1 /*
2 * Copyright (c) 2001, 2010, 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 #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_HPP
26 #define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_HPP
27
28 #include "gc_implementation/g1/heapRegion.hpp"
29 #include "utilities/taskqueue.hpp"
30
31 class G1CollectedHeap;
32 class CMTask;
33 typedef GenericTaskQueue<oop> CMTaskQueue;
34 typedef GenericTaskQueueSet<CMTaskQueue> CMTaskQueueSet;
35
36 // A generic CM bit map. This is essentially a wrapper around the BitMap
37 // class, with one bit per (1<<_shifter) HeapWords.
38
39 class CMBitMapRO VALUE_OBJ_CLASS_SPEC {
40 protected:
41 HeapWord* _bmStartWord; // base address of range covered by map
42 size_t _bmWordSize; // map size (in #HeapWords covered)
43 const int _shifter; // map to char or bit
44 VirtualSpace _virtual_space; // underlying the bit map
45 BitMap _bm; // the bit map itself
46
47 public:
48 // constructor
49 CMBitMapRO(ReservedSpace rs, int shifter);
50
51 enum { do_yield = true };
52
53 // inquiries
54 HeapWord* startWord() const { return _bmStartWord; }
55 size_t sizeInWords() const { return _bmWordSize; }
56 // the following is one past the last word in space
57 HeapWord* endWord() const { return _bmStartWord + _bmWordSize; }
58
59 // read marks
60
61 bool isMarked(HeapWord* addr) const {
62 assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
63 "outside underlying space?");
64 return _bm.at(heapWordToOffset(addr));
65 }
66
67 // iteration
68 bool iterate(BitMapClosure* cl) { return _bm.iterate(cl); }
69 bool iterate(BitMapClosure* cl, MemRegion mr);
70
71 // Return the address corresponding to the next marked bit at or after
72 // "addr", and before "limit", if "limit" is non-NULL. If there is no
73 // such bit, returns "limit" if that is non-NULL, or else "endWord()".
74 HeapWord* getNextMarkedWordAddress(HeapWord* addr,
75 HeapWord* limit = NULL) const;
76 // Return the address corresponding to the next unmarked bit at or after
77 // "addr", and before "limit", if "limit" is non-NULL. If there is no
78 // such bit, returns "limit" if that is non-NULL, or else "endWord()".
79 HeapWord* getNextUnmarkedWordAddress(HeapWord* addr,
80 HeapWord* limit = NULL) const;
81
82 // conversion utilities
83 // XXX Fix these so that offsets are size_t's...
84 HeapWord* offsetToHeapWord(size_t offset) const {
85 return _bmStartWord + (offset << _shifter);
86 }
87 size_t heapWordToOffset(HeapWord* addr) const {
88 return pointer_delta(addr, _bmStartWord) >> _shifter;
89 }
90 int heapWordDiffToOffsetDiff(size_t diff) const;
91 HeapWord* nextWord(HeapWord* addr) {
92 return offsetToHeapWord(heapWordToOffset(addr) + 1);
93 }
94
95 void mostly_disjoint_range_union(BitMap* from_bitmap,
96 size_t from_start_index,
97 HeapWord* to_start_word,
98 size_t word_num);
99
100 // debugging
101 NOT_PRODUCT(bool covers(ReservedSpace rs) const;)
102 };
103
104 class CMBitMap : public CMBitMapRO {
105
106 public:
107 // constructor
108 CMBitMap(ReservedSpace rs, int shifter) :
109 CMBitMapRO(rs, shifter) {}
110
111 // write marks
112 void mark(HeapWord* addr) {
113 assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
114 "outside underlying space?");
115 _bm.at_put(heapWordToOffset(addr), true);
116 }
117 void clear(HeapWord* addr) {
118 assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
119 "outside underlying space?");
120 _bm.at_put(heapWordToOffset(addr), false);
121 }
122 bool parMark(HeapWord* addr) {
123 assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
124 "outside underlying space?");
125 return _bm.par_at_put(heapWordToOffset(addr), true);
126 }
127 bool parClear(HeapWord* addr) {
128 assert(_bmStartWord <= addr && addr < (_bmStartWord + _bmWordSize),
129 "outside underlying space?");
130 return _bm.par_at_put(heapWordToOffset(addr), false);
131 }
132 void markRange(MemRegion mr);
133 void clearAll();
134 void clearRange(MemRegion mr);
135
136 // Starting at the bit corresponding to "addr" (inclusive), find the next
137 // "1" bit, if any. This bit starts some run of consecutive "1"'s; find
138 // the end of this run (stopping at "end_addr"). Return the MemRegion
139 // covering from the start of the region corresponding to the first bit
140 // of the run to the end of the region corresponding to the last bit of
141 // the run. If there is no "1" bit at or after "addr", return an empty
142 // MemRegion.
143 MemRegion getAndClearMarkedRegion(HeapWord* addr, HeapWord* end_addr);
144 };
145
146 // Represents a marking stack used by the CM collector.
147 // Ideally this should be GrowableArray<> just like MSC's marking stack(s).
148 class CMMarkStack VALUE_OBJ_CLASS_SPEC {
149 ConcurrentMark* _cm;
150 oop* _base; // bottom of stack
151 jint _index; // one more than last occupied index
152 jint _capacity; // max #elements
153 jint _oops_do_bound; // Number of elements to include in next iteration.
