1 /*
2 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 # include "incls/_precompiled.incl"
26 # include "incls/_compactibleFreeListSpace.cpp.incl"
27
28 /////////////////////////////////////////////////////////////////////////
29 //// CompactibleFreeListSpace
30 /////////////////////////////////////////////////////////////////////////
31
32 // highest ranked free list lock rank
33 int CompactibleFreeListSpace::_lockRank = Mutex::leaf + 3;
34
35 // Defaults are 0 so things will break badly if incorrectly initialized.
36 int CompactibleFreeListSpace::IndexSetStart = 0;
37 int CompactibleFreeListSpace::IndexSetStride = 0;
38
39 size_t MinChunkSize = 0;
40
41 void CompactibleFreeListSpace::set_cms_values() {
42 // Set CMS global values
43 assert(MinChunkSize == 0, "already set");
44 #define numQuanta(x,y) ((x+y-1)/y)
45 MinChunkSize = numQuanta(sizeof(FreeChunk), MinObjAlignmentInBytes) * MinObjAlignment;
46
47 assert(IndexSetStart == 0 && IndexSetStride == 0, "already set");
48 IndexSetStart = MinObjAlignment;
49 IndexSetStride = MinObjAlignment;
50 }
51
52 // Constructor
53 CompactibleFreeListSpace::CompactibleFreeListSpace(BlockOffsetSharedArray* bs,
54 MemRegion mr, bool use_adaptive_freelists,
55 FreeBlockDictionary::DictionaryChoice dictionaryChoice) :
56 _dictionaryChoice(dictionaryChoice),
57 _adaptive_freelists(use_adaptive_freelists),
58 _bt(bs, mr),
59 // free list locks are in the range of values taken by _lockRank
60 // This range currently is [_leaf+2, _leaf+3]
61 // Note: this requires that CFLspace c'tors
62 // are called serially in the order in which the locks are
63 // are acquired in the program text. This is true today.
64 _freelistLock(_lockRank--, "CompactibleFreeListSpace._lock", true),
65 _parDictionaryAllocLock(Mutex::leaf - 1, // == rank(ExpandHeap_lock) - 1
66 "CompactibleFreeListSpace._dict_par_lock", true),
67 _rescan_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
68 CMSRescanMultiple),
69 _marking_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
70 CMSConcMarkMultiple),
71 _collector(NULL)
72 {
73 _bt.set_space(this);
74 initialize(mr, SpaceDecorator::Clear, SpaceDecorator::Mangle);
75 // We have all of "mr", all of which we place in the dictionary
76 // as one big chunk. We'll need to decide here which of several
77 // possible alternative dictionary implementations to use. For
78 // now the choice is easy, since we have only one working
79 // implementation, namely, the simple binary tree (splaying
80 // temporarily disabled).
81 switch (dictionaryChoice) {
82 case FreeBlockDictionary::dictionarySplayTree:
83 case FreeBlockDictionary::dictionarySkipList:
84 default:
85 warning("dictionaryChoice: selected option not understood; using"
86 " default BinaryTreeDictionary implementation instead.");
87 case FreeBlockDictionary::dictionaryBinaryTree:
88 _dictionary = new BinaryTreeDictionary(mr);
89 break;
90 }
91 assert(_dictionary != NULL, "CMS dictionary initialization");
92 // The indexed free lists are initially all empty and are lazily
93 // filled in on demand. Initialize the array elements to NULL.
94 initializeIndexedFreeListArray();
95
96 // Not using adaptive free lists assumes that allocation is first
97 // from the linAB's. Also a cms perm gen which can be compacted
98 // has to have the klass's klassKlass allocated at a lower
99 // address in the heap than the klass so that the klassKlass is
100 // moved to its new location before the klass is moved.
101 // Set the _refillSize for the linear allocation blocks
102 if (!use_adaptive_freelists) {
103 FreeChunk* fc = _dictionary->getChunk(mr.word_size());
104 // The small linAB initially has all the space and will allocate
105 // a chunk of any size.
106 HeapWord* addr = (HeapWord*) fc;
107 _smallLinearAllocBlock.set(addr, fc->size() ,
108 1024*SmallForLinearAlloc, fc->size());
109 // Note that _unallocated_block is not updated here.
110 // Allocations from the linear allocation block should
111 // update it.
112 } else {
113 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc,
114 SmallForLinearAlloc);
115 }
116 // CMSIndexedFreeListReplenish should be at least 1
117 CMSIndexedFreeListReplenish = MAX2((uintx)1, CMSIndexedFreeListReplenish);
118 _promoInfo.setSpace(this);
119 if (UseCMSBestFit) {
120 _fitStrategy = FreeBlockBestFitFirst;
121 } else {
122 _fitStrategy = FreeBlockStrategyNone;
123 }
124 checkFreeListConsistency();
125
126 // Initialize locks for parallel case.
127
128 if (CollectedHeap::use_parallel_gc_threads()) {
129 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
130 _indexedFreeListParLocks[i] = new Mutex(Mutex::leaf - 1, // == ExpandHeap_lock - 1
131 "a freelist par lock",
132 true);
133 if (_indexedFreeListParLocks[i] == NULL)
134 vm_exit_during_initialization("Could not allocate a par lock");
135 DEBUG_ONLY(
136 _indexedFreeList[i].set_protecting_lock(_indexedFreeListParLocks[i]);
137 )
138 }
139 _dictionary->set_par_lock(&_parDictionaryAllocLock);
140 }
141 }
142
143 // Like CompactibleSpace forward() but always calls cross_threshold() to
144 // update the block offset table. Removed initialize_threshold call because
145 // CFLS does not use a block offset array for contiguous spaces.
146 HeapWord* CompactibleFreeListSpace::forward(oop q, size_t size,
147 CompactPoint* cp, HeapWord* compact_top) {
148 // q is alive
149 // First check if we should switch compaction space
150 assert(this == cp->space, "'this' should be current compaction space.");
151 size_t compaction_max_size = pointer_delta(end(), compact_top);
152 assert(adjustObjectSize(size) == cp->space->adjust_object_size_v(size),
153 "virtual adjustObjectSize_v() method is not correct");
154 size_t adjusted_size = adjustObjectSize(size);
155 assert(compaction_max_size >= MinChunkSize || compaction_max_size == 0,
156 "no small fragments allowed");
157 assert(minimum_free_block_size() == MinChunkSize,
158 "for de-virtualized reference below");
159 // Can't leave a nonzero size, residual fragment smaller than MinChunkSize
160 if (adjusted_size + MinChunkSize > compaction_max_size &&
161 adjusted_size != compaction_max_size) {
162 do {
163 // switch to next compaction space
164 cp->space->set_compaction_top(compact_top);
165 cp->space = cp->space->next_compaction_space();
166 if (cp->space == NULL) {
167 cp->gen = GenCollectedHeap::heap()->prev_gen(cp->gen);
168 assert(cp->gen != NULL, "compaction must succeed");
169 cp->space = cp->gen->first_compaction_space();
170 assert(cp->space != NULL, "generation must have a first compaction space");
171 }
172 compact_top = cp->space->bottom();
173 cp->space->set_compaction_top(compact_top);
174 // The correct adjusted_size may not be the same as that for this method
175 // (i.e., cp->space may no longer be "this" so adjust the size again.
176 // Use the virtual method which is not used above to save the virtual
177 // dispatch.
178 adjusted_size = cp->space->adjust_object_size_v(size);
179 compaction_max_size = pointer_delta(cp->space->end(), compact_top);
180 assert(cp->space->minimum_free_block_size() == 0, "just checking");
181 } while (adjusted_size > compaction_max_size);
182 }
183
184 // store the forwarding pointer into the mark word
185 if ((HeapWord*)q != compact_top) {
186 q->forward_to(oop(compact_top));
187 assert(q->is_gc_marked(), "encoding the pointer should preserve the mark");
188 } else {
189 // if the object isn't moving we can just set the mark to the default
190 // mark and handle it specially later on.
191 q->init_mark();
192 assert(q->forwardee() == NULL, "should be forwarded to NULL");
193 }
194
195 VALIDATE_MARK_SWEEP_ONLY(MarkSweep::register_live_oop(q, adjusted_size));
196 compact_top += adjusted_size;
197
198 // we need to update the offset table so that the beginnings of objects can be
199 // found during scavenge. Note that we are updating the offset table based on
200 // where the object will be once the compaction phase finishes.
201
202 // Always call cross_threshold(). A contiguous space can only call it when
203 // the compaction_top exceeds the current threshold but not for an
204 // non-contiguous space.
205 cp->threshold =
206 cp->space->cross_threshold(compact_top - adjusted_size, compact_top);
207 return compact_top;
208 }
209
210 // A modified copy of OffsetTableContigSpace::cross_threshold() with _offsets -> _bt
211 // and use of single_block instead of alloc_block. The name here is not really
212 // appropriate - maybe a more general name could be invented for both the
213 // contiguous and noncontiguous spaces.
214
215 HeapWord* CompactibleFreeListSpace::cross_threshold(HeapWord* start, HeapWord* the_end) {
216 _bt.single_block(start, the_end);
217 return end();
218 }
219
220 // Initialize them to NULL.
221 void CompactibleFreeListSpace::initializeIndexedFreeListArray() {
222 for (size_t i = 0; i < IndexSetSize; i++) {
223 // Note that on platforms where objects are double word aligned,
224 // the odd array elements are not used. It is convenient, however,
225 // to map directly from the object size to the array element.
226 _indexedFreeList[i].reset(IndexSetSize);
227 _indexedFreeList[i].set_size(i);
228 assert(_indexedFreeList[i].count() == 0, "reset check failed");
229 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
230 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
231 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
232 }
233 }
234
235 void CompactibleFreeListSpace::resetIndexedFreeListArray() {
236 for (int i = 1; i < IndexSetSize; i++) {
237 assert(_indexedFreeList[i].size() == (size_t) i,
238 "Indexed free list sizes are incorrect");
239 _indexedFreeList[i].reset(IndexSetSize);
240 assert(_indexedFreeList[i].count() == 0, "reset check failed");
241 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
242 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
243 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
244 }
245 }
246
247 void CompactibleFreeListSpace::reset(MemRegion mr) {
248 resetIndexedFreeListArray();
249 dictionary()->reset();
250 if (BlockOffsetArrayUseUnallocatedBlock) {
251 assert(end() == mr.end(), "We are compacting to the bottom of CMS gen");
252 // Everything's allocated until proven otherwise.
253 _bt.set_unallocated_block(end());
254 }
255 if (!mr.is_empty()) {
256 assert(mr.word_size() >= MinChunkSize, "Chunk size is too small");
257 _bt.single_block(mr.start(), mr.word_size());
258 FreeChunk* fc = (FreeChunk*) mr.start();
259 fc->setSize(mr.word_size());
260 if (mr.word_size() >= IndexSetSize ) {
261 returnChunkToDictionary(fc);
262 } else {
263 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
264 _indexedFreeList[mr.word_size()].returnChunkAtHead(fc);
265 }
266 }
267 _promoInfo.reset();
268 _smallLinearAllocBlock._ptr = NULL;
269 _smallLinearAllocBlock._word_size = 0;
270 }
271
272 void CompactibleFreeListSpace::reset_after_compaction() {
273 // Reset the space to the new reality - one free chunk.
274 MemRegion mr(compaction_top(), end());
275 reset(mr);
276 // Now refill the linear allocation block(s) if possible.
277 if (_adaptive_freelists) {
278 refillLinearAllocBlocksIfNeeded();
279 } else {
280 // Place as much of mr in the linAB as we can get,
281 // provided it was big enough to go into the dictionary.
282 FreeChunk* fc = dictionary()->findLargestDict();
283 if (fc != NULL) {
284 assert(fc->size() == mr.word_size(),
285 "Why was the chunk broken up?");
286 removeChunkFromDictionary(fc);
287 HeapWord* addr = (HeapWord*) fc;
288 _smallLinearAllocBlock.set(addr, fc->size() ,
289 1024*SmallForLinearAlloc, fc->size());
290 // Note that _unallocated_block is not updated here.
291 }
292 }
293 }
294
295 // Walks the entire dictionary, returning a coterminal
296 // chunk, if it exists. Use with caution since it involves
297 // a potentially complete walk of a potentially large tree.
298 FreeChunk* CompactibleFreeListSpace::find_chunk_at_end() {
299
300 assert_lock_strong(&_freelistLock);
301
302 return dictionary()->find_chunk_ends_at(end());
303 }
304
305
306 #ifndef PRODUCT
307 void CompactibleFreeListSpace::initializeIndexedFreeListArrayReturnedBytes() {
308 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
309 _indexedFreeList[i].allocation_stats()->set_returnedBytes(0);
310 }
311 }
312
313 size_t CompactibleFreeListSpace::sumIndexedFreeListArrayReturnedBytes() {
314 size_t sum = 0;
315 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
316 sum += _indexedFreeList[i].allocation_stats()->returnedBytes();
317 }
318 return sum;
319 }
320
321 size_t CompactibleFreeListSpace::totalCountInIndexedFreeLists() const {
322 size_t count = 0;
323 for (int i = (int)MinChunkSize; i < IndexSetSize; i++) {
324 debug_only(
325 ssize_t total_list_count = 0;
326 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
327 fc = fc->next()) {
328 total_list_count++;
329 }
330 assert(total_list_count == _indexedFreeList[i].count(),
331 "Count in list is incorrect");
332 )
333 count += _indexedFreeList[i].count();
334 }
335 return count;
336 }
337
338 size_t CompactibleFreeListSpace::totalCount() {
339 size_t num = totalCountInIndexedFreeLists();
340 num += dictionary()->totalCount();
341 if (_smallLinearAllocBlock._word_size != 0) {
342 num++;
343 }
344 return num;
345 }
346 #endif
347
348 bool CompactibleFreeListSpace::is_free_block(const HeapWord* p) const {
349 FreeChunk* fc = (FreeChunk*) p;
350 return fc->isFree();
351 }
352
353 size_t CompactibleFreeListSpace::used() const {
354 return capacity() - free();
355 }
356
357 size_t CompactibleFreeListSpace::free() const {
358 // "MT-safe, but not MT-precise"(TM), if you will: i.e.
359 // if you do this while the structures are in flux you
360 // may get an approximate answer only; for instance
361 // because there is concurrent allocation either
362 // directly by mutators or for promotion during a GC.
363 // It's "MT-safe", however, in the sense that you are guaranteed
364 // not to crash and burn, for instance, because of walking
365 // pointers that could disappear as you were walking them.
366 // The approximation is because the various components
367 // that are read below are not read atomically (and
368 // further the computation of totalSizeInIndexedFreeLists()
369 // is itself a non-atomic computation. The normal use of
370 // this is during a resize operation at the end of GC
371 // and at that time you are guaranteed to get the
372 // correct actual value. However, for instance, this is
373 // also read completely asynchronously by the "perf-sampler"
374 // that supports jvmstat, and you are apt to see the values
375 // flicker in such cases.
