Tue, 28 Sep 2010 15:56:15 -0700
6423256: GC stacks should use a better data structure
6942771: SEGV in ParScanThreadState::take_from_overflow_stack
Reviewed-by: apetrusenko, ysr, pbk
1 /*
2 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 class ChunkArray;
26 class ParScanWithoutBarrierClosure;
27 class ParScanWithBarrierClosure;
28 class ParRootScanWithoutBarrierClosure;
29 class ParRootScanWithBarrierTwoGensClosure;
30 class ParEvacuateFollowersClosure;
32 // It would be better if these types could be kept local to the .cpp file,
33 // but they must be here to allow ParScanClosure::do_oop_work to be defined
34 // in genOopClosures.inline.hpp.
36 typedef Padded<OopTaskQueue> ObjToScanQueue;
37 typedef GenericTaskQueueSet<ObjToScanQueue> ObjToScanQueueSet;
39 class ParKeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
40 private:
41 ParScanWeakRefClosure* _par_cl;
42 protected:
43 template <class T> void do_oop_work(T* p);
44 public:
45 ParKeepAliveClosure(ParScanWeakRefClosure* cl);
46 virtual void do_oop(oop* p);
47 virtual void do_oop(narrowOop* p);
48 };
50 // The state needed by thread performing parallel young-gen collection.
51 class ParScanThreadState {
52 friend class ParScanThreadStateSet;
53 private:
54 ObjToScanQueue *_work_queue;
55 Stack<oop>* const _overflow_stack;
57 ParGCAllocBuffer _to_space_alloc_buffer;
59 ParScanWithoutBarrierClosure _to_space_closure; // scan_without_gc_barrier
60 ParScanWithBarrierClosure _old_gen_closure; // scan_with_gc_barrier
61 ParRootScanWithoutBarrierClosure _to_space_root_closure; // scan_root_without_gc_barrier
62 // One of these two will be passed to process_strong_roots, which will
63 // set its generation. The first is for two-gen configs where the
64 // old gen collects the perm gen; the second is for arbitrary configs.
65 // The second isn't used right now (it used to be used for the train, an
66 // incremental collector) but the declaration has been left as a reminder.
67 ParRootScanWithBarrierTwoGensClosure _older_gen_closure;
68 // This closure will always be bound to the old gen; it will be used
69 // in evacuate_followers.
70 ParRootScanWithBarrierTwoGensClosure _old_gen_root_closure; // scan_old_root_with_gc_barrier
71 ParEvacuateFollowersClosure _evacuate_followers;
72 DefNewGeneration::IsAliveClosure _is_alive_closure;
73 ParScanWeakRefClosure _scan_weak_ref_closure;
74 ParKeepAliveClosure _keep_alive_closure;
77 Space* _to_space;
78 Space* to_space() { return _to_space; }
80 ParNewGeneration* _young_gen;
81 ParNewGeneration* young_gen() const { return _young_gen; }
83 Generation* _old_gen;
84 Generation* old_gen() { return _old_gen; }
86 HeapWord *_young_old_boundary;
88 int _hash_seed;
89 int _thread_num;
90 ageTable _ageTable;
92 bool _to_space_full;
94 #if TASKQUEUE_STATS
95 size_t _term_attempts;
96 size_t _overflow_refills;
97 size_t _overflow_refill_objs;
98 #endif // TASKQUEUE_STATS
100 // Stats for promotion failure
101 size_t _promotion_failure_size;
103 // Timing numbers.
104 double _start;
105 double _start_strong_roots;
106 double _strong_roots_time;
107 double _start_term;
108 double _term_time;
110 // Helper for trim_queues. Scans subset of an array and makes
111 // remainder available for work stealing.
112 void scan_partial_array_and_push_remainder(oop obj);
114 // In support of CMS' parallel rescan of survivor space.
115 ChunkArray* _survivor_chunk_array;
116 ChunkArray* survivor_chunk_array() { return _survivor_chunk_array; }
118 void record_survivor_plab(HeapWord* plab_start, size_t plab_word_size);
120 ParScanThreadState(Space* to_space_, ParNewGeneration* gen_,
121 Generation* old_gen_, int thread_num_,
122 ObjToScanQueueSet* work_queue_set_,
123 Stack<oop>* overflow_stacks_,
124 size_t desired_plab_sz_,
125 ParallelTaskTerminator& term_);
127 public:
128 ageTable* age_table() {return &_ageTable;}
130 ObjToScanQueue* work_queue() { return _work_queue; }
132 ParGCAllocBuffer* to_space_alloc_buffer() {
133 return &_to_space_alloc_buffer;
134 }
136 ParEvacuateFollowersClosure& evacuate_followers_closure() { return _evacuate_followers; }
137 DefNewGeneration::IsAliveClosure& is_alive_closure() { return _is_alive_closure; }
138 ParScanWeakRefClosure& scan_weak_ref_closure() { return _scan_weak_ref_closure; }
139 ParKeepAliveClosure& keep_alive_closure() { return _keep_alive_closure; }
140 ParScanClosure& older_gen_closure() { return _older_gen_closure; }
141 ParRootScanWithoutBarrierClosure& to_space_root_closure() { return _to_space_root_closure; };
143 // Decrease queue size below "max_size".
