(CMTask *task, ConcurrentMark* cm, CMBitMap* nextMarkBitMap) :
3201 _task(task), _cm(cm), _nextMarkBitMap(nextMarkBitMap) { }
3202
3203 bool do_bit(size_t offset) {
3204 HeapWord* addr = _nextMarkBitMap->offsetToHeapWord(offset);
3205 assert(_nextMarkBitMap->isMarked(addr), "invariant");
3206 assert( addr < _cm->finger(), "invariant");
3207
3208 statsOnly( _task->increase_objs_found_on_bitmap() );
3209 assert(addr >= _task->finger(), "invariant");
3210
3211 // We move that task's local finger along.
3212 _task->move_finger_to(addr);
3213
3214 _task->scan_object(oop(addr));
3215 // we only partially drain the local queue and global stack
3216 _task->drain_local_queue(true);
3217 _task->drain_global_stack(true);
3218
3219 // if the has_aborted flag has been raised, we need to bail out of
3220 // the iteration
3221 return !_task->has_aborted();
3222 }
3223 };
3224
3225 G1CMOopClosure::G1CMOopClosure(G1CollectedHeap* g1h,
3226 ConcurrentMark* cm,
3227 CMTask* task)
3228 : _g1h(g1h), _cm(cm), _task(task) {
3229 assert(_ref_processor == NULL, "should be initialized to NULL");
3230
3231 if (G1UseConcMarkReferenceProcessing) {
3232 _ref_processor = g1h->ref_processor_cm();
3233 assert(_ref_processor != NULL, "should not be NULL");
3234 }
3235 }
3236
3237 void CMTask::setup_for_region(HeapRegion* hr) {
3238 assert(hr != NULL,
3239 "claim_region() should have filtered out NULL regions");
3240 assert(!hr->is_continues_humongous(),
3241 "claim_region() should have filtered out continues humongous regions");
3242
3243 if (_cm->verbose_low()) {
3244 gclog_or_tty->print_cr("[%u] setting up for region " PTR_FORMAT,
3245 _worker_id, p2i(hr));
3246 }
3247
3248 _curr_region = hr;
3249 _finger = hr->bottom();
3250 update_region_limit();
3251 }
3252
3253 void CMTask::update_region_limit() {
3254 HeapRegion* hr = _curr_region;
3255 HeapWord* bottom = hr->bottom();
3256 HeapWord* limit = hr->next_top_at_mark_start();
3257
3258 if (limit == bottom) {
3259 if (_cm->verbose_low()) {
3260 gclog_or_tty->print_cr("[%u] found an empty region "
3261 "[" PTR_FORMAT ", " PTR_FORMAT ")",
3262 _worker_id, p2i(bottom), p2i(limit));
3263 }
3264 // The region was collected underneath our feet.
3265 // We set the finger to bottom to ensure that the bitmap
3266 // iteration that will follow this will not do anything.
3267 // (this is not a condition that holds when we set the region up,
3268 // as the region is not supposed to be empty in the first place)
3269 _finger = bottom;
3270 } else if (limit >= _region_limit) {
3271 assert(limit >= _finger, "peace of mind");
3272 } else {
3273 assert(limit < _region_limit, "only way to get here");
3274 // This can happen under some pretty unusual circumstances. An
3275 // evacuation pause empties the region underneath our feet (NTAMS
3276 // at bottom). We then do some allocation in the region (NTAMS
3277 // stays at bottom), followed by the region being used as a GC
3278 // alloc region (NTAMS will move to top() and the objects
3279 // originally below it will be grayed). All objects now marked in
3280 // the region are explicitly grayed, if below the global finger,
3281 // and we do not need in fact to scan anything else. So, we simply
3282 // set _finger to be limit to ensure that the bitmap iteration
3283 // doesn't do anything.
3284 _finger = limit;
3285 }
3286
3287 _region_limit = limit;
3288 }
3289
3290 void CMTask::giveup_current_region() {
3291 assert(_curr_region != NULL, "invariant");
3292 if (_cm->verbose_low()) {
3293 gclog_or_tty->print_cr("[%u] giving up region " PTR_FORMAT,
3294 _worker_id, p2i(_curr_region));
3295 }
3296 clear_region_fields();
3297 }
3298
3299 void CMTask::clear_region_fields() {
3300 // Values for these three fields that indicate that we're not
3301 // holding on to a region.
3302 _curr_region = NULL;
3303 _finger = NULL;
3304 _region_limit = NULL;
3305 }
3306
3307 void CMTask::set_cm_oop_closure(G1CMOopClosure* cm_oop_closure) {
3308 if (cm_oop_closure == NULL) {
3309 assert(_cm_oop_closure != NULL, "invariant");
3310 } else {
3311 assert(_cm_oop_closure == NULL, "invariant");
3312 }
3313 _cm_oop_closure = cm_oop_closure;
3314 }
3315
3316 void CMTask::reset(CMBitMap* nextMarkBitMap) {
3317 guarantee(nextMarkBitMap != NULL, "invariant");
3318
3319 if (_cm->verbose_low()) {
3320 gclog_or_tty->print_cr("[%u] resetting", _worker_id);
3321 }
3322
3323 _nextMarkBitMap = nextMarkBitMap;
3324 clear_region_fields();
3325
3326 _calls = 0;
3327 _elapsed_time_ms = 0.0;
3328 _termination_time_ms = 0.0;
3329 _termination_start_time_ms = 0.0;
3330
3331 #if _MARKING_STATS_
3332 _aborted = 0;
3333 _aborted_overflow = 0;
3334 _aborted_cm_aborted = 0;
3335 _aborted_yield = 0;
3336 _aborted_timed_out = 0;
3337 _aborted_satb = 0;
3338 _aborted_termination = 0;
3339 _steal_attempts = 0;
3340 _steals = 0;
3341 _local_pushes = 0;
3342 _local_pops = 0;
3343 _local_max_size = 0;
3344 _objs_scanned = 0;
3345 _global_pushes = 0;
3346 _global_pops = 0;
3347 _global_max_size = 0;
3348 _global_transfers_to = 0;
3349 _global_transfers_from = 0;
3350 _regions_claimed = 0;
3351 _objs_found_on_bitmap = 0;
3352 _satb_buffers_processed = 0;
3353 #endif // _MARKING_STATS_
3354 }
3355
3356 bool CMTask::should_exit_termination() {
3357 regular_clock_call();
3358 // This is called when we are in the termination protocol. We should
3359 // quit if, for some reason, this task wants to abort or the global
3360 // stack is not empty (this means that we can get work from it).
3361 return !_cm->mark_stack_empty() || has_aborted();
3362 }
3363
3364 void CMTask::reached_limit() {
3365 assert(_words_scanned >= _words_scanned_limit ||
3366 _refs_reached >= _refs_reached_limit ,
3367 "shouldn't have been called otherwise");
3368 regular_clock_call();
3369 }
3370
3371 void CMTask::regular_clock_call() {
3372 if (has_aborted()) return;
3373
3374 // First, we need to recalculate the words scanned and refs reached
3375 // limits for the next clock call.