154 NOT_PRODUCT(jint _max_depth;) // max depth plumbed during run
155
156 bool _overflow;
157 DEBUG_ONLY(bool _drain_in_progress;)
158 DEBUG_ONLY(bool _drain_in_progress_yields;)
159
160 public:
161 CMMarkStack(ConcurrentMark* cm);
162 ~CMMarkStack();
163
164 void allocate(size_t size);
165
166 oop pop() {
167 if (!isEmpty()) {
168 return _base[--_index] ;
169 }
170 return NULL;
171 }
172
173 // If overflow happens, don't do the push, and record the overflow.
174 // *Requires* that "ptr" is already marked.
175 void push(oop ptr) {
176 if (isFull()) {
177 // Record overflow.
178 _overflow = true;
179 return;
180 } else {
181 _base[_index++] = ptr;
182 NOT_PRODUCT(_max_depth = MAX2(_max_depth, _index));
183 }
184 }
185 // Non-block impl. Note: concurrency is allowed only with other
186 // "par_push" operations, not with "pop" or "drain". We would need
187 // parallel versions of them if such concurrency was desired.
188 void par_push(oop ptr);
189
190 // Pushes the first "n" elements of "ptr_arr" on the stack.
191 // Non-block impl. Note: concurrency is allowed only with other
192 // "par_adjoin_arr" or "push" operations, not with "pop" or "drain".
193 void par_adjoin_arr(oop* ptr_arr, int n);
194
195 // Pushes the first "n" elements of "ptr_arr" on the stack.
196 // Locking impl: concurrency is allowed only with
197 // "par_push_arr" and/or "par_pop_arr" operations, which use the same
198 // locking strategy.
199 void par_push_arr(oop* ptr_arr, int n);
200
201 // If returns false, the array was empty. Otherwise, removes up to "max"
202 // elements from the stack, and transfers them to "ptr_arr" in an
203 // unspecified order. The actual number transferred is given in "n" ("n
204 // == 0" is deliberately redundant with the return value.) Locking impl:
205 // concurrency is allowed only with "par_push_arr" and/or "par_pop_arr"
206 // operations, which use the same locking strategy.
207 bool par_pop_arr(oop* ptr_arr, int max, int* n);
208
209 // Drain the mark stack, applying the given closure to all fields of
210 // objects on the stack. (That is, continue until the stack is empty,
211 // even if closure applications add entries to the stack.) The "bm"
212 // argument, if non-null, may be used to verify that only marked objects
213 // are on the mark stack. If "yield_after" is "true", then the
214 // concurrent marker performing the drain offers to yield after
215 // processing each object. If a yield occurs, stops the drain operation
216 // and returns false. Otherwise, returns true.
217 template<class OopClosureClass>
218 bool drain(OopClosureClass* cl, CMBitMap* bm, bool yield_after = false);
219
220 bool isEmpty() { return _index == 0; }
221 bool isFull() { return _index == _capacity; }
222 int maxElems() { return _capacity; }
223
224 bool overflow() { return _overflow; }
225 void clear_overflow() { _overflow = false; }
226
227 int size() { return _index; }
228
229 void setEmpty() { _index = 0; clear_overflow(); }
230
231 // Record the current size; a subsequent "oops_do" will iterate only over
232 // indices valid at the time of this call.
233 void set_oops_do_bound(jint bound = -1) {
234 if (bound == -1) {
235 _oops_do_bound = _index;
236 } else {
237 _oops_do_bound = bound;
238 }
239 }
240 jint oops_do_bound() { return _oops_do_bound; }
241 // iterate over the oops in the mark stack, up to the bound recorded via
242 // the call above.
243 void oops_do(OopClosure* f);
244 };
245
246 class CMRegionStack VALUE_OBJ_CLASS_SPEC {
247 MemRegion* _base;
248 jint _capacity;
249 jint _index;
250 jint _oops_do_bound;
251 bool _overflow;
252 public:
253 CMRegionStack();
254 ~CMRegionStack();
255 void allocate(size_t size);
256
257 // This is lock-free; assumes that it will only be called in parallel
258 // with other "push" operations (no pops).
259 void push_lock_free(MemRegion mr);
260
261 // Lock-free; assumes that it will only be called in parallel
262 // with other "pop" operations (no pushes).
263 MemRegion pop_lock_free();
264
265 #if 0
266 // The routines that manipulate the region stack with a lock are
267 // not currently used. They should be retained, however, as a
268 // diagnostic aid.
269
270 // These two are the implementations that use a lock. They can be
271 // called concurrently with each other but they should not be called
272 // concurrently with the lock-free versions (push() / pop()).
273 void push_with_lock(MemRegion mr);
274 MemRegion pop_with_lock();
275 #endif
276
277 bool isEmpty() { return _index == 0; }
278 bool isFull() { return _index == _capacity; }
279
280 bool overflow() { return _overflow; }
281 void clear_overflow() { _overflow = false; }
282
283 int size() { return _index; }
284
285 // It iterates over the entries in the region stack and it
286 // invalidates (i.e. assigns MemRegion()) the ones that point to
287 // regions in the collection set.
288 bool invalidate_entries_into_cset();
289
290 // This gives an upper bound up to which the iteration in
291 // invalidate_entries_into_cset() will reach. This prevents
292 // newly-added entries to be unnecessarily scanned.
