1 /*
2 * Copyright (c) 2005, 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_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP
26 #define SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP
27
28 #include "gc_implementation/parallelScavenge/objectStartArray.hpp"
29 #include "gc_implementation/parallelScavenge/parMarkBitMap.hpp"
30 #include "gc_implementation/parallelScavenge/psCompactionManager.hpp"
31 #include "gc_implementation/shared/collectorCounters.hpp"
32 #include "gc_implementation/shared/markSweep.hpp"
33 #include "gc_implementation/shared/mutableSpace.hpp"
34 #include "memory/sharedHeap.hpp"
35 #include "oops/oop.hpp"
36
37 class ParallelScavengeHeap;
38 class PSAdaptiveSizePolicy;
39 class PSYoungGen;
40 class PSOldGen;
41 class PSPermGen;
42 class ParCompactionManager;
43 class ParallelTaskTerminator;
44 class PSParallelCompact;
45 class GCTaskManager;
46 class GCTaskQueue;
47 class PreGCValues;
48 class MoveAndUpdateClosure;
49 class RefProcTaskExecutor;
50
51 // The SplitInfo class holds the information needed to 'split' a source region
52 // so that the live data can be copied to two destination *spaces*. Normally,
53 // all the live data in a region is copied to a single destination space (e.g.,
54 // everything live in a region in eden is copied entirely into the old gen).
55 // However, when the heap is nearly full, all the live data in eden may not fit
56 // into the old gen. Copying only some of the regions from eden to old gen
57 // requires finding a region that does not contain a partial object (i.e., no
58 // live object crosses the region boundary) somewhere near the last object that
59 // does fit into the old gen. Since it's not always possible to find such a
60 // region, splitting is necessary for predictable behavior.
61 //
62 // A region is always split at the end of the partial object. This avoids
63 // additional tests when calculating the new location of a pointer, which is a
64 // very hot code path. The partial object and everything to its left will be
65 // copied to another space (call it dest_space_1). The live data to the right
66 // of the partial object will be copied either within the space itself, or to a
67 // different destination space (distinct from dest_space_1).
68 //
69 // Split points are identified during the summary phase, when region
70 // destinations are computed: data about the split, including the
71 // partial_object_size, is recorded in a SplitInfo record and the
72 // partial_object_size field in the summary data is set to zero. The zeroing is
73 // possible (and necessary) since the partial object will move to a different
74 // destination space than anything to its right, thus the partial object should
75 // not affect the locations of any objects to its right.
76 //
77 // The recorded data is used during the compaction phase, but only rarely: when
78 // the partial object on the split region will be copied across a destination
79 // region boundary. This test is made once each time a region is filled, and is
80 // a simple address comparison, so the overhead is negligible (see
81 // PSParallelCompact::first_src_addr()).
82 //
83 // Notes:
84 //
85 // Only regions with partial objects are split; a region without a partial
86 // object does not need any extra bookkeeping.
87 //
88 // At most one region is split per space, so the amount of data required is
89 // constant.
90 //
91 // A region is split only when the destination space would overflow. Once that
92 // happens, the destination space is abandoned and no other data (even from
93 // other source spaces) is targeted to that destination space. Abandoning the
94 // destination space may leave a somewhat large unused area at the end, if a
95 // large object caused the overflow.
96 //
97 // Future work:
98 //
99 // More bookkeeping would be required to continue to use the destination space.
100 // The most general solution would allow data from regions in two different
101 // source spaces to be "joined" in a single destination region. At the very
102 // least, additional code would be required in next_src_region() to detect the
103 // join and skip to an out-of-order source region. If the join region was also
104 // the last destination region to which a split region was copied (the most
105 // likely case), then additional work would be needed to get fill_region() to
106 // stop iteration and switch to a new source region at the right point. Basic
107 // idea would be to use a fake value for the top of the source space. It is
108 // doable, if a bit tricky.
109 //
110 // A simpler (but less general) solution would fill the remainder of the
111 // destination region with a dummy object and continue filling the next
112 // destination region.
113
114 class SplitInfo
115 {
116 public:
117 // Return true if this split info is valid (i.e., if a split has been
118 // recorded). The very first region cannot have a partial object and thus is
119 // never split, so 0 is the 'invalid' value.
120 bool is_valid() const { return _src_region_idx > 0; }
121
122 // Return true if this split holds data for the specified source region.
123 inline bool is_split(size_t source_region) const;
124
125 // The index of the split region, the size of the partial object on that
126 // region and the destination of the partial object.
127 size_t src_region_idx() const { return _src_region_idx; }
128 size_t partial_obj_size() const { return _partial_obj_size; }
129 HeapWord* destination() const { return _destination; }
130
131 // The destination count of the partial object referenced by this split
132 // (either 1 or 2). This must be added to the destination count of the
133 // remainder of the source region.
134 unsigned int destination_count() const { return _destination_count; }
135
136 // If a word within the partial object will be written to the first word of a
137 // destination region, this is the address of the destination region;
138 // otherwise this is NULL.
139 HeapWord* dest_region_addr() const { return _dest_region_addr; }
140
141 // If a word within the partial object will be written to the first word of a
142 // destination region, this is the address of that word within the partial
143 // object; otherwise this is NULL.
144 HeapWord* first_src_addr() const { return _first_src_addr; }
145
146 // Record the data necessary to split the region src_region_idx.
147 void record(size_t src_region_idx, size_t partial_obj_size,
148 HeapWord* destination);
149
150 void clear();
151
152 DEBUG_ONLY(void verify_clear();)
153
154 private:
155 size_t _src_region_idx;
156 size_t _partial_obj_size;
157 HeapWord* _destination;
158 unsigned int _destination_count;
159 HeapWord* _dest_region_addr;
160 HeapWord* _first_src_addr;
161 };
162
163 inline bool SplitInfo::is_split(size_t region_idx) const
164 {
165 return _src_region_idx == region_idx && is_valid();
166 }
167
168 class SpaceInfo
169 {
170 public:
171 MutableSpace* space() const { return _space; }
172
173 // Where the free space will start after the collection. Valid only after the
174 // summary phase completes.
175 HeapWord* new_top() const { return _new_top; }
176
177 // Allows new_top to be set.
178 HeapWord** new_top_addr() { return &_new_top; }
179
180 // Where the smallest allowable dense prefix ends (used only for perm gen).
181 HeapWord* min_dense_prefix() const { return _min_dense_prefix; }
182
183 // Where the dense prefix ends, or the compacted region begins.
184 HeapWord* dense_prefix() const { return _dense_prefix; }
185
186 // The start array for the (generation containing the) space, or NULL if there
187 // is no start array.