376 assert(_dictionary != NULL, "No _dictionary?");
377 return (_dictionary->totalChunkSize(DEBUG_ONLY(freelistLock())) +
378 totalSizeInIndexedFreeLists() +
379 _smallLinearAllocBlock._word_size) * HeapWordSize;
380 }
381
382 size_t CompactibleFreeListSpace::max_alloc_in_words() const {
383 assert(_dictionary != NULL, "No _dictionary?");
384 assert_locked();
385 size_t res = _dictionary->maxChunkSize();
386 res = MAX2(res, MIN2(_smallLinearAllocBlock._word_size,
387 (size_t) SmallForLinearAlloc - 1));
388 // XXX the following could potentially be pretty slow;
389 // should one, pesimally for the rare cases when res
390 // caclulated above is less than IndexSetSize,
391 // just return res calculated above? My reasoning was that
392 // those cases will be so rare that the extra time spent doesn't
393 // really matter....
394 // Note: do not change the loop test i >= res + IndexSetStride
395 // to i > res below, because i is unsigned and res may be zero.
396 for (size_t i = IndexSetSize - 1; i >= res + IndexSetStride;
397 i -= IndexSetStride) {
398 if (_indexedFreeList[i].head() != NULL) {
399 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
400 return i;
401 }
402 }
403 return res;
404 }
405
406 void LinearAllocBlock::print_on(outputStream* st) const {
407 st->print_cr(" LinearAllocBlock: ptr = " PTR_FORMAT ", word_size = " SIZE_FORMAT
408 ", refillsize = " SIZE_FORMAT ", allocation_size_limit = " SIZE_FORMAT,
409 _ptr, _word_size, _refillSize, _allocation_size_limit);
410 }
411
412 void CompactibleFreeListSpace::print_on(outputStream* st) const {
413 st->print_cr("COMPACTIBLE FREELIST SPACE");
414 st->print_cr(" Space:");
415 Space::print_on(st);
416
417 st->print_cr("promoInfo:");
418 _promoInfo.print_on(st);
419
420 st->print_cr("_smallLinearAllocBlock");
421 _smallLinearAllocBlock.print_on(st);
422
423 // dump_memory_block(_smallLinearAllocBlock->_ptr, 128);
424
425 st->print_cr(" _fitStrategy = %s, _adaptive_freelists = %s",
426 _fitStrategy?"true":"false", _adaptive_freelists?"true":"false");
427 }
428
429 void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st)
430 const {
431 reportIndexedFreeListStatistics();
432 gclog_or_tty->print_cr("Layout of Indexed Freelists");
433 gclog_or_tty->print_cr("---------------------------");
434 FreeList::print_labels_on(st, "size");
435 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
436 _indexedFreeList[i].print_on(gclog_or_tty);
437 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
438 fc = fc->next()) {
439 gclog_or_tty->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s",
440 fc, (HeapWord*)fc + i,
441 fc->cantCoalesce() ? "\t CC" : "");
442 }
443 }
444 }
445
446 void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st)
447 const {
448 _promoInfo.print_on(st);
449 }
450
451 void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st)
452 const {
453 _dictionary->reportStatistics();
454 st->print_cr("Layout of Freelists in Tree");
455 st->print_cr("---------------------------");
456 _dictionary->print_free_lists(st);
457 }
458
459 class BlkPrintingClosure: public BlkClosure {
460 const CMSCollector* _collector;
461 const CompactibleFreeListSpace* _sp;
462 const CMSBitMap* _live_bit_map;
463 const bool _post_remark;
464 outputStream* _st;
465 public:
466 BlkPrintingClosure(const CMSCollector* collector,
467 const CompactibleFreeListSpace* sp,
468 const CMSBitMap* live_bit_map,
469 outputStream* st):
470 _collector(collector),
471 _sp(sp),
472 _live_bit_map(live_bit_map),
473 _post_remark(collector->abstract_state() > CMSCollector::FinalMarking),
474 _st(st) { }
475 size_t do_blk(HeapWord* addr);
476 };
477
478 size_t BlkPrintingClosure::do_blk(HeapWord* addr) {
479 size_t sz = _sp->block_size_no_stall(addr, _collector);
480 assert(sz != 0, "Should always be able to compute a size");
481 if (_sp->block_is_obj(addr)) {
482 const bool dead = _post_remark && !_live_bit_map->isMarked(addr);
483 _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s",
484 addr,
485 dead ? "dead" : "live",
486 sz,
487 (!dead && CMSPrintObjectsInDump) ? ":" : ".");
488 if (CMSPrintObjectsInDump && !dead) {
489 oop(addr)->print_on(_st);
490 _st->print_cr("--------------------------------------");
491 }
492 } else { // free block
493 _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s",
494 addr, sz, CMSPrintChunksInDump ? ":" : ".");
495 if (CMSPrintChunksInDump) {
496 ((FreeChunk*)addr)->print_on(_st);
497 _st->print_cr("--------------------------------------");
498 }
499 }
500 return sz;
501 }
502
503 void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c,
504 outputStream* st) {
505 st->print_cr("\n=========================");
506 st->print_cr("Block layout in CMS Heap:");
507 st->print_cr("=========================");
508 BlkPrintingClosure bpcl(c, this, c->markBitMap(), st);
509 blk_iterate(&bpcl);
510
511 st->print_cr("\n=======================================");
512 st->print_cr("Order & Layout of Promotion Info Blocks");
513 st->print_cr("=======================================");
514 print_promo_info_blocks(st);
515
516 st->print_cr("\n===========================");
517 st->print_cr("Order of Indexed Free Lists");
518 st->print_cr("=========================");
519 print_indexed_free_lists(st);
520
521 st->print_cr("\n=================================");
522 st->print_cr("Order of Free Lists in Dictionary");
523 st->print_cr("=================================");
524 print_dictionary_free_lists(st);
525 }
526
527
528 void CompactibleFreeListSpace::reportFreeListStatistics() const {
529 assert_lock_strong(&_freelistLock);
530 assert(PrintFLSStatistics != 0, "Reporting error");
531 _dictionary->reportStatistics();
532 if (PrintFLSStatistics > 1) {
533 reportIndexedFreeListStatistics();
534 size_t totalSize = totalSizeInIndexedFreeLists() +
535 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock()));
536 gclog_or_tty->print(" free=%ld frag=%1.4f\n", totalSize, flsFrag());
537 }
538 }
539
540 void CompactibleFreeListSpace::reportIndexedFreeListStatistics() const {
541 assert_lock_strong(&_freelistLock);
542 gclog_or_tty->print("Statistics for IndexedFreeLists:\n"
543 "--------------------------------\n");
544 size_t totalSize = totalSizeInIndexedFreeLists();
545 size_t freeBlocks = numFreeBlocksInIndexedFreeLists();
546 gclog_or_tty->print("Total Free Space: %d\n", totalSize);
547 gclog_or_tty->print("Max Chunk Size: %d\n", maxChunkSizeInIndexedFreeLists());
548 gclog_or_tty->print("Number of Blocks: %d\n", freeBlocks);
549 if (freeBlocks != 0) {
550 gclog_or_tty->print("Av. Block Size: %d\n", totalSize/freeBlocks);
551 }
552 }
553
554 size_t CompactibleFreeListSpace::numFreeBlocksInIndexedFreeLists() const {
555 size_t res = 0;
556 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
557 debug_only(
558 ssize_t recount = 0;
559 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
560 fc = fc->next()) {
561 recount += 1;
562 }
563 assert(recount == _indexedFreeList[i].count(),
564 "Incorrect count in list");
565 )
566 res += _indexedFreeList[i].count();
567 }
568 return res;
569 }
570
571 size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const {
572 for (size_t i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
573 if (_indexedFreeList[i].head() != NULL) {
574 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
575 return (size_t)i;
576 }
577 }
578 return 0;
579 }
580
581 void CompactibleFreeListSpace::set_end(HeapWord* value) {
582 HeapWord* prevEnd = end();
583 assert(prevEnd != value, "unnecessary set_end call");
584 assert(prevEnd == NULL || !BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
585 "New end is below unallocated block");
586 _end = value;
587 if (prevEnd != NULL) {
588 // Resize the underlying block offset table.
589 _bt.resize(pointer_delta(value, bottom()));
590 if (value <= prevEnd) {
591 assert(!BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
592 "New end is below unallocated block");
593 } else {
594 // Now, take this new chunk and add it to the free blocks.
595 // Note that the BOT has not yet been updated for this block.
596 size_t newFcSize = pointer_delta(value, prevEnd);
597 // XXX This is REALLY UGLY and should be fixed up. XXX
598 if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) {
599 // Mark the boundary of the new block in BOT
600 _bt.mark_block(prevEnd, value);
601 // put it all in the linAB
602 if (ParallelGCThreads == 0) {
603 _smallLinearAllocBlock._ptr = prevEnd;
604 _smallLinearAllocBlock._word_size = newFcSize;
605 repairLinearAllocBlock(&_smallLinearAllocBlock);
606 } else { // ParallelGCThreads > 0
607 MutexLockerEx x(parDictionaryAllocLock(),
608 Mutex::_no_safepoint_check_flag);
609 _smallLinearAllocBlock._ptr = prevEnd;
610 _smallLinearAllocBlock._word_size = newFcSize;
611 repairLinearAllocBlock(&_smallLinearAllocBlock);
612 }
613 // Births of chunks put into a LinAB are not recorded. Births
614 // of chunks as they are allocated out of a LinAB are.
615 } else {
616 // Add the block to the free lists, if possible coalescing it
617 // with the last free block, and update the BOT and census data.
618 addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize);
619 }
620 }
621 }
622 }
623
624 class FreeListSpace_DCTOC : public Filtering_DCTOC {
625 CompactibleFreeListSpace* _cfls;
626 CMSCollector* _collector;
627 protected:
628 // Override.
629 #define walk_mem_region_with_cl_DECL(ClosureType) \
630 virtual void walk_mem_region_with_cl(MemRegion mr, \
631 HeapWord* bottom, HeapWord* top, \
632 ClosureType* cl); \
633 void walk_mem_region_with_cl_par(MemRegion mr, \
634 HeapWord* bottom, HeapWord* top, \
635 ClosureType* cl); \
636 void walk_mem_region_with_cl_nopar(MemRegion mr, \
637 HeapWord* bottom, HeapWord* top, \
638 ClosureType* cl)
639 walk_mem_region_with_cl_DECL(OopClosure);
640 walk_mem_region_with_cl_DECL(FilteringClosure);
641
642 public:
643 FreeListSpace_DCTOC(CompactibleFreeListSpace* sp,
644 CMSCollector* collector,
645 OopClosure* cl,
646 CardTableModRefBS::PrecisionStyle precision,
647 HeapWord* boundary) :
648 Filtering_DCTOC(sp, cl, precision, boundary),
649 _cfls(sp), _collector(collector) {}
650 };
651
652 // We de-virtualize the block-related calls below, since we know that our
653 // space is a CompactibleFreeListSpace.
654 #define FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ClosureType) \
655 void FreeListSpace_DCTOC::walk_mem_region_with_cl(MemRegion mr, \
656 HeapWord* bottom, \
657 HeapWord* top, \
658 ClosureType* cl) { \
659 if (SharedHeap::heap()->n_par_threads() > 0) { \
660 walk_mem_region_with_cl_par(mr, bottom, top, cl); \
661 } else { \
662 walk_mem_region_with_cl_nopar(mr, bottom, top, cl); \
663 } \
664 } \
665 void FreeListSpace_DCTOC::walk_mem_region_with_cl_par(MemRegion mr, \
666 HeapWord* bottom, \
667 HeapWord* top, \
668 ClosureType* cl) { \
669 /* Skip parts that are before "mr", in case "block_start" sent us \
670 back too far. */ \
671 HeapWord* mr_start = mr.start(); \
672 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
673 HeapWord* next = bottom + bot_size; \
674 while (next < mr_start) { \
675 bottom = next; \
676 bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
677 next = bottom + bot_size; \
678 } \
679 \
680 while (bottom < top) { \
681 if (_cfls->CompactibleFreeListSpace::block_is_obj(bottom) && \
682 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
683 oop(bottom)) && \
684 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
685 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
686 bottom += _cfls->adjustObjectSize(word_sz); \
687 } else { \
688 bottom += _cfls->CompactibleFreeListSpace::block_size(bottom); \
689 } \
690 } \
691 } \
692 void FreeListSpace_DCTOC::walk_mem_region_with_cl_nopar(MemRegion mr, \
693 HeapWord* bottom, \
694 HeapWord* top, \
695 ClosureType* cl) { \
696 /* Skip parts that are before "mr", in case "block_start" sent us \
697 back too far. */ \
698 HeapWord* mr_start = mr.start(); \
699 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
700 HeapWord* next = bottom + bot_size; \
701 while (next < mr_start) { \
702 bottom = next; \
703 bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
704 next = bottom + bot_size; \
705 } \
706 \
707 while (bottom < top) { \
708 if (_cfls->CompactibleFreeListSpace::block_is_obj_nopar(bottom) && \
709 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
710 oop(bottom)) && \
711 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
712 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
713 bottom += _cfls->adjustObjectSize(word_sz); \
714 } else { \
715 bottom += _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
716 } \
717 } \
718 }
719
720 // (There are only two of these, rather than N, because the split is due
721 // only to the introduction of the FilteringClosure, a local part of the
722 // impl of this abstraction.)
723 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(OopClosure)
724 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(FilteringClosure)
725
726 DirtyCardToOopClosure*
727 CompactibleFreeListSpace::new_dcto_cl(OopClosure* cl,
728 CardTableModRefBS::PrecisionStyle precision,
729 HeapWord* boundary) {
730 return new FreeListSpace_DCTOC(this, _collector, cl, precision, boundary);
731 }
732
733
734 // Note on locking for the space iteration functions:
735 // since the collector's iteration activities are concurrent with
736 // allocation activities by mutators, absent a suitable mutual exclusion
737 // mechanism the iterators may go awry. For instace a block being iterated
738 // may suddenly be allocated or divided up and part of it allocated and
739 // so on.
740
741 // Apply the given closure to each block in the space.
742 void CompactibleFreeListSpace::blk_iterate_careful(BlkClosureCareful* cl) {
743 assert_lock_strong(freelistLock());
744 HeapWord *cur, *limit;
745 for (cur = bottom(), limit = end(); cur < limit;
746 cur += cl->do_blk_careful(cur));
747 }
748
749 // Apply the given closure to each block in the space.
750 void CompactibleFreeListSpace::blk_iterate(BlkClosure* cl) {
751 assert_lock_strong(freelistLock());
752 HeapWord *cur, *limit;
753 for (cur = bottom(), limit = end(); cur < limit;
754 cur += cl->do_blk(cur));
755 }
756
757 // Apply the given closure to each oop in the space.
758 void CompactibleFreeListSpace::oop_iterate(OopClosure* cl) {
759 assert_lock_strong(freelistLock());
760 HeapWord *cur, *limit;
761 size_t curSize;
762 for (cur = bottom(), limit = end(); cur < limit;
763 cur += curSize) {
764 curSize = block_size(cur);
765 if (block_is_obj(cur)) {
766 oop(cur)->oop_iterate(cl);
767 }
768 }
769 }
770
771 // Apply the given closure to each oop in the space \intersect memory region.
