1 /*
2 * Copyright (c) 2005, 2019, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP
26 #define SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP
27
28 #include "gc/parallel/mutableSpace.hpp"
29 #include "gc/parallel/objectStartArray.hpp"
30 #include "gc/parallel/parMarkBitMap.hpp"
31 #include "gc/parallel/parallelScavengeHeap.hpp"
32 #include "gc/shared/collectedHeap.hpp"
33 #include "gc/shared/collectorCounters.hpp"
34 #include "oops/oop.hpp"
35
36 class ParallelScavengeHeap;
37 class PSAdaptiveSizePolicy;
38 class PSYoungGen;
39 class PSOldGen;
40 class ParCompactionManager;
41 class ParallelTaskTerminator;
42 class PSParallelCompact;
43 class PreGCValues;
44 class MoveAndUpdateClosure;
45 class RefProcTaskExecutor;
46 class ParallelOldTracer;
47 class STWGCTimer;
48
49 // The SplitInfo class holds the information needed to 'split' a source region
50 // so that the live data can be copied to two destination *spaces*. Normally,
51 // all the live data in a region is copied to a single destination space (e.g.,
52 // everything live in a region in eden is copied entirely into the old gen).
53 // However, when the heap is nearly full, all the live data in eden may not fit
54 // into the old gen. Copying only some of the regions from eden to old gen
55 // requires finding a region that does not contain a partial object (i.e., no
56 // live object crosses the region boundary) somewhere near the last object that
57 // does fit into the old gen. Since it's not always possible to find such a
58 // region, splitting is necessary for predictable behavior.
59 //
60 // A region is always split at the end of the partial object. This avoids
61 // additional tests when calculating the new location of a pointer, which is a
62 // very hot code path. The partial object and everything to its left will be
63 // copied to another space (call it dest_space_1). The live data to the right
64 // of the partial object will be copied either within the space itself, or to a
65 // different destination space (distinct from dest_space_1).
66 //
67 // Split points are identified during the summary phase, when region
68 // destinations are computed: data about the split, including the
69 // partial_object_size, is recorded in a SplitInfo record and the
70 // partial_object_size field in the summary data is set to zero. The zeroing is
71 // possible (and necessary) since the partial object will move to a different
72 // destination space than anything to its right, thus the partial object should
73 // not affect the locations of any objects to its right.
74 //
75 // The recorded data is used during the compaction phase, but only rarely: when
76 // the partial object on the split region will be copied across a destination
77 // region boundary. This test is made once each time a region is filled, and is
78 // a simple address comparison, so the overhead is negligible (see
79 // PSParallelCompact::first_src_addr()).
80 //
81 // Notes:
82 //
83 // Only regions with partial objects are split; a region without a partial
84 // object does not need any extra bookkeeping.
85 //
86 // At most one region is split per space, so the amount of data required is
87 // constant.
88 //
89 // A region is split only when the destination space would overflow. Once that
90 // happens, the destination space is abandoned and no other data (even from
91 // other source spaces) is targeted to that destination space. Abandoning the
92 // destination space may leave a somewhat large unused area at the end, if a
93 // large object caused the overflow.
94 //
95 // Future work:
96 //
97 // More bookkeeping would be required to continue to use the destination space.
98 // The most general solution would allow data from regions in two different
99 // source spaces to be "joined" in a single destination region. At the very
100 // least, additional code would be required in next_src_region() to detect the
101 // join and skip to an out-of-order source region. If the join region was also
102 // the last destination region to which a split region was copied (the most
103 // likely case), then additional work would be needed to get fill_region() to
104 // stop iteration and switch to a new source region at the right point. Basic
105 // idea would be to use a fake value for the top of the source space. It is
106 // doable, if a bit tricky.
107 //
108 // A simpler (but less general) solution would fill the remainder of the
109 // destination region with a dummy object and continue filling the next
110 // destination region.
111
112 class SplitInfo
113 {
114 public:
115 // Return true if this split info is valid (i.e., if a split has been
116 // recorded). The very first region cannot have a partial object and thus is
117 // never split, so 0 is the 'invalid' value.
118 bool is_valid() const { return _src_region_idx > 0; }
119
120 // Return true if this split holds data for the specified source region.
121 inline bool is_split(size_t source_region) const;
122
123 // The index of the split region, the size of the partial object on that
124 // region and the destination of the partial object.
125 size_t src_region_idx() const { return _src_region_idx; }
126 size_t partial_obj_size() const { return _partial_obj_size; }
127 HeapWord* destination() const { return _destination; }
128
129 // The destination count of the partial object referenced by this split
130 // (either 1 or 2). This must be added to the destination count of the
131 // remainder of the source region.
132 unsigned int destination_count() const { return _destination_count; }
133
134 // If a word within the partial object will be written to the first word of a
135 // destination region, this is the address of the destination region;
136 // otherwise this is NULL.
137 HeapWord* dest_region_addr() const { return _dest_region_addr; }
138
139 // If a word within the partial object will be written to the first word of a
140 // destination region, this is the address of that word within the partial
141 // object; otherwise this is NULL.
142 HeapWord* first_src_addr() const { return _first_src_addr; }
143
144 // Record the data necessary to split the region src_region_idx.
145 void record(size_t src_region_idx, size_t partial_obj_size,
146 HeapWord* destination);
147
148 void clear();
149
150 DEBUG_ONLY(void verify_clear();)
151
152 private:
153 size_t _src_region_idx;
154 size_t _partial_obj_size;
155 HeapWord* _destination;
156 unsigned int _destination_count;
157 HeapWord* _dest_region_addr;
158 HeapWord* _first_src_addr;
159 };
160
161 inline bool SplitInfo::is_split(size_t region_idx) const
162 {
163 return _src_region_idx == region_idx && is_valid();
164 }
165
166 class SpaceInfo
167 {
168 public:
169 MutableSpace* space() const { return _space; }
170
171 // Where the free space will start after the collection. Valid only after the
172 // summary phase completes.
173 HeapWord* new_top() const { return _new_top; }
174
175 // Allows new_top to be set.
176 HeapWord** new_top_addr() { return &_new_top; }
177
178 // Where the smallest allowable dense prefix ends (used only for perm gen).
179 HeapWord* min_dense_prefix() const { return _min_dense_prefix; }
180
181 // Where the dense prefix ends, or the compacted region begins.