3376 recalculate_limits();
3377
3378 // During the regular clock call we do the following
3379
3380 // (1) If an overflow has been flagged, then we abort.
3381 if (_cm->has_overflown()) {
3382 set_has_aborted();
3383 return;
3384 }
3385
3386 // If we are not concurrent (i.e. we're doing remark) we don't need
3387 // to check anything else. The other steps are only needed during
3388 // the concurrent marking phase.
3389 if (!concurrent()) return;
3390
3391 // (2) If marking has been aborted for Full GC, then we also abort.
3392 if (_cm->has_aborted()) {
3393 set_has_aborted();
3394 statsOnly( ++_aborted_cm_aborted );
3395 return;
3396 }
3397
3398 double curr_time_ms = os::elapsedVTime() * 1000.0;
3399
3400 // (3) If marking stats are enabled, then we update the step history.
3401 #if _MARKING_STATS_
3402 if (_words_scanned >= _words_scanned_limit) {
3403 ++_clock_due_to_scanning;
3404 }
3405 if (_refs_reached >= _refs_reached_limit) {
3406 ++_clock_due_to_marking;
3407 }
3408
3409 double last_interval_ms = curr_time_ms - _interval_start_time_ms;
3410 _interval_start_time_ms = curr_time_ms;
3411 _all_clock_intervals_ms.add(last_interval_ms);
3412
3413 if (_cm->verbose_medium()) {
3414 gclog_or_tty->print_cr("[%u] regular clock, interval = %1.2lfms, "
3415 "scanned = " SIZE_FORMAT "%s, refs reached = " SIZE_FORMAT "%s",
3416 _worker_id, last_interval_ms,
3417 _words_scanned,
3418 (_words_scanned >= _words_scanned_limit) ? " (*)" : "",
3419 _refs_reached,
3420 (_refs_reached >= _refs_reached_limit) ? " (*)" : "");
3421 }
3422 #endif // _MARKING_STATS_
3423
3424 // (4) We check whether we should yield. If we have to, then we abort.
3425 if (SuspendibleThreadSet::should_yield()) {
3426 // We should yield. To do this we abort the task. The caller is
3427 // responsible for yielding.
3428 set_has_aborted();
3429 statsOnly( ++_aborted_yield );
3430 return;
3431 }
3432
3433 // (5) We check whether we've reached our time quota. If we have,
3434 // then we abort.
3435 double elapsed_time_ms = curr_time_ms - _start_time_ms;
3436 if (elapsed_time_ms > _time_target_ms) {
3437 set_has_aborted();
3438 _has_timed_out = true;
3439 statsOnly( ++_aborted_timed_out );
3440 return;
3441 }
3442
3443 // (6) Finally, we check whether there are enough completed STAB
3444 // buffers available for processing. If there are, we abort.
3445 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
3446 if (!_draining_satb_buffers && satb_mq_set.process_completed_buffers()) {
3447 if (_cm->verbose_low()) {
3448 gclog_or_tty->print_cr("[%u] aborting to deal with pending SATB buffers",
3449 _worker_id);
3450 }
3451 // we do need to process SATB buffers, we'll abort and restart
3452 // the marking task to do so
3453 set_has_aborted();
3454 statsOnly( ++_aborted_satb );
3455 return;
3456 }
3457 }
3458
3459 void CMTask::recalculate_limits() {
3460 _real_words_scanned_limit = _words_scanned + words_scanned_period;
3461 _words_scanned_limit = _real_words_scanned_limit;
3462
3463 _real_refs_reached_limit = _refs_reached + refs_reached_period;
3464 _refs_reached_limit = _real_refs_reached_limit;
3465 }
3466
3467 void CMTask::decrease_limits() {
3468 // This is called when we believe that we're going to do an infrequent
3469 // operation which will increase the per byte scanned cost (i.e. move
3470 // entries to/from the global stack). It basically tries to decrease the
3471 // scanning limit so that the clock is called earlier.
3472
3473 if (_cm->verbose_medium()) {
3474 gclog_or_tty->print_cr("[%u] decreasing limits", _worker_id);
3475 }
3476
3477 _words_scanned_limit = _real_words_scanned_limit -
3478 3 * words_scanned_period / 4;
3479 _refs_reached_limit = _real_refs_reached_limit -
3480 3 * refs_reached_period / 4;
3481 }
3482
3483 void CMTask::move_entries_to_global_stack() {
3484 // local array where we'll store the entries that will be popped
3485 // from the local queue
3486 oop buffer[global_stack_transfer_size];
3487
3488 int n = 0;
3489 oop obj;
3490 while (n < global_stack_transfer_size && _task_queue->pop_local(obj)) {
3491 buffer[n] = obj;
3492 ++n;
3493 }
3494
3495 if (n > 0) {
3496 // we popped at least one entry from the local queue
3497
3498 statsOnly( ++_global_transfers_to; _local_pops += n );
3499
3500 if (!_cm->mark_stack_push(buffer, n)) {
3501 if (_cm->verbose_low()) {
3502 gclog_or_tty->print_cr("[%u] aborting due to global stack overflow",
3503 _worker_id);
3504 }
3505 set_has_aborted();
3506 } else {
3507 // the transfer was successful
3508
3509 if (_cm->verbose_medium()) {
3510 gclog_or_tty->print_cr("[%u] pushed %d entries to the global stack",
3511 _worker_id, n);
3512 }
3513 statsOnly( size_t tmp_size = _cm->mark_stack_size();
3514 if (tmp_size > _global_max_size) {
3515 _global_max_size = tmp_size;
3516 }
3517 _global_pushes += n );
3518 }
3519 }
3520
3521 // this operation was quite expensive, so decrease the limits
3522 decrease_limits();
3523 }
3524
3525 void CMTask::get_entries_from_global_stack() {
3526 // local array where we'll store the entries that will be popped
3527 // from the global stack.
3528 oop buffer[global_stack_transfer_size];
3529 int n;
3530 _cm->mark_stack_pop(buffer, global_stack_transfer_size, &n);
3531 assert(n <= global_stack_transfer_size,
3532 "we should not pop more than the given limit");
3533 if (n > 0) {
3534 // yes, we did actually pop at least one entry
3535
3536 statsOnly( ++_global_transfers_from; _global_pops += n );
3537 if (_cm->verbose_medium()) {
3538 gclog_or_tty->print_cr("[%u] popped %d entries from the global stack",
3539 _worker_id, n);
3540 }
3541 for (int i = 0; i < n; ++i) {
3542 bool success = _task_queue->push(buffer[i]);
3543 // We only call this when the local queue is empty or under a
3544 // given target limit. So, we do not expect this push to fail.
3545 assert(success, "invariant");
3546 }
3547
3548 statsOnly( size_t tmp_size = (size_t)_task_queue->size();
3549 if (tmp_size > _local_max_size) {
3550 _local_max_size = tmp_size;
3551 }
3552 _local_pushes += n );
3553 }
3554
3555 // this operation was quite expensive, so decrease the limits
3556 decrease_limits();
3557 }
3558
3559 void CMTask::drain_local_queue(bool partially) {
3560 if (has_aborted()) return;
3561
3562 // Decide what the target size is, depending whether we're going to
3563 // drain it partially (so that other tasks can steal if they run out
3564 // of things to do) or totally (at the very end).