144 void trim_queues(int max_size);
146 // Private overflow stack usage
147 Stack<oop>* overflow_stack() { return _overflow_stack; }
148 bool take_from_overflow_stack();
149 void push_on_overflow_stack(oop p);
151 // Is new_obj a candidate for scan_partial_array_and_push_remainder method.
152 inline bool should_be_partially_scanned(oop new_obj, oop old_obj) const;
154 int* hash_seed() { return &_hash_seed; }
155 int thread_num() { return _thread_num; }
157 // Allocate a to-space block of size "sz", or else return NULL.
158 HeapWord* alloc_in_to_space_slow(size_t word_sz);
160 HeapWord* alloc_in_to_space(size_t word_sz) {
161 HeapWord* obj = to_space_alloc_buffer()->allocate(word_sz);
162 if (obj != NULL) return obj;
163 else return alloc_in_to_space_slow(word_sz);
164 }
166 HeapWord* young_old_boundary() { return _young_old_boundary; }
168 void set_young_old_boundary(HeapWord *boundary) {
169 _young_old_boundary = boundary;
170 }
172 // Undo the most recent allocation ("obj", of "word_sz").
173 void undo_alloc_in_to_space(HeapWord* obj, size_t word_sz);
175 // Promotion failure stats
176 size_t promotion_failure_size() { return promotion_failure_size(); }
177 void log_promotion_failure(size_t sz) {
178 if (_promotion_failure_size == 0) {
179 _promotion_failure_size = sz;
180 }
181 }
182 void print_and_clear_promotion_failure_size();
184 #if TASKQUEUE_STATS
185 TaskQueueStats & taskqueue_stats() const { return _work_queue->stats; }
187 size_t term_attempts() const { return _term_attempts; }
188 size_t overflow_refills() const { return _overflow_refills; }
189 size_t overflow_refill_objs() const { return _overflow_refill_objs; }
191 void note_term_attempt() { ++_term_attempts; }
192 void note_overflow_refill(size_t objs) {
193 ++_overflow_refills; _overflow_refill_objs += objs;
194 }
196 void reset_stats();
197 #endif // TASKQUEUE_STATS
199 void start_strong_roots() {
200 _start_strong_roots = os::elapsedTime();
201 }
202 void end_strong_roots() {
203 _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
204 }
205 double strong_roots_time() const { return _strong_roots_time; }
206 void start_term_time() {
207 TASKQUEUE_STATS_ONLY(note_term_attempt());
208 _start_term = os::elapsedTime();
209 }
210 void end_term_time() {
211 _term_time += (os::elapsedTime() - _start_term);
212 }
213 double term_time() const { return _term_time; }
215 double elapsed_time() const {
216 return os::elapsedTime() - _start;
217 }
218 };
220 class ParNewGenTask: public AbstractGangTask {
221 private:
222 ParNewGeneration* _gen;
223 Generation* _next_gen;
224 HeapWord* _young_old_boundary;
225 class ParScanThreadStateSet* _state_set;
227 public:
228 ParNewGenTask(ParNewGeneration* gen,
229 Generation* next_gen,
230 HeapWord* young_old_boundary,
231 ParScanThreadStateSet* state_set);
233 HeapWord* young_old_boundary() { return _young_old_boundary; }
235 void work(int i);
236 };
238 class KeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
239 protected:
240 template <class T> void do_oop_work(T* p);
241 public:
242 KeepAliveClosure(ScanWeakRefClosure* cl);
243 virtual void do_oop(oop* p);
244 virtual void do_oop(narrowOop* p);
245 };
247 class EvacuateFollowersClosureGeneral: public VoidClosure {
248 private:
249 GenCollectedHeap* _gch;
250 int _level;
251 OopsInGenClosure* _scan_cur_or_nonheap;
252 OopsInGenClosure* _scan_older;
253 public:
254 EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level,
255 OopsInGenClosure* cur,
256 OopsInGenClosure* older);
257 virtual void do_void();
258 };
260 // Closure for scanning ParNewGeneration.
261 // Same as ScanClosure, except does parallel GC barrier.
262 class ScanClosureWithParBarrier: public ScanClosure {
263 protected:
264 template <class T> void do_oop_work(T* p);
265 public:
266 ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier);
267 virtual void do_oop(oop* p);
268 virtual void do_oop(narrowOop* p);
269 };
271 // Implements AbstractRefProcTaskExecutor for ParNew.
272 class ParNewRefProcTaskExecutor: public AbstractRefProcTaskExecutor {
273 private:
274 ParNewGeneration& _generation;
275 ParScanThreadStateSet& _state_set;
276 public:
277 ParNewRefProcTaskExecutor(ParNewGeneration& generation,
278 ParScanThreadStateSet& state_set)
279 : _generation(generation), _state_set(state_set)
280 { }
282 // Executes a task using worker threads.