182 HeapWord* dense_prefix() const { return _dense_prefix; }
183
184 // The start array for the (generation containing the) space, or NULL if there
185 // is no start array.
186 ObjectStartArray* start_array() const { return _start_array; }
187
188 SplitInfo& split_info() { return _split_info; }
189
190 void set_space(MutableSpace* s) { _space = s; }
191 void set_new_top(HeapWord* addr) { _new_top = addr; }
192 void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; }
193 void set_dense_prefix(HeapWord* addr) { _dense_prefix = addr; }
194 void set_start_array(ObjectStartArray* s) { _start_array = s; }
195
196 void publish_new_top() const { _space->set_top(_new_top); }
197
198 private:
199 MutableSpace* _space;
200 HeapWord* _new_top;
201 HeapWord* _min_dense_prefix;
202 HeapWord* _dense_prefix;
203 ObjectStartArray* _start_array;
204 SplitInfo _split_info;
205 };
206
207 class ParallelCompactData
208 {
209 public:
210 // Sizes are in HeapWords, unless indicated otherwise.
211 static const size_t Log2RegionSize;
212 static const size_t RegionSize;
213 static const size_t RegionSizeBytes;
214
215 // Mask for the bits in a size_t to get an offset within a region.
216 static const size_t RegionSizeOffsetMask;
217 // Mask for the bits in a pointer to get an offset within a region.
218 static const size_t RegionAddrOffsetMask;
219 // Mask for the bits in a pointer to get the address of the start of a region.
220 static const size_t RegionAddrMask;
221
222 static const size_t Log2BlockSize;
223 static const size_t BlockSize;
224 static const size_t BlockSizeBytes;
225
226 static const size_t BlockSizeOffsetMask;
227 static const size_t BlockAddrOffsetMask;
228 static const size_t BlockAddrMask;
229
230 static const size_t BlocksPerRegion;
231 static const size_t Log2BlocksPerRegion;
232
233 class RegionData
234 {
235 public:
236 // Destination address of the region.
237 HeapWord* destination() const { return _destination; }
238
239 // The first region containing data destined for this region.
240 size_t source_region() const { return _source_region; }
241
242 // Reuse _source_region to store the corresponding shadow region index
243 size_t shadow_region() const { return _source_region; }
244
245 // The object (if any) starting in this region and ending in a different
246 // region that could not be updated during the main (parallel) compaction
247 // phase. This is different from _partial_obj_addr, which is an object that
248 // extends onto a source region. However, the two uses do not overlap in
249 // time, so the same field is used to save space.
250 HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
251
252 // The starting address of the partial object extending onto the region.
253 HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
254
255 // Size of the partial object extending onto the region (words).
256 size_t partial_obj_size() const { return _partial_obj_size; }
257
258 // Size of live data that lies within this region due to objects that start
259 // in this region (words). This does not include the partial object
260 // extending onto the region (if any), or the part of an object that extends
261 // onto the next region (if any).
262 size_t live_obj_size() const { return _dc_and_los & los_mask; }
263
264 // Total live data that lies within the region (words).
265 size_t data_size() const { return partial_obj_size() + live_obj_size(); }
266
267 // The destination_count is the number of other regions to which data from
268 // this region will be copied. At the end of the summary phase, the valid
269 // values of destination_count are
270 //
271 // 0 - data from the region will be compacted completely into itself, or the
272 // region is empty. The region can be claimed and then filled.
273 // 1 - data from the region will be compacted into 1 other region; some
274 // data from the region may also be compacted into the region itself.
275 // 2 - data from the region will be copied to 2 other regions.
276 //
277 // During compaction as regions are emptied, the destination_count is
278 // decremented (atomically) and when it reaches 0, it can be claimed and
279 // then filled.
280 //
281 // A region is claimed for processing by atomically changing the
282 // destination_count to the claimed value (dc_claimed). After a region has
283 // been filled, the destination_count should be set to the completed value
284 // (dc_completed).
285 inline uint destination_count() const;
286 inline uint destination_count_raw() const;
287
288 // Whether the block table for this region has been filled.
289 inline bool blocks_filled() const;
290
291 // Number of times the block table was filled.
292 DEBUG_ONLY(inline size_t blocks_filled_count() const;)
293
294 // The location of the java heap data that corresponds to this region.
295 inline HeapWord* data_location() const;
296
297 // The highest address referenced by objects in this region.
298 inline HeapWord* highest_ref() const;
299
300 // Whether this region is available to be claimed, has been claimed, or has
301 // been completed.
302 //
303 // Minor subtlety: claimed() returns true if the region is marked
304 // completed(), which is desirable since a region must be claimed before it
305 // can be completed.
306 bool available() const { return _dc_and_los < dc_one; }
307 bool claimed() const { return _dc_and_los >= dc_claimed; }
308 bool completed() const { return _dc_and_los >= dc_completed; }
309
310 // These are not atomic.
311 void set_destination(HeapWord* addr) { _destination = addr; }
312 void set_source_region(size_t region) { _source_region = region; }
313 void set_shadow_region(size_t region) { _source_region = region; }
314 void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
315 void set_partial_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
316 void set_partial_obj_size(size_t words) {
317 _partial_obj_size = (region_sz_t) words;
318 }
319 inline void set_blocks_filled();
320
321 inline void set_destination_count(uint count);
322 inline void set_live_obj_size(size_t words);
323 inline void set_data_location(HeapWord* addr);
324 inline void set_completed();
325 inline bool claim_unsafe();
326
327 // These are atomic.
328 inline void add_live_obj(size_t words);
329 inline void set_highest_ref(HeapWord* addr);
330 inline void decrement_destination_count();
331 inline bool claim();
332
333 // Possible values of _shadow_state, and transition is as follows
334 // Normal Path:
335 // UNUSED -> try_push() -> FINISHED
336 // Steal Path:
337 // UNUSED -> try_steal() -> SHADOW -> mark_filled() -> FILLED -> try_copy() -> FINISHED
338 static const int UNUSED; // Original state
339 static const int SHADOW; // Stolen by an idle thread, and a shadow region is created for it
340 static const int FILLED; // Its shadow region has been filled and ready to be copied back
341 static const int FINISH; // Work has been done
342
343 // Preempt the region to avoid double processes
344 inline bool try_push();
345 inline bool try_steal();
346 // Mark the region as filled and ready to be copied back
347 inline bool mark_filled();
348 // Preempt the region to copy the shadow region content back
349 inline bool try_copy();
350
351 int shadow_state() { return _shadow_state; }
352
353 private:
354 // The type used to represent object sizes within a region.