293 void set_oops_do_bound() {
294 _oops_do_bound = _index;
295 }
296
297 void setEmpty() { _index = 0; clear_overflow(); }
298 };
299
300 // this will enable a variety of different statistics per GC task
301 #define _MARKING_STATS_ 0
302 // this will enable the higher verbose levels
303 #define _MARKING_VERBOSE_ 0
304
305 #if _MARKING_STATS_
306 #define statsOnly(statement) \
307 do { \
308 statement ; \
309 } while (0)
310 #else // _MARKING_STATS_
311 #define statsOnly(statement) \
312 do { \
313 } while (0)
314 #endif // _MARKING_STATS_
315
316 typedef enum {
317 no_verbose = 0, // verbose turned off
318 stats_verbose, // only prints stats at the end of marking
319 low_verbose, // low verbose, mostly per region and per major event
320 medium_verbose, // a bit more detailed than low
321 high_verbose // per object verbose
322 } CMVerboseLevel;
323
324
325 class ConcurrentMarkThread;
326
327 class ConcurrentMark: public CHeapObj {
328 friend class ConcurrentMarkThread;
329 friend class CMTask;
330 friend class CMBitMapClosure;
331 friend class CSMarkOopClosure;
332 friend class CMGlobalObjectClosure;
333 friend class CMRemarkTask;
334 friend class CMConcurrentMarkingTask;
335 friend class G1ParNoteEndTask;
336 friend class CalcLiveObjectsClosure;
337
338 protected:
339 ConcurrentMarkThread* _cmThread; // the thread doing the work
340 G1CollectedHeap* _g1h; // the heap.
341 size_t _parallel_marking_threads; // the number of marking
342 // threads we'll use
343 double _sleep_factor; // how much we have to sleep, with
344 // respect to the work we just did, to
345 // meet the marking overhead goal
346 double _marking_task_overhead; // marking target overhead for
347 // a single task
348
349 // same as the two above, but for the cleanup task
350 double _cleanup_sleep_factor;
351 double _cleanup_task_overhead;
352
353 // Stuff related to age cohort processing.
354 struct ParCleanupThreadState {
355 char _pre[64];
356 UncleanRegionList list;
357 char _post[64];
358 };
359 ParCleanupThreadState** _par_cleanup_thread_state;
360
361 // CMS marking support structures
362 CMBitMap _markBitMap1;
363 CMBitMap _markBitMap2;
364 CMBitMapRO* _prevMarkBitMap; // completed mark bitmap
365 CMBitMap* _nextMarkBitMap; // under-construction mark bitmap
366 bool _at_least_one_mark_complete;
367
368 BitMap _region_bm;
369 BitMap _card_bm;
370
371 // Heap bounds
372 HeapWord* _heap_start;
373 HeapWord* _heap_end;
374
375 // For gray objects
376 CMMarkStack _markStack; // Grey objects behind global finger.
377 CMRegionStack _regionStack; // Grey regions behind global finger.
378 HeapWord* volatile _finger; // the global finger, region aligned,
379 // always points to the end of the
380 // last claimed region
381
382 // marking tasks
383 size_t _max_task_num; // maximum task number
384 size_t _active_tasks; // task num currently active
385 CMTask** _tasks; // task queue array (max_task_num len)
386 CMTaskQueueSet* _task_queues; // task queue set
387 ParallelTaskTerminator _terminator; // for termination
388
389 // Two sync barriers that are used to synchronise tasks when an
390 // overflow occurs. The algorithm is the following. All tasks enter
391 // the first one to ensure that they have all stopped manipulating
392 // the global data structures. After they exit it, they re-initialise
393 // their data structures and task 0 re-initialises the global data
394 // structures. Then, they enter the second sync barrier. This
395 // ensure, that no task starts doing work before all data
396 // structures (local and global) have been re-initialised. When they
397 // exit it, they are free to start working again.
398 WorkGangBarrierSync _first_overflow_barrier_sync;
399 WorkGangBarrierSync _second_overflow_barrier_sync;
400
401
402 // this is set by any task, when an overflow on the global data
403 // structures is detected.
404 volatile bool _has_overflown;
405 // true: marking is concurrent, false: we're in remark
406 volatile bool _concurrent;
407 // set at the end of a Full GC so that marking aborts
408 volatile bool _has_aborted;
409
410 // used when remark aborts due to an overflow to indicate that
411 // another concurrent marking phase should start
412 volatile bool _restart_for_overflow;
413
414 // This is true from the very start of concurrent marking until the
415 // point when all the tasks complete their work. It is really used
416 // to determine the points between the end of concurrent marking and
417 // time of remark.
418 volatile bool _concurrent_marking_in_progress;
419
420 // verbose level
421 CMVerboseLevel _verbose_level;
422
423 // These two fields are used to implement the optimisation that
424 // avoids pushing objects on the global/region stack if there are
425 // no collection set regions above the lowest finger.
426
427 // This is the lowest finger (among the global and local fingers),
428 // which is calculated before a new collection set is chosen.
429 HeapWord* _min_finger;
430 // If this flag is true, objects/regions that are marked below the
431 // finger should be pushed on the stack(s). If this is flag is
432 // false, it is safe not to push them on the stack(s).
433 bool _should_gray_objects;
434
435 // All of these times are in ms.
