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 */ 24 25 #include "incls/_precompiled.incl" 26 #include "incls/_g1CollectorPolicy.cpp.incl" 27 28 #define PREDICTIONS_VERBOSE 0 29 30 // <NEW PREDICTION> 31 32 // Different defaults for different number of GC threads 33 // They were chosen by running GCOld and SPECjbb on debris with different 34 // numbers of GC threads and choosing them based on the results 35 36 // all the same 37 static double rs_length_diff_defaults[] = { 38 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 39 }; 40 41 static double cost_per_card_ms_defaults[] = { 42 0.01, 0.005, 0.005, 0.003, 0.003, 0.002, 0.002, 0.0015 43 }; 44 45 // all the same 46 static double fully_young_cards_per_entry_ratio_defaults[] = { 47 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 48 }; 49 50 static double cost_per_entry_ms_defaults[] = { 51 0.015, 0.01, 0.01, 0.008, 0.008, 0.0055, 0.0055, 0.005 52 }; 53 54 static double cost_per_byte_ms_defaults[] = { 55 0.00006, 0.00003, 0.00003, 0.000015, 0.000015, 0.00001, 0.00001, 0.000009 56 }; 57 58 // these should be pretty consistent 59 static double constant_other_time_ms_defaults[] = { 60 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0 61 }; 62 63 64 static double young_other_cost_per_region_ms_defaults[] = { 65 0.3, 0.2, 0.2, 0.15, 0.15, 0.12, 0.12, 0.1 66 }; 67 68 static double non_young_other_cost_per_region_ms_defaults[] = { 69 1.0, 0.7, 0.7, 0.5, 0.5, 0.42, 0.42, 0.30 70 }; 71 72 // </NEW PREDICTION> 73 74 G1CollectorPolicy::G1CollectorPolicy() : 75 _parallel_gc_threads((ParallelGCThreads > 0) ? ParallelGCThreads : 1), 76 _n_pauses(0), 77 _recent_CH_strong_roots_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 78 _recent_G1_strong_roots_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 79 _recent_evac_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 80 _recent_pause_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 81 _recent_rs_sizes(new TruncatedSeq(NumPrevPausesForHeuristics)), 82 _recent_gc_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 83 _all_pause_times_ms(new NumberSeq()), 84 _stop_world_start(0.0), 85 _all_stop_world_times_ms(new NumberSeq()), 86 _all_yield_times_ms(new NumberSeq()), 87 88 _all_mod_union_times_ms(new NumberSeq()), 89 90 _summary(new Summary()), 91 92 #ifndef PRODUCT 93 _cur_clear_ct_time_ms(0.0), 94 _min_clear_cc_time_ms(-1.0), 95 _max_clear_cc_time_ms(-1.0), 96 _cur_clear_cc_time_ms(0.0), 97 _cum_clear_cc_time_ms(0.0), 98 _num_cc_clears(0L), 99 #endif 100 101 _region_num_young(0), 102 _region_num_tenured(0), 103 _prev_region_num_young(0), 104 _prev_region_num_tenured(0), 105 106 _aux_num(10), 107 _all_aux_times_ms(new NumberSeq[_aux_num]), 108 _cur_aux_start_times_ms(new double[_aux_num]), 109 _cur_aux_times_ms(new double[_aux_num]), 110 _cur_aux_times_set(new bool[_aux_num]), 111 112 _concurrent_mark_init_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 113 _concurrent_mark_remark_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 114 _concurrent_mark_cleanup_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)), 115 116 // <NEW PREDICTION> 117 118 _alloc_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 119 _prev_collection_pause_end_ms(0.0), 120 _pending_card_diff_seq(new TruncatedSeq(TruncatedSeqLength)), 121 _rs_length_diff_seq(new TruncatedSeq(TruncatedSeqLength)), 122 _cost_per_card_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 123 _fully_young_cards_per_entry_ratio_seq(new TruncatedSeq(TruncatedSeqLength)), 124 _partially_young_cards_per_entry_ratio_seq( 125 new TruncatedSeq(TruncatedSeqLength)), 126 _cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 127 _partially_young_cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 128 _cost_per_byte_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 129 _cost_per_byte_ms_during_cm_seq(new TruncatedSeq(TruncatedSeqLength)), 130 _constant_other_time_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 131 _young_other_cost_per_region_ms_seq(new TruncatedSeq(TruncatedSeqLength)), 132 _non_young_other_cost_per_region_ms_seq( 133 new TruncatedSeq(TruncatedSeqLength)), 134 135 _pending_cards_seq(new TruncatedSeq(TruncatedSeqLength)), 136 _scanned_cards_seq(new TruncatedSeq(TruncatedSeqLength)), 137 _rs_lengths_seq(new TruncatedSeq(TruncatedSeqLength)), 138 139 _pause_time_target_ms((double) MaxGCPauseMillis), 140 141 // </NEW PREDICTION> 142 143 _in_young_gc_mode(false), 144 _full_young_gcs(true), 145 _full_young_pause_num(0), 146 _partial_young_pause_num(0), 147 148 _during_marking(false), 149 _in_marking_window(false), 150 _in_marking_window_im(false), 151 152 _known_garbage_ratio(0.0), 153 _known_garbage_bytes(0), 154 155 _young_gc_eff_seq(new TruncatedSeq(TruncatedSeqLength)), 156 157 _recent_prev_end_times_for_all_gcs_sec(new TruncatedSeq(NumPrevPausesForHeuristics)), 158 159 _recent_CS_bytes_used_before(new TruncatedSeq(NumPrevPausesForHeuristics)), 160 _recent_CS_bytes_surviving(new TruncatedSeq(NumPrevPausesForHeuristics)), 161 162 _recent_avg_pause_time_ratio(0.0), 163 _num_markings(0), 164 _n_marks(0), 165 _n_pauses_at_mark_end(0), 166 167 _all_full_gc_times_ms(new NumberSeq()), 168 169 // G1PausesBtwnConcMark defaults to -1 170 // so the hack is to do the cast QQQ FIXME 171 _pauses_btwn_concurrent_mark((size_t)G1PausesBtwnConcMark), 172 _n_marks_since_last_pause(0), 173 _initiate_conc_mark_if_possible(false), 174 _during_initial_mark_pause(false), 175 _should_revert_to_full_young_gcs(false), 176 _last_full_young_gc(false), 177 178 _prev_collection_pause_used_at_end_bytes(0), 179 180 _collection_set(NULL), 181 _collection_set_size(0), 182 _collection_set_bytes_used_before(0), 183 184 // Incremental CSet attributes 185 _inc_cset_build_state(Inactive), 186 _inc_cset_head(NULL), 187 _inc_cset_tail(NULL), 188 _inc_cset_size(0), 189 _inc_cset_young_index(0), 190 _inc_cset_bytes_used_before(0), 191 _inc_cset_max_finger(NULL), 192 _inc_cset_recorded_young_bytes(0), 193 _inc_cset_recorded_rs_lengths(0), 194 _inc_cset_predicted_elapsed_time_ms(0.0), 195 _inc_cset_predicted_bytes_to_copy(0), 196 197 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away 198 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list 199 #endif // _MSC_VER 200 201 _short_lived_surv_rate_group(new SurvRateGroup(this, "Short Lived", 202 G1YoungSurvRateNumRegionsSummary)), 203 _survivor_surv_rate_group(new SurvRateGroup(this, "Survivor", 204 G1YoungSurvRateNumRegionsSummary)), 205 // add here any more surv rate groups 206 _recorded_survivor_regions(0), 207 _recorded_survivor_head(NULL), 208 _recorded_survivor_tail(NULL), 209 _survivors_age_table(true), 210 211 _gc_overhead_perc(0.0) 212 213 { 214 // Set up the region size and associated fields. Given that the 215 // policy is created before the heap, we have to set this up here, 216 // so it's done as soon as possible. 217 HeapRegion::setup_heap_region_size(Arguments::min_heap_size()); 218 HeapRegionRemSet::setup_remset_size(); 219 220 // Verify PLAB sizes 221 const uint region_size = HeapRegion::GrainWords; 222 if (YoungPLABSize > region_size || OldPLABSize > region_size) { 223 char buffer[128]; 224 jio_snprintf(buffer, sizeof(buffer), "%sPLABSize should be at most %u", 225 OldPLABSize > region_size ? "Old" : "Young", region_size); 226 vm_exit_during_initialization(buffer); 227 } 228 229 _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime()); 230 _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0; 231 232 _par_last_gc_worker_start_times_ms = new double[_parallel_gc_threads]; 233 _par_last_ext_root_scan_times_ms = new double[_parallel_gc_threads]; 234 _par_last_mark_stack_scan_times_ms = new double[_parallel_gc_threads]; 235 236 _par_last_update_rs_times_ms = new double[_parallel_gc_threads]; 237 _par_last_update_rs_processed_buffers = new double[_parallel_gc_threads]; 238 239 _par_last_scan_rs_times_ms = new double[_parallel_gc_threads]; 240 241 _par_last_obj_copy_times_ms = new double[_parallel_gc_threads]; 242 243 _par_last_termination_times_ms = new double[_parallel_gc_threads]; 244 _par_last_termination_attempts = new double[_parallel_gc_threads]; 245 _par_last_gc_worker_end_times_ms = new double[_parallel_gc_threads]; 246 247 // start conservatively 248 _expensive_region_limit_ms = 0.5 * (double) MaxGCPauseMillis; 249 250 // <NEW PREDICTION> 251 252 int index; 253 if (ParallelGCThreads == 0) 254 index = 0; 255 else if (ParallelGCThreads > 8) 256 index = 7; 257 else 258 index = ParallelGCThreads - 1; 259 260 _pending_card_diff_seq->add(0.0); 261 _rs_length_diff_seq->add(rs_length_diff_defaults[index]); 262 _cost_per_card_ms_seq->add(cost_per_card_ms_defaults[index]); 263 _fully_young_cards_per_entry_ratio_seq->add( 264 fully_young_cards_per_entry_ratio_defaults[index]); 265 _cost_per_entry_ms_seq->add(cost_per_entry_ms_defaults[index]); 266 _cost_per_byte_ms_seq->add(cost_per_byte_ms_defaults[index]); 267 _constant_other_time_ms_seq->add(constant_other_time_ms_defaults[index]); 268 _young_other_cost_per_region_ms_seq->add( 269 young_other_cost_per_region_ms_defaults[index]); 270 _non_young_other_cost_per_region_ms_seq->add( 271 non_young_other_cost_per_region_ms_defaults[index]); 272 273 // </NEW PREDICTION> 274 275 // Below, we might need to calculate the pause time target based on 276 // the pause interval. When we do so we are going to give G1 maximum 277 // flexibility and allow it to do pauses when it needs to. So, we'll 278 // arrange that the pause interval to be pause time target + 1 to 279 // ensure that a) the pause time target is maximized with respect to 280 // the pause interval and b) we maintain the invariant that pause 281 // time target < pause interval. If the user does not want this 282 // maximum flexibility, they will have to set the pause interval 283 // explicitly. 284 285 // First make sure that, if either parameter is set, its value is 286 // reasonable. 287 if (!FLAG_IS_DEFAULT(MaxGCPauseMillis)) { 288 if (MaxGCPauseMillis < 1) { 289 vm_exit_during_initialization("MaxGCPauseMillis should be " 290 "greater than 0"); 291 } 292 } 293 if (!FLAG_IS_DEFAULT(GCPauseIntervalMillis)) { 294 if (GCPauseIntervalMillis < 1) { 295 vm_exit_during_initialization("GCPauseIntervalMillis should be " 296 "greater than 0"); 297 } 298 } 299 300 // Then, if the pause time target parameter was not set, set it to 301 // the default value. 302 if (FLAG_IS_DEFAULT(MaxGCPauseMillis)) { 303 if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) { 304 // The default pause time target in G1 is 200ms 305 FLAG_SET_DEFAULT(MaxGCPauseMillis, 200); 306 } else { 307 // We do not allow the pause interval to be set without the 308 // pause time target 309 vm_exit_during_initialization("GCPauseIntervalMillis cannot be set " 310 "without setting MaxGCPauseMillis"); 311 } 312 } 313 314 // Then, if the interval parameter was not set, set it according to 315 // the pause time target (this will also deal with the case when the 316 // pause time target is the default value). 317 if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) { 318 FLAG_SET_DEFAULT(GCPauseIntervalMillis, MaxGCPauseMillis + 1); 319 } 320 321 // Finally, make sure that the two parameters are consistent. 322 if (MaxGCPauseMillis >= GCPauseIntervalMillis) { 323 char buffer[256]; 324 jio_snprintf(buffer, 256, 325 "MaxGCPauseMillis (%u) should be less than " 326 "GCPauseIntervalMillis (%u)", 327 MaxGCPauseMillis, GCPauseIntervalMillis); 328 vm_exit_during_initialization(buffer); 329 } 330 331 double max_gc_time = (double) MaxGCPauseMillis / 1000.0; 332 double time_slice = (double) GCPauseIntervalMillis / 1000.0; 333 _mmu_tracker = new G1MMUTrackerQueue(time_slice, max_gc_time); 334 _sigma = (double) G1ConfidencePercent / 100.0; 335 336 // start conservatively (around 50ms is about right) 337 _concurrent_mark_init_times_ms->add(0.05); 338 _concurrent_mark_remark_times_ms->add(0.05); 339 _concurrent_mark_cleanup_times_ms->add(0.20); 340 _tenuring_threshold = MaxTenuringThreshold; 341 342 // if G1FixedSurvivorSpaceSize is 0 which means the size is not 343 // fixed, then _max_survivor_regions will be calculated at 344 // calculate_young_list_target_length during initialization 345 _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; 346 347 assert(GCTimeRatio > 0, 348 "we should have set it to a default value set_g1_gc_flags() " 349 "if a user set it to 0"); 350 _gc_overhead_perc = 100.0 * (1.0 / (1.0 + GCTimeRatio)); 351 352 initialize_all(); 353 } 354 355 // Increment "i", mod "len" 356 static void inc_mod(int& i, int len) { 357 i++; if (i == len) i = 0; 358 } 359 360 void G1CollectorPolicy::initialize_flags() { 361 set_min_alignment(HeapRegion::GrainBytes); 362 set_max_alignment(GenRemSet::max_alignment_constraint(rem_set_name())); 363 if (SurvivorRatio < 1) { 364 vm_exit_during_initialization("Invalid survivor ratio specified"); 365 } 366 CollectorPolicy::initialize_flags(); 367 } 368 369 // The easiest way to deal with the parsing of the NewSize / 370 // MaxNewSize / etc. parameteres is to re-use the code in the 371 // TwoGenerationCollectorPolicy class. This is similar to what 372 // ParallelScavenge does with its GenerationSizer class (see 373 // ParallelScavengeHeap::initialize()). We might change this in the 374 // future, but it's a good start. 375 class G1YoungGenSizer : public TwoGenerationCollectorPolicy { 376 size_t size_to_region_num(size_t byte_size) { 377 return MAX2((size_t) 1, byte_size / HeapRegion::GrainBytes); 378 } 379 380 public: 381 G1YoungGenSizer() { 382 initialize_flags(); 383 initialize_size_info(); 384 } 385 386 size_t min_young_region_num() { 387 return size_to_region_num(_min_gen0_size); 388 } 389 size_t initial_young_region_num() { 390 return size_to_region_num(_initial_gen0_size); 391 } 392 size_t max_young_region_num() { 393 return size_to_region_num(_max_gen0_size); 394 } 395 }; 396 397 void G1CollectorPolicy::init() { 398 // Set aside an initial future to_space. 