355 typedef uint region_sz_t;
356
357 // Constants for manipulating the _dc_and_los field, which holds both the
358 // destination count and live obj size. The live obj size lives at the
359 // least significant end so no masking is necessary when adding.
360 static const region_sz_t dc_shift; // Shift amount.
361 static const region_sz_t dc_mask; // Mask for destination count.
362 static const region_sz_t dc_one; // 1, shifted appropriately.
363 static const region_sz_t dc_claimed; // Region has been claimed.
364 static const region_sz_t dc_completed; // Region has been completed.
365 static const region_sz_t los_mask; // Mask for live obj size.
366
367 HeapWord* _destination;
368 size_t _source_region;
369 HeapWord* _partial_obj_addr;
370 region_sz_t _partial_obj_size;
371 region_sz_t volatile _dc_and_los;
372 bool volatile _blocks_filled;
373 int volatile _shadow_state;
374
375 #ifdef ASSERT
376 size_t _blocks_filled_count; // Number of block table fills.
377
378 // These enable optimizations that are only partially implemented. Use
379 // debug builds to prevent the code fragments from breaking.
380 HeapWord* _data_location;
381 HeapWord* _highest_ref;
382 #endif // #ifdef ASSERT
383
384 #ifdef ASSERT
385 public:
386 uint _pushed; // 0 until region is pushed onto a stack
387 private:
388 #endif
389 };
390
391 // "Blocks" allow shorter sections of the bitmap to be searched. Each Block
392 // holds an offset, which is the amount of live data in the Region to the left
393 // of the first live object that starts in the Block.
394 class BlockData
395 {
396 public:
397 typedef unsigned short int blk_ofs_t;
398
399 blk_ofs_t offset() const { return _offset; }
400 void set_offset(size_t val) { _offset = (blk_ofs_t)val; }
401
402 private:
403 blk_ofs_t _offset;
404 };
405
406 public:
407 ParallelCompactData();
408 bool initialize(MemRegion covered_region);
409
410 size_t region_count() const { return _region_count; }
411 size_t reserved_byte_size() const { return _reserved_byte_size; }
412
413 // Convert region indices to/from RegionData pointers.
414 inline RegionData* region(size_t region_idx) const;
415 inline size_t region(const RegionData* const region_ptr) const;
416
417 size_t block_count() const { return _block_count; }
418 inline BlockData* block(size_t block_idx) const;
419 inline size_t block(const BlockData* block_ptr) const;
420
421 void add_obj(HeapWord* addr, size_t len);
422 void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
423
424 // Fill in the regions covering [beg, end) so that no data moves; i.e., the
425 // destination of region n is simply the start of region n. The argument beg
426 // must be region-aligned; end need not be.
427 void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
428
429 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
430 HeapWord* destination, HeapWord* target_end,
431 HeapWord** target_next);
432 bool summarize(SplitInfo& split_info,
433 HeapWord* source_beg, HeapWord* source_end,
434 HeapWord** source_next,
435 HeapWord* target_beg, HeapWord* target_end,
436 HeapWord** target_next);
437
438 void clear();
439 void clear_range(size_t beg_region, size_t end_region);
440 void clear_range(HeapWord* beg, HeapWord* end) {
441 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
442 }
443
444 // Return the number of words between addr and the start of the region
445 // containing addr.
446 inline size_t region_offset(const HeapWord* addr) const;
447
448 // Convert addresses to/from a region index or region pointer.
449 inline size_t addr_to_region_idx(const HeapWord* addr) const;
450 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
451 inline HeapWord* region_to_addr(size_t region) const;
452 inline HeapWord* region_to_addr(size_t region, size_t offset) const;
453 inline HeapWord* region_to_addr(const RegionData* region) const;
454
455 inline HeapWord* region_align_down(HeapWord* addr) const;
456 inline HeapWord* region_align_up(HeapWord* addr) const;
457 inline bool is_region_aligned(HeapWord* addr) const;
458
459 // Analogous to region_offset() for blocks.
460 size_t block_offset(const HeapWord* addr) const;
461 size_t addr_to_block_idx(const HeapWord* addr) const;
462 size_t addr_to_block_idx(const oop obj) const {
463 return addr_to_block_idx((HeapWord*) obj);
464 }
465 inline BlockData* addr_to_block_ptr(const HeapWord* addr) const;
466 inline HeapWord* block_to_addr(size_t block) const;
467 inline size_t region_to_block_idx(size_t region) const;
468
469 inline HeapWord* block_align_down(HeapWord* addr) const;
470 inline HeapWord* block_align_up(HeapWord* addr) const;
471 inline bool is_block_aligned(HeapWord* addr) const;
472
473 // Return the address one past the end of the partial object.
474 HeapWord* partial_obj_end(size_t region_idx) const;
475
476 // Return the location of the object after compaction.
477 HeapWord* calc_new_pointer(HeapWord* addr, ParCompactionManager* cm);
478
479 HeapWord* calc_new_pointer(oop p, ParCompactionManager* cm) {
480 return calc_new_pointer((HeapWord*) p, cm);
481 }
482
483 #ifdef ASSERT
484 void verify_clear(const PSVirtualSpace* vspace);
485 void verify_clear();
486 #endif // #ifdef ASSERT
487
488 private:
489 bool initialize_block_data();
490 bool initialize_region_data(size_t region_size);
491 PSVirtualSpace* create_vspace(size_t count, size_t element_size);
492
493 private:
494 HeapWord* _region_start;
495 #ifdef ASSERT
496 HeapWord* _region_end;
497 #endif // #ifdef ASSERT
498
499 PSVirtualSpace* _region_vspace;
500 size_t _reserved_byte_size;
501 RegionData* _region_data;
502 size_t _region_count;
503
504 PSVirtualSpace* _block_vspace;
505 BlockData* _block_data;
506 size_t _block_count;
507 };
508
509 inline uint
510 ParallelCompactData::RegionData::destination_count_raw() const
511 {
512 return _dc_and_los & dc_mask;
513 }
514
515 inline uint
516 ParallelCompactData::RegionData::destination_count() const
517 {
518 return destination_count_raw() >> dc_shift;
519 }
520
521 inline bool
522 ParallelCompactData::RegionData::blocks_filled() const
523 {
524 bool result = _blocks_filled;
525 OrderAccess::acquire();
526 return result;
527 }
528
529 #ifdef ASSERT
530 inline size_t
531 ParallelCompactData::RegionData::blocks_filled_count() const
532 {
533 return _blocks_filled_count;
534 }
535 #endif // #ifdef ASSERT
536
537 inline void
538 ParallelCompactData::RegionData::set_blocks_filled()
539 {
540 OrderAccess::release();
541 _blocks_filled = true;
542 // Debug builds count the number of times the table was filled.