188 ObjectStartArray* start_array() const { return _start_array; }
189
190 SplitInfo& split_info() { return _split_info; }
191
192 void set_space(MutableSpace* s) { _space = s; }
193 void set_new_top(HeapWord* addr) { _new_top = addr; }
194 void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; }
195 void set_dense_prefix(HeapWord* addr) { _dense_prefix = addr; }
196 void set_start_array(ObjectStartArray* s) { _start_array = s; }
197
198 void publish_new_top() const { _space->set_top(_new_top); }
199
200 private:
201 MutableSpace* _space;
202 HeapWord* _new_top;
203 HeapWord* _min_dense_prefix;
204 HeapWord* _dense_prefix;
205 ObjectStartArray* _start_array;
206 SplitInfo _split_info;
207 };
208
209 class ParallelCompactData
210 {
211 public:
212 // Sizes are in HeapWords, unless indicated otherwise.
213 static const size_t Log2RegionSize;
214 static const size_t RegionSize;
215 static const size_t RegionSizeBytes;
216
217 // Mask for the bits in a size_t to get an offset within a region.
218 static const size_t RegionSizeOffsetMask;
219 // Mask for the bits in a pointer to get an offset within a region.
220 static const size_t RegionAddrOffsetMask;
221 // Mask for the bits in a pointer to get the address of the start of a region.
222 static const size_t RegionAddrMask;
223
224 class RegionData
225 {
226 public:
227 // Destination address of the region.
228 HeapWord* destination() const { return _destination; }
229
230 // The first region containing data destined for this region.
231 size_t source_region() const { return _source_region; }
232
233 // The object (if any) starting in this region and ending in a different
234 // region that could not be updated during the main (parallel) compaction
235 // phase. This is different from _partial_obj_addr, which is an object that
236 // extends onto a source region. However, the two uses do not overlap in
237 // time, so the same field is used to save space.
238 HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
239
240 // The starting address of the partial object extending onto the region.
241 HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
242
243 // Size of the partial object extending onto the region (words).
244 size_t partial_obj_size() const { return _partial_obj_size; }
245
246 // Size of live data that lies within this region due to objects that start
247 // in this region (words). This does not include the partial object
248 // extending onto the region (if any), or the part of an object that extends
249 // onto the next region (if any).
250 size_t live_obj_size() const { return _dc_and_los & los_mask; }
251
252 // Total live data that lies within the region (words).
253 size_t data_size() const { return partial_obj_size() + live_obj_size(); }
254
255 // The destination_count is the number of other regions to which data from
256 // this region will be copied. At the end of the summary phase, the valid
257 // values of destination_count are
258 //
259 // 0 - data from the region will be compacted completely into itself, or the
260 // region is empty. The region can be claimed and then filled.
261 // 1 - data from the region will be compacted into 1 other region; some
262 // data from the region may also be compacted into the region itself.
263 // 2 - data from the region will be copied to 2 other regions.
264 //
265 // During compaction as regions are emptied, the destination_count is
266 // decremented (atomically) and when it reaches 0, it can be claimed and
267 // then filled.
268 //
269 // A region is claimed for processing by atomically changing the
270 // destination_count to the claimed value (dc_claimed). After a region has
271 // been filled, the destination_count should be set to the completed value
272 // (dc_completed).
273 inline uint destination_count() const;
274 inline uint destination_count_raw() const;
275
276 // The location of the java heap data that corresponds to this region.
277 inline HeapWord* data_location() const;
278
279 // The highest address referenced by objects in this region.
280 inline HeapWord* highest_ref() const;
281
282 // Whether this region is available to be claimed, has been claimed, or has
283 // been completed.
284 //
285 // Minor subtlety: claimed() returns true if the region is marked
286 // completed(), which is desirable since a region must be claimed before it
287 // can be completed.
288 bool available() const { return _dc_and_los < dc_one; }
289 bool claimed() const { return _dc_and_los >= dc_claimed; }
290 bool completed() const { return _dc_and_los >= dc_completed; }
291
292 // These are not atomic.
293 void set_destination(HeapWord* addr) { _destination = addr; }
294 void set_source_region(size_t region) { _source_region = region; }
295 void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
296 void set_partial_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
297 void set_partial_obj_size(size_t words) {
298 _partial_obj_size = (region_sz_t) words;
299 }
300
301 inline void set_destination_count(uint count);
302 inline void set_live_obj_size(size_t words);
303 inline void set_data_location(HeapWord* addr);
304 inline void set_completed();
305 inline bool claim_unsafe();
306
307 // These are atomic.
308 inline void add_live_obj(size_t words);
309 inline void set_highest_ref(HeapWord* addr);
310 inline void decrement_destination_count();
311 inline bool claim();
312
313 private:
314 // The type used to represent object sizes within a region.
315 typedef uint region_sz_t;
316
317 // Constants for manipulating the _dc_and_los field, which holds both the
318 // destination count and live obj size. The live obj size lives at the
319 // least significant end so no masking is necessary when adding.
320 static const region_sz_t dc_shift; // Shift amount.
321 static const region_sz_t dc_mask; // Mask for destination count.
322 static const region_sz_t dc_one; // 1, shifted appropriately.
323 static const region_sz_t dc_claimed; // Region has been claimed.
324 static const region_sz_t dc_completed; // Region has been completed.
325 static const region_sz_t los_mask; // Mask for live obj size.
326
327 HeapWord* _destination;
328 size_t _source_region;
329 HeapWord* _partial_obj_addr;
330 region_sz_t _partial_obj_size;
331 region_sz_t volatile _dc_and_los;
332 #ifdef ASSERT
333 // These enable optimizations that are only partially implemented. Use
334 // debug builds to prevent the code fragments from breaking.
335 HeapWord* _data_location;
336 HeapWord* _highest_ref;
337 #endif // #ifdef ASSERT
338
339 #ifdef ASSERT
340 public:
341 uint _pushed; // 0 until region is pushed onto a worker's stack
342 private:
343 #endif
344 };
345
346 public:
347 ParallelCompactData();
348 bool initialize(MemRegion covered_region);
349
350 size_t region_count() const { return _region_count; }
351
352 // Convert region indices to/from RegionData pointers.
353 inline RegionData* region(size_t region_idx) const;
354 inline size_t region(const RegionData* const region_ptr) const;
355
356 // Returns true if the given address is contained within the region
357 bool region_contains(size_t region_index, HeapWord* addr);
358
359 void add_obj(HeapWord* addr, size_t len);
360 void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
361
362 // Fill in the regions covering [beg, end) so that no data moves; i.e., the
363 // destination of region n is simply the start of region n. The argument beg
364 // must be region-aligned; end need not be.
365 void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
366
367 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
368 HeapWord* destination, HeapWord* target_end,
369 HeapWord** target_next);
370 bool summarize(SplitInfo& split_info,
371 HeapWord* source_beg, HeapWord* source_end,
372 HeapWord** source_next,
373 HeapWord* target_beg, HeapWord* target_end,
374 HeapWord** target_next);
375
376 void clear();
377 void clear_range(size_t beg_region, size_t end_region);
378 void clear_range(HeapWord* beg, HeapWord* end) {
379 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
380 }
381
382 // Return the number of words between addr and the start of the region
383 // containing addr.