3565 size_t target_size;
3566 if (partially) {
3567 target_size = MIN2((size_t)_task_queue->max_elems()/3, GCDrainStackTargetSize);
3568 } else {
3569 target_size = 0;
3570 }
3571
3572 if (_task_queue->size() > target_size) {
3573 if (_cm->verbose_high()) {
3574 gclog_or_tty->print_cr("[%u] draining local queue, target size = " SIZE_FORMAT,
3575 _worker_id, target_size);
3576 }
3577
3578 oop obj;
3579 bool ret = _task_queue->pop_local(obj);
3580 while (ret) {
3581 statsOnly( ++_local_pops );
3582
3583 if (_cm->verbose_high()) {
3584 gclog_or_tty->print_cr("[%u] popped " PTR_FORMAT, _worker_id,
3585 p2i((void*) obj));
3586 }
3587
3588 assert(_g1h->is_in_g1_reserved((HeapWord*) obj), "invariant" );
3589 assert(!_g1h->is_on_master_free_list(
3590 _g1h->heap_region_containing((HeapWord*) obj)), "invariant");
3591
3592 scan_object(obj);
3593
3594 if (_task_queue->size() <= target_size || has_aborted()) {
3595 ret = false;
3596 } else {
3597 ret = _task_queue->pop_local(obj);
3598 }
3599 }
3600
3601 if (_cm->verbose_high()) {
3602 gclog_or_tty->print_cr("[%u] drained local queue, size = %u",
3603 _worker_id, _task_queue->size());
3604 }
3605 }
3606 }
3607
3608 void CMTask::drain_global_stack(bool partially) {
3609 if (has_aborted()) return;
3610
3611 // We have a policy to drain the local queue before we attempt to
3612 // drain the global stack.
3613 assert(partially || _task_queue->size() == 0, "invariant");
3614
3615 // Decide what the target size is, depending whether we're going to
3616 // drain it partially (so that other tasks can steal if they run out
3617 // of things to do) or totally (at the very end). Notice that,
3618 // because we move entries from the global stack in chunks or
3619 // because another task might be doing the same, we might in fact
3620 // drop below the target. But, this is not a problem.
3621 size_t target_size;
3622 if (partially) {
3623 target_size = _cm->partial_mark_stack_size_target();
3624 } else {
3625 target_size = 0;
3626 }
3627
3628 if (_cm->mark_stack_size() > target_size) {
3629 if (_cm->verbose_low()) {
3630 gclog_or_tty->print_cr("[%u] draining global_stack, target size " SIZE_FORMAT,
3631 _worker_id, target_size);
3632 }
3633
3634 while (!has_aborted() && _cm->mark_stack_size() > target_size) {
3635 get_entries_from_global_stack();
3636 drain_local_queue(partially);
3637 }
3638
3639 if (_cm->verbose_low()) {
3640 gclog_or_tty->print_cr("[%u] drained global stack, size = " SIZE_FORMAT,
3641 _worker_id, _cm->mark_stack_size());
3642 }
3643 }
3644 }
3645
3646 // SATB Queue has several assumptions on whether to call the par or
3647 // non-par versions of the methods. this is why some of the code is
3648 // replicated. We should really get rid of the single-threaded version
3649 // of the code to simplify things.
3650 void CMTask::drain_satb_buffers() {
3651 if (has_aborted()) return;
3652
3653 // We set this so that the regular clock knows that we're in the
3654 // middle of draining buffers and doesn't set the abort flag when it
3655 // notices that SATB buffers are available for draining. It'd be
3656 // very counter productive if it did that. :-)
3657 _draining_satb_buffers = true;
3658
3659 CMSATBBufferClosure satb_cl(this, _g1h);
3660 SATBMarkQueueSet& satb_mq_set = JavaThread::satb_mark_queue_set();
3661
3662 // This keeps claiming and applying the closure to completed buffers
3663 // until we run out of buffers or we need to abort.
3664 while (!has_aborted() &&
3665 satb_mq_set.apply_closure_to_completed_buffer(&satb_cl)) {
3666 if (_cm->verbose_medium()) {
3667 gclog_or_tty->print_cr("[%u] processed an SATB buffer", _worker_id);
3668 }
3669 statsOnly( ++_satb_buffers_processed );
3670 regular_clock_call();
3671 }
3672
3673 _draining_satb_buffers = false;
3674
3675 assert(has_aborted() ||
3676 concurrent() ||
3677 satb_mq_set.completed_buffers_num() == 0, "invariant");
3678
3679 // again, this was a potentially expensive operation, decrease the
3680 // limits to get the regular clock call early
3681 decrease_limits();
3682 }
3683
3684 void CMTask::print_stats() {
3685 gclog_or_tty->print_cr("Marking Stats, task = %u, calls = %d",
3686 _worker_id, _calls);
3687 gclog_or_tty->print_cr(" Elapsed time = %1.2lfms, Termination time = %1.2lfms",
3688 _elapsed_time_ms, _termination_time_ms);
3689 gclog_or_tty->print_cr(" Step Times (cum): num = %d, avg = %1.2lfms, sd = %1.2lfms",
3690 _step_times_ms.num(), _step_times_ms.avg(),
3691 _step_times_ms.sd());
3692 gclog_or_tty->print_cr(" max = %1.2lfms, total = %1.2lfms",
3693 _step_times_ms.maximum(), _step_times_ms.sum());
3694
3695 #if _MARKING_STATS_
3696 gclog_or_tty->print_cr(" Clock Intervals (cum): num = %d, avg = %1.2lfms, sd = %1.2lfms",
3697 _all_clock_intervals_ms.num(), _all_clock_intervals_ms.avg(),
3698 _all_clock_intervals_ms.sd());
3699 gclog_or_tty->print_cr(" max = %1.2lfms, total = %1.2lfms",
3700 _all_clock_intervals_ms.maximum(),
3701 _all_clock_intervals_ms.sum());
3702 gclog_or_tty->print_cr(" Clock Causes (cum): scanning = " SIZE_FORMAT ", marking = " SIZE_FORMAT,
3703 _clock_due_to_scanning, _clock_due_to_marking);
3704 gclog_or_tty->print_cr(" Objects: scanned = " SIZE_FORMAT ", found on the bitmap = " SIZE_FORMAT,
3705 _objs_scanned, _objs_found_on_bitmap);
3706 gclog_or_tty->print_cr(" Local Queue: pushes = " SIZE_FORMAT ", pops = " SIZE_FORMAT ", max size = " SIZE_FORMAT,
3707 _local_pushes, _local_pops, _local_max_size);
3708 gclog_or_tty->print_cr(" Global Stack: pushes = " SIZE_FORMAT ", pops = " SIZE_FORMAT ", max size = " SIZE_FORMAT,
3709 _global_pushes, _global_pops, _global_max_size);
3710 gclog_or_tty->print_cr(" transfers to = " SIZE_FORMAT ", transfers from = " SIZE_FORMAT,
3711 _global_transfers_to,_global_transfers_from);
3712 gclog_or_tty->print_cr(" Regions: claimed = " SIZE_FORMAT, _regions_claimed);
3713 gclog_or_tty->print_cr(" SATB buffers: processed = " SIZE_FORMAT, _satb_buffers_processed);
3714 gclog_or_tty->print_cr(" Steals: attempts = " SIZE_FORMAT ", successes = " SIZE_FORMAT,
3715 _steal_attempts, _steals);
3716 gclog_or_tty->print_cr(" Aborted: " SIZE_FORMAT ", due to", _aborted);
3717 gclog_or_tty->print_cr(" overflow: " SIZE_FORMAT ", global abort: " SIZE_FORMAT ", yield: " SIZE_FORMAT,
3718 _aborted_overflow, _aborted_cm_aborted, _aborted_yield);
3719 gclog_or_tty->print_cr(" time out: " SIZE_FORMAT ", SATB: " SIZE_FORMAT ", termination: " SIZE_FORMAT,
3720 _aborted_timed_out, _aborted_satb, _aborted_termination);
3721 #endif // _MARKING_STATS_
3722 }
3723
3724 bool ConcurrentMark::try_stealing(uint worker_id, int* hash_seed, oop& obj) {
3725 return _task_queues->steal(worker_id, hash_seed, obj);
3726 }
3727
3728 /*****************************************************************************
3729
3730 The do_marking_step(time_target_ms, ...) method is the building
3731 block of the parallel marking framework. It can be called in parallel
3732 with other invocations of do_marking_step() on different tasks
3733 (but only one per task, obviously) and concurrently with the
3734 mutator threads, or during remark, hence it eliminates the need
3735 for two versions of the code. When called during remark, it will
3736 pick up from where the task left off during the concurrent marking
3737 phase. Interestingly, tasks are also claimable during evacuation
3738 pauses too, since do_marking_step() ensures that it aborts before
3739 it needs to yield.
