src/share/vm/gc_implementation/parNew/parNewGeneration.hpp

Sat, 28 Mar 2009 15:47:29 -0700

author
ysr
date
Sat, 28 Mar 2009 15:47:29 -0700
changeset 1114
cea947c8a988
parent 1014
0fbdb4381b99
child 1130
becb17ad5e51
permissions
-rw-r--r--

6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
Summary: When using compressed oops, rather than chaining the overflowed grey objects' pre-images through their klass words, we use GC-worker thread-local overflow stacks.
Reviewed-by: jcoomes, jmasa

duke@435 1 /*
xdono@1014 2 * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved.
duke@435 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435 4 *
duke@435 5 * This code is free software; you can redistribute it and/or modify it
duke@435 6 * under the terms of the GNU General Public License version 2 only, as
duke@435 7 * published by the Free Software Foundation.
duke@435 8 *
duke@435 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@435 13 * accompanied this code).
duke@435 14 *
duke@435 15 * You should have received a copy of the GNU General Public License version
duke@435 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@435 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435 18 *
duke@435 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435 20 * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435 21 * have any questions.
duke@435 22 *
duke@435 23 */
duke@435 24
duke@435 25 class ChunkArray;
duke@435 26 class ParScanWithoutBarrierClosure;
duke@435 27 class ParScanWithBarrierClosure;
duke@435 28 class ParRootScanWithoutBarrierClosure;
duke@435 29 class ParRootScanWithBarrierTwoGensClosure;
duke@435 30 class ParEvacuateFollowersClosure;
duke@435 31
duke@435 32 // It would be better if these types could be kept local to the .cpp file,
duke@435 33 // but they must be here to allow ParScanClosure::do_oop_work to be defined
duke@435 34 // in genOopClosures.inline.hpp.
duke@435 35
duke@435 36 typedef OopTaskQueue ObjToScanQueue;
duke@435 37 typedef OopTaskQueueSet ObjToScanQueueSet;
duke@435 38
duke@435 39 // Enable this to get push/pop/steal stats.
duke@435 40 const int PAR_STATS_ENABLED = 0;
duke@435 41
duke@435 42 class ParKeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
coleenp@548 43 private:
duke@435 44 ParScanWeakRefClosure* _par_cl;
coleenp@548 45 protected:
coleenp@548 46 template <class T> void do_oop_work(T* p);
duke@435 47 public:
duke@435 48 ParKeepAliveClosure(ParScanWeakRefClosure* cl);
coleenp@548 49 virtual void do_oop(oop* p);
coleenp@548 50 virtual void do_oop(narrowOop* p);
duke@435 51 };
duke@435 52
duke@435 53 // The state needed by thread performing parallel young-gen collection.
duke@435 54 class ParScanThreadState {
duke@435 55 friend class ParScanThreadStateSet;
coleenp@548 56 private:
duke@435 57 ObjToScanQueue *_work_queue;
ysr@1114 58 GrowableArray<oop>* _overflow_stack;
duke@435 59
duke@435 60 ParGCAllocBuffer _to_space_alloc_buffer;
duke@435 61
duke@435 62 ParScanWithoutBarrierClosure _to_space_closure; // scan_without_gc_barrier
duke@435 63 ParScanWithBarrierClosure _old_gen_closure; // scan_with_gc_barrier
duke@435 64 ParRootScanWithoutBarrierClosure _to_space_root_closure; // scan_root_without_gc_barrier
duke@435 65 // One of these two will be passed to process_strong_roots, which will
duke@435 66 // set its generation. The first is for two-gen configs where the
duke@435 67 // old gen collects the perm gen; the second is for arbitrary configs.
duke@435 68 // The second isn't used right now (it used to be used for the train, an
duke@435 69 // incremental collector) but the declaration has been left as a reminder.
duke@435 70 ParRootScanWithBarrierTwoGensClosure _older_gen_closure;
duke@435 71 // This closure will always be bound to the old gen; it will be used
duke@435 72 // in evacuate_followers.