772 void CompactibleFreeListSpace::oop_iterate(MemRegion mr, OopClosure* cl) {
773 assert_lock_strong(freelistLock());
774 if (is_empty()) {
775 return;
776 }
777 MemRegion cur = MemRegion(bottom(), end());
778 mr = mr.intersection(cur);
779 if (mr.is_empty()) {
780 return;
781 }
782 if (mr.equals(cur)) {
783 oop_iterate(cl);
784 return;
785 }
786 assert(mr.end() <= end(), "just took an intersection above");
787 HeapWord* obj_addr = block_start(mr.start());
788 HeapWord* t = mr.end();
789
790 SpaceMemRegionOopsIterClosure smr_blk(cl, mr);
791 if (block_is_obj(obj_addr)) {
792 // Handle first object specially.
793 oop obj = oop(obj_addr);
794 obj_addr += adjustObjectSize(obj->oop_iterate(&smr_blk));
795 } else {
796 FreeChunk* fc = (FreeChunk*)obj_addr;
797 obj_addr += fc->size();
798 }
799 while (obj_addr < t) {
800 HeapWord* obj = obj_addr;
801 obj_addr += block_size(obj_addr);
802 // If "obj_addr" is not greater than top, then the
803 // entire object "obj" is within the region.
804 if (obj_addr <= t) {
805 if (block_is_obj(obj)) {
806 oop(obj)->oop_iterate(cl);
807 }
808 } else {
809 // "obj" extends beyond end of region
810 if (block_is_obj(obj)) {
811 oop(obj)->oop_iterate(&smr_blk);
812 }
813 break;
814 }
815 }
816 }
817
818 // NOTE: In the following methods, in order to safely be able to
819 // apply the closure to an object, we need to be sure that the
820 // object has been initialized. We are guaranteed that an object
821 // is initialized if we are holding the Heap_lock with the
822 // world stopped.
823 void CompactibleFreeListSpace::verify_objects_initialized() const {
824 if (is_init_completed()) {
825 assert_locked_or_safepoint(Heap_lock);
826 if (Universe::is_fully_initialized()) {
827 guarantee(SafepointSynchronize::is_at_safepoint(),
828 "Required for objects to be initialized");
829 }
830 } // else make a concession at vm start-up
831 }
832
833 // Apply the given closure to each object in the space
834 void CompactibleFreeListSpace::object_iterate(ObjectClosure* blk) {
835 assert_lock_strong(freelistLock());
836 NOT_PRODUCT(verify_objects_initialized());
837 HeapWord *cur, *limit;
838 size_t curSize;
839 for (cur = bottom(), limit = end(); cur < limit;
840 cur += curSize) {
841 curSize = block_size(cur);
842 if (block_is_obj(cur)) {
843 blk->do_object(oop(cur));
844 }
845 }
846 }
847
848 // Apply the given closure to each live object in the space
849 // The usage of CompactibleFreeListSpace
850 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
851 // objects in the space with references to objects that are no longer
852 // valid. For example, an object may reference another object
853 // that has already been sweep up (collected). This method uses
854 // obj_is_alive() to determine whether it is safe to apply the closure to
855 // an object. See obj_is_alive() for details on how liveness of an
856 // object is decided.
857
858 void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) {
859 assert_lock_strong(freelistLock());
860 NOT_PRODUCT(verify_objects_initialized());
861 HeapWord *cur, *limit;
862 size_t curSize;
863 for (cur = bottom(), limit = end(); cur < limit;
864 cur += curSize) {
865 curSize = block_size(cur);
866 if (block_is_obj(cur) && obj_is_alive(cur)) {
867 blk->do_object(oop(cur));
868 }
869 }
870 }
871
872 void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr,
873 UpwardsObjectClosure* cl) {
874 assert_locked(freelistLock());
875 NOT_PRODUCT(verify_objects_initialized());
876 Space::object_iterate_mem(mr, cl);
877 }
878
879 // Callers of this iterator beware: The closure application should
880 // be robust in the face of uninitialized objects and should (always)
881 // return a correct size so that the next addr + size below gives us a
882 // valid block boundary. [See for instance,
883 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
884 // in ConcurrentMarkSweepGeneration.cpp.]
885 HeapWord*
886 CompactibleFreeListSpace::object_iterate_careful(ObjectClosureCareful* cl) {
887 assert_lock_strong(freelistLock());
888 HeapWord *addr, *last;
889 size_t size;
890 for (addr = bottom(), last = end();
891 addr < last; addr += size) {
892 FreeChunk* fc = (FreeChunk*)addr;
893 if (fc->isFree()) {
894 // Since we hold the free list lock, which protects direct
895 // allocation in this generation by mutators, a free object
896 // will remain free throughout this iteration code.
897 size = fc->size();
898 } else {
899 // Note that the object need not necessarily be initialized,
900 // because (for instance) the free list lock does NOT protect
901 // object initialization. The closure application below must
902 // therefore be correct in the face of uninitialized objects.
903 size = cl->do_object_careful(oop(addr));
904 if (size == 0) {
905 // An unparsable object found. Signal early termination.
906 return addr;
907 }
908 }
909 }
910 return NULL;
911 }
912
913 // Callers of this iterator beware: The closure application should
914 // be robust in the face of uninitialized objects and should (always)
915 // return a correct size so that the next addr + size below gives us a
916 // valid block boundary. [See for instance,
917 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
918 // in ConcurrentMarkSweepGeneration.cpp.]
919 HeapWord*
920 CompactibleFreeListSpace::object_iterate_careful_m(MemRegion mr,
921 ObjectClosureCareful* cl) {
922 assert_lock_strong(freelistLock());
923 // Can't use used_region() below because it may not necessarily
924 // be the same as [bottom(),end()); although we could
925 // use [used_region().start(),round_to(used_region().end(),CardSize)),
926 // that appears too cumbersome, so we just do the simpler check
927 // in the assertion below.
928 assert(!mr.is_empty() && MemRegion(bottom(),end()).contains(mr),
929 "mr should be non-empty and within used space");
930 HeapWord *addr, *end;
931 size_t size;
932 for (addr = block_start_careful(mr.start()), end = mr.end();
933 addr < end; addr += size) {
934 FreeChunk* fc = (FreeChunk*)addr;
935 if (fc->isFree()) {
936 // Since we hold the free list lock, which protects direct
937 // allocation in this generation by mutators, a free object
938 // will remain free throughout this iteration code.
939 size = fc->size();
940 } else {
941 // Note that the object need not necessarily be initialized,
942 // because (for instance) the free list lock does NOT protect
943 // object initialization. The closure application below must
944 // therefore be correct in the face of uninitialized objects.
945 size = cl->do_object_careful_m(oop(addr), mr);
946 if (size == 0) {
947 // An unparsable object found. Signal early termination.
948 return addr;
949 }
950 }
951 }
952 return NULL;
953 }
954
955
956 HeapWord* CompactibleFreeListSpace::block_start_const(const void* p) const {
957 NOT_PRODUCT(verify_objects_initialized());
958 return _bt.block_start(p);
959 }
960
961 HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const {
962 return _bt.block_start_careful(p);
963 }
964
965 size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const {
966 NOT_PRODUCT(verify_objects_initialized());
967 // This must be volatile, or else there is a danger that the compiler
968 // will compile the code below into a sometimes-infinite loop, by keeping
969 // the value read the first time in a register.
970 while (true) {
971 // We must do this until we get a consistent view of the object.
972 if (FreeChunk::indicatesFreeChunk(p)) {
973 volatile FreeChunk* fc = (volatile FreeChunk*)p;
974 size_t res = fc->size();
975 // If the object is still a free chunk, return the size, else it
976 // has been allocated so try again.
977 if (FreeChunk::indicatesFreeChunk(p)) {
978 assert(res != 0, "Block size should not be 0");
979 return res;
980 }
981 } else {
982 // must read from what 'p' points to in each loop.
983 klassOop k = ((volatile oopDesc*)p)->klass_or_null();
984 if (k != NULL) {
985 assert(k->is_oop(true /* ignore mark word */), "Should be klass oop");
986 oop o = (oop)p;
987 assert(o->is_parsable(), "Should be parsable");
988 assert(o->is_oop(true /* ignore mark word */), "Should be an oop.");
989 size_t res = o->size_given_klass(k->klass_part());
990 res = adjustObjectSize(res);
991 assert(res != 0, "Block size should not be 0");
992 return res;
993 }
994 }
995 }
996 }
997
998 // A variant of the above that uses the Printezis bits for
999 // unparsable but allocated objects. This avoids any possible
1000 // stalls waiting for mutators to initialize objects, and is
1001 // thus potentially faster than the variant above. However,
1002 // this variant may return a zero size for a block that is
1003 // under mutation and for which a consistent size cannot be
1004 // inferred without stalling; see CMSCollector::block_size_if_printezis_bits().
1005 size_t CompactibleFreeListSpace::block_size_no_stall(HeapWord* p,
1006 const CMSCollector* c)
1007 const {
1008 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1009 // This must be volatile, or else there is a danger that the compiler
1010 // will compile the code below into a sometimes-infinite loop, by keeping
1011 // the value read the first time in a register.
1012 DEBUG_ONLY(uint loops = 0;)
1013 while (true) {
1014 // We must do this until we get a consistent view of the object.
1015 if (FreeChunk::indicatesFreeChunk(p)) {
1016 volatile FreeChunk* fc = (volatile FreeChunk*)p;
1017 size_t res = fc->size();
1018 if (FreeChunk::indicatesFreeChunk(p)) {
1019 assert(res != 0, "Block size should not be 0");
1020 assert(loops == 0, "Should be 0");
1021 return res;
1022 }
1023 } else {
1024 // must read from what 'p' points to in each loop.
1025 klassOop k = ((volatile oopDesc*)p)->klass_or_null();
1026 if (k != NULL &&
1027 ((oopDesc*)p)->is_parsable() &&
1028 ((oopDesc*)p)->is_conc_safe()) {
1029 assert(k->is_oop(), "Should really be klass oop.");
1030 oop o = (oop)p;
1031 assert(o->is_oop(), "Should be an oop");
1032 size_t res = o->size_given_klass(k->klass_part());
1033 res = adjustObjectSize(res);
1034 assert(res != 0, "Block size should not be 0");
1035 return res;
1036 } else {
1037 return c->block_size_if_printezis_bits(p);
1038 }
1039 }
1040 assert(loops == 0, "Can loop at most once");
1041 DEBUG_ONLY(loops++;)
1042 }
1043 }
1044
1045 size_t CompactibleFreeListSpace::block_size_nopar(const HeapWord* p) const {
1046 NOT_PRODUCT(verify_objects_initialized());
1047 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1048 FreeChunk* fc = (FreeChunk*)p;
1049 if (fc->isFree()) {
1050 return fc->size();
1051 } else {
1052 // Ignore mark word because this may be a recently promoted
1053 // object whose mark word is used to chain together grey
1054 // objects (the last one would have a null value).
1055 assert(oop(p)->is_oop(true), "Should be an oop");
1056 return adjustObjectSize(oop(p)->size());
1057 }
1058 }
1059
1060 // This implementation assumes that the property of "being an object" is
1061 // stable. But being a free chunk may not be (because of parallel
1062 // promotion.)
1063 bool CompactibleFreeListSpace::block_is_obj(const HeapWord* p) const {
1064 FreeChunk* fc = (FreeChunk*)p;
1065 assert(is_in_reserved(p), "Should be in space");
1066 // When doing a mark-sweep-compact of the CMS generation, this
1067 // assertion may fail because prepare_for_compaction() uses
1068 // space that is garbage to maintain information on ranges of
1069 // live objects so that these live ranges can be moved as a whole.
1070 // Comment out this assertion until that problem can be solved
1071 // (i.e., that the block start calculation may look at objects
1072 // at address below "p" in finding the object that contains "p"
1073 // and those objects (if garbage) may have been modified to hold
1074 // live range information.
1075 // assert(CollectedHeap::use_parallel_gc_threads() || _bt.block_start(p) == p,
1076 // "Should be a block boundary");
1077 if (FreeChunk::indicatesFreeChunk(p)) return false;
1078 klassOop k = oop(p)->klass_or_null();
1079 if (k != NULL) {
1080 // Ignore mark word because it may have been used to
1081 // chain together promoted objects (the last one
1082 // would have a null value).
1083 assert(oop(p)->is_oop(true), "Should be an oop");
1084 return true;
1085 } else {
1086 return false; // Was not an object at the start of collection.
1087 }
1088 }
1089
1090 // Check if the object is alive. This fact is checked either by consulting
1091 // the main marking bitmap in the sweeping phase or, if it's a permanent
1092 // generation and we're not in the sweeping phase, by checking the
1093 // perm_gen_verify_bit_map where we store the "deadness" information if
1094 // we did not sweep the perm gen in the most recent previous GC cycle.
1095 bool CompactibleFreeListSpace::obj_is_alive(const HeapWord* p) const {
1096 assert (block_is_obj(p), "The address should point to an object");
1097
1098 // If we're sweeping, we use object liveness information from the main bit map
1099 // for both perm gen and old gen.
1100 // We don't need to lock the bitmap (live_map or dead_map below), because
1101 // EITHER we are in the middle of the sweeping phase, and the
1102 // main marking bit map (live_map below) is locked,
1103 // OR we're in other phases and perm_gen_verify_bit_map (dead_map below)
1104 // is stable, because it's mutated only in the sweeping phase.
1105 if (_collector->abstract_state() == CMSCollector::Sweeping) {
1106 CMSBitMap* live_map = _collector->markBitMap();
1107 return live_map->isMarked((HeapWord*) p);
1108 } else {
1109 // If we're not currently sweeping and we haven't swept the perm gen in
1110 // the previous concurrent cycle then we may have dead but unswept objects
1111 // in the perm gen. In this case, we use the "deadness" information
1112 // that we had saved in perm_gen_verify_bit_map at the last sweep.
1113 if (!CMSClassUnloadingEnabled && _collector->_permGen->reserved().contains(p)) {
1114 if (_collector->verifying()) {
1115 CMSBitMap* dead_map = _collector->perm_gen_verify_bit_map();
1116 // Object is marked in the dead_map bitmap at the previous sweep
1117 // when we know that it's dead; if the bitmap is not allocated then
1118 // the object is alive.
1119 return (dead_map->sizeInBits() == 0) // bit_map has been allocated
1120 || !dead_map->par_isMarked((HeapWord*) p);
1121 } else {
1122 return false; // We can't say for sure if it's live, so we say that it's dead.
1123 }
1124 }
1125 }
1126 return true;
1127 }
1128
1129 bool CompactibleFreeListSpace::block_is_obj_nopar(const HeapWord* p) const {
1130 FreeChunk* fc = (FreeChunk*)p;
1131 assert(is_in_reserved(p), "Should be in space");
1132 assert(_bt.block_start(p) == p, "Should be a block boundary");
1133 if (!fc->isFree()) {
1134 // Ignore mark word because it may have been used to
1135 // chain together promoted objects (the last one
1136 // would have a null value).
1137 assert(oop(p)->is_oop(true), "Should be an oop");
1138 return true;
1139 }
1140 return false;
1141 }
1142
1143 // "MT-safe but not guaranteed MT-precise" (TM); you may get an
1144 // approximate answer if you don't hold the freelistlock when you call this.