436 NumberSeq _init_times;
437 NumberSeq _remark_times;
438 NumberSeq _remark_mark_times;
439 NumberSeq _remark_weak_ref_times;
440 NumberSeq _cleanup_times;
441 double _total_counting_time;
442 double _total_rs_scrub_time;
443
444 double* _accum_task_vtime; // accumulated task vtime
445
446 WorkGang* _parallel_workers;
447
448 void weakRefsWork(bool clear_all_soft_refs);
449
450 void swapMarkBitMaps();
451
452 // It resets the global marking data structures, as well as the
453 // task local ones; should be called during initial mark.
454 void reset();
455 // It resets all the marking data structures.
456 void clear_marking_state();
457
458 // It should be called to indicate which phase we're in (concurrent
459 // mark or remark) and how many threads are currently active.
460 void set_phase(size_t active_tasks, bool concurrent);
461 // We do this after we're done with marking so that the marking data
462 // structures are initialised to a sensible and predictable state.
463 void set_non_marking_state();
464
465 // prints all gathered CM-related statistics
466 void print_stats();
467
468 // accessor methods
469 size_t parallel_marking_threads() { return _parallel_marking_threads; }
470 double sleep_factor() { return _sleep_factor; }
471 double marking_task_overhead() { return _marking_task_overhead;}
472 double cleanup_sleep_factor() { return _cleanup_sleep_factor; }
473 double cleanup_task_overhead() { return _cleanup_task_overhead;}
474
475 HeapWord* finger() { return _finger; }
476 bool concurrent() { return _concurrent; }
477 size_t active_tasks() { return _active_tasks; }
478 ParallelTaskTerminator* terminator() { return &_terminator; }
479
480 // It claims the next available region to be scanned by a marking
481 // task. It might return NULL if the next region is empty or we have
482 // run out of regions. In the latter case, out_of_regions()
483 // determines whether we've really run out of regions or the task
484 // should call claim_region() again. This might seem a bit
485 // awkward. Originally, the code was written so that claim_region()
486 // either successfully returned with a non-empty region or there
487 // were no more regions to be claimed. The problem with this was
488 // that, in certain circumstances, it iterated over large chunks of
489 // the heap finding only empty regions and, while it was working, it
490 // was preventing the calling task to call its regular clock
491 // method. So, this way, each task will spend very little time in
492 // claim_region() and is allowed to call the regular clock method
493 // frequently.
494 HeapRegion* claim_region(int task);
495
496 // It determines whether we've run out of regions to scan.
497 bool out_of_regions() { return _finger == _heap_end; }
498
499 // Returns the task with the given id
500 CMTask* task(int id) {
501 assert(0 <= id && id < (int) _active_tasks,
502 "task id not within active bounds");
503 return _tasks[id];
504 }
505
506 // Returns the task queue with the given id
507 CMTaskQueue* task_queue(int id) {
508 assert(0 <= id && id < (int) _active_tasks,
509 "task queue id not within active bounds");
510 return (CMTaskQueue*) _task_queues->queue(id);
511 }
512
513 // Returns the task queue set
514 CMTaskQueueSet* task_queues() { return _task_queues; }
515
516 // Access / manipulation of the overflow flag which is set to
517 // indicate that the global stack or region stack has overflown
518 bool has_overflown() { return _has_overflown; }
519 void set_has_overflown() { _has_overflown = true; }
520 void clear_has_overflown() { _has_overflown = false; }
521
522 bool has_aborted() { return _has_aborted; }
523 bool restart_for_overflow() { return _restart_for_overflow; }
524
525 // Methods to enter the two overflow sync barriers
526 void enter_first_sync_barrier(int task_num);
527 void enter_second_sync_barrier(int task_num);
528
529 public:
530 // Manipulation of the global mark stack.
531 // Notice that the first mark_stack_push is CAS-based, whereas the
532 // two below are Mutex-based. This is OK since the first one is only
533 // called during evacuation pauses and doesn't compete with the
534 // other two (which are called by the marking tasks during
535 // concurrent marking or remark).
536 bool mark_stack_push(oop p) {
537 _markStack.par_push(p);
538 if (_markStack.overflow()) {
539 set_has_overflown();
540 return false;
541 }
542 return true;
543 }
544 bool mark_stack_push(oop* arr, int n) {
545 _markStack.par_push_arr(arr, n);
546 if (_markStack.overflow()) {
547 set_has_overflown();
548 return false;
549 }
550 return true;
551 }
552 void mark_stack_pop(oop* arr, int max, int* n) {
553 _markStack.par_pop_arr(arr, max, n);
554 }
555 size_t mark_stack_size() { return _markStack.size(); }
556 size_t partial_mark_stack_size_target() { return _markStack.maxElems()/3; }
557 bool mark_stack_overflow() { return _markStack.overflow(); }
558 bool mark_stack_empty() { return _markStack.isEmpty(); }
559
560 // (Lock-free) Manipulation of the region stack
561 bool region_stack_push_lock_free(MemRegion mr) {
562 // Currently we only call the lock-free version during evacuation
563 // pauses.
564 assert(SafepointSynchronize::is_at_safepoint(), "world should be stopped");
565
566 _regionStack.push_lock_free(mr);
567 if (_regionStack.overflow()) {
568 set_has_overflown();
569 return false;
570 }
571 return true;
572 }
573
574 // Lock-free version of region-stack pop. Should only be
575 // called in tandem with other lock-free pops.