3740
3741 The data structures that it uses to do marking work are the
3742 following:
3743
3744 (1) Marking Bitmap. If there are gray objects that appear only
3745 on the bitmap (this happens either when dealing with an overflow
3746 or when the initial marking phase has simply marked the roots
3747 and didn't push them on the stack), then tasks claim heap
3748 regions whose bitmap they then scan to find gray objects. A
3749 global finger indicates where the end of the last claimed region
3750 is. A local finger indicates how far into the region a task has
3751 scanned. The two fingers are used to determine how to gray an
3752 object (i.e. whether simply marking it is OK, as it will be
3753 visited by a task in the future, or whether it needs to be also
3754 pushed on a stack).
3755
3756 (2) Local Queue. The local queue of the task which is accessed
3757 reasonably efficiently by the task. Other tasks can steal from
3758 it when they run out of work. Throughout the marking phase, a
3759 task attempts to keep its local queue short but not totally
3760 empty, so that entries are available for stealing by other
3761 tasks. Only when there is no more work, a task will totally
3762 drain its local queue.
3763
3764 (3) Global Mark Stack. This handles local queue overflow. During
3765 marking only sets of entries are moved between it and the local
3766 queues, as access to it requires a mutex and more fine-grain
3767 interaction with it which might cause contention. If it
3768 overflows, then the marking phase should restart and iterate
3769 over the bitmap to identify gray objects. Throughout the marking
3770 phase, tasks attempt to keep the global mark stack at a small
3771 length but not totally empty, so that entries are available for
3772 popping by other tasks. Only when there is no more work, tasks
3773 will totally drain the global mark stack.
3774
3775 (4) SATB Buffer Queue. This is where completed SATB buffers are
3776 made available. Buffers are regularly removed from this queue
3777 and scanned for roots, so that the queue doesn't get too
3778 long. During remark, all completed buffers are processed, as
3779 well as the filled in parts of any uncompleted buffers.
3780
3781 The do_marking_step() method tries to abort when the time target
3782 has been reached. There are a few other cases when the
3783 do_marking_step() method also aborts:
3784
3785 (1) When the marking phase has been aborted (after a Full GC).
3786
3787 (2) When a global overflow (on the global stack) has been
3788 triggered. Before the task aborts, it will actually sync up with
3789 the other tasks to ensure that all the marking data structures
3790 (local queues, stacks, fingers etc.) are re-initialized so that
3791 when do_marking_step() completes, the marking phase can
3792 immediately restart.
3793
3794 (3) When enough completed SATB buffers are available. The
3795 do_marking_step() method only tries to drain SATB buffers right
3796 at the beginning. So, if enough buffers are available, the
3797 marking step aborts and the SATB buffers are processed at
3798 the beginning of the next invocation.
3799
3800 (4) To yield. when we have to yield then we abort and yield
3801 right at the end of do_marking_step(). This saves us from a lot
3802 of hassle as, by yielding we might allow a Full GC. If this
3803 happens then objects will be compacted underneath our feet, the
3804 heap might shrink, etc. We save checking for this by just
3805 aborting and doing the yield right at the end.
3806
3807 From the above it follows that the do_marking_step() method should
3808 be called in a loop (or, otherwise, regularly) until it completes.
3809
3810 If a marking step completes without its has_aborted() flag being
3811 true, it means it has completed the current marking phase (and
3812 also all other marking tasks have done so and have all synced up).
3813
3814 A method called regular_clock_call() is invoked "regularly" (in
3815 sub ms intervals) throughout marking. It is this clock method that
3816 checks all the abort conditions which were mentioned above and
3817 decides when the task should abort. A work-based scheme is used to
3818 trigger this clock method: when the number of object words the
3819 marking phase has scanned or the number of references the marking
3820 phase has visited reach a given limit. Additional invocations to
3821 the method clock have been planted in a few other strategic places
3822 too. The initial reason for the clock method was to avoid calling
3823 vtime too regularly, as it is quite expensive. So, once it was in
3824 place, it was natural to piggy-back all the other conditions on it
3825 too and not constantly check them throughout the code.
3826
3827 If do_termination is true then do_marking_step will enter its
3828 termination protocol.
3829
3830 The value of is_serial must be true when do_marking_step is being
3831 called serially (i.e. by the VMThread) and do_marking_step should
3832 skip any synchronization in the termination and overflow code.
3833 Examples include the serial remark code and the serial reference
3834 processing closures.
3835
3836 The value of is_serial must be false when do_marking_step is
3837 being called by any of the worker threads in a work gang.
3838 Examples include the concurrent marking code (CMMarkingTask),
3839 the MT remark code, and the MT reference processing closures.
3840
3841 *****************************************************************************/
3842
3843 void CMTask::do_marking_step(double time_target_ms,
3844 bool do_termination,
3845 bool is_serial) {
3846 assert(time_target_ms >= 1.0, "minimum granularity is 1ms");
3847 assert(concurrent() == _cm->concurrent(), "they should be the same");
3848
3849 G1CollectorPolicy* g1_policy = _g1h->g1_policy();
3850 assert(_task_queues != NULL, "invariant");
3851 assert(_task_queue != NULL, "invariant");
3852 assert(_task_queues->queue(_worker_id) == _task_queue, "invariant");
3853
3854 assert(!_claimed,
3855 "only one thread should claim this task at any one time");
3856
3857 // OK, this doesn't safeguard again all possible scenarios, as it is
3858 // possible for two threads to set the _claimed flag at the same
3859 // time. But it is only for debugging purposes anyway and it will
3860 // catch most problems.