384 inline size_t region_offset(const HeapWord* addr) const;
385
386 // Convert addresses to/from a region index or region pointer.
387 inline size_t addr_to_region_idx(const HeapWord* addr) const;
388 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
389 inline HeapWord* region_to_addr(size_t region) const;
390 inline HeapWord* region_to_addr(size_t region, size_t offset) const;
391 inline HeapWord* region_to_addr(const RegionData* region) const;
392
393 inline HeapWord* region_align_down(HeapWord* addr) const;
394 inline HeapWord* region_align_up(HeapWord* addr) const;
395 inline bool is_region_aligned(HeapWord* addr) const;
396
397 // Return the address one past the end of the partial object.
398 HeapWord* partial_obj_end(size_t region_idx) const;
399
400 // Return the new location of the object p after the
401 // the compaction.
402 HeapWord* calc_new_pointer(HeapWord* addr);
403
404 HeapWord* calc_new_pointer(oop p) {
405 return calc_new_pointer((HeapWord*) p);
406 }
407
408 // Return the updated address for the given klass
409 klassOop calc_new_klass(klassOop);
410
411 #ifdef ASSERT
412 void verify_clear(const PSVirtualSpace* vspace);
413 void verify_clear();
414 #endif // #ifdef ASSERT
415
416 private:
417 bool initialize_region_data(size_t region_size);
418 PSVirtualSpace* create_vspace(size_t count, size_t element_size);
419
420 private:
421 HeapWord* _region_start;
422 #ifdef ASSERT
423 HeapWord* _region_end;
424 #endif // #ifdef ASSERT
425
426 PSVirtualSpace* _region_vspace;
427 RegionData* _region_data;
428 size_t _region_count;
429 };
430
431 inline uint
432 ParallelCompactData::RegionData::destination_count_raw() const
433 {
434 return _dc_and_los & dc_mask;
435 }
436
437 inline uint
438 ParallelCompactData::RegionData::destination_count() const
439 {
440 return destination_count_raw() >> dc_shift;
441 }
442
443 inline void
444 ParallelCompactData::RegionData::set_destination_count(uint count)
445 {
446 assert(count <= (dc_completed >> dc_shift), "count too large");
447 const region_sz_t live_sz = (region_sz_t) live_obj_size();
448 _dc_and_los = (count << dc_shift) | live_sz;
449 }
450
451 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
452 {
453 assert(words <= los_mask, "would overflow");
454 _dc_and_los = destination_count_raw() | (region_sz_t)words;
455 }
456
457 inline void ParallelCompactData::RegionData::decrement_destination_count()
458 {
459 assert(_dc_and_los < dc_claimed, "already claimed");
460 assert(_dc_and_los >= dc_one, "count would go negative");
461 Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los);
462 }
463
464 inline HeapWord* ParallelCompactData::RegionData::data_location() const
465 {
466 DEBUG_ONLY(return _data_location;)
467 NOT_DEBUG(return NULL;)
468 }
469
470 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
471 {
472 DEBUG_ONLY(return _highest_ref;)
473 NOT_DEBUG(return NULL;)
474 }
475
476 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
477 {
478 DEBUG_ONLY(_data_location = addr;)
479 }
480
481 inline void ParallelCompactData::RegionData::set_completed()
482 {
483 assert(claimed(), "must be claimed first");
484 _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
485 }
486
487 // MT-unsafe claiming of a region. Should only be used during single threaded
488 // execution.
489 inline bool ParallelCompactData::RegionData::claim_unsafe()
490 {
491 if (available()) {
492 _dc_and_los |= dc_claimed;
493 return true;
494 }
495 return false;
496 }
497
498 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
499 {
500 assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
501 Atomic::add((int) words, (volatile int*) &_dc_and_los);
502 }
503
504 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
505 {
506 #ifdef ASSERT
507 HeapWord* tmp = _highest_ref;
508 while (addr > tmp) {
509 tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp);
510 }
511 #endif // #ifdef ASSERT
512 }
513
514 inline bool ParallelCompactData::RegionData::claim()
515 {
516 const int los = (int) live_obj_size();
517 const int old = Atomic::cmpxchg(dc_claimed | los,
518 (volatile int*) &_dc_and_los, los);
519 return old == los;
520 }
521
522 inline ParallelCompactData::RegionData*
523 ParallelCompactData::region(size_t region_idx) const
524 {
525 assert(region_idx <= region_count(), "bad arg");
526 return _region_data + region_idx;
527 }
528
529 inline size_t
530 ParallelCompactData::region(const RegionData* const region_ptr) const
531 {
532 assert(region_ptr >= _region_data, "bad arg");
533 assert(region_ptr <= _region_data + region_count(), "bad arg");
534 return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
535 }
536
537 inline size_t
538 ParallelCompactData::region_offset(const HeapWord* addr) const
539 {
540 assert(addr >= _region_start, "bad addr");
541 assert(addr <= _region_end, "bad addr");
542 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
543 }
544
545 inline size_t
546 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
547 {
548 assert(addr >= _region_start, "bad addr");
549 assert(addr <= _region_end, "bad addr");
550 return pointer_delta(addr, _region_start) >> Log2RegionSize;
551 }
552
553 inline ParallelCompactData::RegionData*
554 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
555 {
556 return region(addr_to_region_idx(addr));
557 }
558
559 inline HeapWord*
560 ParallelCompactData::region_to_addr(size_t region) const
561 {
562 assert(region <= _region_count, "region out of range");
563 return _region_start + (region << Log2RegionSize);
564 }
565
566 inline HeapWord*
567 ParallelCompactData::region_to_addr(const RegionData* region) const
568 {
569 return region_to_addr(pointer_delta(region, _region_data,
570 sizeof(RegionData)));
571 }
572
573 inline HeapWord*
574 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
575 {
576 assert(region <= _region_count, "region out of range");
577 assert(offset < RegionSize, "offset too big"); // This may be too strict.
578 return region_to_addr(region) + offset;
579 }
580
581 inline HeapWord*
582 ParallelCompactData::region_align_down(HeapWord* addr) const
583 {
584 assert(addr >= _region_start, "bad addr");
585 assert(addr < _region_end + RegionSize, "bad addr");
586 return (HeapWord*)(size_t(addr) & RegionAddrMask);
587 }
588
589 inline HeapWord*
590 ParallelCompactData::region_align_up(HeapWord* addr) const
591 {
592 assert(addr >= _region_start, "bad addr");
593 assert(addr <= _region_end, "bad addr");
594 return region_align_down(addr + RegionSizeOffsetMask);
595 }
596
597 inline bool
598 ParallelCompactData::is_region_aligned(HeapWord* addr) const
599 {
600 return region_offset(addr) == 0;
601 }
602
603 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
604 // do_addr() method.