576 MemRegion region_stack_pop_lock_free() {
577 return _regionStack.pop_lock_free();
578 }
579
580 #if 0
581 // The routines that manipulate the region stack with a lock are
582 // not currently used. They should be retained, however, as a
583 // diagnostic aid.
584
585 bool region_stack_push_with_lock(MemRegion mr) {
586 // Currently we only call the lock-based version during either
587 // concurrent marking or remark.
588 assert(!SafepointSynchronize::is_at_safepoint() || !concurrent(),
589 "if we are at a safepoint it should be the remark safepoint");
590
591 _regionStack.push_with_lock(mr);
592 if (_regionStack.overflow()) {
593 set_has_overflown();
594 return false;
595 }
596 return true;
597 }
598
599 MemRegion region_stack_pop_with_lock() {
600 // Currently we only call the lock-based version during either
601 // concurrent marking or remark.
602 assert(!SafepointSynchronize::is_at_safepoint() || !concurrent(),
603 "if we are at a safepoint it should be the remark safepoint");
604
605 return _regionStack.pop_with_lock();
606 }
607 #endif
608
609 int region_stack_size() { return _regionStack.size(); }
610 bool region_stack_overflow() { return _regionStack.overflow(); }
611 bool region_stack_empty() { return _regionStack.isEmpty(); }
612
613 // Iterate over any regions that were aborted while draining the
614 // region stack (any such regions are saved in the corresponding
615 // CMTask) and invalidate (i.e. assign to the empty MemRegion())
616 // any regions that point into the collection set.
617 bool invalidate_aborted_regions_in_cset();
618
619 // Returns true if there are any aborted memory regions.
620 bool has_aborted_regions();
621
622 bool concurrent_marking_in_progress() {
623 return _concurrent_marking_in_progress;
624 }
625 void set_concurrent_marking_in_progress() {
626 _concurrent_marking_in_progress = true;
627 }
628 void clear_concurrent_marking_in_progress() {
629 _concurrent_marking_in_progress = false;
630 }
631
632 void update_accum_task_vtime(int i, double vtime) {
633 _accum_task_vtime[i] += vtime;
634 }
635
636 double all_task_accum_vtime() {
637 double ret = 0.0;
638 for (int i = 0; i < (int)_max_task_num; ++i)
639 ret += _accum_task_vtime[i];
640 return ret;
641 }
642
643 // Attempts to steal an object from the task queues of other tasks
644 bool try_stealing(int task_num, int* hash_seed, oop& obj) {
645 return _task_queues->steal(task_num, hash_seed, obj);
646 }
647
648 // It grays an object by first marking it. Then, if it's behind the
649 // global finger, it also pushes it on the global stack.
650 void deal_with_reference(oop obj);
651
652 ConcurrentMark(ReservedSpace rs, int max_regions);
653 ~ConcurrentMark();
654 ConcurrentMarkThread* cmThread() { return _cmThread; }
655
656 CMBitMapRO* prevMarkBitMap() const { return _prevMarkBitMap; }
657 CMBitMap* nextMarkBitMap() const { return _nextMarkBitMap; }
658
659 // The following three are interaction between CM and
660 // G1CollectedHeap
661
662 // This notifies CM that a root during initial-mark needs to be
663 // grayed and it's MT-safe. Currently, we just mark it. But, in the
664 // future, we can experiment with pushing it on the stack and we can
665 // do this without changing G1CollectedHeap.
666 void grayRoot(oop p);
667 // It's used during evacuation pauses to gray a region, if
668 // necessary, and it's MT-safe. It assumes that the caller has
669 // marked any objects on that region. If _should_gray_objects is
670 // true and we're still doing concurrent marking, the region is
671 // pushed on the region stack, if it is located below the global
672 // finger, otherwise we do nothing.
673 void grayRegionIfNecessary(MemRegion mr);
674 // It's used during evacuation pauses to mark and, if necessary,
675 // gray a single object and it's MT-safe. It assumes the caller did
676 // not mark the object. If _should_gray_objects is true and we're
677 // still doing concurrent marking, the objects is pushed on the
678 // global stack, if it is located below the global finger, otherwise
679 // we do nothing.
680 void markAndGrayObjectIfNecessary(oop p);
681
682 // It iterates over the heap and for each object it comes across it
683 // will dump the contents of its reference fields, as well as
684 // liveness information for the object and its referents. The dump
685 // will be written to a file with the following name:
686 // G1PrintReachableBaseFile + "." + str. use_prev_marking decides
687 // whether the prev (use_prev_marking == true) or next
688 // (use_prev_marking == false) marking information will be used to
689 // determine the liveness of each object / referent. If all is true,
690 // all objects in the heap will be dumped, otherwise only the live
691 // ones. In the dump the following symbols / abbreviations are used:
692 // M : an explicitly live object (its bitmap bit is set)
693 // > : an implicitly live object (over tams)
694 // O : an object outside the G1 heap (typically: in the perm gen)
695 // NOT : a reference field whose referent is not live
696 // AND MARKED : indicates that an object is both explicitly and
697 // implicitly live (it should be one or the other, not both)
698 void print_reachable(const char* str,
699 bool use_prev_marking, bool all) PRODUCT_RETURN;
700
701 // Clear the next marking bitmap (will be called concurrently).