3861 _claimed = true;
3862
3863 _start_time_ms = os::elapsedVTime() * 1000.0;
3864 statsOnly( _interval_start_time_ms = _start_time_ms );
3865
3866 // If do_stealing is true then do_marking_step will attempt to
3867 // steal work from the other CMTasks. It only makes sense to
3868 // enable stealing when the termination protocol is enabled
3869 // and do_marking_step() is not being called serially.
3870 bool do_stealing = do_termination && !is_serial;
3871
3872 double diff_prediction_ms =
3873 g1_policy->get_new_prediction(&_marking_step_diffs_ms);
3874 _time_target_ms = time_target_ms - diff_prediction_ms;
3875
3876 // set up the variables that are used in the work-based scheme to
3877 // call the regular clock method
3878 _words_scanned = 0;
3879 _refs_reached = 0;
3880 recalculate_limits();
3881
3882 // clear all flags
3883 clear_has_aborted();
3884 _has_timed_out = false;
3885 _draining_satb_buffers = false;
3886
3887 ++_calls;
3888
3889 if (_cm->verbose_low()) {
3890 gclog_or_tty->print_cr("[%u] >>>>>>>>>> START, call = %d, "
3891 "target = %1.2lfms >>>>>>>>>>",
3892 _worker_id, _calls, _time_target_ms);
3893 }
3894
3895 // Set up the bitmap and oop closures. Anything that uses them is
3896 // eventually called from this method, so it is OK to allocate these
3897 // statically.
3898 CMBitMapClosure bitmap_closure(this, _cm, _nextMarkBitMap);
3899 G1CMOopClosure cm_oop_closure(_g1h, _cm, this);
3900 set_cm_oop_closure(&cm_oop_closure);
3901
3902 if (_cm->has_overflown()) {
3903 // This can happen if the mark stack overflows during a GC pause
3904 // and this task, after a yield point, restarts. We have to abort
3905 // as we need to get into the overflow protocol which happens
3906 // right at the end of this task.
3907 set_has_aborted();
3908 }
3909
3910 // First drain any available SATB buffers. After this, we will not
3911 // look at SATB buffers before the next invocation of this method.
3912 // If enough completed SATB buffers are queued up, the regular clock
3913 // will abort this task so that it restarts.
3914 drain_satb_buffers();
3915 // ...then partially drain the local queue and the global stack
3916 drain_local_queue(true);
3917 drain_global_stack(true);
3918
3919 do {
3920 if (!has_aborted() && _curr_region != NULL) {
3921 // This means that we're already holding on to a region.
3922 assert(_finger != NULL, "if region is not NULL, then the finger "
3923 "should not be NULL either");
3924
3925 // We might have restarted this task after an evacuation pause
3926 // which might have evacuated the region we're holding on to
3927 // underneath our feet. Let's read its limit again to make sure
3928 // that we do not iterate over a region of the heap that
3929 // contains garbage (update_region_limit() will also move
3930 // _finger to the start of the region if it is found empty).
3931 update_region_limit();
3932 // We will start from _finger not from the start of the region,
3933 // as we might be restarting this task after aborting half-way
3934 // through scanning this region. In this case, _finger points to
3935 // the address where we last found a marked object. If this is a
3936 // fresh region, _finger points to start().
3937 MemRegion mr = MemRegion(_finger, _region_limit);
3938
3939 if (_cm->verbose_low()) {
3940 gclog_or_tty->print_cr("[%u] we're scanning part "
3941 "[" PTR_FORMAT ", " PTR_FORMAT ") "
3942 "of region " HR_FORMAT,
3943 _worker_id, p2i(_finger), p2i(_region_limit),
3944 HR_FORMAT_PARAMS(_curr_region));
3945 }
3946
3947 assert(!_curr_region->is_humongous() || mr.start() == _curr_region->bottom(),
3948 "humongous regions should go around loop once only");
3949
3950 // Some special cases:
3951 // If the memory region is empty, we can just give up the region.
3952 // If the current region is humongous then we only need to check
3953 // the bitmap for the bit associated with the start of the object,
3954 // scan the object if it's live, and give up the region.
3955 // Otherwise, let's iterate over the bitmap of the part of the region
3956 // that is left.
3957 // If the iteration is successful, give up the region.
3958 if (mr.is_empty()) {
3959 giveup_current_region();
3960 regular_clock_call();
3961 } else if (_curr_region->is_humongous() && mr.start() == _curr_region->bottom()) {
3962 if (_nextMarkBitMap->isMarked(mr.start())) {
3963 // The object is marked - apply the closure
3964 BitMap::idx_t offset = _nextMarkBitMap->heapWordToOffset(mr.start());
3965 bitmap_closure.do_bit(offset);
3966 }
3967 // Even if this task aborted while scanning the humongous object
3968 // we can (and should) give up the current region.
3969 giveup_current_region();
3970 regular_clock_call();
3971 } else if (_nextMarkBitMap->iterate(&bitmap_closure, mr)) {
3972 giveup_current_region();
3973 regular_clock_call();
3974 } else {
3975 assert(has_aborted(), "currently the only way to do so");
3976 // The only way to abort the bitmap iteration is to return
3977 // false from the do_bit() method. However, inside the
3978 // do_bit() method we move the _finger to point to the
3979 // object currently being looked at. So, if we bail out, we
3980 // have definitely set _finger to something non-null.
3981 assert(_finger != NULL, "invariant");
3982
3983 // Region iteration was actually aborted. So now _finger
3984 // points to the address of the object we last scanned. If we
3985 // leave it there, when we restart this task, we will rescan
3986 // the object. It is easy to avoid this. We move the finger by
3987 // enough to point to the next possible object header (the
3988 // bitmap knows by how much we need to move it as it knows its
3989 // granularity).
3990 assert(_finger < _region_limit, "invariant");
3991 HeapWord* new_finger = _nextMarkBitMap->nextObject(_finger);
3992 // Check if bitmap iteration was aborted while scanning the last object
3993 if (new_finger >= _region_limit) {
3994 giveup_current_region();
3995 } else {
3996 move_finger_to(new_finger);
3997 }
3998 }
3999 }
4000 // At this point we have either completed iterating over the
4001 // region we were holding on to, or we have aborted.
4002
4003 // We then partially drain the local queue and the global stack.
4004 // (Do we really need this?)
4005 drain_local_queue(true);
4006 drain_global_stack(true);
4007
4008 // Read the note on the claim_region() method on why it might
4009 // return NULL with potentially more regions available for
4010 // claiming and why we have to check out_of_regions() to determine
4011 // whether we're done or not.
4012 while (!has_aborted() && _curr_region == NULL && !_cm->out_of_regions()) {
4013 // We are going to try to claim a new region. We should have
4014 // given up on the previous one.
4015 // Separated the asserts so that we know which one fires.
