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 "precompiled.hpp"
26 #include "gc_implementation/concurrentMarkSweep/cmsLockVerifier.hpp"
27 #include "gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp"
28 #include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.inline.hpp"
29 #include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepThread.hpp"
30 #include "gc_implementation/shared/liveRange.hpp"
31 #include "gc_implementation/shared/spaceDecorator.hpp"
32 #include "gc_interface/collectedHeap.hpp"
33 #include "memory/allocation.inline.hpp"
34 #include "memory/blockOffsetTable.inline.hpp"
35 #include "memory/resourceArea.hpp"
36 #include "memory/universe.inline.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "runtime/globals.hpp"
39 #include "runtime/handles.inline.hpp"
40 #include "runtime/init.hpp"
41 #include "runtime/java.hpp"
42 #include "runtime/vmThread.hpp"
43 #include "utilities/copy.hpp"
44
45 /////////////////////////////////////////////////////////////////////////
46 //// CompactibleFreeListSpace
47 /////////////////////////////////////////////////////////////////////////
48
49 // highest ranked free list lock rank
50 int CompactibleFreeListSpace::_lockRank = Mutex::leaf + 3;
51
52 // Defaults are 0 so things will break badly if incorrectly initialized.
53 int CompactibleFreeListSpace::IndexSetStart = 0;
54 int CompactibleFreeListSpace::IndexSetStride = 0;
55
56 size_t MinChunkSize = 0;
57
58 void CompactibleFreeListSpace::set_cms_values() {
59 // Set CMS global values
60 assert(MinChunkSize == 0, "already set");
61 #define numQuanta(x,y) ((x+y-1)/y)
62 MinChunkSize = numQuanta(sizeof(FreeChunk), MinObjAlignmentInBytes) * MinObjAlignment;
63
64 assert(IndexSetStart == 0 && IndexSetStride == 0, "already set");
65 IndexSetStart = MinObjAlignment;
66 IndexSetStride = MinObjAlignment;
67 }
68
69 // Constructor
70 CompactibleFreeListSpace::CompactibleFreeListSpace(BlockOffsetSharedArray* bs,
71 MemRegion mr, bool use_adaptive_freelists,
72 FreeBlockDictionary::DictionaryChoice dictionaryChoice) :
73 _dictionaryChoice(dictionaryChoice),
74 _adaptive_freelists(use_adaptive_freelists),
75 _bt(bs, mr),
76 // free list locks are in the range of values taken by _lockRank
77 // This range currently is [_leaf+2, _leaf+3]
78 // Note: this requires that CFLspace c'tors
79 // are called serially in the order in which the locks are
80 // are acquired in the program text. This is true today.
81 _freelistLock(_lockRank--, "CompactibleFreeListSpace._lock", true),
82 _parDictionaryAllocLock(Mutex::leaf - 1, // == rank(ExpandHeap_lock) - 1
83 "CompactibleFreeListSpace._dict_par_lock", true),
84 _rescan_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
85 CMSRescanMultiple),
86 _marking_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
87 CMSConcMarkMultiple),
88 _collector(NULL)
89 {
90 _bt.set_space(this);
91 initialize(mr, SpaceDecorator::Clear, SpaceDecorator::Mangle);
92 // We have all of "mr", all of which we place in the dictionary
93 // as one big chunk. We'll need to decide here which of several
94 // possible alternative dictionary implementations to use. For
95 // now the choice is easy, since we have only one working
96 // implementation, namely, the simple binary tree (splaying
97 // temporarily disabled).
98 switch (dictionaryChoice) {
99 case FreeBlockDictionary::dictionarySplayTree:
100 case FreeBlockDictionary::dictionarySkipList:
101 default:
102 warning("dictionaryChoice: selected option not understood; using"
103 " default BinaryTreeDictionary implementation instead.");
104 case FreeBlockDictionary::dictionaryBinaryTree:
105 _dictionary = new BinaryTreeDictionary(mr);
106 break;
107 }
108 assert(_dictionary != NULL, "CMS dictionary initialization");
109 // The indexed free lists are initially all empty and are lazily
110 // filled in on demand. Initialize the array elements to NULL.
111 initializeIndexedFreeListArray();
112
113 // Not using adaptive free lists assumes that allocation is first
114 // from the linAB's. Also a cms perm gen which can be compacted
115 // has to have the klass's klassKlass allocated at a lower
116 // address in the heap than the klass so that the klassKlass is
117 // moved to its new location before the klass is moved.
118 // Set the _refillSize for the linear allocation blocks
119 if (!use_adaptive_freelists) {
120 FreeChunk* fc = _dictionary->getChunk(mr.word_size());
121 // The small linAB initially has all the space and will allocate
122 // a chunk of any size.
123 HeapWord* addr = (HeapWord*) fc;
124 _smallLinearAllocBlock.set(addr, fc->size() ,
125 1024*SmallForLinearAlloc, fc->size());
126 // Note that _unallocated_block is not updated here.
127 // Allocations from the linear allocation block should
128 // update it.
129 } else {
130 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc,
131 SmallForLinearAlloc);
132 }
133 // CMSIndexedFreeListReplenish should be at least 1
134 CMSIndexedFreeListReplenish = MAX2((uintx)1, CMSIndexedFreeListReplenish);
135 _promoInfo.setSpace(this);
136 if (UseCMSBestFit) {
137 _fitStrategy = FreeBlockBestFitFirst;
138 } else {
139 _fitStrategy = FreeBlockStrategyNone;
140 }
141 checkFreeListConsistency();
142
143 // Initialize locks for parallel case.
144
145 if (CollectedHeap::use_parallel_gc_threads()) {
146 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
147 _indexedFreeListParLocks[i] = new Mutex(Mutex::leaf - 1, // == ExpandHeap_lock - 1
148 "a freelist par lock",
149 true);
150 if (_indexedFreeListParLocks[i] == NULL)
151 vm_exit_during_initialization("Could not allocate a par lock");
152 DEBUG_ONLY(
153 _indexedFreeList[i].set_protecting_lock(_indexedFreeListParLocks[i]);
154 )
155 }
156 _dictionary->set_par_lock(&_parDictionaryAllocLock);
157 }
158 }
159
160 // Like CompactibleSpace forward() but always calls cross_threshold() to
161 // update the block offset table. Removed initialize_threshold call because
162 // CFLS does not use a block offset array for contiguous spaces.
163 HeapWord* CompactibleFreeListSpace::forward(oop q, size_t size,
164 CompactPoint* cp, HeapWord* compact_top) {
165 // q is alive
166 // First check if we should switch compaction space
167 assert(this == cp->space, "'this' should be current compaction space.");
168 size_t compaction_max_size = pointer_delta(end(), compact_top);
169 assert(adjustObjectSize(size) == cp->space->adjust_object_size_v(size),
170 "virtual adjustObjectSize_v() method is not correct");
171 size_t adjusted_size = adjustObjectSize(size);
172 assert(compaction_max_size >= MinChunkSize || compaction_max_size == 0,
173 "no small fragments allowed");
174 assert(minimum_free_block_size() == MinChunkSize,
175 "for de-virtualized reference below");
176 // Can't leave a nonzero size, residual fragment smaller than MinChunkSize
177 if (adjusted_size + MinChunkSize > compaction_max_size &&
178 adjusted_size != compaction_max_size) {
179 do {
180 // switch to next compaction space
181 cp->space->set_compaction_top(compact_top);
182 cp->space = cp->space->next_compaction_space();
183 if (cp->space == NULL) {
184 cp->gen = GenCollectedHeap::heap()->prev_gen(cp->gen);
185 assert(cp->gen != NULL, "compaction must succeed");
186 cp->space = cp->gen->first_compaction_space();
187 assert(cp->space != NULL, "generation must have a first compaction space");
188 }
189 compact_top = cp->space->bottom();
190 cp->space->set_compaction_top(compact_top);
191 // The correct adjusted_size may not be the same as that for this method
192 // (i.e., cp->space may no longer be "this" so adjust the size again.
193 // Use the virtual method which is not used above to save the virtual
194 // dispatch.
195 adjusted_size = cp->space->adjust_object_size_v(size);
196 compaction_max_size = pointer_delta(cp->space->end(), compact_top);
197 assert(cp->space->minimum_free_block_size() == 0, "just checking");
198 } while (adjusted_size > compaction_max_size);
199 }
200
201 // store the forwarding pointer into the mark word
202 if ((HeapWord*)q != compact_top) {
203 q->forward_to(oop(compact_top));
204 assert(q->is_gc_marked(), "encoding the pointer should preserve the mark");
205 } else {
206 // if the object isn't moving we can just set the mark to the default
207 // mark and handle it specially later on.
208 q->init_mark();
209 assert(q->forwardee() == NULL, "should be forwarded to NULL");
210 }
211
212 VALIDATE_MARK_SWEEP_ONLY(MarkSweep::register_live_oop(q, adjusted_size));
213 compact_top += adjusted_size;
214
215 // we need to update the offset table so that the beginnings of objects can be
216 // found during scavenge. Note that we are updating the offset table based on
217 // where the object will be once the compaction phase finishes.
218
219 // Always call cross_threshold(). A contiguous space can only call it when
220 // the compaction_top exceeds the current threshold but not for an
221 // non-contiguous space.
222 cp->threshold =
223 cp->space->cross_threshold(compact_top - adjusted_size, compact_top);
224 return compact_top;
225 }
226
227 // A modified copy of OffsetTableContigSpace::cross_threshold() with _offsets -> _bt
228 // and use of single_block instead of alloc_block. The name here is not really
229 // appropriate - maybe a more general name could be invented for both the
230 // contiguous and noncontiguous spaces.
231
232 HeapWord* CompactibleFreeListSpace::cross_threshold(HeapWord* start, HeapWord* the_end) {
233 _bt.single_block(start, the_end);
234 return end();
235 }
236
237 // Initialize them to NULL.
238 void CompactibleFreeListSpace::initializeIndexedFreeListArray() {
239 for (size_t i = 0; i < IndexSetSize; i++) {
240 // Note that on platforms where objects are double word aligned,
241 // the odd array elements are not used. It is convenient, however,
242 // to map directly from the object size to the array element.
243 _indexedFreeList[i].reset(IndexSetSize);
244 _indexedFreeList[i].set_size(i);
245 assert(_indexedFreeList[i].count() == 0, "reset check failed");
246 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
247 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
248 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
249 }
250 }
251
252 void CompactibleFreeListSpace::resetIndexedFreeListArray() {
253 for (int i = 1; i < IndexSetSize; i++) {
254 assert(_indexedFreeList[i].size() == (size_t) i,
255 "Indexed free list sizes are incorrect");
256 _indexedFreeList[i].reset(IndexSetSize);
257 assert(_indexedFreeList[i].count() == 0, "reset check failed");
258 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
259 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
260 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
261 }
262 }
263
264 void CompactibleFreeListSpace::reset(MemRegion mr) {
265 resetIndexedFreeListArray();
266 dictionary()->reset();
267 if (BlockOffsetArrayUseUnallocatedBlock) {
268 assert(end() == mr.end(), "We are compacting to the bottom of CMS gen");
269 // Everything's allocated until proven otherwise.
270 _bt.set_unallocated_block(end());
271 }
272 if (!mr.is_empty()) {
273 assert(mr.word_size() >= MinChunkSize, "Chunk size is too small");
274 _bt.single_block(mr.start(), mr.word_size());
275 FreeChunk* fc = (FreeChunk*) mr.start();
276 fc->setSize(mr.word_size());
277 if (mr.word_size() >= IndexSetSize ) {
278 returnChunkToDictionary(fc);
279 } else {
280 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
281 _indexedFreeList[mr.word_size()].returnChunkAtHead(fc);
282 }
283 }
284 _promoInfo.reset();
285 _smallLinearAllocBlock._ptr = NULL;
286 _smallLinearAllocBlock._word_size = 0;
287 }
288
289 void CompactibleFreeListSpace::reset_after_compaction() {
290 // Reset the space to the new reality - one free chunk.
291 MemRegion mr(compaction_top(), end());
292 reset(mr);
293 // Now refill the linear allocation block(s) if possible.
294 if (_adaptive_freelists) {
295 refillLinearAllocBlocksIfNeeded();
296 } else {
297 // Place as much of mr in the linAB as we can get,
298 // provided it was big enough to go into the dictionary.
299 FreeChunk* fc = dictionary()->findLargestDict();
300 if (fc != NULL) {
301 assert(fc->size() == mr.word_size(),
302 "Why was the chunk broken up?");
303 removeChunkFromDictionary(fc);
304 HeapWord* addr = (HeapWord*) fc;
305 _smallLinearAllocBlock.set(addr, fc->size() ,
306 1024*SmallForLinearAlloc, fc->size());
307 // Note that _unallocated_block is not updated here.
308 }
309 }
310 }
311
312 // Walks the entire dictionary, returning a coterminal
313 // chunk, if it exists. Use with caution since it involves
314 // a potentially complete walk of a potentially large tree.
315 FreeChunk* CompactibleFreeListSpace::find_chunk_at_end() {
316
317 assert_lock_strong(&_freelistLock);
318
319 return dictionary()->find_chunk_ends_at(end());
320 }
321
322
323 #ifndef PRODUCT
324 void CompactibleFreeListSpace::initializeIndexedFreeListArrayReturnedBytes() {
325 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
326 _indexedFreeList[i].allocation_stats()->set_returnedBytes(0);
327 }
328 }
329
330 size_t CompactibleFreeListSpace::sumIndexedFreeListArrayReturnedBytes() {
331 size_t sum = 0;
332 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
333 sum += _indexedFreeList[i].allocation_stats()->returnedBytes();
334 }
335 return sum;
336 }
337
338 size_t CompactibleFreeListSpace::totalCountInIndexedFreeLists() const {
339 size_t count = 0;
340 for (int i = (int)MinChunkSize; i < IndexSetSize; i++) {
341 debug_only(
342 ssize_t total_list_count = 0;
343 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
344 fc = fc->next()) {
345 total_list_count++;
346 }
347 assert(total_list_count == _indexedFreeList[i].count(),
348 "Count in list is incorrect");
349 )
350 count += _indexedFreeList[i].count();
351 }
352 return count;
353 }
354
355 size_t CompactibleFreeListSpace::totalCount() {
356 size_t num = totalCountInIndexedFreeLists();
357 num += dictionary()->totalCount();
358 if (_smallLinearAllocBlock._word_size != 0) {
359 num++;
360 }
361 return num;
362 }
363 #endif
364
365 bool CompactibleFreeListSpace::is_free_block(const HeapWord* p) const {
366 FreeChunk* fc = (FreeChunk*) p;
367 return fc->isFree();
368 }
369
370 size_t CompactibleFreeListSpace::used() const {
371 return capacity() - free();
372 }
373
374 size_t CompactibleFreeListSpace::free() const {
375 // "MT-safe, but not MT-precise"(TM), if you will: i.e.
376 // if you do this while the structures are in flux you
377 // may get an approximate answer only; for instance
378 // because there is concurrent allocation either
379 // directly by mutators or for promotion during a GC.
380 // It's "MT-safe", however, in the sense that you are guaranteed
381 // not to crash and burn, for instance, because of walking
382 // pointers that could disappear as you were walking them.