605 //
606 // The closure is initialized with the number of heap words to process
607 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr()
608 // methods in subclasses should update the total as words are processed. Since
609 // only one subclass actually uses this mechanism to terminate iteration, the
610 // default initial value is > 0. The implementation is here and not in the
611 // single subclass that uses it to avoid making is_full() virtual, and thus
612 // adding a virtual call per live object.
613
614 class ParMarkBitMapClosure: public StackObj {
615 public:
616 typedef ParMarkBitMap::idx_t idx_t;
617 typedef ParMarkBitMap::IterationStatus IterationStatus;
618
619 public:
620 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
621 size_t words = max_uintx);
622
623 inline ParCompactionManager* compaction_manager() const;
624 inline ParMarkBitMap* bitmap() const;
625 inline size_t words_remaining() const;
626 inline bool is_full() const;
627 inline HeapWord* source() const;
628
629 inline void set_source(HeapWord* addr);
630
631 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
632
633 protected:
634 inline void decrement_words_remaining(size_t words);
635
636 private:
637 ParMarkBitMap* const _bitmap;
638 ParCompactionManager* const _compaction_manager;
639 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger.
640 size_t _words_remaining; // Words left to copy.
641
642 protected:
643 HeapWord* _source; // Next addr that would be read.
644 };
645
646 inline
647 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
648 ParCompactionManager* cm,
649 size_t words):
650 _bitmap(bitmap), _compaction_manager(cm)
651 #ifdef ASSERT
652 , _initial_words_remaining(words)
653 #endif
654 {
655 _words_remaining = words;
656 _source = NULL;
657 }
658
659 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
660 return _compaction_manager;
661 }
662
663 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
664 return _bitmap;
665 }
666
667 inline size_t ParMarkBitMapClosure::words_remaining() const {
668 return _words_remaining;
669 }
670
671 inline bool ParMarkBitMapClosure::is_full() const {
672 return words_remaining() == 0;
673 }
674
675 inline HeapWord* ParMarkBitMapClosure::source() const {
676 return _source;
677 }
678
679 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
680 _source = addr;
681 }
682
683 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
684 assert(_words_remaining >= words, "processed too many words");
685 _words_remaining -= words;
686 }
687
688 // The UseParallelOldGC collector is a stop-the-world garbage collector that
689 // does parts of the collection using parallel threads. The collection includes
690 // the tenured generation and the young generation. The permanent generation is
691 // collected at the same time as the other two generations but the permanent
692 // generation is collect by a single GC thread. The permanent generation is
693 // collected serially because of the requirement that during the processing of a
694 // klass AAA, any objects reference by AAA must already have been processed.
695 // This requirement is enforced by a left (lower address) to right (higher
696 // address) sliding compaction.
697 //
698 // There are four phases of the collection.
699 //
700 // - marking phase
701 // - summary phase
702 // - compacting phase
703 // - clean up phase
704 //
705 // Roughly speaking these phases correspond, respectively, to
706 // - mark all the live objects
707 // - calculate the destination of each object at the end of the collection
708 // - move the objects to their destination
709 // - update some references and reinitialize some variables
710 //
711 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The
712 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
713 // complete marking of the heap. The summary phase is implemented in
714 // PSParallelCompact::summary_phase(). The move and update phase is implemented
715 // in PSParallelCompact::compact().
716 //
717 // A space that is being collected is divided into regions and with each region
718 // is associated an object of type ParallelCompactData. Each region is of a
719 // fixed size and typically will contain more than 1 object and may have parts
720 // of objects at the front and back of the region.
721 //
722 // region -----+---------------------+----------
723 // objects covered [ AAA )[ BBB )[ CCC )[ DDD )
724 //
725 // The marking phase does a complete marking of all live objects in the heap.
726 // The marking also compiles the size of the data for all live objects covered
727 // by the region. This size includes the part of any live object spanning onto
728 // the region (part of AAA if it is live) from the front, all live objects
729 // contained in the region (BBB and/or CCC if they are live), and the part of
730 // any live objects covered by the region that extends off the region (part of
731 // DDD if it is live). The marking phase uses multiple GC threads and marking
732 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is
733 // done atomically as is the accumulation of the size of the live objects
734 // covered by a region.
735 //
736 // The summary phase calculates the total live data to the left of each region
737 // XXX. Based on that total and the bottom of the space, it can calculate the
738 // starting location of the live data in XXX. The summary phase calculates for
739 // each region XXX quantites such as
740 //
741 // - the amount of live data at the beginning of a region from an object
742 // entering the region.
743 // - the location of the first live data on the region
744 // - a count of the number of regions receiving live data from XXX.
745 //
746 // See ParallelCompactData for precise details. The summary phase also
747 // calculates the dense prefix for the compaction. The dense prefix is a
748 // portion at the beginning of the space that is not moved. The objects in the
749 // dense prefix do need to have their object references updated. See method
750 // summarize_dense_prefix().
751 //
752 // The summary phase is done using 1 GC thread.
753 //
754 // The compaction phase moves objects to their new location and updates all
755 // references in the object.
756 //
757 // A current exception is that objects that cross a region boundary are moved
758 // but do not have their references updated. References are not updated because
759 // it cannot easily be determined if the klass pointer KKK for the object AAA
760 // has been updated. KKK likely resides in a region to the left of the region
761 // containing AAA. These AAA's have there references updated at the end in a
762 // clean up phase. See the method PSParallelCompact::update_deferred_objects().
763 // An alternate strategy is being investigated for this deferral of updating.
764 //
765 // Compaction is done on a region basis. A region that is ready to be filled is
766 // put on a ready list and GC threads take region off the list and fill them. A
767 // region is ready to be filled if it empty of live objects. Such a region may
768 // have been initially empty (only contained dead objects) or may have had all
769 // its live objects copied out already. A region that compacts into itself is
770 // also ready for filling. The ready list is initially filled with empty
771 // regions and regions compacting into themselves. There is always at least 1
772 // region that can be put on the ready list. The regions are atomically added
773 // and removed from the ready list.
774
775 class PSParallelCompact : AllStatic {
776 public:
777 // Convenient access to type names.