543 DEBUG_ONLY(Atomic::inc(&_blocks_filled_count));
544 }
545
546 inline void
547 ParallelCompactData::RegionData::set_destination_count(uint count)
548 {
549 assert(count <= (dc_completed >> dc_shift), "count too large");
550 const region_sz_t live_sz = (region_sz_t) live_obj_size();
551 _dc_and_los = (count << dc_shift) | live_sz;
552 }
553
554 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
555 {
556 assert(words <= los_mask, "would overflow");
557 _dc_and_los = destination_count_raw() | (region_sz_t)words;
558 }
559
560 inline void ParallelCompactData::RegionData::decrement_destination_count()
561 {
562 assert(_dc_and_los < dc_claimed, "already claimed");
563 assert(_dc_and_los >= dc_one, "count would go negative");
564 Atomic::add(dc_mask, &_dc_and_los);
565 }
566
567 inline HeapWord* ParallelCompactData::RegionData::data_location() const
568 {
569 DEBUG_ONLY(return _data_location;)
570 NOT_DEBUG(return NULL;)
571 }
572
573 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
574 {
575 DEBUG_ONLY(return _highest_ref;)
576 NOT_DEBUG(return NULL;)
577 }
578
579 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
580 {
581 DEBUG_ONLY(_data_location = addr;)
582 }
583
584 inline void ParallelCompactData::RegionData::set_completed()
585 {
586 assert(claimed(), "must be claimed first");
587 _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
588 }
589
590 // MT-unsafe claiming of a region. Should only be used during single threaded
591 // execution.
592 inline bool ParallelCompactData::RegionData::claim_unsafe()
593 {
594 if (available()) {
595 _dc_and_los |= dc_claimed;
596 return true;
597 }
598 return false;
599 }
600
601 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
602 {
603 assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
604 Atomic::add(static_cast<region_sz_t>(words), &_dc_and_los);
605 }
606
607 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
608 {
609 #ifdef ASSERT
610 HeapWord* tmp = _highest_ref;
611 while (addr > tmp) {
612 tmp = Atomic::cmpxchg(addr, &_highest_ref, tmp);
613 }
614 #endif // #ifdef ASSERT
615 }
616
617 inline bool ParallelCompactData::RegionData::claim()
618 {
619 const region_sz_t los = static_cast<region_sz_t>(live_obj_size());
620 const region_sz_t old = Atomic::cmpxchg(dc_claimed | los, &_dc_and_los, los);
621 return old == los;
622 }
623
624 inline bool ParallelCompactData::RegionData::try_push() {
625 return Atomic::cmpxchg(FINISH, &_shadow_state, UNUSED) == UNUSED;
626 }
627
628 inline bool ParallelCompactData::RegionData::try_steal() {
629 return Atomic::cmpxchg(SHADOW, &_shadow_state, UNUSED) == UNUSED;
630 }
631
632 inline bool ParallelCompactData::RegionData::mark_filled() {
633 return Atomic::cmpxchg(FILLED, &_shadow_state, SHADOW) == SHADOW;
634 }
635
636 inline bool ParallelCompactData::RegionData::try_copy() {
637 return Atomic::cmpxchg(FINISH, &_shadow_state, FILLED) == FILLED;
638 }
639
640 inline ParallelCompactData::RegionData*
641 ParallelCompactData::region(size_t region_idx) const
642 {
643 assert(region_idx <= region_count(), "bad arg");
644 return _region_data + region_idx;
645 }
646
647 inline size_t
648 ParallelCompactData::region(const RegionData* const region_ptr) const
649 {
650 assert(region_ptr >= _region_data, "bad arg");
651 assert(region_ptr <= _region_data + region_count(), "bad arg");
652 return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
653 }
654
655 inline ParallelCompactData::BlockData*
656 ParallelCompactData::block(size_t n) const {
657 assert(n < block_count(), "bad arg");
658 return _block_data + n;
659 }
660
661 inline size_t
662 ParallelCompactData::region_offset(const HeapWord* addr) const
663 {
664 assert(addr >= _region_start, "bad addr");
665 assert(addr <= _region_end, "bad addr");
666 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
667 }
668
669 inline size_t
670 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
671 {
672 assert(addr >= _region_start, "bad addr " PTR_FORMAT " _region_start " PTR_FORMAT, p2i(addr), p2i(_region_start));
673 assert(addr <= _region_end, "bad addr " PTR_FORMAT " _region_end " PTR_FORMAT, p2i(addr), p2i(_region_end));
674 return pointer_delta(addr, _region_start) >> Log2RegionSize;
675 }
676
677 inline ParallelCompactData::RegionData*
678 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
679 {
680 return region(addr_to_region_idx(addr));
681 }
682
683 inline HeapWord*
684 ParallelCompactData::region_to_addr(size_t region) const
685 {
686 assert(region <= _region_count, "region out of range");
687 return _region_start + (region << Log2RegionSize);
688 }
689
690 inline HeapWord*
691 ParallelCompactData::region_to_addr(const RegionData* region) const
692 {
693 return region_to_addr(pointer_delta(region, _region_data,
694 sizeof(RegionData)));
695 }
696
697 inline HeapWord*
698 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
699 {
700 assert(region <= _region_count, "region out of range");
701 assert(offset < RegionSize, "offset too big"); // This may be too strict.