383 // The approximation is because the various components
384 // that are read below are not read atomically (and
385 // further the computation of totalSizeInIndexedFreeLists()
386 // is itself a non-atomic computation. The normal use of
387 // this is during a resize operation at the end of GC
388 // and at that time you are guaranteed to get the
389 // correct actual value. However, for instance, this is
390 // also read completely asynchronously by the "perf-sampler"
391 // that supports jvmstat, and you are apt to see the values
392 // flicker in such cases.
393 assert(_dictionary != NULL, "No _dictionary?");
394 return (_dictionary->totalChunkSize(DEBUG_ONLY(freelistLock())) +
395 totalSizeInIndexedFreeLists() +
396 _smallLinearAllocBlock._word_size) * HeapWordSize;
397 }
398
399 size_t CompactibleFreeListSpace::max_alloc_in_words() const {
400 assert(_dictionary != NULL, "No _dictionary?");
401 assert_locked();
402 size_t res = _dictionary->maxChunkSize();
403 res = MAX2(res, MIN2(_smallLinearAllocBlock._word_size,
404 (size_t) SmallForLinearAlloc - 1));
405 // XXX the following could potentially be pretty slow;
406 // should one, pesimally for the rare cases when res
407 // caclulated above is less than IndexSetSize,
408 // just return res calculated above? My reasoning was that
409 // those cases will be so rare that the extra time spent doesn't
410 // really matter....
411 // Note: do not change the loop test i >= res + IndexSetStride
412 // to i > res below, because i is unsigned and res may be zero.
413 for (size_t i = IndexSetSize - 1; i >= res + IndexSetStride;
414 i -= IndexSetStride) {
415 if (_indexedFreeList[i].head() != NULL) {
416 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
417 return i;
418 }
419 }
420 return res;
421 }
422
423 void LinearAllocBlock::print_on(outputStream* st) const {
424 st->print_cr(" LinearAllocBlock: ptr = " PTR_FORMAT ", word_size = " SIZE_FORMAT
425 ", refillsize = " SIZE_FORMAT ", allocation_size_limit = " SIZE_FORMAT,
426 _ptr, _word_size, _refillSize, _allocation_size_limit);
427 }
428
429 void CompactibleFreeListSpace::print_on(outputStream* st) const {
430 st->print_cr("COMPACTIBLE FREELIST SPACE");
431 st->print_cr(" Space:");
432 Space::print_on(st);
433
434 st->print_cr("promoInfo:");
435 _promoInfo.print_on(st);
436
437 st->print_cr("_smallLinearAllocBlock");
438 _smallLinearAllocBlock.print_on(st);
439
440 // dump_memory_block(_smallLinearAllocBlock->_ptr, 128);
441
442 st->print_cr(" _fitStrategy = %s, _adaptive_freelists = %s",
443 _fitStrategy?"true":"false", _adaptive_freelists?"true":"false");
444 }
445
446 void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st)
447 const {
448 reportIndexedFreeListStatistics();
449 gclog_or_tty->print_cr("Layout of Indexed Freelists");
450 gclog_or_tty->print_cr("---------------------------");
451 FreeList::print_labels_on(st, "size");
452 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
453 _indexedFreeList[i].print_on(gclog_or_tty);
454 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
455 fc = fc->next()) {
456 gclog_or_tty->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s",
457 fc, (HeapWord*)fc + i,
458 fc->cantCoalesce() ? "\t CC" : "");
459 }
460 }
461 }
462
463 void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st)
464 const {
465 _promoInfo.print_on(st);
466 }
467
468 void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st)
469 const {
470 _dictionary->reportStatistics();
471 st->print_cr("Layout of Freelists in Tree");
472 st->print_cr("---------------------------");
473 _dictionary->print_free_lists(st);
474 }
475
476 class BlkPrintingClosure: public BlkClosure {
477 const CMSCollector* _collector;
478 const CompactibleFreeListSpace* _sp;
479 const CMSBitMap* _live_bit_map;
480 const bool _post_remark;
481 outputStream* _st;
482 public:
483 BlkPrintingClosure(const CMSCollector* collector,
484 const CompactibleFreeListSpace* sp,
485 const CMSBitMap* live_bit_map,
486 outputStream* st):
487 _collector(collector),
488 _sp(sp),
489 _live_bit_map(live_bit_map),
490 _post_remark(collector->abstract_state() > CMSCollector::FinalMarking),
491 _st(st) { }
492 size_t do_blk(HeapWord* addr);
493 };
494
495 size_t BlkPrintingClosure::do_blk(HeapWord* addr) {
496 size_t sz = _sp->block_size_no_stall(addr, _collector);
497 assert(sz != 0, "Should always be able to compute a size");
498 if (_sp->block_is_obj(addr)) {
499 const bool dead = _post_remark && !_live_bit_map->isMarked(addr);
500 _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s",
501 addr,
502 dead ? "dead" : "live",
503 sz,
504 (!dead && CMSPrintObjectsInDump) ? ":" : ".");
505 if (CMSPrintObjectsInDump && !dead) {
506 oop(addr)->print_on(_st);
507 _st->print_cr("--------------------------------------");
508 }
509 } else { // free block
510 _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s",
511 addr, sz, CMSPrintChunksInDump ? ":" : ".");
512 if (CMSPrintChunksInDump) {
513 ((FreeChunk*)addr)->print_on(_st);
514 _st->print_cr("--------------------------------------");
515 }
516 }
517 return sz;
518 }
519
520 void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c,
521 outputStream* st) {
522 st->print_cr("\n=========================");
523 st->print_cr("Block layout in CMS Heap:");
524 st->print_cr("=========================");
525 BlkPrintingClosure bpcl(c, this, c->markBitMap(), st);
526 blk_iterate(&bpcl);
527
528 st->print_cr("\n=======================================");
529 st->print_cr("Order & Layout of Promotion Info Blocks");
530 st->print_cr("=======================================");
531 print_promo_info_blocks(st);
532
533 st->print_cr("\n===========================");
534 st->print_cr("Order of Indexed Free Lists");
535 st->print_cr("=========================");
536 print_indexed_free_lists(st);
537
538 st->print_cr("\n=================================");
539 st->print_cr("Order of Free Lists in Dictionary");
540 st->print_cr("=================================");
541 print_dictionary_free_lists(st);
542 }
543
544
545 void CompactibleFreeListSpace::reportFreeListStatistics() const {
546 assert_lock_strong(&_freelistLock);
547 assert(PrintFLSStatistics != 0, "Reporting error");
548 _dictionary->reportStatistics();
549 if (PrintFLSStatistics > 1) {
550 reportIndexedFreeListStatistics();
551 size_t totalSize = totalSizeInIndexedFreeLists() +
552 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock()));
553 gclog_or_tty->print(" free=%ld frag=%1.4f\n", totalSize, flsFrag());
554 }
555 }
556
557 void CompactibleFreeListSpace::reportIndexedFreeListStatistics() const {
558 assert_lock_strong(&_freelistLock);
559 gclog_or_tty->print("Statistics for IndexedFreeLists:\n"
560 "--------------------------------\n");
561 size_t totalSize = totalSizeInIndexedFreeLists();
562 size_t freeBlocks = numFreeBlocksInIndexedFreeLists();
563 gclog_or_tty->print("Total Free Space: %d\n", totalSize);
564 gclog_or_tty->print("Max Chunk Size: %d\n", maxChunkSizeInIndexedFreeLists());
565 gclog_or_tty->print("Number of Blocks: %d\n", freeBlocks);
566 if (freeBlocks != 0) {
567 gclog_or_tty->print("Av. Block Size: %d\n", totalSize/freeBlocks);
568 }
569 }
570
571 size_t CompactibleFreeListSpace::numFreeBlocksInIndexedFreeLists() const {
572 size_t res = 0;
573 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
574 debug_only(
575 ssize_t recount = 0;
576 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
577 fc = fc->next()) {
578 recount += 1;
579 }
580 assert(recount == _indexedFreeList[i].count(),
581 "Incorrect count in list");
582 )
583 res += _indexedFreeList[i].count();
584 }
585 return res;
586 }
587
588 size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const {
589 for (size_t i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
590 if (_indexedFreeList[i].head() != NULL) {
591 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
592 return (size_t)i;
593 }
594 }
595 return 0;
596 }
597
598 void CompactibleFreeListSpace::set_end(HeapWord* value) {
599 HeapWord* prevEnd = end();
600 assert(prevEnd != value, "unnecessary set_end call");
601 assert(prevEnd == NULL || !BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
602 "New end is below unallocated block");
603 _end = value;
604 if (prevEnd != NULL) {
605 // Resize the underlying block offset table.
606 _bt.resize(pointer_delta(value, bottom()));
607 if (value <= prevEnd) {
608 assert(!BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
609 "New end is below unallocated block");
610 } else {
611 // Now, take this new chunk and add it to the free blocks.
612 // Note that the BOT has not yet been updated for this block.
613 size_t newFcSize = pointer_delta(value, prevEnd);
614 // XXX This is REALLY UGLY and should be fixed up. XXX
615 if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) {
616 // Mark the boundary of the new block in BOT
617 _bt.mark_block(prevEnd, value);
618 // put it all in the linAB
619 if (ParallelGCThreads == 0) {
620 _smallLinearAllocBlock._ptr = prevEnd;
621 _smallLinearAllocBlock._word_size = newFcSize;
622 repairLinearAllocBlock(&_smallLinearAllocBlock);
623 } else { // ParallelGCThreads > 0
624 MutexLockerEx x(parDictionaryAllocLock(),
625 Mutex::_no_safepoint_check_flag);
626 _smallLinearAllocBlock._ptr = prevEnd;
627 _smallLinearAllocBlock._word_size = newFcSize;
628 repairLinearAllocBlock(&_smallLinearAllocBlock);
629 }
630 // Births of chunks put into a LinAB are not recorded. Births
631 // of chunks as they are allocated out of a LinAB are.
632 } else {
633 // Add the block to the free lists, if possible coalescing it
634 // with the last free block, and update the BOT and census data.
635 addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize);
636 }
637 }
638 }
639 }
640
641 class FreeListSpace_DCTOC : public Filtering_DCTOC {
642 CompactibleFreeListSpace* _cfls;
643 CMSCollector* _collector;
644 protected:
645 // Override.
646 #define walk_mem_region_with_cl_DECL(ClosureType) \
647 virtual void walk_mem_region_with_cl(MemRegion mr, \
648 HeapWord* bottom, HeapWord* top, \
649 ClosureType* cl); \
650 void walk_mem_region_with_cl_par(MemRegion mr, \
651 HeapWord* bottom, HeapWord* top, \
652 ClosureType* cl); \
653 void walk_mem_region_with_cl_nopar(MemRegion mr, \
654 HeapWord* bottom, HeapWord* top, \
655 ClosureType* cl)
656 walk_mem_region_with_cl_DECL(OopClosure);
657 walk_mem_region_with_cl_DECL(FilteringClosure);
658
659 public:
660 FreeListSpace_DCTOC(CompactibleFreeListSpace* sp,
661 CMSCollector* collector,
662 OopClosure* cl,
663 CardTableModRefBS::PrecisionStyle precision,
664 HeapWord* boundary) :
665 Filtering_DCTOC(sp, cl, precision, boundary),
666 _cfls(sp), _collector(collector) {}
667 };
668
669 // We de-virtualize the block-related calls below, since we know that our
670 // space is a CompactibleFreeListSpace.
671 #define FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ClosureType) \
672 void FreeListSpace_DCTOC::walk_mem_region_with_cl(MemRegion mr, \
673 HeapWord* bottom, \
674 HeapWord* top, \
675 ClosureType* cl) { \
676 if (SharedHeap::heap()->n_par_threads() > 0) { \
677 walk_mem_region_with_cl_par(mr, bottom, top, cl); \
678 } else { \
679 walk_mem_region_with_cl_nopar(mr, bottom, top, cl); \
680 } \
681 } \
682 void FreeListSpace_DCTOC::walk_mem_region_with_cl_par(MemRegion mr, \
683 HeapWord* bottom, \
684 HeapWord* top, \
685 ClosureType* cl) { \
686 /* Skip parts that are before "mr", in case "block_start" sent us \
687 back too far. */ \
688 HeapWord* mr_start = mr.start(); \
689 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
690 HeapWord* next = bottom + bot_size; \
691 while (next < mr_start) { \
692 bottom = next; \
693 bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
694 next = bottom + bot_size; \
695 } \
696 \
697 while (bottom < top) { \
698 if (_cfls->CompactibleFreeListSpace::block_is_obj(bottom) && \
699 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
700 oop(bottom)) && \
701 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
702 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
703 bottom += _cfls->adjustObjectSize(word_sz); \
704 } else { \
705 bottom += _cfls->CompactibleFreeListSpace::block_size(bottom); \
706 } \
707 } \
708 } \
709 void FreeListSpace_DCTOC::walk_mem_region_with_cl_nopar(MemRegion mr, \
710 HeapWord* bottom, \
711 HeapWord* top, \
712 ClosureType* cl) { \
713 /* Skip parts that are before "mr", in case "block_start" sent us \
714 back too far. */ \
715 HeapWord* mr_start = mr.start(); \
716 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
717 HeapWord* next = bottom + bot_size; \
718 while (next < mr_start) { \
719 bottom = next; \
720 bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
721 next = bottom + bot_size; \
722 } \
723 \
724 while (bottom < top) { \
725 if (_cfls->CompactibleFreeListSpace::block_is_obj_nopar(bottom) && \
726 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
727 oop(bottom)) && \
728 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
729 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
730 bottom += _cfls->adjustObjectSize(word_sz); \
731 } else { \
732 bottom += _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
733 } \
734 } \
735 }
736
737 // (There are only two of these, rather than N, because the split is due
738 // only to the introduction of the FilteringClosure, a local part of the
739 // impl of this abstraction.)
740 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(OopClosure)
741 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(FilteringClosure)
742
743 DirtyCardToOopClosure*
744 CompactibleFreeListSpace::new_dcto_cl(OopClosure* cl,
745 CardTableModRefBS::PrecisionStyle precision,
746 HeapWord* boundary) {
747 return new FreeListSpace_DCTOC(this, _collector, cl, precision, boundary);
748 }
749
750
751 // Note on locking for the space iteration functions:
752 // since the collector's iteration activities are concurrent with
753 // allocation activities by mutators, absent a suitable mutual exclusion
754 // mechanism the iterators may go awry. For instace a block being iterated
755 // may suddenly be allocated or divided up and part of it allocated and
756 // so on.
757
758 // Apply the given closure to each block in the space.
759 void CompactibleFreeListSpace::blk_iterate_careful(BlkClosureCareful* cl) {
760 assert_lock_strong(freelistLock());
761 HeapWord *cur, *limit;
762 for (cur = bottom(), limit = end(); cur < limit;
763 cur += cl->do_blk_careful(cur));
764 }
765
766 // Apply the given closure to each block in the space.
767 void CompactibleFreeListSpace::blk_iterate(BlkClosure* cl) {
768 assert_lock_strong(freelistLock());
769 HeapWord *cur, *limit;
770 for (cur = bottom(), limit = end(); cur < limit;
771 cur += cl->do_blk(cur));
772 }
773
774 // Apply the given closure to each oop in the space.