778 typedef ParMarkBitMap::idx_t idx_t;
779 typedef ParallelCompactData::RegionData RegionData;
780
781 typedef enum {
782 perm_space_id, old_space_id, eden_space_id,
783 from_space_id, to_space_id, last_space_id
784 } SpaceId;
785
786 public:
787 // Inline closure decls
788 //
789 class IsAliveClosure: public BoolObjectClosure {
790 public:
791 virtual void do_object(oop p);
792 virtual bool do_object_b(oop p);
793 };
794
795 class KeepAliveClosure: public OopClosure {
796 private:
797 ParCompactionManager* _compaction_manager;
798 protected:
799 template <class T> inline void do_oop_work(T* p);
800 public:
801 KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
802 virtual void do_oop(oop* p);
803 virtual void do_oop(narrowOop* p);
804 };
805
806 // Current unused
807 class FollowRootClosure: public OopsInGenClosure {
808 private:
809 ParCompactionManager* _compaction_manager;
810 public:
811 FollowRootClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
812 virtual void do_oop(oop* p);
813 virtual void do_oop(narrowOop* p);
814 };
815
816 class FollowStackClosure: public VoidClosure {
817 private:
818 ParCompactionManager* _compaction_manager;
819 public:
820 FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
821 virtual void do_void();
822 };
823
824 class AdjustPointerClosure: public OopsInGenClosure {
825 private:
826 bool _is_root;
827 public:
828 AdjustPointerClosure(bool is_root) : _is_root(is_root) { }
829 virtual void do_oop(oop* p);
830 virtual void do_oop(narrowOop* p);
831 // do not walk from thread stacks to the code cache on this phase
832 virtual void do_code_blob(CodeBlob* cb) const { }
833 };
834
835 // Closure for verifying update of pointers. Does not
836 // have any side effects.
837 class VerifyUpdateClosure: public ParMarkBitMapClosure {
838 const MutableSpace* _space; // Is this ever used?
839
840 public:
841 VerifyUpdateClosure(ParCompactionManager* cm, const MutableSpace* sp) :
842 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm), _space(sp)
843 { }
844
845 virtual IterationStatus do_addr(HeapWord* addr, size_t words);
846
847 const MutableSpace* space() { return _space; }
848 };
849
850 // Closure for updating objects altered for debug checking
851 class ResetObjectsClosure: public ParMarkBitMapClosure {
852 public:
853 ResetObjectsClosure(ParCompactionManager* cm):
854 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm)
855 { }
856
857 virtual IterationStatus do_addr(HeapWord* addr, size_t words);
858 };
859
860 friend class KeepAliveClosure;
861 friend class FollowStackClosure;
862 friend class AdjustPointerClosure;
863 friend class FollowRootClosure;
864 friend class instanceKlassKlass;
865 friend class RefProcTaskProxy;
866
867 private:
868 static elapsedTimer _accumulated_time;
869 static unsigned int _total_invocations;
870 static unsigned int _maximum_compaction_gc_num;
871 static jlong _time_of_last_gc; // ms
872 static CollectorCounters* _counters;
873 static ParMarkBitMap _mark_bitmap;
874 static ParallelCompactData _summary_data;
875 static IsAliveClosure _is_alive_closure;
876 static SpaceInfo _space_info[last_space_id];
877 static bool _print_phases;
878 static AdjustPointerClosure _adjust_root_pointer_closure;
879 static AdjustPointerClosure _adjust_pointer_closure;
880
881 // Reference processing (used in ...follow_contents)
882 static ReferenceProcessor* _ref_processor;
883
884 // Updated location of intArrayKlassObj.
885 static klassOop _updated_int_array_klass_obj;
886
887 // Values computed at initialization and used by dead_wood_limiter().
888 static double _dwl_mean;
889 static double _dwl_std_dev;
890 static double _dwl_first_term;
891 static double _dwl_adjustment;
892 #ifdef ASSERT
893 static bool _dwl_initialized;
894 #endif // #ifdef ASSERT
895
896 private:
897 // Closure accessors
898 static OopClosure* adjust_pointer_closure() { return (OopClosure*)&_adjust_pointer_closure; }
899 static OopClosure* adjust_root_pointer_closure() { return (OopClosure*)&_adjust_root_pointer_closure; }
900 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; }
901
902 static void initialize_space_info();
903
904 // Return true if details about individual phases should be printed.
905 static inline bool print_phases();
906
907 // Clear the marking bitmap and summary data that cover the specified space.
908 static void clear_data_covering_space(SpaceId id);
909
910 static void pre_compact(PreGCValues* pre_gc_values);
911 static void post_compact();
912
913 // Mark live objects
914 static void marking_phase(ParCompactionManager* cm,
915 bool maximum_heap_compaction);
916 static void follow_weak_klass_links();
917 static void follow_mdo_weak_refs();
918
919 template <class T> static inline void adjust_pointer(T* p, bool is_root);
920 static void adjust_root_pointer(oop* p) { adjust_pointer(p, true); }
921
922 template <class T>
923 static inline void follow_root(ParCompactionManager* cm, T* p);
924
925 // Compute the dense prefix for the designated space. This is an experimental
926 // implementation currently not used in production.
927 static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
928 bool maximum_compaction);
929
930 // Methods used to compute the dense prefix.
931
932 // Compute the value of the normal distribution at x = density. The mean and
933 // standard deviation are values saved by initialize_dead_wood_limiter().
934 static inline double normal_distribution(double density);
935
936 // Initialize the static vars used by dead_wood_limiter().
937 static void initialize_dead_wood_limiter();
938
939 // Return the percentage of space that can be treated as "dead wood" (i.e.,
940 // not reclaimed).
941 static double dead_wood_limiter(double density, size_t min_percent);
942
943 // Find the first (left-most) region in the range [beg, end) that has at least
944 // dead_words of dead space to the left. The argument beg must be the first
945 // region in the space that is not completely live.
946 static RegionData* dead_wood_limit_region(const RegionData* beg,
947 const RegionData* end,
948 size_t dead_words);
949
950 // Return a pointer to the first region in the range [beg, end) that is not
951 // completely full.
952 static RegionData* first_dead_space_region(const RegionData* beg,
953 const RegionData* end);
954
955 // Return a value indicating the benefit or 'yield' if the compacted region
956 // were to start (or equivalently if the dense prefix were to end) at the
957 // candidate region. Higher values are better.
958 //
959 // The value is based on the amount of space reclaimed vs. the costs of (a)
960 // updating references in the dense prefix plus (b) copying objects and
961 // updating references in the compacted region.
962 static inline double reclaimed_ratio(const RegionData* const candidate,
963 HeapWord* const bottom,
964 HeapWord* const top,
965 HeapWord* const new_top);
966
967 // Compute the dense prefix for the designated space.
968 static HeapWord* compute_dense_prefix(const SpaceId id,
969 bool maximum_compaction);
970
971 // Return true if dead space crosses onto the specified Region; bit must be
972 // the bit index corresponding to the first word of the Region.
973 static inline bool dead_space_crosses_boundary(const RegionData* region,
974 idx_t bit);
975
976 // Summary phase utility routine to fill dead space (if any) at the dense
977 // prefix boundary. Should only be called if the the dense prefix is
978 // non-empty.
979 static void fill_dense_prefix_end(SpaceId id);
980
981 // Clear the summary data source_region field for the specified addresses.
982 static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr);
983
984 #ifndef PRODUCT
985 // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot).
986
987 // Fill the region [start, start + words) with live object(s). Only usable
988 // for the old and permanent generations.