1145 size_t CompactibleFreeListSpace::totalSizeInIndexedFreeLists() const {
1146 size_t size = 0;
1147 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
1148 debug_only(
1149 // We may be calling here without the lock in which case we
1150 // won't do this modest sanity check.
1151 if (freelistLock()->owned_by_self()) {
1152 size_t total_list_size = 0;
1153 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
1154 fc = fc->next()) {
1155 total_list_size += i;
1156 }
1157 assert(total_list_size == i * _indexedFreeList[i].count(),
1158 "Count in list is incorrect");
1159 }
1160 )
1161 size += i * _indexedFreeList[i].count();
1162 }
1163 return size;
1164 }
1165
1166 HeapWord* CompactibleFreeListSpace::par_allocate(size_t size) {
1167 MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
1168 return allocate(size);
1169 }
1170
1171 HeapWord*
1172 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlockRemainder(size_t size) {
1173 return getChunkFromLinearAllocBlockRemainder(&_smallLinearAllocBlock, size);
1174 }
1175
1176 HeapWord* CompactibleFreeListSpace::allocate(size_t size) {
1177 assert_lock_strong(freelistLock());
1178 HeapWord* res = NULL;
1179 assert(size == adjustObjectSize(size),
1180 "use adjustObjectSize() before calling into allocate()");
1181
1182 if (_adaptive_freelists) {
1183 res = allocate_adaptive_freelists(size);
1184 } else { // non-adaptive free lists
1185 res = allocate_non_adaptive_freelists(size);
1186 }
1187
1188 if (res != NULL) {
1189 // check that res does lie in this space!
1190 assert(is_in_reserved(res), "Not in this space!");
1191 assert(is_aligned((void*)res), "alignment check");
1192
1193 FreeChunk* fc = (FreeChunk*)res;
1194 fc->markNotFree();
1195 assert(!fc->isFree(), "shouldn't be marked free");
1196 assert(oop(fc)->klass_or_null() == NULL, "should look uninitialized");
1197 // Verify that the block offset table shows this to
1198 // be a single block, but not one which is unallocated.
1199 _bt.verify_single_block(res, size);
1200 _bt.verify_not_unallocated(res, size);
1201 // mangle a just allocated object with a distinct pattern.
1202 debug_only(fc->mangleAllocated(size));
1203 }
1204
1205 return res;
1206 }
1207
1208 HeapWord* CompactibleFreeListSpace::allocate_non_adaptive_freelists(size_t size) {
1209 HeapWord* res = NULL;
1210 // try and use linear allocation for smaller blocks
1211 if (size < _smallLinearAllocBlock._allocation_size_limit) {
1212 // if successful, the following also adjusts block offset table
1213 res = getChunkFromSmallLinearAllocBlock(size);
1214 }
1215 // Else triage to indexed lists for smaller sizes
1216 if (res == NULL) {
1217 if (size < SmallForDictionary) {
1218 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1219 } else {
1220 // else get it from the big dictionary; if even this doesn't
1221 // work we are out of luck.
1222 res = (HeapWord*)getChunkFromDictionaryExact(size);
1223 }
1224 }
1225
1226 return res;
1227 }
1228
1229 HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) {
1230 assert_lock_strong(freelistLock());
1231 HeapWord* res = NULL;
1232 assert(size == adjustObjectSize(size),
1233 "use adjustObjectSize() before calling into allocate()");
1234
1235 // Strategy
1236 // if small
1237 // exact size from small object indexed list if small
1238 // small or large linear allocation block (linAB) as appropriate
1239 // take from lists of greater sized chunks
1240 // else
1241 // dictionary
1242 // small or large linear allocation block if it has the space
1243 // Try allocating exact size from indexTable first
1244 if (size < IndexSetSize) {
1245 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1246 if(res != NULL) {
1247 assert(res != (HeapWord*)_indexedFreeList[size].head(),
1248 "Not removed from free list");
1249 // no block offset table adjustment is necessary on blocks in
1250 // the indexed lists.
1251
1252 // Try allocating from the small LinAB
1253 } else if (size < _smallLinearAllocBlock._allocation_size_limit &&
1254 (res = getChunkFromSmallLinearAllocBlock(size)) != NULL) {
1255 // if successful, the above also adjusts block offset table
1256 // Note that this call will refill the LinAB to
1257 // satisfy the request. This is different that
1258 // evm.
1259 // Don't record chunk off a LinAB? smallSplitBirth(size);
1260 } else {
1261 // Raid the exact free lists larger than size, even if they are not
1262 // overpopulated.
1263 res = (HeapWord*) getChunkFromGreater(size);
1264 }
1265 } else {
1266 // Big objects get allocated directly from the dictionary.
1267 res = (HeapWord*) getChunkFromDictionaryExact(size);
1268 if (res == NULL) {
1269 // Try hard not to fail since an allocation failure will likely
1270 // trigger a synchronous GC. Try to get the space from the
1271 // allocation blocks.
1272 res = getChunkFromSmallLinearAllocBlockRemainder(size);
1273 }
1274 }
1275
1276 return res;
1277 }
1278
1279 // A worst-case estimate of the space required (in HeapWords) to expand the heap
1280 // when promoting obj.
1281 size_t CompactibleFreeListSpace::expansionSpaceRequired(size_t obj_size) const {
1282 // Depending on the object size, expansion may require refilling either a
1283 // bigLAB or a smallLAB plus refilling a PromotionInfo object. MinChunkSize
1284 // is added because the dictionary may over-allocate to avoid fragmentation.
1285 size_t space = obj_size;
1286 if (!_adaptive_freelists) {
1287 space = MAX2(space, _smallLinearAllocBlock._refillSize);
1288 }
1289 space += _promoInfo.refillSize() + 2 * MinChunkSize;
1290 return space;
1291 }
1292
1293 FreeChunk* CompactibleFreeListSpace::getChunkFromGreater(size_t numWords) {
1294 FreeChunk* ret;
1295
1296 assert(numWords >= MinChunkSize, "Size is less than minimum");
1297 assert(linearAllocationWouldFail() || bestFitFirst(),
1298 "Should not be here");
1299
1300 size_t i;
1301 size_t currSize = numWords + MinChunkSize;
1302 assert(currSize % MinObjAlignment == 0, "currSize should be aligned");
1303 for (i = currSize; i < IndexSetSize; i += IndexSetStride) {
1304 FreeList* fl = &_indexedFreeList[i];
1305 if (fl->head()) {
1306 ret = getFromListGreater(fl, numWords);
1307 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk");
1308 return ret;
1309 }
1310 }
1311
1312 currSize = MAX2((size_t)SmallForDictionary,
1313 (size_t)(numWords + MinChunkSize));
1314
1315 /* Try to get a chunk that satisfies request, while avoiding
1316 fragmentation that can't be handled. */
1317 {
1318 ret = dictionary()->getChunk(currSize);
1319 if (ret != NULL) {
1320 assert(ret->size() - numWords >= MinChunkSize,
1321 "Chunk is too small");
1322 _bt.allocated((HeapWord*)ret, ret->size());
1323 /* Carve returned chunk. */
1324 (void) splitChunkAndReturnRemainder(ret, numWords);
1325 /* Label this as no longer a free chunk. */
1326 assert(ret->isFree(), "This chunk should be free");
1327 ret->linkPrev(NULL);
1328 }
1329 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk");
1330 return ret;
1331 }
1332 ShouldNotReachHere();
1333 }
1334
1335 bool CompactibleFreeListSpace::verifyChunkInIndexedFreeLists(FreeChunk* fc)
1336 const {
1337 assert(fc->size() < IndexSetSize, "Size of chunk is too large");
1338 return _indexedFreeList[fc->size()].verifyChunkInFreeLists(fc);
1339 }
1340
1341 bool CompactibleFreeListSpace::verifyChunkInFreeLists(FreeChunk* fc) const {
1342 if (fc->size() >= IndexSetSize) {
1343 return dictionary()->verifyChunkInFreeLists(fc);
1344 } else {
1345 return verifyChunkInIndexedFreeLists(fc);
1346 }
1347 }
1348
1349 #ifndef PRODUCT
1350 void CompactibleFreeListSpace::assert_locked() const {
1351 CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock());
1352 }
1353
1354 void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const {
1355 CMSLockVerifier::assert_locked(lock);
1356 }
1357 #endif
1358
1359 FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) {
1360 // In the parallel case, the main thread holds the free list lock
1361 // on behalf the parallel threads.
1362 FreeChunk* fc;
1363 {
1364 // If GC is parallel, this might be called by several threads.
1365 // This should be rare enough that the locking overhead won't affect
1366 // the sequential code.
1367 MutexLockerEx x(parDictionaryAllocLock(),
1368 Mutex::_no_safepoint_check_flag);
1369 fc = getChunkFromDictionary(size);
1370 }
1371 if (fc != NULL) {
1372 fc->dontCoalesce();
1373 assert(fc->isFree(), "Should be free, but not coalescable");
1374 // Verify that the block offset table shows this to
1375 // be a single block, but not one which is unallocated.
1376 _bt.verify_single_block((HeapWord*)fc, fc->size());
1377 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
1378 }
1379 return fc;
1380 }
1381
1382 oop CompactibleFreeListSpace::promote(oop obj, size_t obj_size) {
1383 assert(obj_size == (size_t)obj->size(), "bad obj_size passed in");
1384 assert_locked();
1385
1386 // if we are tracking promotions, then first ensure space for
1387 // promotion (including spooling space for saving header if necessary).
1388 // then allocate and copy, then track promoted info if needed.
1389 // When tracking (see PromotionInfo::track()), the mark word may
1390 // be displaced and in this case restoration of the mark word
1391 // occurs in the (oop_since_save_marks_)iterate phase.
1392 if (_promoInfo.tracking() && !_promoInfo.ensure_spooling_space()) {
1393 return NULL;
1394 }
1395 // Call the allocate(size_t, bool) form directly to avoid the
1396 // additional call through the allocate(size_t) form. Having
1397 // the compile inline the call is problematic because allocate(size_t)
1398 // is a virtual method.
1399 HeapWord* res = allocate(adjustObjectSize(obj_size));
1400 if (res != NULL) {
1401 Copy::aligned_disjoint_words((HeapWord*)obj, res, obj_size);
1402 // if we should be tracking promotions, do so.
1403 if (_promoInfo.tracking()) {
1404 _promoInfo.track((PromotedObject*)res);
1405 }
1406 }
1407 return oop(res);
1408 }
1409
1410 HeapWord*
1411 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlock(size_t size) {
1412 assert_locked();
1413 assert(size >= MinChunkSize, "minimum chunk size");
1414 assert(size < _smallLinearAllocBlock._allocation_size_limit,
1415 "maximum from smallLinearAllocBlock");
1416 return getChunkFromLinearAllocBlock(&_smallLinearAllocBlock, size);
1417 }
1418
1419 HeapWord*
1420 CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk,
1421 size_t size) {
1422 assert_locked();
1423 assert(size >= MinChunkSize, "too small");
1424 HeapWord* res = NULL;
1425 // Try to do linear allocation from blk, making sure that
1426 if (blk->_word_size == 0) {
1427 // We have probably been unable to fill this either in the prologue or
1428 // when it was exhausted at the last linear allocation. Bail out until
1429 // next time.
1430 assert(blk->_ptr == NULL, "consistency check");
1431 return NULL;
1432 }
1433 assert(blk->_word_size != 0 && blk->_ptr != NULL, "consistency check");
1434 res = getChunkFromLinearAllocBlockRemainder(blk, size);
1435 if (res != NULL) return res;
1436
1437 // about to exhaust this linear allocation block
1438 if (blk->_word_size == size) { // exactly satisfied
1439 res = blk->_ptr;
1440 _bt.allocated(res, blk->_word_size);
1441 } else if (size + MinChunkSize <= blk->_refillSize) {
1442 size_t sz = blk->_word_size;
1443 // Update _unallocated_block if the size is such that chunk would be
1444 // returned to the indexed free list. All other chunks in the indexed
1445 // free lists are allocated from the dictionary so that _unallocated_block
1446 // has already been adjusted for them. Do it here so that the cost
1447 // for all chunks added back to the indexed free lists.
1448 if (sz < SmallForDictionary) {
1449 _bt.allocated(blk->_ptr, sz);
1450 }
1451 // Return the chunk that isn't big enough, and then refill below.
1452 addChunkToFreeLists(blk->_ptr, sz);
1453 splitBirth(sz);
1454 // Don't keep statistics on adding back chunk from a LinAB.
1455 } else {
1456 // A refilled block would not satisfy the request.
1457 return NULL;
1458 }
1459
1460 blk->_ptr = NULL; blk->_word_size = 0;
1461 refillLinearAllocBlock(blk);
1462 assert(blk->_ptr == NULL || blk->_word_size >= size + MinChunkSize,
1463 "block was replenished");
1464 if (res != NULL) {
1465 splitBirth(size);
1466 repairLinearAllocBlock(blk);
1467 } else if (blk->_ptr != NULL) {
1468 res = blk->_ptr;
1469 size_t blk_size = blk->_word_size;
1470 blk->_word_size -= size;
1471 blk->_ptr += size;
1472 splitBirth(size);
1473 repairLinearAllocBlock(blk);
1474 // Update BOT last so that other (parallel) GC threads see a consistent
1475 // view of the BOT and free blocks.
1476 // Above must occur before BOT is updated below.
1477 OrderAccess::storestore();
1478 _bt.split_block(res, blk_size, size); // adjust block offset table
1479 }
1480 return res;
1481 }
1482
1483 HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder(
1484 LinearAllocBlock* blk,
1485 size_t size) {
1486 assert_locked();
1487 assert(size >= MinChunkSize, "too small");
1488
1489 HeapWord* res = NULL;
1490 // This is the common case. Keep it simple.
1491 if (blk->_word_size >= size + MinChunkSize) {
1492 assert(blk->_ptr != NULL, "consistency check");
1493 res = blk->_ptr;
1494 // Note that the BOT is up-to-date for the linAB before allocation. It
1495 // indicates the start of the linAB. The split_block() updates the
1496 // BOT for the linAB after the allocation (indicates the start of the
1497 // next chunk to be allocated).
1498 size_t blk_size = blk->_word_size;
1499 blk->_word_size -= size;
1500 blk->_ptr += size;
1501 splitBirth(size);
1502 repairLinearAllocBlock(blk);
1503 // Update BOT last so that other (parallel) GC threads see a consistent
1504 // view of the BOT and free blocks.
1505 // Above must occur before BOT is updated below.
1506 OrderAccess::storestore();
1507 _bt.split_block(res, blk_size, size); // adjust block offset table
1508 _bt.allocated(res, size);
1509 }
1510 return res;
1511 }
1512
1513 FreeChunk*
1514 CompactibleFreeListSpace::getChunkFromIndexedFreeList(size_t size) {
1515 assert_locked();
1516 assert(size < SmallForDictionary, "just checking");
1517 FreeChunk* res;
1518 res = _indexedFreeList[size].getChunkAtHead();
1519 if (res == NULL) {
1520 res = getChunkFromIndexedFreeListHelper(size);
1521 }
1522 _bt.verify_not_unallocated((HeapWord*) res, size);
1523 assert(res == NULL || res->size() == size, "Incorrect block size");
1524 return res;
1525 }
1526
1527 FreeChunk*
1528 CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size,
1529 bool replenish) {
1530 assert_locked();
1531 FreeChunk* fc = NULL;
1532 if (size < SmallForDictionary) {
1533 assert(_indexedFreeList[size].head() == NULL ||
1534 _indexedFreeList[size].surplus() <= 0,
1535 "List for this size should be empty or under populated");
1536 // Try best fit in exact lists before replenishing the list
1537 if (!bestFitFirst() || (fc = bestFitSmall(size)) == NULL) {
1538 // Replenish list.