702 return region_to_addr(region) + offset;
703 }
704
705 inline HeapWord*
706 ParallelCompactData::region_align_down(HeapWord* addr) const
707 {
708 assert(addr >= _region_start, "bad addr");
709 assert(addr < _region_end + RegionSize, "bad addr");
710 return (HeapWord*)(size_t(addr) & RegionAddrMask);
711 }
712
713 inline HeapWord*
714 ParallelCompactData::region_align_up(HeapWord* addr) const
715 {
716 assert(addr >= _region_start, "bad addr");
717 assert(addr <= _region_end, "bad addr");
718 return region_align_down(addr + RegionSizeOffsetMask);
719 }
720
721 inline bool
722 ParallelCompactData::is_region_aligned(HeapWord* addr) const
723 {
724 return region_offset(addr) == 0;
725 }
726
727 inline size_t
728 ParallelCompactData::block_offset(const HeapWord* addr) const
729 {
730 assert(addr >= _region_start, "bad addr");
731 assert(addr <= _region_end, "bad addr");
732 return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize;
733 }
734
735 inline size_t
736 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const
737 {
738 assert(addr >= _region_start, "bad addr");
739 assert(addr <= _region_end, "bad addr");
740 return pointer_delta(addr, _region_start) >> Log2BlockSize;
741 }
742
743 inline ParallelCompactData::BlockData*
744 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const
745 {
746 return block(addr_to_block_idx(addr));
747 }
748
749 inline HeapWord*
750 ParallelCompactData::block_to_addr(size_t block) const
751 {
752 assert(block < _block_count, "block out of range");
753 return _region_start + (block << Log2BlockSize);
754 }
755
756 inline size_t
757 ParallelCompactData::region_to_block_idx(size_t region) const
758 {
759 return region << Log2BlocksPerRegion;
760 }
761
762 inline HeapWord*
763 ParallelCompactData::block_align_down(HeapWord* addr) const
764 {
765 assert(addr >= _region_start, "bad addr");
766 assert(addr < _region_end + RegionSize, "bad addr");
767 return (HeapWord*)(size_t(addr) & BlockAddrMask);
768 }
769
770 inline HeapWord*
771 ParallelCompactData::block_align_up(HeapWord* addr) const
772 {
773 assert(addr >= _region_start, "bad addr");
774 assert(addr <= _region_end, "bad addr");
775 return block_align_down(addr + BlockSizeOffsetMask);
776 }
777
778 inline bool
779 ParallelCompactData::is_block_aligned(HeapWord* addr) const
780 {
781 return block_offset(addr) == 0;
782 }
783
784 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
785 // do_addr() method.
786 //
787 // The closure is initialized with the number of heap words to process
788 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr()
789 // methods in subclasses should update the total as words are processed. Since
790 // only one subclass actually uses this mechanism to terminate iteration, the
791 // default initial value is > 0. The implementation is here and not in the
792 // single subclass that uses it to avoid making is_full() virtual, and thus
793 // adding a virtual call per live object.
794
795 class ParMarkBitMapClosure: public StackObj {
796 public:
797 typedef ParMarkBitMap::idx_t idx_t;
798 typedef ParMarkBitMap::IterationStatus IterationStatus;
799
800 public:
801 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
802 size_t words = max_uintx);
803
804 inline ParCompactionManager* compaction_manager() const;
805 inline ParMarkBitMap* bitmap() const;
806 inline size_t words_remaining() const;
807 inline bool is_full() const;
808 inline HeapWord* source() const;
809
810 inline void set_source(HeapWord* addr);
811
812 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
813
814 protected:
815 inline void decrement_words_remaining(size_t words);
816
817 private:
818 ParMarkBitMap* const _bitmap;
819 ParCompactionManager* const _compaction_manager;
820 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger.
821 size_t _words_remaining; // Words left to copy.
822
823 protected:
824 HeapWord* _source; // Next addr that would be read.
825 };
826
827 inline
828 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
829 ParCompactionManager* cm,
830 size_t words):
831 _bitmap(bitmap), _compaction_manager(cm)
832 #ifdef ASSERT
833 , _initial_words_remaining(words)
834 #endif
835 {
836 _words_remaining = words;
837 _source = NULL;
838 }
839
840 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
841 return _compaction_manager;
842 }
843
844 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
845 return _bitmap;
846 }
847
848 inline size_t ParMarkBitMapClosure::words_remaining() const {
849 return _words_remaining;
850 }
851
852 inline bool ParMarkBitMapClosure::is_full() const {
853 return words_remaining() == 0;
854 }
855
856 inline HeapWord* ParMarkBitMapClosure::source() const {
857 return _source;
858 }
859
860 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
861 _source = addr;
862 }
863
864 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
865 assert(_words_remaining >= words, "processed too many words");
866 _words_remaining -= words;
867 }
868
869 // The UseParallelOldGC collector is a stop-the-world garbage collector that
870 // does parts of the collection using parallel threads. The collection includes
871 // the tenured generation and the young generation. The permanent generation is
872 // collected at the same time as the other two generations but the permanent
873 // generation is collect by a single GC thread. The permanent generation is
874 // collected serially because of the requirement that during the processing of a
875 // klass AAA, any objects reference by AAA must already have been processed.
876 // This requirement is enforced by a left (lower address) to right (higher
877 // address) sliding compaction.
878 //
879 // There are four phases of the collection.
880 //
881 // - marking phase
882 // - summary phase
883 // - compacting phase
884 // - clean up phase
885 //
886 // Roughly speaking these phases correspond, respectively, to
887 // - mark all the live objects
888 // - calculate the destination of each object at the end of the collection
889 // - move the objects to their destination
890 // - update some references and reinitialize some variables
891 //
892 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The
893 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
894 // complete marking of the heap. The summary phase is implemented in
895 // PSParallelCompact::summary_phase(). The move and update phase is implemented
896 // in PSParallelCompact::compact().
897 //
898 // A space that is being collected is divided into regions and with each region
899 // is associated an object of type ParallelCompactData. Each region is of a
900 // fixed size and typically will contain more than 1 object and may have parts
901 // of objects at the front and back of the region.
902 //
903 // region -----+---------------------+----------
904 // objects covered [ AAA )[ BBB )[ CCC )[ DDD )
905 //
906 // The marking phase does a complete marking of all live objects in the heap.
907 // The marking also compiles the size of the data for all live objects covered
908 // by the region. This size includes the part of any live object spanning onto
909 // the region (part of AAA if it is live) from the front, all live objects
910 // contained in the region (BBB and/or CCC if they are live), and the part of
911 // any live objects covered by the region that extends off the region (part of
912 // DDD if it is live). The marking phase uses multiple GC threads and marking
913 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is
914 // done atomically as is the accumulation of the size of the live objects
915 // covered by a region.
916 //
917 // The summary phase calculates the total live data to the left of each region
918 // XXX. Based on that total and the bottom of the space, it can calculate the
919 // starting location of the live data in XXX. The summary phase calculates for
920 // each region XXX quantities such as
921 //
922 // - the amount of live data at the beginning of a region from an object
923 // entering the region.