775 void CompactibleFreeListSpace::oop_iterate(OopClosure* cl) {
776 assert_lock_strong(freelistLock());
777 HeapWord *cur, *limit;
778 size_t curSize;
779 for (cur = bottom(), limit = end(); cur < limit;
780 cur += curSize) {
781 curSize = block_size(cur);
782 if (block_is_obj(cur)) {
783 oop(cur)->oop_iterate(cl);
784 }
785 }
786 }
787
788 // Apply the given closure to each oop in the space \intersect memory region.
789 void CompactibleFreeListSpace::oop_iterate(MemRegion mr, OopClosure* cl) {
790 assert_lock_strong(freelistLock());
791 if (is_empty()) {
792 return;
793 }
794 MemRegion cur = MemRegion(bottom(), end());
795 mr = mr.intersection(cur);
796 if (mr.is_empty()) {
797 return;
798 }
799 if (mr.equals(cur)) {
800 oop_iterate(cl);
801 return;
802 }
803 assert(mr.end() <= end(), "just took an intersection above");
804 HeapWord* obj_addr = block_start(mr.start());
805 HeapWord* t = mr.end();
806
807 SpaceMemRegionOopsIterClosure smr_blk(cl, mr);
808 if (block_is_obj(obj_addr)) {
809 // Handle first object specially.
810 oop obj = oop(obj_addr);
811 obj_addr += adjustObjectSize(obj->oop_iterate(&smr_blk));
812 } else {
813 FreeChunk* fc = (FreeChunk*)obj_addr;
814 obj_addr += fc->size();
815 }
816 while (obj_addr < t) {
817 HeapWord* obj = obj_addr;
818 obj_addr += block_size(obj_addr);
819 // If "obj_addr" is not greater than top, then the
820 // entire object "obj" is within the region.
821 if (obj_addr <= t) {
822 if (block_is_obj(obj)) {
823 oop(obj)->oop_iterate(cl);
824 }
825 } else {
826 // "obj" extends beyond end of region
827 if (block_is_obj(obj)) {
828 oop(obj)->oop_iterate(&smr_blk);
829 }
830 break;
831 }
832 }
833 }
834
835 // NOTE: In the following methods, in order to safely be able to
836 // apply the closure to an object, we need to be sure that the
837 // object has been initialized. We are guaranteed that an object
838 // is initialized if we are holding the Heap_lock with the
839 // world stopped.
840 void CompactibleFreeListSpace::verify_objects_initialized() const {
841 if (is_init_completed()) {
842 assert_locked_or_safepoint(Heap_lock);
843 if (Universe::is_fully_initialized()) {
844 guarantee(SafepointSynchronize::is_at_safepoint(),
845 "Required for objects to be initialized");
846 }
847 } // else make a concession at vm start-up
848 }
849
850 // Apply the given closure to each object in the space
851 void CompactibleFreeListSpace::object_iterate(ObjectClosure* blk) {
852 assert_lock_strong(freelistLock());
853 NOT_PRODUCT(verify_objects_initialized());
854 HeapWord *cur, *limit;
855 size_t curSize;
856 for (cur = bottom(), limit = end(); cur < limit;
857 cur += curSize) {
858 curSize = block_size(cur);
859 if (block_is_obj(cur)) {
860 blk->do_object(oop(cur));
861 }
862 }
863 }
864
865 // Apply the given closure to each live object in the space
866 // The usage of CompactibleFreeListSpace
867 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
868 // objects in the space with references to objects that are no longer
869 // valid. For example, an object may reference another object
870 // that has already been sweep up (collected). This method uses
871 // obj_is_alive() to determine whether it is safe to apply the closure to
872 // an object. See obj_is_alive() for details on how liveness of an
873 // object is decided.
874
875 void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) {
876 assert_lock_strong(freelistLock());
877 NOT_PRODUCT(verify_objects_initialized());
878 HeapWord *cur, *limit;
879 size_t curSize;
880 for (cur = bottom(), limit = end(); cur < limit;
881 cur += curSize) {
882 curSize = block_size(cur);
883 if (block_is_obj(cur) && obj_is_alive(cur)) {
884 blk->do_object(oop(cur));
885 }
886 }
887 }
888
889 void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr,
890 UpwardsObjectClosure* cl) {
891 assert_locked(freelistLock());
892 NOT_PRODUCT(verify_objects_initialized());
893 Space::object_iterate_mem(mr, cl);
894 }
895
896 // Callers of this iterator beware: The closure application should
897 // be robust in the face of uninitialized objects and should (always)
898 // return a correct size so that the next addr + size below gives us a
899 // valid block boundary. [See for instance,
900 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
901 // in ConcurrentMarkSweepGeneration.cpp.]
902 HeapWord*
903 CompactibleFreeListSpace::object_iterate_careful(ObjectClosureCareful* cl) {
904 assert_lock_strong(freelistLock());
905 HeapWord *addr, *last;
906 size_t size;
907 for (addr = bottom(), last = end();
908 addr < last; addr += size) {
909 FreeChunk* fc = (FreeChunk*)addr;
910 if (fc->isFree()) {
911 // Since we hold the free list lock, which protects direct
912 // allocation in this generation by mutators, a free object
913 // will remain free throughout this iteration code.
914 size = fc->size();
915 } else {
916 // Note that the object need not necessarily be initialized,
917 // because (for instance) the free list lock does NOT protect
918 // object initialization. The closure application below must
919 // therefore be correct in the face of uninitialized objects.
920 size = cl->do_object_careful(oop(addr));
921 if (size == 0) {
922 // An unparsable object found. Signal early termination.
923 return addr;
924 }
925 }
926 }
927 return NULL;
928 }
929
930 // Callers of this iterator beware: The closure application should
931 // be robust in the face of uninitialized objects and should (always)
932 // return a correct size so that the next addr + size below gives us a
933 // valid block boundary. [See for instance,
934 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
935 // in ConcurrentMarkSweepGeneration.cpp.]
936 HeapWord*
937 CompactibleFreeListSpace::object_iterate_careful_m(MemRegion mr,
938 ObjectClosureCareful* cl) {
939 assert_lock_strong(freelistLock());
940 // Can't use used_region() below because it may not necessarily
941 // be the same as [bottom(),end()); although we could
942 // use [used_region().start(),round_to(used_region().end(),CardSize)),
943 // that appears too cumbersome, so we just do the simpler check
944 // in the assertion below.
945 assert(!mr.is_empty() && MemRegion(bottom(),end()).contains(mr),
946 "mr should be non-empty and within used space");
947 HeapWord *addr, *end;
948 size_t size;
949 for (addr = block_start_careful(mr.start()), end = mr.end();
950 addr < end; addr += size) {
951 FreeChunk* fc = (FreeChunk*)addr;
952 if (fc->isFree()) {
953 // Since we hold the free list lock, which protects direct
954 // allocation in this generation by mutators, a free object
955 // will remain free throughout this iteration code.
956 size = fc->size();
957 } else {
958 // Note that the object need not necessarily be initialized,
959 // because (for instance) the free list lock does NOT protect
960 // object initialization. The closure application below must
961 // therefore be correct in the face of uninitialized objects.
962 size = cl->do_object_careful_m(oop(addr), mr);
963 if (size == 0) {
964 // An unparsable object found. Signal early termination.
965 return addr;
966 }
967 }
968 }
969 return NULL;
970 }
971
972
973 HeapWord* CompactibleFreeListSpace::block_start_const(const void* p) const {
974 NOT_PRODUCT(verify_objects_initialized());
975 return _bt.block_start(p);
976 }
977
978 HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const {
979 return _bt.block_start_careful(p);
980 }
981
982 size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const {
983 NOT_PRODUCT(verify_objects_initialized());
984 // This must be volatile, or else there is a danger that the compiler
985 // will compile the code below into a sometimes-infinite loop, by keeping
986 // the value read the first time in a register.
987 while (true) {
988 // We must do this until we get a consistent view of the object.
989 if (FreeChunk::indicatesFreeChunk(p)) {
990 volatile FreeChunk* fc = (volatile FreeChunk*)p;
991 size_t res = fc->size();
992 // If the object is still a free chunk, return the size, else it
993 // has been allocated so try again.
994 if (FreeChunk::indicatesFreeChunk(p)) {
995 assert(res != 0, "Block size should not be 0");
996 return res;
997 }
998 } else {
999 // must read from what 'p' points to in each loop.
1000 klassOop k = ((volatile oopDesc*)p)->klass_or_null();
1001 if (k != NULL) {
1002 assert(k->is_oop(true /* ignore mark word */), "Should be klass oop");
1003 oop o = (oop)p;
1004 assert(o->is_parsable(), "Should be parsable");
1005 assert(o->is_oop(true /* ignore mark word */), "Should be an oop.");
1006 size_t res = o->size_given_klass(k->klass_part());
1007 res = adjustObjectSize(res);
1008 assert(res != 0, "Block size should not be 0");
1009 return res;
1010 }
1011 }
1012 }
1013 }
1014
1015 // A variant of the above that uses the Printezis bits for
1016 // unparsable but allocated objects. This avoids any possible
1017 // stalls waiting for mutators to initialize objects, and is
1018 // thus potentially faster than the variant above. However,
1019 // this variant may return a zero size for a block that is
1020 // under mutation and for which a consistent size cannot be
1021 // inferred without stalling; see CMSCollector::block_size_if_printezis_bits().
1022 size_t CompactibleFreeListSpace::block_size_no_stall(HeapWord* p,
1023 const CMSCollector* c)
1024 const {
1025 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1026 // This must be volatile, or else there is a danger that the compiler
1027 // will compile the code below into a sometimes-infinite loop, by keeping
1028 // the value read the first time in a register.
1029 DEBUG_ONLY(uint loops = 0;)
1030 while (true) {
1031 // We must do this until we get a consistent view of the object.
1032 if (FreeChunk::indicatesFreeChunk(p)) {
1033 volatile FreeChunk* fc = (volatile FreeChunk*)p;
1034 size_t res = fc->size();
1035 if (FreeChunk::indicatesFreeChunk(p)) {
1036 assert(res != 0, "Block size should not be 0");
1037 assert(loops == 0, "Should be 0");
1038 return res;
1039 }
1040 } else {
1041 // must read from what 'p' points to in each loop.
1042 klassOop k = ((volatile oopDesc*)p)->klass_or_null();
1043 if (k != NULL &&
1044 ((oopDesc*)p)->is_parsable() &&
1045 ((oopDesc*)p)->is_conc_safe()) {
1046 assert(k->is_oop(), "Should really be klass oop.");
1047 oop o = (oop)p;
1048 assert(o->is_oop(), "Should be an oop");
1049 size_t res = o->size_given_klass(k->klass_part());
1050 res = adjustObjectSize(res);
1051 assert(res != 0, "Block size should not be 0");
1052 return res;
1053 } else {
1054 return c->block_size_if_printezis_bits(p);
1055 }
1056 }
1057 assert(loops == 0, "Can loop at most once");
1058 DEBUG_ONLY(loops++;)
1059 }
1060 }
1061
1062 size_t CompactibleFreeListSpace::block_size_nopar(const HeapWord* p) const {
1063 NOT_PRODUCT(verify_objects_initialized());
1064 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1065 FreeChunk* fc = (FreeChunk*)p;
1066 if (fc->isFree()) {
1067 return fc->size();
1068 } else {
1069 // Ignore mark word because this may be a recently promoted
1070 // object whose mark word is used to chain together grey
1071 // objects (the last one would have a null value).
1072 assert(oop(p)->is_oop(true), "Should be an oop");
1073 return adjustObjectSize(oop(p)->size());
1074 }
1075 }
1076
1077 // This implementation assumes that the property of "being an object" is
1078 // stable. But being a free chunk may not be (because of parallel
1079 // promotion.)
1080 bool CompactibleFreeListSpace::block_is_obj(const HeapWord* p) const {
1081 FreeChunk* fc = (FreeChunk*)p;
1082 assert(is_in_reserved(p), "Should be in space");
1083 // When doing a mark-sweep-compact of the CMS generation, this
1084 // assertion may fail because prepare_for_compaction() uses
1085 // space that is garbage to maintain information on ranges of
1086 // live objects so that these live ranges can be moved as a whole.
1087 // Comment out this assertion until that problem can be solved
1088 // (i.e., that the block start calculation may look at objects
1089 // at address below "p" in finding the object that contains "p"
1090 // and those objects (if garbage) may have been modified to hold
1091 // live range information.
1092 // assert(CollectedHeap::use_parallel_gc_threads() || _bt.block_start(p) == p,
1093 // "Should be a block boundary");
1094 if (FreeChunk::indicatesFreeChunk(p)) return false;
1095 klassOop k = oop(p)->klass_or_null();
1096 if (k != NULL) {
1097 // Ignore mark word because it may have been used to
1098 // chain together promoted objects (the last one
1099 // would have a null value).
1100 assert(oop(p)->is_oop(true), "Should be an oop");
1101 return true;
1102 } else {
1103 return false; // Was not an object at the start of collection.
1104 }
1105 }
1106
1107 // Check if the object is alive. This fact is checked either by consulting
1108 // the main marking bitmap in the sweeping phase or, if it's a permanent
1109 // generation and we're not in the sweeping phase, by checking the
1110 // perm_gen_verify_bit_map where we store the "deadness" information if
1111 // we did not sweep the perm gen in the most recent previous GC cycle.
1112 bool CompactibleFreeListSpace::obj_is_alive(const HeapWord* p) const {
1113 assert (block_is_obj(p), "The address should point to an object");
1114
1115 // If we're sweeping, we use object liveness information from the main bit map
1116 // for both perm gen and old gen.
1117 // We don't need to lock the bitmap (live_map or dead_map below), because
1118 // EITHER we are in the middle of the sweeping phase, and the
1119 // main marking bit map (live_map below) is locked,
1120 // OR we're in other phases and perm_gen_verify_bit_map (dead_map below)
1121 // is stable, because it's mutated only in the sweeping phase.
1122 if (_collector->abstract_state() == CMSCollector::Sweeping) {
1123 CMSBitMap* live_map = _collector->markBitMap();
1124 return live_map->isMarked((HeapWord*) p);
1125 } else {
1126 // If we're not currently sweeping and we haven't swept the perm gen in
1127 // the previous concurrent cycle then we may have dead but unswept objects
1128 // in the perm gen. In this case, we use the "deadness" information
1129 // that we had saved in perm_gen_verify_bit_map at the last sweep.
1130 if (!CMSClassUnloadingEnabled && _collector->_permGen->reserved().contains(p)) {
1131 if (_collector->verifying()) {
1132 CMSBitMap* dead_map = _collector->perm_gen_verify_bit_map();
1133 // Object is marked in the dead_map bitmap at the previous sweep
1134 // when we know that it's dead; if the bitmap is not allocated then
1135 // the object is alive.
1136 return (dead_map->sizeInBits() == 0) // bit_map has been allocated
1137 || !dead_map->par_isMarked((HeapWord*) p);
1138 } else {
1139 return false; // We can't say for sure if it's live, so we say that it's dead.
1140 }
1141 }
1142 }
1143 return true;
1144 }
1145
1146 bool CompactibleFreeListSpace::block_is_obj_nopar(const HeapWord* p) const {
1147 FreeChunk* fc = (FreeChunk*)p;
1148 assert(is_in_reserved(p), "Should be in space");
1149 assert(_bt.block_start(p) == p, "Should be a block boundary");
1150 if (!fc->isFree()) {
1151 // Ignore mark word because it may have been used to
1152 // chain together promoted objects (the last one
1153 // would have a null value).
