Mon, 26 Jan 2009 12:47:21 -0800
6786503: Overflow list performance can be improved
Summary: Avoid overflow list walk in CMS & ParNew when it is unnecessary. Fix a couple of correctness issues, including a C-heap leak, in ParNew at the intersection of promotion failure, work queue overflow and object array chunking. Add stress testing option and related assertion checking.
Reviewed-by: jmasa
1 /*
2 * Copyright 2001-2008 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_genMarkSweep.cpp.incl"
28 void GenMarkSweep::invoke_at_safepoint(int level, ReferenceProcessor* rp,
29 bool clear_all_softrefs) {
30 assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint");
32 // hook up weak ref data so it can be used during Mark-Sweep
33 assert(ref_processor() == NULL, "no stomping");
34 assert(rp != NULL, "should be non-NULL");
35 _ref_processor = rp;
36 rp->setup_policy(clear_all_softrefs);
38 TraceTime t1("Full GC", PrintGC && !PrintGCDetails, true, gclog_or_tty);
40 // When collecting the permanent generation methodOops may be moving,
41 // so we either have to flush all bcp data or convert it into bci.
42 CodeCache::gc_prologue();
43 Threads::gc_prologue();
45 // Increment the invocation count for the permanent generation, since it is
46 // implicitly collected whenever we do a full mark sweep collection.
47 GenCollectedHeap* gch = GenCollectedHeap::heap();
48 gch->perm_gen()->stat_record()->invocations++;
50 // Capture heap size before collection for printing.
51 size_t gch_prev_used = gch->used();
53 // Some of the card table updates below assume that the perm gen is
54 // also being collected.
55 assert(level == gch->n_gens() - 1,
56 "All generations are being collected, ergo perm gen too.");
58 // Capture used regions for each generation that will be
59 // subject to collection, so that card table adjustments can
60 // be made intelligently (see clear / invalidate further below).
61 gch->save_used_regions(level, true /* perm */);
63 allocate_stacks();
65 mark_sweep_phase1(level, clear_all_softrefs);
67 mark_sweep_phase2();
69 // Don't add any more derived pointers during phase3
70 COMPILER2_PRESENT(assert(DerivedPointerTable::is_active(), "Sanity"));
71 COMPILER2_PRESENT(DerivedPointerTable::set_active(false));
73 mark_sweep_phase3(level);
75 VALIDATE_MARK_SWEEP_ONLY(
76 if (ValidateMarkSweep) {
77 guarantee(_root_refs_stack->length() == 0, "should be empty by now");
78 }
79 )
81 mark_sweep_phase4();
83 VALIDATE_MARK_SWEEP_ONLY(
84 if (ValidateMarkSweep) {
85 guarantee(_live_oops->length() == _live_oops_moved_to->length(),
86 "should be the same size");
87 }
88 )
90 restore_marks();
92 // Set saved marks for allocation profiler (and other things? -- dld)
93 // (Should this be in general part?)
94 gch->save_marks();
96 deallocate_stacks();
98 // If compaction completely evacuated all generations younger than this
99 // one, then we can clear the card table. Otherwise, we must invalidate
100 // it (consider all cards dirty). In the future, we might consider doing
101 // compaction within generations only, and doing card-table sliding.
102 bool all_empty = true;
103 for (int i = 0; all_empty && i < level; i++) {
104 Generation* g = gch->get_gen(i);
105 all_empty = all_empty && gch->get_gen(i)->used() == 0;
106 }
107 GenRemSet* rs = gch->rem_set();
108 // Clear/invalidate below make use of the "prev_used_regions" saved earlier.
109 if (all_empty) {
110 // We've evacuated all generations below us.
111 Generation* g = gch->get_gen(level);
112 rs->clear_into_younger(g, true /* perm */);
113 } else {
114 // Invalidate the cards corresponding to the currently used
115 // region and clear those corresponding to the evacuated region
116 // of all generations just collected (i.e. level and younger).
117 rs->invalidate_or_clear(gch->get_gen(level),
118 true /* younger */,
119 true /* perm */);
120 }
122 Threads::gc_epilogue();
123 CodeCache::gc_epilogue();
125 if (PrintGC && !PrintGCDetails) {
126 gch->print_heap_change(gch_prev_used);
127 }
129 // refs processing: clean slate
130 _ref_processor = NULL;
132 // Update heap occupancy information which is used as
133 // input to soft ref clearing policy at the next gc.
134 Universe::update_heap_info_at_gc();
136 // Update time of last gc for all generations we collected
137 // (which curently is all the generations in the heap).
138 gch->update_time_of_last_gc(os::javaTimeMillis());
139 }
141 void GenMarkSweep::allocate_stacks() {
142 GenCollectedHeap* gch = GenCollectedHeap::heap();
143 // Scratch request on behalf of oldest generation; will do no
144 // allocation.
145 ScratchBlock* scratch = gch->gather_scratch(gch->_gens[gch->_n_gens-1], 0);
147 // $$$ To cut a corner, we'll only use the first scratch block, and then
148 // revert to malloc.