924 // - the location of the first live data on the region
925 // - a count of the number of regions receiving live data from XXX.
926 //
927 // See ParallelCompactData for precise details. The summary phase also
928 // calculates the dense prefix for the compaction. The dense prefix is a
929 // portion at the beginning of the space that is not moved. The objects in the
930 // dense prefix do need to have their object references updated. See method
931 // summarize_dense_prefix().
932 //
933 // The summary phase is done using 1 GC thread.
934 //
935 // The compaction phase moves objects to their new location and updates all
936 // references in the object.
937 //
938 // A current exception is that objects that cross a region boundary are moved
939 // but do not have their references updated. References are not updated because
940 // it cannot easily be determined if the klass pointer KKK for the object AAA
941 // has been updated. KKK likely resides in a region to the left of the region
942 // containing AAA. These AAA's have there references updated at the end in a
943 // clean up phase. See the method PSParallelCompact::update_deferred_objects().
944 // An alternate strategy is being investigated for this deferral of updating.
945 //
946 // Compaction is done on a region basis. A region that is ready to be filled is
947 // put on a ready list and GC threads take region off the list and fill them. A
948 // region is ready to be filled if it empty of live objects. Such a region may
949 // have been initially empty (only contained dead objects) or may have had all
950 // its live objects copied out already. A region that compacts into itself is
951 // also ready for filling. The ready list is initially filled with empty
952 // regions and regions compacting into themselves. There is always at least 1
953 // region that can be put on the ready list. The regions are atomically added
954 // and removed from the ready list.
955
956 class TaskQueue;
957
958 class PSParallelCompact : AllStatic {
959 public:
960 // Convenient access to type names.
961 typedef ParMarkBitMap::idx_t idx_t;
962 typedef ParallelCompactData::RegionData RegionData;
963 typedef ParallelCompactData::BlockData BlockData;
964
965 typedef enum {
966 old_space_id, eden_space_id,
967 from_space_id, to_space_id, last_space_id
968 } SpaceId;
969
970 struct UpdateDensePrefixTask : public CHeapObj<mtGC> {
971 SpaceId _space_id;
972 size_t _region_index_start;
973 size_t _region_index_end;
974
975 UpdateDensePrefixTask() :
976 _space_id(SpaceId(0)),
977 _region_index_start(0),
978 _region_index_end(0) {}
979
980 UpdateDensePrefixTask(SpaceId space_id,
981 size_t region_index_start,
982 size_t region_index_end) :
983 _space_id(space_id),
984 _region_index_start(region_index_start),
985 _region_index_end(region_index_end) {}
986 };
987
988 public:
989 // Inline closure decls
990 //
991 class IsAliveClosure: public BoolObjectClosure {
992 public:
993 virtual bool do_object_b(oop p);
994 };
995
996 friend class RefProcTaskProxy;
997 friend class PSParallelCompactTest;
998
999 private:
1000 static STWGCTimer _gc_timer;
1001 static ParallelOldTracer _gc_tracer;
1002 static elapsedTimer _accumulated_time;
1003 static unsigned int _total_invocations;
1004 static unsigned int _maximum_compaction_gc_num;
1005 static jlong _time_of_last_gc; // ms
1006 static CollectorCounters* _counters;
1007 static ParMarkBitMap _mark_bitmap;
1008 static ParallelCompactData _summary_data;
1009 static IsAliveClosure _is_alive_closure;
1010 static SpaceInfo _space_info[last_space_id];
1011
1012 // Reference processing (used in ...follow_contents)
1013 static SpanSubjectToDiscoveryClosure _span_based_discoverer;
1014 static ReferenceProcessor* _ref_processor;
1015
1016 // Values computed at initialization and used by dead_wood_limiter().
1017 static double _dwl_mean;
1018 static double _dwl_std_dev;
1019 static double _dwl_first_term;
1020 static double _dwl_adjustment;
1021 #ifdef ASSERT
1022 static bool _dwl_initialized;
1023 #endif // #ifdef ASSERT
1024
1025 public:
1026 static ParallelOldTracer* gc_tracer() { return &_gc_tracer; }
1027
1028 private:
1029
1030 static void initialize_space_info();
1031
1032 // Clear the marking bitmap and summary data that cover the specified space.
1033 static void clear_data_covering_space(SpaceId id);
1034
1035 static void pre_compact();
1036 static void post_compact();
1037
1038 // Mark live objects
1039 static void marking_phase(ParCompactionManager* cm,
1040 bool maximum_heap_compaction,
1041 ParallelOldTracer *gc_tracer);
1042
1043 // Compute the dense prefix for the designated space. This is an experimental
1044 // implementation currently not used in production.
1045 static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
1046 bool maximum_compaction);
1047
1048 // Methods used to compute the dense prefix.
1049
1050 // Compute the value of the normal distribution at x = density. The mean and
1051 // standard deviation are values saved by initialize_dead_wood_limiter().
1052 static inline double normal_distribution(double density);
1053
1054 // Initialize the static vars used by dead_wood_limiter().
1055 static void initialize_dead_wood_limiter();
1056
1057 // Return the percentage of space that can be treated as "dead wood" (i.e.,
1058 // not reclaimed).
1059 static double dead_wood_limiter(double density, size_t min_percent);
1060
1061 // Find the first (left-most) region in the range [beg, end) that has at least
1062 // dead_words of dead space to the left. The argument beg must be the first
1063 // region in the space that is not completely live.
1064 static RegionData* dead_wood_limit_region(const RegionData* beg,
1065 const RegionData* end,
1066 size_t dead_words);
1067
1068 // Return a pointer to the first region in the range [beg, end) that is not
1069 // completely full.
1070 static RegionData* first_dead_space_region(const RegionData* beg,
1071 const RegionData* end);
1072
1073 // Return a value indicating the benefit or 'yield' if the compacted region
1074 // were to start (or equivalently if the dense prefix were to end) at the
1075 // candidate region. Higher values are better.
1076 //
1077 // The value is based on the amount of space reclaimed vs. the costs of (a)
1078 // updating references in the dense prefix plus (b) copying objects and
1079 // updating references in the compacted region.
1080 static inline double reclaimed_ratio(const RegionData* const candidate,
1081 HeapWord* const bottom,
1082 HeapWord* const top,
1083 HeapWord* const new_top);
1084
1085 // Compute the dense prefix for the designated space.