1154 assert(oop(p)->is_oop(true), "Should be an oop");
1155 return true;
1156 }
1157 return false;
1158 }
1159
1160 // "MT-safe but not guaranteed MT-precise" (TM); you may get an
1161 // approximate answer if you don't hold the freelistlock when you call this.
1162 size_t CompactibleFreeListSpace::totalSizeInIndexedFreeLists() const {
1163 size_t size = 0;
1164 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
1165 debug_only(
1166 // We may be calling here without the lock in which case we
1167 // won't do this modest sanity check.
1168 if (freelistLock()->owned_by_self()) {
1169 size_t total_list_size = 0;
1170 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
1171 fc = fc->next()) {
1172 total_list_size += i;
1173 }
1174 assert(total_list_size == i * _indexedFreeList[i].count(),
1175 "Count in list is incorrect");
1176 }
1177 )
1178 size += i * _indexedFreeList[i].count();
1179 }
1180 return size;
1181 }
1182
1183 HeapWord* CompactibleFreeListSpace::par_allocate(size_t size) {
1184 MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
1185 return allocate(size);
1186 }
1187
1188 HeapWord*
1189 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlockRemainder(size_t size) {
1190 return getChunkFromLinearAllocBlockRemainder(&_smallLinearAllocBlock, size);
1191 }
1192
1193 HeapWord* CompactibleFreeListSpace::allocate(size_t size) {
1194 assert_lock_strong(freelistLock());
1195 HeapWord* res = NULL;
1196 assert(size == adjustObjectSize(size),
1197 "use adjustObjectSize() before calling into allocate()");
1198
1199 if (_adaptive_freelists) {
1200 res = allocate_adaptive_freelists(size);
1201 } else { // non-adaptive free lists
1202 res = allocate_non_adaptive_freelists(size);
1203 }
1204
1205 if (res != NULL) {
1206 // check that res does lie in this space!
1207 assert(is_in_reserved(res), "Not in this space!");
1208 assert(is_aligned((void*)res), "alignment check");
1209
1210 FreeChunk* fc = (FreeChunk*)res;
1211 fc->markNotFree();
1212 assert(!fc->isFree(), "shouldn't be marked free");
1213 assert(oop(fc)->klass_or_null() == NULL, "should look uninitialized");
1214 // Verify that the block offset table shows this to
1215 // be a single block, but not one which is unallocated.
1216 _bt.verify_single_block(res, size);
1217 _bt.verify_not_unallocated(res, size);
1218 // mangle a just allocated object with a distinct pattern.
1219 debug_only(fc->mangleAllocated(size));
1220 }
1221
1222 return res;
1223 }
1224
1225 HeapWord* CompactibleFreeListSpace::allocate_non_adaptive_freelists(size_t size) {
1226 HeapWord* res = NULL;
1227 // try and use linear allocation for smaller blocks
1228 if (size < _smallLinearAllocBlock._allocation_size_limit) {
1229 // if successful, the following also adjusts block offset table
1230 res = getChunkFromSmallLinearAllocBlock(size);
1231 }
1232 // Else triage to indexed lists for smaller sizes
1233 if (res == NULL) {
1234 if (size < SmallForDictionary) {
1235 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1236 } else {
1237 // else get it from the big dictionary; if even this doesn't
1238 // work we are out of luck.
1239 res = (HeapWord*)getChunkFromDictionaryExact(size);
1240 }
1241 }
1242
1243 return res;
1244 }
1245
1246 HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) {
1247 assert_lock_strong(freelistLock());
1248 HeapWord* res = NULL;
1249 assert(size == adjustObjectSize(size),
1250 "use adjustObjectSize() before calling into allocate()");
1251
1252 // Strategy
1253 // if small
1254 // exact size from small object indexed list if small
1255 // small or large linear allocation block (linAB) as appropriate
1256 // take from lists of greater sized chunks
1257 // else
1258 // dictionary
1259 // small or large linear allocation block if it has the space
1260 // Try allocating exact size from indexTable first
1261 if (size < IndexSetSize) {
1262 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1263 if(res != NULL) {
1264 assert(res != (HeapWord*)_indexedFreeList[size].head(),
1265 "Not removed from free list");
1266 // no block offset table adjustment is necessary on blocks in
1267 // the indexed lists.
1268
1269 // Try allocating from the small LinAB
1270 } else if (size < _smallLinearAllocBlock._allocation_size_limit &&
1271 (res = getChunkFromSmallLinearAllocBlock(size)) != NULL) {
1272 // if successful, the above also adjusts block offset table
1273 // Note that this call will refill the LinAB to
1274 // satisfy the request. This is different that
1275 // evm.
1276 // Don't record chunk off a LinAB? smallSplitBirth(size);
1277 } else {
1278 // Raid the exact free lists larger than size, even if they are not
1279 // overpopulated.
1280 res = (HeapWord*) getChunkFromGreater(size);
1281 }
1282 } else {
1283 // Big objects get allocated directly from the dictionary.
1284 res = (HeapWord*) getChunkFromDictionaryExact(size);
1285 if (res == NULL) {
1286 // Try hard not to fail since an allocation failure will likely
1287 // trigger a synchronous GC. Try to get the space from the
1288 // allocation blocks.
1289 res = getChunkFromSmallLinearAllocBlockRemainder(size);
1290 }
1291 }
1292
1293 return res;
1294 }
1295
1296 // A worst-case estimate of the space required (in HeapWords) to expand the heap
1297 // when promoting obj.
1298 size_t CompactibleFreeListSpace::expansionSpaceRequired(size_t obj_size) const {
1299 // Depending on the object size, expansion may require refilling either a
1300 // bigLAB or a smallLAB plus refilling a PromotionInfo object. MinChunkSize
1301 // is added because the dictionary may over-allocate to avoid fragmentation.
1302 size_t space = obj_size;
1303 if (!_adaptive_freelists) {
1304 space = MAX2(space, _smallLinearAllocBlock._refillSize);
1305 }
1306 space += _promoInfo.refillSize() + 2 * MinChunkSize;
1307 return space;
1308 }
1309
1310 FreeChunk* CompactibleFreeListSpace::getChunkFromGreater(size_t numWords) {
1311 FreeChunk* ret;
1312
1313 assert(numWords >= MinChunkSize, "Size is less than minimum");
1314 assert(linearAllocationWouldFail() || bestFitFirst(),
1315 "Should not be here");
1316
1317 size_t i;
1318 size_t currSize = numWords + MinChunkSize;
1319 assert(currSize % MinObjAlignment == 0, "currSize should be aligned");
1320 for (i = currSize; i < IndexSetSize; i += IndexSetStride) {
1321 FreeList* fl = &_indexedFreeList[i];
1322 if (fl->head()) {
1323 ret = getFromListGreater(fl, numWords);
1324 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk");
1325 return ret;
1326 }
1327 }
1328
1329 currSize = MAX2((size_t)SmallForDictionary,
1330 (size_t)(numWords + MinChunkSize));
1331
1332 /* Try to get a chunk that satisfies request, while avoiding
1333 fragmentation that can't be handled. */
1334 {
1335 ret = dictionary()->getChunk(currSize);
1336 if (ret != NULL) {
1337 assert(ret->size() - numWords >= MinChunkSize,
1338 "Chunk is too small");
1339 _bt.allocated((HeapWord*)ret, ret->size());
1340 /* Carve returned chunk. */
1341 (void) splitChunkAndReturnRemainder(ret, numWords);
1342 /* Label this as no longer a free chunk. */
1343 assert(ret->isFree(), "This chunk should be free");
1344 ret->linkPrev(NULL);
1345 }
1346 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk");
1347 return ret;
1348 }
1349 ShouldNotReachHere();
1350 }
1351
1352 bool CompactibleFreeListSpace::verifyChunkInIndexedFreeLists(FreeChunk* fc)
1353 const {
1354 assert(fc->size() < IndexSetSize, "Size of chunk is too large");
1355 return _indexedFreeList[fc->size()].verifyChunkInFreeLists(fc);
1356 }
1357
1358 bool CompactibleFreeListSpace::verifyChunkInFreeLists(FreeChunk* fc) const {
1359 if (fc->size() >= IndexSetSize) {
1360 return dictionary()->verifyChunkInFreeLists(fc);
1361 } else {
1362 return verifyChunkInIndexedFreeLists(fc);
1363 }
1364 }
1365
1366 #ifndef PRODUCT
1367 void CompactibleFreeListSpace::assert_locked() const {
1368 CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock());
1369 }
1370
1371 void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const {
1372 CMSLockVerifier::assert_locked(lock);
1373 }
1374 #endif
1375
1376 FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) {
1377 // In the parallel case, the main thread holds the free list lock
1378 // on behalf the parallel threads.
1379 FreeChunk* fc;
1380 {
1381 // If GC is parallel, this might be called by several threads.
1382 // This should be rare enough that the locking overhead won't affect
1383 // the sequential code.
1384 MutexLockerEx x(parDictionaryAllocLock(),
1385 Mutex::_no_safepoint_check_flag);
1386 fc = getChunkFromDictionary(size);
1387 }
1388 if (fc != NULL) {
1389 fc->dontCoalesce();
1390 assert(fc->isFree(), "Should be free, but not coalescable");
1391 // Verify that the block offset table shows this to
1392 // be a single block, but not one which is unallocated.
1393 _bt.verify_single_block((HeapWord*)fc, fc->size());
1394 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
1395 }
1396 return fc;
1397 }
1398
1399 oop CompactibleFreeListSpace::promote(oop obj, size_t obj_size) {
1400 assert(obj_size == (size_t)obj->size(), "bad obj_size passed in");
1401 assert_locked();
1402
1403 // if we are tracking promotions, then first ensure space for
1404 // promotion (including spooling space for saving header if necessary).
1405 // then allocate and copy, then track promoted info if needed.
1406 // When tracking (see PromotionInfo::track()), the mark word may
1407 // be displaced and in this case restoration of the mark word
1408 // occurs in the (oop_since_save_marks_)iterate phase.
1409 if (_promoInfo.tracking() && !_promoInfo.ensure_spooling_space()) {
1410 return NULL;
1411 }
1412 // Call the allocate(size_t, bool) form directly to avoid the
1413 // additional call through the allocate(size_t) form. Having
1414 // the compile inline the call is problematic because allocate(size_t)
1415 // is a virtual method.
1416 HeapWord* res = allocate(adjustObjectSize(obj_size));
1417 if (res != NULL) {
1418 Copy::aligned_disjoint_words((HeapWord*)obj, res, obj_size);
1419 // if we should be tracking promotions, do so.
1420 if (_promoInfo.tracking()) {
1421 _promoInfo.track((PromotedObject*)res);
1422 }
1423 }
1424 return oop(res);
1425 }
1426
1427 HeapWord*
1428 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlock(size_t size) {
1429 assert_locked();
1430 assert(size >= MinChunkSize, "minimum chunk size");
1431 assert(size < _smallLinearAllocBlock._allocation_size_limit,
1432 "maximum from smallLinearAllocBlock");
1433 return getChunkFromLinearAllocBlock(&_smallLinearAllocBlock, size);
1434 }
1435
1436 HeapWord*
1437 CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk,
1438 size_t size) {
1439 assert_locked();
1440 assert(size >= MinChunkSize, "too small");
1441 HeapWord* res = NULL;
1442 // Try to do linear allocation from blk, making sure that
1443 if (blk->_word_size == 0) {
1444 // We have probably been unable to fill this either in the prologue or
1445 // when it was exhausted at the last linear allocation. Bail out until
1446 // next time.
1447 assert(blk->_ptr == NULL, "consistency check");
1448 return NULL;
1449 }
1450 assert(blk->_word_size != 0 && blk->_ptr != NULL, "consistency check");
1451 res = getChunkFromLinearAllocBlockRemainder(blk, size);
1452 if (res != NULL) return res;
1453
1454 // about to exhaust this linear allocation block
1455 if (blk->_word_size == size) { // exactly satisfied
1456 res = blk->_ptr;
1457 _bt.allocated(res, blk->_word_size);
1458 } else if (size + MinChunkSize <= blk->_refillSize) {
1459 size_t sz = blk->_word_size;
1460 // Update _unallocated_block if the size is such that chunk would be
1461 // returned to the indexed free list. All other chunks in the indexed
1462 // free lists are allocated from the dictionary so that _unallocated_block
1463 // has already been adjusted for them. Do it here so that the cost
1464 // for all chunks added back to the indexed free lists.
1465 if (sz < SmallForDictionary) {
1466 _bt.allocated(blk->_ptr, sz);
1467 }
1468 // Return the chunk that isn't big enough, and then refill below.
1469 addChunkToFreeLists(blk->_ptr, sz);
1470 splitBirth(sz);
1471 // Don't keep statistics on adding back chunk from a LinAB.
1472 } else {
1473 // A refilled block would not satisfy the request.
1474 return NULL;
1475 }
1476
1477 blk->_ptr = NULL; blk->_word_size = 0;
1478 refillLinearAllocBlock(blk);
1479 assert(blk->_ptr == NULL || blk->_word_size >= size + MinChunkSize,
1480 "block was replenished");
1481 if (res != NULL) {
1482 splitBirth(size);
1483 repairLinearAllocBlock(blk);
1484 } else if (blk->_ptr != NULL) {
1485 res = blk->_ptr;
1486 size_t blk_size = blk->_word_size;
1487 blk->_word_size -= size;
1488 blk->_ptr += size;
1489 splitBirth(size);
1490 repairLinearAllocBlock(blk);
1491 // Update BOT last so that other (parallel) GC threads see a consistent
1492 // view of the BOT and free blocks.
1493 // Above must occur before BOT is updated below.
1494 OrderAccess::storestore();
1495 _bt.split_block(res, blk_size, size); // adjust block offset table
1496 }
1497 return res;
1498 }
1499
1500 HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder(
1501 LinearAllocBlock* blk,
1502 size_t size) {
1503 assert_locked();
1504 assert(size >= MinChunkSize, "too small");
1505
1506 HeapWord* res = NULL;
1507 // This is the common case. Keep it simple.
1508 if (blk->_word_size >= size + MinChunkSize) {
1509 assert(blk->_ptr != NULL, "consistency check");
1510 res = blk->_ptr;
1511 // Note that the BOT is up-to-date for the linAB before allocation. It
1512 // indicates the start of the linAB. The split_block() updates the
1513 // BOT for the linAB after the allocation (indicates the start of the
1514 // next chunk to be allocated).
1515 size_t blk_size = blk->_word_size;
1516 blk->_word_size -= size;
1517 blk->_ptr += size;
1518 splitBirth(size);
1519 repairLinearAllocBlock(blk);
1520 // Update BOT last so that other (parallel) GC threads see a consistent
1521 // view of the BOT and free blocks.
1522 // Above must occur before BOT is updated below.