149 if (scratch != NULL) {
150 _preserved_count_max =
151 scratch->num_words * HeapWordSize / sizeof(PreservedMark);
152 } else {
153 _preserved_count_max = 0;
154 }
156 _preserved_marks = (PreservedMark*)scratch;
157 _preserved_count = 0;
158 _preserved_mark_stack = NULL;
159 _preserved_oop_stack = NULL;
161 _marking_stack = new (ResourceObj::C_HEAP) GrowableArray<oop>(4000, true);
163 int size = SystemDictionary::number_of_classes() * 2;
164 _revisit_klass_stack = new (ResourceObj::C_HEAP) GrowableArray<Klass*>(size, true);
166 #ifdef VALIDATE_MARK_SWEEP
167 if (ValidateMarkSweep) {
168 _root_refs_stack = new (ResourceObj::C_HEAP) GrowableArray<void*>(100, true);
169 _other_refs_stack = new (ResourceObj::C_HEAP) GrowableArray<void*>(100, true);
170 _adjusted_pointers = new (ResourceObj::C_HEAP) GrowableArray<void*>(100, true);
171 _live_oops = new (ResourceObj::C_HEAP) GrowableArray<oop>(100, true);
172 _live_oops_moved_to = new (ResourceObj::C_HEAP) GrowableArray<oop>(100, true);
173 _live_oops_size = new (ResourceObj::C_HEAP) GrowableArray<size_t>(100, true);
174 }
175 if (RecordMarkSweepCompaction) {
176 if (_cur_gc_live_oops == NULL) {
177 _cur_gc_live_oops = new(ResourceObj::C_HEAP) GrowableArray<HeapWord*>(100, true);
178 _cur_gc_live_oops_moved_to = new(ResourceObj::C_HEAP) GrowableArray<HeapWord*>(100, true);
179 _cur_gc_live_oops_size = new(ResourceObj::C_HEAP) GrowableArray<size_t>(100, true);
180 _last_gc_live_oops = new(ResourceObj::C_HEAP) GrowableArray<HeapWord*>(100, true);
181 _last_gc_live_oops_moved_to = new(ResourceObj::C_HEAP) GrowableArray<HeapWord*>(100, true);
182 _last_gc_live_oops_size = new(ResourceObj::C_HEAP) GrowableArray<size_t>(100, true);
183 } else {
184 _cur_gc_live_oops->clear();
185 _cur_gc_live_oops_moved_to->clear();
186 _cur_gc_live_oops_size->clear();
187 }
188 }
189 #endif
190 }
193 void GenMarkSweep::deallocate_stacks() {
195 if (!UseG1GC) {
196 GenCollectedHeap* gch = GenCollectedHeap::heap();
197 gch->release_scratch();
198 }
200 if (_preserved_oop_stack) {
201 delete _preserved_mark_stack;
202 _preserved_mark_stack = NULL;
203 delete _preserved_oop_stack;
204 _preserved_oop_stack = NULL;
205 }
207 delete _marking_stack;
208 delete _revisit_klass_stack;
210 #ifdef VALIDATE_MARK_SWEEP
211 if (ValidateMarkSweep) {
212 delete _root_refs_stack;
213 delete _other_refs_stack;
214 delete _adjusted_pointers;
215 delete _live_oops;
216 delete _live_oops_size;
217 delete _live_oops_moved_to;
218 _live_oops_index = 0;
219 _live_oops_index_at_perm = 0;
220 }
221 #endif
222 }
224 void GenMarkSweep::mark_sweep_phase1(int level,
225 bool clear_all_softrefs) {
226 // Recursively traverse all live objects and mark them
227 EventMark m("1 mark object");
228 TraceTime tm("phase 1", PrintGC && Verbose, true, gclog_or_tty);
229 trace(" 1");
231 VALIDATE_MARK_SWEEP_ONLY(reset_live_oop_tracking(false));
233 GenCollectedHeap* gch = GenCollectedHeap::heap();
235 // Because follow_root_closure is created statically, cannot
236 // use OopsInGenClosure constructor which takes a generation,
237 // as the Universe has not been created when the static constructors
238 // are run.
239 follow_root_closure.set_orig_generation(gch->get_gen(level));
241 gch->gen_process_strong_roots(level,
242 false, // Younger gens are not roots.
243 true, // Collecting permanent generation.
244 SharedHeap::SO_SystemClasses,
245 &follow_root_closure, &follow_root_closure);
247 // Process reference objects found during marking
248 {
249 ref_processor()->setup_policy(clear_all_softrefs);
250 ref_processor()->process_discovered_references(
251 &is_alive, &keep_alive, &follow_stack_closure, NULL);
252 }
254 // Follow system dictionary roots and unload classes
255 bool purged_class = SystemDictionary::do_unloading(&is_alive);
257 // Follow code cache roots
258 CodeCache::do_unloading(&is_alive, &keep_alive, purged_class);
259 follow_stack(); // Flush marking stack
261 // Update subklass/sibling/implementor links of live klasses
262 follow_weak_klass_links();
263 assert(_marking_stack->is_empty(), "just drained");
265 // Visit symbol and interned string tables and delete unmarked oops
266 SymbolTable::unlink(&is_alive);
267 StringTable::unlink(&is_alive);
269 assert(_marking_stack->is_empty(), "stack should be empty by now");
270 }
273 void GenMarkSweep::mark_sweep_phase2() {
274 // Now all live objects are marked, compute the new object addresses.
