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