1523 OrderAccess::storestore();
1524 _bt.split_block(res, blk_size, size); // adjust block offset table
1525 _bt.allocated(res, size);
1526 }
1527 return res;
1528 }
1529
1530 FreeChunk*
1531 CompactibleFreeListSpace::getChunkFromIndexedFreeList(size_t size) {
1532 assert_locked();
1533 assert(size < SmallForDictionary, "just checking");
1534 FreeChunk* res;
1535 res = _indexedFreeList[size].getChunkAtHead();
1536 if (res == NULL) {
1537 res = getChunkFromIndexedFreeListHelper(size);
1538 }
1539 _bt.verify_not_unallocated((HeapWord*) res, size);
1540 assert(res == NULL || res->size() == size, "Incorrect block size");
1541 return res;
1542 }
1543
1544 FreeChunk*
1545 CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size,
1546 bool replenish) {
1547 assert_locked();
1548 FreeChunk* fc = NULL;
1549 if (size < SmallForDictionary) {
1550 assert(_indexedFreeList[size].head() == NULL ||
1551 _indexedFreeList[size].surplus() <= 0,
1552 "List for this size should be empty or under populated");
1553 // Try best fit in exact lists before replenishing the list
1554 if (!bestFitFirst() || (fc = bestFitSmall(size)) == NULL) {
1555 // Replenish list.
1556 //
1557 // Things tried that failed.
1558 // Tried allocating out of the two LinAB's first before
1559 // replenishing lists.
1560 // Tried small linAB of size 256 (size in indexed list)
1561 // and replenishing indexed lists from the small linAB.
1562 //
1563 FreeChunk* newFc = NULL;
1564 const size_t replenish_size = CMSIndexedFreeListReplenish * size;
1565 if (replenish_size < SmallForDictionary) {
1566 // Do not replenish from an underpopulated size.
1567 if (_indexedFreeList[replenish_size].surplus() > 0 &&
1568 _indexedFreeList[replenish_size].head() != NULL) {
1569 newFc = _indexedFreeList[replenish_size].getChunkAtHead();
1570 } else if (bestFitFirst()) {
1571 newFc = bestFitSmall(replenish_size);
1572 }
1573 }
1574 if (newFc == NULL && replenish_size > size) {
1575 assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant");
1576 newFc = getChunkFromIndexedFreeListHelper(replenish_size, false);
1577 }
1578 // Note: The stats update re split-death of block obtained above
1579 // will be recorded below precisely when we know we are going to
1580 // be actually splitting it into more than one pieces below.
1581 if (newFc != NULL) {
1582 if (replenish || CMSReplenishIntermediate) {
1583 // Replenish this list and return one block to caller.
1584 size_t i;
1585 FreeChunk *curFc, *nextFc;
1586 size_t num_blk = newFc->size() / size;
1587 assert(num_blk >= 1, "Smaller than requested?");
1588 assert(newFc->size() % size == 0, "Should be integral multiple of request");
1589 if (num_blk > 1) {
1590 // we are sure we will be splitting the block just obtained
1591 // into multiple pieces; record the split-death of the original
1592 splitDeath(replenish_size);
1593 }
1594 // carve up and link blocks 0, ..., num_blk - 2
1595 // The last chunk is not added to the lists but is returned as the
1596 // free chunk.
1597 for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size),
1598 i = 0;
1599 i < (num_blk - 1);
1600 curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size),
1601 i++) {
1602 curFc->setSize(size);
1603 // Don't record this as a return in order to try and
1604 // determine the "returns" from a GC.
1605 _bt.verify_not_unallocated((HeapWord*) fc, size);
1606 _indexedFreeList[size].returnChunkAtTail(curFc, false);
1607 _bt.mark_block((HeapWord*)curFc, size);
1608 splitBirth(size);
1609 // Don't record the initial population of the indexed list
1610 // as a split birth.
1611 }
1612
1613 // check that the arithmetic was OK above
1614 assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size,
1615 "inconsistency in carving newFc");
1616 curFc->setSize(size);
1617 _bt.mark_block((HeapWord*)curFc, size);
1618 splitBirth(size);
1619 fc = curFc;
1620 } else {
1621 // Return entire block to caller
1622 fc = newFc;
1623 }
1624 }
1625 }
1626 } else {
1627 // Get a free chunk from the free chunk dictionary to be returned to
1628 // replenish the indexed free list.
1629 fc = getChunkFromDictionaryExact(size);
1630 }
1631 // assert(fc == NULL || fc->isFree(), "Should be returning a free chunk");
1632 return fc;
1633 }
1634
1635 FreeChunk*
1636 CompactibleFreeListSpace::getChunkFromDictionary(size_t size) {
1637 assert_locked();
1638 FreeChunk* fc = _dictionary->getChunk(size);
1639 if (fc == NULL) {
1640 return NULL;
1641 }
1642 _bt.allocated((HeapWord*)fc, fc->size());
1643 if (fc->size() >= size + MinChunkSize) {
1644 fc = splitChunkAndReturnRemainder(fc, size);
1645 }
1646 assert(fc->size() >= size, "chunk too small");
1647 assert(fc->size() < size + MinChunkSize, "chunk too big");
1648 _bt.verify_single_block((HeapWord*)fc, fc->size());
1649 return fc;
1650 }
1651
1652 FreeChunk*
1653 CompactibleFreeListSpace::getChunkFromDictionaryExact(size_t size) {
1654 assert_locked();
1655 FreeChunk* fc = _dictionary->getChunk(size);
1656 if (fc == NULL) {
1657 return fc;
1658 }
1659 _bt.allocated((HeapWord*)fc, fc->size());
1660 if (fc->size() == size) {
1661 _bt.verify_single_block((HeapWord*)fc, size);
1662 return fc;
1663 }
1664 assert(fc->size() > size, "getChunk() guarantee");
1665 if (fc->size() < size + MinChunkSize) {
1666 // Return the chunk to the dictionary and go get a bigger one.
1667 returnChunkToDictionary(fc);
1668 fc = _dictionary->getChunk(size + MinChunkSize);
1669 if (fc == NULL) {
1670 return NULL;
1671 }
1672 _bt.allocated((HeapWord*)fc, fc->size());
1673 }
1674 assert(fc->size() >= size + MinChunkSize, "tautology");
1675 fc = splitChunkAndReturnRemainder(fc, size);
1676 assert(fc->size() == size, "chunk is wrong size");
1677 _bt.verify_single_block((HeapWord*)fc, size);
1678 return fc;
1679 }
1680
1681 void
1682 CompactibleFreeListSpace::returnChunkToDictionary(FreeChunk* chunk) {
1683 assert_locked();
1684
1685 size_t size = chunk->size();
1686 _bt.verify_single_block((HeapWord*)chunk, size);
1687 // adjust _unallocated_block downward, as necessary
1688 _bt.freed((HeapWord*)chunk, size);
1689 _dictionary->returnChunk(chunk);
1690 #ifndef PRODUCT
1691 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1692 TreeChunk::as_TreeChunk(chunk)->list()->verify_stats();
1693 }
1694 #endif // PRODUCT
1695 }
1696
1697 void
1698 CompactibleFreeListSpace::returnChunkToFreeList(FreeChunk* fc) {
1699 assert_locked();
1700 size_t size = fc->size();
1701 _bt.verify_single_block((HeapWord*) fc, size);
1702 _bt.verify_not_unallocated((HeapWord*) fc, size);
1703 if (_adaptive_freelists) {
1704 _indexedFreeList[size].returnChunkAtTail(fc);
1705 } else {
1706 _indexedFreeList[size].returnChunkAtHead(fc);
1707 }
1708 #ifndef PRODUCT
1709 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1710 _indexedFreeList[size].verify_stats();
1711 }
1712 #endif // PRODUCT
1713 }
1714
1715 // Add chunk to end of last block -- if it's the largest
1716 // block -- and update BOT and census data. We would
1717 // of course have preferred to coalesce it with the
1718 // last block, but it's currently less expensive to find the
1719 // largest block than it is to find the last.
1720 void
1721 CompactibleFreeListSpace::addChunkToFreeListsAtEndRecordingStats(
1722 HeapWord* chunk, size_t size) {
1723 // check that the chunk does lie in this space!
1724 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1725 // One of the parallel gc task threads may be here
1726 // whilst others are allocating.
1727 Mutex* lock = NULL;
1728 if (ParallelGCThreads != 0) {
1729 lock = &_parDictionaryAllocLock;
1730 }
1731 FreeChunk* ec;
1732 {
1733 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1734 ec = dictionary()->findLargestDict(); // get largest block
1735 if (ec != NULL && ec->end() == chunk) {
1736 // It's a coterminal block - we can coalesce.
1737 size_t old_size = ec->size();
1738 coalDeath(old_size);
1739 removeChunkFromDictionary(ec);
1740 size += old_size;
1741 } else {
1742 ec = (FreeChunk*)chunk;
1743 }
1744 }
1745 ec->setSize(size);
1746 debug_only(ec->mangleFreed(size));
1747 if (size < SmallForDictionary) {
1748 lock = _indexedFreeListParLocks[size];
1749 }
1750 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1751 addChunkAndRepairOffsetTable((HeapWord*)ec, size, true);
1752 // record the birth under the lock since the recording involves
1753 // manipulation of the list on which the chunk lives and
1754 // if the chunk is allocated and is the last on the list,
1755 // the list can go away.
1756 coalBirth(size);
1757 }
1758
1759 void
1760 CompactibleFreeListSpace::addChunkToFreeLists(HeapWord* chunk,
1761 size_t size) {
1762 // check that the chunk does lie in this space!
1763 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1764 assert_locked();
1765 _bt.verify_single_block(chunk, size);
1766
1767 FreeChunk* fc = (FreeChunk*) chunk;
1768 fc->setSize(size);
1769 debug_only(fc->mangleFreed(size));
1770 if (size < SmallForDictionary) {
1771 returnChunkToFreeList(fc);
1772 } else {
1773 returnChunkToDictionary(fc);
1774 }
1775 }
1776
1777 void
1778 CompactibleFreeListSpace::addChunkAndRepairOffsetTable(HeapWord* chunk,
1779 size_t size, bool coalesced) {
1780 assert_locked();
1781 assert(chunk != NULL, "null chunk");
1782 if (coalesced) {
1783 // repair BOT
1784 _bt.single_block(chunk, size);
1785 }
1786 addChunkToFreeLists(chunk, size);
1787 }
1788
1789 // We _must_ find the purported chunk on our free lists;
1790 // we assert if we don't.
1791 void
1792 CompactibleFreeListSpace::removeFreeChunkFromFreeLists(FreeChunk* fc) {
1793 size_t size = fc->size();
1794 assert_locked();
1795 debug_only(verifyFreeLists());
1796 if (size < SmallForDictionary) {
1797 removeChunkFromIndexedFreeList(fc);
1798 } else {
1799 removeChunkFromDictionary(fc);
1800 }
1801 _bt.verify_single_block((HeapWord*)fc, size);
1802 debug_only(verifyFreeLists());
1803 }
1804
1805 void
1806 CompactibleFreeListSpace::removeChunkFromDictionary(FreeChunk* fc) {
1807 size_t size = fc->size();
1808 assert_locked();
1809 assert(fc != NULL, "null chunk");
1810 _bt.verify_single_block((HeapWord*)fc, size);
1811 _dictionary->removeChunk(fc);
1812 // adjust _unallocated_block upward, as necessary
1813 _bt.allocated((HeapWord*)fc, size);
1814 }
1815
1816 void
1817 CompactibleFreeListSpace::removeChunkFromIndexedFreeList(FreeChunk* fc) {
1818 assert_locked();
1819 size_t size = fc->size();
1820 _bt.verify_single_block((HeapWord*)fc, size);
1821 NOT_PRODUCT(
1822 if (FLSVerifyIndexTable) {
1823 verifyIndexedFreeList(size);
1824 }
1825 )
1826 _indexedFreeList[size].removeChunk(fc);
1827 debug_only(fc->clearNext());
1828 debug_only(fc->clearPrev());
1829 NOT_PRODUCT(
1830 if (FLSVerifyIndexTable) {
1831 verifyIndexedFreeList(size);
1832 }
1833 )
1834 }
1835
1836 FreeChunk* CompactibleFreeListSpace::bestFitSmall(size_t numWords) {
1837 /* A hint is the next larger size that has a surplus.
1838 Start search at a size large enough to guarantee that
1839 the excess is >= MIN_CHUNK. */
1840 size_t start = align_object_size(numWords + MinChunkSize);
1841 if (start < IndexSetSize) {
1842 FreeList* it = _indexedFreeList;
1843 size_t hint = _indexedFreeList[start].hint();
1844 while (hint < IndexSetSize) {
1845 assert(hint % MinObjAlignment == 0, "hint should be aligned");
1846 FreeList *fl = &_indexedFreeList[hint];
1847 if (fl->surplus() > 0 && fl->head() != NULL) {
1848 // Found a list with surplus, reset original hint
1849 // and split out a free chunk which is returned.
1850 _indexedFreeList[start].set_hint(hint);
1851 FreeChunk* res = getFromListGreater(fl, numWords);
1852 assert(res == NULL || res->isFree(),
1853 "Should be returning a free chunk");
1854 return res;
1855 }
1856 hint = fl->hint(); /* keep looking */
1857 }
1858 /* None found. */
1859 it[start].set_hint(IndexSetSize);
1860 }
1861 return NULL;
1862 }
1863
1864 /* Requires fl->size >= numWords + MinChunkSize */
1865 FreeChunk* CompactibleFreeListSpace::getFromListGreater(FreeList* fl,
1866 size_t numWords) {
1867 FreeChunk *curr = fl->head();
1868 size_t oldNumWords = curr->size();
1869 assert(numWords >= MinChunkSize, "Word size is too small");
1870 assert(curr != NULL, "List is empty");
1871 assert(oldNumWords >= numWords + MinChunkSize,
1872 "Size of chunks in the list is too small");
1873
1874 fl->removeChunk(curr);
1875 // recorded indirectly by splitChunkAndReturnRemainder -
1876 // smallSplit(oldNumWords, numWords);
1877 FreeChunk* new_chunk = splitChunkAndReturnRemainder(curr, numWords);
1878 // Does anything have to be done for the remainder in terms of
1879 // fixing the card table?
1880 assert(new_chunk == NULL || new_chunk->isFree(),
1881 "Should be returning a free chunk");
1882 return new_chunk;
1883 }
1884
1885 FreeChunk*
1886 CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk,
1887 size_t new_size) {
1888 assert_locked();
1889 size_t size = chunk->size();
1890 assert(size > new_size, "Split from a smaller block?");
1891 assert(is_aligned(chunk), "alignment problem");
1892 assert(size == adjustObjectSize(size), "alignment problem");
1893 size_t rem_size = size - new_size;
1894 assert(rem_size == adjustObjectSize(rem_size), "alignment problem");
1895 assert(rem_size >= MinChunkSize, "Free chunk smaller than minimum");
1896 FreeChunk* ffc = (FreeChunk*)((HeapWord*)chunk + new_size);
1897 assert(is_aligned(ffc), "alignment problem");
1898 ffc->setSize(rem_size);
1899 ffc->linkNext(NULL);
1900 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
1901 // Above must occur before BOT is updated below.