1539 //
1540 // Things tried that failed.
1541 // Tried allocating out of the two LinAB's first before
1542 // replenishing lists.
1543 // Tried small linAB of size 256 (size in indexed list)
1544 // and replenishing indexed lists from the small linAB.
1545 //
1546 FreeChunk* newFc = NULL;
1547 const size_t replenish_size = CMSIndexedFreeListReplenish * size;
1548 if (replenish_size < SmallForDictionary) {
1549 // Do not replenish from an underpopulated size.
1550 if (_indexedFreeList[replenish_size].surplus() > 0 &&
1551 _indexedFreeList[replenish_size].head() != NULL) {
1552 newFc = _indexedFreeList[replenish_size].getChunkAtHead();
1553 } else if (bestFitFirst()) {
1554 newFc = bestFitSmall(replenish_size);
1555 }
1556 }
1557 if (newFc == NULL && replenish_size > size) {
1558 assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant");
1559 newFc = getChunkFromIndexedFreeListHelper(replenish_size, false);
1560 }
1561 // Note: The stats update re split-death of block obtained above
1562 // will be recorded below precisely when we know we are going to
1563 // be actually splitting it into more than one pieces below.
1564 if (newFc != NULL) {
1565 if (replenish || CMSReplenishIntermediate) {
1566 // Replenish this list and return one block to caller.
1567 size_t i;
1568 FreeChunk *curFc, *nextFc;
1569 size_t num_blk = newFc->size() / size;
1570 assert(num_blk >= 1, "Smaller than requested?");
1571 assert(newFc->size() % size == 0, "Should be integral multiple of request");
1572 if (num_blk > 1) {
1573 // we are sure we will be splitting the block just obtained
1574 // into multiple pieces; record the split-death of the original
1575 splitDeath(replenish_size);
1576 }
1577 // carve up and link blocks 0, ..., num_blk - 2
1578 // The last chunk is not added to the lists but is returned as the
1579 // free chunk.
1580 for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size),
1581 i = 0;
1582 i < (num_blk - 1);
1583 curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size),
1584 i++) {
1585 curFc->setSize(size);
1586 // Don't record this as a return in order to try and
1587 // determine the "returns" from a GC.
1588 _bt.verify_not_unallocated((HeapWord*) fc, size);
1589 _indexedFreeList[size].returnChunkAtTail(curFc, false);
1590 _bt.mark_block((HeapWord*)curFc, size);
1591 splitBirth(size);
1592 // Don't record the initial population of the indexed list
1593 // as a split birth.
1594 }
1595
1596 // check that the arithmetic was OK above
1597 assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size,
1598 "inconsistency in carving newFc");
1599 curFc->setSize(size);
1600 _bt.mark_block((HeapWord*)curFc, size);
1601 splitBirth(size);
1602 fc = curFc;
1603 } else {
1604 // Return entire block to caller
1605 fc = newFc;
1606 }
1607 }
1608 }
1609 } else {
1610 // Get a free chunk from the free chunk dictionary to be returned to
1611 // replenish the indexed free list.
1612 fc = getChunkFromDictionaryExact(size);
1613 }
1614 // assert(fc == NULL || fc->isFree(), "Should be returning a free chunk");
1615 return fc;
1616 }
1617
1618 FreeChunk*
1619 CompactibleFreeListSpace::getChunkFromDictionary(size_t size) {
1620 assert_locked();
1621 FreeChunk* fc = _dictionary->getChunk(size);
1622 if (fc == NULL) {
1623 return NULL;
1624 }
1625 _bt.allocated((HeapWord*)fc, fc->size());
1626 if (fc->size() >= size + MinChunkSize) {
1627 fc = splitChunkAndReturnRemainder(fc, size);
1628 }
1629 assert(fc->size() >= size, "chunk too small");
1630 assert(fc->size() < size + MinChunkSize, "chunk too big");
1631 _bt.verify_single_block((HeapWord*)fc, fc->size());
1632 return fc;
1633 }
1634
1635 FreeChunk*
1636 CompactibleFreeListSpace::getChunkFromDictionaryExact(size_t size) {
1637 assert_locked();
1638 FreeChunk* fc = _dictionary->getChunk(size);
1639 if (fc == NULL) {
1640 return fc;
1641 }
1642 _bt.allocated((HeapWord*)fc, fc->size());
1643 if (fc->size() == size) {
1644 _bt.verify_single_block((HeapWord*)fc, size);
1645 return fc;
1646 }
1647 assert(fc->size() > size, "getChunk() guarantee");
1648 if (fc->size() < size + MinChunkSize) {
1649 // Return the chunk to the dictionary and go get a bigger one.
1650 returnChunkToDictionary(fc);
1651 fc = _dictionary->getChunk(size + MinChunkSize);
1652 if (fc == NULL) {
1653 return NULL;
1654 }
1655 _bt.allocated((HeapWord*)fc, fc->size());
1656 }
1657 assert(fc->size() >= size + MinChunkSize, "tautology");
1658 fc = splitChunkAndReturnRemainder(fc, size);
1659 assert(fc->size() == size, "chunk is wrong size");
1660 _bt.verify_single_block((HeapWord*)fc, size);
1661 return fc;
1662 }
1663
1664 void
1665 CompactibleFreeListSpace::returnChunkToDictionary(FreeChunk* chunk) {
1666 assert_locked();
1667
1668 size_t size = chunk->size();
1669 _bt.verify_single_block((HeapWord*)chunk, size);
1670 // adjust _unallocated_block downward, as necessary
1671 _bt.freed((HeapWord*)chunk, size);
1672 _dictionary->returnChunk(chunk);
1673 #ifndef PRODUCT
1674 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1675 TreeChunk::as_TreeChunk(chunk)->list()->verify_stats();
1676 }
1677 #endif // PRODUCT
1678 }
1679
1680 void
1681 CompactibleFreeListSpace::returnChunkToFreeList(FreeChunk* fc) {
1682 assert_locked();
1683 size_t size = fc->size();
1684 _bt.verify_single_block((HeapWord*) fc, size);
1685 _bt.verify_not_unallocated((HeapWord*) fc, size);
1686 if (_adaptive_freelists) {
1687 _indexedFreeList[size].returnChunkAtTail(fc);
1688 } else {
1689 _indexedFreeList[size].returnChunkAtHead(fc);
1690 }
1691 #ifndef PRODUCT
1692 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1693 _indexedFreeList[size].verify_stats();
1694 }
1695 #endif // PRODUCT
1696 }
1697
1698 // Add chunk to end of last block -- if it's the largest
1699 // block -- and update BOT and census data. We would
1700 // of course have preferred to coalesce it with the
1701 // last block, but it's currently less expensive to find the
1702 // largest block than it is to find the last.
1703 void
1704 CompactibleFreeListSpace::addChunkToFreeListsAtEndRecordingStats(
1705 HeapWord* chunk, size_t size) {
1706 // check that the chunk does lie in this space!
1707 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1708 // One of the parallel gc task threads may be here
1709 // whilst others are allocating.
1710 Mutex* lock = NULL;
1711 if (ParallelGCThreads != 0) {
1712 lock = &_parDictionaryAllocLock;
1713 }
1714 FreeChunk* ec;
1715 {
1716 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1717 ec = dictionary()->findLargestDict(); // get largest block
1718 if (ec != NULL && ec->end() == chunk) {
1719 // It's a coterminal block - we can coalesce.
1720 size_t old_size = ec->size();
1721 coalDeath(old_size);
1722 removeChunkFromDictionary(ec);
1723 size += old_size;
1724 } else {
1725 ec = (FreeChunk*)chunk;
1726 }
1727 }
1728 ec->setSize(size);
1729 debug_only(ec->mangleFreed(size));
1730 if (size < SmallForDictionary) {
1731 lock = _indexedFreeListParLocks[size];
1732 }
1733 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1734 addChunkAndRepairOffsetTable((HeapWord*)ec, size, true);
1735 // record the birth under the lock since the recording involves
1736 // manipulation of the list on which the chunk lives and
1737 // if the chunk is allocated and is the last on the list,
1738 // the list can go away.
1739 coalBirth(size);
1740 }
1741
1742 void
1743 CompactibleFreeListSpace::addChunkToFreeLists(HeapWord* chunk,
1744 size_t size) {
1745 // check that the chunk does lie in this space!
1746 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1747 assert_locked();
1748 _bt.verify_single_block(chunk, size);
1749
1750 FreeChunk* fc = (FreeChunk*) chunk;
1751 fc->setSize(size);
1752 debug_only(fc->mangleFreed(size));
1753 if (size < SmallForDictionary) {
1754 returnChunkToFreeList(fc);
1755 } else {
1756 returnChunkToDictionary(fc);
1757 }
1758 }
1759
1760 void
1761 CompactibleFreeListSpace::addChunkAndRepairOffsetTable(HeapWord* chunk,
1762 size_t size, bool coalesced) {
1763 assert_locked();
1764 assert(chunk != NULL, "null chunk");
1765 if (coalesced) {
1766 // repair BOT
1767 _bt.single_block(chunk, size);
1768 }
1769 addChunkToFreeLists(chunk, size);
1770 }
1771
1772 // We _must_ find the purported chunk on our free lists;
1773 // we assert if we don't.
1774 void
1775 CompactibleFreeListSpace::removeFreeChunkFromFreeLists(FreeChunk* fc) {
1776 size_t size = fc->size();
1777 assert_locked();
1778 debug_only(verifyFreeLists());
1779 if (size < SmallForDictionary) {
1780 removeChunkFromIndexedFreeList(fc);
1781 } else {
1782 removeChunkFromDictionary(fc);
1783 }
1784 _bt.verify_single_block((HeapWord*)fc, size);
1785 debug_only(verifyFreeLists());
1786 }
1787
1788 void
1789 CompactibleFreeListSpace::removeChunkFromDictionary(FreeChunk* fc) {
1790 size_t size = fc->size();
1791 assert_locked();
1792 assert(fc != NULL, "null chunk");
1793 _bt.verify_single_block((HeapWord*)fc, size);
1794 _dictionary->removeChunk(fc);
1795 // adjust _unallocated_block upward, as necessary
1796 _bt.allocated((HeapWord*)fc, size);
1797 }
1798
1799 void
1800 CompactibleFreeListSpace::removeChunkFromIndexedFreeList(FreeChunk* fc) {
1801 assert_locked();
1802 size_t size = fc->size();
1803 _bt.verify_single_block((HeapWord*)fc, size);
1804 NOT_PRODUCT(
1805 if (FLSVerifyIndexTable) {
1806 verifyIndexedFreeList(size);
1807 }
1808 )
1809 _indexedFreeList[size].removeChunk(fc);
1810 debug_only(fc->clearNext());
1811 debug_only(fc->clearPrev());
1812 NOT_PRODUCT(
1813 if (FLSVerifyIndexTable) {
1814 verifyIndexedFreeList(size);
1815 }
1816 )
1817 }
1818
1819 FreeChunk* CompactibleFreeListSpace::bestFitSmall(size_t numWords) {
1820 /* A hint is the next larger size that has a surplus.
1821 Start search at a size large enough to guarantee that
1822 the excess is >= MIN_CHUNK. */
1823 size_t start = align_object_size(numWords + MinChunkSize);
1824 if (start < IndexSetSize) {
1825 FreeList* it = _indexedFreeList;
1826 size_t hint = _indexedFreeList[start].hint();
1827 while (hint < IndexSetSize) {
1828 assert(hint % MinObjAlignment == 0, "hint should be aligned");
1829 FreeList *fl = &_indexedFreeList[hint];
1830 if (fl->surplus() > 0 && fl->head() != NULL) {
1831 // Found a list with surplus, reset original hint
1832 // and split out a free chunk which is returned.
1833 _indexedFreeList[start].set_hint(hint);
1834 FreeChunk* res = getFromListGreater(fl, numWords);
1835 assert(res == NULL || res->isFree(),
1836 "Should be returning a free chunk");
1837 return res;
1838 }
1839 hint = fl->hint(); /* keep looking */
1840 }
1841 /* None found. */
1842 it[start].set_hint(IndexSetSize);
1843 }
1844 return NULL;
1845 }
1846
1847 /* Requires fl->size >= numWords + MinChunkSize */
1848 FreeChunk* CompactibleFreeListSpace::getFromListGreater(FreeList* fl,
1849 size_t numWords) {
1850 FreeChunk *curr = fl->head();
1851 size_t oldNumWords = curr->size();
1852 assert(numWords >= MinChunkSize, "Word size is too small");
1853 assert(curr != NULL, "List is empty");
1854 assert(oldNumWords >= numWords + MinChunkSize,
1855 "Size of chunks in the list is too small");
1856
1857 fl->removeChunk(curr);
1858 // recorded indirectly by splitChunkAndReturnRemainder -
1859 // smallSplit(oldNumWords, numWords);
1860 FreeChunk* new_chunk = splitChunkAndReturnRemainder(curr, numWords);
1861 // Does anything have to be done for the remainder in terms of
1862 // fixing the card table?
1863 assert(new_chunk == NULL || new_chunk->isFree(),
1864 "Should be returning a free chunk");
1865 return new_chunk;
1866 }
1867
1868 FreeChunk*
1869 CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk,
1870 size_t new_size) {
1871 assert_locked();
1872 size_t size = chunk->size();
1873 assert(size > new_size, "Split from a smaller block?");
1874 assert(is_aligned(chunk), "alignment problem");
1875 assert(size == adjustObjectSize(size), "alignment problem");
1876 size_t rem_size = size - new_size;
1877 assert(rem_size == adjustObjectSize(rem_size), "alignment problem");
1878 assert(rem_size >= MinChunkSize, "Free chunk smaller than minimum");
1879 FreeChunk* ffc = (FreeChunk*)((HeapWord*)chunk + new_size);
1880 assert(is_aligned(ffc), "alignment problem");
1881 ffc->setSize(rem_size);
1882 ffc->linkNext(NULL);
1883 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
1884 // Above must occur before BOT is updated below.
1885 // adjust block offset table
1886 OrderAccess::storestore();
1887 assert(chunk->isFree() && ffc->isFree(), "Error");
1888 _bt.split_block((HeapWord*)chunk, chunk->size(), new_size);
1889 if (rem_size < SmallForDictionary) {
1890 bool is_par = (SharedHeap::heap()->n_par_threads() > 0);
1891 if (is_par) _indexedFreeListParLocks[rem_size]->lock();
1892 returnChunkToFreeList(ffc);
1893 split(size, rem_size);
1894 if (is_par) _indexedFreeListParLocks[rem_size]->unlock();
1895 } else {
1896 returnChunkToDictionary(ffc);
1897 split(size ,rem_size);
1898 }
1899 chunk->setSize(new_size);
1900 return chunk;
1901 }
1902
1903 void
1904 CompactibleFreeListSpace::sweep_completed() {
1905 // Now that space is probably plentiful, refill linear
1906 // allocation blocks as needed.