4016 assert(_curr_region == NULL, "invariant");
4017 assert(_finger == NULL, "invariant");
4018 assert(_region_limit == NULL, "invariant");
4019 if (_cm->verbose_low()) {
4020 gclog_or_tty->print_cr("[%u] trying to claim a new region", _worker_id);
4021 }
4022 HeapRegion* claimed_region = _cm->claim_region(_worker_id);
4023 if (claimed_region != NULL) {
4024 // Yes, we managed to claim one
4025 statsOnly( ++_regions_claimed );
4026
4027 if (_cm->verbose_low()) {
4028 gclog_or_tty->print_cr("[%u] we successfully claimed "
4029 "region " PTR_FORMAT,
4030 _worker_id, p2i(claimed_region));
4031 }
4032
4033 setup_for_region(claimed_region);
4034 assert(_curr_region == claimed_region, "invariant");
4035 }
4036 // It is important to call the regular clock here. It might take
4037 // a while to claim a region if, for example, we hit a large
4038 // block of empty regions. So we need to call the regular clock
4039 // method once round the loop to make sure it's called
4040 // frequently enough.
4041 regular_clock_call();
4042 }
4043
4044 if (!has_aborted() && _curr_region == NULL) {
4045 assert(_cm->out_of_regions(),
4046 "at this point we should be out of regions");
4047 }
4048 } while ( _curr_region != NULL && !has_aborted());
4049
4050 if (!has_aborted()) {
4051 // We cannot check whether the global stack is empty, since other
4052 // tasks might be pushing objects to it concurrently.
4053 assert(_cm->out_of_regions(),
4054 "at this point we should be out of regions");
4055
4056 if (_cm->verbose_low()) {
4057 gclog_or_tty->print_cr("[%u] all regions claimed", _worker_id);
4058 }
4059
4060 // Try to reduce the number of available SATB buffers so that
4061 // remark has less work to do.
4062 drain_satb_buffers();
4063 }
4064
4065 // Since we've done everything else, we can now totally drain the
4066 // local queue and global stack.
4067 drain_local_queue(false);
4068 drain_global_stack(false);
4069
4070 // Attempt at work stealing from other task's queues.
4071 if (do_stealing && !has_aborted()) {
4072 // We have not aborted. This means that we have finished all that
4073 // we could. Let's try to do some stealing...
4074
4075 // We cannot check whether the global stack is empty, since other
4076 // tasks might be pushing objects to it concurrently.
4077 assert(_cm->out_of_regions() && _task_queue->size() == 0,
4078 "only way to reach here");
4079
4080 if (_cm->verbose_low()) {
4081 gclog_or_tty->print_cr("[%u] starting to steal", _worker_id);
4082 }
4083
4084 while (!has_aborted()) {
4085 oop obj;
4086 statsOnly( ++_steal_attempts );
4087
4088 if (_cm->try_stealing(_worker_id, &_hash_seed, obj)) {
4089 if (_cm->verbose_medium()) {
4090 gclog_or_tty->print_cr("[%u] stolen " PTR_FORMAT " successfully",
4091 _worker_id, p2i((void*) obj));
4092 }
4093
4094 statsOnly( ++_steals );
4095
4096 assert(_nextMarkBitMap->isMarked((HeapWord*) obj),
4097 "any stolen object should be marked");
4098 scan_object(obj);
4099
4100 // And since we're towards the end, let's totally drain the
4101 // local queue and global stack.
4102 drain_local_queue(false);
4103 drain_global_stack(false);
4104 } else {
4105 break;
4106 }
4107 }
4108 }
4109
4110 // If we are about to wrap up and go into termination, check if we
4111 // should raise the overflow flag.
4112 if (do_termination && !has_aborted()) {
4113 if (_cm->force_overflow()->should_force()) {
4114 _cm->set_has_overflown();
4115 regular_clock_call();
4116 }
4117 }
4118
4119 // We still haven't aborted. Now, let's try to get into the
4120 // termination protocol.
4121 if (do_termination && !has_aborted()) {
4122 // We cannot check whether the global stack is empty, since other
4123 // tasks might be concurrently pushing objects on it.
4124 // Separated the asserts so that we know which one fires.
4125 assert(_cm->out_of_regions(), "only way to reach here");
4126 assert(_task_queue->size() == 0, "only way to reach here");
4127
4128 if (_cm->verbose_low()) {
4129 gclog_or_tty->print_cr("[%u] starting termination protocol", _worker_id);
4130 }
4131
4132 _termination_start_time_ms = os::elapsedVTime() * 1000.0;
4133
4134 // The CMTask class also extends the TerminatorTerminator class,
4135 // hence its should_exit_termination() method will also decide
4136 // whether to exit the termination protocol or not.
4137 bool finished = (is_serial ||
4138 _cm->terminator()->offer_termination(this));
4139 double termination_end_time_ms = os::elapsedVTime() * 1000.0;
4140 _termination_time_ms +=
4141 termination_end_time_ms - _termination_start_time_ms;
4142
4143 if (finished) {
4144 // We're all done.
4145
4146 if (_worker_id == 0) {
4147 // let's allow task 0 to do this
4148 if (concurrent()) {
4149 assert(_cm->concurrent_marking_in_progress(), "invariant");
4150 // we need to set this to false before the next
4151 // safepoint. This way we ensure that the marking phase
4152 // doesn't observe any more heap expansions.
4153 _cm->clear_concurrent_marking_in_progress();
4154 }
4155 }
4156
4157 // We can now guarantee that the global stack is empty, since
4158 // all other tasks have finished. We separated the guarantees so
4159 // that, if a condition is false, we can immediately find out
4160 // which one.
4161 guarantee(_cm->out_of_regions(), "only way to reach here");
4162 guarantee(_cm->mark_stack_empty(), "only way to reach here");
4163 guarantee(_task_queue->size() == 0, "only way to reach here");
4164 guarantee(!_cm->has_overflown(), "only way to reach here");
4165 guarantee(!_cm->mark_stack_overflow(), "only way to reach here");
4166
4167 if (_cm->verbose_low()) {
4168 gclog_or_tty->print_cr("[%u] all tasks terminated", _worker_id);
4169 }
4170 } else {
4171 // Apparently there's more work to do. Let's abort this task. It
4172 // will restart it and we can hopefully find more things to do.
4173
4174 if (_cm->verbose_low()) {
4175 gclog_or_tty->print_cr("[%u] apparently there is more work to do",
4176 _worker_id);
4177 }
4178
4179 set_has_aborted();
4180 statsOnly( ++_aborted_termination );
4181 }
4182 }
4183
4184 // Mainly for debugging purposes to make sure that a pointer to the
4185 // closure which was statically allocated in this frame doesn't
4186 // escape it by accident.
4187 set_cm_oop_closure(NULL);
4188 double end_time_ms = os::elapsedVTime() * 1000.0;
4189 double elapsed_time_ms = end_time_ms - _start_time_ms;
4190 // Update the step history.
4191 _step_times_ms.add(elapsed_time_ms);
4192
4193 if (has_aborted()) {
4194 // The task was aborted for some reason.
4195
4196 statsOnly( ++_aborted );
4197
4198 if (_has_timed_out) {
4199 double diff_ms = elapsed_time_ms - _time_target_ms;
4200 // Keep statistics of how well we did with respect to hitting
4201 // our target only if we actually timed out (if we aborted for
4202 // other reasons, then the results might get skewed).