duke@435 73 ParRootScanWithBarrierTwoGensClosure _old_gen_root_closure; // scan_old_root_with_gc_barrier
duke@435 74 ParEvacuateFollowersClosure _evacuate_followers;
duke@435 75 DefNewGeneration::IsAliveClosure _is_alive_closure;
duke@435 76 ParScanWeakRefClosure _scan_weak_ref_closure;
duke@435 77 ParKeepAliveClosure _keep_alive_closure;
duke@435 78
duke@435 79
duke@435 80 Space* _to_space;
duke@435 81 Space* to_space() { return _to_space; }
duke@435 82
ysr@1114 83 ParNewGeneration* _young_gen;
ysr@1114 84 ParNewGeneration* young_gen() const { return _young_gen; }
ysr@1114 85
duke@435 86 Generation* _old_gen;
duke@435 87 Generation* old_gen() { return _old_gen; }
duke@435 88
duke@435 89 HeapWord *_young_old_boundary;
duke@435 90
duke@435 91 int _hash_seed;
duke@435 92 int _thread_num;
duke@435 93 ageTable _ageTable;
duke@435 94
duke@435 95 bool _to_space_full;
duke@435 96
duke@435 97 int _pushes, _pops, _steals, _steal_attempts, _term_attempts;
duke@435 98 int _overflow_pushes, _overflow_refills, _overflow_refill_objs;
duke@435 99
duke@435 100 // Timing numbers.
duke@435 101 double _start;
duke@435 102 double _start_strong_roots;
duke@435 103 double _strong_roots_time;
duke@435 104 double _start_term;
duke@435 105 double _term_time;
duke@435 106
duke@435 107 // Helper for trim_queues. Scans subset of an array and makes
duke@435 108 // remainder available for work stealing.
duke@435 109 void scan_partial_array_and_push_remainder(oop obj);
duke@435 110
duke@435 111 // In support of CMS' parallel rescan of survivor space.
duke@435 112 ChunkArray* _survivor_chunk_array;
duke@435 113 ChunkArray* survivor_chunk_array() { return _survivor_chunk_array; }
duke@435 114
duke@435 115 void record_survivor_plab(HeapWord* plab_start, size_t plab_word_size);
duke@435 116
duke@435 117 ParScanThreadState(Space* to_space_, ParNewGeneration* gen_,
duke@435 118 Generation* old_gen_, int thread_num_,
duke@435 119 ObjToScanQueueSet* work_queue_set_, size_t desired_plab_sz_,
duke@435 120 ParallelTaskTerminator& term_);
duke@435 121
coleenp@548 122 public:
duke@435 123 ageTable* age_table() {return &_ageTable;}
duke@435 124
duke@435 125 ObjToScanQueue* work_queue() { return _work_queue; }
duke@435 126
duke@435 127 ParGCAllocBuffer* to_space_alloc_buffer() {
duke@435 128 return &_to_space_alloc_buffer;
duke@435 129 }
duke@435 130
duke@435 131 ParEvacuateFollowersClosure& evacuate_followers_closure() { return _evacuate_followers; }
duke@435 132 DefNewGeneration::IsAliveClosure& is_alive_closure() { return _is_alive_closure; }
duke@435 133 ParScanWeakRefClosure& scan_weak_ref_closure() { return _scan_weak_ref_closure; }
duke@435 134 ParKeepAliveClosure& keep_alive_closure() { return _keep_alive_closure; }
duke@435 135 ParScanClosure& older_gen_closure() { return _older_gen_closure; }
duke@435 136 ParRootScanWithoutBarrierClosure& to_space_root_closure() { return _to_space_root_closure; };
duke@435 137
duke@435 138 // Decrease queue size below "max_size".
duke@435 139 void trim_queues(int max_size);
duke@435 140
ysr@1114 141 // Private overflow stack usage
ysr@1114 142 GrowableArray<oop>* overflow_stack() { return _overflow_stack; }
ysr@1114 143 bool take_from_overflow_stack();
ysr@1114 144 void push_on_overflow_stack(oop p);
ysr@1114 145
duke@435 146 // Is new_obj a candidate for scan_partial_array_and_push_remainder method.
duke@435 147 inline bool should_be_partially_scanned(oop new_obj, oop old_obj) const;
duke@435 148
duke@435 149 int* hash_seed() { return &_hash_seed; }
duke@435 150 int thread_num() { return _thread_num; }
duke@435 151
duke@435 152 // Allocate a to-space block of size "sz", or else return NULL.