276 // It is imperative that we traverse perm_gen LAST. If dead space is
277 // allowed a range of dead object may get overwritten by a dead int
278 // array. If perm_gen is not traversed last a klassOop may get
279 // overwritten. This is fine since it is dead, but if the class has dead
280 // instances we have to skip them, and in order to find their size we
281 // need the klassOop!
282 //
283 // It is not required that we traverse spaces in the same order in
284 // phase2, phase3 and phase4, but the ValidateMarkSweep live oops
285 // tracking expects us to do so. See comment under phase4.
287 GenCollectedHeap* gch = GenCollectedHeap::heap();
288 Generation* pg = gch->perm_gen();
290 EventMark m("2 compute new addresses");
291 TraceTime tm("phase 2", PrintGC && Verbose, true, gclog_or_tty);
292 trace("2");
294 VALIDATE_MARK_SWEEP_ONLY(reset_live_oop_tracking(false));
296 gch->prepare_for_compaction();
298 VALIDATE_MARK_SWEEP_ONLY(_live_oops_index_at_perm = _live_oops_index);
299 CompactPoint perm_cp(pg, NULL, NULL);
300 pg->prepare_for_compaction(&perm_cp);
301 }
303 class GenAdjustPointersClosure: public GenCollectedHeap::GenClosure {
304 public:
305 void do_generation(Generation* gen) {
306 gen->adjust_pointers();
307 }
308 };
310 void GenMarkSweep::mark_sweep_phase3(int level) {
311 GenCollectedHeap* gch = GenCollectedHeap::heap();
312 Generation* pg = gch->perm_gen();
314 // Adjust the pointers to reflect the new locations
315 EventMark m("3 adjust pointers");
316 TraceTime tm("phase 3", PrintGC && Verbose, true, gclog_or_tty);
317 trace("3");
319 VALIDATE_MARK_SWEEP_ONLY(reset_live_oop_tracking(false));
321 // Needs to be done before the system dictionary is adjusted.
322 pg->pre_adjust_pointers();
324 // Because the two closures below are created statically, cannot
325 // use OopsInGenClosure constructor which takes a generation,
326 // as the Universe has not been created when the static constructors
327 // are run.
328 adjust_root_pointer_closure.set_orig_generation(gch->get_gen(level));
329 adjust_pointer_closure.set_orig_generation(gch->get_gen(level));
331 gch->gen_process_strong_roots(level,
332 false, // Younger gens are not roots.
333 true, // Collecting permanent generation.
334 SharedHeap::SO_AllClasses,
335 &adjust_root_pointer_closure,
336 &adjust_root_pointer_closure);
338 // Now adjust pointers in remaining weak roots. (All of which should
339 // have been cleared if they pointed to non-surviving objects.)
340 gch->gen_process_weak_roots(&adjust_root_pointer_closure,
341 &adjust_pointer_closure);
343 adjust_marks();
344 GenAdjustPointersClosure blk;
345 gch->generation_iterate(&blk, true);
346 pg->adjust_pointers();
347 }
349 class GenCompactClosure: public GenCollectedHeap::GenClosure {
350 public:
351 void do_generation(Generation* gen) {
352 gen->compact();
353 }
354 };
356 void GenMarkSweep::mark_sweep_phase4() {
357 // All pointers are now adjusted, move objects accordingly
359 // It is imperative that we traverse perm_gen first in phase4. All
360 // classes must be allocated earlier than their instances, and traversing
361 // perm_gen first makes sure that all klassOops have moved to their new
362 // location before any instance does a dispatch through it's klass!
364 // The ValidateMarkSweep live oops tracking expects us to traverse spaces
365 // in the same order in phase2, phase3 and phase4. We don't quite do that
366 // here (perm_gen first rather than last), so we tell the validate code
367 // to use a higher index (saved from phase2) when verifying perm_gen.
368 GenCollectedHeap* gch = GenCollectedHeap::heap();
369 Generation* pg = gch->perm_gen();
371 EventMark m("4 compact heap");
372 TraceTime tm("phase 4", PrintGC && Verbose, true, gclog_or_tty);
373 trace("4");
375 VALIDATE_MARK_SWEEP_ONLY(reset_live_oop_tracking(true));
377 pg->compact();
379 VALIDATE_MARK_SWEEP_ONLY(reset_live_oop_tracking(false));
381 GenCompactClosure blk;
382 gch->generation_iterate(&blk, true);
384 VALIDATE_MARK_SWEEP_ONLY(compaction_complete());
386 pg->post_compact(); // Shared spaces verification.
387 }