1907 refillLinearAllocBlocksIfNeeded();
1908 }
1909
1910 void
1911 CompactibleFreeListSpace::gc_prologue() {
1912 assert_locked();
1913 if (PrintFLSStatistics != 0) {
1914 gclog_or_tty->print("Before GC:\n");
1915 reportFreeListStatistics();
1916 }
1917 refillLinearAllocBlocksIfNeeded();
1918 }
1919
1920 void
1921 CompactibleFreeListSpace::gc_epilogue() {
1922 assert_locked();
1923 if (PrintGCDetails && Verbose && !_adaptive_freelists) {
1924 if (_smallLinearAllocBlock._word_size == 0)
1925 warning("CompactibleFreeListSpace(epilogue):: Linear allocation failure");
1926 }
1927 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1928 _promoInfo.stopTrackingPromotions();
1929 repairLinearAllocationBlocks();
1930 // Print Space's stats
1931 if (PrintFLSStatistics != 0) {
1932 gclog_or_tty->print("After GC:\n");
1933 reportFreeListStatistics();
1934 }
1935 }
1936
1937 // Iteration support, mostly delegated from a CMS generation
1938
1939 void CompactibleFreeListSpace::save_marks() {
1940 // mark the "end" of the used space at the time of this call;
1941 // note, however, that promoted objects from this point
1942 // on are tracked in the _promoInfo below.
1943 set_saved_mark_word(unallocated_block());
1944 // inform allocator that promotions should be tracked.
1945 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1946 _promoInfo.startTrackingPromotions();
1947 }
1948
1949 bool CompactibleFreeListSpace::no_allocs_since_save_marks() {
1950 assert(_promoInfo.tracking(), "No preceding save_marks?");
1951 assert(SharedHeap::heap()->n_par_threads() == 0,
1952 "Shouldn't be called if using parallel gc.");
1953 return _promoInfo.noPromotions();
1954 }
1955
1956 #define CFLS_OOP_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix) \
1957 \
1958 void CompactibleFreeListSpace:: \
1959 oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk) { \
1960 assert(SharedHeap::heap()->n_par_threads() == 0, \
1961 "Shouldn't be called (yet) during parallel part of gc."); \
1962 _promoInfo.promoted_oops_iterate##nv_suffix(blk); \
1963 /* \
1964 * This also restores any displaced headers and removes the elements from \
1965 * the iteration set as they are processed, so that we have a clean slate \
1966 * at the end of the iteration. Note, thus, that if new objects are \
1967 * promoted as a result of the iteration they are iterated over as well. \
1968 */ \
1969 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); \
1970 }
1971
1972 ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DEFN)
1973
1974
1975 void CompactibleFreeListSpace::object_iterate_since_last_GC(ObjectClosure* cl) {
1976 // ugghh... how would one do this efficiently for a non-contiguous space?
1977 guarantee(false, "NYI");
1978 }
1979
1980 bool CompactibleFreeListSpace::linearAllocationWouldFail() const {
1981 return _smallLinearAllocBlock._word_size == 0;
1982 }
1983
1984 void CompactibleFreeListSpace::repairLinearAllocationBlocks() {
1985 // Fix up linear allocation blocks to look like free blocks
1986 repairLinearAllocBlock(&_smallLinearAllocBlock);
1987 }
1988
1989 void CompactibleFreeListSpace::repairLinearAllocBlock(LinearAllocBlock* blk) {
1990 assert_locked();
1991 if (blk->_ptr != NULL) {
1992 assert(blk->_word_size != 0 && blk->_word_size >= MinChunkSize,
1993 "Minimum block size requirement");
1994 FreeChunk* fc = (FreeChunk*)(blk->_ptr);
1995 fc->setSize(blk->_word_size);
1996 fc->linkPrev(NULL); // mark as free
1997 fc->dontCoalesce();
1998 assert(fc->isFree(), "just marked it free");
1999 assert(fc->cantCoalesce(), "just marked it uncoalescable");
2000 }
2001 }
2002
2003 void CompactibleFreeListSpace::refillLinearAllocBlocksIfNeeded() {
2004 assert_locked();
2005 if (_smallLinearAllocBlock._ptr == NULL) {
2006 assert(_smallLinearAllocBlock._word_size == 0,
2007 "Size of linAB should be zero if the ptr is NULL");
2008 // Reset the linAB refill and allocation size limit.
2009 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, SmallForLinearAlloc);
2010 }
2011 refillLinearAllocBlockIfNeeded(&_smallLinearAllocBlock);
2012 }
2013
2014 void
2015 CompactibleFreeListSpace::refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk) {
2016 assert_locked();
2017 assert((blk->_ptr == NULL && blk->_word_size == 0) ||
2018 (blk->_ptr != NULL && blk->_word_size >= MinChunkSize),
2019 "blk invariant");
2020 if (blk->_ptr == NULL) {
2021 refillLinearAllocBlock(blk);
2022 }
2023 if (PrintMiscellaneous && Verbose) {
2024 if (blk->_word_size == 0) {
2025 warning("CompactibleFreeListSpace(prologue):: Linear allocation failure");
2026 }
2027 }
2028 }
2029
2030 void
2031 CompactibleFreeListSpace::refillLinearAllocBlock(LinearAllocBlock* blk) {
2032 assert_locked();
2033 assert(blk->_word_size == 0 && blk->_ptr == NULL,
2034 "linear allocation block should be empty");
2035 FreeChunk* fc;
2036 if (blk->_refillSize < SmallForDictionary &&
2037 (fc = getChunkFromIndexedFreeList(blk->_refillSize)) != NULL) {
2038 // A linAB's strategy might be to use small sizes to reduce
2039 // fragmentation but still get the benefits of allocation from a
2040 // linAB.
2041 } else {
2042 fc = getChunkFromDictionary(blk->_refillSize);
2043 }
2044 if (fc != NULL) {
2045 blk->_ptr = (HeapWord*)fc;
2046 blk->_word_size = fc->size();
2047 fc->dontCoalesce(); // to prevent sweeper from sweeping us up
2048 }
2049 }
2050
2051 // Support for concurrent collection policy decisions.
2052 bool CompactibleFreeListSpace::should_concurrent_collect() const {
2053 // In the future we might want to add in frgamentation stats --
2054 // including erosion of the "mountain" into this decision as well.
2055 return !adaptive_freelists() && linearAllocationWouldFail();
2056 }
2057
2058 // Support for compaction
2059
2060 void CompactibleFreeListSpace::prepare_for_compaction(CompactPoint* cp) {
2061 SCAN_AND_FORWARD(cp,end,block_is_obj,block_size);
2062 // prepare_for_compaction() uses the space between live objects
2063 // so that later phase can skip dead space quickly. So verification
2064 // of the free lists doesn't work after.
2065 }
2066
2067 #define obj_size(q) adjustObjectSize(oop(q)->size())
2068 #define adjust_obj_size(s) adjustObjectSize(s)
2069
2070 void CompactibleFreeListSpace::adjust_pointers() {
2071 // In other versions of adjust_pointers(), a bail out
2072 // based on the amount of live data in the generation
2073 // (i.e., if 0, bail out) may be used.
2074 // Cannot test used() == 0 here because the free lists have already
2075 // been mangled by the compaction.
2076
2077 SCAN_AND_ADJUST_POINTERS(adjust_obj_size);
2078 // See note about verification in prepare_for_compaction().
2079 }
2080
2081 void CompactibleFreeListSpace::compact() {
2082 SCAN_AND_COMPACT(obj_size);
2083 }
2084
2085 // fragmentation_metric = 1 - [sum of (fbs**2) / (sum of fbs)**2]
2086 // where fbs is free block sizes
2087 double CompactibleFreeListSpace::flsFrag() const {
2088 size_t itabFree = totalSizeInIndexedFreeLists();
2089 double frag = 0.0;
2090 size_t i;
2091
2092 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2093 double sz = i;
2094 frag += _indexedFreeList[i].count() * (sz * sz);
2095 }
2096
2097 double totFree = itabFree +
2098 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock()));
2099 if (totFree > 0) {
2100 frag = ((frag + _dictionary->sum_of_squared_block_sizes()) /
2101 (totFree * totFree));
2102 frag = (double)1.0 - frag;
2103 } else {
2104 assert(frag == 0.0, "Follows from totFree == 0");
2105 }
2106 return frag;
2107 }
2108
2109 void CompactibleFreeListSpace::beginSweepFLCensus(
2110 float inter_sweep_current,
2111 float inter_sweep_estimate,
2112 float intra_sweep_estimate) {
2113 assert_locked();
2114 size_t i;
2115 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2116 FreeList* fl = &_indexedFreeList[i];
2117 if (PrintFLSStatistics > 1) {
2118 gclog_or_tty->print("size[%d] : ", i);
2119 }
2120 fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate);
2121 fl->set_coalDesired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent));
2122 fl->set_beforeSweep(fl->count());
2123 fl->set_bfrSurp(fl->surplus());
2124 }
2125 _dictionary->beginSweepDictCensus(CMSLargeCoalSurplusPercent,
2126 inter_sweep_current,
2127 inter_sweep_estimate,
2128 intra_sweep_estimate);
2129 }
2130
2131 void CompactibleFreeListSpace::setFLSurplus() {
2132 assert_locked();
2133 size_t i;
2134 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2135 FreeList *fl = &_indexedFreeList[i];
2136 fl->set_surplus(fl->count() -
2137 (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent));
2138 }
2139 }
2140
2141 void CompactibleFreeListSpace::setFLHints() {
2142 assert_locked();
2143 size_t i;
2144 size_t h = IndexSetSize;
2145 for (i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
2146 FreeList *fl = &_indexedFreeList[i];
2147 fl->set_hint(h);
2148 if (fl->surplus() > 0) {
2149 h = i;
2150 }
2151 }
2152 }
2153
2154 void CompactibleFreeListSpace::clearFLCensus() {
2155 assert_locked();
2156 int i;
2157 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2158 FreeList *fl = &_indexedFreeList[i];
2159 fl->set_prevSweep(fl->count());
2160 fl->set_coalBirths(0);
2161 fl->set_coalDeaths(0);
2162 fl->set_splitBirths(0);
2163 fl->set_splitDeaths(0);
2164 }
2165 }
2166
2167 void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) {
2168 if (PrintFLSStatistics > 0) {
2169 HeapWord* largestAddr = (HeapWord*) dictionary()->findLargestDict();
2170 gclog_or_tty->print_cr("CMS: Large block " PTR_FORMAT,
2171 largestAddr);
2172 }
2173 setFLSurplus();
2174 setFLHints();
2175 if (PrintGC && PrintFLSCensus > 0) {
2176 printFLCensus(sweep_count);
2177 }
2178 clearFLCensus();
2179 assert_locked();
2180 _dictionary->endSweepDictCensus(CMSLargeSplitSurplusPercent);
2181 }
2182
2183 bool CompactibleFreeListSpace::coalOverPopulated(size_t size) {
2184 if (size < SmallForDictionary) {
2185 FreeList *fl = &_indexedFreeList[size];
2186 return (fl->coalDesired() < 0) ||
2187 ((int)fl->count() > fl->coalDesired());
2188 } else {
2189 return dictionary()->coalDictOverPopulated(size);
2190 }
2191 }
2192
2193 void CompactibleFreeListSpace::smallCoalBirth(size_t size) {
2194 assert(size < SmallForDictionary, "Size too large for indexed list");
2195 FreeList *fl = &_indexedFreeList[size];
2196 fl->increment_coalBirths();
2197 fl->increment_surplus();
2198 }
2199
2200 void CompactibleFreeListSpace::smallCoalDeath(size_t size) {
2201 assert(size < SmallForDictionary, "Size too large for indexed list");
2202 FreeList *fl = &_indexedFreeList[size];
2203 fl->increment_coalDeaths();
2204 fl->decrement_surplus();
2205 }
2206
2207 void CompactibleFreeListSpace::coalBirth(size_t size) {
2208 if (size < SmallForDictionary) {
2209 smallCoalBirth(size);
2210 } else {
2211 dictionary()->dictCensusUpdate(size,
2212 false /* split */,
2213 true /* birth */);
2214 }
2215 }
2216
2217 void CompactibleFreeListSpace::coalDeath(size_t size) {
2218 if(size < SmallForDictionary) {
2219 smallCoalDeath(size);
2220 } else {
2221 dictionary()->dictCensusUpdate(size,
2222 false /* split */,
2223 false /* birth */);
2224 }
2225 }
2226
2227 void CompactibleFreeListSpace::smallSplitBirth(size_t size) {
2228 assert(size < SmallForDictionary, "Size too large for indexed list");
2229 FreeList *fl = &_indexedFreeList[size];
2230 fl->increment_splitBirths();
2231 fl->increment_surplus();
2232 }
2233
2234 void CompactibleFreeListSpace::smallSplitDeath(size_t size) {
2235 assert(size < SmallForDictionary, "Size too large for indexed list");
2236 FreeList *fl = &_indexedFreeList[size];
2237 fl->increment_splitDeaths();
2238 fl->decrement_surplus();
2239 }
2240
2241 void CompactibleFreeListSpace::splitBirth(size_t size) {
2242 if (size < SmallForDictionary) {
2243 smallSplitBirth(size);
2244 } else {
2245 dictionary()->dictCensusUpdate(size,
2246 true /* split */,
2247 true /* birth */);
2248 }
2249 }
2250
2251 void CompactibleFreeListSpace::splitDeath(size_t size) {
2252 if (size < SmallForDictionary) {
2253 smallSplitDeath(size);
2254 } else {
2255 dictionary()->dictCensusUpdate(size,
2256 true /* split */,
2257 false /* birth */);
2258 }
2259 }
2260
2261 void CompactibleFreeListSpace::split(size_t from, size_t to1) {
2262 size_t to2 = from - to1;
2263 splitDeath(from);
2264 splitBirth(to1);
2265 splitBirth(to2);
2266 }
2267
2268 void CompactibleFreeListSpace::print() const {
2269 Space::print_on(tty);
2270 }
2271
2272 void CompactibleFreeListSpace::prepare_for_verify() {
2273 assert_locked();
2274 repairLinearAllocationBlocks();
2275 // Verify that the SpoolBlocks look like free blocks of
2276 // appropriate sizes... To be done ...