duke@435 153 HeapWord* alloc_in_to_space_slow(size_t word_sz);
duke@435 154
duke@435 155 HeapWord* alloc_in_to_space(size_t word_sz) {
duke@435 156 HeapWord* obj = to_space_alloc_buffer()->allocate(word_sz);
duke@435 157 if (obj != NULL) return obj;
duke@435 158 else return alloc_in_to_space_slow(word_sz);
duke@435 159 }
duke@435 160
duke@435 161 HeapWord* young_old_boundary() { return _young_old_boundary; }
duke@435 162
duke@435 163 void set_young_old_boundary(HeapWord *boundary) {
duke@435 164 _young_old_boundary = boundary;
duke@435 165 }
duke@435 166
duke@435 167 // Undo the most recent allocation ("obj", of "word_sz").
duke@435 168 void undo_alloc_in_to_space(HeapWord* obj, size_t word_sz);
duke@435 169
duke@435 170 int pushes() { return _pushes; }
duke@435 171 int pops() { return _pops; }
duke@435 172 int steals() { return _steals; }
duke@435 173 int steal_attempts() { return _steal_attempts; }
duke@435 174 int term_attempts() { return _term_attempts; }
duke@435 175 int overflow_pushes() { return _overflow_pushes; }
duke@435 176 int overflow_refills() { return _overflow_refills; }
duke@435 177 int overflow_refill_objs() { return _overflow_refill_objs; }
duke@435 178
duke@435 179 void note_push() { if (PAR_STATS_ENABLED) _pushes++; }
duke@435 180 void note_pop() { if (PAR_STATS_ENABLED) _pops++; }
duke@435 181 void note_steal() { if (PAR_STATS_ENABLED) _steals++; }
duke@435 182 void note_steal_attempt() { if (PAR_STATS_ENABLED) _steal_attempts++; }
duke@435 183 void note_term_attempt() { if (PAR_STATS_ENABLED) _term_attempts++; }
duke@435 184 void note_overflow_push() { if (PAR_STATS_ENABLED) _overflow_pushes++; }
duke@435 185 void note_overflow_refill(int objs) {
duke@435 186 if (PAR_STATS_ENABLED) {
duke@435 187 _overflow_refills++;
duke@435 188 _overflow_refill_objs += objs;
duke@435 189 }
duke@435 190 }
duke@435 191
duke@435 192 void start_strong_roots() {
duke@435 193 _start_strong_roots = os::elapsedTime();
duke@435 194 }
duke@435 195 void end_strong_roots() {
duke@435 196 _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
duke@435 197 }
duke@435 198 double strong_roots_time() { return _strong_roots_time; }
duke@435 199 void start_term_time() {
duke@435 200 note_term_attempt();
duke@435 201 _start_term = os::elapsedTime();
duke@435 202 }
duke@435 203 void end_term_time() {
duke@435 204 _term_time += (os::elapsedTime() - _start_term);
duke@435 205 }
duke@435 206 double term_time() { return _term_time; }
duke@435 207
duke@435 208 double elapsed() {
duke@435 209 return os::elapsedTime() - _start;
duke@435 210 }
duke@435 211 };
duke@435 212
duke@435 213 class ParNewGenTask: public AbstractGangTask {
coleenp@548 214 private:
coleenp@548 215 ParNewGeneration* _gen;
coleenp@548 216 Generation* _next_gen;
coleenp@548 217 HeapWord* _young_old_boundary;
duke@435 218 class ParScanThreadStateSet* _state_set;
duke@435 219
duke@435 220 public:
duke@435 221 ParNewGenTask(ParNewGeneration* gen,
duke@435 222 Generation* next_gen,
duke@435 223 HeapWord* young_old_boundary,
duke@435 224 ParScanThreadStateSet* state_set);
duke@435 225
duke@435 226 HeapWord* young_old_boundary() { return _young_old_boundary; }
duke@435 227
duke@435 228 void work(int i);
duke@435 229 };
duke@435 230
duke@435 231 class KeepAliveClosure: public DefNewGeneration::KeepAliveClosure {
coleenp@548 232 protected:
coleenp@548 233 template <class T> void do_oop_work(T* p);
duke@435 234 public:
duke@435 235 KeepAliveClosure(ScanWeakRefClosure* cl);
coleenp@548 236 virtual void do_oop(oop* p);
coleenp@548 237 virtual void do_oop(narrowOop* p);
duke@435 238 };
duke@435 239
duke@435 240 class EvacuateFollowersClosureGeneral: public VoidClosure {
coleenp@548 241 private:
coleenp@548 242 GenCollectedHeap* _gch;
coleenp@548 243 int _level;
coleenp@548 244 OopsInGenClosure* _scan_cur_or_nonheap;
coleenp@548 245 OopsInGenClosure* _scan_older;
coleenp@548 246 public:
coleenp@548 247 EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level,
coleenp@548 248 OopsInGenClosure* cur,
coleenp@548 249 OopsInGenClosure* older);
coleenp@548 250 virtual void do_void();
duke@435 251 };
duke@435 252
duke@435 253 // Closure for scanning ParNewGeneration.