1902 // adjust block offset table
1903 OrderAccess::storestore();
1904 assert(chunk->isFree() && ffc->isFree(), "Error");
1905 _bt.split_block((HeapWord*)chunk, chunk->size(), new_size);
1906 if (rem_size < SmallForDictionary) {
1907 bool is_par = (SharedHeap::heap()->n_par_threads() > 0);
1908 if (is_par) _indexedFreeListParLocks[rem_size]->lock();
1909 returnChunkToFreeList(ffc);
1910 split(size, rem_size);
1911 if (is_par) _indexedFreeListParLocks[rem_size]->unlock();
1912 } else {
1913 returnChunkToDictionary(ffc);
1914 split(size ,rem_size);
1915 }
1916 chunk->setSize(new_size);
1917 return chunk;
1918 }
1919
1920 void
1921 CompactibleFreeListSpace::sweep_completed() {
1922 // Now that space is probably plentiful, refill linear
1923 // allocation blocks as needed.
1924 refillLinearAllocBlocksIfNeeded();
1925 }
1926
1927 void
1928 CompactibleFreeListSpace::gc_prologue() {
1929 assert_locked();
1930 if (PrintFLSStatistics != 0) {
1931 gclog_or_tty->print("Before GC:\n");
1932 reportFreeListStatistics();
1933 }
1934 refillLinearAllocBlocksIfNeeded();
1935 }
1936
1937 void
1938 CompactibleFreeListSpace::gc_epilogue() {
1939 assert_locked();
1940 if (PrintGCDetails && Verbose && !_adaptive_freelists) {
1941 if (_smallLinearAllocBlock._word_size == 0)
1942 warning("CompactibleFreeListSpace(epilogue):: Linear allocation failure");
1943 }
1944 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1945 _promoInfo.stopTrackingPromotions();
1946 repairLinearAllocationBlocks();
1947 // Print Space's stats
1948 if (PrintFLSStatistics != 0) {
1949 gclog_or_tty->print("After GC:\n");
1950 reportFreeListStatistics();
1951 }
1952 }
1953
1954 // Iteration support, mostly delegated from a CMS generation
1955
1956 void CompactibleFreeListSpace::save_marks() {
1957 // mark the "end" of the used space at the time of this call;
1958 // note, however, that promoted objects from this point
1959 // on are tracked in the _promoInfo below.
1960 set_saved_mark_word(unallocated_block());
1961 // inform allocator that promotions should be tracked.
1962 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1963 _promoInfo.startTrackingPromotions();
1964 }
1965
1966 bool CompactibleFreeListSpace::no_allocs_since_save_marks() {
1967 assert(_promoInfo.tracking(), "No preceding save_marks?");
1968 assert(SharedHeap::heap()->n_par_threads() == 0,
1969 "Shouldn't be called if using parallel gc.");
1970 return _promoInfo.noPromotions();
1971 }
1972
1973 #define CFLS_OOP_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix) \
1974 \
1975 void CompactibleFreeListSpace:: \
1976 oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk) { \
1977 assert(SharedHeap::heap()->n_par_threads() == 0, \
1978 "Shouldn't be called (yet) during parallel part of gc."); \
1979 _promoInfo.promoted_oops_iterate##nv_suffix(blk); \
1980 /* \
1981 * This also restores any displaced headers and removes the elements from \
1982 * the iteration set as they are processed, so that we have a clean slate \
1983 * at the end of the iteration. Note, thus, that if new objects are \
1984 * promoted as a result of the iteration they are iterated over as well. \
1985 */ \
1986 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); \
1987 }
1988
1989 ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DEFN)
1990
1991
1992 void CompactibleFreeListSpace::object_iterate_since_last_GC(ObjectClosure* cl) {
1993 // ugghh... how would one do this efficiently for a non-contiguous space?
1994 guarantee(false, "NYI");
1995 }
1996
1997 bool CompactibleFreeListSpace::linearAllocationWouldFail() const {
1998 return _smallLinearAllocBlock._word_size == 0;
1999 }
2000
2001 void CompactibleFreeListSpace::repairLinearAllocationBlocks() {
2002 // Fix up linear allocation blocks to look like free blocks
2003 repairLinearAllocBlock(&_smallLinearAllocBlock);
2004 }
2005
2006 void CompactibleFreeListSpace::repairLinearAllocBlock(LinearAllocBlock* blk) {
2007 assert_locked();
2008 if (blk->_ptr != NULL) {
2009 assert(blk->_word_size != 0 && blk->_word_size >= MinChunkSize,
2010 "Minimum block size requirement");
2011 FreeChunk* fc = (FreeChunk*)(blk->_ptr);
2012 fc->setSize(blk->_word_size);
2013 fc->linkPrev(NULL); // mark as free
2014 fc->dontCoalesce();
2015 assert(fc->isFree(), "just marked it free");
2016 assert(fc->cantCoalesce(), "just marked it uncoalescable");
2017 }
2018 }
2019
2020 void CompactibleFreeListSpace::refillLinearAllocBlocksIfNeeded() {
2021 assert_locked();
2022 if (_smallLinearAllocBlock._ptr == NULL) {
2023 assert(_smallLinearAllocBlock._word_size == 0,
2024 "Size of linAB should be zero if the ptr is NULL");
2025 // Reset the linAB refill and allocation size limit.
2026 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, SmallForLinearAlloc);
2027 }
2028 refillLinearAllocBlockIfNeeded(&_smallLinearAllocBlock);
2029 }
2030
2031 void
2032 CompactibleFreeListSpace::refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk) {
2033 assert_locked();
2034 assert((blk->_ptr == NULL && blk->_word_size == 0) ||
2035 (blk->_ptr != NULL && blk->_word_size >= MinChunkSize),
2036 "blk invariant");
2037 if (blk->_ptr == NULL) {
2038 refillLinearAllocBlock(blk);
2039 }
2040 if (PrintMiscellaneous && Verbose) {
2041 if (blk->_word_size == 0) {
2042 warning("CompactibleFreeListSpace(prologue):: Linear allocation failure");
2043 }
2044 }
2045 }
2046
2047 void
2048 CompactibleFreeListSpace::refillLinearAllocBlock(LinearAllocBlock* blk) {
2049 assert_locked();
2050 assert(blk->_word_size == 0 && blk->_ptr == NULL,
2051 "linear allocation block should be empty");
2052 FreeChunk* fc;
2053 if (blk->_refillSize < SmallForDictionary &&
2054 (fc = getChunkFromIndexedFreeList(blk->_refillSize)) != NULL) {
2055 // A linAB's strategy might be to use small sizes to reduce
2056 // fragmentation but still get the benefits of allocation from a
2057 // linAB.
2058 } else {
2059 fc = getChunkFromDictionary(blk->_refillSize);
2060 }
2061 if (fc != NULL) {
2062 blk->_ptr = (HeapWord*)fc;
2063 blk->_word_size = fc->size();
2064 fc->dontCoalesce(); // to prevent sweeper from sweeping us up
2065 }
2066 }
2067
2068 // Support for concurrent collection policy decisions.
2069 bool CompactibleFreeListSpace::should_concurrent_collect() const {
2070 // In the future we might want to add in frgamentation stats --
2071 // including erosion of the "mountain" into this decision as well.
2072 return !adaptive_freelists() && linearAllocationWouldFail();
2073 }
2074
2075 // Support for compaction
2076
2077 void CompactibleFreeListSpace::prepare_for_compaction(CompactPoint* cp) {
2078 SCAN_AND_FORWARD(cp,end,block_is_obj,block_size);
2079 // prepare_for_compaction() uses the space between live objects
2080 // so that later phase can skip dead space quickly. So verification
2081 // of the free lists doesn't work after.
2082 }
2083
2084 #define obj_size(q) adjustObjectSize(oop(q)->size())
2085 #define adjust_obj_size(s) adjustObjectSize(s)
2086
2087 void CompactibleFreeListSpace::adjust_pointers() {
2088 // In other versions of adjust_pointers(), a bail out
2089 // based on the amount of live data in the generation
2090 // (i.e., if 0, bail out) may be used.
2091 // Cannot test used() == 0 here because the free lists have already
2092 // been mangled by the compaction.
2093
2094 SCAN_AND_ADJUST_POINTERS(adjust_obj_size);
2095 // See note about verification in prepare_for_compaction().
2096 }
2097
2098 void CompactibleFreeListSpace::compact() {
2099 SCAN_AND_COMPACT(obj_size);
2100 }
2101
2102 // fragmentation_metric = 1 - [sum of (fbs**2) / (sum of fbs)**2]
2103 // where fbs is free block sizes
2104 double CompactibleFreeListSpace::flsFrag() const {
2105 size_t itabFree = totalSizeInIndexedFreeLists();
2106 double frag = 0.0;
2107 size_t i;
2108
2109 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2110 double sz = i;
2111 frag += _indexedFreeList[i].count() * (sz * sz);
2112 }
2113
2114 double totFree = itabFree +
2115 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock()));
2116 if (totFree > 0) {
2117 frag = ((frag + _dictionary->sum_of_squared_block_sizes()) /
2118 (totFree * totFree));
2119 frag = (double)1.0 - frag;
2120 } else {
2121 assert(frag == 0.0, "Follows from totFree == 0");
2122 }
2123 return frag;
2124 }
2125
2126 void CompactibleFreeListSpace::beginSweepFLCensus(
2127 float inter_sweep_current,
2128 float inter_sweep_estimate,
2129 float intra_sweep_estimate) {
2130 assert_locked();
2131 size_t i;
2132 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2133 FreeList* fl = &_indexedFreeList[i];
2134 if (PrintFLSStatistics > 1) {
2135 gclog_or_tty->print("size[%d] : ", i);
2136 }
2137 fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate);
2138 fl->set_coalDesired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent));
2139 fl->set_beforeSweep(fl->count());
2140 fl->set_bfrSurp(fl->surplus());
2141 }
2142 _dictionary->beginSweepDictCensus(CMSLargeCoalSurplusPercent,
2143 inter_sweep_current,
2144 inter_sweep_estimate,
2145 intra_sweep_estimate);
2146 }
2147
2148 void CompactibleFreeListSpace::setFLSurplus() {
2149 assert_locked();
2150 size_t i;
2151 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2152 FreeList *fl = &_indexedFreeList[i];
2153 fl->set_surplus(fl->count() -
2154 (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent));
2155 }
2156 }
2157
2158 void CompactibleFreeListSpace::setFLHints() {
2159 assert_locked();
2160 size_t i;
2161 size_t h = IndexSetSize;
2162 for (i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
2163 FreeList *fl = &_indexedFreeList[i];
2164 fl->set_hint(h);
2165 if (fl->surplus() > 0) {
2166 h = i;
2167 }
2168 }
2169 }
2170
2171 void CompactibleFreeListSpace::clearFLCensus() {
2172 assert_locked();
2173 int i;
2174 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2175 FreeList *fl = &_indexedFreeList[i];
2176 fl->set_prevSweep(fl->count());
2177 fl->set_coalBirths(0);
2178 fl->set_coalDeaths(0);
2179 fl->set_splitBirths(0);
2180 fl->set_splitDeaths(0);
2181 }
2182 }
2183
2184 void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) {
2185 if (PrintFLSStatistics > 0) {
2186 HeapWord* largestAddr = (HeapWord*) dictionary()->findLargestDict();
2187 gclog_or_tty->print_cr("CMS: Large block " PTR_FORMAT,
2188 largestAddr);
2189 }
2190 setFLSurplus();
2191 setFLHints();
2192 if (PrintGC && PrintFLSCensus > 0) {
2193 printFLCensus(sweep_count);
2194 }
2195 clearFLCensus();
2196 assert_locked();
2197 _dictionary->endSweepDictCensus(CMSLargeSplitSurplusPercent);
2198 }
2199
2200 bool CompactibleFreeListSpace::coalOverPopulated(size_t size) {
2201 if (size < SmallForDictionary) {
2202 FreeList *fl = &_indexedFreeList[size];
2203 return (fl->coalDesired() < 0) ||
2204 ((int)fl->count() > fl->coalDesired());
2205 } else {
2206 return dictionary()->coalDictOverPopulated(size);
2207 }
2208 }
2209
2210 void CompactibleFreeListSpace::smallCoalBirth(size_t size) {
2211 assert(size < SmallForDictionary, "Size too large for indexed list");
2212 FreeList *fl = &_indexedFreeList[size];
2213 fl->increment_coalBirths();
2214 fl->increment_surplus();
2215 }
2216
2217 void CompactibleFreeListSpace::smallCoalDeath(size_t size) {
2218 assert(size < SmallForDictionary, "Size too large for indexed list");
2219 FreeList *fl = &_indexedFreeList[size];
2220 fl->increment_coalDeaths();
2221 fl->decrement_surplus();
2222 }
2223
2224 void CompactibleFreeListSpace::coalBirth(size_t size) {
2225 if (size < SmallForDictionary) {
2226 smallCoalBirth(size);
2227 } else {
2228 dictionary()->dictCensusUpdate(size,
2229 false /* split */,
2230 true /* birth */);
2231 }
2232 }
2233
2234 void CompactibleFreeListSpace::coalDeath(size_t size) {
2235 if(size < SmallForDictionary) {
2236 smallCoalDeath(size);
2237 } else {
2238 dictionary()->dictCensusUpdate(size,
2239 false /* split */,
2240 false /* birth */);
2241 }
2242 }
2243
2244 void CompactibleFreeListSpace::smallSplitBirth(size_t size) {
2245 assert(size < SmallForDictionary, "Size too large for indexed list");
2246 FreeList *fl = &_indexedFreeList[size];
2247 fl->increment_splitBirths();
2248 fl->increment_surplus();
2249 }
2250
2251 void CompactibleFreeListSpace::smallSplitDeath(size_t size) {
2252 assert(size < SmallForDictionary, "Size too large for indexed list");
2253 FreeList *fl = &_indexedFreeList[size];
2254 fl->increment_splitDeaths();
2255 fl->decrement_surplus();
2256 }
2257
2258 void CompactibleFreeListSpace::splitBirth(size_t size) {
2259 if (size < SmallForDictionary) {
2260 smallSplitBirth(size);
2261 } else {
2262 dictionary()->dictCensusUpdate(size,
2263 true /* split */,
2264 true /* birth */);
2265 }
2266 }
2267
2268 void CompactibleFreeListSpace::splitDeath(size_t size) {
2269 if (size < SmallForDictionary) {
2270 smallSplitDeath(size);
2271 } else {
2272 dictionary()->dictCensusUpdate(size,
2273 true /* split */,
2274 false /* birth */);
2275 }
2276 }
2277
2278 void CompactibleFreeListSpace::split(size_t from, size_t to1) {
2279 size_t to2 = from - to1;
2280 splitDeath(from);
2281 splitBirth(to1);
2282 splitBirth(to2);
2283 }
2284
2285 void CompactibleFreeListSpace::print() const {
2286 Space::print_on(tty);
2287 }
2288
2289 void CompactibleFreeListSpace::prepare_for_verify() {
2290 assert_locked();
2291 repairLinearAllocationBlocks();
2292 // Verify that the SpoolBlocks look like free blocks of
2293 // appropriate sizes... To be done ...