399 _g1 = G1CollectedHeap::heap(); 400 401 assert(Heap_lock->owned_by_self(), "Locking discipline."); 402 403 initialize_gc_policy_counters(); 404 405 if (G1Gen) { 406 _in_young_gc_mode = true; 407 408 G1YoungGenSizer sizer; 409 size_t initial_region_num = sizer.initial_young_region_num(); 410 411 if (UseAdaptiveSizePolicy) { 412 set_adaptive_young_list_length(true); 413 _young_list_fixed_length = 0; 414 } else { 415 set_adaptive_young_list_length(false); 416 _young_list_fixed_length = initial_region_num; 417 } 418 _free_regions_at_end_of_collection = _g1->free_regions(); 419 calculate_young_list_min_length(); 420 guarantee( _young_list_min_length == 0, "invariant, not enough info" ); 421 calculate_young_list_target_length(); 422 } else { 423 _young_list_fixed_length = 0; 424 _in_young_gc_mode = false; 425 } 426 427 // We may immediately start allocating regions and placing them on the 428 // collection set list. Initialize the per-collection set info 429 start_incremental_cset_building(); 430 } 431 432 // Create the jstat counters for the policy. 433 void G1CollectorPolicy::initialize_gc_policy_counters() 434 { 435 _gc_policy_counters = new GCPolicyCounters("GarbageFirst", 1, 2 + G1Gen); 436 } 437 438 void G1CollectorPolicy::calculate_young_list_min_length() { 439 _young_list_min_length = 0; 440 441 if (!adaptive_young_list_length()) 442 return; 443 444 if (_alloc_rate_ms_seq->num() > 3) { 445 double now_sec = os::elapsedTime(); 446 double when_ms = _mmu_tracker->when_max_gc_sec(now_sec) * 1000.0; 447 double alloc_rate_ms = predict_alloc_rate_ms(); 448 int min_regions = (int) ceil(alloc_rate_ms * when_ms); 449 int current_region_num = (int) _g1->young_list()->length(); 450 _young_list_min_length = min_regions + current_region_num; 451 } 452 } 453 454 void G1CollectorPolicy::calculate_young_list_target_length() { 455 if (adaptive_young_list_length()) { 456 size_t rs_lengths = (size_t) get_new_prediction(_rs_lengths_seq); 457 calculate_young_list_target_length(rs_lengths); 458 } else { 459 if (full_young_gcs()) 460 _young_list_target_length = _young_list_fixed_length; 461 else 462 _young_list_target_length = _young_list_fixed_length / 2; 463 464 _young_list_target_length = MAX2(_young_list_target_length, (size_t)1); 465 } 466 calculate_survivors_policy(); 467 } 468 469 void G1CollectorPolicy::calculate_young_list_target_length(size_t rs_lengths) { 470 guarantee( adaptive_young_list_length(), "pre-condition" ); 471 guarantee( !_in_marking_window || !_last_full_young_gc, "invariant" ); 472 473 double start_time_sec = os::elapsedTime(); 474 size_t min_reserve_perc = MAX2((size_t)2, (size_t)G1ReservePercent); 475 min_reserve_perc = MIN2((size_t) 50, min_reserve_perc); 476 size_t reserve_regions = 477 (size_t) ((double) min_reserve_perc * (double) _g1->n_regions() / 100.0); 478 479 if (full_young_gcs() && _free_regions_at_end_of_collection > 0) { 480 // we are in fully-young mode and there are free regions in the heap 481 482 double survivor_regions_evac_time = 483 predict_survivor_regions_evac_time(); 484 485 double target_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0; 486 size_t pending_cards = (size_t) get_new_prediction(_pending_cards_seq); 487 size_t adj_rs_lengths = rs_lengths + predict_rs_length_diff(); 488 size_t scanned_cards = predict_young_card_num(adj_rs_lengths); 489 double base_time_ms = predict_base_elapsed_time_ms(pending_cards, scanned_cards) 490 + survivor_regions_evac_time; 491 492 // the result 493 size_t final_young_length = 0; 494 495 size_t init_free_regions = 496 MAX2((size_t)0, _free_regions_at_end_of_collection - reserve_regions); 497 498 // if we're still under the pause target... 499 if (base_time_ms <= target_pause_time_ms) { 500 // We make sure that the shortest young length that makes sense 501 // fits within the target pause time. 502 size_t min_young_length = 1; 503 504 if (predict_will_fit(min_young_length, base_time_ms, 505 init_free_regions, target_pause_time_ms)) { 506 // The shortest young length will fit within the target pause time; 507 // we'll now check whether the absolute maximum number of young 508 // regions will fit in the target pause time. If not, we'll do 509 // a binary search between min_young_length and max_young_length 510 size_t abs_max_young_length = _free_regions_at_end_of_collection - 1; 511 size_t max_young_length = abs_max_young_length; 512 513 if (max_young_length > min_young_length) { 514 // Let's check if the initial max young length will fit within the 515 // target pause. If so then there is no need to search for a maximal 516 // young length - we'll return the initial maximum 517 518 if (predict_will_fit(max_young_length, base_time_ms, 519 init_free_regions, target_pause_time_ms)) { 520 // The maximum young length will satisfy the target pause time. 521 // We are done so set min young length to this maximum length. 522 // The code after the loop will then set final_young_length using 523 // the value cached in the minimum length. 524 min_young_length = max_young_length; 525 } else { 526 // The maximum possible number of young regions will not fit within 527 // the target pause time so let's search.... 528 529 size_t diff = (max_young_length - min_young_length) / 2; 530 max_young_length = min_young_length + diff; 531 532 while (max_young_length > min_young_length) { 533 if (predict_will_fit(max_young_length, base_time_ms, 534 init_free_regions, target_pause_time_ms)) { 535 536 // The current max young length will fit within the target 537 // pause time. Note we do not exit the loop here. By setting 538 // min = max, and then increasing the max below means that 539 // we will continue searching for an upper bound in the 540 // range [max..max+diff] 541 min_young_length = max_young_length; 542 } 543 diff = (max_young_length - min_young_length) / 2; 544 max_young_length = min_young_length + diff; 545 } 546 // the above loop found a maximal young length that will fit 547 // within the target pause time. 548 } 549 assert(min_young_length <= abs_max_young_length, "just checking"); 550 } 551 final_young_length = min_young_length; 552 } 553 } 554 // and we're done! 555 556 // we should have at least one region in the target young length 557 _young_list_target_length = 558 MAX2((size_t) 1, final_young_length + _recorded_survivor_regions); 559 560 // let's keep an eye of how long we spend on this calculation 561 // right now, I assume that we'll print it when we need it; we 562 // should really adde it to the breakdown of a pause 563 double end_time_sec = os::elapsedTime(); 564 double elapsed_time_ms = (end_time_sec - start_time_sec) * 1000.0; 565 566 #ifdef TRACE_CALC_YOUNG_LENGTH 567 // leave this in for debugging, just in case 568 gclog_or_tty->print_cr("target = %1.1lf ms, young = " SIZE_FORMAT ", " 569 "elapsed %1.2lf ms, (%s%s) " SIZE_FORMAT SIZE_FORMAT, 570 target_pause_time_ms, 571 _young_list_target_length 572 elapsed_time_ms, 573 full_young_gcs() ? "full" : "partial", 574 during_initial_mark_pause() ? " i-m" : "", 575 _in_marking_window, 576 _in_marking_window_im); 577 #endif // TRACE_CALC_YOUNG_LENGTH 578 579 if (_young_list_target_length < _young_list_min_length) { 580 // bummer; this means that, if we do a pause when the maximal 581 // length dictates, we'll violate the pause spacing target (the 582 // min length was calculate based on the application's current 583 // alloc rate); 584 585 // so, we have to bite the bullet, and allocate the minimum 586 // number. We'll violate our target, but we just can't meet it. 587 588 #ifdef TRACE_CALC_YOUNG_LENGTH 589 // leave this in for debugging, just in case 590 gclog_or_tty->print_cr("adjusted target length from " 591 SIZE_FORMAT " to " SIZE_FORMAT, 592 _young_list_target_length, _young_list_min_length); 593 #endif // TRACE_CALC_YOUNG_LENGTH 594 595 _young_list_target_length = _young_list_min_length; 596 } 597 } else { 598 // we are in a partially-young mode or we've run out of regions (due 599 // to evacuation failure) 600 601 #ifdef TRACE_CALC_YOUNG_LENGTH 602 // leave this in for debugging, just in case 603 gclog_or_tty->print_cr("(partial) setting target to " SIZE_FORMAT 604 _young_list_min_length); 605 #endif // TRACE_CALC_YOUNG_LENGTH 606 // we'll do the pause as soon as possible by choosing the minimum 607 _young_list_target_length = 608 MAX2(_young_list_min_length, (size_t) 1); 609 } 610 611 _rs_lengths_prediction = rs_lengths; 612 } 613 614 // This is used by: calculate_young_list_target_length(rs_length). It 615 // returns true iff: 616 // the predicted pause time for the given young list will not overflow 617 // the target pause time 618 // and: 619 // the predicted amount of surviving data will not overflow the 620 // the amount of free space available for survivor regions. 621 // 622 bool 623 G1CollectorPolicy::predict_will_fit(size_t young_length, 624 double base_time_ms, 625 size_t init_free_regions, 626 double target_pause_time_ms) { 627 628 if (young_length >= init_free_regions) 629 // end condition 1: not enough space for the young regions 630 return false; 631 632 double accum_surv_rate_adj = 0.0; 633 double accum_surv_rate = 634 accum_yg_surv_rate_pred((int)(young_length - 1)) - accum_surv_rate_adj; 635 636 size_t bytes_to_copy = 637 (size_t) (accum_surv_rate * (double) HeapRegion::GrainBytes); 638 639 double copy_time_ms = predict_object_copy_time_ms(bytes_to_copy); 640 641 double young_other_time_ms = 642 predict_young_other_time_ms(young_length); 643 644 double pause_time_ms = 645 base_time_ms + copy_time_ms + young_other_time_ms; 646 647 if (pause_time_ms > target_pause_time_ms) 648 // end condition 2: over the target pause time 649 return false; 650 651 size_t free_bytes = 652 (init_free_regions - young_length) * HeapRegion::GrainBytes; 653 654 if ((2.0 + sigma()) * (double) bytes_to_copy > (double) free_bytes) 655 // end condition 3: out of to-space (conservatively) 656 return false; 657 658 // success! 659 return true; 660 } 661 662 double G1CollectorPolicy::predict_survivor_regions_evac_time() { 663 double survivor_regions_evac_time = 0.0; 664 for (HeapRegion * r = _recorded_survivor_head; 665 r != NULL && r != _recorded_survivor_tail->get_next_young_region(); 666 r = r->get_next_young_region()) { 667 survivor_regions_evac_time += predict_region_elapsed_time_ms(r, true); 668 } 669 return survivor_regions_evac_time; 670 } 671 672 void G1CollectorPolicy::check_prediction_validity() { 673 guarantee( adaptive_young_list_length(), "should not call this otherwise" ); 674 675 size_t rs_lengths = _g1->young_list()->sampled_rs_lengths(); 676 if (rs_lengths > _rs_lengths_prediction) { 677 // add 10% to avoid having to recalculate often 678 size_t rs_lengths_prediction = rs_lengths * 1100 / 1000; 679 calculate_young_list_target_length(rs_lengths_prediction); 680 } 681 } 682 683 HeapWord* G1CollectorPolicy::mem_allocate_work(size_t size, 684 bool is_tlab, 685 bool* gc_overhead_limit_was_exceeded) { 686 guarantee(false, "Not using this policy feature yet."); 687 return NULL; 688 } 689 690 // This method controls how a collector handles one or more 691 // of its generations being fully allocated. 692 HeapWord* G1CollectorPolicy::satisfy_failed_allocation(size_t size, 693 bool is_tlab) { 694 guarantee(false, "Not using this policy feature yet."); 695 return NULL; 696 } 697 698 699 #ifndef PRODUCT 700 bool G1CollectorPolicy::verify_young_ages() { 701 HeapRegion* head = _g1->young_list()->first_region(); 702 return 703 verify_young_ages(head, _short_lived_surv_rate_group); 704 // also call verify_young_ages on any additional surv rate groups 705 } 706 707 bool 708 G1CollectorPolicy::verify_young_ages(HeapRegion* head, 709 SurvRateGroup *surv_rate_group) { 710 guarantee( surv_rate_group != NULL, "pre-condition" ); 711 712 const char* name = surv_rate_group->name(); 713 bool ret = true; 714 int prev_age = -1; 715 716 for (HeapRegion* curr = head; 717 curr != NULL; 718 curr = curr->get_next_young_region()) { 719 SurvRateGroup* group = curr->surv_rate_group(); 720 if (group == NULL && !curr->is_survivor()) { 721 gclog_or_tty->print_cr("## %s: encountered NULL surv_rate_group", name); 722 ret = false; 723 } 724 725 if (surv_rate_group == group) { 726 int age = curr->age_in_surv_rate_group(); 727 728 if (age < 0) { 729 gclog_or_tty->print_cr("## %s: encountered negative age", name); 730 ret = false; 731 } 732 733 if (age <= prev_age) { 734 gclog_or_tty->print_cr("## %s: region ages are not strictly increasing " 735 "(%d, %d)", name, age, prev_age); 736 ret = false; 737 } 738 prev_age = age; 739 } 740 } 741 742 return ret; 743 } 744 #endif // PRODUCT 745 746 void G1CollectorPolicy::record_full_collection_start() { 747 _cur_collection_start_sec = os::elapsedTime(); 748 // Release the future to-space so that it is available for compaction into. 749 _g1->set_full_collection(); 750 } 751 752 void G1CollectorPolicy::record_full_collection_end() { 753 // Consider this like a collection pause for the purposes of allocation 754 // since last pause. 