duke@435 254 // Same as ScanClosure, except does parallel GC barrier.
duke@435 255 class ScanClosureWithParBarrier: public ScanClosure {
coleenp@548 256 protected:
coleenp@548 257 template <class T> void do_oop_work(T* p);
coleenp@548 258 public:
duke@435 259 ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier);
coleenp@548 260 virtual void do_oop(oop* p);
coleenp@548 261 virtual void do_oop(narrowOop* p);
duke@435 262 };
duke@435 263
duke@435 264 // Implements AbstractRefProcTaskExecutor for ParNew.
duke@435 265 class ParNewRefProcTaskExecutor: public AbstractRefProcTaskExecutor {
coleenp@548 266 private:
coleenp@548 267 ParNewGeneration& _generation;
coleenp@548 268 ParScanThreadStateSet& _state_set;
coleenp@548 269 public:
duke@435 270 ParNewRefProcTaskExecutor(ParNewGeneration& generation,
duke@435 271 ParScanThreadStateSet& state_set)
duke@435 272 : _generation(generation), _state_set(state_set)
duke@435 273 { }
duke@435 274
duke@435 275 // Executes a task using worker threads.
duke@435 276 virtual void execute(ProcessTask& task);
duke@435 277 virtual void execute(EnqueueTask& task);
duke@435 278 // Switch to single threaded mode.
duke@435 279 virtual void set_single_threaded_mode();
duke@435 280 };
duke@435 281
duke@435 282
duke@435 283 // A Generation that does parallel young-gen collection.
duke@435 284
duke@435 285 class ParNewGeneration: public DefNewGeneration {
duke@435 286 friend class ParNewGenTask;
duke@435 287 friend class ParNewRefProcTask;
duke@435 288 friend class ParNewRefProcTaskExecutor;
duke@435 289 friend class ParScanThreadStateSet;
ysr@969 290 friend class ParEvacuateFollowersClosure;
duke@435 291
coleenp@548 292 private:
duke@435 293 // XXX use a global constant instead of 64!
duke@435 294 struct ObjToScanQueuePadded {
duke@435 295 ObjToScanQueue work_queue;
duke@435 296 char pad[64 - sizeof(ObjToScanQueue)]; // prevent false sharing
duke@435 297 };
duke@435 298
duke@435 299 // The per-thread work queues, available here for stealing.
duke@435 300 ObjToScanQueueSet* _task_queues;
duke@435 301
duke@435 302 // Desired size of survivor space plab's
duke@435 303 PLABStats _plab_stats;
duke@435 304
duke@435 305 // A list of from-space images of to-be-scanned objects, threaded through
duke@435 306 // klass-pointers (klass information already copied to the forwarded
duke@435 307 // image.) Manipulated with CAS.
duke@435 308 oop _overflow_list;
ysr@969 309 NOT_PRODUCT(ssize_t _num_par_pushes;)
duke@435 310
duke@435 311 // If true, older generation does not support promotion undo, so avoid.
duke@435 312 static bool _avoid_promotion_undo;
duke@435 313
duke@435 314 // This closure is used by the reference processor to filter out
duke@435 315 // references to live referent.
duke@435 316 DefNewGeneration::IsAliveClosure _is_alive_closure;
duke@435 317
duke@435 318 static oop real_forwardee_slow(oop obj);
duke@435 319 static void waste_some_time();
duke@435 320
duke@435 321 // Preserve the mark of "obj", if necessary, in preparation for its mark
duke@435 322 // word being overwritten with a self-forwarding-pointer.
duke@435 323 void preserve_mark_if_necessary(oop obj, markOop m);
duke@435 324
duke@435 325 protected:
duke@435 326
duke@435 327 bool _survivor_overflow;
duke@435 328
duke@435 329 bool avoid_promotion_undo() { return _avoid_promotion_undo; }
duke@435 330 void set_avoid_promotion_undo(bool v) { _avoid_promotion_undo = v; }
duke@435 331
duke@435 332 bool survivor_overflow() { return _survivor_overflow; }
duke@435 333 void set_survivor_overflow(bool v) { _survivor_overflow = v; }
duke@435 334
duke@435 335 // Adjust the tenuring threshold. See the implementation for
duke@435 336 // the details of the policy.