1086 static HeapWord* compute_dense_prefix(const SpaceId id,
1087 bool maximum_compaction);
1088
1089 // Return true if dead space crosses onto the specified Region; bit must be
1090 // the bit index corresponding to the first word of the Region.
1091 static inline bool dead_space_crosses_boundary(const RegionData* region,
1092 idx_t bit);
1093
1094 // Summary phase utility routine to fill dead space (if any) at the dense
1095 // prefix boundary. Should only be called if the the dense prefix is
1096 // non-empty.
1097 static void fill_dense_prefix_end(SpaceId id);
1098
1099 static void summarize_spaces_quick();
1100 static void summarize_space(SpaceId id, bool maximum_compaction);
1101 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
1102
1103 // Adjust addresses in roots. Does not adjust addresses in heap.
1104 static void adjust_roots(ParCompactionManager* cm);
1105
1106 DEBUG_ONLY(static void write_block_fill_histogram();)
1107
1108 // Move objects to new locations.
1109 static void compact_perm(ParCompactionManager* cm);
1110 static void compact();
1111
1112 // Add available regions to the stack and draining tasks to the task queue.
1113 static void prepare_region_draining_tasks(uint parallel_gc_threads);
1114
1115 // Add dense prefix update tasks to the task queue.
1116 static void enqueue_dense_prefix_tasks(TaskQueue& task_queue,
1117 uint parallel_gc_threads);
1118
1119 // If objects are left in eden after a collection, try to move the boundary
1120 // and absorb them into the old gen. Returns true if eden was emptied.
1121 static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
1122 PSYoungGen* young_gen,
1123 PSOldGen* old_gen);
1124
1125 // Reset time since last full gc
1126 static void reset_millis_since_last_gc();
1127
1128 #ifndef PRODUCT
1129 // Print generic summary data
1130 static void print_generic_summary_data(ParallelCompactData& summary_data,
1131 HeapWord* const beg_addr,
1132 HeapWord* const end_addr);
1133 #endif // #ifndef PRODUCT
1134
1135 public:
1136
1137 PSParallelCompact();
1138
1139 static void invoke(bool maximum_heap_compaction);
1140 static bool invoke_no_policy(bool maximum_heap_compaction);
1141
1142 static void post_initialize();
1143 // Perform initialization for PSParallelCompact that requires
1144 // allocations. This should be called during the VM initialization
1145 // at a pointer where it would be appropriate to return a JNI_ENOMEM
1146 // in the event of a failure.
1147 static bool initialize();
1148
1149 // Closure accessors
1150 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; }
1151
1152 // Public accessors
1153 static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1154 static unsigned int total_invocations() { return _total_invocations; }
1155 static CollectorCounters* counters() { return _counters; }
1156
1157 // Marking support
1158 static inline bool mark_obj(oop obj);
1159 static inline bool is_marked(oop obj);
1160
1161 template <class T> static inline void adjust_pointer(T* p, ParCompactionManager* cm);
1162
1163 // Compaction support.
1164 // Return true if p is in the range [beg_addr, end_addr).
1165 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1166 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1167
1168 // Convenience wrappers for per-space data kept in _space_info.
1169 static inline MutableSpace* space(SpaceId space_id);
1170 static inline HeapWord* new_top(SpaceId space_id);
1171 static inline HeapWord* dense_prefix(SpaceId space_id);
1172 static inline ObjectStartArray* start_array(SpaceId space_id);
1173
1174 // Process the end of the given region range in the dense prefix.
1175 // This includes saving any object not updated.
1176 static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1177 size_t region_start_index,
1178 size_t region_end_index,
1179 idx_t exiting_object_offset,
1180 idx_t region_offset_start,
1181 idx_t region_offset_end);
1182
1183 // Update a region in the dense prefix. For each live object
1184 // in the region, update it's interior references. For each
1185 // dead object, fill it with deadwood. Dead space at the end
1186 // of a region range will be filled to the start of the next
1187 // live object regardless of the region_index_end. None of the
1188 // objects in the dense prefix move and dead space is dead
1189 // (holds only dead objects that don't need any processing), so
1190 // dead space can be filled in any order.
1191 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1192 SpaceId space_id,
1193 size_t region_index_start,
1194 size_t region_index_end);
1195
1196 // Return the address of the count + 1st live word in the range [beg, end).
1197 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1198
1199 // Return the address of the word to be copied to dest_addr, which must be
1200 // aligned to a region boundary.
1201 static HeapWord* first_src_addr(HeapWord* const dest_addr,
1202 SpaceId src_space_id,
1203 size_t src_region_idx);
1204
1205 // Determine the next source region, set closure.source() to the start of the
1206 // new region return the region index. Parameter end_addr is the address one
1207 // beyond the end of source range just processed. If necessary, switch to a
1208 // new source space and set src_space_id (in-out parameter) and src_space_top
1209 // (out parameter) accordingly.
1210 static size_t next_src_region(MoveAndUpdateClosure& closure,
1211 SpaceId& src_space_id,
1212 HeapWord*& src_space_top,
1213 HeapWord* end_addr);
1214
1215 // Decrement the destination count for each non-empty source region in the
1216 // range [beg_region, region(region_align_up(end_addr))). If the destination
1217 // count for a region goes to 0 and it needs to be filled, enqueue it.
1218 static void decrement_destination_counts(ParCompactionManager* cm,
1219 SpaceId src_space_id,
1220 size_t beg_region,
1221 HeapWord* end_addr);
1222
1223 static void fill_region(ParCompactionManager* cm, MoveAndUpdateClosure& closure, size_t region);
1224 static void fill_and_update_region(ParCompactionManager* cm, size_t region);
1225
1226 static bool steal_shadow_region(ParCompactionManager* cm, size_t& region_idx);
1227 static void fill_shadow_region(ParCompactionManager* cm, size_t region_idx);
1228 static void fill_and_update_shadow_region(ParCompactionManager* cm, size_t region) {
1229 fill_shadow_region(cm, region);
1230 }
1231 // Copy the content of a shadow region back to its corresponding heap region
1232 static void copy_back(HeapWord* shadow_addr, HeapWord* region_addr);
1233 // Initialize the steal record of a GC thread
1234 static void initialize_steal_record(uint which);
1235 // Reuse the empty heap regions as shadow regions, like to-space regions
1236 static void enqueue_shadow_region();
1237
1238 // Fill in the block table for the specified region.