755 double end_sec = os::elapsedTime(); 756 double full_gc_time_sec = end_sec - _cur_collection_start_sec; 757 double full_gc_time_ms = full_gc_time_sec * 1000.0; 758 759 _all_full_gc_times_ms->add(full_gc_time_ms); 760 761 update_recent_gc_times(end_sec, full_gc_time_ms); 762 763 _g1->clear_full_collection(); 764 765 // "Nuke" the heuristics that control the fully/partially young GC 766 // transitions and make sure we start with fully young GCs after the 767 // Full GC. 768 set_full_young_gcs(true); 769 _last_full_young_gc = false; 770 _should_revert_to_full_young_gcs = false; 771 clear_initiate_conc_mark_if_possible(); 772 clear_during_initial_mark_pause(); 773 _known_garbage_bytes = 0; 774 _known_garbage_ratio = 0.0; 775 _in_marking_window = false; 776 _in_marking_window_im = false; 777 778 _short_lived_surv_rate_group->start_adding_regions(); 779 // also call this on any additional surv rate groups 780 781 record_survivor_regions(0, NULL, NULL); 782 783 _prev_region_num_young = _region_num_young; 784 _prev_region_num_tenured = _region_num_tenured; 785 786 _free_regions_at_end_of_collection = _g1->free_regions(); 787 // Reset survivors SurvRateGroup. 788 _survivor_surv_rate_group->reset(); 789 calculate_young_list_min_length(); 790 calculate_young_list_target_length(); 791 } 792 793 void G1CollectorPolicy::record_before_bytes(size_t bytes) { 794 _bytes_in_to_space_before_gc += bytes; 795 } 796 797 void G1CollectorPolicy::record_after_bytes(size_t bytes) { 798 _bytes_in_to_space_after_gc += bytes; 799 } 800 801 void G1CollectorPolicy::record_stop_world_start() { 802 _stop_world_start = os::elapsedTime(); 803 } 804 805 void G1CollectorPolicy::record_collection_pause_start(double start_time_sec, 806 size_t start_used) { 807 if (PrintGCDetails) { 808 gclog_or_tty->stamp(PrintGCTimeStamps); 809 gclog_or_tty->print("[GC pause"); 810 if (in_young_gc_mode()) 811 gclog_or_tty->print(" (%s)", full_young_gcs() ? "young" : "partial"); 812 } 813 814 assert(_g1->used_regions() == _g1->recalculate_used_regions(), 815 "sanity"); 816 assert(_g1->used() == _g1->recalculate_used(), "sanity"); 817 818 double s_w_t_ms = (start_time_sec - _stop_world_start) * 1000.0; 819 _all_stop_world_times_ms->add(s_w_t_ms); 820 _stop_world_start = 0.0; 821 822 _cur_collection_start_sec = start_time_sec; 823 _cur_collection_pause_used_at_start_bytes = start_used; 824 _cur_collection_pause_used_regions_at_start = _g1->used_regions(); 825 _pending_cards = _g1->pending_card_num(); 826 _max_pending_cards = _g1->max_pending_card_num(); 827 828 _bytes_in_to_space_before_gc = 0; 829 _bytes_in_to_space_after_gc = 0; 830 _bytes_in_collection_set_before_gc = 0; 831 832 #ifdef DEBUG 833 // initialise these to something well known so that we can spot 834 // if they are not set properly 835 836 for (int i = 0; i < _parallel_gc_threads; ++i) { 837 _par_last_gc_worker_start_times_ms[i] = -1234.0; 838 _par_last_ext_root_scan_times_ms[i] = -1234.0; 839 _par_last_mark_stack_scan_times_ms[i] = -1234.0; 840 _par_last_update_rs_times_ms[i] = -1234.0; 841 _par_last_update_rs_processed_buffers[i] = -1234.0; 842 _par_last_scan_rs_times_ms[i] = -1234.0; 843 _par_last_obj_copy_times_ms[i] = -1234.0; 844 _par_last_termination_times_ms[i] = -1234.0; 845 _par_last_termination_attempts[i] = -1234.0; 846 _par_last_gc_worker_end_times_ms[i] = -1234.0; 847 } 848 #endif 849 850 for (int i = 0; i < _aux_num; ++i) { 851 _cur_aux_times_ms[i] = 0.0; 852 _cur_aux_times_set[i] = false; 853 } 854 855 _satb_drain_time_set = false; 856 _last_satb_drain_processed_buffers = -1; 857 858 if (in_young_gc_mode()) 859 _last_young_gc_full = false; 860 861 // do that for any other surv rate groups 862 _short_lived_surv_rate_group->stop_adding_regions(); 863 _survivors_age_table.clear(); 864 865 assert( verify_young_ages(), "region age verification" ); 866 } 867 868 void G1CollectorPolicy::record_mark_closure_time(double mark_closure_time_ms) { 869 _mark_closure_time_ms = mark_closure_time_ms; 870 } 871 872 void G1CollectorPolicy::record_concurrent_mark_init_start() { 873 _mark_init_start_sec = os::elapsedTime(); 874 guarantee(!in_young_gc_mode(), "should not do be here in young GC mode"); 875 } 876 877 void G1CollectorPolicy::record_concurrent_mark_init_end_pre(double 878 mark_init_elapsed_time_ms) { 879 _during_marking = true; 880 assert(!initiate_conc_mark_if_possible(), "we should have cleared it by now"); 881 clear_during_initial_mark_pause(); 882 _cur_mark_stop_world_time_ms = mark_init_elapsed_time_ms; 883 } 884 885 void G1CollectorPolicy::record_concurrent_mark_init_end() { 886 double end_time_sec = os::elapsedTime(); 887 double elapsed_time_ms = (end_time_sec - _mark_init_start_sec) * 1000.0; 888 _concurrent_mark_init_times_ms->add(elapsed_time_ms); 889 record_concurrent_mark_init_end_pre(elapsed_time_ms); 890 891 _mmu_tracker->add_pause(_mark_init_start_sec, end_time_sec, true); 892 } 893 894 void G1CollectorPolicy::record_concurrent_mark_remark_start() { 895 _mark_remark_start_sec = os::elapsedTime(); 896 _during_marking = false; 897 } 898 899 void G1CollectorPolicy::record_concurrent_mark_remark_end() { 900 double end_time_sec = os::elapsedTime(); 901 double elapsed_time_ms = (end_time_sec - _mark_remark_start_sec)*1000.0; 902 _concurrent_mark_remark_times_ms->add(elapsed_time_ms); 903 _cur_mark_stop_world_time_ms += elapsed_time_ms; 904 _prev_collection_pause_end_ms += elapsed_time_ms; 905 906 _mmu_tracker->add_pause(_mark_remark_start_sec, end_time_sec, true); 907 } 908 909 void G1CollectorPolicy::record_concurrent_mark_cleanup_start() { 910 _mark_cleanup_start_sec = os::elapsedTime(); 911 } 912 913 void 914 G1CollectorPolicy::record_concurrent_mark_cleanup_end(size_t freed_bytes, 915 size_t max_live_bytes) { 916 record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes); 917 record_concurrent_mark_cleanup_end_work2(); 918 } 919 920 void 921 G1CollectorPolicy:: 922 record_concurrent_mark_cleanup_end_work1(size_t freed_bytes, 923 size_t max_live_bytes) { 924 if (_n_marks < 2) _n_marks++; 925 if (G1PolicyVerbose > 0) 926 gclog_or_tty->print_cr("At end of marking, max_live is " SIZE_FORMAT " MB " 927 " (of " SIZE_FORMAT " MB heap).", 928 max_live_bytes/M, _g1->capacity()/M); 929 } 930 931 // The important thing about this is that it includes "os::elapsedTime". 932 void G1CollectorPolicy::record_concurrent_mark_cleanup_end_work2() { 933 double end_time_sec = os::elapsedTime(); 934 double elapsed_time_ms = (end_time_sec - _mark_cleanup_start_sec)*1000.0; 935 _concurrent_mark_cleanup_times_ms->add(elapsed_time_ms); 936 _cur_mark_stop_world_time_ms += elapsed_time_ms; 937 _prev_collection_pause_end_ms += elapsed_time_ms; 938 939 _mmu_tracker->add_pause(_mark_cleanup_start_sec, end_time_sec, true); 940 941 _num_markings++; 942 943 // We did a marking, so reset the "since_last_mark" variables. 944 double considerConcMarkCost = 1.0; 945 // If there are available processors, concurrent activity is free... 946 if (Threads::number_of_non_daemon_threads() * 2 < 947 os::active_processor_count()) { 948 considerConcMarkCost = 0.0; 949 } 950 _n_pauses_at_mark_end = _n_pauses; 951 _n_marks_since_last_pause++; 952 } 953 954 void 955 G1CollectorPolicy::record_concurrent_mark_cleanup_completed() { 956 if (in_young_gc_mode()) { 957 _should_revert_to_full_young_gcs = false; 958 _last_full_young_gc = true; 959 _in_marking_window = false; 960 if (adaptive_young_list_length()) 961 calculate_young_list_target_length(); 962 } 963 } 964 965 void G1CollectorPolicy::record_concurrent_pause() { 966 if (_stop_world_start > 0.0) { 967 double yield_ms = (os::elapsedTime() - _stop_world_start) * 1000.0; 968 _all_yield_times_ms->add(yield_ms); 969 } 970 } 971 972 void G1CollectorPolicy::record_concurrent_pause_end() { 973 } 974 975 void G1CollectorPolicy::record_collection_pause_end_CH_strong_roots() { 976 _cur_CH_strong_roots_end_sec = os::elapsedTime(); 977 _cur_CH_strong_roots_dur_ms = 978 (_cur_CH_strong_roots_end_sec - _cur_collection_start_sec) * 1000.0; 979 } 980 981 void G1CollectorPolicy::record_collection_pause_end_G1_strong_roots() { 982 _cur_G1_strong_roots_end_sec = os::elapsedTime(); 983 _cur_G1_strong_roots_dur_ms = 984 (_cur_G1_strong_roots_end_sec - _cur_CH_strong_roots_end_sec) * 1000.0; 985 } 986 987 template<class T> 988 T sum_of(T* sum_arr, int start, int n, int N) { 989 T sum = (T)0; 990 for (int i = 0; i < n; i++) { 991 int j = (start + i) % N; 992 sum += sum_arr[j]; 993 } 994 return sum; 995 } 996 997 void G1CollectorPolicy::print_par_stats(int level, 998 const char* str, 999 double* data, 1000 bool summary) { 1001 double min = data[0], max = data[0]; 1002 double total = 0.0; 1003 int j; 1004 for (j = 0; j < level; ++j) 1005 gclog_or_tty->print(" "); 1006 gclog_or_tty->print("[%s (ms):", str); 1007 for (uint i = 0; i < ParallelGCThreads; ++i) { 1008 double val = data[i]; 1009 if (val < min) 1010 min = val; 1011 if (val > max) 1012 max = val; 1013 total += val; 1014 gclog_or_tty->print(" %3.1lf", val); 1015 } 1016 if (summary) { 1017 gclog_or_tty->print_cr(""); 1018 double avg = total / (double) ParallelGCThreads; 1019 gclog_or_tty->print(" "); 1020 for (j = 0; j < level; ++j) 1021 gclog_or_tty->print(" "); 1022 gclog_or_tty->print("Avg: %5.1lf, Min: %5.1lf, Max: %5.1lf", 1023 avg, min, max); 1024 } 1025 gclog_or_tty->print_cr("]"); 1026 } 1027 1028 void G1CollectorPolicy::print_par_sizes(int level, 1029 const char* str, 1030 double* data, 1031 bool summary) { 1032 double min = data[0], max = data[0]; 1033 double total = 0.0; 1034 int j; 1035 for (j = 0; j < level; ++j) 1036 gclog_or_tty->print(" "); 1037 gclog_or_tty->print("[%s :", str); 1038 for (uint i = 0; i < ParallelGCThreads; ++i) { 1039 double val = data[i]; 1040 if (val < min) 1041 min = val; 1042 if (val > max) 1043 max = val; 1044 total += val; 1045 gclog_or_tty->print(" %d", (int) val); 1046 } 1047 if (summary) { 1048 gclog_or_tty->print_cr(""); 1049 double avg = total / (double) ParallelGCThreads; 1050 gclog_or_tty->print(" "); 1051 for (j = 0; j < level; ++j) 1052 gclog_or_tty->print(" "); 1053 gclog_or_tty->print("Sum: %d, Avg: %d, Min: %d, Max: %d", 1054 (int)total, (int)avg, (int)min, (int)max); 1055 } 1056 gclog_or_tty->print_cr("]"); 1057 } 1058 1059 void G1CollectorPolicy::print_stats (int level, 1060 const char* str, 1061 double value) { 1062 for (int j = 0; j < level; ++j) 1063 gclog_or_tty->print(" "); 1064 gclog_or_tty->print_cr("[%s: %5.1lf ms]", str, value); 1065 } 1066 1067 void G1CollectorPolicy::print_stats (int level, 1068 const char* str, 1069 int value) { 1070 for (int j = 0; j < level; ++j) 1071 gclog_or_tty->print(" "); 1072 gclog_or_tty->print_cr("[%s: %d]", str, value); 1073 } 1074 1075 double G1CollectorPolicy::avg_value (double* data) { 1076 if (ParallelGCThreads > 0) { 1077 double ret = 0.0; 1078 for (uint i = 0; i < ParallelGCThreads; ++i) 1079 ret += data[i]; 1080 return ret / (double) ParallelGCThreads; 1081 } else { 1082 return data[0]; 1083 } 1084 } 1085 1086 double G1CollectorPolicy::max_value (double* data) { 1087 if (ParallelGCThreads > 0) { 1088 double ret = data[0]; 1089 for (uint i = 1; i < ParallelGCThreads; ++i) 1090 if (data[i] > ret) 1091 ret = data[i]; 1092 return ret; 1093 } else { 1094 return data[0]; 1095 } 1096 } 1097 1098 double G1CollectorPolicy::sum_of_values (double* data) { 1099 if (ParallelGCThreads > 0) { 1100 double sum = 0.0; 1101 for (uint i = 0; i < ParallelGCThreads; i++) 1102 sum += data[i]; 1103 return sum; 1104 } else { 1105 return data[0]; 1106 } 1107 } 1108 1109 double G1CollectorPolicy::max_sum (double* data1, 1110 double* data2) { 1111 double ret = data1[0] + data2[0]; 1112 1113 if (ParallelGCThreads > 0) { 1114 for (uint i = 1; i < ParallelGCThreads; ++i) { 1115 double data = data1[i] + data2[i]; 1116 if (data > ret) 1117 ret = data; 1118 } 1119 } 1120 return ret; 1121 } 1122 1123 // Anything below that is considered to be zero 1124 #define MIN_TIMER_GRANULARITY 0.0000001 1125 1126 void G1CollectorPolicy::record_collection_pause_end() { 1127 double end_time_sec = os::elapsedTime(); 1128 double elapsed_ms = _last_pause_time_ms; 1129 bool parallel = ParallelGCThreads > 0; 1130 double evac_ms = (end_time_sec - _cur_G1_strong_roots_end_sec) * 1000.0; 1131 size_t rs_size = 1132 _cur_collection_pause_used_regions_at_start - collection_set_size(); 1133 size_t cur_used_bytes = _g1->used(); 1134 assert(cur_used_bytes == _g1->recalculate_used(), "It should!"); 1135 bool last_pause_included_initial_mark = false; 1136 bool update_stats = !_g1->evacuation_failed(); 1137 1138 #ifndef PRODUCT 1139 if (G1YoungSurvRateVerbose) { 1140 gclog_or_tty->print_cr(""); 1141 _short_lived_surv_rate_group->print(); 1142 // do that for any other surv rate groups too 1143 } 1144 #endif // PRODUCT 1145 1146 if (in_young_gc_mode()) { 1147 last_pause_included_initial_mark = during_initial_mark_pause(); 1148 if (last_pause_included_initial_mark) 1149 record_concurrent_mark_init_end_pre(0.0); 1150 1151 size_t min_used_targ = 1152 (_g1->capacity() / 100) * InitiatingHeapOccupancyPercent; 1153 1154 1155 if (!_g1->mark_in_progress() && !_last_full_young_gc) { 1156 assert(!last_pause_included_initial_mark, "invariant"); 1157 if (cur_used_bytes > min_used_targ && 1158 cur_used_bytes > _prev_collection_pause_used_at_end_bytes) { 1159 assert(!during_initial_mark_pause(), "we should not see this here"); 1160 1161 // Note: this might have already been set, if during the last 1162 // pause we decided to start a cycle but at the beginning of 1163 // this pause we decided to postpone it. That's OK. 1164 set_initiate_conc_mark_if_possible(); 1165 } 1166 } 1167 1168 _prev_collection_pause_used_at_end_bytes = cur_used_bytes; 1169 } 1170 1171 _mmu_tracker->add_pause(end_time_sec - elapsed_ms/1000.0, 1172 end_time_sec, false); 1173 1174 guarantee(_cur_collection_pause_used_regions_at_start >= 1175 collection_set_size(), 1176 "Negative RS size?"); 1177 1178 // This assert is exempted when we're doing parallel collection pauses, 1179 // because the fragmentation caused by the parallel GC allocation buffers 1180 // can lead to more memory being used during collection than was used 1181 // before. Best leave this out until the fragmentation problem is fixed. 