4203 _marking_step_diffs_ms.add(diff_ms);
4204 }
4205
4206 if (_cm->has_overflown()) {
4207 // This is the interesting one. We aborted because a global
4208 // overflow was raised. This means we have to restart the
4209 // marking phase and start iterating over regions. However, in
4210 // order to do this we have to make sure that all tasks stop
4211 // what they are doing and re-initialize in a safe manner. We
4212 // will achieve this with the use of two barrier sync points.
4213
4214 if (_cm->verbose_low()) {
4215 gclog_or_tty->print_cr("[%u] detected overflow", _worker_id);
4216 }
4217
4218 if (!is_serial) {
4219 // We only need to enter the sync barrier if being called
4220 // from a parallel context
4221 _cm->enter_first_sync_barrier(_worker_id);
4222
4223 // When we exit this sync barrier we know that all tasks have
4224 // stopped doing marking work. So, it's now safe to
4225 // re-initialize our data structures. At the end of this method,
4226 // task 0 will clear the global data structures.
4227 }
4228
4229 statsOnly( ++_aborted_overflow );
4230
4231 // We clear the local state of this task...
4232 clear_region_fields();
4233
4234 if (!is_serial) {
4235 // ...and enter the second barrier.
4236 _cm->enter_second_sync_barrier(_worker_id);
4237 }
4238 // At this point, if we're during the concurrent phase of
4239 // marking, everything has been re-initialized and we're
4240 // ready to restart.
4241 }
4242
4243 if (_cm->verbose_low()) {
4244 gclog_or_tty->print_cr("[%u] <<<<<<<<<< ABORTING, target = %1.2lfms, "
4245 "elapsed = %1.2lfms <<<<<<<<<<",
4246 _worker_id, _time_target_ms, elapsed_time_ms);
4247 if (_cm->has_aborted()) {
4248 gclog_or_tty->print_cr("[%u] ========== MARKING ABORTED ==========",
4249 _worker_id);
4250 }
4251 }
4252 } else {
4253 if (_cm->verbose_low()) {
4254 gclog_or_tty->print_cr("[%u] <<<<<<<<<< FINISHED, target = %1.2lfms, "
4255 "elapsed = %1.2lfms <<<<<<<<<<",
4256 _worker_id, _time_target_ms, elapsed_time_ms);
4257 }
4258 }
4259
4260 _claimed = false;
4261 }
4262
4263 CMTask::CMTask(uint worker_id,
4264 ConcurrentMark* cm,
4265 size_t* marked_bytes,
4266 BitMap* card_bm,
4267 CMTaskQueue* task_queue,
4268 CMTaskQueueSet* task_queues)
4269 : _g1h(G1CollectedHeap::heap()),
4270 _worker_id(worker_id), _cm(cm),
4271 _claimed(false),
4272 _nextMarkBitMap(NULL), _hash_seed(17),
4273 _task_queue(task_queue),
4274 _task_queues(task_queues),
4275 _cm_oop_closure(NULL),
4276 _marked_bytes_array(marked_bytes),
4277 _card_bm(card_bm) {
4278 guarantee(task_queue != NULL, "invariant");
4279 guarantee(task_queues != NULL, "invariant");
4280
4281 statsOnly( _clock_due_to_scanning = 0;
4282 _clock_due_to_marking = 0 );
4283
4284 _marking_step_diffs_ms.add(0.5);
4285 }
4286
4287 // These are formatting macros that are used below to ensure
4288 // consistent formatting. The *_H_* versions are used to format the
4289 // header for a particular value and they should be kept consistent
4290 // with the corresponding macro. Also note that most of the macros add
4291 // the necessary white space (as a prefix) which makes them a bit
4292 // easier to compose.
4293
4294 // All the output lines are prefixed with this string to be able to
4295 // identify them easily in a large log file.
4296 #define G1PPRL_LINE_PREFIX "###"
4297
4298 #define G1PPRL_ADDR_BASE_FORMAT " " PTR_FORMAT "-" PTR_FORMAT
4299 #ifdef _LP64
4300 #define G1PPRL_ADDR_BASE_H_FORMAT " %37s"
4301 #else // _LP64
4302 #define G1PPRL_ADDR_BASE_H_FORMAT " %21s"
4303 #endif // _LP64
4304
4305 // For per-region info
4306 #define G1PPRL_TYPE_FORMAT " %-4s"
4307 #define G1PPRL_TYPE_H_FORMAT " %4s"
4308 #define G1PPRL_BYTE_FORMAT " " SIZE_FORMAT_W(9)
4309 #define G1PPRL_BYTE_H_FORMAT " %9s"
4310 #define G1PPRL_DOUBLE_FORMAT " %14.1f"
4311 #define G1PPRL_DOUBLE_H_FORMAT " %14s"
4312
4313 // For summary info
4314 #define G1PPRL_SUM_ADDR_FORMAT(tag) " " tag ":" G1PPRL_ADDR_BASE_FORMAT
4315 #define G1PPRL_SUM_BYTE_FORMAT(tag) " " tag ": " SIZE_FORMAT
4316 #define G1PPRL_SUM_MB_FORMAT(tag) " " tag ": %1.2f MB"
4317 #define G1PPRL_SUM_MB_PERC_FORMAT(tag) G1PPRL_SUM_MB_FORMAT(tag) " / %1.2f %%"
4318
4319 G1PrintRegionLivenessInfoClosure::
4320 G1PrintRegionLivenessInfoClosure(outputStream* out, const char* phase_name)
4321 : _out(out),
4322 _total_used_bytes(0), _total_capacity_bytes(0),
4323 _total_prev_live_bytes(0), _total_next_live_bytes(0),
4324 _hum_used_bytes(0), _hum_capacity_bytes(0),
4325 _hum_prev_live_bytes(0), _hum_next_live_bytes(0),
4326 _total_remset_bytes(0), _total_strong_code_roots_bytes(0) {
4327 G1CollectedHeap* g1h = G1CollectedHeap::heap();
4328 MemRegion g1_reserved = g1h->g1_reserved();
4329 double now = os::elapsedTime();
4330
4331 // Print the header of the output.
