Fri, 16 Jul 2010 21:33:21 -0700
6962947: shared TaskQueue statistics
Reviewed-by: tonyp, ysr
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 // Enable this to get push/pop/steal stats.
40 const int PAR_STATS_ENABLED = 0;
42 class ParKeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
43 private:
44 ParScanWeakRefClosure* _par_cl;
45 protected:
46 template <class T> void do_oop_work(T* p);
47 public:
48 ParKeepAliveClosure(ParScanWeakRefClosure* cl);
49 virtual void do_oop(oop* p);
50 virtual void do_oop(narrowOop* p);
51 };
53 // The state needed by thread performing parallel young-gen collection.
54 class ParScanThreadState {
55 friend class ParScanThreadStateSet;
56 private:
57 ObjToScanQueue *_work_queue;
58 GrowableArray<oop>* _overflow_stack;
60 ParGCAllocBuffer _to_space_alloc_buffer;
62 ParScanWithoutBarrierClosure _to_space_closure; // scan_without_gc_barrier
63 ParScanWithBarrierClosure _old_gen_closure; // scan_with_gc_barrier
64 ParRootScanWithoutBarrierClosure _to_space_root_closure; // scan_root_without_gc_barrier
65 // One of these two will be passed to process_strong_roots, which will
66 // set its generation. The first is for two-gen configs where the
67 // old gen collects the perm gen; the second is for arbitrary configs.
68 // The second isn't used right now (it used to be used for the train, an
69 // incremental collector) but the declaration has been left as a reminder.
70 ParRootScanWithBarrierTwoGensClosure _older_gen_closure;
71 // This closure will always be bound to the old gen; it will be used
72 // in evacuate_followers.
73 ParRootScanWithBarrierTwoGensClosure _old_gen_root_closure; // scan_old_root_with_gc_barrier
74 ParEvacuateFollowersClosure _evacuate_followers;
75 DefNewGeneration::IsAliveClosure _is_alive_closure;
76 ParScanWeakRefClosure _scan_weak_ref_closure;
77 ParKeepAliveClosure _keep_alive_closure;
80 Space* _to_space;
81 Space* to_space() { return _to_space; }
83 ParNewGeneration* _young_gen;
84 ParNewGeneration* young_gen() const { return _young_gen; }
86 Generation* _old_gen;
87 Generation* old_gen() { return _old_gen; }
89 HeapWord *_young_old_boundary;
91 int _hash_seed;
92 int _thread_num;
93 ageTable _ageTable;
95 bool _to_space_full;
97 int _pushes, _pops, _steals, _steal_attempts, _term_attempts;
98 int _overflow_pushes, _overflow_refills, _overflow_refill_objs;
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 GrowableArray<oop>** overflow_stack_set_,
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 GrowableArray<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 int pushes() { return _pushes; }
185 int pops() { return _pops; }
186 int steals() { return _steals; }
187 int steal_attempts() { return _steal_attempts; }
188 int term_attempts() { return _term_attempts; }
189 int overflow_pushes() { return _overflow_pushes; }
190 int overflow_refills() { return _overflow_refills; }
191 int overflow_refill_objs() { return _overflow_refill_objs; }
193 void note_push() { if (PAR_STATS_ENABLED) _pushes++; }
194 void note_pop() { if (PAR_STATS_ENABLED) _pops++; }
195 void note_steal() { if (PAR_STATS_ENABLED) _steals++; }
196 void note_steal_attempt() { if (PAR_STATS_ENABLED) _steal_attempts++; }
197 void note_term_attempt() { if (PAR_STATS_ENABLED) _term_attempts++; }
198 void note_overflow_push() { if (PAR_STATS_ENABLED) _overflow_pushes++; }
199 void note_overflow_refill(int objs) {
200 if (PAR_STATS_ENABLED) {
201 _overflow_refills++;
202 _overflow_refill_objs += objs;
203 }
204 }
206 void start_strong_roots() {
207 _start_strong_roots = os::elapsedTime();
208 }
209 void end_strong_roots() {
210 _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
211 }
212 double strong_roots_time() { return _strong_roots_time; }
213 void start_term_time() {
214 note_term_attempt();
215 _start_term = os::elapsedTime();
216 }
217 void end_term_time() {
218 _term_time += (os::elapsedTime() - _start_term);
219 }
220 double term_time() { return _term_time; }
222 double elapsed() {
223 return os::elapsedTime() - _start;
224 }
225 };
227 class ParNewGenTask: public AbstractGangTask {
228 private:
229 ParNewGeneration* _gen;
230 Generation* _next_gen;
231 HeapWord* _young_old_boundary;
232 class ParScanThreadStateSet* _state_set;
234 public:
235 ParNewGenTask(ParNewGeneration* gen,
236 Generation* next_gen,
237 HeapWord* young_old_boundary,
238 ParScanThreadStateSet* state_set);
240 HeapWord* young_old_boundary() { return _young_old_boundary; }
242 void work(int i);
243 };
245 class KeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
246 protected:
247 template <class T> void do_oop_work(T* p);
248 public:
249 KeepAliveClosure(ScanWeakRefClosure* cl);
250 virtual void do_oop(oop* p);
251 virtual void do_oop(narrowOop* p);
252 };
254 class EvacuateFollowersClosureGeneral: public VoidClosure {
255 private:
256 GenCollectedHeap* _gch;
257 int _level;
258 OopsInGenClosure* _scan_cur_or_nonheap;
259 OopsInGenClosure* _scan_older;
260 public:
261 EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level,
262 OopsInGenClosure* cur,
263 OopsInGenClosure* older);
264 virtual void do_void();
265 };
267 // Closure for scanning ParNewGeneration.
268 // Same as ScanClosure, except does parallel GC barrier.