2294 }
2295
2296 class VerifyAllBlksClosure: public BlkClosure {
2297 private:
2298 const CompactibleFreeListSpace* _sp;
2299 const MemRegion _span;
2300 HeapWord* _last_addr;
2301 size_t _last_size;
2302 bool _last_was_obj;
2303 bool _last_was_live;
2304
2305 public:
2306 VerifyAllBlksClosure(const CompactibleFreeListSpace* sp,
2307 MemRegion span) : _sp(sp), _span(span),
2308 _last_addr(NULL), _last_size(0),
2309 _last_was_obj(false), _last_was_live(false) { }
2310
2311 virtual size_t do_blk(HeapWord* addr) {
2312 size_t res;
2313 bool was_obj = false;
2314 bool was_live = false;
2315 if (_sp->block_is_obj(addr)) {
2316 was_obj = true;
2317 oop p = oop(addr);
2318 guarantee(p->is_oop(), "Should be an oop");
2319 res = _sp->adjustObjectSize(p->size());
2320 if (_sp->obj_is_alive(addr)) {
2321 was_live = true;
2322 p->verify();
2323 }
2324 } else {
2325 FreeChunk* fc = (FreeChunk*)addr;
2326 res = fc->size();
2327 if (FLSVerifyLists && !fc->cantCoalesce()) {
2328 guarantee(_sp->verifyChunkInFreeLists(fc),
2329 "Chunk should be on a free list");
2330 }
2331 }
2332 if (res == 0) {
2333 gclog_or_tty->print_cr("Livelock: no rank reduction!");
2334 gclog_or_tty->print_cr(
2335 " Current: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n"
2336 " Previous: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n",
2337 addr, res, was_obj ?"true":"false", was_live ?"true":"false",
2338 _last_addr, _last_size, _last_was_obj?"true":"false", _last_was_live?"true":"false");
2339 _sp->print_on(gclog_or_tty);
2340 guarantee(false, "Seppuku!");
2341 }
2342 _last_addr = addr;
2343 _last_size = res;
2344 _last_was_obj = was_obj;
2345 _last_was_live = was_live;
2346 return res;
2347 }
2348 };
2349
2350 class VerifyAllOopsClosure: public OopClosure {
2351 private:
2352 const CMSCollector* _collector;
2353 const CompactibleFreeListSpace* _sp;
2354 const MemRegion _span;
2355 const bool _past_remark;
2356 const CMSBitMap* _bit_map;
2357
2358 protected:
2359 void do_oop(void* p, oop obj) {
2360 if (_span.contains(obj)) { // the interior oop points into CMS heap
2361 if (!_span.contains(p)) { // reference from outside CMS heap
2362 // Should be a valid object; the first disjunct below allows
2363 // us to sidestep an assertion in block_is_obj() that insists
2364 // that p be in _sp. Note that several generations (and spaces)
2365 // are spanned by _span (CMS heap) above.
2366 guarantee(!_sp->is_in_reserved(obj) ||
2367 _sp->block_is_obj((HeapWord*)obj),
2368 "Should be an object");
2369 guarantee(obj->is_oop(), "Should be an oop");
2370 obj->verify();
2371 if (_past_remark) {
2372 // Remark has been completed, the object should be marked
2373 _bit_map->isMarked((HeapWord*)obj);
2374 }
2375 } else { // reference within CMS heap
2376 if (_past_remark) {
2377 // Remark has been completed -- so the referent should have
2378 // been marked, if referring object is.
2379 if (_bit_map->isMarked(_collector->block_start(p))) {
2380 guarantee(_bit_map->isMarked((HeapWord*)obj), "Marking error?");
2381 }
2382 }
2383 }
2384 } else if (_sp->is_in_reserved(p)) {
2385 // the reference is from FLS, and points out of FLS
2386 guarantee(obj->is_oop(), "Should be an oop");
2387 obj->verify();
2388 }
2389 }
2390
2391 template <class T> void do_oop_work(T* p) {
2392 T heap_oop = oopDesc::load_heap_oop(p);
2393 if (!oopDesc::is_null(heap_oop)) {
2394 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
2395 do_oop(p, obj);
2396 }
2397 }
2398
2399 public:
2400 VerifyAllOopsClosure(const CMSCollector* collector,
2401 const CompactibleFreeListSpace* sp, MemRegion span,
2402 bool past_remark, CMSBitMap* bit_map) :
2403 OopClosure(), _collector(collector), _sp(sp), _span(span),
2404 _past_remark(past_remark), _bit_map(bit_map) { }
2405
2406 virtual void do_oop(oop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2407 virtual void do_oop(narrowOop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2408 };
2409
2410 void CompactibleFreeListSpace::verify(bool ignored) const {
2411 assert_lock_strong(&_freelistLock);
2412 verify_objects_initialized();
2413 MemRegion span = _collector->_span;
2414 bool past_remark = (_collector->abstract_state() ==
2415 CMSCollector::Sweeping);
2416
2417 ResourceMark rm;
2418 HandleMark hm;
2419
2420 // Check integrity of CFL data structures
2421 _promoInfo.verify();
2422 _dictionary->verify();
2423 if (FLSVerifyIndexTable) {
2424 verifyIndexedFreeLists();
2425 }
2426 // Check integrity of all objects and free blocks in space
2427 {
2428 VerifyAllBlksClosure cl(this, span);
2429 ((CompactibleFreeListSpace*)this)->blk_iterate(&cl); // cast off const
2430 }
2431 // Check that all references in the heap to FLS
2432 // are to valid objects in FLS or that references in
2433 // FLS are to valid objects elsewhere in the heap
2434 if (FLSVerifyAllHeapReferences)
2435 {
2436 VerifyAllOopsClosure cl(_collector, this, span, past_remark,
2437 _collector->markBitMap());
2438 CollectedHeap* ch = Universe::heap();
2439 ch->oop_iterate(&cl); // all oops in generations
2440 ch->permanent_oop_iterate(&cl); // all oops in perm gen
2441 }
2442
2443 if (VerifyObjectStartArray) {
2444 // Verify the block offset table
2445 _bt.verify();
2446 }
2447 }
2448
2449 #ifndef PRODUCT
2450 void CompactibleFreeListSpace::verifyFreeLists() const {
2451 if (FLSVerifyLists) {
2452 _dictionary->verify();
2453 verifyIndexedFreeLists();
2454 } else {
2455 if (FLSVerifyDictionary) {
2456 _dictionary->verify();
2457 }
2458 if (FLSVerifyIndexTable) {
2459 verifyIndexedFreeLists();
2460 }
2461 }
2462 }
2463 #endif
2464
2465 void CompactibleFreeListSpace::verifyIndexedFreeLists() const {
2466 size_t i = 0;
2467 for (; i < MinChunkSize; i++) {
2468 guarantee(_indexedFreeList[i].head() == NULL, "should be NULL");
2469 }
2470 for (; i < IndexSetSize; i++) {
2471 verifyIndexedFreeList(i);
2472 }
2473 }
2474
2475 void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const {
2476 FreeChunk* fc = _indexedFreeList[size].head();
2477 FreeChunk* tail = _indexedFreeList[size].tail();
2478 size_t num = _indexedFreeList[size].count();
2479 size_t n = 0;
2480 guarantee((size % 2 == 0) || fc == NULL, "Odd slots should be empty");
2481 for (; fc != NULL; fc = fc->next(), n++) {
2482 guarantee(fc->size() == size, "Size inconsistency");
2483 guarantee(fc->isFree(), "!free?");
2484 guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list");
2485 guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail");
2486 }
2487 guarantee(n == num, "Incorrect count");
2488 }
2489
2490 #ifndef PRODUCT
2491 void CompactibleFreeListSpace::checkFreeListConsistency() const {
2492 assert(_dictionary->minSize() <= IndexSetSize,
2493 "Some sizes can't be allocated without recourse to"
2494 " linear allocation buffers");
2495 assert(MIN_TREE_CHUNK_SIZE*HeapWordSize == sizeof(TreeChunk),
2496 "else MIN_TREE_CHUNK_SIZE is wrong");
2497 assert((IndexSetStride == 2 && IndexSetStart == 2) ||
2498 (IndexSetStride == 1 && IndexSetStart == 1), "just checking");
2499 assert((IndexSetStride != 2) || (MinChunkSize % 2 == 0),
2500 "Some for-loops may be incorrectly initialized");
2501 assert((IndexSetStride != 2) || (IndexSetSize % 2 == 1),
2502 "For-loops that iterate over IndexSet with stride 2 may be wrong");
2503 }
2504 #endif
2505
2506 void CompactibleFreeListSpace::printFLCensus(size_t sweep_count) const {
2507 assert_lock_strong(&_freelistLock);
2508 FreeList total;
2509 gclog_or_tty->print("end sweep# " SIZE_FORMAT "\n", sweep_count);
2510 FreeList::print_labels_on(gclog_or_tty, "size");
2511 size_t totalFree = 0;
2512 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2513 const FreeList *fl = &_indexedFreeList[i];
2514 totalFree += fl->count() * fl->size();
2515 if (i % (40*IndexSetStride) == 0) {
2516 FreeList::print_labels_on(gclog_or_tty, "size");
2517 }
2518 fl->print_on(gclog_or_tty);
2519 total.set_bfrSurp( total.bfrSurp() + fl->bfrSurp() );
2520 total.set_surplus( total.surplus() + fl->surplus() );
2521 total.set_desired( total.desired() + fl->desired() );
2522 total.set_prevSweep( total.prevSweep() + fl->prevSweep() );
2523 total.set_beforeSweep(total.beforeSweep() + fl->beforeSweep());
2524 total.set_count( total.count() + fl->count() );
2525 total.set_coalBirths( total.coalBirths() + fl->coalBirths() );
2526 total.set_coalDeaths( total.coalDeaths() + fl->coalDeaths() );
2527 total.set_splitBirths(total.splitBirths() + fl->splitBirths());
2528 total.set_splitDeaths(total.splitDeaths() + fl->splitDeaths());
2529 }
2530 total.print_on(gclog_or_tty, "TOTAL");
2531 gclog_or_tty->print_cr("Total free in indexed lists "
2532 SIZE_FORMAT " words", totalFree);
2533 gclog_or_tty->print("growth: %8.5f deficit: %8.5f\n",
2534 (double)(total.splitBirths()+total.coalBirths()-total.splitDeaths()-total.coalDeaths())/
2535 (total.prevSweep() != 0 ? (double)total.prevSweep() : 1.0),
2536 (double)(total.desired() - total.count())/(total.desired() != 0 ? (double)total.desired() : 1.0));
2537 _dictionary->printDictCensus();
2538 }
2539
2540 ///////////////////////////////////////////////////////////////////////////
2541 // CFLS_LAB
2542 ///////////////////////////////////////////////////////////////////////////
2543
2544 #define VECTOR_257(x) \
2545 /* 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 */ \
2546 { 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, \
2547 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, \
2548 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, \
2549 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, \
2550 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, \
2551 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, \
2552 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, \
2553 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, \
2554 x }
2555
2556 // Initialize with default setting of CMSParPromoteBlocksToClaim, _not_
2557 // OldPLABSize, whose static default is different; if overridden at the
2558 // command-line, this will get reinitialized via a call to
2559 // modify_initialization() below.
2560 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] =
2561 VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim));
2562 size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0);
2563 int CFLS_LAB::_global_num_workers[] = VECTOR_257(0);
2564
2565 CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) :
2566 _cfls(cfls)
2567 {
2568 assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above");
2569 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2570 i < CompactibleFreeListSpace::IndexSetSize;
2571 i += CompactibleFreeListSpace::IndexSetStride) {
2572 _indexedFreeList[i].set_size(i);
2573 _num_blocks[i] = 0;
2574 }
2575 }
2576
2577 static bool _CFLS_LAB_modified = false;
2578
2579 void CFLS_LAB::modify_initialization(size_t n, unsigned wt) {
2580 assert(!_CFLS_LAB_modified, "Call only once");
2581 _CFLS_LAB_modified = true;
2582 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2583 i < CompactibleFreeListSpace::IndexSetSize;
2584 i += CompactibleFreeListSpace::IndexSetStride) {
2585 _blocks_to_claim[i].modify(n, wt, true /* force */);
2586 }
2587 }
2588
2589 HeapWord* CFLS_LAB::alloc(size_t word_sz) {
2590 FreeChunk* res;
2591 assert(word_sz == _cfls->adjustObjectSize(word_sz), "Error");
2592 if (word_sz >= CompactibleFreeListSpace::IndexSetSize) {
2593 // This locking manages sync with other large object allocations.
2594 MutexLockerEx x(_cfls->parDictionaryAllocLock(),
2595 Mutex::_no_safepoint_check_flag);
2596 res = _cfls->getChunkFromDictionaryExact(word_sz);
2597 if (res == NULL) return NULL;
2598 } else {
2599 FreeList* fl = &_indexedFreeList[word_sz];
2600 if (fl->count() == 0) {
2601 // Attempt to refill this local free list.
2602 get_from_global_pool(word_sz, fl);
2603 // If it didn't work, give up.
2604 if (fl->count() == 0) return NULL;
2605 }
2606 res = fl->getChunkAtHead();
2607 assert(res != NULL, "Why was count non-zero?");
2608 }
2609 res->markNotFree();
2610 assert(!res->isFree(), "shouldn't be marked free");
2611 assert(oop(res)->klass_or_null() == NULL, "should look uninitialized");
2612 // mangle a just allocated object with a distinct pattern.
2613 debug_only(res->mangleAllocated(word_sz));
2614 return (HeapWord*)res;
2615 }
2616
2617 // Get a chunk of blocks of the right size and update related
2618 // book-keeping stats
2619 void CFLS_LAB::get_from_global_pool(size_t word_sz, FreeList* fl) {
2620 // Get the #blocks we want to claim
2621 size_t n_blks = (size_t)_blocks_to_claim[word_sz].average();
2622 assert(n_blks > 0, "Error");
2623 assert(ResizePLAB || n_blks == OldPLABSize, "Error");
2624 // In some cases, when the application has a phase change,
2625 // there may be a sudden and sharp shift in the object survival
2626 // profile, and updating the counts at the end of a scavenge
2627 // may not be quick enough, giving rise to large scavenge pauses
2628 // during these phase changes. It is beneficial to detect such
2629 // changes on-the-fly during a scavenge and avoid such a phase-change
2630 // pothole. The following code is a heuristic attempt to do that.
2631 // It is protected by a product flag until we have gained
2632 // enough experience with this heuristic and fine-tuned its behaviour.
2633 // WARNING: This might increase fragmentation if we overreact to
2634 // small spikes, so some kind of historical smoothing based on
2635 // previous experience with the greater reactivity might be useful.
2636 // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by
2637 // default.
2638 if (ResizeOldPLAB && CMSOldPLABResizeQuicker) {
2639 size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks);
2640 n_blks += CMSOldPLABReactivityFactor*multiple*n_blks;
2641 n_blks = MIN2(n_blks, CMSOldPLABMax);
2642 }
2643 assert(n_blks > 0, "Error");
2644 _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl);
2645 // Update stats table entry for this block size
2646 _num_blocks[word_sz] += fl->count();
2647 }
2648
2649 void CFLS_LAB::compute_desired_plab_size() {
2650 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2651 i < CompactibleFreeListSpace::IndexSetSize;
2652 i += CompactibleFreeListSpace::IndexSetStride) {
2653 assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0),
2654 "Counter inconsistency");
2655 if (_global_num_workers[i] > 0) {
2656 // Need to smooth wrt historical average
2657 if (ResizeOldPLAB) {
2658 _blocks_to_claim[i].sample(
2659 MAX2((size_t)CMSOldPLABMin,
2660 MIN2((size_t)CMSOldPLABMax,
2661 _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills))));
2662 }
2663 // Reset counters for next round
2664 _global_num_workers[i] = 0;
2665 _global_num_blocks[i] = 0;
2666 if (PrintOldPLAB) {
2667 gclog_or_tty->print_cr("[%d]: %d", i, (size_t)_blocks_to_claim[i].average());
2668 }
2669 }
2670 }
2671 }
2672
2673 void CFLS_LAB::retire(int tid) {
2674 // We run this single threaded with the world stopped;
2675 // so no need for locks and such.