4332 _out->cr();
4333 _out->print_cr(G1PPRL_LINE_PREFIX" PHASE %s @ %1.3f", phase_name, now);
4334 _out->print_cr(G1PPRL_LINE_PREFIX" HEAP"
4335 G1PPRL_SUM_ADDR_FORMAT("reserved")
4336 G1PPRL_SUM_BYTE_FORMAT("region-size"),
4337 p2i(g1_reserved.start()), p2i(g1_reserved.end()),
4338 HeapRegion::GrainBytes);
4339 _out->print_cr(G1PPRL_LINE_PREFIX);
4340 _out->print_cr(G1PPRL_LINE_PREFIX
4341 G1PPRL_TYPE_H_FORMAT
4342 G1PPRL_ADDR_BASE_H_FORMAT
4343 G1PPRL_BYTE_H_FORMAT
4344 G1PPRL_BYTE_H_FORMAT
4345 G1PPRL_BYTE_H_FORMAT
4346 G1PPRL_DOUBLE_H_FORMAT
4347 G1PPRL_BYTE_H_FORMAT
4348 G1PPRL_BYTE_H_FORMAT,
4349 "type", "address-range",
4350 "used", "prev-live", "next-live", "gc-eff",
4351 "remset", "code-roots");
4352 _out->print_cr(G1PPRL_LINE_PREFIX
4353 G1PPRL_TYPE_H_FORMAT
4354 G1PPRL_ADDR_BASE_H_FORMAT
4355 G1PPRL_BYTE_H_FORMAT
4356 G1PPRL_BYTE_H_FORMAT
4357 G1PPRL_BYTE_H_FORMAT
4358 G1PPRL_DOUBLE_H_FORMAT
4359 G1PPRL_BYTE_H_FORMAT
4360 G1PPRL_BYTE_H_FORMAT,
4361 "", "",
4362 "(bytes)", "(bytes)", "(bytes)", "(bytes/ms)",
4363 "(bytes)", "(bytes)");
4364 }
4365
4366 // It takes as a parameter a reference to one of the _hum_* fields, it
4367 // deduces the corresponding value for a region in a humongous region
4368 // series (either the region size, or what's left if the _hum_* field
4369 // is < the region size), and updates the _hum_* field accordingly.
4370 size_t G1PrintRegionLivenessInfoClosure::get_hum_bytes(size_t* hum_bytes) {
4371 size_t bytes = 0;
4372 // The > 0 check is to deal with the prev and next live bytes which
4373 // could be 0.
4374 if (*hum_bytes > 0) {
4375 bytes = MIN2(HeapRegion::GrainBytes, *hum_bytes);
4376 *hum_bytes -= bytes;
4377 }
4378 return bytes;
4379 }
4380
4381 // It deduces the values for a region in a humongous region series
4382 // from the _hum_* fields and updates those accordingly. It assumes
4383 // that that _hum_* fields have already been set up from the "starts
4384 // humongous" region and we visit the regions in address order.
4385 void G1PrintRegionLivenessInfoClosure::get_hum_bytes(size_t* used_bytes,
4386 size_t* capacity_bytes,
4387 size_t* prev_live_bytes,
4388 size_t* next_live_bytes) {
4389 assert(_hum_used_bytes > 0 && _hum_capacity_bytes > 0, "pre-condition");
4390 *used_bytes = get_hum_bytes(&_hum_used_bytes);
4391 *capacity_bytes = get_hum_bytes(&_hum_capacity_bytes);
4392 *prev_live_bytes = get_hum_bytes(&_hum_prev_live_bytes);
4393 *next_live_bytes = get_hum_bytes(&_hum_next_live_bytes);
4394 }
4395
4396 bool G1PrintRegionLivenessInfoClosure::doHeapRegion(HeapRegion* r) {
4397 const char* type = r->get_type_str();
4398 HeapWord* bottom = r->bottom();
4399 HeapWord* end = r->end();
4400 size_t capacity_bytes = r->capacity();
4401 size_t used_bytes = r->used();
4402 size_t prev_live_bytes = r->live_bytes();
4403 size_t next_live_bytes = r->next_live_bytes();
4404 double gc_eff = r->gc_efficiency();
4405 size_t remset_bytes = r->rem_set()->mem_size();
4406 size_t strong_code_roots_bytes = r->rem_set()->strong_code_roots_mem_size();
4407
4408 if (r->is_starts_humongous()) {
4409 assert(_hum_used_bytes == 0 && _hum_capacity_bytes == 0 &&
4410 _hum_prev_live_bytes == 0 && _hum_next_live_bytes == 0,
4411 "they should have been zeroed after the last time we used them");
4412 // Set up the _hum_* fields.
4413 _hum_capacity_bytes = capacity_bytes;
4414 _hum_used_bytes = used_bytes;
4415 _hum_prev_live_bytes = prev_live_bytes;
4416 _hum_next_live_bytes = next_live_bytes;
4417 get_hum_bytes(&used_bytes, &capacity_bytes,
4418 &prev_live_bytes, &next_live_bytes);
4419 end = bottom + HeapRegion::GrainWords;
4420 } else if (r->is_continues_humongous()) {
4421 get_hum_bytes(&used_bytes, &capacity_bytes,
4422 &prev_live_bytes, &next_live_bytes);
4423 assert(end == bottom + HeapRegion::GrainWords, "invariant");
4424 }
4425
4426 _total_used_bytes += used_bytes;
4427 _total_capacity_bytes += capacity_bytes;
4428 _total_prev_live_bytes += prev_live_bytes;
4429 _total_next_live_bytes += next_live_bytes;
4430 _total_remset_bytes += remset_bytes;
4431 _total_strong_code_roots_bytes += strong_code_roots_bytes;
4432
4433 // Print a line for this particular region.
4434 _out->print_cr(G1PPRL_LINE_PREFIX
4435 G1PPRL_TYPE_FORMAT
4436 G1PPRL_ADDR_BASE_FORMAT
4437 G1PPRL_BYTE_FORMAT
4438 G1PPRL_BYTE_FORMAT
4439 G1PPRL_BYTE_FORMAT
4440 G1PPRL_DOUBLE_FORMAT
4441 G1PPRL_BYTE_FORMAT
4442 G1PPRL_BYTE_FORMAT,
4443 type, p2i(bottom), p2i(end),
4444 used_bytes, prev_live_bytes, next_live_bytes, gc_eff,
4445 remset_bytes, strong_code_roots_bytes);
4446
4447 return false;
4448 }
4449
4450 G1PrintRegionLivenessInfoClosure::~G1PrintRegionLivenessInfoClosure() {
4451 // add static memory usages to remembered set sizes
4452 _total_remset_bytes += HeapRegionRemSet::fl_mem_size() + HeapRegionRemSet::static_mem_size();
4453 // Print the footer of the output.
4454 _out->print_cr(G1PPRL_LINE_PREFIX);
4455 _out->print_cr(G1PPRL_LINE_PREFIX
4456 " SUMMARY"
4457 G1PPRL_SUM_MB_FORMAT("capacity")
4458 G1PPRL_SUM_MB_PERC_FORMAT("used")
4459 G1PPRL_SUM_MB_PERC_FORMAT("prev-live")
4460 G1PPRL_SUM_MB_PERC_FORMAT("next-live")
4461 G1PPRL_SUM_MB_FORMAT("remset")
4462 G1PPRL_SUM_MB_FORMAT("code-roots"),
4463 bytes_to_mb(_total_capacity_bytes),
4464 bytes_to_mb(_total_used_bytes),
4465 perc(_total_used_bytes, _total_capacity_bytes),
4466 bytes_to_mb(_total_prev_live_bytes),
4467 perc(_total_prev_live_bytes, _total_capacity_bytes),
4468 bytes_to_mb(_total_next_live_bytes),
4469 perc(_total_next_live_bytes, _total_capacity_bytes),
4470 bytes_to_mb(_total_remset_bytes),
4471 bytes_to_mb(_total_strong_code_roots_bytes));
4472 _out->cr();
4473 }