duke@435 337 virtual void adjust_desired_tenuring_threshold();
duke@435 338
coleenp@548 339 public:
duke@435 340 ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level);
duke@435 341
duke@435 342 ~ParNewGeneration() {
duke@435 343 for (uint i = 0; i < ParallelGCThreads; i++)
duke@435 344 delete _task_queues->queue(i);
duke@435 345
duke@435 346 delete _task_queues;
duke@435 347 }
duke@435 348
duke@435 349 virtual void ref_processor_init();
duke@435 350 virtual Generation::Name kind() { return Generation::ParNew; }
duke@435 351 virtual const char* name() const;
duke@435 352 virtual const char* short_name() const { return "ParNew"; }
duke@435 353
duke@435 354 // override
duke@435 355 virtual bool refs_discovery_is_mt() const {
duke@435 356 assert(UseParNewGC, "ParNewGeneration only when UseParNewGC");
duke@435 357 return ParallelGCThreads > 1;
duke@435 358 }
duke@435 359
duke@435 360 // Make the collection virtual.
duke@435 361 virtual void collect(bool full,
duke@435 362 bool clear_all_soft_refs,
duke@435 363 size_t size,
duke@435 364 bool is_tlab);
duke@435 365
duke@435 366 // This needs to be visible to the closure function.
duke@435 367 // "obj" is the object to be copied, "m" is a recent value of its mark
duke@435 368 // that must not contain a forwarding pointer (though one might be
duke@435 369 // inserted in "obj"s mark word by a parallel thread).
duke@435 370 inline oop copy_to_survivor_space(ParScanThreadState* par_scan_state,
duke@435 371 oop obj, size_t obj_sz, markOop m) {
duke@435 372 if (_avoid_promotion_undo) {
duke@435 373 return copy_to_survivor_space_avoiding_promotion_undo(par_scan_state,
duke@435 374 obj, obj_sz, m);
duke@435 375 }
duke@435 376
duke@435 377 return copy_to_survivor_space_with_undo(par_scan_state, obj, obj_sz, m);
duke@435 378 }
duke@435 379
duke@435 380 oop copy_to_survivor_space_avoiding_promotion_undo(ParScanThreadState* par_scan_state,
duke@435 381 oop obj, size_t obj_sz, markOop m);
duke@435 382
duke@435 383 oop copy_to_survivor_space_with_undo(ParScanThreadState* par_scan_state,
duke@435 384 oop obj, size_t obj_sz, markOop m);
duke@435 385
ysr@969 386 // in support of testing overflow code
ysr@969 387 NOT_PRODUCT(int _overflow_counter;)
ysr@969 388 NOT_PRODUCT(bool should_simulate_overflow();)
ysr@969 389
ysr@1114 390 // Accessor for overflow list
ysr@1114 391 oop overflow_list() { return _overflow_list; }
ysr@1114 392
duke@435 393 // Push the given (from-space) object on the global overflow list.
ysr@969 394 void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state);
duke@435 395
duke@435 396 // If the global overflow list is non-empty, move some tasks from it
ysr@1114 397 // onto "work_q" (which need not be empty). No more than 1/4 of the
ysr@1114 398 // available space on "work_q" is used.
duke@435 399 bool take_from_overflow_list(ParScanThreadState* par_scan_state);
ysr@1114 400 bool take_from_overflow_list_work(ParScanThreadState* par_scan_state);
duke@435 401
duke@435 402 // The task queues to be used by parallel GC threads.
duke@435 403 ObjToScanQueueSet* task_queues() {
duke@435 404 return _task_queues;
duke@435 405 }
duke@435 406
duke@435 407 PLABStats* plab_stats() {
duke@435 408 return &_plab_stats;
duke@435 409 }
duke@435 410
duke@435 411 size_t desired_plab_sz() {
duke@435 412 return _plab_stats.desired_plab_sz();
duke@435 413 }
duke@435 414
duke@435 415 static oop real_forwardee(oop obj);
duke@435 416
duke@435 417 DEBUG_ONLY(static bool is_legal_forward_ptr(oop p);)
duke@435 418 };

mercurial