269 class ScanClosureWithParBarrier: public ScanClosure {
270 protected:
271 template <class T> void do_oop_work(T* p);
272 public:
273 ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier);
274 virtual void do_oop(oop* p);
275 virtual void do_oop(narrowOop* p);
276 };
278 // Implements AbstractRefProcTaskExecutor for ParNew.
279 class ParNewRefProcTaskExecutor: public AbstractRefProcTaskExecutor {
280 private:
281 ParNewGeneration& _generation;
282 ParScanThreadStateSet& _state_set;
283 public:
284 ParNewRefProcTaskExecutor(ParNewGeneration& generation,
285 ParScanThreadStateSet& state_set)
286 : _generation(generation), _state_set(state_set)
287 { }
289 // Executes a task using worker threads.
290 virtual void execute(ProcessTask& task);
291 virtual void execute(EnqueueTask& task);
292 // Switch to single threaded mode.
293 virtual void set_single_threaded_mode();
294 };
297 // A Generation that does parallel young-gen collection.
299 class ParNewGeneration: public DefNewGeneration {
300 friend class ParNewGenTask;
301 friend class ParNewRefProcTask;
302 friend class ParNewRefProcTaskExecutor;
303 friend class ParScanThreadStateSet;
304 friend class ParEvacuateFollowersClosure;
306 private:
307 // The per-worker-thread work queues
308 ObjToScanQueueSet* _task_queues;
310 // Per-worker-thread local overflow stacks
311 GrowableArray<oop>** _overflow_stacks;
313 // Desired size of survivor space plab's
314 PLABStats _plab_stats;
316 // A list of from-space images of to-be-scanned objects, threaded through
317 // klass-pointers (klass information already copied to the forwarded
318 // image.) Manipulated with CAS.
319 oop _overflow_list;
320 NOT_PRODUCT(ssize_t _num_par_pushes;)
322 // If true, older generation does not support promotion undo, so avoid.
323 static bool _avoid_promotion_undo;
325 // This closure is used by the reference processor to filter out
326 // references to live referent.
327 DefNewGeneration::IsAliveClosure _is_alive_closure;
329 static oop real_forwardee_slow(oop obj);
330 static void waste_some_time();
332 // Preserve the mark of "obj", if necessary, in preparation for its mark
333 // word being overwritten with a self-forwarding-pointer.
334 void preserve_mark_if_necessary(oop obj, markOop m);
336 protected:
338 bool _survivor_overflow;
340 bool avoid_promotion_undo() { return _avoid_promotion_undo; }
341 void set_avoid_promotion_undo(bool v) { _avoid_promotion_undo = v; }
343 bool survivor_overflow() { return _survivor_overflow; }
344 void set_survivor_overflow(bool v) { _survivor_overflow = v; }
346 // Adjust the tenuring threshold. See the implementation for
347 // the details of the policy.
348 virtual void adjust_desired_tenuring_threshold();
350 public:
351 ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level);
353 ~ParNewGeneration() {
354 for (uint i = 0; i < ParallelGCThreads; i++)
355 delete _task_queues->queue(i);
357 delete _task_queues;
358 }
360 virtual void ref_processor_init();
361 virtual Generation::Name kind() { return Generation::ParNew; }
362 virtual const char* name() const;
363 virtual const char* short_name() const { return "ParNew"; }
365 // override
366 virtual bool refs_discovery_is_mt() const {
367 assert(UseParNewGC, "ParNewGeneration only when UseParNewGC");
368 return ParallelGCThreads > 1;
369 }
371 // Make the collection virtual.
372 virtual void collect(bool full,
373 bool clear_all_soft_refs,
374 size_t size,
375 bool is_tlab);
377 // This needs to be visible to the closure function.
378 // "obj" is the object to be copied, "m" is a recent value of its mark
379 // that must not contain a forwarding pointer (though one might be
380 // inserted in "obj"s mark word by a parallel thread).
381 inline oop copy_to_survivor_space(ParScanThreadState* par_scan_state,
382 oop obj, size_t obj_sz, markOop m) {
383 if (_avoid_promotion_undo) {
384 return copy_to_survivor_space_avoiding_promotion_undo(par_scan_state,
385 obj, obj_sz, m);
386 }
388 return copy_to_survivor_space_with_undo(par_scan_state, obj, obj_sz, m);
389 }
391 oop copy_to_survivor_space_avoiding_promotion_undo(ParScanThreadState* par_scan_state,
392 oop obj, size_t obj_sz, markOop m);
394 oop copy_to_survivor_space_with_undo(ParScanThreadState* par_scan_state,
395 oop obj, size_t obj_sz, markOop m);
397 // in support of testing overflow code
398 NOT_PRODUCT(int _overflow_counter;)
399 NOT_PRODUCT(bool should_simulate_overflow();)
401 // Accessor for overflow list
402 oop overflow_list() { return _overflow_list; }
404 // Push the given (from-space) object on the global overflow list.
405 void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state);
407 // If the global overflow list is non-empty, move some tasks from it
408 // onto "work_q" (which need not be empty). No more than 1/4 of the
409 // available space on "work_q" is used.
410 bool take_from_overflow_list(ParScanThreadState* par_scan_state);
411 bool take_from_overflow_list_work(ParScanThreadState* par_scan_state);
413 // The task queues to be used by parallel GC threads.
414 ObjToScanQueueSet* task_queues() {
415 return _task_queues;
416 }
418 PLABStats* plab_stats() {
419 return &_plab_stats;
420 }
422 size_t desired_plab_sz() {
423 return _plab_stats.desired_plab_sz();
424 }
426 static oop real_forwardee(oop obj);
428 DEBUG_ONLY(static bool is_legal_forward_ptr(oop p);)
429 };