2277 }
2278
2279 class VerifyAllBlksClosure: public BlkClosure {
2280 private:
2281 const CompactibleFreeListSpace* _sp;
2282 const MemRegion _span;
2283 HeapWord* _last_addr;
2284 size_t _last_size;
2285 bool _last_was_obj;
2286 bool _last_was_live;
2287
2288 public:
2289 VerifyAllBlksClosure(const CompactibleFreeListSpace* sp,
2290 MemRegion span) : _sp(sp), _span(span),
2291 _last_addr(NULL), _last_size(0),
2292 _last_was_obj(false), _last_was_live(false) { }
2293
2294 virtual size_t do_blk(HeapWord* addr) {
2295 size_t res;
2296 bool was_obj = false;
2297 bool was_live = false;
2298 if (_sp->block_is_obj(addr)) {
2299 was_obj = true;
2300 oop p = oop(addr);
2301 guarantee(p->is_oop(), "Should be an oop");
2302 res = _sp->adjustObjectSize(p->size());
2303 if (_sp->obj_is_alive(addr)) {
2304 was_live = true;
2305 p->verify();
2306 }
2307 } else {
2308 FreeChunk* fc = (FreeChunk*)addr;
2309 res = fc->size();
2310 if (FLSVerifyLists && !fc->cantCoalesce()) {
2311 guarantee(_sp->verifyChunkInFreeLists(fc),
2312 "Chunk should be on a free list");
2313 }
2314 }
2315 if (res == 0) {
2316 gclog_or_tty->print_cr("Livelock: no rank reduction!");
2317 gclog_or_tty->print_cr(
2318 " Current: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n"
2319 " Previous: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n",
2320 addr, res, was_obj ?"true":"false", was_live ?"true":"false",
2321 _last_addr, _last_size, _last_was_obj?"true":"false", _last_was_live?"true":"false");
2322 _sp->print_on(gclog_or_tty);
2323 guarantee(false, "Seppuku!");
2324 }
2325 _last_addr = addr;
2326 _last_size = res;
2327 _last_was_obj = was_obj;
2328 _last_was_live = was_live;
2329 return res;
2330 }
2331 };
2332
2333 class VerifyAllOopsClosure: public OopClosure {
2334 private:
2335 const CMSCollector* _collector;
2336 const CompactibleFreeListSpace* _sp;
2337 const MemRegion _span;
2338 const bool _past_remark;
2339 const CMSBitMap* _bit_map;
2340
2341 protected:
2342 void do_oop(void* p, oop obj) {
2343 if (_span.contains(obj)) { // the interior oop points into CMS heap
2344 if (!_span.contains(p)) { // reference from outside CMS heap
2345 // Should be a valid object; the first disjunct below allows
2346 // us to sidestep an assertion in block_is_obj() that insists
2347 // that p be in _sp. Note that several generations (and spaces)
2348 // are spanned by _span (CMS heap) above.
2349 guarantee(!_sp->is_in_reserved(obj) ||
2350 _sp->block_is_obj((HeapWord*)obj),
2351 "Should be an object");
2352 guarantee(obj->is_oop(), "Should be an oop");
2353 obj->verify();
2354 if (_past_remark) {
2355 // Remark has been completed, the object should be marked
2356 _bit_map->isMarked((HeapWord*)obj);
2357 }
2358 } else { // reference within CMS heap
2359 if (_past_remark) {
2360 // Remark has been completed -- so the referent should have
2361 // been marked, if referring object is.
2362 if (_bit_map->isMarked(_collector->block_start(p))) {
2363 guarantee(_bit_map->isMarked((HeapWord*)obj), "Marking error?");
2364 }
2365 }
2366 }
2367 } else if (_sp->is_in_reserved(p)) {
2368 // the reference is from FLS, and points out of FLS
2369 guarantee(obj->is_oop(), "Should be an oop");
2370 obj->verify();
2371 }
2372 }
2373
2374 template <class T> void do_oop_work(T* p) {
2375 T heap_oop = oopDesc::load_heap_oop(p);
2376 if (!oopDesc::is_null(heap_oop)) {
2377 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
2378 do_oop(p, obj);
2379 }
2380 }
2381
2382 public:
2383 VerifyAllOopsClosure(const CMSCollector* collector,
2384 const CompactibleFreeListSpace* sp, MemRegion span,
2385 bool past_remark, CMSBitMap* bit_map) :
2386 OopClosure(), _collector(collector), _sp(sp), _span(span),
2387 _past_remark(past_remark), _bit_map(bit_map) { }
2388
2389 virtual void do_oop(oop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2390 virtual void do_oop(narrowOop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2391 };
2392
2393 void CompactibleFreeListSpace::verify(bool ignored) const {
2394 assert_lock_strong(&_freelistLock);
2395 verify_objects_initialized();
2396 MemRegion span = _collector->_span;
2397 bool past_remark = (_collector->abstract_state() ==
2398 CMSCollector::Sweeping);
2399
2400 ResourceMark rm;
2401 HandleMark hm;
2402
2403 // Check integrity of CFL data structures
2404 _promoInfo.verify();
2405 _dictionary->verify();
2406 if (FLSVerifyIndexTable) {
2407 verifyIndexedFreeLists();
2408 }
2409 // Check integrity of all objects and free blocks in space
2410 {
2411 VerifyAllBlksClosure cl(this, span);
2412 ((CompactibleFreeListSpace*)this)->blk_iterate(&cl); // cast off const
2413 }
2414 // Check that all references in the heap to FLS
2415 // are to valid objects in FLS or that references in
2416 // FLS are to valid objects elsewhere in the heap
2417 if (FLSVerifyAllHeapReferences)
2418 {
2419 VerifyAllOopsClosure cl(_collector, this, span, past_remark,
2420 _collector->markBitMap());
2421 CollectedHeap* ch = Universe::heap();
2422 ch->oop_iterate(&cl); // all oops in generations
2423 ch->permanent_oop_iterate(&cl); // all oops in perm gen
2424 }
2425
2426 if (VerifyObjectStartArray) {
2427 // Verify the block offset table
2428 _bt.verify();
2429 }
2430 }
2431
2432 #ifndef PRODUCT
2433 void CompactibleFreeListSpace::verifyFreeLists() const {
2434 if (FLSVerifyLists) {
2435 _dictionary->verify();
2436 verifyIndexedFreeLists();
2437 } else {
2438 if (FLSVerifyDictionary) {
2439 _dictionary->verify();
2440 }
2441 if (FLSVerifyIndexTable) {
2442 verifyIndexedFreeLists();
2443 }
2444 }
2445 }
2446 #endif
2447
2448 void CompactibleFreeListSpace::verifyIndexedFreeLists() const {
2449 size_t i = 0;
2450 for (; i < MinChunkSize; i++) {
2451 guarantee(_indexedFreeList[i].head() == NULL, "should be NULL");
2452 }
2453 for (; i < IndexSetSize; i++) {
2454 verifyIndexedFreeList(i);
2455 }
2456 }
2457
2458 void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const {
2459 FreeChunk* fc = _indexedFreeList[size].head();
2460 FreeChunk* tail = _indexedFreeList[size].tail();
2461 size_t num = _indexedFreeList[size].count();
2462 size_t n = 0;
2463 guarantee((size % 2 == 0) || fc == NULL, "Odd slots should be empty");
2464 for (; fc != NULL; fc = fc->next(), n++) {
2465 guarantee(fc->size() == size, "Size inconsistency");
2466 guarantee(fc->isFree(), "!free?");
2467 guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list");
2468 guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail");
2469 }
2470 guarantee(n == num, "Incorrect count");
2471 }
2472
2473 #ifndef PRODUCT
2474 void CompactibleFreeListSpace::checkFreeListConsistency() const {
2475 assert(_dictionary->minSize() <= IndexSetSize,
2476 "Some sizes can't be allocated without recourse to"
2477 " linear allocation buffers");
2478 assert(MIN_TREE_CHUNK_SIZE*HeapWordSize == sizeof(TreeChunk),
2479 "else MIN_TREE_CHUNK_SIZE is wrong");
2480 assert((IndexSetStride == 2 && IndexSetStart == 2) ||
2481 (IndexSetStride == 1 && IndexSetStart == 1), "just checking");
2482 assert((IndexSetStride != 2) || (MinChunkSize % 2 == 0),
2483 "Some for-loops may be incorrectly initialized");
2484 assert((IndexSetStride != 2) || (IndexSetSize % 2 == 1),
2485 "For-loops that iterate over IndexSet with stride 2 may be wrong");
2486 }
2487 #endif
2488
2489 void CompactibleFreeListSpace::printFLCensus(size_t sweep_count) const {
2490 assert_lock_strong(&_freelistLock);
2491 FreeList total;
2492 gclog_or_tty->print("end sweep# " SIZE_FORMAT "\n", sweep_count);
2493 FreeList::print_labels_on(gclog_or_tty, "size");
2494 size_t totalFree = 0;
2495 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2496 const FreeList *fl = &_indexedFreeList[i];
2497 totalFree += fl->count() * fl->size();
2498 if (i % (40*IndexSetStride) == 0) {
2499 FreeList::print_labels_on(gclog_or_tty, "size");
2500 }
2501 fl->print_on(gclog_or_tty);
2502 total.set_bfrSurp( total.bfrSurp() + fl->bfrSurp() );
2503 total.set_surplus( total.surplus() + fl->surplus() );
2504 total.set_desired( total.desired() + fl->desired() );
2505 total.set_prevSweep( total.prevSweep() + fl->prevSweep() );
2506 total.set_beforeSweep(total.beforeSweep() + fl->beforeSweep());
2507 total.set_count( total.count() + fl->count() );
2508 total.set_coalBirths( total.coalBirths() + fl->coalBirths() );
2509 total.set_coalDeaths( total.coalDeaths() + fl->coalDeaths() );
2510 total.set_splitBirths(total.splitBirths() + fl->splitBirths());
2511 total.set_splitDeaths(total.splitDeaths() + fl->splitDeaths());
2512 }
2513 total.print_on(gclog_or_tty, "TOTAL");
2514 gclog_or_tty->print_cr("Total free in indexed lists "
2515 SIZE_FORMAT " words", totalFree);
2516 gclog_or_tty->print("growth: %8.5f deficit: %8.5f\n",
2517 (double)(total.splitBirths()+total.coalBirths()-total.splitDeaths()-total.coalDeaths())/
2518 (total.prevSweep() != 0 ? (double)total.prevSweep() : 1.0),
2519 (double)(total.desired() - total.count())/(total.desired() != 0 ? (double)total.desired() : 1.0));
2520 _dictionary->printDictCensus();
2521 }
2522
2523 ///////////////////////////////////////////////////////////////////////////
2524 // CFLS_LAB
2525 ///////////////////////////////////////////////////////////////////////////
2526
2527 #define VECTOR_257(x) \
2528 /* 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 */ \
2529 { 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, \
2530 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, \
2531 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, \
2532 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, \
2533 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, \
2534 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, \
2535 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, \
2536 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, \
2537 x }
2538
2539 // Initialize with default setting of CMSParPromoteBlocksToClaim, _not_
2540 // OldPLABSize, whose static default is different; if overridden at the
2541 // command-line, this will get reinitialized via a call to
2542 // modify_initialization() below.
2543 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] =
2544 VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim));
2545 size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0);
2546 int CFLS_LAB::_global_num_workers[] = VECTOR_257(0);
2547
2548 CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) :
2549 _cfls(cfls)
2550 {
2551 assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above");
2552 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2553 i < CompactibleFreeListSpace::IndexSetSize;
2554 i += CompactibleFreeListSpace::IndexSetStride) {
2555 _indexedFreeList[i].set_size(i);
2556 _num_blocks[i] = 0;
2557 }
2558 }
2559
2560 static bool _CFLS_LAB_modified = false;
2561
2562 void CFLS_LAB::modify_initialization(size_t n, unsigned wt) {
2563 assert(!_CFLS_LAB_modified, "Call only once");
2564 _CFLS_LAB_modified = true;
2565 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2566 i < CompactibleFreeListSpace::IndexSetSize;
2567 i += CompactibleFreeListSpace::IndexSetStride) {
2568 _blocks_to_claim[i].modify(n, wt, true /* force */);
2569 }
2570 }
2571
2572 HeapWord* CFLS_LAB::alloc(size_t word_sz) {
2573 FreeChunk* res;
2574 assert(word_sz == _cfls->adjustObjectSize(word_sz), "Error");
2575 if (word_sz >= CompactibleFreeListSpace::IndexSetSize) {
2576 // This locking manages sync with other large object allocations.
2577 MutexLockerEx x(_cfls->parDictionaryAllocLock(),
2578 Mutex::_no_safepoint_check_flag);
2579 res = _cfls->getChunkFromDictionaryExact(word_sz);
2580 if (res == NULL) return NULL;
2581 } else {
2582 FreeList* fl = &_indexedFreeList[word_sz];
2583 if (fl->count() == 0) {
2584 // Attempt to refill this local free list.
2585 get_from_global_pool(word_sz, fl);
2586 // If it didn't work, give up.
2587 if (fl->count() == 0) return NULL;
2588 }
2589 res = fl->getChunkAtHead();
2590 assert(res != NULL, "Why was count non-zero?");
2591 }
2592 res->markNotFree();
2593 assert(!res->isFree(), "shouldn't be marked free");
2594 assert(oop(res)->klass_or_null() == NULL, "should look uninitialized");
2595 // mangle a just allocated object with a distinct pattern.
2596 debug_only(res->mangleAllocated(word_sz));
2597 return (HeapWord*)res;
2598 }
2599
2600 // Get a chunk of blocks of the right size and update related
2601 // book-keeping stats
2602 void CFLS_LAB::get_from_global_pool(size_t word_sz, FreeList* fl) {
2603 // Get the #blocks we want to claim
2604 size_t n_blks = (size_t)_blocks_to_claim[word_sz].average();
2605 assert(n_blks > 0, "Error");
2606 assert(ResizePLAB || n_blks == OldPLABSize, "Error");
2607 // In some cases, when the application has a phase change,
2608 // there may be a sudden and sharp shift in the object survival
2609 // profile, and updating the counts at the end of a scavenge
2610 // may not be quick enough, giving rise to large scavenge pauses
2611 // during these phase changes. It is beneficial to detect such
2612 // changes on-the-fly during a scavenge and avoid such a phase-change
2613 // pothole. The following code is a heuristic attempt to do that.
2614 // It is protected by a product flag until we have gained
2615 // enough experience with this heuristic and fine-tuned its behaviour.
2616 // WARNING: This might increase fragmentation if we overreact to
2617 // small spikes, so some kind of historical smoothing based on
2618 // previous experience with the greater reactivity might be useful.
2619 // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by
2620 // default.
2621 if (ResizeOldPLAB && CMSOldPLABResizeQuicker) {
2622 size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks);
2623 n_blks += CMSOldPLABReactivityFactor*multiple*n_blks;
2624 n_blks = MIN2(n_blks, CMSOldPLABMax);
2625 }
2626 assert(n_blks > 0, "Error");
2627 _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl);
2628 // Update stats table entry for this block size
2629 _num_blocks[word_sz] += fl->count();
2630 }
2631
2632 void CFLS_LAB::compute_desired_plab_size() {
2633 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2634 i < CompactibleFreeListSpace::IndexSetSize;
2635 i += CompactibleFreeListSpace::IndexSetStride) {
2636 assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0),
2637 "Counter inconsistency");
2638 if (_global_num_workers[i] > 0) {
2639 // Need to smooth wrt historical average
2640 if (ResizeOldPLAB) {
2641 _blocks_to_claim[i].sample(
2642 MAX2((size_t)CMSOldPLABMin,
2643 MIN2((size_t)CMSOldPLABMax,
2644 _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills))));
2645 }
2646 // Reset counters for next round
2647 _global_num_workers[i] = 0;
2648 _global_num_blocks[i] = 0;
2649 if (PrintOldPLAB) {
2650 gclog_or_tty->print_cr("[%d]: %d", i, (size_t)_blocks_to_claim[i].average());
2651 }
2652 }
2653 }
2654 }
2655
2656 void CFLS_LAB::retire(int tid) {
2657 // We run this single threaded with the world stopped;
2658 // so no need for locks and such.