1182 // Pauses in which evacuation failed can also lead to negative 1183 // collections, since no space is reclaimed from a region containing an 1184 // object whose evacuation failed. 1185 // Further, we're now always doing parallel collection. But I'm still 1186 // leaving this here as a placeholder for a more precise assertion later. 1187 // (DLD, 10/05.) 1188 assert((true || parallel) // Always using GC LABs now. 1189 || _g1->evacuation_failed() 1190 || _cur_collection_pause_used_at_start_bytes >= cur_used_bytes, 1191 "Negative collection"); 1192 1193 size_t freed_bytes = 1194 _cur_collection_pause_used_at_start_bytes - cur_used_bytes; 1195 size_t surviving_bytes = _collection_set_bytes_used_before - freed_bytes; 1196 1197 double survival_fraction = 1198 (double)surviving_bytes/ 1199 (double)_collection_set_bytes_used_before; 1200 1201 _n_pauses++; 1202 1203 if (update_stats) { 1204 _recent_CH_strong_roots_times_ms->add(_cur_CH_strong_roots_dur_ms); 1205 _recent_G1_strong_roots_times_ms->add(_cur_G1_strong_roots_dur_ms); 1206 _recent_evac_times_ms->add(evac_ms); 1207 _recent_pause_times_ms->add(elapsed_ms); 1208 1209 _recent_rs_sizes->add(rs_size); 1210 1211 // We exempt parallel collection from this check because Alloc Buffer 1212 // fragmentation can produce negative collections. Same with evac 1213 // failure. 1214 // Further, we're now always doing parallel collection. But I'm still 1215 // leaving this here as a placeholder for a more precise assertion later. 1216 // (DLD, 10/05. 1217 assert((true || parallel) 1218 || _g1->evacuation_failed() 1219 || surviving_bytes <= _collection_set_bytes_used_before, 1220 "Or else negative collection!"); 1221 _recent_CS_bytes_used_before->add(_collection_set_bytes_used_before); 1222 _recent_CS_bytes_surviving->add(surviving_bytes); 1223 1224 // this is where we update the allocation rate of the application 1225 double app_time_ms = 1226 (_cur_collection_start_sec * 1000.0 - _prev_collection_pause_end_ms); 1227 if (app_time_ms < MIN_TIMER_GRANULARITY) { 1228 // This usually happens due to the timer not having the required 1229 // granularity. Some Linuxes are the usual culprits. 1230 // We'll just set it to something (arbitrarily) small. 1231 app_time_ms = 1.0; 1232 } 1233 size_t regions_allocated = 1234 (_region_num_young - _prev_region_num_young) + 1235 (_region_num_tenured - _prev_region_num_tenured); 1236 double alloc_rate_ms = (double) regions_allocated / app_time_ms; 1237 _alloc_rate_ms_seq->add(alloc_rate_ms); 1238 _prev_region_num_young = _region_num_young; 1239 _prev_region_num_tenured = _region_num_tenured; 1240 1241 double interval_ms = 1242 (end_time_sec - _recent_prev_end_times_for_all_gcs_sec->oldest()) * 1000.0; 1243 update_recent_gc_times(end_time_sec, elapsed_ms); 1244 _recent_avg_pause_time_ratio = _recent_gc_times_ms->sum()/interval_ms; 1245 if (recent_avg_pause_time_ratio() < 0.0 || 1246 (recent_avg_pause_time_ratio() - 1.0 > 0.0)) { 1247 #ifndef PRODUCT 1248 // Dump info to allow post-facto debugging 1249 gclog_or_tty->print_cr("recent_avg_pause_time_ratio() out of bounds"); 1250 gclog_or_tty->print_cr("-------------------------------------------"); 1251 gclog_or_tty->print_cr("Recent GC Times (ms):"); 1252 _recent_gc_times_ms->dump(); 1253 gclog_or_tty->print_cr("(End Time=%3.3f) Recent GC End Times (s):", end_time_sec); 1254 _recent_prev_end_times_for_all_gcs_sec->dump(); 1255 gclog_or_tty->print_cr("GC = %3.3f, Interval = %3.3f, Ratio = %3.3f", 1256 _recent_gc_times_ms->sum(), interval_ms, recent_avg_pause_time_ratio()); 1257 // In debug mode, terminate the JVM if the user wants to debug at this point. 1258 assert(!G1FailOnFPError, "Debugging data for CR 6898948 has been dumped above"); 1259 #endif // !PRODUCT 1260 // Clip ratio between 0.0 and 1.0, and continue. This will be fixed in 1261 // CR 6902692 by redoing the manner in which the ratio is incrementally computed. 1262 if (_recent_avg_pause_time_ratio < 0.0) { 1263 _recent_avg_pause_time_ratio = 0.0; 1264 } else { 1265 assert(_recent_avg_pause_time_ratio - 1.0 > 0.0, "Ctl-point invariant"); 1266 _recent_avg_pause_time_ratio = 1.0; 1267 } 1268 } 1269 } 1270 1271 if (G1PolicyVerbose > 1) { 1272 gclog_or_tty->print_cr(" Recording collection pause(%d)", _n_pauses); 1273 } 1274 1275 PauseSummary* summary = _summary; 1276 1277 double ext_root_scan_time = avg_value(_par_last_ext_root_scan_times_ms); 1278 double mark_stack_scan_time = avg_value(_par_last_mark_stack_scan_times_ms); 1279 double update_rs_time = avg_value(_par_last_update_rs_times_ms); 1280 double update_rs_processed_buffers = 1281 sum_of_values(_par_last_update_rs_processed_buffers); 1282 double scan_rs_time = avg_value(_par_last_scan_rs_times_ms); 1283 double obj_copy_time = avg_value(_par_last_obj_copy_times_ms); 1284 double termination_time = avg_value(_par_last_termination_times_ms); 1285 1286 double parallel_other_time = _cur_collection_par_time_ms - 1287 (update_rs_time + ext_root_scan_time + mark_stack_scan_time + 1288 scan_rs_time + obj_copy_time + termination_time); 1289 if (update_stats) { 1290 MainBodySummary* body_summary = summary->main_body_summary(); 1291 guarantee(body_summary != NULL, "should not be null!"); 1292 1293 if (_satb_drain_time_set) 1294 body_summary->record_satb_drain_time_ms(_cur_satb_drain_time_ms); 1295 else 1296 body_summary->record_satb_drain_time_ms(0.0); 1297 body_summary->record_ext_root_scan_time_ms(ext_root_scan_time); 1298 body_summary->record_mark_stack_scan_time_ms(mark_stack_scan_time); 1299 body_summary->record_update_rs_time_ms(update_rs_time); 1300 body_summary->record_scan_rs_time_ms(scan_rs_time); 1301 body_summary->record_obj_copy_time_ms(obj_copy_time); 1302 if (parallel) { 1303 body_summary->record_parallel_time_ms(_cur_collection_par_time_ms); 1304 body_summary->record_clear_ct_time_ms(_cur_clear_ct_time_ms); 1305 body_summary->record_termination_time_ms(termination_time); 1306 body_summary->record_parallel_other_time_ms(parallel_other_time); 1307 } 1308 body_summary->record_mark_closure_time_ms(_mark_closure_time_ms); 1309 } 1310 1311 if (G1PolicyVerbose > 1) { 1312 gclog_or_tty->print_cr(" ET: %10.6f ms (avg: %10.6f ms)\n" 1313 " CH Strong: %10.6f ms (avg: %10.6f ms)\n" 1314 " G1 Strong: %10.6f ms (avg: %10.6f ms)\n" 1315 " Evac: %10.6f ms (avg: %10.6f ms)\n" 1316 " ET-RS: %10.6f ms (avg: %10.6f ms)\n" 1317 " |RS|: " SIZE_FORMAT, 1318 elapsed_ms, recent_avg_time_for_pauses_ms(), 1319 _cur_CH_strong_roots_dur_ms, recent_avg_time_for_CH_strong_ms(), 1320 _cur_G1_strong_roots_dur_ms, recent_avg_time_for_G1_strong_ms(), 1321 evac_ms, recent_avg_time_for_evac_ms(), 1322 scan_rs_time, 1323 recent_avg_time_for_pauses_ms() - 1324 recent_avg_time_for_G1_strong_ms(), 1325 rs_size); 1326 1327 gclog_or_tty->print_cr(" Used at start: " SIZE_FORMAT"K" 1328 " At end " SIZE_FORMAT "K\n" 1329 " garbage : " SIZE_FORMAT "K" 1330 " of " SIZE_FORMAT "K\n" 1331 " survival : %6.2f%% (%6.2f%% avg)", 1332 _cur_collection_pause_used_at_start_bytes/K, 1333 _g1->used()/K, freed_bytes/K, 1334 _collection_set_bytes_used_before/K, 1335 survival_fraction*100.0, 1336 recent_avg_survival_fraction()*100.0); 1337 gclog_or_tty->print_cr(" Recent %% gc pause time: %6.2f", 1338 recent_avg_pause_time_ratio() * 100.0); 1339 } 1340 1341 double other_time_ms = elapsed_ms; 1342 1343 if (_satb_drain_time_set) { 1344 other_time_ms -= _cur_satb_drain_time_ms; 1345 } 1346 1347 if (parallel) { 1348 other_time_ms -= _cur_collection_par_time_ms + _cur_clear_ct_time_ms; 1349 } else { 1350 other_time_ms -= 1351 update_rs_time + 1352 ext_root_scan_time + mark_stack_scan_time + 1353 scan_rs_time + obj_copy_time; 1354 } 1355 1356 if (PrintGCDetails) { 1357 gclog_or_tty->print_cr("%s, %1.8lf secs]", 1358 (last_pause_included_initial_mark) ? " (initial-mark)" : "", 1359 elapsed_ms / 1000.0); 1360 1361 if (_satb_drain_time_set) { 1362 print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms); 1363 } 1364 if (_last_satb_drain_processed_buffers >= 0) { 1365 print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers); 1366 } 1367 if (parallel) { 1368 print_stats(1, "Parallel Time", _cur_collection_par_time_ms); 1369 print_par_stats(2, "GC Worker Start Time", 1370 _par_last_gc_worker_start_times_ms, false); 1371 print_par_stats(2, "Update RS", _par_last_update_rs_times_ms); 1372 print_par_sizes(3, "Processed Buffers", 1373 _par_last_update_rs_processed_buffers, true); 1374 print_par_stats(2, "Ext Root Scanning", 1375 _par_last_ext_root_scan_times_ms); 1376 print_par_stats(2, "Mark Stack Scanning", 1377 _par_last_mark_stack_scan_times_ms); 1378 print_par_stats(2, "Scan RS", _par_last_scan_rs_times_ms); 1379 print_par_stats(2, "Object Copy", _par_last_obj_copy_times_ms); 1380 print_par_stats(2, "Termination", _par_last_termination_times_ms); 1381 print_par_sizes(3, "Termination Attempts", 1382 _par_last_termination_attempts, true); 1383 print_par_stats(2, "GC Worker End Time", 1384 _par_last_gc_worker_end_times_ms, false); 1385 print_stats(2, "Other", parallel_other_time); 1386 print_stats(1, "Clear CT", _cur_clear_ct_time_ms); 1387 } else { 1388 print_stats(1, "Update RS", update_rs_time); 1389 print_stats(2, "Processed Buffers", 1390 (int)update_rs_processed_buffers); 1391 print_stats(1, "Ext Root Scanning", ext_root_scan_time); 1392 print_stats(1, "Mark Stack Scanning", mark_stack_scan_time); 1393 print_stats(1, "Scan RS", scan_rs_time); 1394 print_stats(1, "Object Copying", obj_copy_time); 1395 } 1396 #ifndef PRODUCT 1397 print_stats(1, "Cur Clear CC", _cur_clear_cc_time_ms); 1398 print_stats(1, "Cum Clear CC", _cum_clear_cc_time_ms); 1399 print_stats(1, "Min Clear CC", _min_clear_cc_time_ms); 1400 print_stats(1, "Max Clear CC", _max_clear_cc_time_ms); 1401 if (_num_cc_clears > 0) { 1402 print_stats(1, "Avg Clear CC", _cum_clear_cc_time_ms / ((double)_num_cc_clears)); 1403 } 1404 #endif 1405 print_stats(1, "Other", other_time_ms); 1406 print_stats(2, "Choose CSet", _recorded_young_cset_choice_time_ms); 1407 1408 for (int i = 0; i < _aux_num; ++i) { 1409 if (_cur_aux_times_set[i]) { 1410 char buffer[96]; 1411 sprintf(buffer, "Aux%d", i); 1412 print_stats(1, buffer, _cur_aux_times_ms[i]); 1413 } 1414 } 1415 } 1416 if (PrintGCDetails) 1417 gclog_or_tty->print(" ["); 1418 if (PrintGC || PrintGCDetails) 1419 _g1->print_size_transition(gclog_or_tty, 1420 _cur_collection_pause_used_at_start_bytes, 1421 _g1->used(), _g1->capacity()); 1422 if (PrintGCDetails) 1423 gclog_or_tty->print_cr("]"); 1424 1425 _all_pause_times_ms->add(elapsed_ms); 1426 if (update_stats) { 1427 summary->record_total_time_ms(elapsed_ms); 1428 summary->record_other_time_ms(other_time_ms); 1429 } 1430 for (int i = 0; i < _aux_num; ++i) 1431 if (_cur_aux_times_set[i]) 1432 _all_aux_times_ms[i].add(_cur_aux_times_ms[i]); 1433 1434 // Reset marks-between-pauses counter. 1435 _n_marks_since_last_pause = 0; 1436 1437 // Update the efficiency-since-mark vars. 1438 double proc_ms = elapsed_ms * (double) _parallel_gc_threads; 1439 if (elapsed_ms < MIN_TIMER_GRANULARITY) { 1440 // This usually happens due to the timer not having the required 1441 // granularity. Some Linuxes are the usual culprits. 1442 // We'll just set it to something (arbitrarily) small. 1443 proc_ms = 1.0; 1444 } 1445 double cur_efficiency = (double) freed_bytes / proc_ms; 1446 1447 bool new_in_marking_window = _in_marking_window; 1448 bool new_in_marking_window_im = false; 1449 if (during_initial_mark_pause()) { 1450 new_in_marking_window = true; 1451 new_in_marking_window_im = true; 1452 } 1453 1454 if (in_young_gc_mode()) { 1455 if (_last_full_young_gc) { 1456 set_full_young_gcs(false); 1457 _last_full_young_gc = false; 1458 } 1459 1460 if ( !_last_young_gc_full ) { 1461 if ( _should_revert_to_full_young_gcs || 1462 _known_garbage_ratio < 0.05 || 1463 (adaptive_young_list_length() && 1464 (get_gc_eff_factor() * cur_efficiency < predict_young_gc_eff())) ) { 1465 set_full_young_gcs(true); 1466 } 1467 } 1468 _should_revert_to_full_young_gcs = false; 1469 1470 if (_last_young_gc_full && !_during_marking) 1471 _young_gc_eff_seq->add(cur_efficiency); 1472 } 1473 1474 _short_lived_surv_rate_group->start_adding_regions(); 1475 // do that for any other surv rate groupsx 1476 1477 // <NEW PREDICTION> 1478 1479 if (update_stats) { 1480 double pause_time_ms = elapsed_ms; 1481 1482 size_t diff = 0; 1483 if (_max_pending_cards >= _pending_cards) 1484 diff = _max_pending_cards - _pending_cards; 1485 _pending_card_diff_seq->add((double) diff); 1486 1487 double cost_per_card_ms = 0.0; 1488 if (_pending_cards > 0) { 1489 cost_per_card_ms = update_rs_time / (double) _pending_cards; 1490 _cost_per_card_ms_seq->add(cost_per_card_ms); 1491 } 1492 1493 size_t cards_scanned = _g1->cards_scanned(); 1494 1495 double cost_per_entry_ms = 0.0; 1496 if (cards_scanned > 10) { 1497 cost_per_entry_ms = scan_rs_time / (double) cards_scanned; 1498 if (_last_young_gc_full) 1499 _cost_per_entry_ms_seq->add(cost_per_entry_ms); 1500 else 1501 _partially_young_cost_per_entry_ms_seq->add(cost_per_entry_ms); 1502 } 1503 1504 if (_max_rs_lengths > 0) { 1505 double cards_per_entry_ratio = 1506 (double) cards_scanned / (double) _max_rs_lengths; 1507 if (_last_young_gc_full) 1508 _fully_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio); 1509 else 1510 _partially_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio); 1511 } 1512 1513 size_t rs_length_diff = _max_rs_lengths - _recorded_rs_lengths; 1514 if (rs_length_diff >= 0) 1515 _rs_length_diff_seq->add((double) rs_length_diff); 1516 1517 size_t copied_bytes = surviving_bytes; 1518 double cost_per_byte_ms = 0.0; 1519 if (copied_bytes > 0) { 1520 cost_per_byte_ms = obj_copy_time / (double) copied_bytes; 1521 if (_in_marking_window) 1522 _cost_per_byte_ms_during_cm_seq->add(cost_per_byte_ms); 1523 else 1524 _cost_per_byte_ms_seq->add(cost_per_byte_ms); 1525 } 1526 1527 double all_other_time_ms = pause_time_ms - 1528 (update_rs_time + scan_rs_time + obj_copy_time + 1529 _mark_closure_time_ms + termination_time); 1530 1531 double young_other_time_ms = 0.0; 1532 if (_recorded_young_regions > 0) { 1533 young_other_time_ms = 1534 _recorded_young_cset_choice_time_ms + 1535 _recorded_young_free_cset_time_ms; 1536 _young_other_cost_per_region_ms_seq->add(young_other_time_ms / 1537 (double) _recorded_young_regions); 1538 } 1539 double non_young_other_time_ms = 0.0; 1540 if (_recorded_non_young_regions > 0) { 1541 non_young_other_time_ms = 1542 _recorded_non_young_cset_choice_time_ms + 1543 _recorded_non_young_free_cset_time_ms; 1544 1545 _non_young_other_cost_per_region_ms_seq->add(non_young_other_time_ms / 1546 (double) _recorded_non_young_regions); 1547 } 1548 1549 double constant_other_time_ms = all_other_time_ms - 1550 (young_other_time_ms + non_young_other_time_ms); 1551 _constant_other_time_ms_seq->add(constant_other_time_ms); 1552 1553 double survival_ratio = 0.0; 1554 if (_bytes_in_collection_set_before_gc > 0) { 1555 survival_ratio = (double) bytes_in_to_space_during_gc() / 1556 (double) _bytes_in_collection_set_before_gc; 1557 } 1558 1559 _pending_cards_seq->add((double) _pending_cards); 1560 _scanned_cards_seq->add((double) cards_scanned); 1561 _rs_lengths_seq->add((double) _max_rs_lengths); 1562 1563 double expensive_region_limit_ms = 1564 (double) MaxGCPauseMillis - predict_constant_other_time_ms(); 1565 if (expensive_region_limit_ms < 0.0) { 1566 // this means that the other time was predicted to be longer than 1567 // than the max pause time 1568 expensive_region_limit_ms = (double) MaxGCPauseMillis; 1569 } 1570 _expensive_region_limit_ms = expensive_region_limit_ms; 1571 1572 if (PREDICTIONS_VERBOSE) { 1573 gclog_or_tty->print_cr(""); 1574 gclog_or_tty->print_cr("PREDICTIONS %1.4lf %d " 1575 "REGIONS %d %d %d " 1576 "PENDING_CARDS %d %d " 1577 "CARDS_SCANNED %d %d " 1578 "RS_LENGTHS %d %d " 1579 "RS_UPDATE %1.6lf %1.6lf RS_SCAN %1.6lf %1.6lf " 1580 "SURVIVAL_RATIO %1.6lf %1.6lf " 1581 "OBJECT_COPY %1.6lf %1.6lf OTHER_CONSTANT %1.6lf %1.6lf " 1582 "OTHER_YOUNG %1.6lf %1.6lf " 1583 "OTHER_NON_YOUNG %1.6lf %1.6lf " 1584 "VTIME_DIFF %1.6lf TERMINATION %1.6lf " 1585 "ELAPSED %1.6lf %1.6lf ", 1586 _cur_collection_start_sec, 1587 (!_last_young_gc_full) ? 2 : 1588 (last_pause_included_initial_mark) ? 1 : 0, 1589 _recorded_region_num, 1590 _recorded_young_regions, 1591 _recorded_non_young_regions, 1592 _predicted_pending_cards, _pending_cards, 1593 _predicted_cards_scanned, cards_scanned, 1594 _predicted_rs_lengths, _max_rs_lengths, 1595 _predicted_rs_update_time_ms, update_rs_time, 1596 _predicted_rs_scan_time_ms, scan_rs_time, 1597 _predicted_survival_ratio, survival_ratio, 1598 _predicted_object_copy_time_ms, obj_copy_time, 1599 _predicted_constant_other_time_ms, constant_other_time_ms, 1600 _predicted_young_other_time_ms, young_other_time_ms, 1601 _predicted_non_young_other_time_ms, 1602 non_young_other_time_ms, 1603 _vtime_diff_ms, termination_time, 1604 _predicted_pause_time_ms, elapsed_ms); 1605 } 1606 1607 if (G1PolicyVerbose > 0) { 1608 gclog_or_tty->print_cr("Pause Time, predicted: %1.4lfms (predicted %s), actual: %1.4lfms", 1609 _predicted_pause_time_ms, 1610 (_within_target) ? "within" : "outside", 1611 elapsed_ms); 1612 } 1613 1614 } 1615 1616 _in_marking_window = new_in_marking_window; 1617 _in_marking_window_im = new_in_marking_window_im; 1618 _free_regions_at_end_of_collection = _g1->free_regions(); 1619 calculate_young_list_min_length(); 1620 calculate_young_list_target_length(); 1621 1622 // Note that _mmu_tracker->max_gc_time() returns the time in seconds. 1623 double update_rs_time_goal_ms = _mmu_tracker->max_gc_time() * MILLIUNITS * G1RSetUpdatingPauseTimePercent / 100.0; 1624 adjust_concurrent_refinement(update_rs_time, update_rs_processed_buffers, update_rs_time_goal_ms); 1625 // </NEW PREDICTION> 1626 } 1627 1628 // <NEW PREDICTION> 1629 1630 void G1CollectorPolicy::adjust_concurrent_refinement(double update_rs_time, 1631 double update_rs_processed_buffers, 1632 double goal_ms) { 1633 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); 1634 ConcurrentG1Refine *cg1r = G1CollectedHeap::heap()->concurrent_g1_refine(); 1635 1636 if (G1UseAdaptiveConcRefinement) { 1637 const int k_gy = 3, k_gr = 6; 1638 const double inc_k = 1.1, dec_k = 0.9; 1639 1640 int g = cg1r->green_zone(); 1641 if (update_rs_time > goal_ms) { 1642 g = (int)(g * dec_k); // Can become 0, that's OK. That would mean a mutator-only processing. 1643 } else { 1644 if (update_rs_time < goal_ms && update_rs_processed_buffers > g) { 1645 g = (int)MAX2(g * inc_k, g + 1.0); 1646 } 1647 } 1648 // Change the refinement threads params 1649 cg1r->set_green_zone(g); 1650 cg1r->set_yellow_zone(g * k_gy); 1651 cg1r->set_red_zone(g * k_gr); 1652 cg1r->reinitialize_threads(); 1653 1654 int processing_threshold_delta = MAX2((int)(cg1r->green_zone() * sigma()), 1); 1655 int processing_threshold = MIN2(cg1r->green_zone() + processing_threshold_delta, 1656 cg1r->yellow_zone()); 1657 // Change the barrier params 1658 dcqs.set_process_completed_threshold(processing_threshold); 1659 dcqs.set_max_completed_queue(cg1r->red_zone()); 1660 } 1661 1662 int curr_queue_size = dcqs.completed_buffers_num(); 1663 if (curr_queue_size >= cg1r->yellow_zone()) { 1664 dcqs.set_completed_queue_padding(curr_queue_size); 1665 } else { 1666 dcqs.set_completed_queue_padding(0); 1667 } 1668 dcqs.notify_if_necessary(); 1669 } 1670 1671 double 1672 G1CollectorPolicy:: 1673 predict_young_collection_elapsed_time_ms(size_t adjustment) { 1674 guarantee( adjustment == 0 || adjustment == 1, "invariant" ); 1675 1676 G1CollectedHeap* g1h = G1CollectedHeap::heap(); 1677 size_t young_num = g1h->young_list()->length(); 1678 if (young_num == 0) 1679 return 0.0; 1680 1681 young_num += adjustment; 1682 size_t pending_cards = predict_pending_cards(); 1683 size_t rs_lengths = g1h->young_list()->sampled_rs_lengths() + 1684 predict_rs_length_diff(); 1685 size_t card_num; 1686 if (full_young_gcs()) 1687 card_num = predict_young_card_num(rs_lengths); 1688 else 1689 card_num = predict_non_young_card_num(rs_lengths); 1690 size_t young_byte_size = young_num * HeapRegion::GrainBytes; 1691 double accum_yg_surv_rate = 1692 _short_lived_surv_rate_group->accum_surv_rate(adjustment); 1693 1694 size_t bytes_to_copy = 1695 (size_t) (accum_yg_surv_rate * (double) HeapRegion::GrainBytes); 1696 1697 return 1698 predict_rs_update_time_ms(pending_cards) + 1699 predict_rs_scan_time_ms(card_num) + 1700 predict_object_copy_time_ms(bytes_to_copy) + 1701 predict_young_other_time_ms(young_num) + 1702 predict_constant_other_time_ms(); 1703 } 1704 1705 double 1706 G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards) { 1707 size_t rs_length = predict_rs_length_diff(); 1708 size_t card_num; 1709 if (full_young_gcs()) 1710 card_num = predict_young_card_num(rs_length); 1711 else 1712 card_num = predict_non_young_card_num(rs_length); 1713 return predict_base_elapsed_time_ms(pending_cards, card_num); 1714 } 1715 1716 double 1717 G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards, 1718 size_t scanned_cards) { 1719 return 1720 predict_rs_update_time_ms(pending_cards) + 1721 predict_rs_scan_time_ms(scanned_cards) + 1722 predict_constant_other_time_ms(); 1723 } 1724 1725 double 1726 G1CollectorPolicy::predict_region_elapsed_time_ms(HeapRegion* hr, 1727 bool young) { 1728 size_t rs_length = hr->rem_set()->occupied(); 1729 size_t card_num; 1730 if (full_young_gcs()) 1731 card_num = predict_young_card_num(rs_length); 1732 else 1733 card_num = predict_non_young_card_num(rs_length); 1734 size_t bytes_to_copy = predict_bytes_to_copy(hr); 1735 1736 double region_elapsed_time_ms = 1737 predict_rs_scan_time_ms(card_num) + 1738 predict_object_copy_time_ms(bytes_to_copy); 1739 1740 if (young) 1741 region_elapsed_time_ms += predict_young_other_time_ms(1); 1742 else 1743 region_elapsed_time_ms += predict_non_young_other_time_ms(1); 1744 1745 return region_elapsed_time_ms; 1746 } 1747 1748 size_t 1749 G1CollectorPolicy::predict_bytes_to_copy(HeapRegion* hr) { 1750 size_t bytes_to_copy; 1751 if (hr->is_marked()) 1752 bytes_to_copy = hr->max_live_bytes(); 1753 else { 1754 guarantee( hr->is_young() && hr->age_in_surv_rate_group() != -1, 1755 "invariant" ); 1756 int age = hr->age_in_surv_rate_group(); 1757 double yg_surv_rate = predict_yg_surv_rate(age, hr->surv_rate_group()); 1758 bytes_to_copy = (size_t) ((double) hr->used() * yg_surv_rate); 1759 } 1760 1761 return bytes_to_copy; 1762 } 1763 1764 void 1765 G1CollectorPolicy::start_recording_regions() { 1766 _recorded_rs_lengths = 0; 1767 _recorded_young_regions = 0; 1768 _recorded_non_young_regions = 0; 1769 1770 #if PREDICTIONS_VERBOSE 1771 _recorded_marked_bytes = 0; 1772 _recorded_young_bytes = 0; 1773 _predicted_bytes_to_copy = 0; 1774 _predicted_rs_lengths = 0; 1775 _predicted_cards_scanned = 0; 1776 #endif // PREDICTIONS_VERBOSE 1777 } 1778 1779 void 1780 G1CollectorPolicy::record_cset_region_info(HeapRegion* hr, bool young) { 1781 #if PREDICTIONS_VERBOSE 1782 if (!young) { 1783 _recorded_marked_bytes += hr->max_live_bytes(); 1784 } 1785 _predicted_bytes_to_copy += predict_bytes_to_copy(hr); 1786 #endif // PREDICTIONS_VERBOSE 1787 1788 size_t rs_length = hr->rem_set()->occupied(); 1789 _recorded_rs_lengths += rs_length; 1790 } 1791 1792 void 1793 G1CollectorPolicy::record_non_young_cset_region(HeapRegion* hr) { 1794 assert(!hr->is_young(), "should not call this"); 1795 ++_recorded_non_young_regions; 1796 record_cset_region_info(hr, false); 1797 } 1798 1799 void 1800 G1CollectorPolicy::set_recorded_young_regions(size_t n_regions) { 1801 _recorded_young_regions = n_regions; 1802 } 1803 1804 void G1CollectorPolicy::set_recorded_young_bytes(size_t bytes) { 1805 #if PREDICTIONS_VERBOSE 1806 _recorded_young_bytes = bytes; 1807 #endif // PREDICTIONS_VERBOSE 1808 } 1809 1810 void G1CollectorPolicy::set_recorded_rs_lengths(size_t rs_lengths) { 1811 _recorded_rs_lengths = rs_lengths; 1812 } 1813 1814 void G1CollectorPolicy::set_predicted_bytes_to_copy(size_t bytes) { 1815 _predicted_bytes_to_copy = bytes; 1816 } 1817 1818 void 1819 G1CollectorPolicy::end_recording_regions() { 1820 // The _predicted_pause_time_ms field is referenced in code 1821 // not under PREDICTIONS_VERBOSE. Let's initialize it. 1822 _predicted_pause_time_ms = -1.0; 1823 1824 #if PREDICTIONS_VERBOSE 1825 _predicted_pending_cards = predict_pending_cards(); 1826 _predicted_rs_lengths = _recorded_rs_lengths + predict_rs_length_diff(); 1827 if (full_young_gcs()) 1828 _predicted_cards_scanned += predict_young_card_num(_predicted_rs_lengths); 1829 else 1830 _predicted_cards_scanned += 1831 predict_non_young_card_num(_predicted_rs_lengths); 1832 _recorded_region_num = _recorded_young_regions + _recorded_non_young_regions; 1833 1834 _predicted_rs_update_time_ms = 1835 predict_rs_update_time_ms(_g1->pending_card_num()); 1836 _predicted_rs_scan_time_ms = 1837 predict_rs_scan_time_ms(_predicted_cards_scanned); 1838 _predicted_object_copy_time_ms = 1839 predict_object_copy_time_ms(_predicted_bytes_to_copy); 1840 _predicted_constant_other_time_ms = 1841 predict_constant_other_time_ms(); 1842 _predicted_young_other_time_ms = 1843 predict_young_other_time_ms(_recorded_young_regions); 1844 _predicted_non_young_other_time_ms = 1845 predict_non_young_other_time_ms(_recorded_non_young_regions); 1846 1847 _predicted_pause_time_ms = 1848 _predicted_rs_update_time_ms + 1849 _predicted_rs_scan_time_ms + 1850 _predicted_object_copy_time_ms + 1851 _predicted_constant_other_time_ms + 1852 _predicted_young_other_time_ms + 1853 _predicted_non_young_other_time_ms; 1854 #endif // PREDICTIONS_VERBOSE 1855 } 1856 1857 void G1CollectorPolicy::check_if_region_is_too_expensive(double 1858 predicted_time_ms) { 1859 // I don't think we need to do this when in young GC mode since 1860 // marking will be initiated next time we hit the soft limit anyway... 1861 if (predicted_time_ms > _expensive_region_limit_ms) { 1862 if (!in_young_gc_mode()) { 1863 set_full_young_gcs(true); 1864 // We might want to do something different here. However, 1865 // right now we don't support the non-generational G1 mode 1866 // (and in fact we are planning to remove the associated code, 1867 // see CR 6814390). So, let's leave it as is and this will be 1868 // removed some time in the future 1869 ShouldNotReachHere(); 1870 set_during_initial_mark_pause(); 1871 } else 1872 // no point in doing another partial one 1873 _should_revert_to_full_young_gcs = true; 1874 } 1875 } 1876 1877 // </NEW PREDICTION> 1878 1879 1880 void G1CollectorPolicy::update_recent_gc_times(double end_time_sec, 1881 double elapsed_ms) { 1882 _recent_gc_times_ms->add(elapsed_ms); 1883 _recent_prev_end_times_for_all_gcs_sec->add(end_time_sec); 1884 _prev_collection_pause_end_ms = end_time_sec * 1000.0; 1885 } 1886 1887 double G1CollectorPolicy::recent_avg_time_for_pauses_ms() { 1888 if (_recent_pause_times_ms->num() == 0) return (double) MaxGCPauseMillis; 1889 else return _recent_pause_times_ms->avg(); 1890 } 1891 1892 double G1CollectorPolicy::recent_avg_time_for_CH_strong_ms() { 1893 if (_recent_CH_strong_roots_times_ms->num() == 0) 1894 return (double)MaxGCPauseMillis/3.0; 1895 else return _recent_CH_strong_roots_times_ms->avg(); 1896 } 1897 1898 double G1CollectorPolicy::recent_avg_time_for_G1_strong_ms() { 1899 if (_recent_G1_strong_roots_times_ms->num() == 0) 1900 return (double)MaxGCPauseMillis/3.0; 1901 else return _recent_G1_strong_roots_times_ms->avg(); 1902 } 1903 1904 double G1CollectorPolicy::recent_avg_time_for_evac_ms() { 1905 if (_recent_evac_times_ms->num() == 0) return (double)MaxGCPauseMillis/3.0; 1906 else return _recent_evac_times_ms->avg(); 1907 } 1908 1909 int G1CollectorPolicy::number_of_recent_gcs() { 1910 assert(_recent_CH_strong_roots_times_ms->num() == 1911 _recent_G1_strong_roots_times_ms->num(), "Sequence out of sync"); 1912 assert(_recent_G1_strong_roots_times_ms->num() == 1913 _recent_evac_times_ms->num(), "Sequence out of sync"); 1914 assert(_recent_evac_times_ms->num() == 1915 _recent_pause_times_ms->num(), "Sequence out of sync"); 1916 assert(_recent_pause_times_ms->num() == 1917 _recent_CS_bytes_used_before->num(), "Sequence out of sync"); 1918 assert(_recent_CS_bytes_used_before->num() == 1919 _recent_CS_bytes_surviving->num(), "Sequence out of sync"); 1920 return _recent_pause_times_ms->num(); 1921 } 1922 1923 double G1CollectorPolicy::recent_avg_survival_fraction() { 1924 return recent_avg_survival_fraction_work(_recent_CS_bytes_surviving, 1925 _recent_CS_bytes_used_before); 1926 } 1927 1928 double G1CollectorPolicy::last_survival_fraction() { 1929 return last_survival_fraction_work(_recent_CS_bytes_surviving, 1930 _recent_CS_bytes_used_before); 1931 } 1932 1933 double 1934 G1CollectorPolicy::recent_avg_survival_fraction_work(TruncatedSeq* surviving, 1935 TruncatedSeq* before) { 1936 assert(surviving->num() == before->num(), "Sequence out of sync"); 1937 if (before->sum() > 0.0) { 1938 double recent_survival_rate = surviving->sum() / before->sum(); 1939 // We exempt parallel collection from this check because Alloc Buffer 1940 // fragmentation can produce negative collections. 1941 // Further, we're now always doing parallel collection. But I'm still 1942 // leaving this here as a placeholder for a more precise assertion later. 1943 // (DLD, 10/05.) 1944 assert((true || ParallelGCThreads > 0) || 1945 _g1->evacuation_failed() || 1946 recent_survival_rate <= 1.0, "Or bad frac"); 1947 return recent_survival_rate; 1948 } else { 1949 return 1.0; // Be conservative. 1950 } 1951 } 1952 1953 double 1954 G1CollectorPolicy::last_survival_fraction_work(TruncatedSeq* surviving, 1955 TruncatedSeq* before) { 1956 assert(surviving->num() == before->num(), "Sequence out of sync"); 1957 if (surviving->num() > 0 && before->last() > 0.0) { 1958 double last_survival_rate = surviving->last() / before->last(); 1959 // We exempt parallel collection from this check because Alloc Buffer 1960 // fragmentation can produce negative collections. 1961 // Further, we're now always doing parallel collection. But I'm still 1962 // leaving this here as a placeholder for a more precise assertion later. 1963 // (DLD, 10/05.) 1964 assert((true || ParallelGCThreads > 0) || 1965 last_survival_rate <= 1.0, "Or bad frac"); 1966 return last_survival_rate; 1967 } else { 1968 return 1.0; 1969 } 1970 } 1971 1972 static const int survival_min_obs = 5; 1973 static double survival_min_obs_limits[] = { 0.9, 0.7, 0.5, 0.3, 0.1 }; 1974 static const double min_survival_rate = 0.1; 1975 1976 double 1977 G1CollectorPolicy::conservative_avg_survival_fraction_work(double avg, 1978 double latest) { 1979 double res = avg; 1980 if (number_of_recent_gcs() < survival_min_obs) { 1981 res = MAX2(res, survival_min_obs_limits[number_of_recent_gcs()]); 1982 } 1983 res = MAX2(res, latest); 1984 res = MAX2(res, min_survival_rate); 1985 // In the parallel case, LAB fragmentation can produce "negative 1986 // collections"; so can evac failure. Cap at 1.0 1987 res = MIN2(res, 1.0); 1988 return res; 1989 } 1990 1991 size_t G1CollectorPolicy::expansion_amount() { 1992 if ((recent_avg_pause_time_ratio() * 100.0) > _gc_overhead_perc) { 1993 // We will double the existing space, or take 1994 // G1ExpandByPercentOfAvailable % of the available expansion 1995 // space, whichever is smaller, bounded below by a minimum 1996 // expansion (unless that's all that's left.) 1997 const size_t min_expand_bytes = 1*M; 1998 size_t reserved_bytes = _g1->g1_reserved_obj_bytes(); 1999 size_t committed_bytes = _g1->capacity(); 2000 size_t uncommitted_bytes = reserved_bytes - committed_bytes; 2001 size_t expand_bytes; 2002 size_t expand_bytes_via_pct = 2003 uncommitted_bytes * G1ExpandByPercentOfAvailable / 100; 2004 expand_bytes = MIN2(expand_bytes_via_pct, committed_bytes); 2005 expand_bytes = MAX2(expand_bytes, min_expand_bytes); 2006 expand_bytes = MIN2(expand_bytes, uncommitted_bytes); 2007 if (G1PolicyVerbose > 1) { 2008 gclog_or_tty->print("Decided to expand: ratio = %5.2f, " 2009 "committed = %d%s, uncommited = %d%s, via pct = %d%s.\n" 2010 " Answer = %d.\n", 2011 recent_avg_pause_time_ratio(), 2012 byte_size_in_proper_unit(committed_bytes), 2013 proper_unit_for_byte_size(committed_bytes), 2014 byte_size_in_proper_unit(uncommitted_bytes), 2015 proper_unit_for_byte_size(uncommitted_bytes), 2016 byte_size_in_proper_unit(expand_bytes_via_pct), 2017 proper_unit_for_byte_size(expand_bytes_via_pct), 2018 byte_size_in_proper_unit(expand_bytes), 2019 proper_unit_for_byte_size(expand_bytes)); 2020 } 2021 return expand_bytes; 2022 } else { 2023 return 0; 2024 } 2025 } 2026 2027 void G1CollectorPolicy::note_start_of_mark_thread() { 2028 _mark_thread_startup_sec = os::elapsedTime(); 2029 } 2030 2031 class CountCSClosure: public HeapRegionClosure { 2032 G1CollectorPolicy* _g1_policy; 2033 public: 2034 CountCSClosure(G1CollectorPolicy* g1_policy) : 2035 _g1_policy(g1_policy) {} 2036 bool doHeapRegion(HeapRegion* r) { 2037 _g1_policy->_bytes_in_collection_set_before_gc += r->used(); 2038 return false; 2039 } 2040 }; 2041 2042 void G1CollectorPolicy::count_CS_bytes_used() { 2043 CountCSClosure cs_closure(this); 2044 _g1->collection_set_iterate(&cs_closure); 2045 } 2046 2047 static void print_indent(int level) { 2048 for (int j = 0; j < level+1; ++j) 2049 gclog_or_tty->print(" "); 2050 } 2051 2052 void G1CollectorPolicy::print_summary (int level, 2053 const char* str, 2054 NumberSeq* seq) const { 2055 double sum = seq->sum(); 2056 print_indent(level); 2057 gclog_or_tty->print_cr("%-24s = %8.2lf s (avg = %8.2lf ms)", 2058 str, sum / 1000.0, seq->avg()); 2059 } 2060 2061 void G1CollectorPolicy::print_summary_sd (int level, 2062 const char* str, 2063 NumberSeq* seq) const { 2064 print_summary(level, str, seq); 2065 print_indent(level + 5); 2066 gclog_or_tty->print_cr("(num = %5d, std dev = %8.2lf ms, max = %8.2lf ms)", 2067 seq->num(), seq->sd(), seq->maximum()); 2068 } 2069 2070 void G1CollectorPolicy::check_other_times(int level, 2071 NumberSeq* other_times_ms, 2072 NumberSeq* calc_other_times_ms) const { 2073 bool should_print = false; 2074 2075 double max_sum = MAX2(fabs(other_times_ms->sum()), 2076 fabs(calc_other_times_ms->sum())); 2077 double min_sum = MIN2(fabs(other_times_ms->sum()), 2078 fabs(calc_other_times_ms->sum())); 2079 double sum_ratio = max_sum / min_sum; 2080 if (sum_ratio > 1.1) { 2081 should_print = true; 2082 print_indent(level + 1); 2083 gclog_or_tty->print_cr("## CALCULATED OTHER SUM DOESN'T MATCH RECORDED ###"); 2084 } 2085 2086 double max_avg = MAX2(fabs(other_times_ms->avg()), 2087 fabs(calc_other_times_ms->avg())); 2088 double min_avg = MIN2(fabs(other_times_ms->avg()), 2089 fabs(calc_other_times_ms->avg())); 2090 double avg_ratio = max_avg / min_avg; 2091 if (avg_ratio > 1.1) { 2092 should_print = true; 2093 print_indent(level + 1); 2094 gclog_or_tty->print_cr("## CALCULATED OTHER AVG DOESN'T MATCH RECORDED ###"); 2095 } 2096 2097 if (other_times_ms->sum() < -0.01) { 2098 print_indent(level + 1); 2099 gclog_or_tty->print_cr("## RECORDED OTHER SUM IS NEGATIVE ###"); 2100 } 2101 2102 if (other_times_ms->avg() < -0.01) { 2103 print_indent(level + 1); 2104 gclog_or_tty->print_cr("## RECORDED OTHER AVG IS NEGATIVE ###"); 2105 } 2106 2107 if (calc_other_times_ms->sum() < -0.01) { 2108 should_print = true; 2109 print_indent(level + 1); 2110 gclog_or_tty->print_cr("## CALCULATED OTHER SUM IS NEGATIVE ###"); 2111 } 2112 2113 if (calc_other_times_ms->avg() < -0.01) { 2114 should_print = true; 2115 print_indent(level + 1); 2116 gclog_or_tty->print_cr("## CALCULATED OTHER AVG IS NEGATIVE ###"); 2117 } 2118 2119 if (should_print) 2120 print_summary(level, "Other(Calc)", calc_other_times_ms); 2121 } 2122 2123 void G1CollectorPolicy::print_summary(PauseSummary* summary) const { 2124 bool parallel = ParallelGCThreads > 0; 2125 MainBodySummary* body_summary = summary->main_body_summary(); 2126 if (summary->get_total_seq()->num() > 0) { 2127 print_summary_sd(0, "Evacuation Pauses", summary->get_total_seq()); 2128 if (body_summary != NULL) { 2129 print_summary(1, "SATB Drain", body_summary->get_satb_drain_seq()); 2130 if (parallel) { 2131 print_summary(1, "Parallel Time", body_summary->get_parallel_seq()); 2132 print_summary(2, "Update RS", body_summary->get_update_rs_seq()); 2133 print_summary(2, "Ext Root Scanning", 2134 body_summary->get_ext_root_scan_seq()); 2135 print_summary(2, "Mark Stack Scanning", 2136 body_summary->get_mark_stack_scan_seq()); 2137 print_summary(2, "Scan RS", body_summary->get_scan_rs_seq()); 2138 print_summary(2, "Object Copy", body_summary->get_obj_copy_seq()); 2139 print_summary(2, "Termination", body_summary->get_termination_seq()); 2140 print_summary(2, "Other", body_summary->get_parallel_other_seq()); 2141 { 2142 NumberSeq* other_parts[] = { 2143 body_summary->get_update_rs_seq(), 2144 body_summary->get_ext_root_scan_seq(), 2145 body_summary->get_mark_stack_scan_seq(), 2146 body_summary->get_scan_rs_seq(), 2147 body_summary->get_obj_copy_seq(), 2148 body_summary->get_termination_seq() 2149 }; 2150 NumberSeq calc_other_times_ms(body_summary->get_parallel_seq(), 2151 6, other_parts); 2152 check_other_times(2, body_summary->get_parallel_other_seq(), 2153 &calc_other_times_ms); 2154 } 2155 print_summary(1, "Mark Closure", body_summary->get_mark_closure_seq()); 2156 print_summary(1, "Clear CT", body_summary->get_clear_ct_seq()); 2157 } else { 2158 print_summary(1, "Update RS", body_summary->get_update_rs_seq()); 2159 print_summary(1, "Ext Root Scanning", 2160 body_summary->get_ext_root_scan_seq()); 2161 print_summary(1, "Mark Stack Scanning", 2162 body_summary->get_mark_stack_scan_seq()); 2163 print_summary(1, "Scan RS", body_summary->get_scan_rs_seq()); 2164 print_summary(1, "Object Copy", body_summary->get_obj_copy_seq()); 2165 } 2166 } 2167 print_summary(1, "Other", summary->get_other_seq()); 2168 { 2169 if (body_summary != NULL) { 2170 NumberSeq calc_other_times_ms; 2171 if (parallel) { 2172 // parallel 2173 NumberSeq* other_parts[] = { 2174 body_summary->get_satb_drain_seq(), 2175 body_summary->get_parallel_seq(), 2176 body_summary->get_clear_ct_seq() 2177 }; 2178 calc_other_times_ms = NumberSeq(summary->get_total_seq(), 2179 3, other_parts); 2180 } else { 2181 // serial 2182 NumberSeq* other_parts[] = { 2183 body_summary->get_satb_drain_seq(), 2184 body_summary->get_update_rs_seq(), 2185 body_summary->get_ext_root_scan_seq(), 2186 body_summary->get_mark_stack_scan_seq(), 2187 body_summary->get_scan_rs_seq(), 2188 body_summary->get_obj_copy_seq() 2189 }; 2190 calc_other_times_ms = NumberSeq(summary->get_total_seq(), 2191 6, other_parts); 2192 } 2193 check_other_times(1, summary->get_other_seq(), &calc_other_times_ms); 2194 } 2195 } 2196 } else { 2197 print_indent(0); 2198 gclog_or_tty->print_cr("none"); 2199 } 2200 gclog_or_tty->print_cr(""); 2201 } 2202 2203 void G1CollectorPolicy::print_tracing_info() const { 2204 if (TraceGen0Time) { 2205 gclog_or_tty->print_cr("ALL PAUSES"); 2206 print_summary_sd(0, "Total", _all_pause_times_ms); 2207 gclog_or_tty->print_cr(""); 2208 gclog_or_tty->print_cr(""); 2209 gclog_or_tty->print_cr(" Full Young GC Pauses: %8d", _full_young_pause_num); 2210 gclog_or_tty->print_cr(" Partial Young GC Pauses: %8d", _partial_young_pause_num); 2211 gclog_or_tty->print_cr(""); 2212 2213 gclog_or_tty->print_cr("EVACUATION PAUSES"); 2214 print_summary(_summary); 2215 2216 gclog_or_tty->print_cr("MISC"); 2217 print_summary_sd(0, "Stop World", _all_stop_world_times_ms); 2218 print_summary_sd(0, "Yields", _all_yield_times_ms); 2219 for (int i = 0; i < _aux_num; ++i) { 2220 if (_all_aux_times_ms[i].num() > 0) { 2221 char buffer[96]; 2222 sprintf(buffer, "Aux%d", i); 2223 print_summary_sd(0, buffer, &_all_aux_times_ms[i]); 2224 } 2225 } 2226 2227 size_t all_region_num = _region_num_young + _region_num_tenured; 2228 gclog_or_tty->print_cr(" New Regions %8d, Young %8d (%6.2lf%%), " 2229 "Tenured %8d (%6.2lf%%)", 2230 all_region_num, 2231 _region_num_young, 2232 (double) _region_num_young / (double) all_region_num * 100.0, 2233 _region_num_tenured, 2234 (double) _region_num_tenured / (double) all_region_num * 100.0); 2235 } 2236 if (TraceGen1Time) { 2237 if (_all_full_gc_times_ms->num() > 0) { 2238 gclog_or_tty->print("\n%4d full_gcs: total time = %8.2f s", 2239 _all_full_gc_times_ms->num(), 2240 _all_full_gc_times_ms->sum() / 1000.0); 2241 gclog_or_tty->print_cr(" (avg = %8.2fms).", _all_full_gc_times_ms->avg()); 2242 gclog_or_tty->print_cr(" [std. dev = %8.2f ms, max = %8.2f ms]", 2243 _all_full_gc_times_ms->sd(), 2244 _all_full_gc_times_ms->maximum()); 2245 } 2246 } 2247 } 2248 2249 void G1CollectorPolicy::print_yg_surv_rate_info() const { 2250 #ifndef PRODUCT 2251 _short_lived_surv_rate_group->print_surv_rate_summary(); 2252 // add this call for any other surv rate groups 2253 #endif // PRODUCT 2254 } 2255 2256 bool 2257 G1CollectorPolicy::should_add_next_region_to_young_list() { 2258 assert(in_young_gc_mode(), "should be in young GC mode"); 2259 bool ret; 2260 size_t young_list_length = _g1->young_list()->length(); 2261 size_t young_list_max_length = _young_list_target_length; 2262 if (G1FixedEdenSize) { 2263 young_list_max_length -= _max_survivor_regions; 2264 } 2265 if (young_list_length < young_list_max_length) { 2266 ret = true; 2267 ++_region_num_young; 2268 } else { 2269 ret = false; 2270 ++_region_num_tenured; 2271 } 2272 2273 return ret; 2274 } 2275 2276 #ifndef PRODUCT 2277 // for debugging, bit of a hack... 2278 static char* 2279 region_num_to_mbs(int length) { 2280 static char buffer[64]; 2281 double bytes = (double) (length * HeapRegion::GrainBytes); 2282 double mbs = bytes / (double) (1024 * 1024); 2283 sprintf(buffer, "%7.2lfMB", mbs); 2284 return buffer; 2285 } 2286 #endif // PRODUCT 2287 2288 size_t G1CollectorPolicy::max_regions(int purpose) { 2289 switch (purpose) { 2290 case GCAllocForSurvived: 2291 return _max_survivor_regions; 2292 case GCAllocForTenured: 2293 return REGIONS_UNLIMITED; 2294 default: 2295 ShouldNotReachHere(); 2296 return REGIONS_UNLIMITED; 2297 }; 2298 } 2299 2300 // Calculates survivor space parameters. 2301 void G1CollectorPolicy::calculate_survivors_policy() 2302 { 2303 if (G1FixedSurvivorSpaceSize == 0) { 2304 _max_survivor_regions = _young_list_target_length / SurvivorRatio; 2305 } else { 2306 _max_survivor_regions = G1FixedSurvivorSpaceSize / HeapRegion::GrainBytes; 2307 } 2308 2309 if (G1FixedTenuringThreshold) { 2310 _tenuring_threshold = MaxTenuringThreshold; 2311 } else { 2312 _tenuring_threshold = _survivors_age_table.compute_tenuring_threshold( 2313 HeapRegion::GrainWords * _max_survivor_regions); 2314 } 2315 } 2316 2317 bool 2318 G1CollectorPolicy_BestRegionsFirst::should_do_collection_pause(size_t 2319 word_size) { 2320 assert(_g1->regions_accounted_for(), "Region leakage!"); 2321 double max_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0; 2322 2323 size_t young_list_length = _g1->young_list()->length(); 2324 size_t young_list_max_length = _young_list_target_length; 2325 if (G1FixedEdenSize) { 2326 young_list_max_length -= _max_survivor_regions; 2327 } 2328 bool reached_target_length = young_list_length >= young_list_max_length; 2329 2330 if (in_young_gc_mode()) { 2331 if (reached_target_length) { 2332 assert( young_list_length > 0 && _g1->young_list()->length() > 0, 2333 "invariant" ); 2334 return true; 2335 } 2336 } else { 2337 guarantee( false, "should not reach here" ); 2338 } 2339 2340 return false; 2341 } 2342 2343 #ifndef PRODUCT 2344 class HRSortIndexIsOKClosure: public HeapRegionClosure { 2345 CollectionSetChooser* _chooser; 2346 public: 2347 HRSortIndexIsOKClosure(CollectionSetChooser* chooser) : 2348 _chooser(chooser) {} 2349 2350 bool doHeapRegion(HeapRegion* r) { 2351 if (!r->continuesHumongous()) { 2352 assert(_chooser->regionProperlyOrdered(r), "Ought to be."); 2353 } 2354 return false; 2355 } 2356 }; 2357 2358 bool G1CollectorPolicy_BestRegionsFirst::assertMarkedBytesDataOK() { 2359 HRSortIndexIsOKClosure cl(_collectionSetChooser); 2360 _g1->heap_region_iterate(&cl); 2361 return true; 2362 } 2363 #endif 2364 2365 bool 2366 G1CollectorPolicy::force_initial_mark_if_outside_cycle() { 2367 bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle(); 2368 if (!during_cycle) { 2369 set_initiate_conc_mark_if_possible(); 2370 return true; 2371 } else { 2372 return false; 2373 } 2374 } 2375 2376 void 2377 G1CollectorPolicy::decide_on_conc_mark_initiation() { 2378 // We are about to decide on whether this pause will be an 2379 // initial-mark pause. 2380 2381 // First, during_initial_mark_pause() should not be already set. We 2382 // will set it here if we have to. However, it should be cleared by 2383 // the end of the pause (it's only set for the duration of an 2384 // initial-mark pause). 2385 assert(!during_initial_mark_pause(), "pre-condition"); 2386 2387 if (initiate_conc_mark_if_possible()) { 2388 // We had noticed on a previous pause that the heap occupancy has 2389 // gone over the initiating threshold and we should start a 2390 // concurrent marking cycle. So we might initiate one. 2391 2392 bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle(); 2393 if (!during_cycle) { 2394 // The concurrent marking thread is not "during a cycle", i.e., 2395 // it has completed the last one. So we can go ahead and 2396 // initiate a new cycle. 2397 2398 set_during_initial_mark_pause(); 2399 2400 // And we can now clear initiate_conc_mark_if_possible() as 2401 // we've already acted on it. 2402 clear_initiate_conc_mark_if_possible(); 2403 } else { 2404 // The concurrent marking thread is still finishing up the 2405 // previous cycle. If we start one right now the two cycles 2406 // overlap. In particular, the concurrent marking thread might 2407 // be in the process of clearing the next marking bitmap (which 2408 // we will use for the next cycle if we start one). Starting a 2409 // cycle now will be bad given that parts of the marking 2410 // information might get cleared by the marking thread. And we 2411 // cannot wait for the marking thread to finish the cycle as it 2412 // periodically yields while clearing the next marking bitmap 2413 // and, if it's in a yield point, it's waiting for us to 2414 // finish. So, at this point we will not start a cycle and we'll 2415 // let the concurrent marking thread complete the last one. 2416 } 2417 } 2418 } 2419 2420 void 2421 G1CollectorPolicy_BestRegionsFirst:: 2422 record_collection_pause_start(double start_time_sec, size_t start_used) { 2423 G1CollectorPolicy::record_collection_pause_start(start_time_sec, start_used); 2424 } 2425 2426 class NextNonCSElemFinder: public HeapRegionClosure { 2427 HeapRegion* _res; 2428 public: 2429 NextNonCSElemFinder(): _res(NULL) {} 2430 bool doHeapRegion(HeapRegion* r) { 2431 if (!r->in_collection_set()) { 2432 _res = r; 2433 return true; 2434 } else { 2435 return false; 2436 } 2437 } 2438 HeapRegion* res() { return _res; } 2439 }; 2440 2441 class KnownGarbageClosure: public HeapRegionClosure { 2442 CollectionSetChooser* _hrSorted; 2443 2444 public: 2445 KnownGarbageClosure(CollectionSetChooser* hrSorted) : 2446 _hrSorted(hrSorted) 2447 {} 2448 2449 bool doHeapRegion(HeapRegion* r) { 2450 // We only include humongous regions in collection 2451 // sets when concurrent mark shows that their contained object is 2452 // unreachable. 2453 2454 // Do we have any marking information for this region? 2455 if (r->is_marked()) { 2456 // We don't include humongous regions in collection 2457 // sets because we collect them immediately at the end of a marking 2458 // cycle. We also don't include young regions because we *must* 2459 // include them in the next collection pause. 2460 if (!r->isHumongous() && !r->is_young()) { 2461 _hrSorted->addMarkedHeapRegion(r); 2462 } 2463 } 2464 return false; 2465 } 2466 }; 2467 2468 class ParKnownGarbageHRClosure: public HeapRegionClosure { 2469 CollectionSetChooser* _hrSorted; 2470 jint _marked_regions_added; 2471 jint _chunk_size; 2472 jint _cur_chunk_idx; 2473 jint _cur_chunk_end; // Cur chunk [_cur_chunk_idx, _cur_chunk_end) 2474 int _worker; 2475 int _invokes; 2476 2477 void get_new_chunk() { 2478 _cur_chunk_idx = _hrSorted->getParMarkedHeapRegionChunk(_chunk_size); 2479 _cur_chunk_end = _cur_chunk_idx + _chunk_size; 2480 } 2481 void add_region(HeapRegion* r) { 2482 if (_cur_chunk_idx == _cur_chunk_end) { 2483 get_new_chunk(); 2484 } 2485 assert(_cur_chunk_idx < _cur_chunk_end, "postcondition"); 2486 _hrSorted->setMarkedHeapRegion(_cur_chunk_idx, r); 2487 _marked_regions_added++; 2488 _cur_chunk_idx++; 2489 } 2490 2491 public: 2492 ParKnownGarbageHRClosure(CollectionSetChooser* hrSorted, 2493 jint chunk_size, 2494 int worker) : 2495 _hrSorted(hrSorted), _chunk_size(chunk_size), _worker(worker), 2496 _marked_regions_added(0), _cur_chunk_idx(0), _cur_chunk_end(0), 2497 _invokes(0) 2498 {} 2499 2500 bool doHeapRegion(HeapRegion* r) { 2501 // We only include humongous regions in collection 2502 // sets when concurrent mark shows that their contained object is 2503 // unreachable. 2504 _invokes++; 2505 2506 // Do we have any marking information for this region? 2507 if (r->is_marked()) { 2508 // We don't include humongous regions in collection 2509 // sets because we collect them immediately at the end of a marking 2510 // cycle. 2511 // We also do not include young regions in collection sets 2512 if (!r->isHumongous() && !r->is_young()) { 2513 add_region(r); 2514 } 2515 } 2516 return false; 2517 } 2518 jint marked_regions_added() { return _marked_regions_added; } 2519 int invokes() { return _invokes; } 2520 }; 2521 2522 class ParKnownGarbageTask: public AbstractGangTask { 2523 CollectionSetChooser* _hrSorted; 2524 jint _chunk_size; 2525 G1CollectedHeap* _g1; 2526 public: 2527 ParKnownGarbageTask(CollectionSetChooser* hrSorted, jint chunk_size) : 2528 AbstractGangTask("ParKnownGarbageTask"), 2529 _hrSorted(hrSorted), _chunk_size(chunk_size), 2530 _g1(G1CollectedHeap::heap()) 2531 {} 2532 2533 void work(int i) { 2534 ParKnownGarbageHRClosure parKnownGarbageCl(_hrSorted, _chunk_size, i); 2535 // Back to zero for the claim value. 2536 _g1->heap_region_par_iterate_chunked(&parKnownGarbageCl, i, 2537 HeapRegion::InitialClaimValue); 2538 jint regions_added = parKnownGarbageCl.marked_regions_added(); 2539 _hrSorted->incNumMarkedHeapRegions(regions_added); 2540 if (G1PrintParCleanupStats) { 2541 gclog_or_tty->print(" Thread %d called %d times, added %d regions to list.\n", 2542 i, parKnownGarbageCl.invokes(), regions_added); 2543 } 2544 } 2545 }; 2546 2547 void 2548 G1CollectorPolicy_BestRegionsFirst:: 2549 record_concurrent_mark_cleanup_end(size_t freed_bytes, 2550 size_t max_live_bytes) { 2551 double start; 2552 if (G1PrintParCleanupStats) start = os::elapsedTime(); 2553 record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes); 2554 2555 _collectionSetChooser->clearMarkedHeapRegions(); 2556 double clear_marked_end; 2557 if (G1PrintParCleanupStats) { 2558 clear_marked_end = os::elapsedTime(); 2559 gclog_or_tty->print_cr(" clear marked regions + work1: %8.3f ms.", 2560 (clear_marked_end - start)*1000.0); 2561 } 2562 if (ParallelGCThreads > 0) { 2563 const size_t OverpartitionFactor = 4; 2564 const size_t MinWorkUnit = 8; 2565 const size_t WorkUnit = 2566 MAX2(_g1->n_regions() / (ParallelGCThreads * OverpartitionFactor), 2567 MinWorkUnit); 2568 _collectionSetChooser->prepareForAddMarkedHeapRegionsPar(_g1->n_regions(), 2569 WorkUnit); 2570 ParKnownGarbageTask parKnownGarbageTask(_collectionSetChooser, 2571 (int) WorkUnit); 2572 _g1->workers()->run_task(&parKnownGarbageTask); 2573 2574 assert(_g1->check_heap_region_claim_values(HeapRegion::InitialClaimValue), 2575 "sanity check"); 2576 } else { 2577 KnownGarbageClosure knownGarbagecl(_collectionSetChooser); 2578 _g1->heap_region_iterate(&knownGarbagecl); 2579 } 2580 double known_garbage_end; 2581 if (G1PrintParCleanupStats) { 2582 known_garbage_end = os::elapsedTime(); 2583 gclog_or_tty->print_cr(" compute known garbage: %8.3f ms.", 2584 (known_garbage_end - clear_marked_end)*1000.0); 2585 } 2586 _collectionSetChooser->sortMarkedHeapRegions(); 2587 double sort_end; 2588 if (G1PrintParCleanupStats) { 2589 sort_end = os::elapsedTime(); 2590 gclog_or_tty->print_cr(" sorting: %8.3f ms.", 2591 (sort_end - known_garbage_end)*1000.0); 2592 } 2593 2594 record_concurrent_mark_cleanup_end_work2(); 2595 double work2_end; 2596 if (G1PrintParCleanupStats) { 2597 work2_end = os::elapsedTime(); 2598 gclog_or_tty->print_cr(" work2: %8.3f ms.", 2599 (work2_end - sort_end)*1000.0); 2600 } 2601 } 2602 2603 // Add the heap region at the head of the non-incremental collection set 2604 void G1CollectorPolicy:: 2605 add_to_collection_set(HeapRegion* hr) { 2606 assert(_inc_cset_build_state == Active, "Precondition"); 2607 assert(!hr->is_young(), "non-incremental add of young region"); 2608 2609 if (G1PrintHeapRegions) { 2610 gclog_or_tty->print_cr("added region to cset " 2611 "%d:["PTR_FORMAT", "PTR_FORMAT"], " 2612 "top "PTR_FORMAT", %s", 2613 hr->hrs_index(), hr->bottom(), hr->end(), 2614 hr->top(), hr->is_young() ? "YOUNG" : "NOT_YOUNG"); 2615 } 2616 2617 if (_g1->mark_in_progress()) 2618 _g1->concurrent_mark()->registerCSetRegion(hr); 2619 2620 assert(!hr->in_collection_set(), "should not already be in the CSet"); 2621 hr->set_in_collection_set(true); 2622 hr->set_next_in_collection_set(_collection_set); 2623 _collection_set = hr; 2624 _collection_set_size++; 2625 _collection_set_bytes_used_before += hr->used(); 2626 _g1->register_region_with_in_cset_fast_test(hr); 2627 } 2628 2629 // Initialize the per-collection-set information 2630 void G1CollectorPolicy::start_incremental_cset_building() { 2631 assert(_inc_cset_build_state == Inactive, "Precondition"); 2632 2633 _inc_cset_head = NULL; 2634 _inc_cset_tail = NULL; 2635 _inc_cset_size = 0; 2636 _inc_cset_bytes_used_before = 0; 2637 2638 if (in_young_gc_mode()) { 2639 _inc_cset_young_index = 0; 2640 } 2641 2642 _inc_cset_max_finger = 0; 2643 _inc_cset_recorded_young_bytes = 0; 2644 _inc_cset_recorded_rs_lengths = 0; 2645 _inc_cset_predicted_elapsed_time_ms = 0; 2646 _inc_cset_predicted_bytes_to_copy = 0; 2647 _inc_cset_build_state = Active; 2648 } 2649 2650 void G1CollectorPolicy::add_to_incremental_cset_info(HeapRegion* hr, size_t rs_length) { 2651 // This routine is used when: 2652 // * adding survivor regions to the incremental cset at the end of an 2653 // evacuation pause, 2654 // * adding the current allocation region to the incremental cset 2655 // when it is retired, and 2656 // * updating existing policy information for a region in the 2657 // incremental cset via young list RSet sampling. 2658 // Therefore this routine may be called at a safepoint by the 2659 // VM thread, or in-between safepoints by mutator threads (when 2660 // retiring the current allocation region) or a concurrent 2661 // refine thread (RSet sampling). 2662 2663 double region_elapsed_time_ms = predict_region_elapsed_time_ms(hr, true); 2664 size_t used_bytes = hr->used(); 2665 2666 _inc_cset_recorded_rs_lengths += rs_length; 2667 _inc_cset_predicted_elapsed_time_ms += region_elapsed_time_ms; 2668 2669 _inc_cset_bytes_used_before += used_bytes; 2670 2671 // Cache the values we have added to the aggregated informtion 2672 // in the heap region in case we have to remove this region from 2673 // the incremental collection set, or it is updated by the 2674 // rset sampling code 2675 hr->set_recorded_rs_length(rs_length); 2676 hr->set_predicted_elapsed_time_ms(region_elapsed_time_ms); 2677 2678 #if PREDICTIONS_VERBOSE 2679 size_t bytes_to_copy = predict_bytes_to_copy(hr); 2680 _inc_cset_predicted_bytes_to_copy += bytes_to_copy; 2681 2682 // Record the number of bytes used in this region 2683 _inc_cset_recorded_young_bytes += used_bytes; 2684 2685 // Cache the values we have added to the aggregated informtion 2686 // in the heap region in case we have to remove this region from 2687 // the incremental collection set, or it is updated by the 2688 // rset sampling code 2689 hr->set_predicted_bytes_to_copy(bytes_to_copy); 2690 #endif // PREDICTIONS_VERBOSE 2691 } 2692 2693 void G1CollectorPolicy::remove_from_incremental_cset_info(HeapRegion* hr) { 2694 // This routine is currently only called as part of the updating of 2695 // existing policy information for regions in the incremental cset that 2696 // is performed by the concurrent refine thread(s) as part of young list 2697 // RSet sampling. Therefore we should not be at a safepoint. 2698 2699 assert(!SafepointSynchronize::is_at_safepoint(), "should not be at safepoint"); 2700 assert(hr->is_young(), "it should be"); 2701 2702 size_t used_bytes = hr->used(); 2703 size_t old_rs_length = hr->recorded_rs_length(); 2704 double old_elapsed_time_ms = hr->predicted_elapsed_time_ms(); 2705 2706 // Subtract the old recorded/predicted policy information for 2707 // the given heap region from the collection set info. 2708 _inc_cset_recorded_rs_lengths -= old_rs_length; 2709 _inc_cset_predicted_elapsed_time_ms -= old_elapsed_time_ms; 2710 2711 _inc_cset_bytes_used_before -= used_bytes; 2712 2713 // Clear the values cached in the heap region 2714 hr->set_recorded_rs_length(0); 2715 hr->set_predicted_elapsed_time_ms(0); 2716 2717 #if PREDICTIONS_VERBOSE 2718 size_t old_predicted_bytes_to_copy = hr->predicted_bytes_to_copy(); 2719 _inc_cset_predicted_bytes_to_copy -= old_predicted_bytes_to_copy; 2720 2721 // Subtract the number of bytes used in this region 2722 _inc_cset_recorded_young_bytes -= used_bytes; 2723 2724 // Clear the values cached in the heap region 2725 hr->set_predicted_bytes_to_copy(0); 2726 #endif // PREDICTIONS_VERBOSE 2727 } 2728 2729 void G1CollectorPolicy::update_incremental_cset_info(HeapRegion* hr, size_t new_rs_length) { 2730 // Update the collection set information that is dependent on the new RS length 2731 assert(hr->is_young(), "Precondition"); 2732 2733 remove_from_incremental_cset_info(hr); 2734 add_to_incremental_cset_info(hr, new_rs_length); 2735 } 2736 2737 void G1CollectorPolicy::add_region_to_incremental_cset_common(HeapRegion* hr) { 2738 assert( hr->is_young(), "invariant"); 2739 assert( hr->young_index_in_cset() == -1, "invariant" ); 2740 assert(_inc_cset_build_state == Active, "Precondition"); 2741 2742 // We need to clear and set the cached recorded/cached collection set 2743 // information in the heap region here (before the region gets added 2744 // to the collection set). An individual heap region's cached values 2745 // are calculated, aggregated with the policy collection set info, 2746 // and cached in the heap region here (initially) and (subsequently) 2747 // by the Young List sampling code. 2748 2749 size_t rs_length = hr->rem_set()->occupied(); 2750 add_to_incremental_cset_info(hr, rs_length); 2751 2752 HeapWord* hr_end = hr->end(); 2753 _inc_cset_max_finger = MAX2(_inc_cset_max_finger, hr_end); 2754 2755 assert(!hr->in_collection_set(), "invariant"); 2756 hr->set_in_collection_set(true); 2757 assert( hr->next_in_collection_set() == NULL, "invariant"); 2758 2759 _inc_cset_size++; 2760 _g1->register_region_with_in_cset_fast_test(hr); 2761 2762 hr->set_young_index_in_cset((int) _inc_cset_young_index); 2763 ++_inc_cset_young_index; 2764 } 2765 2766 // Add the region at the RHS of the incremental cset 2767 void G1CollectorPolicy::add_region_to_incremental_cset_rhs(HeapRegion* hr) { 2768 // We should only ever be appending survivors at the end of a pause 2769 assert( hr->is_survivor(), "Logic"); 2770 2771 // Do the 'common' stuff 2772 add_region_to_incremental_cset_common(hr); 2773 2774 // Now add the region at the right hand side 2775 if (_inc_cset_tail == NULL) { 2776 assert(_inc_cset_head == NULL, "invariant"); 2777 _inc_cset_head = hr; 2778 } else { 2779 _inc_cset_tail->set_next_in_collection_set(hr); 2780 } 2781 _inc_cset_tail = hr; 2782 2783 if (G1PrintHeapRegions) { 2784 gclog_or_tty->print_cr(" added region to incremental cset (RHS) " 2785 "%d:["PTR_FORMAT", "PTR_FORMAT"], " 2786 "top "PTR_FORMAT", young %s", 2787 hr->hrs_index(), hr->bottom(), hr->end(), 2788 hr->top(), (hr->is_young()) ? "YES" : "NO"); 2789 } 2790 } 2791 2792 // Add the region to the LHS of the incremental cset 2793 void G1CollectorPolicy::add_region_to_incremental_cset_lhs(HeapRegion* hr) { 2794 // Survivors should be added to the RHS at the end of a pause 2795 assert(!hr->is_survivor(), "Logic"); 2796 2797 // Do the 'common' stuff 2798 add_region_to_incremental_cset_common(hr); 2799 2800 // Add the region at the left hand side 2801 hr->set_next_in_collection_set(_inc_cset_head); 2802 if (_inc_cset_head == NULL) { 2803 assert(_inc_cset_tail == NULL, "Invariant"); 2804 _inc_cset_tail = hr; 2805 } 2806 _inc_cset_head = hr; 2807 2808 if (G1PrintHeapRegions) { 2809 gclog_or_tty->print_cr(" added region to incremental cset (LHS) " 2810 "%d:["PTR_FORMAT", "PTR_FORMAT"], " 2811 "top "PTR_FORMAT", young %s", 2812 hr->hrs_index(), hr->bottom(), hr->end(), 2813 hr->top(), (hr->is_young()) ? "YES" : "NO"); 2814 } 2815 } 2816 2817 #ifndef PRODUCT 2818 void G1CollectorPolicy::print_collection_set(HeapRegion* list_head, outputStream* st) { 2819 assert(list_head == inc_cset_head() || list_head == collection_set(), "must be"); 2820 2821 st->print_cr("\nCollection_set:"); 2822 HeapRegion* csr = list_head; 2823 while (csr != NULL) { 2824 HeapRegion* next = csr->next_in_collection_set(); 2825 assert(csr->in_collection_set(), "bad CS"); 2826 st->print_cr(" [%08x-%08x], t: %08x, P: %08x, N: %08x, C: %08x, " 2827 "age: %4d, y: %d, surv: %d", 2828 csr->bottom(), csr->end(), 2829 csr->top(), 2830 csr->prev_top_at_mark_start(), 2831 csr->next_top_at_mark_start(), 2832 csr->top_at_conc_mark_count(), 2833 csr->age_in_surv_rate_group_cond(), 2834 csr->is_young(), 2835 csr->is_survivor()); 2836 csr = next; 2837 } 2838 } 2839 #endif // !PRODUCT 2840 2841 void 2842 G1CollectorPolicy_BestRegionsFirst::choose_collection_set( 2843 double target_pause_time_ms) { 2844 // Set this here - in case we're not doing young collections. 2845 double non_young_start_time_sec = os::elapsedTime(); 2846 2847 start_recording_regions(); 2848 2849 guarantee(target_pause_time_ms > 0.0, 2850 err_msg("target_pause_time_ms = %1.6lf should be positive", 2851 target_pause_time_ms)); 2852 guarantee(_collection_set == NULL, "Precondition"); 2853 2854 double base_time_ms = predict_base_elapsed_time_ms(_pending_cards); 2855 double predicted_pause_time_ms = base_time_ms; 2856 2857 double time_remaining_ms = target_pause_time_ms - base_time_ms; 2858 2859 // the 10% and 50% values are arbitrary... 2860 if (time_remaining_ms < 0.10 * target_pause_time_ms) { 2861 time_remaining_ms = 0.50 * target_pause_time_ms; 2862 _within_target = false; 2863 } else { 2864 _within_target = true; 2865 } 2866 2867 // We figure out the number of bytes available for future to-space. 2868 // For new regions without marking information, we must assume the 2869 // worst-case of complete survival. If we have marking information for a 2870 // region, we can bound the amount of live data. We can add a number of 2871 // such regions, as long as the sum of the live data bounds does not 2872 // exceed the available evacuation space. 2873 size_t max_live_bytes = _g1->free_regions() * HeapRegion::GrainBytes; 2874 2875 size_t expansion_bytes = 2876 _g1->expansion_regions() * HeapRegion::GrainBytes; 2877 2878 _collection_set_bytes_used_before = 0; 2879 _collection_set_size = 0; 2880 2881 // Adjust for expansion and slop. 2882 max_live_bytes = max_live_bytes + expansion_bytes; 2883 2884 assert(_g1->regions_accounted_for(), "Region leakage!"); 2885 2886 HeapRegion* hr; 2887 if (in_young_gc_mode()) { 2888 double young_start_time_sec = os::elapsedTime(); 2889 2890 if (G1PolicyVerbose > 0) { 2891 gclog_or_tty->print_cr("Adding %d young regions to the CSet", 2892 _g1->young_list()->length()); 2893 } 2894 2895 _young_cset_length = 0; 2896 _last_young_gc_full = full_young_gcs() ? true : false; 2897 2898 if (_last_young_gc_full) 2899 ++_full_young_pause_num; 2900 else 2901 ++_partial_young_pause_num; 2902 2903 // The young list is laid with the survivor regions from the previous 2904 // pause are appended to the RHS of the young list, i.e. 2905 // [Newly Young Regions ++ Survivors from last pause]. 2906 2907 hr = _g1->young_list()->first_survivor_region(); 2908 while (hr != NULL) { 2909 assert(hr->is_survivor(), "badly formed young list"); 2910 hr->set_young(); 2911 hr = hr->get_next_young_region(); 2912 } 2913 2914 // Clear the fields that point to the survivor list - they are 2915 // all young now. 2916 _g1->young_list()->clear_survivors(); 2917 2918 if (_g1->mark_in_progress()) 2919 _g1->concurrent_mark()->register_collection_set_finger(_inc_cset_max_finger); 2920 2921 _young_cset_length = _inc_cset_young_index; 2922 _collection_set = _inc_cset_head; 2923 _collection_set_size = _inc_cset_size; 2924 _collection_set_bytes_used_before = _inc_cset_bytes_used_before; 2925 2926 // For young regions in the collection set, we assume the worst 2927 // case of complete survival 2928 max_live_bytes -= _inc_cset_size * HeapRegion::GrainBytes; 2929 2930 time_remaining_ms -= _inc_cset_predicted_elapsed_time_ms; 2931 predicted_pause_time_ms += _inc_cset_predicted_elapsed_time_ms; 2932 2933 // The number of recorded young regions is the incremental 2934 // collection set's current size 2935 set_recorded_young_regions(_inc_cset_size); 2936 set_recorded_rs_lengths(_inc_cset_recorded_rs_lengths); 2937 set_recorded_young_bytes(_inc_cset_recorded_young_bytes); 2938 #if PREDICTIONS_VERBOSE 2939 set_predicted_bytes_to_copy(_inc_cset_predicted_bytes_to_copy); 2940 #endif // PREDICTIONS_VERBOSE 2941 2942 if (G1PolicyVerbose > 0) { 2943 gclog_or_tty->print_cr(" Added " PTR_FORMAT " Young Regions to CS.", 2944 _inc_cset_size); 2945 gclog_or_tty->print_cr(" (" SIZE_FORMAT " KB left in heap.)", 2946 max_live_bytes/K); 2947 } 2948 2949 assert(_inc_cset_size == _g1->young_list()->length(), "Invariant"); 2950 2951 double young_end_time_sec = os::elapsedTime(); 2952 _recorded_young_cset_choice_time_ms = 2953 (young_end_time_sec - young_start_time_sec) * 1000.0; 2954 2955 // We are doing young collections so reset this. 2956 non_young_start_time_sec = young_end_time_sec; 2957 2958 // Note we can use either _collection_set_size or 2959 // _young_cset_length here 2960 if (_collection_set_size > 0 && _last_young_gc_full) { 2961 // don't bother adding more regions... 2962 goto choose_collection_set_end; 2963 } 2964 } 2965 2966 if (!in_young_gc_mode() || !full_young_gcs()) { 2967 bool should_continue = true; 2968 NumberSeq seq; 2969 double avg_prediction = 100000000000000000.0; // something very large 2970 2971 do { 2972 hr = _collectionSetChooser->getNextMarkedRegion(time_remaining_ms, 2973 avg_prediction); 2974 if (hr != NULL) { 2975 double predicted_time_ms = predict_region_elapsed_time_ms(hr, false); 2976 time_remaining_ms -= predicted_time_ms; 2977 predicted_pause_time_ms += predicted_time_ms; 2978 add_to_collection_set(hr); 2979 record_non_young_cset_region(hr); 2980 max_live_bytes -= MIN2(hr->max_live_bytes(), max_live_bytes); 2981 if (G1PolicyVerbose > 0) { 2982 gclog_or_tty->print_cr(" (" SIZE_FORMAT " KB left in heap.)", 2983 max_live_bytes/K); 2984 } 2985 seq.add(predicted_time_ms); 2986 avg_prediction = seq.avg() + seq.sd(); 2987 } 2988 should_continue = 2989 ( hr != NULL) && 2990 ( (adaptive_young_list_length()) ? time_remaining_ms > 0.0 2991 : _collection_set_size < _young_list_fixed_length ); 2992 } while (should_continue); 2993 2994 if (!adaptive_young_list_length() && 2995 _collection_set_size < _young_list_fixed_length) 2996 _should_revert_to_full_young_gcs = true; 2997 } 2998 2999 choose_collection_set_end: 3000 stop_incremental_cset_building(); 3001 3002 count_CS_bytes_used(); 3003 3004 end_recording_regions(); 3005 3006 double non_young_end_time_sec = os::elapsedTime(); 3007 _recorded_non_young_cset_choice_time_ms = 3008 (non_young_end_time_sec - non_young_start_time_sec) * 1000.0; 3009 } 3010 3011 void G1CollectorPolicy_BestRegionsFirst::record_full_collection_end() { 3012 G1CollectorPolicy::record_full_collection_end(); 3013 _collectionSetChooser->updateAfterFullCollection(); 3014 } 3015 3016 void G1CollectorPolicy_BestRegionsFirst:: 3017 expand_if_possible(size_t numRegions) { 3018 size_t expansion_bytes = numRegions * HeapRegion::GrainBytes; 3019 _g1->expand(expansion_bytes); 3020 } 3021 3022 void G1CollectorPolicy_BestRegionsFirst:: 3023 record_collection_pause_end() { 3024 G1CollectorPolicy::record_collection_pause_end(); 3025 assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end."); 3026 }