702 void clearNextBitmap();
703
704 // main CMS steps and related support
705 void checkpointRootsInitial();
706
707 // These two do the work that needs to be done before and after the
708 // initial root checkpoint. Since this checkpoint can be done at two
709 // different points (i.e. an explicit pause or piggy-backed on a
710 // young collection), then it's nice to be able to easily share the
711 // pre/post code. It might be the case that we can put everything in
712 // the post method. TP
713 void checkpointRootsInitialPre();
714 void checkpointRootsInitialPost();
715
716 // Do concurrent phase of marking, to a tentative transitive closure.
717 void markFromRoots();
718
719 // Process all unprocessed SATB buffers. It is called at the
720 // beginning of an evacuation pause.
721 void drainAllSATBBuffers();
722
723 void checkpointRootsFinal(bool clear_all_soft_refs);
724 void checkpointRootsFinalWork();
725 void calcDesiredRegions();
726 void cleanup();
727 void completeCleanup();
728
729 // Mark in the previous bitmap. NB: this is usually read-only, so use
730 // this carefully!
731 void markPrev(oop p);
732 void clear(oop p);
733 // Clears marks for all objects in the given range, for both prev and
734 // next bitmaps. NB: the previous bitmap is usually read-only, so use
735 // this carefully!
736 void clearRangeBothMaps(MemRegion mr);
737
738 // Record the current top of the mark and region stacks; a
739 // subsequent oops_do() on the mark stack and
740 // invalidate_entries_into_cset() on the region stack will iterate
741 // only over indices valid at the time of this call.
742 void set_oops_do_bound() {
743 _markStack.set_oops_do_bound();
744 _regionStack.set_oops_do_bound();
745 }
746 // Iterate over the oops in the mark stack and all local queues. It
747 // also calls invalidate_entries_into_cset() on the region stack.
748 void oops_do(OopClosure* f);
749 // It is called at the end of an evacuation pause during marking so
750 // that CM is notified of where the new end of the heap is. It
751 // doesn't do anything if concurrent_marking_in_progress() is false,
752 // unless the force parameter is true.
753 void update_g1_committed(bool force = false);
754
755 void complete_marking_in_collection_set();
756
757 // It indicates that a new collection set is being chosen.
758 void newCSet();
759 // It registers a collection set heap region with CM. This is used
760 // to determine whether any heap regions are located above the finger.
761 void registerCSetRegion(HeapRegion* hr);
762
763 // Registers the maximum region-end associated with a set of
764 // regions with CM. Again this is used to determine whether any
765 // heap regions are located above the finger.
766 void register_collection_set_finger(HeapWord* max_finger) {
767 // max_finger is the highest heap region end of the regions currently
768 // contained in the collection set. If this value is larger than
769 // _min_finger then we need to gray objects.
770 // This routine is like registerCSetRegion but for an entire
771 // collection of regions.
772 if (max_finger > _min_finger)
773 _should_gray_objects = true;
774 }
775
776 // Returns "true" if at least one mark has been completed.
777 bool at_least_one_mark_complete() { return _at_least_one_mark_complete; }
778
779 bool isMarked(oop p) const {
780 assert(p != NULL && p->is_oop(), "expected an oop");
781 HeapWord* addr = (HeapWord*)p;
782 assert(addr >= _nextMarkBitMap->startWord() ||
783 addr < _nextMarkBitMap->endWord(), "in a region");
784
785 return _nextMarkBitMap->isMarked(addr);
786 }
787
788 inline bool not_yet_marked(oop p) const;
789
790 // XXX Debug code
791 bool containing_card_is_marked(void* p);
792 bool containing_cards_are_marked(void* start, void* last);
793
794 bool isPrevMarked(oop p) const {
795 assert(p != NULL && p->is_oop(), "expected an oop");
796 HeapWord* addr = (HeapWord*)p;
797 assert(addr >= _prevMarkBitMap->startWord() ||
798 addr < _prevMarkBitMap->endWord(), "in a region");
799
800 return _prevMarkBitMap->isMarked(addr);
801 }
802
803 inline bool do_yield_check(int worker_i = 0);
804 inline bool should_yield();
805
806 // Called to abort the marking cycle after a Full GC takes palce.
807 void abort();
808
809 // This prints the global/local fingers. It is used for debugging.
810 NOT_PRODUCT(void print_finger();)
811
812 void print_summary_info();
813
814 void print_worker_threads_on(outputStream* st) const;
815
816 // The following indicate whether a given verbose level has been
817 // set. Notice that anything above stats is conditional to
818 // _MARKING_VERBOSE_ having been set to 1
819 bool verbose_stats()
820 { return _verbose_level >= stats_verbose; }
821 bool verbose_low()
822 { return _MARKING_VERBOSE_ && _verbose_level >= low_verbose; }
823 bool verbose_medium()
824 { return _MARKING_VERBOSE_ && _verbose_level >= medium_verbose; }
825 bool verbose_high()
826 { return _MARKING_VERBOSE_ && _verbose_level >= high_verbose; }
827 };
828
829 // A class representing a marking task.