2659 #define CFLS_LAB_PARALLEL_ACCESS 0
2660 NOT_PRODUCT(Thread* t = Thread::current();)
2661 assert(Thread::current()->is_VM_thread(), "Error");
2662 assert(CompactibleFreeListSpace::IndexSetStart == CompactibleFreeListSpace::IndexSetStride,
2663 "Will access to uninitialized slot below");
2664 #if CFLS_LAB_PARALLEL_ACCESS
2665 for (size_t i = CompactibleFreeListSpace::IndexSetSize - 1;
2666 i > 0;
2667 i -= CompactibleFreeListSpace::IndexSetStride) {
2668 #else // CFLS_LAB_PARALLEL_ACCESS
2669 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2670 i < CompactibleFreeListSpace::IndexSetSize;
2671 i += CompactibleFreeListSpace::IndexSetStride) {
2672 #endif // !CFLS_LAB_PARALLEL_ACCESS
2673 assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(),
2674 "Can't retire more than what we obtained");
2675 if (_num_blocks[i] > 0) {
2676 size_t num_retire = _indexedFreeList[i].count();
2677 assert(_num_blocks[i] > num_retire, "Should have used at least one");
2678 {
2679 #if CFLS_LAB_PARALLEL_ACCESS
2680 MutexLockerEx x(_cfls->_indexedFreeListParLocks[i],
2681 Mutex::_no_safepoint_check_flag);
2682 #endif // CFLS_LAB_PARALLEL_ACCESS
2683 // Update globals stats for num_blocks used
2684 _global_num_blocks[i] += (_num_blocks[i] - num_retire);
2685 _global_num_workers[i]++;
2686 assert(_global_num_workers[i] <= (ssize_t)ParallelGCThreads, "Too big");
2687 if (num_retire > 0) {
2688 _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]);
2689 // Reset this list.
2690 _indexedFreeList[i] = FreeList();
2691 _indexedFreeList[i].set_size(i);
2692 }
2693 }
2694 if (PrintOldPLAB) {
2695 gclog_or_tty->print_cr("%d[%d]: %d/%d/%d",
2696 tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average());
2697 }
2698 // Reset stats for next round
2699 _num_blocks[i] = 0;
2700 }
2701 }
2702 }
2703
2704 void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl) {
2705 assert(fl->count() == 0, "Precondition.");
2706 assert(word_sz < CompactibleFreeListSpace::IndexSetSize,
2707 "Precondition");
2708
2709 // We'll try all multiples of word_sz in the indexed set, starting with
2710 // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples,
2711 // then try getting a big chunk and splitting it.
2712 {
2713 bool found;
2714 int k;
2715 size_t cur_sz;
2716 for (k = 1, cur_sz = k * word_sz, found = false;
2717 (cur_sz < CompactibleFreeListSpace::IndexSetSize) &&
2718 (CMSSplitIndexedFreeListBlocks || k <= 1);
2719 k++, cur_sz = k * word_sz) {
2720 FreeList fl_for_cur_sz; // Empty.
2721 fl_for_cur_sz.set_size(cur_sz);
2722 {
2723 MutexLockerEx x(_indexedFreeListParLocks[cur_sz],
2724 Mutex::_no_safepoint_check_flag);
2725 FreeList* gfl = &_indexedFreeList[cur_sz];
2726 if (gfl->count() != 0) {
2727 // nn is the number of chunks of size cur_sz that
2728 // we'd need to split k-ways each, in order to create
2729 // "n" chunks of size word_sz each.
2730 const size_t nn = MAX2(n/k, (size_t)1);
2731 gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz);
2732 found = true;
2733 if (k > 1) {
2734 // Update split death stats for the cur_sz-size blocks list:
2735 // we increment the split death count by the number of blocks
2736 // we just took from the cur_sz-size blocks list and which
2737 // we will be splitting below.
2738 ssize_t deaths = gfl->splitDeaths() +
2739 fl_for_cur_sz.count();
2740 gfl->set_splitDeaths(deaths);
2741 }
2742 }
2743 }
2744 // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1.
2745 if (found) {
2746 if (k == 1) {
2747 fl->prepend(&fl_for_cur_sz);
2748 } else {
2749 // Divide each block on fl_for_cur_sz up k ways.
2750 FreeChunk* fc;
2751 while ((fc = fl_for_cur_sz.getChunkAtHead()) != NULL) {
2752 // Must do this in reverse order, so that anybody attempting to
2753 // access the main chunk sees it as a single free block until we
2754 // change it.
2755 size_t fc_size = fc->size();
2756 assert(fc->isFree(), "Error");
2757 for (int i = k-1; i >= 0; i--) {
2758 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2759 assert((i != 0) ||
2760 ((fc == ffc) && ffc->isFree() &&
2761 (ffc->size() == k*word_sz) && (fc_size == word_sz)),
2762 "Counting error");
2763 ffc->setSize(word_sz);
2764 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2765 ffc->linkNext(NULL);
2766 // Above must occur before BOT is updated below.
2767 OrderAccess::storestore();
2768 // splitting from the right, fc_size == i * word_sz
2769 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2770 fc_size -= word_sz;
2771 assert(fc_size == i*word_sz, "Error");
2772 _bt.verify_not_unallocated((HeapWord*)ffc, word_sz);
2773 _bt.verify_single_block((HeapWord*)fc, fc_size);
2774 _bt.verify_single_block((HeapWord*)ffc, word_sz);
2775 // Push this on "fl".
2776 fl->returnChunkAtHead(ffc);
2777 }
2778 // TRAP
2779 assert(fl->tail()->next() == NULL, "List invariant.");
2780 }
2781 }
2782 // Update birth stats for this block size.
2783 size_t num = fl->count();
2784 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2785 Mutex::_no_safepoint_check_flag);
2786 ssize_t births = _indexedFreeList[word_sz].splitBirths() + num;
2787 _indexedFreeList[word_sz].set_splitBirths(births);
2788 return;
2789 }
2790 }
2791 }
2792 // Otherwise, we'll split a block from the dictionary.
2793 FreeChunk* fc = NULL;
2794 FreeChunk* rem_fc = NULL;
2795 size_t rem;
2796 {
2797 MutexLockerEx x(parDictionaryAllocLock(),
2798 Mutex::_no_safepoint_check_flag);
2799 while (n > 0) {
2800 fc = dictionary()->getChunk(MAX2(n * word_sz,
2801 _dictionary->minSize()),
2802 FreeBlockDictionary::atLeast);
2803 if (fc != NULL) {
2804 _bt.allocated((HeapWord*)fc, fc->size(), true /* reducing */); // update _unallocated_blk
2805 dictionary()->dictCensusUpdate(fc->size(),
2806 true /*split*/,
2807 false /*birth*/);
2808 break;
2809 } else {
2810 n--;
2811 }
2812 }
2813 if (fc == NULL) return;
2814 // Otherwise, split up that block.
2815 assert((ssize_t)n >= 1, "Control point invariant");
2816 assert(fc->isFree(), "Error: should be a free block");
2817 _bt.verify_single_block((HeapWord*)fc, fc->size());
2818 const size_t nn = fc->size() / word_sz;
2819 n = MIN2(nn, n);
2820 assert((ssize_t)n >= 1, "Control point invariant");
2821 rem = fc->size() - n * word_sz;
2822 // If there is a remainder, and it's too small, allocate one fewer.
2823 if (rem > 0 && rem < MinChunkSize) {
2824 n--; rem += word_sz;
2825 }
2826 // Note that at this point we may have n == 0.
2827 assert((ssize_t)n >= 0, "Control point invariant");
2828
2829 // If n is 0, the chunk fc that was found is not large
2830 // enough to leave a viable remainder. We are unable to
2831 // allocate even one block. Return fc to the
2832 // dictionary and return, leaving "fl" empty.
2833 if (n == 0) {
2834 returnChunkToDictionary(fc);
2835 assert(fl->count() == 0, "We never allocated any blocks");
2836 return;
2837 }
2838
2839 // First return the remainder, if any.
2840 // Note that we hold the lock until we decide if we're going to give
2841 // back the remainder to the dictionary, since a concurrent allocation
2842 // may otherwise see the heap as empty. (We're willing to take that
2843 // hit if the block is a small block.)
2844 if (rem > 0) {
2845 size_t prefix_size = n * word_sz;
2846 rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size);
2847 rem_fc->setSize(rem);
2848 rem_fc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2849 rem_fc->linkNext(NULL);
2850 // Above must occur before BOT is updated below.
2851 assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error");
2852 OrderAccess::storestore();
2853 _bt.split_block((HeapWord*)fc, fc->size(), prefix_size);
2854 assert(fc->isFree(), "Error");
2855 fc->setSize(prefix_size);
2856 if (rem >= IndexSetSize) {
2857 returnChunkToDictionary(rem_fc);
2858 dictionary()->dictCensusUpdate(rem, true /*split*/, true /*birth*/);
2859 rem_fc = NULL;
2860 }
2861 // Otherwise, return it to the small list below.
2862 }
2863 }
2864 if (rem_fc != NULL) {
2865 MutexLockerEx x(_indexedFreeListParLocks[rem],
2866 Mutex::_no_safepoint_check_flag);
2867 _bt.verify_not_unallocated((HeapWord*)rem_fc, rem_fc->size());
2868 _indexedFreeList[rem].returnChunkAtHead(rem_fc);
2869 smallSplitBirth(rem);
2870 }
2871 assert((ssize_t)n > 0 && fc != NULL, "Consistency");
2872 // Now do the splitting up.
2873 // Must do this in reverse order, so that anybody attempting to
2874 // access the main chunk sees it as a single free block until we
2875 // change it.
2876 size_t fc_size = n * word_sz;
2877 // All but first chunk in this loop
2878 for (ssize_t i = n-1; i > 0; i--) {
2879 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2880 ffc->setSize(word_sz);
2881 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2882 ffc->linkNext(NULL);
2883 // Above must occur before BOT is updated below.
2884 OrderAccess::storestore();
2885 // splitting from the right, fc_size == (n - i + 1) * wordsize
2886 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2887 fc_size -= word_sz;
2888 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size());
2889 _bt.verify_single_block((HeapWord*)ffc, ffc->size());
2890 _bt.verify_single_block((HeapWord*)fc, fc_size);
2891 // Push this on "fl".
2892 fl->returnChunkAtHead(ffc);
2893 }
2894 // First chunk
2895 assert(fc->isFree() && fc->size() == n*word_sz, "Error: should still be a free block");
2896 // The blocks above should show their new sizes before the first block below
2897 fc->setSize(word_sz);
2898 fc->linkPrev(NULL); // idempotent wrt free-ness, see assert above
2899 fc->linkNext(NULL);
2900 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
2901 _bt.verify_single_block((HeapWord*)fc, fc->size());
2902 fl->returnChunkAtHead(fc);
2903
2904 assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks");
2905 {
2906 // Update the stats for this block size.
2907 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2908 Mutex::_no_safepoint_check_flag);
2909 const ssize_t births = _indexedFreeList[word_sz].splitBirths() + n;
2910 _indexedFreeList[word_sz].set_splitBirths(births);
2911 // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n;
2912 // _indexedFreeList[word_sz].set_surplus(new_surplus);
2913 }
2914
2915 // TRAP
2916 assert(fl->tail()->next() == NULL, "List invariant.");
2917 }
2918
2919 // Set up the space's par_seq_tasks structure for work claiming
2920 // for parallel rescan. See CMSParRemarkTask where this is currently used.
2921 // XXX Need to suitably abstract and generalize this and the next
2922 // method into one.
2923 void
2924 CompactibleFreeListSpace::
2925 initialize_sequential_subtasks_for_rescan(int n_threads) {
2926 // The "size" of each task is fixed according to rescan_task_size.
2927 assert(n_threads > 0, "Unexpected n_threads argument");
2928 const size_t task_size = rescan_task_size();
2929 size_t n_tasks = (used_region().word_size() + task_size - 1)/task_size;
2930 assert((n_tasks == 0) == used_region().is_empty(), "n_tasks incorrect");
2931 assert(n_tasks == 0 ||
2932 ((used_region().start() + (n_tasks - 1)*task_size < used_region().end()) &&
2933 (used_region().start() + n_tasks*task_size >= used_region().end())),
2934 "n_tasks calculation incorrect");
2935 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2936 assert(!pst->valid(), "Clobbering existing data?");
2937 // Sets the condition for completion of the subtask (how many threads
2938 // need to finish in order to be done).
2939 pst->set_n_threads(n_threads);
2940 pst->set_n_tasks((int)n_tasks);
2941 }
2942
2943 // Set up the space's par_seq_tasks structure for work claiming
2944 // for parallel concurrent marking. See CMSConcMarkTask where this is currently used.
2945 void
2946 CompactibleFreeListSpace::
2947 initialize_sequential_subtasks_for_marking(int n_threads,
2948 HeapWord* low) {
2949 // The "size" of each task is fixed according to rescan_task_size.
2950 assert(n_threads > 0, "Unexpected n_threads argument");
2951 const size_t task_size = marking_task_size();
2952 assert(task_size > CardTableModRefBS::card_size_in_words &&
2953 (task_size % CardTableModRefBS::card_size_in_words == 0),
2954 "Otherwise arithmetic below would be incorrect");
2955 MemRegion span = _gen->reserved();
2956 if (low != NULL) {
2957 if (span.contains(low)) {
2958 // Align low down to a card boundary so that
2959 // we can use block_offset_careful() on span boundaries.
2960 HeapWord* aligned_low = (HeapWord*)align_size_down((uintptr_t)low,
2961 CardTableModRefBS::card_size);
2962 // Clip span prefix at aligned_low
2963 span = span.intersection(MemRegion(aligned_low, span.end()));
2964 } else if (low > span.end()) {
2965 span = MemRegion(low, low); // Null region
2966 } // else use entire span
2967 }
2968 assert(span.is_empty() ||
2969 ((uintptr_t)span.start() % CardTableModRefBS::card_size == 0),
2970 "span should start at a card boundary");
2971 size_t n_tasks = (span.word_size() + task_size - 1)/task_size;
2972 assert((n_tasks == 0) == span.is_empty(), "Inconsistency");
2973 assert(n_tasks == 0 ||
2974 ((span.start() + (n_tasks - 1)*task_size < span.end()) &&
2975 (span.start() + n_tasks*task_size >= span.end())),
2976 "n_tasks calculation incorrect");
2977 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2978 assert(!pst->valid(), "Clobbering existing data?");
2979 // Sets the condition for completion of the subtask (how many threads
2980 // need to finish in order to be done).
2981 pst->set_n_threads(n_threads);
2982 pst->set_n_tasks((int)n_tasks);
2983 }