2676 #define CFLS_LAB_PARALLEL_ACCESS 0
2677 NOT_PRODUCT(Thread* t = Thread::current();)
2678 assert(Thread::current()->is_VM_thread(), "Error");
2679 assert(CompactibleFreeListSpace::IndexSetStart == CompactibleFreeListSpace::IndexSetStride,
2680 "Will access to uninitialized slot below");
2681 #if CFLS_LAB_PARALLEL_ACCESS
2682 for (size_t i = CompactibleFreeListSpace::IndexSetSize - 1;
2683 i > 0;
2684 i -= CompactibleFreeListSpace::IndexSetStride) {
2685 #else // CFLS_LAB_PARALLEL_ACCESS
2686 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2687 i < CompactibleFreeListSpace::IndexSetSize;
2688 i += CompactibleFreeListSpace::IndexSetStride) {
2689 #endif // !CFLS_LAB_PARALLEL_ACCESS
2690 assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(),
2691 "Can't retire more than what we obtained");
2692 if (_num_blocks[i] > 0) {
2693 size_t num_retire = _indexedFreeList[i].count();
2694 assert(_num_blocks[i] > num_retire, "Should have used at least one");
2695 {
2696 #if CFLS_LAB_PARALLEL_ACCESS
2697 MutexLockerEx x(_cfls->_indexedFreeListParLocks[i],
2698 Mutex::_no_safepoint_check_flag);
2699 #endif // CFLS_LAB_PARALLEL_ACCESS
2700 // Update globals stats for num_blocks used
2701 _global_num_blocks[i] += (_num_blocks[i] - num_retire);
2702 _global_num_workers[i]++;
2703 assert(_global_num_workers[i] <= (ssize_t)ParallelGCThreads, "Too big");
2704 if (num_retire > 0) {
2705 _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]);
2706 // Reset this list.
2707 _indexedFreeList[i] = FreeList();
2708 _indexedFreeList[i].set_size(i);
2709 }
2710 }
2711 if (PrintOldPLAB) {
2712 gclog_or_tty->print_cr("%d[%d]: %d/%d/%d",
2713 tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average());
2714 }
2715 // Reset stats for next round
2716 _num_blocks[i] = 0;
2717 }
2718 }
2719 }
2720
2721 void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl) {
2722 assert(fl->count() == 0, "Precondition.");
2723 assert(word_sz < CompactibleFreeListSpace::IndexSetSize,
2724 "Precondition");
2725
2726 // We'll try all multiples of word_sz in the indexed set, starting with
2727 // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples,
2728 // then try getting a big chunk and splitting it.
2729 {
2730 bool found;
2731 int k;
2732 size_t cur_sz;
2733 for (k = 1, cur_sz = k * word_sz, found = false;
2734 (cur_sz < CompactibleFreeListSpace::IndexSetSize) &&
2735 (CMSSplitIndexedFreeListBlocks || k <= 1);
2736 k++, cur_sz = k * word_sz) {
2737 FreeList fl_for_cur_sz; // Empty.
2738 fl_for_cur_sz.set_size(cur_sz);
2739 {
2740 MutexLockerEx x(_indexedFreeListParLocks[cur_sz],
2741 Mutex::_no_safepoint_check_flag);
2742 FreeList* gfl = &_indexedFreeList[cur_sz];
2743 if (gfl->count() != 0) {
2744 // nn is the number of chunks of size cur_sz that
2745 // we'd need to split k-ways each, in order to create
2746 // "n" chunks of size word_sz each.
2747 const size_t nn = MAX2(n/k, (size_t)1);
2748 gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz);
2749 found = true;
2750 if (k > 1) {
2751 // Update split death stats for the cur_sz-size blocks list:
2752 // we increment the split death count by the number of blocks
2753 // we just took from the cur_sz-size blocks list and which
2754 // we will be splitting below.
2755 ssize_t deaths = gfl->splitDeaths() +
2756 fl_for_cur_sz.count();
2757 gfl->set_splitDeaths(deaths);
2758 }
2759 }
2760 }
2761 // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1.
2762 if (found) {
2763 if (k == 1) {
2764 fl->prepend(&fl_for_cur_sz);
2765 } else {
2766 // Divide each block on fl_for_cur_sz up k ways.
2767 FreeChunk* fc;
2768 while ((fc = fl_for_cur_sz.getChunkAtHead()) != NULL) {
2769 // Must do this in reverse order, so that anybody attempting to
2770 // access the main chunk sees it as a single free block until we
2771 // change it.
2772 size_t fc_size = fc->size();
2773 assert(fc->isFree(), "Error");
2774 for (int i = k-1; i >= 0; i--) {
2775 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2776 assert((i != 0) ||
2777 ((fc == ffc) && ffc->isFree() &&
2778 (ffc->size() == k*word_sz) && (fc_size == word_sz)),
2779 "Counting error");
2780 ffc->setSize(word_sz);
2781 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2782 ffc->linkNext(NULL);
2783 // Above must occur before BOT is updated below.
2784 OrderAccess::storestore();
2785 // splitting from the right, fc_size == i * word_sz
2786 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2787 fc_size -= word_sz;
2788 assert(fc_size == i*word_sz, "Error");
2789 _bt.verify_not_unallocated((HeapWord*)ffc, word_sz);
2790 _bt.verify_single_block((HeapWord*)fc, fc_size);
2791 _bt.verify_single_block((HeapWord*)ffc, word_sz);
2792 // Push this on "fl".
2793 fl->returnChunkAtHead(ffc);
2794 }
2795 // TRAP
2796 assert(fl->tail()->next() == NULL, "List invariant.");
2797 }
2798 }
2799 // Update birth stats for this block size.
2800 size_t num = fl->count();
2801 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2802 Mutex::_no_safepoint_check_flag);
2803 ssize_t births = _indexedFreeList[word_sz].splitBirths() + num;
2804 _indexedFreeList[word_sz].set_splitBirths(births);
2805 return;
2806 }
2807 }
2808 }
2809 // Otherwise, we'll split a block from the dictionary.
2810 FreeChunk* fc = NULL;
2811 FreeChunk* rem_fc = NULL;
2812 size_t rem;
2813 {
2814 MutexLockerEx x(parDictionaryAllocLock(),
2815 Mutex::_no_safepoint_check_flag);
2816 while (n > 0) {
2817 fc = dictionary()->getChunk(MAX2(n * word_sz,
2818 _dictionary->minSize()),
2819 FreeBlockDictionary::atLeast);
2820 if (fc != NULL) {
2821 _bt.allocated((HeapWord*)fc, fc->size(), true /* reducing */); // update _unallocated_blk
2822 dictionary()->dictCensusUpdate(fc->size(),
2823 true /*split*/,
2824 false /*birth*/);
2825 break;
2826 } else {
2827 n--;
2828 }
2829 }
2830 if (fc == NULL) return;
2831 // Otherwise, split up that block.
2832 assert((ssize_t)n >= 1, "Control point invariant");
2833 assert(fc->isFree(), "Error: should be a free block");
2834 _bt.verify_single_block((HeapWord*)fc, fc->size());
2835 const size_t nn = fc->size() / word_sz;
2836 n = MIN2(nn, n);
2837 assert((ssize_t)n >= 1, "Control point invariant");
2838 rem = fc->size() - n * word_sz;
2839 // If there is a remainder, and it's too small, allocate one fewer.
2840 if (rem > 0 && rem < MinChunkSize) {
2841 n--; rem += word_sz;
2842 }
2843 // Note that at this point we may have n == 0.
2844 assert((ssize_t)n >= 0, "Control point invariant");
2845
2846 // If n is 0, the chunk fc that was found is not large
2847 // enough to leave a viable remainder. We are unable to
2848 // allocate even one block. Return fc to the
2849 // dictionary and return, leaving "fl" empty.
2850 if (n == 0) {
2851 returnChunkToDictionary(fc);
2852 assert(fl->count() == 0, "We never allocated any blocks");
2853 return;
2854 }
2855
2856 // First return the remainder, if any.
2857 // Note that we hold the lock until we decide if we're going to give
2858 // back the remainder to the dictionary, since a concurrent allocation
2859 // may otherwise see the heap as empty. (We're willing to take that
2860 // hit if the block is a small block.)
2861 if (rem > 0) {
2862 size_t prefix_size = n * word_sz;
2863 rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size);
2864 rem_fc->setSize(rem);
2865 rem_fc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2866 rem_fc->linkNext(NULL);
2867 // Above must occur before BOT is updated below.
2868 assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error");
2869 OrderAccess::storestore();
2870 _bt.split_block((HeapWord*)fc, fc->size(), prefix_size);
2871 assert(fc->isFree(), "Error");
2872 fc->setSize(prefix_size);
2873 if (rem >= IndexSetSize) {
2874 returnChunkToDictionary(rem_fc);
2875 dictionary()->dictCensusUpdate(rem, true /*split*/, true /*birth*/);
2876 rem_fc = NULL;
2877 }
2878 // Otherwise, return it to the small list below.
2879 }
2880 }
2881 if (rem_fc != NULL) {
2882 MutexLockerEx x(_indexedFreeListParLocks[rem],
2883 Mutex::_no_safepoint_check_flag);
2884 _bt.verify_not_unallocated((HeapWord*)rem_fc, rem_fc->size());
2885 _indexedFreeList[rem].returnChunkAtHead(rem_fc);
2886 smallSplitBirth(rem);
2887 }
2888 assert((ssize_t)n > 0 && fc != NULL, "Consistency");
2889 // Now do the splitting up.
2890 // Must do this in reverse order, so that anybody attempting to
2891 // access the main chunk sees it as a single free block until we
2892 // change it.
2893 size_t fc_size = n * word_sz;
2894 // All but first chunk in this loop
2895 for (ssize_t i = n-1; i > 0; i--) {
2896 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2897 ffc->setSize(word_sz);
2898 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads.
2899 ffc->linkNext(NULL);
2900 // Above must occur before BOT is updated below.
2901 OrderAccess::storestore();
2902 // splitting from the right, fc_size == (n - i + 1) * wordsize
2903 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2904 fc_size -= word_sz;
2905 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size());
2906 _bt.verify_single_block((HeapWord*)ffc, ffc->size());
2907 _bt.verify_single_block((HeapWord*)fc, fc_size);
2908 // Push this on "fl".
2909 fl->returnChunkAtHead(ffc);
2910 }
2911 // First chunk
2912 assert(fc->isFree() && fc->size() == n*word_sz, "Error: should still be a free block");
2913 // The blocks above should show their new sizes before the first block below
2914 fc->setSize(word_sz);
2915 fc->linkPrev(NULL); // idempotent wrt free-ness, see assert above
2916 fc->linkNext(NULL);
2917 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
2918 _bt.verify_single_block((HeapWord*)fc, fc->size());
2919 fl->returnChunkAtHead(fc);
2920
2921 assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks");
2922 {
2923 // Update the stats for this block size.
2924 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2925 Mutex::_no_safepoint_check_flag);
2926 const ssize_t births = _indexedFreeList[word_sz].splitBirths() + n;
2927 _indexedFreeList[word_sz].set_splitBirths(births);
2928 // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n;
2929 // _indexedFreeList[word_sz].set_surplus(new_surplus);
2930 }
2931
2932 // TRAP
2933 assert(fl->tail()->next() == NULL, "List invariant.");
2934 }
2935
2936 // Set up the space's par_seq_tasks structure for work claiming
2937 // for parallel rescan. See CMSParRemarkTask where this is currently used.
2938 // XXX Need to suitably abstract and generalize this and the next
2939 // method into one.
2940 void
2941 CompactibleFreeListSpace::
2942 initialize_sequential_subtasks_for_rescan(int n_threads) {
2943 // The "size" of each task is fixed according to rescan_task_size.
2944 assert(n_threads > 0, "Unexpected n_threads argument");
2945 const size_t task_size = rescan_task_size();
2946 size_t n_tasks = (used_region().word_size() + task_size - 1)/task_size;
2947 assert((n_tasks == 0) == used_region().is_empty(), "n_tasks incorrect");
2948 assert(n_tasks == 0 ||
2949 ((used_region().start() + (n_tasks - 1)*task_size < used_region().end()) &&
2950 (used_region().start() + n_tasks*task_size >= used_region().end())),
2951 "n_tasks calculation incorrect");
2952 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2953 assert(!pst->valid(), "Clobbering existing data?");
2954 // Sets the condition for completion of the subtask (how many threads
2955 // need to finish in order to be done).
2956 pst->set_n_threads(n_threads);
2957 pst->set_n_tasks((int)n_tasks);
2958 }
2959
2960 // Set up the space's par_seq_tasks structure for work claiming
2961 // for parallel concurrent marking. See CMSConcMarkTask where this is currently used.
2962 void
2963 CompactibleFreeListSpace::
2964 initialize_sequential_subtasks_for_marking(int n_threads,
2965 HeapWord* low) {
2966 // The "size" of each task is fixed according to rescan_task_size.
2967 assert(n_threads > 0, "Unexpected n_threads argument");
2968 const size_t task_size = marking_task_size();
2969 assert(task_size > CardTableModRefBS::card_size_in_words &&
2970 (task_size % CardTableModRefBS::card_size_in_words == 0),
2971 "Otherwise arithmetic below would be incorrect");
2972 MemRegion span = _gen->reserved();
2973 if (low != NULL) {
2974 if (span.contains(low)) {
2975 // Align low down to a card boundary so that
2976 // we can use block_offset_careful() on span boundaries.
2977 HeapWord* aligned_low = (HeapWord*)align_size_down((uintptr_t)low,
2978 CardTableModRefBS::card_size);
2979 // Clip span prefix at aligned_low
2980 span = span.intersection(MemRegion(aligned_low, span.end()));
2981 } else if (low > span.end()) {
2982 span = MemRegion(low, low); // Null region
2983 } // else use entire span
2984 }
2985 assert(span.is_empty() ||
2986 ((uintptr_t)span.start() % CardTableModRefBS::card_size == 0),
2987 "span should start at a card boundary");
2988 size_t n_tasks = (span.word_size() + task_size - 1)/task_size;
2989 assert((n_tasks == 0) == span.is_empty(), "Inconsistency");
2990 assert(n_tasks == 0 ||
2991 ((span.start() + (n_tasks - 1)*task_size < span.end()) &&
2992 (span.start() + n_tasks*task_size >= span.end())),
2993 "n_tasks calculation incorrect");
2994 SequentialSubTasksDone* pst = conc_par_seq_tasks();
2995 assert(!pst->valid(), "Clobbering existing data?");
2996 // Sets the condition for completion of the subtask (how many threads
2997 // need to finish in order to be done).
2998 pst->set_n_threads(n_threads);
2999 pst->set_n_tasks((int)n_tasks);
3000 }