830 class CMTask : public TerminatorTerminator {
831 private:
832 enum PrivateConstants {
833 // the regular clock call is called once the scanned words reaches
834 // this limit
835 words_scanned_period = 12*1024,
836 // the regular clock call is called once the number of visited
837 // references reaches this limit
838 refs_reached_period = 384,
839 // initial value for the hash seed, used in the work stealing code
840 init_hash_seed = 17,
841 // how many entries will be transferred between global stack and
842 // local queues
843 global_stack_transfer_size = 16
844 };
845
846 int _task_id;
847 G1CollectedHeap* _g1h;
848 ConcurrentMark* _cm;
849 CMBitMap* _nextMarkBitMap;
850 // the task queue of this task
851 CMTaskQueue* _task_queue;
852 private:
853 // the task queue set---needed for stealing
854 CMTaskQueueSet* _task_queues;
855 // indicates whether the task has been claimed---this is only for
856 // debugging purposes
857 bool _claimed;
858
859 // number of calls to this task
860 int _calls;
861
862 // when the virtual timer reaches this time, the marking step should
863 // exit
864 double _time_target_ms;
865 // the start time of the current marking step
866 double _start_time_ms;
867
868 // the oop closure used for iterations over oops
869 OopClosure* _oop_closure;
870
871 // the region this task is scanning, NULL if we're not scanning any
872 HeapRegion* _curr_region;
873 // the local finger of this task, NULL if we're not scanning a region
874 HeapWord* _finger;
875 // limit of the region this task is scanning, NULL if we're not scanning one
876 HeapWord* _region_limit;
877
878 // This is used only when we scan regions popped from the region
879 // stack. It records what the last object on such a region we
880 // scanned was. It is used to ensure that, if we abort region
881 // iteration, we do not rescan the first part of the region. This
882 // should be NULL when we're not scanning a region from the region
883 // stack.
884 HeapWord* _region_finger;
885
886 // If we abort while scanning a region we record the remaining
887 // unscanned portion and check this field when marking restarts.
888 // This avoids having to push on the region stack while other
889 // marking threads may still be popping regions.
890 // If we were to push the unscanned portion directly to the
891 // region stack then we would need to using locking versions
892 // of the push and pop operations.
893 MemRegion _aborted_region;
894
895 // the number of words this task has scanned
896 size_t _words_scanned;
897 // When _words_scanned reaches this limit, the regular clock is
898 // called. Notice that this might be decreased under certain
899 // circumstances (i.e. when we believe that we did an expensive
900 // operation).
901 size_t _words_scanned_limit;
902 // the initial value of _words_scanned_limit (i.e. what it was
903 // before it was decreased).
904 size_t _real_words_scanned_limit;
905
906 // the number of references this task has visited
907 size_t _refs_reached;
908 // When _refs_reached reaches this limit, the regular clock is
909 // called. Notice this this might be decreased under certain
910 // circumstances (i.e. when we believe that we did an expensive
911 // operation).
912 size_t _refs_reached_limit;
913 // the initial value of _refs_reached_limit (i.e. what it was before
914 // it was decreased).
915 size_t _real_refs_reached_limit;
916
917 // used by the work stealing stuff
918 int _hash_seed;
919 // if this is true, then the task has aborted for some reason
920 bool _has_aborted;
921 // set when the task aborts because it has met its time quota
922 bool _has_aborted_timed_out;
923 // true when we're draining SATB buffers; this avoids the task
924 // aborting due to SATB buffers being available (as we're already
925 // dealing with them)
926 bool _draining_satb_buffers;
927
928 // number sequence of past step times
929 NumberSeq _step_times_ms;
930 // elapsed time of this task
931 double _elapsed_time_ms;
932 // termination time of this task
933 double _termination_time_ms;
934 // when this task got into the termination protocol
935 double _termination_start_time_ms;
936
937 // true when the task is during a concurrent phase, false when it is
938 // in the remark phase (so, in the latter case, we do not have to
939 // check all the things that we have to check during the concurrent
940 // phase, i.e. SATB buffer availability...)
941 bool _concurrent;
942
943 TruncatedSeq _marking_step_diffs_ms;
944
945 // LOTS of statistics related with this task
946 #if _MARKING_STATS_
947 NumberSeq _all_clock_intervals_ms;
948 double _interval_start_time_ms;
949
950 int _aborted;
951 int _aborted_overflow;
952 int _aborted_cm_aborted;
953 int _aborted_yield;
954 int _aborted_timed_out;
955 int _aborted_satb;
956 int _aborted_termination;
957
958 int _steal_attempts;
959 int _steals;
960
961 int _clock_due_to_marking;
962 int _clock_due_to_scanning;
963
964 int _local_pushes;
965 int _local_pops;
966 int _local_max_size;
967 int _objs_scanned;
968
969 int _global_pushes;
970 int _global_pops;
971 int _global_max_size;
972
973 int _global_transfers_to;
974 int _global_transfers_from;
975
976 int _region_stack_pops;
977
978 int _regions_claimed;
979 int _objs_found_on_bitmap;
980
981 int _satb_buffers_processed;
982 #endif // _MARKING_STATS_
983
984 // it updates the local fields after this task has claimed
985 // a new region to scan
986 void setup_for_region(HeapRegion* hr);
987 // it brings up-to-date the limit of the region
988 void update_region_limit();
989 // it resets the local fields after a task has finished scanning a
990 // region
991 void giveup_current_region();
992
993 // called when either the words scanned or the refs visited limit
994 // has been reached
995 void reached_limit();
996 // recalculates the words scanned and refs visited limits
997 void recalculate_limits();
998 // decreases the words scanned and refs visited limits when we reach
999 // an expensive operation
1000 void decrease_limits();
1001 // it checks whether the words scanned or refs visited reached their
1002 // respective limit and calls reached_limit() if they have
1003 void check_limits() {
1004 if (_words_scanned >= _words_scanned_limit ||
1005 _refs_reached >= _refs_reached_limit)
1006 reached_limit();
1007 }
1008 // this is supposed to be called regularly during a marking step as
1009 // it checks a bunch of conditions that might cause the marking step
1010 // to abort
1011 void regular_clock_call();
1012 bool concurrent() { return _concurrent; }
1013
1014 public:
1015 // It resets the task; it should be called right at the beginning of
1016 // a marking phase.