283 virtual void execute(ProcessTask& task);
284 virtual void execute(EnqueueTask& task);
285 // Switch to single threaded mode.
286 virtual void set_single_threaded_mode();
287 };
290 // A Generation that does parallel young-gen collection.
292 class ParNewGeneration: public DefNewGeneration {
293 friend class ParNewGenTask;
294 friend class ParNewRefProcTask;
295 friend class ParNewRefProcTaskExecutor;
296 friend class ParScanThreadStateSet;
297 friend class ParEvacuateFollowersClosure;
299 private:
300 // The per-worker-thread work queues
301 ObjToScanQueueSet* _task_queues;
303 // Per-worker-thread local overflow stacks
304 Stack<oop>* _overflow_stacks;
306 // Desired size of survivor space plab's
307 PLABStats _plab_stats;
309 // A list of from-space images of to-be-scanned objects, threaded through
310 // klass-pointers (klass information already copied to the forwarded
311 // image.) Manipulated with CAS.
312 oop _overflow_list;
313 NOT_PRODUCT(ssize_t _num_par_pushes;)
315 // If true, older generation does not support promotion undo, so avoid.
316 static bool _avoid_promotion_undo;
318 // This closure is used by the reference processor to filter out
319 // references to live referent.
320 DefNewGeneration::IsAliveClosure _is_alive_closure;
322 static oop real_forwardee_slow(oop obj);
323 static void waste_some_time();
325 // Preserve the mark of "obj", if necessary, in preparation for its mark
326 // word being overwritten with a self-forwarding-pointer.
327 void preserve_mark_if_necessary(oop obj, markOop m);
329 protected:
331 bool _survivor_overflow;
333 bool avoid_promotion_undo() { return _avoid_promotion_undo; }
334 void set_avoid_promotion_undo(bool v) { _avoid_promotion_undo = v; }
336 bool survivor_overflow() { return _survivor_overflow; }
337 void set_survivor_overflow(bool v) { _survivor_overflow = v; }
339 // Adjust the tenuring threshold. See the implementation for
340 // the details of the policy.
341 virtual void adjust_desired_tenuring_threshold();
343 public:
344 ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level);
346 ~ParNewGeneration() {
347 for (uint i = 0; i < ParallelGCThreads; i++)
348 delete _task_queues->queue(i);
350 delete _task_queues;
351 }
353 static bool in_use();
355 virtual void ref_processor_init();
356 virtual Generation::Name kind() { return Generation::ParNew; }
357 virtual const char* name() const;
358 virtual const char* short_name() const { return "ParNew"; }
360 // override
361 virtual bool refs_discovery_is_mt() const {
362 assert(UseParNewGC, "ParNewGeneration only when UseParNewGC");
363 return ParallelGCThreads > 1;
364 }
366 // Make the collection virtual.
367 virtual void collect(bool full,
368 bool clear_all_soft_refs,
369 size_t size,
370 bool is_tlab);
372 // This needs to be visible to the closure function.
373 // "obj" is the object to be copied, "m" is a recent value of its mark
374 // that must not contain a forwarding pointer (though one might be
375 // inserted in "obj"s mark word by a parallel thread).
376 inline oop copy_to_survivor_space(ParScanThreadState* par_scan_state,
377 oop obj, size_t obj_sz, markOop m) {
378 if (_avoid_promotion_undo) {
379 return copy_to_survivor_space_avoiding_promotion_undo(par_scan_state,
380 obj, obj_sz, m);
381 }
383 return copy_to_survivor_space_with_undo(par_scan_state, obj, obj_sz, m);
384 }
386 oop copy_to_survivor_space_avoiding_promotion_undo(ParScanThreadState* par_scan_state,
387 oop obj, size_t obj_sz, markOop m);
389 oop copy_to_survivor_space_with_undo(ParScanThreadState* par_scan_state,
390 oop obj, size_t obj_sz, markOop m);
392 // in support of testing overflow code
393 NOT_PRODUCT(int _overflow_counter;)
394 NOT_PRODUCT(bool should_simulate_overflow();)
396 // Accessor for overflow list
397 oop overflow_list() { return _overflow_list; }
399 // Push the given (from-space) object on the global overflow list.
400 void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state);
402 // If the global overflow list is non-empty, move some tasks from it
403 // onto "work_q" (which need not be empty). No more than 1/4 of the
404 // available space on "work_q" is used.
405 bool take_from_overflow_list(ParScanThreadState* par_scan_state);
406 bool take_from_overflow_list_work(ParScanThreadState* par_scan_state);
408 // The task queues to be used by parallel GC threads.
409 ObjToScanQueueSet* task_queues() {
410 return _task_queues;
411 }
413 PLABStats* plab_stats() {
414 return &_plab_stats;
415 }
417 size_t desired_plab_sz() {
418 return _plab_stats.desired_plab_sz();
419 }
421 static oop real_forwardee(oop obj);
423 DEBUG_ONLY(static bool is_legal_forward_ptr(oop p);)
424 };