989 static void fill_with_live_objects(SpaceId id, HeapWord* const start,
990 size_t words);
991 // Include the new objects in the summary data.
992 static void summarize_new_objects(SpaceId id, HeapWord* start);
993
994 // Add live objects to a survivor space since it's rare that both survivors
995 // are non-empty.
996 static void provoke_split_fill_survivor(SpaceId id);
997
998 // Add live objects and/or choose the dense prefix to provoke splitting.
999 static void provoke_split(bool & maximum_compaction);
1000 #endif
1001
1002 static void summarize_spaces_quick();
1003 static void summarize_space(SpaceId id, bool maximum_compaction);
1004 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
1005
1006 // Adjust addresses in roots. Does not adjust addresses in heap.
1007 static void adjust_roots();
1008
1009 // Serial code executed in preparation for the compaction phase.
1010 static void compact_prologue();
1011
1012 // Move objects to new locations.
1013 static void compact_perm(ParCompactionManager* cm);
1014 static void compact();
1015
1016 // Add available regions to the stack and draining tasks to the task queue.
1017 static void enqueue_region_draining_tasks(GCTaskQueue* q,
1018 uint parallel_gc_threads);
1019
1020 // Add dense prefix update tasks to the task queue.
1021 static void enqueue_dense_prefix_tasks(GCTaskQueue* q,
1022 uint parallel_gc_threads);
1023
1024 // Add region stealing tasks to the task queue.
1025 static void enqueue_region_stealing_tasks(
1026 GCTaskQueue* q,
1027 ParallelTaskTerminator* terminator_ptr,
1028 uint parallel_gc_threads);
1029
1030 // For debugging only - compacts the old gen serially
1031 static void compact_serial(ParCompactionManager* cm);
1032
1033 // If objects are left in eden after a collection, try to move the boundary
1034 // and absorb them into the old gen. Returns true if eden was emptied.
1035 static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
1036 PSYoungGen* young_gen,
1037 PSOldGen* old_gen);
1038
1039 // Reset time since last full gc
1040 static void reset_millis_since_last_gc();
1041
1042 protected:
1043 #ifdef VALIDATE_MARK_SWEEP
1044 static GrowableArray<void*>* _root_refs_stack;
1045 static GrowableArray<oop> * _live_oops;
1046 static GrowableArray<oop> * _live_oops_moved_to;
1047 static GrowableArray<size_t>* _live_oops_size;
1048 static size_t _live_oops_index;
1049 static size_t _live_oops_index_at_perm;
1050 static GrowableArray<void*>* _other_refs_stack;
1051 static GrowableArray<void*>* _adjusted_pointers;
1052 static bool _pointer_tracking;
1053 static bool _root_tracking;
1054
1055 // The following arrays are saved since the time of the last GC and
1056 // assist in tracking down problems where someone has done an errant
1057 // store into the heap, usually to an oop that wasn't properly
1058 // handleized across a GC. If we crash or otherwise fail before the
1059 // next GC, we can query these arrays to find out the object we had
1060 // intended to do the store to (assuming it is still alive) and the
1061 // offset within that object. Covered under RecordMarkSweepCompaction.
1062 static GrowableArray<HeapWord*> * _cur_gc_live_oops;
1063 static GrowableArray<HeapWord*> * _cur_gc_live_oops_moved_to;
1064 static GrowableArray<size_t>* _cur_gc_live_oops_size;
1065 static GrowableArray<HeapWord*> * _last_gc_live_oops;
1066 static GrowableArray<HeapWord*> * _last_gc_live_oops_moved_to;
1067 static GrowableArray<size_t>* _last_gc_live_oops_size;
1068 #endif
1069
1070 public:
1071 class MarkAndPushClosure: public OopClosure {
1072 private:
1073 ParCompactionManager* _compaction_manager;
1074 public:
1075 MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
1076 virtual void do_oop(oop* p);
1077 virtual void do_oop(narrowOop* p);
1078 };
1079
1080 PSParallelCompact();
1081
1082 // Convenient accessor for Universe::heap().
1083 static ParallelScavengeHeap* gc_heap() {
1084 return (ParallelScavengeHeap*)Universe::heap();
1085 }
1086
1087 static void invoke(bool maximum_heap_compaction);
1088 static void invoke_no_policy(bool maximum_heap_compaction);
1089
1090 static void post_initialize();
1091 // Perform initialization for PSParallelCompact that requires
1092 // allocations. This should be called during the VM initialization
1093 // at a pointer where it would be appropriate to return a JNI_ENOMEM
1094 // in the event of a failure.
1095 static bool initialize();
1096
1097 // Public accessors
1098 static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1099 static unsigned int total_invocations() { return _total_invocations; }
1100 static CollectorCounters* counters() { return _counters; }
1101
1102 // Used to add tasks
1103 static GCTaskManager* const gc_task_manager();
1104 static klassOop updated_int_array_klass_obj() {
1105 return _updated_int_array_klass_obj;
1106 }
1107
1108 // Marking support
1109 static inline bool mark_obj(oop obj);
1110 // Check mark and maybe push on marking stack
1111 template <class T> static inline void mark_and_push(ParCompactionManager* cm,
1112 T* p);
1113
1114 // Compaction support.
1115 // Return true if p is in the range [beg_addr, end_addr).
1116 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1117 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1118
1119 // Convenience wrappers for per-space data kept in _space_info.
1120 static inline MutableSpace* space(SpaceId space_id);
1121 static inline HeapWord* new_top(SpaceId space_id);
1122 static inline HeapWord* dense_prefix(SpaceId space_id);
1123 static inline ObjectStartArray* start_array(SpaceId space_id);
1124
1125 // Return true if the klass should be updated.
1126 static inline bool should_update_klass(klassOop k);
1127
1128 // Move and update the live objects in the specified space.
1129 static void move_and_update(ParCompactionManager* cm, SpaceId space_id);
1130
1131 // Process the end of the given region range in the dense prefix.
1132 // This includes saving any object not updated.
1133 static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1134 size_t region_start_index,
1135 size_t region_end_index,
1136 idx_t exiting_object_offset,
1137 idx_t region_offset_start,
1138 idx_t region_offset_end);
1139
1140 // Update a region in the dense prefix. For each live object
1141 // in the region, update it's interior references. For each
1142 // dead object, fill it with deadwood. Dead space at the end
1143 // of a region range will be filled to the start of the next
1144 // live object regardless of the region_index_end. None of the
1145 // objects in the dense prefix move and dead space is dead
1146 // (holds only dead objects that don't need any processing), so
1147 // dead space can be filled in any order.
1148 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1149 SpaceId space_id,
1150 size_t region_index_start,
1151 size_t region_index_end);
1152
1153 // Return the address of the count + 1st live word in the range [beg, end).