1239 static void fill_blocks(size_t region_idx);
1240
1241 // Update the deferred objects in the space.
1242 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1243
1244 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1245 static ParallelCompactData& summary_data() { return _summary_data; }
1246
1247 // Reference Processing
1248 static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1249
1250 static STWGCTimer* gc_timer() { return &_gc_timer; }
1251
1252 // Return the SpaceId for the given address.
1253 static SpaceId space_id(HeapWord* addr);
1254
1255 // Time since last full gc (in milliseconds).
1256 static jlong millis_since_last_gc();
1257
1258 static void print_on_error(outputStream* st);
1259
1260 #ifndef PRODUCT
1261 // Debugging support.
1262 static const char* space_names[last_space_id];
1263 static void print_region_ranges();
1264 static void print_dense_prefix_stats(const char* const algorithm,
1265 const SpaceId id,
1266 const bool maximum_compaction,
1267 HeapWord* const addr);
1268 static void summary_phase_msg(SpaceId dst_space_id,
1269 HeapWord* dst_beg, HeapWord* dst_end,
1270 SpaceId src_space_id,
1271 HeapWord* src_beg, HeapWord* src_end);
1272 #endif // #ifndef PRODUCT
1273
1274 #ifdef ASSERT
1275 // Sanity check the new location of a word in the heap.
1276 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1277 // Verify that all the regions have been emptied.
1278 static void verify_complete(SpaceId space_id);
1279 #endif // #ifdef ASSERT
1280 };
1281
1282 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1283 static inline size_t calculate_words_remaining(size_t region);
1284 public:
1285 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1286 size_t region);
1287
1288 // Accessors.
1289 HeapWord* destination() const { return _destination; }
1290
1291 // If the object will fit (size <= words_remaining()), copy it to the current
1292 // destination, update the interior oops and the start array and return either
1293 // full (if the closure is full) or incomplete. If the object will not fit,
1294 // return would_overflow.
1295 IterationStatus do_addr(HeapWord* addr, size_t size);
1296
1297 // Copy enough words to fill this closure, starting at source(). Interior
1298 // oops and the start array are not updated. Return full.
1299 IterationStatus copy_until_full();
1300
1301 // Copy enough words to fill this closure or to the end of an object,
1302 // whichever is smaller, starting at source(). Interior oops and the start
1303 // array are not updated.
1304 void copy_partial_obj();
1305
1306 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr,
1307 PSParallelCompact::RegionData* region_ptr);
1308
1309 virtual void allocate_block();
1310
1311 virtual void set_deferred_obj_addr_for(PSParallelCompact::RegionData* region_ptr);
1312
1313 protected:
1314 // Update variables to indicate that word_count words were processed.
1315 inline void update_state(size_t word_count);
1316
1317 protected:
1318 HeapWord* _destination; // Next addr to be written.
1319 ObjectStartArray* const _start_array;
1320 };
1321
1322 inline size_t MoveAndUpdateClosure::calculate_words_remaining(size_t region) {
1323 HeapWord* dest_addr = PSParallelCompact::summary_data().region_to_addr(region);
1324 PSParallelCompact::SpaceId dest_space_id = PSParallelCompact::space_id(dest_addr);
1325 HeapWord* new_top = PSParallelCompact::new_top(dest_space_id);
1326 assert(dest_addr < new_top, "sanity");
1327
1328 return MIN2(pointer_delta(new_top, dest_addr), ParallelCompactData::RegionSize);
1329 }
1330
1331 inline
1332 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1333 ParCompactionManager* cm,
1334 size_t region_idx) :
1335 ParMarkBitMapClosure(bitmap, cm, calculate_words_remaining(region_idx)),
1336 _destination(PSParallelCompact::summary_data().region_to_addr(region_idx)),
1337 _start_array(PSParallelCompact::start_array(PSParallelCompact::space_id(_destination))) { }
1338
1339
1340 inline void MoveAndUpdateClosure::update_state(size_t words)
1341 {
1342 decrement_words_remaining(words);
1343 _source += words;
1344 _destination += words;
1345 }
1346
1347 class ShadowClosure: public MoveAndUpdateClosure {
1348 inline size_t calculate_shadow_offset(size_t region_idx, size_t shadow_idx);
1349 public:
1350 inline ShadowClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1351 size_t region, size_t shadow);
1352
1353 virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr,
1354 PSParallelCompact::RegionData* region_ptr);
1355
1356 virtual void allocate_block();
1357
1358 virtual void set_deferred_obj_addr_for(PSParallelCompact::RegionData* region_ptr);
1359
1360 private:
1361 size_t _shadow;
1362 size_t _offset;
1363 };
1364
1365 inline size_t ShadowClosure::calculate_shadow_offset(size_t region_idx, size_t shadow_idx) {
1366 ParallelCompactData& sd = PSParallelCompact::summary_data();
1367 HeapWord* dest_addr = sd.region_to_addr(region_idx);
1368 HeapWord* shadow_addr = sd.region_to_addr(shadow_idx);
1369 return pointer_delta(shadow_addr, dest_addr);
1370 }
1371
1372 inline
1373 ShadowClosure::ShadowClosure(ParMarkBitMap *bitmap,
1374 ParCompactionManager *cm,
1375 size_t region,
1376 size_t shadow) :
1377 MoveAndUpdateClosure(bitmap, cm, region),
1378 _shadow(shadow),
1379 _offset(calculate_shadow_offset(region, shadow)) {
1380 ParallelCompactData& sd = PSParallelCompact::summary_data();
1381 _destination = sd.region_to_addr(shadow);
1382 }
1383
1384 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1385 private:
1386 const PSParallelCompact::SpaceId _space_id;
1387 ObjectStartArray* const _start_array;
1388
1389 public:
1390 UpdateOnlyClosure(ParMarkBitMap* mbm,
1391 ParCompactionManager* cm,
1392 PSParallelCompact::SpaceId space_id);
1393
1394 // Update the object.
1395 virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1396
1397 inline void do_addr(HeapWord* addr);
1398 };
1399
1400 class FillClosure: public ParMarkBitMapClosure {
1401 public:
1402 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id);
1403
1404 virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1405
1406 private:
1407 ObjectStartArray* const _start_array;
1408 };
1409
1410 #endif // SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP