1 /*
2 * Copyright (c) 2001, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/cms/cmsLockVerifier.hpp"
27 #include "gc/cms/compactibleFreeListSpace.hpp"
28 #include "gc/cms/concurrentMarkSweepGeneration.inline.hpp"
29 #include "gc/cms/concurrentMarkSweepThread.hpp"
30 #include "gc/shared/blockOffsetTable.inline.hpp"
31 #include "gc/shared/collectedHeap.inline.hpp"
32 #include "gc/shared/genCollectedHeap.hpp"
33 #include "gc/shared/liveRange.hpp"
34 #include "gc/shared/space.inline.hpp"
35 #include "gc/shared/spaceDecorator.hpp"
36 #include "memory/allocation.inline.hpp"
37 #include "memory/resourceArea.hpp"
38 #include "memory/universe.inline.hpp"
39 #include "oops/oop.inline.hpp"
40 #include "runtime/globals.hpp"
41 #include "runtime/handles.inline.hpp"
42 #include "runtime/init.hpp"
43 #include "runtime/java.hpp"
44 #include "runtime/orderAccess.inline.hpp"
45 #include "runtime/vmThread.hpp"
46 #include "utilities/copy.hpp"
47
48 /////////////////////////////////////////////////////////////////////////
49 //// CompactibleFreeListSpace
50 /////////////////////////////////////////////////////////////////////////
51
52 // highest ranked free list lock rank
53 int CompactibleFreeListSpace::_lockRank = Mutex::leaf + 3;
54
55 // Defaults are 0 so things will break badly if incorrectly initialized.
56 size_t CompactibleFreeListSpace::IndexSetStart = 0;
57 size_t CompactibleFreeListSpace::IndexSetStride = 0;
58
59 size_t MinChunkSize = 0;
60
61 void CompactibleFreeListSpace::set_cms_values() {
62 // Set CMS global values
63 assert(MinChunkSize == 0, "already set");
64
65 // MinChunkSize should be a multiple of MinObjAlignment and be large enough
66 // for chunks to contain a FreeChunk.
67 size_t min_chunk_size_in_bytes = align_size_up(sizeof(FreeChunk), MinObjAlignmentInBytes);
68 MinChunkSize = min_chunk_size_in_bytes / BytesPerWord;
69
70 assert(IndexSetStart == 0 && IndexSetStride == 0, "already set");
71 IndexSetStart = MinChunkSize;
72 IndexSetStride = MinObjAlignment;
73 }
74
75 // Constructor
76 CompactibleFreeListSpace::CompactibleFreeListSpace(BlockOffsetSharedArray* bs, MemRegion mr) :
77 _bt(bs, mr),
78 // free list locks are in the range of values taken by _lockRank
79 // This range currently is [_leaf+2, _leaf+3]
80 // Note: this requires that CFLspace c'tors
81 // are called serially in the order in which the locks are
82 // are acquired in the program text. This is true today.
83 _freelistLock(_lockRank--, "CompactibleFreeListSpace._lock", true,
84 Monitor::_safepoint_check_sometimes),
85 _parDictionaryAllocLock(Mutex::leaf - 1, // == rank(ExpandHeap_lock) - 1
86 "CompactibleFreeListSpace._dict_par_lock", true,
87 Monitor::_safepoint_check_never),
88 _rescan_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
89 CMSRescanMultiple),
90 _marking_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
91 CMSConcMarkMultiple),
92 _collector(NULL),
93 _preconsumptionDirtyCardClosure(NULL)
94 {
95 assert(sizeof(FreeChunk) / BytesPerWord <= MinChunkSize,
96 "FreeChunk is larger than expected");
97 _bt.set_space(this);
98 initialize(mr, SpaceDecorator::Clear, SpaceDecorator::Mangle);
99
100 _dictionary = new AFLBinaryTreeDictionary(mr);
101
102 assert(_dictionary != NULL, "CMS dictionary initialization");
103 // The indexed free lists are initially all empty and are lazily
104 // filled in on demand. Initialize the array elements to NULL.
105 initializeIndexedFreeListArray();
106
107 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc,
108 SmallForLinearAlloc);
109
110 // CMSIndexedFreeListReplenish should be at least 1
111 CMSIndexedFreeListReplenish = MAX2((uintx)1, CMSIndexedFreeListReplenish);
112 _promoInfo.setSpace(this);
113 if (UseCMSBestFit) {
114 _fitStrategy = FreeBlockBestFitFirst;
115 } else {
116 _fitStrategy = FreeBlockStrategyNone;
117 }
118 check_free_list_consistency();
119
120 // Initialize locks for parallel case.
121 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
122 _indexedFreeListParLocks[i] = new Mutex(Mutex::leaf - 1, // == ExpandHeap_lock - 1
123 "a freelist par lock", true, Mutex::_safepoint_check_sometimes);
124 DEBUG_ONLY(
125 _indexedFreeList[i].set_protecting_lock(_indexedFreeListParLocks[i]);
126 )
127 }
128 _dictionary->set_par_lock(&_parDictionaryAllocLock);
129 }
130
131 // Like CompactibleSpace forward() but always calls cross_threshold() to
132 // update the block offset table. Removed initialize_threshold call because
133 // CFLS does not use a block offset array for contiguous spaces.
134 HeapWord* CompactibleFreeListSpace::forward(oop q, size_t size,
135 CompactPoint* cp, HeapWord* compact_top) {
136 // q is alive
137 // First check if we should switch compaction space
138 assert(this == cp->space, "'this' should be current compaction space.");
139 size_t compaction_max_size = pointer_delta(end(), compact_top);
140 assert(adjustObjectSize(size) == cp->space->adjust_object_size_v(size),
141 "virtual adjustObjectSize_v() method is not correct");
142 size_t adjusted_size = adjustObjectSize(size);
143 assert(compaction_max_size >= MinChunkSize || compaction_max_size == 0,
144 "no small fragments allowed");
145 assert(minimum_free_block_size() == MinChunkSize,
146 "for de-virtualized reference below");
147 // Can't leave a nonzero size, residual fragment smaller than MinChunkSize
148 if (adjusted_size + MinChunkSize > compaction_max_size &&
149 adjusted_size != compaction_max_size) {
150 do {
151 // switch to next compaction space
152 cp->space->set_compaction_top(compact_top);
153 cp->space = cp->space->next_compaction_space();
154 if (cp->space == NULL) {
155 cp->gen = GenCollectedHeap::heap()->young_gen();
156 assert(cp->gen != NULL, "compaction must succeed");
157 cp->space = cp->gen->first_compaction_space();
158 assert(cp->space != NULL, "generation must have a first compaction space");
159 }
160 compact_top = cp->space->bottom();
161 cp->space->set_compaction_top(compact_top);
162 // The correct adjusted_size may not be the same as that for this method
163 // (i.e., cp->space may no longer be "this" so adjust the size again.
164 // Use the virtual method which is not used above to save the virtual
165 // dispatch.
166 adjusted_size = cp->space->adjust_object_size_v(size);
167 compaction_max_size = pointer_delta(cp->space->end(), compact_top);
168 assert(cp->space->minimum_free_block_size() == 0, "just checking");
169 } while (adjusted_size > compaction_max_size);
170 }
171
172 // store the forwarding pointer into the mark word
173 if ((HeapWord*)q != compact_top) {
174 q->forward_to(oop(compact_top));
175 assert(q->is_gc_marked(), "encoding the pointer should preserve the mark");
176 } else {
177 // if the object isn't moving we can just set the mark to the default
178 // mark and handle it specially later on.
179 q->init_mark();
180 assert(q->forwardee() == NULL, "should be forwarded to NULL");
181 }
182
183 compact_top += adjusted_size;
184
185 // we need to update the offset table so that the beginnings of objects can be
186 // found during scavenge. Note that we are updating the offset table based on
187 // where the object will be once the compaction phase finishes.
188
189 // Always call cross_threshold(). A contiguous space can only call it when
190 // the compaction_top exceeds the current threshold but not for an
191 // non-contiguous space.
192 cp->threshold =
193 cp->space->cross_threshold(compact_top - adjusted_size, compact_top);
194 return compact_top;
195 }
196
197 // A modified copy of OffsetTableContigSpace::cross_threshold() with _offsets -> _bt
198 // and use of single_block instead of alloc_block. The name here is not really
199 // appropriate - maybe a more general name could be invented for both the
200 // contiguous and noncontiguous spaces.
201
202 HeapWord* CompactibleFreeListSpace::cross_threshold(HeapWord* start, HeapWord* the_end) {
203 _bt.single_block(start, the_end);
204 return end();
205 }
206
207 // Initialize them to NULL.
208 void CompactibleFreeListSpace::initializeIndexedFreeListArray() {
209 for (size_t i = 0; i < IndexSetSize; i++) {
210 // Note that on platforms where objects are double word aligned,
211 // the odd array elements are not used. It is convenient, however,
212 // to map directly from the object size to the array element.
213 _indexedFreeList[i].reset(IndexSetSize);
214 _indexedFreeList[i].set_size(i);
215 assert(_indexedFreeList[i].count() == 0, "reset check failed");
216 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
217 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
218 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
219 }
220 }
221
222 void CompactibleFreeListSpace::resetIndexedFreeListArray() {
223 for (size_t i = 1; i < IndexSetSize; i++) {
224 assert(_indexedFreeList[i].size() == (size_t) i,
225 "Indexed free list sizes are incorrect");
226 _indexedFreeList[i].reset(IndexSetSize);
227 assert(_indexedFreeList[i].count() == 0, "reset check failed");
228 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
229 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
230 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
231 }
232 }
233
234 void CompactibleFreeListSpace::reset(MemRegion mr) {
235 resetIndexedFreeListArray();
236 dictionary()->reset();
237 if (BlockOffsetArrayUseUnallocatedBlock) {
238 assert(end() == mr.end(), "We are compacting to the bottom of CMS gen");
239 // Everything's allocated until proven otherwise.
240 _bt.set_unallocated_block(end());
241 }
242 if (!mr.is_empty()) {
243 assert(mr.word_size() >= MinChunkSize, "Chunk size is too small");
244 _bt.single_block(mr.start(), mr.word_size());
245 FreeChunk* fc = (FreeChunk*) mr.start();
246 fc->set_size(mr.word_size());
247 if (mr.word_size() >= IndexSetSize ) {
248 returnChunkToDictionary(fc);
249 } else {
250 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
251 _indexedFreeList[mr.word_size()].return_chunk_at_head(fc);
252 }
253 coalBirth(mr.word_size());
254 }
255 _promoInfo.reset();
256 _smallLinearAllocBlock._ptr = NULL;
257 _smallLinearAllocBlock._word_size = 0;
258 }
259
260 void CompactibleFreeListSpace::reset_after_compaction() {
261 // Reset the space to the new reality - one free chunk.
262 MemRegion mr(compaction_top(), end());
263 reset(mr);
264 // Now refill the linear allocation block(s) if possible.
265 refillLinearAllocBlocksIfNeeded();
266 }
267
268 // Walks the entire dictionary, returning a coterminal
269 // chunk, if it exists. Use with caution since it involves
270 // a potentially complete walk of a potentially large tree.
271 FreeChunk* CompactibleFreeListSpace::find_chunk_at_end() {
272
273 assert_lock_strong(&_freelistLock);
274
275 return dictionary()->find_chunk_ends_at(end());
276 }
277
278
279 #ifndef PRODUCT
280 void CompactibleFreeListSpace::initializeIndexedFreeListArrayReturnedBytes() {
281 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
282 _indexedFreeList[i].allocation_stats()->set_returned_bytes(0);
283 }
284 }
285
286 size_t CompactibleFreeListSpace::sumIndexedFreeListArrayReturnedBytes() {
287 size_t sum = 0;
288 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
289 sum += _indexedFreeList[i].allocation_stats()->returned_bytes();
290 }
291 return sum;
292 }
293
294 size_t CompactibleFreeListSpace::totalCountInIndexedFreeLists() const {
295 size_t count = 0;
296 for (size_t i = IndexSetStart; i < IndexSetSize; i++) {
297 debug_only(
298 ssize_t total_list_count = 0;
299 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
300 fc = fc->next()) {
301 total_list_count++;
302 }
303 assert(total_list_count == _indexedFreeList[i].count(),
304 "Count in list is incorrect");
305 )
306 count += _indexedFreeList[i].count();
307 }
308 return count;
309 }
310
311 size_t CompactibleFreeListSpace::totalCount() {
312 size_t num = totalCountInIndexedFreeLists();
313 num += dictionary()->total_count();
314 if (_smallLinearAllocBlock._word_size != 0) {
315 num++;
316 }
317 return num;
318 }
319 #endif
320
321 bool CompactibleFreeListSpace::is_free_block(const HeapWord* p) const {
322 FreeChunk* fc = (FreeChunk*) p;
323 return fc->is_free();
324 }
325
326 size_t CompactibleFreeListSpace::used() const {
327 return capacity() - free();
328 }
329
330 size_t CompactibleFreeListSpace::free() const {
331 // "MT-safe, but not MT-precise"(TM), if you will: i.e.
332 // if you do this while the structures are in flux you
333 // may get an approximate answer only; for instance
334 // because there is concurrent allocation either
335 // directly by mutators or for promotion during a GC.
336 // It's "MT-safe", however, in the sense that you are guaranteed
337 // not to crash and burn, for instance, because of walking
338 // pointers that could disappear as you were walking them.
339 // The approximation is because the various components
340 // that are read below are not read atomically (and
341 // further the computation of totalSizeInIndexedFreeLists()
342 // is itself a non-atomic computation. The normal use of
343 // this is during a resize operation at the end of GC
344 // and at that time you are guaranteed to get the
345 // correct actual value. However, for instance, this is
346 // also read completely asynchronously by the "perf-sampler"
347 // that supports jvmstat, and you are apt to see the values
348 // flicker in such cases.
349 assert(_dictionary != NULL, "No _dictionary?");
350 return (_dictionary->total_chunk_size(DEBUG_ONLY(freelistLock())) +
351 totalSizeInIndexedFreeLists() +
352 _smallLinearAllocBlock._word_size) * HeapWordSize;
353 }
354
355 size_t CompactibleFreeListSpace::max_alloc_in_words() const {
356 assert(_dictionary != NULL, "No _dictionary?");
357 assert_locked();
358 size_t res = _dictionary->max_chunk_size();
359 res = MAX2(res, MIN2(_smallLinearAllocBlock._word_size,
360 (size_t) SmallForLinearAlloc - 1));
361 // XXX the following could potentially be pretty slow;
362 // should one, pessimistically for the rare cases when res
363 // calculated above is less than IndexSetSize,
364 // just return res calculated above? My reasoning was that
365 // those cases will be so rare that the extra time spent doesn't
366 // really matter....
367 // Note: do not change the loop test i >= res + IndexSetStride
368 // to i > res below, because i is unsigned and res may be zero.
369 for (size_t i = IndexSetSize - 1; i >= res + IndexSetStride;
370 i -= IndexSetStride) {
371 if (_indexedFreeList[i].head() != NULL) {
372 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
373 return i;
374 }
375 }
376 return res;
377 }
378
379 void LinearAllocBlock::print_on(outputStream* st) const {
380 st->print_cr(" LinearAllocBlock: ptr = " PTR_FORMAT ", word_size = " SIZE_FORMAT
381 ", refillsize = " SIZE_FORMAT ", allocation_size_limit = " SIZE_FORMAT,
382 p2i(_ptr), _word_size, _refillSize, _allocation_size_limit);
383 }
384
385 void CompactibleFreeListSpace::print_on(outputStream* st) const {
386 st->print_cr("COMPACTIBLE FREELIST SPACE");
387 st->print_cr(" Space:");
388 Space::print_on(st);
389
390 st->print_cr("promoInfo:");
391 _promoInfo.print_on(st);
392
393 st->print_cr("_smallLinearAllocBlock");
394 _smallLinearAllocBlock.print_on(st);
395
396 // dump_memory_block(_smallLinearAllocBlock->_ptr, 128);
397
398 st->print_cr(" _fitStrategy = %s", BOOL_TO_STR(_fitStrategy));
399 }
400
401 void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st)
402 const {
403 reportIndexedFreeListStatistics(st);
404 st->print_cr("Layout of Indexed Freelists");
405 st->print_cr("---------------------------");
406 AdaptiveFreeList<FreeChunk>::print_labels_on(st, "size");
407 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
408 _indexedFreeList[i].print_on(st);
409 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; fc = fc->next()) {
410 st->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s",
411 p2i(fc), p2i((HeapWord*)fc + i),
412 fc->cantCoalesce() ? "\t CC" : "");
413 }
414 }
415 }
416
417 void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st)
418 const {
419 _promoInfo.print_on(st);
420 }
421
422 void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st)
423 const {
424 _dictionary->report_statistics(st);
425 st->print_cr("Layout of Freelists in Tree");
426 st->print_cr("---------------------------");
427 _dictionary->print_free_lists(st);
428 }
429
430 class BlkPrintingClosure: public BlkClosure {
431 const CMSCollector* _collector;
432 const CompactibleFreeListSpace* _sp;
433 const CMSBitMap* _live_bit_map;
434 const bool _post_remark;
435 outputStream* _st;
436 public:
437 BlkPrintingClosure(const CMSCollector* collector,
438 const CompactibleFreeListSpace* sp,
439 const CMSBitMap* live_bit_map,
440 outputStream* st):
441 _collector(collector),
442 _sp(sp),
443 _live_bit_map(live_bit_map),
444 _post_remark(collector->abstract_state() > CMSCollector::FinalMarking),
445 _st(st) { }
446 size_t do_blk(HeapWord* addr);
447 };
448
449 size_t BlkPrintingClosure::do_blk(HeapWord* addr) {
450 size_t sz = _sp->block_size_no_stall(addr, _collector);
451 assert(sz != 0, "Should always be able to compute a size");
452 if (_sp->block_is_obj(addr)) {
453 const bool dead = _post_remark && !_live_bit_map->isMarked(addr);
454 _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s",
455 p2i(addr),
456 dead ? "dead" : "live",
457 sz,
458 (!dead && CMSPrintObjectsInDump) ? ":" : ".");
459 if (CMSPrintObjectsInDump && !dead) {
460 oop(addr)->print_on(_st);
461 _st->print_cr("--------------------------------------");
462 }
463 } else { // free block
464 _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s",
465 p2i(addr), sz, CMSPrintChunksInDump ? ":" : ".");
466 if (CMSPrintChunksInDump) {
467 ((FreeChunk*)addr)->print_on(_st);
468 _st->print_cr("--------------------------------------");
469 }
470 }
471 return sz;
472 }
473
474 void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c, outputStream* st) {
475 st->print_cr("=========================");
476 st->print_cr("Block layout in CMS Heap:");
477 st->print_cr("=========================");
478 BlkPrintingClosure bpcl(c, this, c->markBitMap(), st);
479 blk_iterate(&bpcl);
480
481 st->print_cr("=======================================");
482 st->print_cr("Order & Layout of Promotion Info Blocks");
483 st->print_cr("=======================================");
484 print_promo_info_blocks(st);
485
486 st->print_cr("===========================");
487 st->print_cr("Order of Indexed Free Lists");
488 st->print_cr("=========================");
489 print_indexed_free_lists(st);
490
491 st->print_cr("=================================");
492 st->print_cr("Order of Free Lists in Dictionary");
493 st->print_cr("=================================");
494 print_dictionary_free_lists(st);
495 }
496
497
498 void CompactibleFreeListSpace::reportFreeListStatistics(const char* title) const {
499 assert_lock_strong(&_freelistLock);
500 LogHandle(gc, freelist, stats) log;
501 if (!log.is_debug()) {
502 return;
503 }
504 log.debug("%s", title);
505 _dictionary->report_statistics(log.debug_stream());
506 if (log.is_trace()) {
507 ResourceMark rm;
508 reportIndexedFreeListStatistics(log.trace_stream());
509 size_t total_size = totalSizeInIndexedFreeLists() +
510 _dictionary->total_chunk_size(DEBUG_ONLY(freelistLock()));
511 log.trace(" free=" SIZE_FORMAT " frag=%1.4f", total_size, flsFrag());
512 }
513 }
514
515 void CompactibleFreeListSpace::reportIndexedFreeListStatistics(outputStream* st) const {
516 assert_lock_strong(&_freelistLock);
517 st->print_cr("Statistics for IndexedFreeLists:");
518 st->print_cr("--------------------------------");
519 size_t total_size = totalSizeInIndexedFreeLists();
520 size_t free_blocks = numFreeBlocksInIndexedFreeLists();
521 st->print_cr("Total Free Space: " SIZE_FORMAT, total_size);
522 st->print_cr("Max Chunk Size: " SIZE_FORMAT, maxChunkSizeInIndexedFreeLists());
523 st->print_cr("Number of Blocks: " SIZE_FORMAT, free_blocks);
524 if (free_blocks != 0) {
525 st->print_cr("Av. Block Size: " SIZE_FORMAT, total_size/free_blocks);
526 }
527 }
528
529 size_t CompactibleFreeListSpace::numFreeBlocksInIndexedFreeLists() const {
530 size_t res = 0;
531 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
532 debug_only(
533 ssize_t recount = 0;
534 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
535 fc = fc->next()) {
536 recount += 1;
537 }
538 assert(recount == _indexedFreeList[i].count(),
539 "Incorrect count in list");
540 )
541 res += _indexedFreeList[i].count();
542 }
543 return res;
544 }
545
546 size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const {
547 for (size_t i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
548 if (_indexedFreeList[i].head() != NULL) {
549 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
550 return (size_t)i;
551 }
552 }
553 return 0;
554 }
555
556 void CompactibleFreeListSpace::set_end(HeapWord* value) {
557 HeapWord* prevEnd = end();
558 assert(prevEnd != value, "unnecessary set_end call");
559 assert(prevEnd == NULL || !BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
560 "New end is below unallocated block");
561 _end = value;
562 if (prevEnd != NULL) {
563 // Resize the underlying block offset table.
564 _bt.resize(pointer_delta(value, bottom()));
565 if (value <= prevEnd) {
566 assert(!BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
567 "New end is below unallocated block");
568 } else {
569 // Now, take this new chunk and add it to the free blocks.
570 // Note that the BOT has not yet been updated for this block.
571 size_t newFcSize = pointer_delta(value, prevEnd);
572 // Add the block to the free lists, if possible coalescing it
573 // with the last free block, and update the BOT and census data.
574 addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize);
575 }
576 }
577 }
578
579 class FreeListSpace_DCTOC : public Filtering_DCTOC {
580 CompactibleFreeListSpace* _cfls;
581 CMSCollector* _collector;
582 bool _parallel;
583 protected:
584 // Override.
585 #define walk_mem_region_with_cl_DECL(ClosureType) \
586 virtual void walk_mem_region_with_cl(MemRegion mr, \
587 HeapWord* bottom, HeapWord* top, \
588 ClosureType* cl); \
589 void walk_mem_region_with_cl_par(MemRegion mr, \
590 HeapWord* bottom, HeapWord* top, \
591 ClosureType* cl); \
592 void walk_mem_region_with_cl_nopar(MemRegion mr, \
593 HeapWord* bottom, HeapWord* top, \
594 ClosureType* cl)
595 walk_mem_region_with_cl_DECL(ExtendedOopClosure);
596 walk_mem_region_with_cl_DECL(FilteringClosure);
597
598 public:
599 FreeListSpace_DCTOC(CompactibleFreeListSpace* sp,
600 CMSCollector* collector,
601 ExtendedOopClosure* cl,
602 CardTableModRefBS::PrecisionStyle precision,
603 HeapWord* boundary,
604 bool parallel) :
605 Filtering_DCTOC(sp, cl, precision, boundary),
606 _cfls(sp), _collector(collector), _parallel(parallel) {}
607 };
608
609 // We de-virtualize the block-related calls below, since we know that our
610 // space is a CompactibleFreeListSpace.
611
612 #define FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ClosureType) \
613 void FreeListSpace_DCTOC::walk_mem_region_with_cl(MemRegion mr, \
614 HeapWord* bottom, \
615 HeapWord* top, \
616 ClosureType* cl) { \
617 if (_parallel) { \
618 walk_mem_region_with_cl_par(mr, bottom, top, cl); \
619 } else { \
620 walk_mem_region_with_cl_nopar(mr, bottom, top, cl); \
621 } \
622 } \
623 void FreeListSpace_DCTOC::walk_mem_region_with_cl_par(MemRegion mr, \
624 HeapWord* bottom, \
625 HeapWord* top, \
626 ClosureType* cl) { \
627 /* Skip parts that are before "mr", in case "block_start" sent us \
628 back too far. */ \
629 HeapWord* mr_start = mr.start(); \
630 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
631 HeapWord* next = bottom + bot_size; \
632 while (next < mr_start) { \
633 bottom = next; \
634 bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
635 next = bottom + bot_size; \
636 } \
637 \
638 while (bottom < top) { \
639 if (_cfls->CompactibleFreeListSpace::block_is_obj(bottom) && \
640 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
641 oop(bottom)) && \
642 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
643 size_t word_sz = oop(bottom)->oop_iterate_size(cl, mr); \
644 bottom += _cfls->adjustObjectSize(word_sz); \
645 } else { \
646 bottom += _cfls->CompactibleFreeListSpace::block_size(bottom); \
647 } \
648 } \
649 } \
650 void FreeListSpace_DCTOC::walk_mem_region_with_cl_nopar(MemRegion mr, \
651 HeapWord* bottom, \
652 HeapWord* top, \
653 ClosureType* cl) { \
654 /* Skip parts that are before "mr", in case "block_start" sent us \
655 back too far. */ \
656 HeapWord* mr_start = mr.start(); \
657 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
658 HeapWord* next = bottom + bot_size; \
659 while (next < mr_start) { \
660 bottom = next; \
661 bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
662 next = bottom + bot_size; \
663 } \
664 \
665 while (bottom < top) { \
666 if (_cfls->CompactibleFreeListSpace::block_is_obj_nopar(bottom) && \
667 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
668 oop(bottom)) && \
669 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
670 size_t word_sz = oop(bottom)->oop_iterate_size(cl, mr); \
671 bottom += _cfls->adjustObjectSize(word_sz); \
672 } else { \
673 bottom += _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
674 } \
675 } \
676 }
677
678 // (There are only two of these, rather than N, because the split is due
679 // only to the introduction of the FilteringClosure, a local part of the
680 // impl of this abstraction.)
681 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ExtendedOopClosure)
682 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(FilteringClosure)
683
684 DirtyCardToOopClosure*
685 CompactibleFreeListSpace::new_dcto_cl(ExtendedOopClosure* cl,
686 CardTableModRefBS::PrecisionStyle precision,
687 HeapWord* boundary,
688 bool parallel) {
689 return new FreeListSpace_DCTOC(this, _collector, cl, precision, boundary, parallel);
690 }
691
692
693 // Note on locking for the space iteration functions:
694 // since the collector's iteration activities are concurrent with
695 // allocation activities by mutators, absent a suitable mutual exclusion
696 // mechanism the iterators may go awry. For instance a block being iterated
697 // may suddenly be allocated or divided up and part of it allocated and
698 // so on.
699
700 // Apply the given closure to each block in the space.
701 void CompactibleFreeListSpace::blk_iterate_careful(BlkClosureCareful* cl) {
702 assert_lock_strong(freelistLock());
703 HeapWord *cur, *limit;
704 for (cur = bottom(), limit = end(); cur < limit;
705 cur += cl->do_blk_careful(cur));
706 }
707
708 // Apply the given closure to each block in the space.
709 void CompactibleFreeListSpace::blk_iterate(BlkClosure* cl) {
710 assert_lock_strong(freelistLock());
711 HeapWord *cur, *limit;
712 for (cur = bottom(), limit = end(); cur < limit;
713 cur += cl->do_blk(cur));
714 }
715
716 // Apply the given closure to each oop in the space.
717 void CompactibleFreeListSpace::oop_iterate(ExtendedOopClosure* cl) {
718 assert_lock_strong(freelistLock());
719 HeapWord *cur, *limit;
720 size_t curSize;
721 for (cur = bottom(), limit = end(); cur < limit;
722 cur += curSize) {
723 curSize = block_size(cur);
724 if (block_is_obj(cur)) {
725 oop(cur)->oop_iterate(cl);
726 }
727 }
728 }
729
730 // NOTE: In the following methods, in order to safely be able to
731 // apply the closure to an object, we need to be sure that the
732 // object has been initialized. We are guaranteed that an object
733 // is initialized if we are holding the Heap_lock with the
734 // world stopped.
735 void CompactibleFreeListSpace::verify_objects_initialized() const {
736 if (is_init_completed()) {
737 assert_locked_or_safepoint(Heap_lock);
738 if (Universe::is_fully_initialized()) {
739 guarantee(SafepointSynchronize::is_at_safepoint(),
740 "Required for objects to be initialized");
741 }
742 } // else make a concession at vm start-up
743 }
744
745 // Apply the given closure to each object in the space
746 void CompactibleFreeListSpace::object_iterate(ObjectClosure* blk) {
747 assert_lock_strong(freelistLock());
748 NOT_PRODUCT(verify_objects_initialized());
749 HeapWord *cur, *limit;
750 size_t curSize;
751 for (cur = bottom(), limit = end(); cur < limit;
752 cur += curSize) {
753 curSize = block_size(cur);
754 if (block_is_obj(cur)) {
755 blk->do_object(oop(cur));
756 }
757 }
758 }
759
760 // Apply the given closure to each live object in the space
761 // The usage of CompactibleFreeListSpace
762 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
763 // objects in the space with references to objects that are no longer
764 // valid. For example, an object may reference another object
765 // that has already been sweep up (collected). This method uses
766 // obj_is_alive() to determine whether it is safe to apply the closure to
767 // an object. See obj_is_alive() for details on how liveness of an
768 // object is decided.
769
770 void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) {
771 assert_lock_strong(freelistLock());
772 NOT_PRODUCT(verify_objects_initialized());
773 HeapWord *cur, *limit;
774 size_t curSize;
775 for (cur = bottom(), limit = end(); cur < limit;
776 cur += curSize) {
777 curSize = block_size(cur);
778 if (block_is_obj(cur) && obj_is_alive(cur)) {
779 blk->do_object(oop(cur));
780 }
781 }
782 }
783
784 void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr,
785 UpwardsObjectClosure* cl) {
786 assert_locked(freelistLock());
787 NOT_PRODUCT(verify_objects_initialized());
788 assert(!mr.is_empty(), "Should be non-empty");
789 // We use MemRegion(bottom(), end()) rather than used_region() below
790 // because the two are not necessarily equal for some kinds of
791 // spaces, in particular, certain kinds of free list spaces.
792 // We could use the more complicated but more precise:
793 // MemRegion(used_region().start(), round_to(used_region().end(), CardSize))
794 // but the slight imprecision seems acceptable in the assertion check.
795 assert(MemRegion(bottom(), end()).contains(mr),
796 "Should be within used space");
797 HeapWord* prev = cl->previous(); // max address from last time
798 if (prev >= mr.end()) { // nothing to do
799 return;
800 }
801 // This assert will not work when we go from cms space to perm
802 // space, and use same closure. Easy fix deferred for later. XXX YSR
803 // assert(prev == NULL || contains(prev), "Should be within space");
804
805 bool last_was_obj_array = false;
806 HeapWord *blk_start_addr, *region_start_addr;
807 if (prev > mr.start()) {
808 region_start_addr = prev;
809 blk_start_addr = prev;
810 // The previous invocation may have pushed "prev" beyond the
811 // last allocated block yet there may be still be blocks
812 // in this region due to a particular coalescing policy.
813 // Relax the assertion so that the case where the unallocated
814 // block is maintained and "prev" is beyond the unallocated
815 // block does not cause the assertion to fire.
816 assert((BlockOffsetArrayUseUnallocatedBlock &&
817 (!is_in(prev))) ||
818 (blk_start_addr == block_start(region_start_addr)), "invariant");
819 } else {
820 region_start_addr = mr.start();
821 blk_start_addr = block_start(region_start_addr);
822 }
823 HeapWord* region_end_addr = mr.end();
824 MemRegion derived_mr(region_start_addr, region_end_addr);
825 while (blk_start_addr < region_end_addr) {
826 const size_t size = block_size(blk_start_addr);
827 if (block_is_obj(blk_start_addr)) {
828 last_was_obj_array = cl->do_object_bm(oop(blk_start_addr), derived_mr);
829 } else {
830 last_was_obj_array = false;
831 }
832 blk_start_addr += size;
833 }
834 if (!last_was_obj_array) {
835 assert((bottom() <= blk_start_addr) && (blk_start_addr <= end()),
836 "Should be within (closed) used space");
837 assert(blk_start_addr > prev, "Invariant");
838 cl->set_previous(blk_start_addr); // min address for next time
839 }
840 }
841
842 // Callers of this iterator beware: The closure application should
843 // be robust in the face of uninitialized objects and should (always)
844 // return a correct size so that the next addr + size below gives us a
845 // valid block boundary. [See for instance,
846 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
847 // in ConcurrentMarkSweepGeneration.cpp.]
848 HeapWord*
849 CompactibleFreeListSpace::object_iterate_careful_m(MemRegion mr,
850 ObjectClosureCareful* cl) {
851 assert_lock_strong(freelistLock());
852 // Can't use used_region() below because it may not necessarily
853 // be the same as [bottom(),end()); although we could
854 // use [used_region().start(),round_to(used_region().end(),CardSize)),
855 // that appears too cumbersome, so we just do the simpler check
856 // in the assertion below.
857 assert(!mr.is_empty() && MemRegion(bottom(),end()).contains(mr),
858 "mr should be non-empty and within used space");
859 HeapWord *addr, *end;
860 size_t size;
861 for (addr = block_start_careful(mr.start()), end = mr.end();
862 addr < end; addr += size) {
863 FreeChunk* fc = (FreeChunk*)addr;
864 if (fc->is_free()) {
865 // Since we hold the free list lock, which protects direct
866 // allocation in this generation by mutators, a free object
867 // will remain free throughout this iteration code.
868 size = fc->size();
869 } else {
870 // Note that the object need not necessarily be initialized,
871 // because (for instance) the free list lock does NOT protect
872 // object initialization. The closure application below must
873 // therefore be correct in the face of uninitialized objects.
874 size = cl->do_object_careful_m(oop(addr), mr);
875 if (size == 0) {
876 // An unparsable object found. Signal early termination.
877 return addr;
878 }
879 }
880 }
881 return NULL;
882 }
883
884
885 HeapWord* CompactibleFreeListSpace::block_start_const(const void* p) const {
886 NOT_PRODUCT(verify_objects_initialized());
887 return _bt.block_start(p);
888 }
889
890 HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const {
891 return _bt.block_start_careful(p);
892 }
893
894 size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const {
895 NOT_PRODUCT(verify_objects_initialized());
896 // This must be volatile, or else there is a danger that the compiler
897 // will compile the code below into a sometimes-infinite loop, by keeping
898 // the value read the first time in a register.
899 while (true) {
900 // We must do this until we get a consistent view of the object.
901 if (FreeChunk::indicatesFreeChunk(p)) {
902 volatile FreeChunk* fc = (volatile FreeChunk*)p;
903 size_t res = fc->size();
904
905 // Bugfix for systems with weak memory model (PPC64/IA64). The
906 // block's free bit was set and we have read the size of the
907 // block. Acquire and check the free bit again. If the block is
908 // still free, the read size is correct.
909 OrderAccess::acquire();
910
911 // If the object is still a free chunk, return the size, else it
912 // has been allocated so try again.
913 if (FreeChunk::indicatesFreeChunk(p)) {
914 assert(res != 0, "Block size should not be 0");
915 return res;
916 }
917 } else {
918 // must read from what 'p' points to in each loop.
919 Klass* k = ((volatile oopDesc*)p)->klass_or_null();
920 if (k != NULL) {
921 assert(k->is_klass(), "Should really be klass oop.");
922 oop o = (oop)p;
923 assert(o->is_oop(true /* ignore mark word */), "Should be an oop.");
924
925 // Bugfix for systems with weak memory model (PPC64/IA64).
926 // The object o may be an array. Acquire to make sure that the array
927 // size (third word) is consistent.
928 OrderAccess::acquire();
929
930 size_t res = o->size_given_klass(k);
931 res = adjustObjectSize(res);
932 assert(res != 0, "Block size should not be 0");
933 return res;
934 }
935 }
936 }
937 }
938
939 // TODO: Now that is_parsable is gone, we should combine these two functions.
940 // A variant of the above that uses the Printezis bits for
941 // unparsable but allocated objects. This avoids any possible
942 // stalls waiting for mutators to initialize objects, and is
943 // thus potentially faster than the variant above. However,
944 // this variant may return a zero size for a block that is
945 // under mutation and for which a consistent size cannot be
946 // inferred without stalling; see CMSCollector::block_size_if_printezis_bits().
947 size_t CompactibleFreeListSpace::block_size_no_stall(HeapWord* p,
948 const CMSCollector* c)
949 const {
950 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
951 // This must be volatile, or else there is a danger that the compiler
952 // will compile the code below into a sometimes-infinite loop, by keeping
953 // the value read the first time in a register.
954 DEBUG_ONLY(uint loops = 0;)
955 while (true) {
956 // We must do this until we get a consistent view of the object.
957 if (FreeChunk::indicatesFreeChunk(p)) {
958 volatile FreeChunk* fc = (volatile FreeChunk*)p;
959 size_t res = fc->size();
960
961 // Bugfix for systems with weak memory model (PPC64/IA64). The
962 // free bit of the block was set and we have read the size of
963 // the block. Acquire and check the free bit again. If the
964 // block is still free, the read size is correct.
965 OrderAccess::acquire();
966
967 if (FreeChunk::indicatesFreeChunk(p)) {
968 assert(res != 0, "Block size should not be 0");
969 assert(loops == 0, "Should be 0");
970 return res;
971 }
972 } else {
973 // must read from what 'p' points to in each loop.
974 Klass* k = ((volatile oopDesc*)p)->klass_or_null();
975 // We trust the size of any object that has a non-NULL
976 // klass and (for those in the perm gen) is parsable
977 // -- irrespective of its conc_safe-ty.
978 if (k != NULL) {
979 assert(k->is_klass(), "Should really be klass oop.");
980 oop o = (oop)p;
981 assert(o->is_oop(), "Should be an oop");
982
983 // Bugfix for systems with weak memory model (PPC64/IA64).
984 // The object o may be an array. Acquire to make sure that the array
985 // size (third word) is consistent.
986 OrderAccess::acquire();
987
988 size_t res = o->size_given_klass(k);
989 res = adjustObjectSize(res);
990 assert(res != 0, "Block size should not be 0");
991 return res;
992 } else {
993 // May return 0 if P-bits not present.
994 return c->block_size_if_printezis_bits(p);
995 }
996 }
997 assert(loops == 0, "Can loop at most once");
998 DEBUG_ONLY(loops++;)
999 }
1000 }
1001
1002 size_t CompactibleFreeListSpace::block_size_nopar(const HeapWord* p) const {
1003 NOT_PRODUCT(verify_objects_initialized());
1004 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1005 FreeChunk* fc = (FreeChunk*)p;
1006 if (fc->is_free()) {
1007 return fc->size();
1008 } else {
1009 // Ignore mark word because this may be a recently promoted
1010 // object whose mark word is used to chain together grey
1011 // objects (the last one would have a null value).
1012 assert(oop(p)->is_oop(true), "Should be an oop");
1013 return adjustObjectSize(oop(p)->size());
1014 }
1015 }
1016
1017 // This implementation assumes that the property of "being an object" is
1018 // stable. But being a free chunk may not be (because of parallel
1019 // promotion.)
1020 bool CompactibleFreeListSpace::block_is_obj(const HeapWord* p) const {
1021 FreeChunk* fc = (FreeChunk*)p;
1022 assert(is_in_reserved(p), "Should be in space");
1023 if (FreeChunk::indicatesFreeChunk(p)) return false;
1024 Klass* k = oop(p)->klass_or_null();
1025 if (k != NULL) {
1026 // Ignore mark word because it may have been used to
1027 // chain together promoted objects (the last one
1028 // would have a null value).
1029 assert(oop(p)->is_oop(true), "Should be an oop");
1030 return true;
1031 } else {
1032 return false; // Was not an object at the start of collection.
1033 }
1034 }
1035
1036 // Check if the object is alive. This fact is checked either by consulting
1037 // the main marking bitmap in the sweeping phase or, if it's a permanent
1038 // generation and we're not in the sweeping phase, by checking the
1039 // perm_gen_verify_bit_map where we store the "deadness" information if
1040 // we did not sweep the perm gen in the most recent previous GC cycle.
1041 bool CompactibleFreeListSpace::obj_is_alive(const HeapWord* p) const {
1042 assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(),
1043 "Else races are possible");
1044 assert(block_is_obj(p), "The address should point to an object");
1045
1046 // If we're sweeping, we use object liveness information from the main bit map
1047 // for both perm gen and old gen.
1048 // We don't need to lock the bitmap (live_map or dead_map below), because
1049 // EITHER we are in the middle of the sweeping phase, and the
1050 // main marking bit map (live_map below) is locked,
1051 // OR we're in other phases and perm_gen_verify_bit_map (dead_map below)
1052 // is stable, because it's mutated only in the sweeping phase.
1053 // NOTE: This method is also used by jmap where, if class unloading is
1054 // off, the results can return "false" for legitimate perm objects,
1055 // when we are not in the midst of a sweeping phase, which can result
1056 // in jmap not reporting certain perm gen objects. This will be moot
1057 // if/when the perm gen goes away in the future.
1058 if (_collector->abstract_state() == CMSCollector::Sweeping) {
1059 CMSBitMap* live_map = _collector->markBitMap();
1060 return live_map->par_isMarked((HeapWord*) p);
1061 }
1062 return true;
1063 }
1064
1065 bool CompactibleFreeListSpace::block_is_obj_nopar(const HeapWord* p) const {
1066 FreeChunk* fc = (FreeChunk*)p;
1067 assert(is_in_reserved(p), "Should be in space");
1068 assert(_bt.block_start(p) == p, "Should be a block boundary");
1069 if (!fc->is_free()) {
1070 // Ignore mark word because it may have been used to
1071 // chain together promoted objects (the last one
1072 // would have a null value).
1073 assert(oop(p)->is_oop(true), "Should be an oop");
1074 return true;
1075 }
1076 return false;
1077 }
1078
1079 // "MT-safe but not guaranteed MT-precise" (TM); you may get an
1080 // approximate answer if you don't hold the freelistlock when you call this.
1081 size_t CompactibleFreeListSpace::totalSizeInIndexedFreeLists() const {
1082 size_t size = 0;
1083 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
1084 debug_only(
1085 // We may be calling here without the lock in which case we
1086 // won't do this modest sanity check.
1087 if (freelistLock()->owned_by_self()) {
1088 size_t total_list_size = 0;
1089 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
1090 fc = fc->next()) {
1091 total_list_size += i;
1092 }
1093 assert(total_list_size == i * _indexedFreeList[i].count(),
1094 "Count in list is incorrect");
1095 }
1096 )
1097 size += i * _indexedFreeList[i].count();
1098 }
1099 return size;
1100 }
1101
1102 HeapWord* CompactibleFreeListSpace::par_allocate(size_t size) {
1103 MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
1104 return allocate(size);
1105 }
1106
1107 HeapWord*
1108 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlockRemainder(size_t size) {
1109 return getChunkFromLinearAllocBlockRemainder(&_smallLinearAllocBlock, size);
1110 }
1111
1112 HeapWord* CompactibleFreeListSpace::allocate(size_t size) {
1113 assert_lock_strong(freelistLock());
1114 HeapWord* res = NULL;
1115 assert(size == adjustObjectSize(size),
1116 "use adjustObjectSize() before calling into allocate()");
1117
1118 res = allocate_adaptive_freelists(size);
1119
1120 if (res != NULL) {
1121 // check that res does lie in this space!
1122 assert(is_in_reserved(res), "Not in this space!");
1123 assert(is_aligned((void*)res), "alignment check");
1124
1125 FreeChunk* fc = (FreeChunk*)res;
1126 fc->markNotFree();
1127 assert(!fc->is_free(), "shouldn't be marked free");
1128 assert(oop(fc)->klass_or_null() == NULL, "should look uninitialized");
1129 // Verify that the block offset table shows this to
1130 // be a single block, but not one which is unallocated.
1131 _bt.verify_single_block(res, size);
1132 _bt.verify_not_unallocated(res, size);
1133 // mangle a just allocated object with a distinct pattern.
1134 debug_only(fc->mangleAllocated(size));
1135 }
1136
1137 return res;
1138 }
1139
1140 HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) {
1141 assert_lock_strong(freelistLock());
1142 HeapWord* res = NULL;
1143 assert(size == adjustObjectSize(size),
1144 "use adjustObjectSize() before calling into allocate()");
1145
1146 // Strategy
1147 // if small
1148 // exact size from small object indexed list if small
1149 // small or large linear allocation block (linAB) as appropriate
1150 // take from lists of greater sized chunks
1151 // else
1152 // dictionary
1153 // small or large linear allocation block if it has the space
1154 // Try allocating exact size from indexTable first
1155 if (size < IndexSetSize) {
1156 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1157 if(res != NULL) {
1158 assert(res != (HeapWord*)_indexedFreeList[size].head(),
1159 "Not removed from free list");
1160 // no block offset table adjustment is necessary on blocks in
1161 // the indexed lists.
1162
1163 // Try allocating from the small LinAB
1164 } else if (size < _smallLinearAllocBlock._allocation_size_limit &&
1165 (res = getChunkFromSmallLinearAllocBlock(size)) != NULL) {
1166 // if successful, the above also adjusts block offset table
1167 // Note that this call will refill the LinAB to
1168 // satisfy the request. This is different that
1169 // evm.
1170 // Don't record chunk off a LinAB? smallSplitBirth(size);
1171 } else {
1172 // Raid the exact free lists larger than size, even if they are not
1173 // overpopulated.
1174 res = (HeapWord*) getChunkFromGreater(size);
1175 }
1176 } else {
1177 // Big objects get allocated directly from the dictionary.
1178 res = (HeapWord*) getChunkFromDictionaryExact(size);
1179 if (res == NULL) {
1180 // Try hard not to fail since an allocation failure will likely
1181 // trigger a synchronous GC. Try to get the space from the
1182 // allocation blocks.
1183 res = getChunkFromSmallLinearAllocBlockRemainder(size);
1184 }
1185 }
1186
1187 return res;
1188 }
1189
1190 // A worst-case estimate of the space required (in HeapWords) to expand the heap
1191 // when promoting obj.
1192 size_t CompactibleFreeListSpace::expansionSpaceRequired(size_t obj_size) const {
1193 // Depending on the object size, expansion may require refilling either a
1194 // bigLAB or a smallLAB plus refilling a PromotionInfo object. MinChunkSize
1195 // is added because the dictionary may over-allocate to avoid fragmentation.
1196 size_t space = obj_size;
1197 space += _promoInfo.refillSize() + 2 * MinChunkSize;
1198 return space;
1199 }
1200
1201 FreeChunk* CompactibleFreeListSpace::getChunkFromGreater(size_t numWords) {
1202 FreeChunk* ret;
1203
1204 assert(numWords >= MinChunkSize, "Size is less than minimum");
1205 assert(linearAllocationWouldFail() || bestFitFirst(),
1206 "Should not be here");
1207
1208 size_t i;
1209 size_t currSize = numWords + MinChunkSize;
1210 assert(currSize % MinObjAlignment == 0, "currSize should be aligned");
1211 for (i = currSize; i < IndexSetSize; i += IndexSetStride) {
1212 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[i];
1213 if (fl->head()) {
1214 ret = getFromListGreater(fl, numWords);
1215 assert(ret == NULL || ret->is_free(), "Should be returning a free chunk");
1216 return ret;
1217 }
1218 }
1219
1220 currSize = MAX2((size_t)SmallForDictionary,
1221 (size_t)(numWords + MinChunkSize));
1222
1223 /* Try to get a chunk that satisfies request, while avoiding
1224 fragmentation that can't be handled. */
1225 {
1226 ret = dictionary()->get_chunk(currSize);
1227 if (ret != NULL) {
1228 assert(ret->size() - numWords >= MinChunkSize,
1229 "Chunk is too small");
1230 _bt.allocated((HeapWord*)ret, ret->size());
1231 /* Carve returned chunk. */
1232 (void) splitChunkAndReturnRemainder(ret, numWords);
1233 /* Label this as no longer a free chunk. */
1234 assert(ret->is_free(), "This chunk should be free");
1235 ret->link_prev(NULL);
1236 }
1237 assert(ret == NULL || ret->is_free(), "Should be returning a free chunk");
1238 return ret;
1239 }
1240 ShouldNotReachHere();
1241 }
1242
1243 bool CompactibleFreeListSpace::verifyChunkInIndexedFreeLists(FreeChunk* fc) const {
1244 assert(fc->size() < IndexSetSize, "Size of chunk is too large");
1245 return _indexedFreeList[fc->size()].verify_chunk_in_free_list(fc);
1246 }
1247
1248 bool CompactibleFreeListSpace::verify_chunk_is_linear_alloc_block(FreeChunk* fc) const {
1249 assert((_smallLinearAllocBlock._ptr != (HeapWord*)fc) ||
1250 (_smallLinearAllocBlock._word_size == fc->size()),
1251 "Linear allocation block shows incorrect size");
1252 return ((_smallLinearAllocBlock._ptr == (HeapWord*)fc) &&
1253 (_smallLinearAllocBlock._word_size == fc->size()));
1254 }
1255
1256 // Check if the purported free chunk is present either as a linear
1257 // allocation block, the size-indexed table of (smaller) free blocks,
1258 // or the larger free blocks kept in the binary tree dictionary.
1259 bool CompactibleFreeListSpace::verify_chunk_in_free_list(FreeChunk* fc) const {
1260 if (verify_chunk_is_linear_alloc_block(fc)) {
1261 return true;
1262 } else if (fc->size() < IndexSetSize) {
1263 return verifyChunkInIndexedFreeLists(fc);
1264 } else {
1265 return dictionary()->verify_chunk_in_free_list(fc);
1266 }
1267 }
1268
1269 #ifndef PRODUCT
1270 void CompactibleFreeListSpace::assert_locked() const {
1271 CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock());
1272 }
1273
1274 void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const {
1275 CMSLockVerifier::assert_locked(lock);
1276 }
1277 #endif
1278
1279 FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) {
1280 // In the parallel case, the main thread holds the free list lock
1281 // on behalf the parallel threads.
1282 FreeChunk* fc;
1283 {
1284 // If GC is parallel, this might be called by several threads.
1285 // This should be rare enough that the locking overhead won't affect
1286 // the sequential code.
1287 MutexLockerEx x(parDictionaryAllocLock(),
1288 Mutex::_no_safepoint_check_flag);
1289 fc = getChunkFromDictionary(size);
1290 }
1291 if (fc != NULL) {
1292 fc->dontCoalesce();
1293 assert(fc->is_free(), "Should be free, but not coalescable");
1294 // Verify that the block offset table shows this to
1295 // be a single block, but not one which is unallocated.
1296 _bt.verify_single_block((HeapWord*)fc, fc->size());
1297 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
1298 }
1299 return fc;
1300 }
1301
1302 oop CompactibleFreeListSpace::promote(oop obj, size_t obj_size) {
1303 assert(obj_size == (size_t)obj->size(), "bad obj_size passed in");
1304 assert_locked();
1305
1306 // if we are tracking promotions, then first ensure space for
1307 // promotion (including spooling space for saving header if necessary).
1308 // then allocate and copy, then track promoted info if needed.
1309 // When tracking (see PromotionInfo::track()), the mark word may
1310 // be displaced and in this case restoration of the mark word
1311 // occurs in the (oop_since_save_marks_)iterate phase.
1312 if (_promoInfo.tracking() && !_promoInfo.ensure_spooling_space()) {
1313 return NULL;
1314 }
1315 // Call the allocate(size_t, bool) form directly to avoid the
1316 // additional call through the allocate(size_t) form. Having
1317 // the compile inline the call is problematic because allocate(size_t)
1318 // is a virtual method.
1319 HeapWord* res = allocate(adjustObjectSize(obj_size));
1320 if (res != NULL) {
1321 Copy::aligned_disjoint_words((HeapWord*)obj, res, obj_size);
1322 // if we should be tracking promotions, do so.
1323 if (_promoInfo.tracking()) {
1324 _promoInfo.track((PromotedObject*)res);
1325 }
1326 }
1327 return oop(res);
1328 }
1329
1330 HeapWord*
1331 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlock(size_t size) {
1332 assert_locked();
1333 assert(size >= MinChunkSize, "minimum chunk size");
1334 assert(size < _smallLinearAllocBlock._allocation_size_limit,
1335 "maximum from smallLinearAllocBlock");
1336 return getChunkFromLinearAllocBlock(&_smallLinearAllocBlock, size);
1337 }
1338
1339 HeapWord*
1340 CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk,
1341 size_t size) {
1342 assert_locked();
1343 assert(size >= MinChunkSize, "too small");
1344 HeapWord* res = NULL;
1345 // Try to do linear allocation from blk, making sure that
1346 if (blk->_word_size == 0) {
1347 // We have probably been unable to fill this either in the prologue or
1348 // when it was exhausted at the last linear allocation. Bail out until
1349 // next time.
1350 assert(blk->_ptr == NULL, "consistency check");
1351 return NULL;
1352 }
1353 assert(blk->_word_size != 0 && blk->_ptr != NULL, "consistency check");
1354 res = getChunkFromLinearAllocBlockRemainder(blk, size);
1355 if (res != NULL) return res;
1356
1357 // about to exhaust this linear allocation block
1358 if (blk->_word_size == size) { // exactly satisfied
1359 res = blk->_ptr;
1360 _bt.allocated(res, blk->_word_size);
1361 } else if (size + MinChunkSize <= blk->_refillSize) {
1362 size_t sz = blk->_word_size;
1363 // Update _unallocated_block if the size is such that chunk would be
1364 // returned to the indexed free list. All other chunks in the indexed
1365 // free lists are allocated from the dictionary so that _unallocated_block
1366 // has already been adjusted for them. Do it here so that the cost
1367 // for all chunks added back to the indexed free lists.
1368 if (sz < SmallForDictionary) {
1369 _bt.allocated(blk->_ptr, sz);
1370 }
1371 // Return the chunk that isn't big enough, and then refill below.
1372 addChunkToFreeLists(blk->_ptr, sz);
1373 split_birth(sz);
1374 // Don't keep statistics on adding back chunk from a LinAB.
1375 } else {
1376 // A refilled block would not satisfy the request.
1377 return NULL;
1378 }
1379
1380 blk->_ptr = NULL; blk->_word_size = 0;
1381 refillLinearAllocBlock(blk);
1382 assert(blk->_ptr == NULL || blk->_word_size >= size + MinChunkSize,
1383 "block was replenished");
1384 if (res != NULL) {
1385 split_birth(size);
1386 repairLinearAllocBlock(blk);
1387 } else if (blk->_ptr != NULL) {
1388 res = blk->_ptr;
1389 size_t blk_size = blk->_word_size;
1390 blk->_word_size -= size;
1391 blk->_ptr += size;
1392 split_birth(size);
1393 repairLinearAllocBlock(blk);
1394 // Update BOT last so that other (parallel) GC threads see a consistent
1395 // view of the BOT and free blocks.
1396 // Above must occur before BOT is updated below.
1397 OrderAccess::storestore();
1398 _bt.split_block(res, blk_size, size); // adjust block offset table
1399 }
1400 return res;
1401 }
1402
1403 HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder(
1404 LinearAllocBlock* blk,
1405 size_t size) {
1406 assert_locked();
1407 assert(size >= MinChunkSize, "too small");
1408
1409 HeapWord* res = NULL;
1410 // This is the common case. Keep it simple.
1411 if (blk->_word_size >= size + MinChunkSize) {
1412 assert(blk->_ptr != NULL, "consistency check");
1413 res = blk->_ptr;
1414 // Note that the BOT is up-to-date for the linAB before allocation. It
1415 // indicates the start of the linAB. The split_block() updates the
1416 // BOT for the linAB after the allocation (indicates the start of the
1417 // next chunk to be allocated).
1418 size_t blk_size = blk->_word_size;
1419 blk->_word_size -= size;
1420 blk->_ptr += size;
1421 split_birth(size);
1422 repairLinearAllocBlock(blk);
1423 // Update BOT last so that other (parallel) GC threads see a consistent
1424 // view of the BOT and free blocks.
1425 // Above must occur before BOT is updated below.
1426 OrderAccess::storestore();
1427 _bt.split_block(res, blk_size, size); // adjust block offset table
1428 _bt.allocated(res, size);
1429 }
1430 return res;
1431 }
1432
1433 FreeChunk*
1434 CompactibleFreeListSpace::getChunkFromIndexedFreeList(size_t size) {
1435 assert_locked();
1436 assert(size < SmallForDictionary, "just checking");
1437 FreeChunk* res;
1438 res = _indexedFreeList[size].get_chunk_at_head();
1439 if (res == NULL) {
1440 res = getChunkFromIndexedFreeListHelper(size);
1441 }
1442 _bt.verify_not_unallocated((HeapWord*) res, size);
1443 assert(res == NULL || res->size() == size, "Incorrect block size");
1444 return res;
1445 }
1446
1447 FreeChunk*
1448 CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size,
1449 bool replenish) {
1450 assert_locked();
1451 FreeChunk* fc = NULL;
1452 if (size < SmallForDictionary) {
1453 assert(_indexedFreeList[size].head() == NULL ||
1454 _indexedFreeList[size].surplus() <= 0,
1455 "List for this size should be empty or under populated");
1456 // Try best fit in exact lists before replenishing the list
1457 if (!bestFitFirst() || (fc = bestFitSmall(size)) == NULL) {
1458 // Replenish list.
1459 //
1460 // Things tried that failed.
1461 // Tried allocating out of the two LinAB's first before
1462 // replenishing lists.
1463 // Tried small linAB of size 256 (size in indexed list)
1464 // and replenishing indexed lists from the small linAB.
1465 //
1466 FreeChunk* newFc = NULL;
1467 const size_t replenish_size = CMSIndexedFreeListReplenish * size;
1468 if (replenish_size < SmallForDictionary) {
1469 // Do not replenish from an underpopulated size.
1470 if (_indexedFreeList[replenish_size].surplus() > 0 &&
1471 _indexedFreeList[replenish_size].head() != NULL) {
1472 newFc = _indexedFreeList[replenish_size].get_chunk_at_head();
1473 } else if (bestFitFirst()) {
1474 newFc = bestFitSmall(replenish_size);
1475 }
1476 }
1477 if (newFc == NULL && replenish_size > size) {
1478 assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant");
1479 newFc = getChunkFromIndexedFreeListHelper(replenish_size, false);
1480 }
1481 // Note: The stats update re split-death of block obtained above
1482 // will be recorded below precisely when we know we are going to
1483 // be actually splitting it into more than one pieces below.
1484 if (newFc != NULL) {
1485 if (replenish || CMSReplenishIntermediate) {
1486 // Replenish this list and return one block to caller.
1487 size_t i;
1488 FreeChunk *curFc, *nextFc;
1489 size_t num_blk = newFc->size() / size;
1490 assert(num_blk >= 1, "Smaller than requested?");
1491 assert(newFc->size() % size == 0, "Should be integral multiple of request");
1492 if (num_blk > 1) {
1493 // we are sure we will be splitting the block just obtained
1494 // into multiple pieces; record the split-death of the original
1495 splitDeath(replenish_size);
1496 }
1497 // carve up and link blocks 0, ..., num_blk - 2
1498 // The last chunk is not added to the lists but is returned as the
1499 // free chunk.
1500 for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size),
1501 i = 0;
1502 i < (num_blk - 1);
1503 curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size),
1504 i++) {
1505 curFc->set_size(size);
1506 // Don't record this as a return in order to try and
1507 // determine the "returns" from a GC.
1508 _bt.verify_not_unallocated((HeapWord*) fc, size);
1509 _indexedFreeList[size].return_chunk_at_tail(curFc, false);
1510 _bt.mark_block((HeapWord*)curFc, size);
1511 split_birth(size);
1512 // Don't record the initial population of the indexed list
1513 // as a split birth.
1514 }
1515
1516 // check that the arithmetic was OK above
1517 assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size,
1518 "inconsistency in carving newFc");
1519 curFc->set_size(size);
1520 _bt.mark_block((HeapWord*)curFc, size);
1521 split_birth(size);
1522 fc = curFc;
1523 } else {
1524 // Return entire block to caller
1525 fc = newFc;
1526 }
1527 }
1528 }
1529 } else {
1530 // Get a free chunk from the free chunk dictionary to be returned to
1531 // replenish the indexed free list.
1532 fc = getChunkFromDictionaryExact(size);
1533 }
1534 // assert(fc == NULL || fc->is_free(), "Should be returning a free chunk");
1535 return fc;
1536 }
1537
1538 FreeChunk*
1539 CompactibleFreeListSpace::getChunkFromDictionary(size_t size) {
1540 assert_locked();
1541 FreeChunk* fc = _dictionary->get_chunk(size,
1542 FreeBlockDictionary<FreeChunk>::atLeast);
1543 if (fc == NULL) {
1544 return NULL;
1545 }
1546 _bt.allocated((HeapWord*)fc, fc->size());
1547 if (fc->size() >= size + MinChunkSize) {
1548 fc = splitChunkAndReturnRemainder(fc, size);
1549 }
1550 assert(fc->size() >= size, "chunk too small");
1551 assert(fc->size() < size + MinChunkSize, "chunk too big");
1552 _bt.verify_single_block((HeapWord*)fc, fc->size());
1553 return fc;
1554 }
1555
1556 FreeChunk*
1557 CompactibleFreeListSpace::getChunkFromDictionaryExact(size_t size) {
1558 assert_locked();
1559 FreeChunk* fc = _dictionary->get_chunk(size,
1560 FreeBlockDictionary<FreeChunk>::atLeast);
1561 if (fc == NULL) {
1562 return fc;
1563 }
1564 _bt.allocated((HeapWord*)fc, fc->size());
1565 if (fc->size() == size) {
1566 _bt.verify_single_block((HeapWord*)fc, size);
1567 return fc;
1568 }
1569 assert(fc->size() > size, "get_chunk() guarantee");
1570 if (fc->size() < size + MinChunkSize) {
1571 // Return the chunk to the dictionary and go get a bigger one.
1572 returnChunkToDictionary(fc);
1573 fc = _dictionary->get_chunk(size + MinChunkSize,
1574 FreeBlockDictionary<FreeChunk>::atLeast);
1575 if (fc == NULL) {
1576 return NULL;
1577 }
1578 _bt.allocated((HeapWord*)fc, fc->size());
1579 }
1580 assert(fc->size() >= size + MinChunkSize, "tautology");
1581 fc = splitChunkAndReturnRemainder(fc, size);
1582 assert(fc->size() == size, "chunk is wrong size");
1583 _bt.verify_single_block((HeapWord*)fc, size);
1584 return fc;
1585 }
1586
1587 void
1588 CompactibleFreeListSpace::returnChunkToDictionary(FreeChunk* chunk) {
1589 assert_locked();
1590
1591 size_t size = chunk->size();
1592 _bt.verify_single_block((HeapWord*)chunk, size);
1593 // adjust _unallocated_block downward, as necessary
1594 _bt.freed((HeapWord*)chunk, size);
1595 _dictionary->return_chunk(chunk);
1596 #ifndef PRODUCT
1597 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1598 TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >* tc = TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::as_TreeChunk(chunk);
1599 TreeList<FreeChunk, AdaptiveFreeList<FreeChunk> >* tl = tc->list();
1600 tl->verify_stats();
1601 }
1602 #endif // PRODUCT
1603 }
1604
1605 void
1606 CompactibleFreeListSpace::returnChunkToFreeList(FreeChunk* fc) {
1607 assert_locked();
1608 size_t size = fc->size();
1609 _bt.verify_single_block((HeapWord*) fc, size);
1610 _bt.verify_not_unallocated((HeapWord*) fc, size);
1611 _indexedFreeList[size].return_chunk_at_tail(fc);
1612 #ifndef PRODUCT
1613 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1614 _indexedFreeList[size].verify_stats();
1615 }
1616 #endif // PRODUCT
1617 }
1618
1619 // Add chunk to end of last block -- if it's the largest
1620 // block -- and update BOT and census data. We would
1621 // of course have preferred to coalesce it with the
1622 // last block, but it's currently less expensive to find the
1623 // largest block than it is to find the last.
1624 void
1625 CompactibleFreeListSpace::addChunkToFreeListsAtEndRecordingStats(
1626 HeapWord* chunk, size_t size) {
1627 // check that the chunk does lie in this space!
1628 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1629 // One of the parallel gc task threads may be here
1630 // whilst others are allocating.
1631 Mutex* lock = &_parDictionaryAllocLock;
1632 FreeChunk* ec;
1633 {
1634 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1635 ec = dictionary()->find_largest_dict(); // get largest block
1636 if (ec != NULL && ec->end() == (uintptr_t*) chunk) {
1637 // It's a coterminal block - we can coalesce.
1638 size_t old_size = ec->size();
1639 coalDeath(old_size);
1640 removeChunkFromDictionary(ec);
1641 size += old_size;
1642 } else {
1643 ec = (FreeChunk*)chunk;
1644 }
1645 }
1646 ec->set_size(size);
1647 debug_only(ec->mangleFreed(size));
1648 if (size < SmallForDictionary) {
1649 lock = _indexedFreeListParLocks[size];
1650 }
1651 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1652 addChunkAndRepairOffsetTable((HeapWord*)ec, size, true);
1653 // record the birth under the lock since the recording involves
1654 // manipulation of the list on which the chunk lives and
1655 // if the chunk is allocated and is the last on the list,
1656 // the list can go away.
1657 coalBirth(size);
1658 }
1659
1660 void
1661 CompactibleFreeListSpace::addChunkToFreeLists(HeapWord* chunk,
1662 size_t size) {
1663 // check that the chunk does lie in this space!
1664 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1665 assert_locked();
1666 _bt.verify_single_block(chunk, size);
1667
1668 FreeChunk* fc = (FreeChunk*) chunk;
1669 fc->set_size(size);
1670 debug_only(fc->mangleFreed(size));
1671 if (size < SmallForDictionary) {
1672 returnChunkToFreeList(fc);
1673 } else {
1674 returnChunkToDictionary(fc);
1675 }
1676 }
1677
1678 void
1679 CompactibleFreeListSpace::addChunkAndRepairOffsetTable(HeapWord* chunk,
1680 size_t size, bool coalesced) {
1681 assert_locked();
1682 assert(chunk != NULL, "null chunk");
1683 if (coalesced) {
1684 // repair BOT
1685 _bt.single_block(chunk, size);
1686 }
1687 addChunkToFreeLists(chunk, size);
1688 }
1689
1690 // We _must_ find the purported chunk on our free lists;
1691 // we assert if we don't.
1692 void
1693 CompactibleFreeListSpace::removeFreeChunkFromFreeLists(FreeChunk* fc) {
1694 size_t size = fc->size();
1695 assert_locked();
1696 debug_only(verifyFreeLists());
1697 if (size < SmallForDictionary) {
1698 removeChunkFromIndexedFreeList(fc);
1699 } else {
1700 removeChunkFromDictionary(fc);
1701 }
1702 _bt.verify_single_block((HeapWord*)fc, size);
1703 debug_only(verifyFreeLists());
1704 }
1705
1706 void
1707 CompactibleFreeListSpace::removeChunkFromDictionary(FreeChunk* fc) {
1708 size_t size = fc->size();
1709 assert_locked();
1710 assert(fc != NULL, "null chunk");
1711 _bt.verify_single_block((HeapWord*)fc, size);
1712 _dictionary->remove_chunk(fc);
1713 // adjust _unallocated_block upward, as necessary
1714 _bt.allocated((HeapWord*)fc, size);
1715 }
1716
1717 void
1718 CompactibleFreeListSpace::removeChunkFromIndexedFreeList(FreeChunk* fc) {
1719 assert_locked();
1720 size_t size = fc->size();
1721 _bt.verify_single_block((HeapWord*)fc, size);
1722 NOT_PRODUCT(
1723 if (FLSVerifyIndexTable) {
1724 verifyIndexedFreeList(size);
1725 }
1726 )
1727 _indexedFreeList[size].remove_chunk(fc);
1728 NOT_PRODUCT(
1729 if (FLSVerifyIndexTable) {
1730 verifyIndexedFreeList(size);
1731 }
1732 )
1733 }
1734
1735 FreeChunk* CompactibleFreeListSpace::bestFitSmall(size_t numWords) {
1736 /* A hint is the next larger size that has a surplus.
1737 Start search at a size large enough to guarantee that
1738 the excess is >= MIN_CHUNK. */
1739 size_t start = align_object_size(numWords + MinChunkSize);
1740 if (start < IndexSetSize) {
1741 AdaptiveFreeList<FreeChunk>* it = _indexedFreeList;
1742 size_t hint = _indexedFreeList[start].hint();
1743 while (hint < IndexSetSize) {
1744 assert(hint % MinObjAlignment == 0, "hint should be aligned");
1745 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[hint];
1746 if (fl->surplus() > 0 && fl->head() != NULL) {
1747 // Found a list with surplus, reset original hint
1748 // and split out a free chunk which is returned.
1749 _indexedFreeList[start].set_hint(hint);
1750 FreeChunk* res = getFromListGreater(fl, numWords);
1751 assert(res == NULL || res->is_free(),
1752 "Should be returning a free chunk");
1753 return res;
1754 }
1755 hint = fl->hint(); /* keep looking */
1756 }
1757 /* None found. */
1758 it[start].set_hint(IndexSetSize);
1759 }
1760 return NULL;
1761 }
1762
1763 /* Requires fl->size >= numWords + MinChunkSize */
1764 FreeChunk* CompactibleFreeListSpace::getFromListGreater(AdaptiveFreeList<FreeChunk>* fl,
1765 size_t numWords) {
1766 FreeChunk *curr = fl->head();
1767 size_t oldNumWords = curr->size();
1768 assert(numWords >= MinChunkSize, "Word size is too small");
1769 assert(curr != NULL, "List is empty");
1770 assert(oldNumWords >= numWords + MinChunkSize,
1771 "Size of chunks in the list is too small");
1772
1773 fl->remove_chunk(curr);
1774 // recorded indirectly by splitChunkAndReturnRemainder -
1775 // smallSplit(oldNumWords, numWords);
1776 FreeChunk* new_chunk = splitChunkAndReturnRemainder(curr, numWords);
1777 // Does anything have to be done for the remainder in terms of
1778 // fixing the card table?
1779 assert(new_chunk == NULL || new_chunk->is_free(),
1780 "Should be returning a free chunk");
1781 return new_chunk;
1782 }
1783
1784 FreeChunk*
1785 CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk,
1786 size_t new_size) {
1787 assert_locked();
1788 size_t size = chunk->size();
1789 assert(size > new_size, "Split from a smaller block?");
1790 assert(is_aligned(chunk), "alignment problem");
1791 assert(size == adjustObjectSize(size), "alignment problem");
1792 size_t rem_sz = size - new_size;
1793 assert(rem_sz == adjustObjectSize(rem_sz), "alignment problem");
1794 assert(rem_sz >= MinChunkSize, "Free chunk smaller than minimum");
1795 FreeChunk* ffc = (FreeChunk*)((HeapWord*)chunk + new_size);
1796 assert(is_aligned(ffc), "alignment problem");
1797 ffc->set_size(rem_sz);
1798 ffc->link_next(NULL);
1799 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
1800 // Above must occur before BOT is updated below.
1801 // adjust block offset table
1802 OrderAccess::storestore();
1803 assert(chunk->is_free() && ffc->is_free(), "Error");
1804 _bt.split_block((HeapWord*)chunk, chunk->size(), new_size);
1805 if (rem_sz < SmallForDictionary) {
1806 // The freeList lock is held, but multiple GC task threads might be executing in parallel.
1807 bool is_par = Thread::current()->is_GC_task_thread();
1808 if (is_par) _indexedFreeListParLocks[rem_sz]->lock();
1809 returnChunkToFreeList(ffc);
1810 split(size, rem_sz);
1811 if (is_par) _indexedFreeListParLocks[rem_sz]->unlock();
1812 } else {
1813 returnChunkToDictionary(ffc);
1814 split(size, rem_sz);
1815 }
1816 chunk->set_size(new_size);
1817 return chunk;
1818 }
1819
1820 void
1821 CompactibleFreeListSpace::sweep_completed() {
1822 // Now that space is probably plentiful, refill linear
1823 // allocation blocks as needed.
1824 refillLinearAllocBlocksIfNeeded();
1825 }
1826
1827 void
1828 CompactibleFreeListSpace::gc_prologue() {
1829 assert_locked();
1830 reportFreeListStatistics("Before GC:");
1831 refillLinearAllocBlocksIfNeeded();
1832 }
1833
1834 void
1835 CompactibleFreeListSpace::gc_epilogue() {
1836 assert_locked();
1837 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1838 _promoInfo.stopTrackingPromotions();
1839 repairLinearAllocationBlocks();
1840 reportFreeListStatistics("After GC:");
1841 }
1842
1843 // Iteration support, mostly delegated from a CMS generation
1844
1845 void CompactibleFreeListSpace::save_marks() {
1846 assert(Thread::current()->is_VM_thread(),
1847 "Global variable should only be set when single-threaded");
1848 // Mark the "end" of the used space at the time of this call;
1849 // note, however, that promoted objects from this point
1850 // on are tracked in the _promoInfo below.
1851 set_saved_mark_word(unallocated_block());
1852 #ifdef ASSERT
1853 // Check the sanity of save_marks() etc.
1854 MemRegion ur = used_region();
1855 MemRegion urasm = used_region_at_save_marks();
1856 assert(ur.contains(urasm),
1857 " Error at save_marks(): [" PTR_FORMAT "," PTR_FORMAT ")"
1858 " should contain [" PTR_FORMAT "," PTR_FORMAT ")",
1859 p2i(ur.start()), p2i(ur.end()), p2i(urasm.start()), p2i(urasm.end()));
1860 #endif
1861 // inform allocator that promotions should be tracked.
1862 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1863 _promoInfo.startTrackingPromotions();
1864 }
1865
1866 bool CompactibleFreeListSpace::no_allocs_since_save_marks() {
1867 assert(_promoInfo.tracking(), "No preceding save_marks?");
1868 return _promoInfo.noPromotions();
1869 }
1870
1871 #define CFLS_OOP_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix) \
1872 \
1873 void CompactibleFreeListSpace:: \
1874 oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk) { \
1875 _promoInfo.promoted_oops_iterate##nv_suffix(blk); \
1876 /* \
1877 * This also restores any displaced headers and removes the elements from \
1878 * the iteration set as they are processed, so that we have a clean slate \
1879 * at the end of the iteration. Note, thus, that if new objects are \
1880 * promoted as a result of the iteration they are iterated over as well. \
1881 */ \
1882 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); \
1883 }
1884
1885 ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DEFN)
1886
1887 bool CompactibleFreeListSpace::linearAllocationWouldFail() const {
1888 return _smallLinearAllocBlock._word_size == 0;
1889 }
1890
1891 void CompactibleFreeListSpace::repairLinearAllocationBlocks() {
1892 // Fix up linear allocation blocks to look like free blocks
1893 repairLinearAllocBlock(&_smallLinearAllocBlock);
1894 }
1895
1896 void CompactibleFreeListSpace::repairLinearAllocBlock(LinearAllocBlock* blk) {
1897 assert_locked();
1898 if (blk->_ptr != NULL) {
1899 assert(blk->_word_size != 0 && blk->_word_size >= MinChunkSize,
1900 "Minimum block size requirement");
1901 FreeChunk* fc = (FreeChunk*)(blk->_ptr);
1902 fc->set_size(blk->_word_size);
1903 fc->link_prev(NULL); // mark as free
1904 fc->dontCoalesce();
1905 assert(fc->is_free(), "just marked it free");
1906 assert(fc->cantCoalesce(), "just marked it uncoalescable");
1907 }
1908 }
1909
1910 void CompactibleFreeListSpace::refillLinearAllocBlocksIfNeeded() {
1911 assert_locked();
1912 if (_smallLinearAllocBlock._ptr == NULL) {
1913 assert(_smallLinearAllocBlock._word_size == 0,
1914 "Size of linAB should be zero if the ptr is NULL");
1915 // Reset the linAB refill and allocation size limit.
1916 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, SmallForLinearAlloc);
1917 }
1918 refillLinearAllocBlockIfNeeded(&_smallLinearAllocBlock);
1919 }
1920
1921 void
1922 CompactibleFreeListSpace::refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk) {
1923 assert_locked();
1924 assert((blk->_ptr == NULL && blk->_word_size == 0) ||
1925 (blk->_ptr != NULL && blk->_word_size >= MinChunkSize),
1926 "blk invariant");
1927 if (blk->_ptr == NULL) {
1928 refillLinearAllocBlock(blk);
1929 }
1930 if (PrintMiscellaneous && Verbose) {
1931 if (blk->_word_size == 0) {
1932 warning("CompactibleFreeListSpace(prologue):: Linear allocation failure");
1933 }
1934 }
1935 }
1936
1937 void
1938 CompactibleFreeListSpace::refillLinearAllocBlock(LinearAllocBlock* blk) {
1939 assert_locked();
1940 assert(blk->_word_size == 0 && blk->_ptr == NULL,
1941 "linear allocation block should be empty");
1942 FreeChunk* fc;
1943 if (blk->_refillSize < SmallForDictionary &&
1944 (fc = getChunkFromIndexedFreeList(blk->_refillSize)) != NULL) {
1945 // A linAB's strategy might be to use small sizes to reduce
1946 // fragmentation but still get the benefits of allocation from a
1947 // linAB.
1948 } else {
1949 fc = getChunkFromDictionary(blk->_refillSize);
1950 }
1951 if (fc != NULL) {
1952 blk->_ptr = (HeapWord*)fc;
1953 blk->_word_size = fc->size();
1954 fc->dontCoalesce(); // to prevent sweeper from sweeping us up
1955 }
1956 }
1957
1958 // Support for compaction
1959 void CompactibleFreeListSpace::prepare_for_compaction(CompactPoint* cp) {
1960 scan_and_forward(this, cp);
1961 // Prepare_for_compaction() uses the space between live objects
1962 // so that later phase can skip dead space quickly. So verification
1963 // of the free lists doesn't work after.
1964 }
1965
1966 void CompactibleFreeListSpace::adjust_pointers() {
1967 // In other versions of adjust_pointers(), a bail out
1968 // based on the amount of live data in the generation
1969 // (i.e., if 0, bail out) may be used.
1970 // Cannot test used() == 0 here because the free lists have already
1971 // been mangled by the compaction.
1972
1973 scan_and_adjust_pointers(this);
1974 // See note about verification in prepare_for_compaction().
1975 }
1976
1977 void CompactibleFreeListSpace::compact() {
1978 scan_and_compact(this);
1979 }
1980
1981 // Fragmentation metric = 1 - [sum of (fbs**2) / (sum of fbs)**2]
1982 // where fbs is free block sizes
1983 double CompactibleFreeListSpace::flsFrag() const {
1984 size_t itabFree = totalSizeInIndexedFreeLists();
1985 double frag = 0.0;
1986 size_t i;
1987
1988 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
1989 double sz = i;
1990 frag += _indexedFreeList[i].count() * (sz * sz);
1991 }
1992
1993 double totFree = itabFree +
1994 _dictionary->total_chunk_size(DEBUG_ONLY(freelistLock()));
1995 if (totFree > 0) {
1996 frag = ((frag + _dictionary->sum_of_squared_block_sizes()) /
1997 (totFree * totFree));
1998 frag = (double)1.0 - frag;
1999 } else {
2000 assert(frag == 0.0, "Follows from totFree == 0");
2001 }
2002 return frag;
2003 }
2004
2005 void CompactibleFreeListSpace::beginSweepFLCensus(
2006 float inter_sweep_current,
2007 float inter_sweep_estimate,
2008 float intra_sweep_estimate) {
2009 assert_locked();
2010 size_t i;
2011 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2012 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[i];
2013 log_trace(gc, freelist)("size[" SIZE_FORMAT "] : ", i);
2014 fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate);
2015 fl->set_coal_desired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent));
2016 fl->set_before_sweep(fl->count());
2017 fl->set_bfr_surp(fl->surplus());
2018 }
2019 _dictionary->begin_sweep_dict_census(CMSLargeCoalSurplusPercent,
2020 inter_sweep_current,
2021 inter_sweep_estimate,
2022 intra_sweep_estimate);
2023 }
2024
2025 void CompactibleFreeListSpace::setFLSurplus() {
2026 assert_locked();
2027 size_t i;
2028 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2029 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2030 fl->set_surplus(fl->count() -
2031 (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent));
2032 }
2033 }
2034
2035 void CompactibleFreeListSpace::setFLHints() {
2036 assert_locked();
2037 size_t i;
2038 size_t h = IndexSetSize;
2039 for (i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
2040 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2041 fl->set_hint(h);
2042 if (fl->surplus() > 0) {
2043 h = i;
2044 }
2045 }
2046 }
2047
2048 void CompactibleFreeListSpace::clearFLCensus() {
2049 assert_locked();
2050 size_t i;
2051 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2052 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2053 fl->set_prev_sweep(fl->count());
2054 fl->set_coal_births(0);
2055 fl->set_coal_deaths(0);
2056 fl->set_split_births(0);
2057 fl->set_split_deaths(0);
2058 }
2059 }
2060
2061 void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) {
2062 log_debug(gc, freelist)("CMS: Large block " PTR_FORMAT, p2i(dictionary()->find_largest_dict()));
2063 setFLSurplus();
2064 setFLHints();
2065 printFLCensus(sweep_count);
2066 clearFLCensus();
2067 assert_locked();
2068 _dictionary->end_sweep_dict_census(CMSLargeSplitSurplusPercent);
2069 }
2070
2071 bool CompactibleFreeListSpace::coalOverPopulated(size_t size) {
2072 if (size < SmallForDictionary) {
2073 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2074 return (fl->coal_desired() < 0) ||
2075 ((int)fl->count() > fl->coal_desired());
2076 } else {
2077 return dictionary()->coal_dict_over_populated(size);
2078 }
2079 }
2080
2081 void CompactibleFreeListSpace::smallCoalBirth(size_t size) {
2082 assert(size < SmallForDictionary, "Size too large for indexed list");
2083 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2084 fl->increment_coal_births();
2085 fl->increment_surplus();
2086 }
2087
2088 void CompactibleFreeListSpace::smallCoalDeath(size_t size) {
2089 assert(size < SmallForDictionary, "Size too large for indexed list");
2090 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2091 fl->increment_coal_deaths();
2092 fl->decrement_surplus();
2093 }
2094
2095 void CompactibleFreeListSpace::coalBirth(size_t size) {
2096 if (size < SmallForDictionary) {
2097 smallCoalBirth(size);
2098 } else {
2099 dictionary()->dict_census_update(size,
2100 false /* split */,
2101 true /* birth */);
2102 }
2103 }
2104
2105 void CompactibleFreeListSpace::coalDeath(size_t size) {
2106 if(size < SmallForDictionary) {
2107 smallCoalDeath(size);
2108 } else {
2109 dictionary()->dict_census_update(size,
2110 false /* split */,
2111 false /* birth */);
2112 }
2113 }
2114
2115 void CompactibleFreeListSpace::smallSplitBirth(size_t size) {
2116 assert(size < SmallForDictionary, "Size too large for indexed list");
2117 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2118 fl->increment_split_births();
2119 fl->increment_surplus();
2120 }
2121
2122 void CompactibleFreeListSpace::smallSplitDeath(size_t size) {
2123 assert(size < SmallForDictionary, "Size too large for indexed list");
2124 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2125 fl->increment_split_deaths();
2126 fl->decrement_surplus();
2127 }
2128
2129 void CompactibleFreeListSpace::split_birth(size_t size) {
2130 if (size < SmallForDictionary) {
2131 smallSplitBirth(size);
2132 } else {
2133 dictionary()->dict_census_update(size,
2134 true /* split */,
2135 true /* birth */);
2136 }
2137 }
2138
2139 void CompactibleFreeListSpace::splitDeath(size_t size) {
2140 if (size < SmallForDictionary) {
2141 smallSplitDeath(size);
2142 } else {
2143 dictionary()->dict_census_update(size,
2144 true /* split */,
2145 false /* birth */);
2146 }
2147 }
2148
2149 void CompactibleFreeListSpace::split(size_t from, size_t to1) {
2150 size_t to2 = from - to1;
2151 splitDeath(from);
2152 split_birth(to1);
2153 split_birth(to2);
2154 }
2155
2156 void CompactibleFreeListSpace::print() const {
2157 print_on(tty);
2158 }
2159
2160 void CompactibleFreeListSpace::prepare_for_verify() {
2161 assert_locked();
2162 repairLinearAllocationBlocks();
2163 // Verify that the SpoolBlocks look like free blocks of
2164 // appropriate sizes... To be done ...
2165 }
2166
2167 class VerifyAllBlksClosure: public BlkClosure {
2168 private:
2169 const CompactibleFreeListSpace* _sp;
2170 const MemRegion _span;
2171 HeapWord* _last_addr;
2172 size_t _last_size;
2173 bool _last_was_obj;
2174 bool _last_was_live;
2175
2176 public:
2177 VerifyAllBlksClosure(const CompactibleFreeListSpace* sp,
2178 MemRegion span) : _sp(sp), _span(span),
2179 _last_addr(NULL), _last_size(0),
2180 _last_was_obj(false), _last_was_live(false) { }
2181
2182 virtual size_t do_blk(HeapWord* addr) {
2183 size_t res;
2184 bool was_obj = false;
2185 bool was_live = false;
2186 if (_sp->block_is_obj(addr)) {
2187 was_obj = true;
2188 oop p = oop(addr);
2189 guarantee(p->is_oop(), "Should be an oop");
2190 res = _sp->adjustObjectSize(p->size());
2191 if (_sp->obj_is_alive(addr)) {
2192 was_live = true;
2193 p->verify();
2194 }
2195 } else {
2196 FreeChunk* fc = (FreeChunk*)addr;
2197 res = fc->size();
2198 if (FLSVerifyLists && !fc->cantCoalesce()) {
2199 guarantee(_sp->verify_chunk_in_free_list(fc),
2200 "Chunk should be on a free list");
2201 }
2202 }
2203 if (res == 0) {
2204 LogHandle(gc, verify) log;
2205 log.info("Livelock: no rank reduction!");
2206 log.info(" Current: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n"
2207 " Previous: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n",
2208 p2i(addr), res, was_obj ?"true":"false", was_live ?"true":"false",
2209 p2i(_last_addr), _last_size, _last_was_obj?"true":"false", _last_was_live?"true":"false");
2210 ResourceMark rm;
2211 _sp->print_on(log.info_stream());
2212 guarantee(false, "Verification failed.");
2213 }
2214 _last_addr = addr;
2215 _last_size = res;
2216 _last_was_obj = was_obj;
2217 _last_was_live = was_live;
2218 return res;
2219 }
2220 };
2221
2222 class VerifyAllOopsClosure: public OopClosure {
2223 private:
2224 const CMSCollector* _collector;
2225 const CompactibleFreeListSpace* _sp;
2226 const MemRegion _span;
2227 const bool _past_remark;
2228 const CMSBitMap* _bit_map;
2229
2230 protected:
2231 void do_oop(void* p, oop obj) {
2232 if (_span.contains(obj)) { // the interior oop points into CMS heap
2233 if (!_span.contains(p)) { // reference from outside CMS heap
2234 // Should be a valid object; the first disjunct below allows
2235 // us to sidestep an assertion in block_is_obj() that insists
2236 // that p be in _sp. Note that several generations (and spaces)
2237 // are spanned by _span (CMS heap) above.
2238 guarantee(!_sp->is_in_reserved(obj) ||
2239 _sp->block_is_obj((HeapWord*)obj),
2240 "Should be an object");
2241 guarantee(obj->is_oop(), "Should be an oop");
2242 obj->verify();
2243 if (_past_remark) {
2244 // Remark has been completed, the object should be marked
2245 _bit_map->isMarked((HeapWord*)obj);
2246 }
2247 } else { // reference within CMS heap
2248 if (_past_remark) {
2249 // Remark has been completed -- so the referent should have
2250 // been marked, if referring object is.
2251 if (_bit_map->isMarked(_collector->block_start(p))) {
2252 guarantee(_bit_map->isMarked((HeapWord*)obj), "Marking error?");
2253 }
2254 }
2255 }
2256 } else if (_sp->is_in_reserved(p)) {
2257 // the reference is from FLS, and points out of FLS
2258 guarantee(obj->is_oop(), "Should be an oop");
2259 obj->verify();
2260 }
2261 }
2262
2263 template <class T> void do_oop_work(T* p) {
2264 T heap_oop = oopDesc::load_heap_oop(p);
2265 if (!oopDesc::is_null(heap_oop)) {
2266 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
2267 do_oop(p, obj);
2268 }
2269 }
2270
2271 public:
2272 VerifyAllOopsClosure(const CMSCollector* collector,
2273 const CompactibleFreeListSpace* sp, MemRegion span,
2274 bool past_remark, CMSBitMap* bit_map) :
2275 _collector(collector), _sp(sp), _span(span),
2276 _past_remark(past_remark), _bit_map(bit_map) { }
2277
2278 virtual void do_oop(oop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2279 virtual void do_oop(narrowOop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2280 };
2281
2282 void CompactibleFreeListSpace::verify() const {
2283 assert_lock_strong(&_freelistLock);
2284 verify_objects_initialized();
2285 MemRegion span = _collector->_span;
2286 bool past_remark = (_collector->abstract_state() ==
2287 CMSCollector::Sweeping);
2288
2289 ResourceMark rm;
2290 HandleMark hm;
2291
2292 // Check integrity of CFL data structures
2293 _promoInfo.verify();
2294 _dictionary->verify();
2295 if (FLSVerifyIndexTable) {
2296 verifyIndexedFreeLists();
2297 }
2298 // Check integrity of all objects and free blocks in space
2299 {
2300 VerifyAllBlksClosure cl(this, span);
2301 ((CompactibleFreeListSpace*)this)->blk_iterate(&cl); // cast off const
2302 }
2303 // Check that all references in the heap to FLS
2304 // are to valid objects in FLS or that references in
2305 // FLS are to valid objects elsewhere in the heap
2306 if (FLSVerifyAllHeapReferences)
2307 {
2308 VerifyAllOopsClosure cl(_collector, this, span, past_remark,
2309 _collector->markBitMap());
2310
2311 // Iterate over all oops in the heap. Uses the _no_header version
2312 // since we are not interested in following the klass pointers.
2313 GenCollectedHeap::heap()->oop_iterate_no_header(&cl);
2314 }
2315
2316 if (VerifyObjectStartArray) {
2317 // Verify the block offset table
2318 _bt.verify();
2319 }
2320 }
2321
2322 #ifndef PRODUCT
2323 void CompactibleFreeListSpace::verifyFreeLists() const {
2324 if (FLSVerifyLists) {
2325 _dictionary->verify();
2326 verifyIndexedFreeLists();
2327 } else {
2328 if (FLSVerifyDictionary) {
2329 _dictionary->verify();
2330 }
2331 if (FLSVerifyIndexTable) {
2332 verifyIndexedFreeLists();
2333 }
2334 }
2335 }
2336 #endif
2337
2338 void CompactibleFreeListSpace::verifyIndexedFreeLists() const {
2339 size_t i = 0;
2340 for (; i < IndexSetStart; i++) {
2341 guarantee(_indexedFreeList[i].head() == NULL, "should be NULL");
2342 }
2343 for (; i < IndexSetSize; i++) {
2344 verifyIndexedFreeList(i);
2345 }
2346 }
2347
2348 void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const {
2349 FreeChunk* fc = _indexedFreeList[size].head();
2350 FreeChunk* tail = _indexedFreeList[size].tail();
2351 size_t num = _indexedFreeList[size].count();
2352 size_t n = 0;
2353 guarantee(((size >= IndexSetStart) && (size % IndexSetStride == 0)) || fc == NULL,
2354 "Slot should have been empty");
2355 for (; fc != NULL; fc = fc->next(), n++) {
2356 guarantee(fc->size() == size, "Size inconsistency");
2357 guarantee(fc->is_free(), "!free?");
2358 guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list");
2359 guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail");
2360 }
2361 guarantee(n == num, "Incorrect count");
2362 }
2363
2364 #ifndef PRODUCT
2365 void CompactibleFreeListSpace::check_free_list_consistency() const {
2366 assert((TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::min_size() <= IndexSetSize),
2367 "Some sizes can't be allocated without recourse to"
2368 " linear allocation buffers");
2369 assert((TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::min_size()*HeapWordSize == sizeof(TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >)),
2370 "else MIN_TREE_CHUNK_SIZE is wrong");
2371 assert(IndexSetStart != 0, "IndexSetStart not initialized");
2372 assert(IndexSetStride != 0, "IndexSetStride not initialized");
2373 }
2374 #endif
2375
2376 void CompactibleFreeListSpace::printFLCensus(size_t sweep_count) const {
2377 assert_lock_strong(&_freelistLock);
2378 LogHandle(gc, freelist, census) log;
2379 if (!log.is_debug()) {
2380 return;
2381 }
2382 AdaptiveFreeList<FreeChunk> total;
2383 log.debug("end sweep# " SIZE_FORMAT, sweep_count);
2384 ResourceMark rm;
2385 outputStream* out = log.debug_stream();
2386 AdaptiveFreeList<FreeChunk>::print_labels_on(out, "size");
2387 size_t total_free = 0;
2388 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2389 const AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2390 total_free += fl->count() * fl->size();
2391 if (i % (40*IndexSetStride) == 0) {
2392 AdaptiveFreeList<FreeChunk>::print_labels_on(out, "size");
2393 }
2394 fl->print_on(out);
2395 total.set_bfr_surp( total.bfr_surp() + fl->bfr_surp() );
2396 total.set_surplus( total.surplus() + fl->surplus() );
2397 total.set_desired( total.desired() + fl->desired() );
2398 total.set_prev_sweep( total.prev_sweep() + fl->prev_sweep() );
2399 total.set_before_sweep(total.before_sweep() + fl->before_sweep());
2400 total.set_count( total.count() + fl->count() );
2401 total.set_coal_births( total.coal_births() + fl->coal_births() );
2402 total.set_coal_deaths( total.coal_deaths() + fl->coal_deaths() );
2403 total.set_split_births(total.split_births() + fl->split_births());
2404 total.set_split_deaths(total.split_deaths() + fl->split_deaths());
2405 }
2406 total.print_on(out, "TOTAL");
2407 log.debug("Total free in indexed lists " SIZE_FORMAT " words", total_free);
2408 log.debug("growth: %8.5f deficit: %8.5f",
2409 (double)(total.split_births()+total.coal_births()-total.split_deaths()-total.coal_deaths())/
2410 (total.prev_sweep() != 0 ? (double)total.prev_sweep() : 1.0),
2411 (double)(total.desired() - total.count())/(total.desired() != 0 ? (double)total.desired() : 1.0));
2412 _dictionary->print_dict_census(out);
2413 }
2414
2415 ///////////////////////////////////////////////////////////////////////////
2416 // CFLS_LAB
2417 ///////////////////////////////////////////////////////////////////////////
2418
2419 #define VECTOR_257(x) \
2420 /* 1 2 3 4 5 6 7 8 9 1x 11 12 13 14 15 16 17 18 19 2x 21 22 23 24 25 26 27 28 29 3x 31 32 */ \
2421 { x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2422 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2423 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2424 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2425 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2426 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2427 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2428 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \
2429 x }
2430
2431 // Initialize with default setting for CMS, _not_
2432 // generic OldPLABSize, whose static default is different; if overridden at the
2433 // command-line, this will get reinitialized via a call to
2434 // modify_initialization() below.
2435 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] =
2436 VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CFLS_LAB::_default_dynamic_old_plab_size));
2437 size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0);
2438 uint CFLS_LAB::_global_num_workers[] = VECTOR_257(0);
2439
2440 CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) :
2441 _cfls(cfls)
2442 {
2443 assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above");
2444 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2445 i < CompactibleFreeListSpace::IndexSetSize;
2446 i += CompactibleFreeListSpace::IndexSetStride) {
2447 _indexedFreeList[i].set_size(i);
2448 _num_blocks[i] = 0;
2449 }
2450 }
2451
2452 static bool _CFLS_LAB_modified = false;
2453
2454 void CFLS_LAB::modify_initialization(size_t n, unsigned wt) {
2455 assert(!_CFLS_LAB_modified, "Call only once");
2456 _CFLS_LAB_modified = true;
2457 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2458 i < CompactibleFreeListSpace::IndexSetSize;
2459 i += CompactibleFreeListSpace::IndexSetStride) {
2460 _blocks_to_claim[i].modify(n, wt, true /* force */);
2461 }
2462 }
2463
2464 HeapWord* CFLS_LAB::alloc(size_t word_sz) {
2465 FreeChunk* res;
2466 assert(word_sz == _cfls->adjustObjectSize(word_sz), "Error");
2467 if (word_sz >= CompactibleFreeListSpace::IndexSetSize) {
2468 // This locking manages sync with other large object allocations.
2469 MutexLockerEx x(_cfls->parDictionaryAllocLock(),
2470 Mutex::_no_safepoint_check_flag);
2471 res = _cfls->getChunkFromDictionaryExact(word_sz);
2472 if (res == NULL) return NULL;
2473 } else {
2474 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[word_sz];
2475 if (fl->count() == 0) {
2476 // Attempt to refill this local free list.
2477 get_from_global_pool(word_sz, fl);
2478 // If it didn't work, give up.
2479 if (fl->count() == 0) return NULL;
2480 }
2481 res = fl->get_chunk_at_head();
2482 assert(res != NULL, "Why was count non-zero?");
2483 }
2484 res->markNotFree();
2485 assert(!res->is_free(), "shouldn't be marked free");
2486 assert(oop(res)->klass_or_null() == NULL, "should look uninitialized");
2487 // mangle a just allocated object with a distinct pattern.
2488 debug_only(res->mangleAllocated(word_sz));
2489 return (HeapWord*)res;
2490 }
2491
2492 // Get a chunk of blocks of the right size and update related
2493 // book-keeping stats
2494 void CFLS_LAB::get_from_global_pool(size_t word_sz, AdaptiveFreeList<FreeChunk>* fl) {
2495 // Get the #blocks we want to claim
2496 size_t n_blks = (size_t)_blocks_to_claim[word_sz].average();
2497 assert(n_blks > 0, "Error");
2498 assert(ResizeOldPLAB || n_blks == OldPLABSize, "Error");
2499 // In some cases, when the application has a phase change,
2500 // there may be a sudden and sharp shift in the object survival
2501 // profile, and updating the counts at the end of a scavenge
2502 // may not be quick enough, giving rise to large scavenge pauses
2503 // during these phase changes. It is beneficial to detect such
2504 // changes on-the-fly during a scavenge and avoid such a phase-change
2505 // pothole. The following code is a heuristic attempt to do that.
2506 // It is protected by a product flag until we have gained
2507 // enough experience with this heuristic and fine-tuned its behavior.
2508 // WARNING: This might increase fragmentation if we overreact to
2509 // small spikes, so some kind of historical smoothing based on
2510 // previous experience with the greater reactivity might be useful.
2511 // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by
2512 // default.
2513 if (ResizeOldPLAB && CMSOldPLABResizeQuicker) {
2514 size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks);
2515 n_blks += CMSOldPLABReactivityFactor*multiple*n_blks;
2516 n_blks = MIN2(n_blks, CMSOldPLABMax);
2517 }
2518 assert(n_blks > 0, "Error");
2519 _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl);
2520 // Update stats table entry for this block size
2521 _num_blocks[word_sz] += fl->count();
2522 }
2523
2524 void CFLS_LAB::compute_desired_plab_size() {
2525 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2526 i < CompactibleFreeListSpace::IndexSetSize;
2527 i += CompactibleFreeListSpace::IndexSetStride) {
2528 assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0),
2529 "Counter inconsistency");
2530 if (_global_num_workers[i] > 0) {
2531 // Need to smooth wrt historical average
2532 if (ResizeOldPLAB) {
2533 _blocks_to_claim[i].sample(
2534 MAX2(CMSOldPLABMin,
2535 MIN2(CMSOldPLABMax,
2536 _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills))));
2537 }
2538 // Reset counters for next round
2539 _global_num_workers[i] = 0;
2540 _global_num_blocks[i] = 0;
2541 log_trace(gc, plab)("[" SIZE_FORMAT "]: " SIZE_FORMAT, i, (size_t)_blocks_to_claim[i].average());
2542 }
2543 }
2544 }
2545
2546 // If this is changed in the future to allow parallel
2547 // access, one would need to take the FL locks and,
2548 // depending on how it is used, stagger access from
2549 // parallel threads to reduce contention.
2550 void CFLS_LAB::retire(int tid) {
2551 // We run this single threaded with the world stopped;
2552 // so no need for locks and such.
2553 NOT_PRODUCT(Thread* t = Thread::current();)
2554 assert(Thread::current()->is_VM_thread(), "Error");
2555 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2556 i < CompactibleFreeListSpace::IndexSetSize;
2557 i += CompactibleFreeListSpace::IndexSetStride) {
2558 assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(),
2559 "Can't retire more than what we obtained");
2560 if (_num_blocks[i] > 0) {
2561 size_t num_retire = _indexedFreeList[i].count();
2562 assert(_num_blocks[i] > num_retire, "Should have used at least one");
2563 {
2564 // MutexLockerEx x(_cfls->_indexedFreeListParLocks[i],
2565 // Mutex::_no_safepoint_check_flag);
2566
2567 // Update globals stats for num_blocks used
2568 _global_num_blocks[i] += (_num_blocks[i] - num_retire);
2569 _global_num_workers[i]++;
2570 assert(_global_num_workers[i] <= ParallelGCThreads, "Too big");
2571 if (num_retire > 0) {
2572 _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]);
2573 // Reset this list.
2574 _indexedFreeList[i] = AdaptiveFreeList<FreeChunk>();
2575 _indexedFreeList[i].set_size(i);
2576 }
2577 }
2578 log_trace(gc, plab)("%d[" SIZE_FORMAT "]: " SIZE_FORMAT "/" SIZE_FORMAT "/" SIZE_FORMAT,
2579 tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average());
2580 // Reset stats for next round
2581 _num_blocks[i] = 0;
2582 }
2583 }
2584 }
2585
2586 // Used by par_get_chunk_of_blocks() for the chunks from the
2587 // indexed_free_lists. Looks for a chunk with size that is a multiple
2588 // of "word_sz" and if found, splits it into "word_sz" chunks and add
2589 // to the free list "fl". "n" is the maximum number of chunks to
2590 // be added to "fl".
2591 bool CompactibleFreeListSpace:: par_get_chunk_of_blocks_IFL(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl) {
2592
2593 // We'll try all multiples of word_sz in the indexed set, starting with
2594 // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples,
2595 // then try getting a big chunk and splitting it.
2596 {
2597 bool found;
2598 int k;
2599 size_t cur_sz;
2600 for (k = 1, cur_sz = k * word_sz, found = false;
2601 (cur_sz < CompactibleFreeListSpace::IndexSetSize) &&
2602 (CMSSplitIndexedFreeListBlocks || k <= 1);
2603 k++, cur_sz = k * word_sz) {
2604 AdaptiveFreeList<FreeChunk> fl_for_cur_sz; // Empty.
2605 fl_for_cur_sz.set_size(cur_sz);
2606 {
2607 MutexLockerEx x(_indexedFreeListParLocks[cur_sz],
2608 Mutex::_no_safepoint_check_flag);
2609 AdaptiveFreeList<FreeChunk>* gfl = &_indexedFreeList[cur_sz];
2610 if (gfl->count() != 0) {
2611 // nn is the number of chunks of size cur_sz that
2612 // we'd need to split k-ways each, in order to create
2613 // "n" chunks of size word_sz each.
2614 const size_t nn = MAX2(n/k, (size_t)1);
2615 gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz);
2616 found = true;
2617 if (k > 1) {
2618 // Update split death stats for the cur_sz-size blocks list:
2619 // we increment the split death count by the number of blocks
2620 // we just took from the cur_sz-size blocks list and which
2621 // we will be splitting below.
2622 ssize_t deaths = gfl->split_deaths() +
2623 fl_for_cur_sz.count();
2624 gfl->set_split_deaths(deaths);
2625 }
2626 }
2627 }
2628 // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1.
2629 if (found) {
2630 if (k == 1) {
2631 fl->prepend(&fl_for_cur_sz);
2632 } else {
2633 // Divide each block on fl_for_cur_sz up k ways.
2634 FreeChunk* fc;
2635 while ((fc = fl_for_cur_sz.get_chunk_at_head()) != NULL) {
2636 // Must do this in reverse order, so that anybody attempting to
2637 // access the main chunk sees it as a single free block until we
2638 // change it.
2639 size_t fc_size = fc->size();
2640 assert(fc->is_free(), "Error");
2641 for (int i = k-1; i >= 0; i--) {
2642 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2643 assert((i != 0) ||
2644 ((fc == ffc) && ffc->is_free() &&
2645 (ffc->size() == k*word_sz) && (fc_size == word_sz)),
2646 "Counting error");
2647 ffc->set_size(word_sz);
2648 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2649 ffc->link_next(NULL);
2650 // Above must occur before BOT is updated below.
2651 OrderAccess::storestore();
2652 // splitting from the right, fc_size == i * word_sz
2653 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2654 fc_size -= word_sz;
2655 assert(fc_size == i*word_sz, "Error");
2656 _bt.verify_not_unallocated((HeapWord*)ffc, word_sz);
2657 _bt.verify_single_block((HeapWord*)fc, fc_size);
2658 _bt.verify_single_block((HeapWord*)ffc, word_sz);
2659 // Push this on "fl".
2660 fl->return_chunk_at_head(ffc);
2661 }
2662 // TRAP
2663 assert(fl->tail()->next() == NULL, "List invariant.");
2664 }
2665 }
2666 // Update birth stats for this block size.
2667 size_t num = fl->count();
2668 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2669 Mutex::_no_safepoint_check_flag);
2670 ssize_t births = _indexedFreeList[word_sz].split_births() + num;
2671 _indexedFreeList[word_sz].set_split_births(births);
2672 return true;
2673 }
2674 }
2675 return found;
2676 }
2677 }
2678
2679 FreeChunk* CompactibleFreeListSpace::get_n_way_chunk_to_split(size_t word_sz, size_t n) {
2680
2681 FreeChunk* fc = NULL;
2682 FreeChunk* rem_fc = NULL;
2683 size_t rem;
2684 {
2685 MutexLockerEx x(parDictionaryAllocLock(),
2686 Mutex::_no_safepoint_check_flag);
2687 while (n > 0) {
2688 fc = dictionary()->get_chunk(MAX2(n * word_sz, _dictionary->min_size()),
2689 FreeBlockDictionary<FreeChunk>::atLeast);
2690 if (fc != NULL) {
2691 break;
2692 } else {
2693 n--;
2694 }
2695 }
2696 if (fc == NULL) return NULL;
2697 // Otherwise, split up that block.
2698 assert((ssize_t)n >= 1, "Control point invariant");
2699 assert(fc->is_free(), "Error: should be a free block");
2700 _bt.verify_single_block((HeapWord*)fc, fc->size());
2701 const size_t nn = fc->size() / word_sz;
2702 n = MIN2(nn, n);
2703 assert((ssize_t)n >= 1, "Control point invariant");
2704 rem = fc->size() - n * word_sz;
2705 // If there is a remainder, and it's too small, allocate one fewer.
2706 if (rem > 0 && rem < MinChunkSize) {
2707 n--; rem += word_sz;
2708 }
2709 // Note that at this point we may have n == 0.
2710 assert((ssize_t)n >= 0, "Control point invariant");
2711
2712 // If n is 0, the chunk fc that was found is not large
2713 // enough to leave a viable remainder. We are unable to
2714 // allocate even one block. Return fc to the
2715 // dictionary and return, leaving "fl" empty.
2716 if (n == 0) {
2717 returnChunkToDictionary(fc);
2718 return NULL;
2719 }
2720
2721 _bt.allocated((HeapWord*)fc, fc->size(), true /* reducing */); // update _unallocated_blk
2722 dictionary()->dict_census_update(fc->size(),
2723 true /*split*/,
2724 false /*birth*/);
2725
2726 // First return the remainder, if any.
2727 // Note that we hold the lock until we decide if we're going to give
2728 // back the remainder to the dictionary, since a concurrent allocation
2729 // may otherwise see the heap as empty. (We're willing to take that
2730 // hit if the block is a small block.)
2731 if (rem > 0) {
2732 size_t prefix_size = n * word_sz;
2733 rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size);
2734 rem_fc->set_size(rem);
2735 rem_fc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2736 rem_fc->link_next(NULL);
2737 // Above must occur before BOT is updated below.
2738 assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error");
2739 OrderAccess::storestore();
2740 _bt.split_block((HeapWord*)fc, fc->size(), prefix_size);
2741 assert(fc->is_free(), "Error");
2742 fc->set_size(prefix_size);
2743 if (rem >= IndexSetSize) {
2744 returnChunkToDictionary(rem_fc);
2745 dictionary()->dict_census_update(rem, true /*split*/, true /*birth*/);
2746 rem_fc = NULL;
2747 }
2748 // Otherwise, return it to the small list below.
2749 }
2750 }
2751 if (rem_fc != NULL) {
2752 MutexLockerEx x(_indexedFreeListParLocks[rem],
2753 Mutex::_no_safepoint_check_flag);
2754 _bt.verify_not_unallocated((HeapWord*)rem_fc, rem_fc->size());
2755 _indexedFreeList[rem].return_chunk_at_head(rem_fc);
2756 smallSplitBirth(rem);
2757 }
2758 assert(n * word_sz == fc->size(),
2759 "Chunk size " SIZE_FORMAT " is not exactly splittable by "
2760 SIZE_FORMAT " sized chunks of size " SIZE_FORMAT,
2761 fc->size(), n, word_sz);
2762 return fc;
2763 }
2764
2765 void CompactibleFreeListSpace:: par_get_chunk_of_blocks_dictionary(size_t word_sz, size_t targetted_number_of_chunks, AdaptiveFreeList<FreeChunk>* fl) {
2766
2767 FreeChunk* fc = get_n_way_chunk_to_split(word_sz, targetted_number_of_chunks);
2768
2769 if (fc == NULL) {
2770 return;
2771 }
2772
2773 size_t n = fc->size() / word_sz;
2774
2775 assert((ssize_t)n > 0, "Consistency");
2776 // Now do the splitting up.
2777 // Must do this in reverse order, so that anybody attempting to
2778 // access the main chunk sees it as a single free block until we
2779 // change it.
2780 size_t fc_size = n * word_sz;
2781 // All but first chunk in this loop
2782 for (ssize_t i = n-1; i > 0; i--) {
2783 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2784 ffc->set_size(word_sz);
2785 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2786 ffc->link_next(NULL);
2787 // Above must occur before BOT is updated below.
2788 OrderAccess::storestore();
2789 // splitting from the right, fc_size == (n - i + 1) * wordsize
2790 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2791 fc_size -= word_sz;
2792 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size());
2793 _bt.verify_single_block((HeapWord*)ffc, ffc->size());
2794 _bt.verify_single_block((HeapWord*)fc, fc_size);
2795 // Push this on "fl".
2796 fl->return_chunk_at_head(ffc);
2797 }
2798 // First chunk
2799 assert(fc->is_free() && fc->size() == n*word_sz, "Error: should still be a free block");
2800 // The blocks above should show their new sizes before the first block below
2801 fc->set_size(word_sz);
2802 fc->link_prev(NULL); // idempotent wrt free-ness, see assert above
2803 fc->link_next(NULL);
2804 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
2805 _bt.verify_single_block((HeapWord*)fc, fc->size());
2806 fl->return_chunk_at_head(fc);
2807
2808 assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks");
2809 {
2810 // Update the stats for this block size.
2811 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2812 Mutex::_no_safepoint_check_flag);
2813 const ssize_t births = _indexedFreeList[word_sz].split_births() + n;
2814 _indexedFreeList[word_sz].set_split_births(births);
2815 // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n;
2816 // _indexedFreeList[word_sz].set_surplus(new_surplus);
2817 }
2818
2819 // TRAP
2820 assert(fl->tail()->next() == NULL, "List invariant.");
2821 }
2822
2823 void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl) {
2824 assert(fl->count() == 0, "Precondition.");
2825 assert(word_sz < CompactibleFreeListSpace::IndexSetSize,
2826 "Precondition");
2827
2828 if (par_get_chunk_of_blocks_IFL(word_sz, n, fl)) {
2829 // Got it
2830 return;
2831 }
2832
2833 // Otherwise, we'll split a block from the dictionary.
2834 par_get_chunk_of_blocks_dictionary(word_sz, n, fl);
2835 }
2836
2837 // Set up the space's par_seq_tasks structure for work claiming
2838 // for parallel rescan. See CMSParRemarkTask where this is currently used.
2839 // XXX Need to suitably abstract and generalize this and the next
2840 // method into one.
2841 void
2842 CompactibleFreeListSpace::
2843 initialize_sequential_subtasks_for_rescan(int n_threads) {
2844 // The "size" of each task is fixed according to rescan_task_size.
2845 assert(n_threads > 0, "Unexpected n_threads argument");
2846 const size_t task_size = rescan_task_size();
2847 size_t n_tasks = (used_region().word_size() + task_size - 1)/task_size;
2848 assert((n_tasks == 0) == used_region().is_empty(), "n_tasks incorrect");
2849 assert(n_tasks == 0 ||
2850 ((used_region().start() + (n_tasks - 1)*task_size < used_region().end()) &&
2851 (used_region().start() + n_tasks*task_size >= used_region().end())),
2852 "n_tasks calculation incorrect");
2853 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2854 assert(!pst->valid(), "Clobbering existing data?");
2855 // Sets the condition for completion of the subtask (how many threads
2856 // need to finish in order to be done).
2857 pst->set_n_threads(n_threads);
2858 pst->set_n_tasks((int)n_tasks);
2859 }
2860
2861 // Set up the space's par_seq_tasks structure for work claiming
2862 // for parallel concurrent marking. See CMSConcMarkTask where this is currently used.
2863 void
2864 CompactibleFreeListSpace::
2865 initialize_sequential_subtasks_for_marking(int n_threads,
2866 HeapWord* low) {
2867 // The "size" of each task is fixed according to rescan_task_size.
2868 assert(n_threads > 0, "Unexpected n_threads argument");
2869 const size_t task_size = marking_task_size();
2870 assert(task_size > CardTableModRefBS::card_size_in_words &&
2871 (task_size % CardTableModRefBS::card_size_in_words == 0),
2872 "Otherwise arithmetic below would be incorrect");
2873 MemRegion span = _old_gen->reserved();
2874 if (low != NULL) {
2875 if (span.contains(low)) {
2876 // Align low down to a card boundary so that
2877 // we can use block_offset_careful() on span boundaries.
2878 HeapWord* aligned_low = (HeapWord*)align_size_down((uintptr_t)low,
2879 CardTableModRefBS::card_size);
2880 // Clip span prefix at aligned_low
2881 span = span.intersection(MemRegion(aligned_low, span.end()));
2882 } else if (low > span.end()) {
2883 span = MemRegion(low, low); // Null region
2884 } // else use entire span
2885 }
2886 assert(span.is_empty() ||
2887 ((uintptr_t)span.start() % CardTableModRefBS::card_size == 0),
2888 "span should start at a card boundary");
2889 size_t n_tasks = (span.word_size() + task_size - 1)/task_size;
2890 assert((n_tasks == 0) == span.is_empty(), "Inconsistency");
2891 assert(n_tasks == 0 ||
2892 ((span.start() + (n_tasks - 1)*task_size < span.end()) &&
2893 (span.start() + n_tasks*task_size >= span.end())),
2894 "n_tasks calculation incorrect");
2895 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2896 assert(!pst->valid(), "Clobbering existing data?");
2897 // Sets the condition for completion of the subtask (how many threads
2898 // need to finish in order to be done).
2899 pst->set_n_threads(n_threads);
2900 pst->set_n_tasks((int)n_tasks);
2901 }