1154 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1155
1156 // Return the address of the word to be copied to dest_addr, which must be
1157 // aligned to a region boundary.
1158 static HeapWord* first_src_addr(HeapWord* const dest_addr,
1159 SpaceId src_space_id,
1160 size_t src_region_idx);
1161
1162 // Determine the next source region, set closure.source() to the start of the
1163 // new region return the region index. Parameter end_addr is the address one
1164 // beyond the end of source range just processed. If necessary, switch to a
1165 // new source space and set src_space_id (in-out parameter) and src_space_top
1166 // (out parameter) accordingly.
1167 static size_t next_src_region(MoveAndUpdateClosure& closure,
1168 SpaceId& src_space_id,
1169 HeapWord*& src_space_top,
1170 HeapWord* end_addr);
1171
1172 // Decrement the destination count for each non-empty source region in the
1173 // range [beg_region, region(region_align_up(end_addr))). If the destination
1174 // count for a region goes to 0 and it needs to be filled, enqueue it.
1175 static void decrement_destination_counts(ParCompactionManager* cm,
1176 SpaceId src_space_id,
1177 size_t beg_region,
1178 HeapWord* end_addr);
1179
1180 // Fill a region, copying objects from one or more source regions.
1181 static void fill_region(ParCompactionManager* cm, size_t region_idx);
1182 static void fill_and_update_region(ParCompactionManager* cm, size_t region) {
1183 fill_region(cm, region);
1184 }
1185
1186 // Update the deferred objects in the space.
1187 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1188
1189 // Mark pointer and follow contents.
1190 template <class T>
1191 static inline void mark_and_follow(ParCompactionManager* cm, T* p);
1192
1193 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1194 static ParallelCompactData& summary_data() { return _summary_data; }
1195
1196 static inline void adjust_pointer(oop* p) { adjust_pointer(p, false); }
1197 static inline void adjust_pointer(narrowOop* p) { adjust_pointer(p, false); }
1198
1199 template <class T>
1200 static inline void adjust_pointer(T* p,
1201 HeapWord* beg_addr,
1202 HeapWord* end_addr);
1203
1204 // Reference Processing
1205 static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1206
1207 // Return the SpaceId for the given address.
1208 static SpaceId space_id(HeapWord* addr);
1209
1210 // Time since last full gc (in milliseconds).
1211 static jlong millis_since_last_gc();
1212
1213 #ifdef VALIDATE_MARK_SWEEP
1214 static void track_adjusted_pointer(void* p, bool isroot);
1215 static void check_adjust_pointer(void* p);
1216 static void track_interior_pointers(oop obj);
1217 static void check_interior_pointers();
1218
1219 static void reset_live_oop_tracking(bool at_perm);
1220 static void register_live_oop(oop p, size_t size);
1221 static void validate_live_oop(oop p, size_t size);
1222 static void live_oop_moved_to(HeapWord* q, size_t size, HeapWord* compaction_top);
1223 static void compaction_complete();
1224
1225 // Querying operation of RecordMarkSweepCompaction results.
1226 // Finds and prints the current base oop and offset for a word
1227 // within an oop that was live during the last GC. Helpful for
1228 // tracking down heap stomps.
1229 static void print_new_location_of_heap_address(HeapWord* q);
1230 #endif // #ifdef VALIDATE_MARK_SWEEP
1231
1232 // Call backs for class unloading
1233 // Update subklass/sibling/implementor links at end of marking.
1234 static void revisit_weak_klass_link(ParCompactionManager* cm, Klass* k);
1235
1236 // Clear unmarked oops in MDOs at the end of marking.
1237 static void revisit_mdo(ParCompactionManager* cm, DataLayout* p);
1238
1239 #ifndef PRODUCT
1240 // Debugging support.
1241 static const char* space_names[last_space_id];
1242 static void print_region_ranges();
1243 static void print_dense_prefix_stats(const char* const algorithm,
1244 const SpaceId id,
1245 const bool maximum_compaction,
1246 HeapWord* const addr);
1247 static void summary_phase_msg(SpaceId dst_space_id,
1248 HeapWord* dst_beg, HeapWord* dst_end,
1249 SpaceId src_space_id,
1250 HeapWord* src_beg, HeapWord* src_end);
1251 #endif // #ifndef PRODUCT
1252
1253 #ifdef ASSERT
1254 // Sanity check the new location of a word in the heap.
1255 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1256 // Verify that all the regions have been emptied.
1257 static void verify_complete(SpaceId space_id);
1258 #endif // #ifdef ASSERT
1259 };
1260
1261 inline bool PSParallelCompact::mark_obj(oop obj) {
1262 const int obj_size = obj->size();
1263 if (mark_bitmap()->mark_obj(obj, obj_size)) {
1264 _summary_data.add_obj(obj, obj_size);
1265 return true;
1266 } else {
1267 return false;
1268 }
1269 }
1270
1271 template <class T>
1272 inline void PSParallelCompact::follow_root(ParCompactionManager* cm, T* p) {
1273 assert(!Universe::heap()->is_in_reserved(p),
1274 "roots shouldn't be things within the heap");
1275 #ifdef VALIDATE_MARK_SWEEP
1276 if (ValidateMarkSweep) {
1277 guarantee(!_root_refs_stack->contains(p), "should only be in here once");
1278 _root_refs_stack->push(p);
1279 }
1280 #endif
1281 T heap_oop = oopDesc::load_heap_oop(p);
1282 if (!oopDesc::is_null(heap_oop)) {
1283 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1284 if (mark_bitmap()->is_unmarked(obj)) {
1285 if (mark_obj(obj)) {
1286 obj->follow_contents(cm);
1287 }
1288 }
1289 }
1290 cm->follow_marking_stacks();
1291 }
1292
1293 template <class T>
1294 inline void PSParallelCompact::mark_and_follow(ParCompactionManager* cm,
1295 T* p) {
1296 T heap_oop = oopDesc::load_heap_oop(p);
1297 if (!oopDesc::is_null(heap_oop)) {
1298 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1299 if (mark_bitmap()->is_unmarked(obj)) {
1300 if (mark_obj(obj)) {
1301 obj->follow_contents(cm);
1302 }
1303 }
1304 }
1305 }
1306
1307 template <class T>
1308 inline void PSParallelCompact::mark_and_push(ParCompactionManager* cm, T* p) {
1309 T heap_oop = oopDesc::load_heap_oop(p);
1310 if (!oopDesc::is_null(heap_oop)) {
1311 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1312 if (mark_bitmap()->is_unmarked(obj) && mark_obj(obj)) {
1313 cm->push(obj);
1314 }
1315 }
1316 }
1317
1318 template <class T>
1319 inline void PSParallelCompact::adjust_pointer(T* p, bool isroot) {
1320 T heap_oop = oopDesc::load_heap_oop(p);
1321 if (!oopDesc::is_null(heap_oop)) {
1322 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1323 oop new_obj = (oop)summary_data().calc_new_pointer(obj);
1324 assert(new_obj != NULL || // is forwarding ptr?