1017 void reset(CMBitMap* _nextMarkBitMap);
1018 // it clears all the fields that correspond to a claimed region.
1019 void clear_region_fields();
1020
1021 void set_concurrent(bool concurrent) { _concurrent = concurrent; }
1022
1023 // The main method of this class which performs a marking step
1024 // trying not to exceed the given duration. However, it might exit
1025 // prematurely, according to some conditions (i.e. SATB buffers are
1026 // available for processing).
1027 void do_marking_step(double target_ms);
1028
1029 // These two calls start and stop the timer
1030 void record_start_time() {
1031 _elapsed_time_ms = os::elapsedTime() * 1000.0;
1032 }
1033 void record_end_time() {
1034 _elapsed_time_ms = os::elapsedTime() * 1000.0 - _elapsed_time_ms;
1035 }
1036
1037 // returns the task ID
1038 int task_id() { return _task_id; }
1039
1040 // From TerminatorTerminator. It determines whether this task should
1041 // exit the termination protocol after it's entered it.
1042 virtual bool should_exit_termination();
1043
1044 HeapWord* finger() { return _finger; }
1045
1046 bool has_aborted() { return _has_aborted; }
1047 void set_has_aborted() { _has_aborted = true; }
1048 void clear_has_aborted() { _has_aborted = false; }
1049 bool claimed() { return _claimed; }
1050
1051 // Support routines for the partially scanned region that may be
1052 // recorded as a result of aborting while draining the CMRegionStack
1053 MemRegion aborted_region() { return _aborted_region; }
1054 void set_aborted_region(MemRegion mr)
1055 { _aborted_region = mr; }
1056
1057 // Clears any recorded partially scanned region
1058 void clear_aborted_region() { set_aborted_region(MemRegion()); }
1059
1060 void set_oop_closure(OopClosure* oop_closure) {
1061 _oop_closure = oop_closure;
1062 }
1063
1064 // It grays the object by marking it and, if necessary, pushing it
1065 // on the local queue
1066 void deal_with_reference(oop obj);
1067
1068 // It scans an object and visits its children.
1069 void scan_object(oop obj) {
1070 assert(_nextMarkBitMap->isMarked((HeapWord*) obj), "invariant");
1071
1072 if (_cm->verbose_high())
1073 gclog_or_tty->print_cr("[%d] we're scanning object "PTR_FORMAT,
1074 _task_id, (void*) obj);
1075
1076 size_t obj_size = obj->size();
1077 _words_scanned += obj_size;
1078
1079 obj->oop_iterate(_oop_closure);
1080 statsOnly( ++_objs_scanned );
1081 check_limits();
1082 }
1083
1084 // It pushes an object on the local queue.
1085 void push(oop obj);
1086
1087 // These two move entries to/from the global stack.
1088 void move_entries_to_global_stack();
1089 void get_entries_from_global_stack();
1090
1091 // It pops and scans objects from the local queue. If partially is
1092 // true, then it stops when the queue size is of a given limit. If
1093 // partially is false, then it stops when the queue is empty.
1094 void drain_local_queue(bool partially);
1095 // It moves entries from the global stack to the local queue and
1096 // drains the local queue. If partially is true, then it stops when
1097 // both the global stack and the local queue reach a given size. If
1098 // partially if false, it tries to empty them totally.
1099 void drain_global_stack(bool partially);
1100 // It keeps picking SATB buffers and processing them until no SATB
1101 // buffers are available.
1102 void drain_satb_buffers();
1103 // It keeps popping regions from the region stack and processing
1104 // them until the region stack is empty.
1105 void drain_region_stack(BitMapClosure* closure);
1106
1107 // moves the local finger to a new location
1108 inline void move_finger_to(HeapWord* new_finger) {
1109 assert(new_finger >= _finger && new_finger < _region_limit, "invariant");
1110 _finger = new_finger;
1111 }
1112
1113 // moves the region finger to a new location
1114 inline void move_region_finger_to(HeapWord* new_finger) {
1115 assert(new_finger < _cm->finger(), "invariant");
1116 _region_finger = new_finger;
1117 }
1118
1119 CMTask(int task_num, ConcurrentMark *cm,
1120 CMTaskQueue* task_queue, CMTaskQueueSet* task_queues);
1121
1122 // it prints statistics associated with this task
1123 void print_stats();
1124
1125 #if _MARKING_STATS_
1126 void increase_objs_found_on_bitmap() { ++_objs_found_on_bitmap; }
1127 #endif // _MARKING_STATS_
1128 };
1129
1130 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_HPP