1325 obj->is_shared(), // never forwarded?
1326 "should be forwarded");
1327 // Just always do the update unconditionally?
1328 if (new_obj != NULL) {
1329 assert(Universe::heap()->is_in_reserved(new_obj),
1330 "should be in object space");
1331 oopDesc::encode_store_heap_oop_not_null(p, new_obj);
1332 }
1333 }
1334 VALIDATE_MARK_SWEEP_ONLY(track_adjusted_pointer(p, isroot));
1335 }
1336
1337 template <class T>
1338 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) {
1339 #ifdef VALIDATE_MARK_SWEEP
1340 if (ValidateMarkSweep) {
1341 if (!Universe::heap()->is_in_reserved(p)) {
1342 _root_refs_stack->push(p);
1343 } else {
1344 _other_refs_stack->push(p);
1345 }
1346 }
1347 #endif
1348 mark_and_push(_compaction_manager, p);
1349 }
1350
1351 inline bool PSParallelCompact::print_phases() {
1352 return _print_phases;
1353 }
1354
1355 inline double PSParallelCompact::normal_distribution(double density) {
1356 assert(_dwl_initialized, "uninitialized");
1357 const double squared_term = (density - _dwl_mean) / _dwl_std_dev;
1358 return _dwl_first_term * exp(-0.5 * squared_term * squared_term);
1359 }
1360
1361 inline bool
1362 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region,
1363 idx_t bit)
1364 {
1365 assert(bit > 0, "cannot call this for the first bit/region");
1366 assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit),
1367 "sanity check");
1368
1369 // Dead space crosses the boundary if (1) a partial object does not extend
1370 // onto the region, (2) an object does not start at the beginning of the
1371 // region, and (3) an object does not end at the end of the prior region.
1372 return region->partial_obj_size() == 0 &&
1373 !_mark_bitmap.is_obj_beg(bit) &&
1374 !_mark_bitmap.is_obj_end(bit - 1);
1375 }
1376
1377 inline bool
1378 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) {
1379 return p >= beg_addr && p < end_addr;
1380 }
1381
1382 inline bool
1383 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) {
1384 return is_in((HeapWord*)p, beg_addr, end_addr);
1385 }
1386
1387 inline MutableSpace* PSParallelCompact::space(SpaceId id) {
1388 assert(id < last_space_id, "id out of range");
1389 return _space_info[id].space();
1390 }
1391
1392 inline HeapWord* PSParallelCompact::new_top(SpaceId id) {
1393 assert(id < last_space_id, "id out of range");
1394 return _space_info[id].new_top();
1395 }
1396
1397 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) {
1398 assert(id < last_space_id, "id out of range");
1399 return _space_info[id].dense_prefix();
1400 }
1401
1402 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) {
1403 assert(id < last_space_id, "id out of range");
1404 return _space_info[id].start_array();
1405 }
1406
1407 inline bool PSParallelCompact::should_update_klass(klassOop k) {
1408 return ((HeapWord*) k) >= dense_prefix(perm_space_id);
1409 }
1410
1411 template <class T>
1412 inline void PSParallelCompact::adjust_pointer(T* p,
1413 HeapWord* beg_addr,
1414 HeapWord* end_addr) {
1415 if (is_in((HeapWord*)p, beg_addr, end_addr)) {
1416 adjust_pointer(p);
1417 }
1418 }
1419
1420 #ifdef ASSERT
1421 inline void
1422 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr)
1423 {
1424 assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr),
1425 "must move left or to a different space");
1426 assert(is_object_aligned((intptr_t)old_addr) && is_object_aligned((intptr_t)new_addr),
1427 "checking alignment");
1428 }
1429 #endif // ASSERT
1430
1431 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1432 public:
1433 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1434 ObjectStartArray* start_array,
1435 HeapWord* destination, size_t words);
1436
1437 // Accessors.
1438 HeapWord* destination() const { return _destination; }
1439
1440 // If the object will fit (size <= words_remaining()), copy it to the current
1441 // destination, update the interior oops and the start array and return either
1442 // full (if the closure is full) or incomplete. If the object will not fit,
1443 // return would_overflow.
1444 virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1445
1446 // Copy enough words to fill this closure, starting at source(). Interior
1447 // oops and the start array are not updated. Return full.
1448 IterationStatus copy_until_full();
1449
1450 // Copy enough words to fill this closure or to the end of an object,
1451 // whichever is smaller, starting at source(). Interior oops and the start
1452 // array are not updated.
1453 void copy_partial_obj();
1454
1455 protected:
1456 // Update variables to indicate that word_count words were processed.
1457 inline void update_state(size_t word_count);
1458
1459 protected:
1460 ObjectStartArray* const _start_array;
1461 HeapWord* _destination; // Next addr to be written.
1462 };
1463
1464 inline
1465 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1466 ParCompactionManager* cm,
1467 ObjectStartArray* start_array,
1468 HeapWord* destination,
1469 size_t words) :
1470 ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array)
1471 {
1472 _destination = destination;
1473 }
1474
1475 inline void MoveAndUpdateClosure::update_state(size_t words)
1476 {
1477 decrement_words_remaining(words);
1478 _source += words;
1479 _destination += words;
1480 }
1481
1482 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1483 private:
1484 const PSParallelCompact::SpaceId _space_id;
1485 ObjectStartArray* const _start_array;
1486
1487 public:
1488 UpdateOnlyClosure(ParMarkBitMap* mbm,
1489 ParCompactionManager* cm,
1490 PSParallelCompact::SpaceId space_id);
1491
1492 // Update the object.
1493 virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1494
1495 inline void do_addr(HeapWord* addr);
1496 };
1497
1498 inline void UpdateOnlyClosure::do_addr(HeapWord* addr)
1499 {
1500 _start_array->allocate_block(addr);
1501 oop(addr)->update_contents(compaction_manager());
1502 }
1503
1504 class FillClosure: public ParMarkBitMapClosure
1505 {
1506 public:
1507 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) :
1508 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm),
1509 _start_array(PSParallelCompact::start_array(space_id))
1510 {
1511 assert(space_id == PSParallelCompact::perm_space_id ||
1512 space_id == PSParallelCompact::old_space_id,
1513 "cannot use FillClosure in the young gen");
1514 }
1515
1516 virtual IterationStatus do_addr(HeapWord* addr, size_t size) {
1517 CollectedHeap::fill_with_objects(addr, size);
1518 HeapWord* const end = addr + size;
1519 do {
1520 _start_array->allocate_block(addr);
1521 addr += oop(addr)->size();
1522 } while (addr < end);
1523 return ParMarkBitMap::incomplete;
1524 }
1525
1526 private:
1527 ObjectStartArray* const _start_array;
1528 };
1529
1530 #endif // SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP