g1CollectorPolicy.cpp 115.8 KB
Newer Older
1
/*
2
 * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
19 20 21
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
22 23 24
 *
 */

25 26 27 28 29 30
#include "precompiled.hpp"
#include "gc_implementation/g1/concurrentG1Refine.hpp"
#include "gc_implementation/g1/concurrentMark.hpp"
#include "gc_implementation/g1/concurrentMarkThread.inline.hpp"
#include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
#include "gc_implementation/g1/g1CollectorPolicy.hpp"
31
#include "gc_implementation/g1/g1ErgoVerbose.hpp"
32 33 34 35 36 37
#include "gc_implementation/g1/heapRegionRemSet.hpp"
#include "gc_implementation/shared/gcPolicyCounters.hpp"
#include "runtime/arguments.hpp"
#include "runtime/java.hpp"
#include "runtime/mutexLocker.hpp"
#include "utilities/debug.hpp"
38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84

#define PREDICTIONS_VERBOSE 0

// <NEW PREDICTION>

// Different defaults for different number of GC threads
// They were chosen by running GCOld and SPECjbb on debris with different
//   numbers of GC threads and choosing them based on the results

// all the same
static double rs_length_diff_defaults[] = {
  0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0
};

static double cost_per_card_ms_defaults[] = {
  0.01, 0.005, 0.005, 0.003, 0.003, 0.002, 0.002, 0.0015
};

// all the same
static double fully_young_cards_per_entry_ratio_defaults[] = {
  1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0
};

static double cost_per_entry_ms_defaults[] = {
  0.015, 0.01, 0.01, 0.008, 0.008, 0.0055, 0.0055, 0.005
};

static double cost_per_byte_ms_defaults[] = {
  0.00006, 0.00003, 0.00003, 0.000015, 0.000015, 0.00001, 0.00001, 0.000009
};

// these should be pretty consistent
static double constant_other_time_ms_defaults[] = {
  5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0
};


static double young_other_cost_per_region_ms_defaults[] = {
  0.3, 0.2, 0.2, 0.15, 0.15, 0.12, 0.12, 0.1
};

static double non_young_other_cost_per_region_ms_defaults[] = {
  1.0, 0.7, 0.7, 0.5, 0.5, 0.42, 0.42, 0.30
};

// </NEW PREDICTION>

85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135
// Help class for avoiding interleaved logging
class LineBuffer: public StackObj {

private:
  static const int BUFFER_LEN = 1024;
  static const int INDENT_CHARS = 3;
  char _buffer[BUFFER_LEN];
  int _indent_level;
  int _cur;

  void vappend(const char* format, va_list ap) {
    int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
    if (res != -1) {
      _cur += res;
    } else {
      DEBUG_ONLY(warning("buffer too small in LineBuffer");)
      _buffer[BUFFER_LEN -1] = 0;
      _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
    }
  }

public:
  explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
    for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
      _buffer[_cur] = ' ';
    }
  }

#ifndef PRODUCT
  ~LineBuffer() {
    assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
  }
#endif

  void append(const char* format, ...) {
    va_list ap;
    va_start(ap, format);
    vappend(format, ap);
    va_end(ap);
  }

  void append_and_print_cr(const char* format, ...) {
    va_list ap;
    va_start(ap, format);
    vappend(format, ap);
    va_end(ap);
    gclog_or_tty->print_cr("%s", _buffer);
    _cur = _indent_level * INDENT_CHARS;
  }
};

136
G1CollectorPolicy::G1CollectorPolicy() :
137
  _parallel_gc_threads(G1CollectedHeap::use_parallel_gc_threads()
138
                        ? ParallelGCThreads : 1),
139

140
  _n_pauses(0),
141
  _recent_rs_scan_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
142 143 144 145 146 147 148
  _recent_pause_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  _recent_rs_sizes(new TruncatedSeq(NumPrevPausesForHeuristics)),
  _recent_gc_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  _all_pause_times_ms(new NumberSeq()),
  _stop_world_start(0.0),
  _all_stop_world_times_ms(new NumberSeq()),
  _all_yield_times_ms(new NumberSeq()),
149
  _using_new_ratio_calculations(false),
150 151 152

  _all_mod_union_times_ms(new NumberSeq()),

153
  _summary(new Summary()),
154 155

  _cur_clear_ct_time_ms(0.0),
156 157 158 159 160

  _cur_ref_proc_time_ms(0.0),
  _cur_ref_enq_time_ms(0.0),

#ifndef PRODUCT
161 162 163 164 165 166
  _min_clear_cc_time_ms(-1.0),
  _max_clear_cc_time_ms(-1.0),
  _cur_clear_cc_time_ms(0.0),
  _cum_clear_cc_time_ms(0.0),
  _num_cc_clears(0L),
#endif
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204

  _region_num_young(0),
  _region_num_tenured(0),
  _prev_region_num_young(0),
  _prev_region_num_tenured(0),

  _aux_num(10),
  _all_aux_times_ms(new NumberSeq[_aux_num]),
  _cur_aux_start_times_ms(new double[_aux_num]),
  _cur_aux_times_ms(new double[_aux_num]),
  _cur_aux_times_set(new bool[_aux_num]),

  _concurrent_mark_remark_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
  _concurrent_mark_cleanup_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),

  // <NEW PREDICTION>

  _alloc_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _prev_collection_pause_end_ms(0.0),
  _pending_card_diff_seq(new TruncatedSeq(TruncatedSeqLength)),
  _rs_length_diff_seq(new TruncatedSeq(TruncatedSeqLength)),
  _cost_per_card_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _fully_young_cards_per_entry_ratio_seq(new TruncatedSeq(TruncatedSeqLength)),
  _partially_young_cards_per_entry_ratio_seq(
                                         new TruncatedSeq(TruncatedSeqLength)),
  _cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _partially_young_cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _cost_per_byte_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _cost_per_byte_ms_during_cm_seq(new TruncatedSeq(TruncatedSeqLength)),
  _constant_other_time_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _young_other_cost_per_region_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
  _non_young_other_cost_per_region_ms_seq(
                                         new TruncatedSeq(TruncatedSeqLength)),

  _pending_cards_seq(new TruncatedSeq(TruncatedSeqLength)),
  _scanned_cards_seq(new TruncatedSeq(TruncatedSeqLength)),
  _rs_lengths_seq(new TruncatedSeq(TruncatedSeqLength)),

J
johnc 已提交
205
  _pause_time_target_ms((double) MaxGCPauseMillis),
206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237

  // </NEW PREDICTION>

  _full_young_gcs(true),
  _full_young_pause_num(0),
  _partial_young_pause_num(0),

  _during_marking(false),
  _in_marking_window(false),
  _in_marking_window_im(false),

  _known_garbage_ratio(0.0),
  _known_garbage_bytes(0),

  _young_gc_eff_seq(new TruncatedSeq(TruncatedSeqLength)),

   _recent_prev_end_times_for_all_gcs_sec(new TruncatedSeq(NumPrevPausesForHeuristics)),

  _recent_CS_bytes_used_before(new TruncatedSeq(NumPrevPausesForHeuristics)),
  _recent_CS_bytes_surviving(new TruncatedSeq(NumPrevPausesForHeuristics)),

  _recent_avg_pause_time_ratio(0.0),
  _num_markings(0),
  _n_marks(0),
  _n_pauses_at_mark_end(0),

  _all_full_gc_times_ms(new NumberSeq()),

  // G1PausesBtwnConcMark defaults to -1
  // so the hack is to do the cast  QQQ FIXME
  _pauses_btwn_concurrent_mark((size_t)G1PausesBtwnConcMark),
  _n_marks_since_last_pause(0),
238 239
  _initiate_conc_mark_if_possible(false),
  _during_initial_mark_pause(false),
240 241 242
  _should_revert_to_full_young_gcs(false),
  _last_full_young_gc(false),

243 244 245 246
  _eden_bytes_before_gc(0),
  _survivor_bytes_before_gc(0),
  _capacity_before_gc(0),

247 248 249
  _prev_collection_pause_used_at_end_bytes(0),

  _collection_set(NULL),
250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265
  _collection_set_size(0),
  _collection_set_bytes_used_before(0),

  // Incremental CSet attributes
  _inc_cset_build_state(Inactive),
  _inc_cset_head(NULL),
  _inc_cset_tail(NULL),
  _inc_cset_size(0),
  _inc_cset_young_index(0),
  _inc_cset_bytes_used_before(0),
  _inc_cset_max_finger(NULL),
  _inc_cset_recorded_young_bytes(0),
  _inc_cset_recorded_rs_lengths(0),
  _inc_cset_predicted_elapsed_time_ms(0.0),
  _inc_cset_predicted_bytes_to_copy(0),

266 267 268 269 270 271 272
#ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
#pragma warning( disable:4355 ) // 'this' : used in base member initializer list
#endif // _MSC_VER

  _short_lived_surv_rate_group(new SurvRateGroup(this, "Short Lived",
                                                 G1YoungSurvRateNumRegionsSummary)),
  _survivor_surv_rate_group(new SurvRateGroup(this, "Survivor",
273
                                              G1YoungSurvRateNumRegionsSummary)),
274
  // add here any more surv rate groups
275 276 277
  _recorded_survivor_regions(0),
  _recorded_survivor_head(NULL),
  _recorded_survivor_tail(NULL),
T
tonyp 已提交
278 279
  _survivors_age_table(true),

280
  _gc_overhead_perc(0.0) {
281

282 283 284 285
  // Set up the region size and associated fields. Given that the
  // policy is created before the heap, we have to set this up here,
  // so it's done as soon as possible.
  HeapRegion::setup_heap_region_size(Arguments::min_heap_size());
286
  HeapRegionRemSet::setup_remset_size();
287

288 289 290 291 292 293 294 295 296 297 298 299
  G1ErgoVerbose::initialize();
  if (PrintAdaptiveSizePolicy) {
    // Currently, we only use a single switch for all the heuristics.
    G1ErgoVerbose::set_enabled(true);
    // Given that we don't currently have a verboseness level
    // parameter, we'll hardcode this to high. This can be easily
    // changed in the future.
    G1ErgoVerbose::set_level(ErgoHigh);
  } else {
    G1ErgoVerbose::set_enabled(false);
  }

300 301 302 303 304 305 306 307 308
  // Verify PLAB sizes
  const uint region_size = HeapRegion::GrainWords;
  if (YoungPLABSize > region_size || OldPLABSize > region_size) {
    char buffer[128];
    jio_snprintf(buffer, sizeof(buffer), "%sPLABSize should be at most %u",
                 OldPLABSize > region_size ? "Old" : "Young", region_size);
    vm_exit_during_initialization(buffer);
  }

309 310 311
  _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime());
  _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0;

312
  _par_last_gc_worker_start_times_ms = new double[_parallel_gc_threads];
313 314 315 316 317 318 319 320 321 322 323
  _par_last_ext_root_scan_times_ms = new double[_parallel_gc_threads];
  _par_last_mark_stack_scan_times_ms = new double[_parallel_gc_threads];

  _par_last_update_rs_times_ms = new double[_parallel_gc_threads];
  _par_last_update_rs_processed_buffers = new double[_parallel_gc_threads];

  _par_last_scan_rs_times_ms = new double[_parallel_gc_threads];

  _par_last_obj_copy_times_ms = new double[_parallel_gc_threads];

  _par_last_termination_times_ms = new double[_parallel_gc_threads];
324 325
  _par_last_termination_attempts = new double[_parallel_gc_threads];
  _par_last_gc_worker_end_times_ms = new double[_parallel_gc_threads];
326
  _par_last_gc_worker_times_ms = new double[_parallel_gc_threads];
327 328

  // start conservatively
J
johnc 已提交
329
  _expensive_region_limit_ms = 0.5 * (double) MaxGCPauseMillis;
330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355

  // <NEW PREDICTION>

  int index;
  if (ParallelGCThreads == 0)
    index = 0;
  else if (ParallelGCThreads > 8)
    index = 7;
  else
    index = ParallelGCThreads - 1;

  _pending_card_diff_seq->add(0.0);
  _rs_length_diff_seq->add(rs_length_diff_defaults[index]);
  _cost_per_card_ms_seq->add(cost_per_card_ms_defaults[index]);
  _fully_young_cards_per_entry_ratio_seq->add(
                            fully_young_cards_per_entry_ratio_defaults[index]);
  _cost_per_entry_ms_seq->add(cost_per_entry_ms_defaults[index]);
  _cost_per_byte_ms_seq->add(cost_per_byte_ms_defaults[index]);
  _constant_other_time_ms_seq->add(constant_other_time_ms_defaults[index]);
  _young_other_cost_per_region_ms_seq->add(
                               young_other_cost_per_region_ms_defaults[index]);
  _non_young_other_cost_per_region_ms_seq->add(
                           non_young_other_cost_per_region_ms_defaults[index]);

  // </NEW PREDICTION>

356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411
  // Below, we might need to calculate the pause time target based on
  // the pause interval. When we do so we are going to give G1 maximum
  // flexibility and allow it to do pauses when it needs to. So, we'll
  // arrange that the pause interval to be pause time target + 1 to
  // ensure that a) the pause time target is maximized with respect to
  // the pause interval and b) we maintain the invariant that pause
  // time target < pause interval. If the user does not want this
  // maximum flexibility, they will have to set the pause interval
  // explicitly.

  // First make sure that, if either parameter is set, its value is
  // reasonable.
  if (!FLAG_IS_DEFAULT(MaxGCPauseMillis)) {
    if (MaxGCPauseMillis < 1) {
      vm_exit_during_initialization("MaxGCPauseMillis should be "
                                    "greater than 0");
    }
  }
  if (!FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
    if (GCPauseIntervalMillis < 1) {
      vm_exit_during_initialization("GCPauseIntervalMillis should be "
                                    "greater than 0");
    }
  }

  // Then, if the pause time target parameter was not set, set it to
  // the default value.
  if (FLAG_IS_DEFAULT(MaxGCPauseMillis)) {
    if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
      // The default pause time target in G1 is 200ms
      FLAG_SET_DEFAULT(MaxGCPauseMillis, 200);
    } else {
      // We do not allow the pause interval to be set without the
      // pause time target
      vm_exit_during_initialization("GCPauseIntervalMillis cannot be set "
                                    "without setting MaxGCPauseMillis");
    }
  }

  // Then, if the interval parameter was not set, set it according to
  // the pause time target (this will also deal with the case when the
  // pause time target is the default value).
  if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
    FLAG_SET_DEFAULT(GCPauseIntervalMillis, MaxGCPauseMillis + 1);
  }

  // Finally, make sure that the two parameters are consistent.
  if (MaxGCPauseMillis >= GCPauseIntervalMillis) {
    char buffer[256];
    jio_snprintf(buffer, 256,
                 "MaxGCPauseMillis (%u) should be less than "
                 "GCPauseIntervalMillis (%u)",
                 MaxGCPauseMillis, GCPauseIntervalMillis);
    vm_exit_during_initialization(buffer);
  }

J
johnc 已提交
412
  double max_gc_time = (double) MaxGCPauseMillis / 1000.0;
413
  double time_slice  = (double) GCPauseIntervalMillis / 1000.0;
414
  _mmu_tracker = new G1MMUTrackerQueue(time_slice, max_gc_time);
J
johnc 已提交
415
  _sigma = (double) G1ConfidencePercent / 100.0;
416 417 418 419 420

  // start conservatively (around 50ms is about right)
  _concurrent_mark_remark_times_ms->add(0.05);
  _concurrent_mark_cleanup_times_ms->add(0.20);
  _tenuring_threshold = MaxTenuringThreshold;
421
  // _max_survivor_regions will be calculated by
422
  // update_young_list_target_length() during initialization.
423
  _max_survivor_regions = 0;
424

T
tonyp 已提交
425 426 427 428 429
  assert(GCTimeRatio > 0,
         "we should have set it to a default value set_g1_gc_flags() "
         "if a user set it to 0");
  _gc_overhead_perc = 100.0 * (1.0 / (1.0 + GCTimeRatio));

430 431 432 433 434 435 436 437
  uintx reserve_perc = G1ReservePercent;
  // Put an artificial ceiling on this so that it's not set to a silly value.
  if (reserve_perc > 50) {
    reserve_perc = 50;
    warning("G1ReservePercent is set to a value that is too large, "
            "it's been updated to %u", reserve_perc);
  }
  _reserve_factor = (double) reserve_perc / 100.0;
438
  // This will be set when the heap is expanded
439 440 441
  // for the first time during initialization.
  _reserve_regions = 0;

442 443 444 445 446 447 448 449 450 451 452
  initialize_all();
}

// Increment "i", mod "len"
static void inc_mod(int& i, int len) {
  i++; if (i == len) i = 0;
}

void G1CollectorPolicy::initialize_flags() {
  set_min_alignment(HeapRegion::GrainBytes);
  set_max_alignment(GenRemSet::max_alignment_constraint(rem_set_name()));
453 454 455
  if (SurvivorRatio < 1) {
    vm_exit_during_initialization("Invalid survivor ratio specified");
  }
456 457 458
  CollectorPolicy::initialize_flags();
}

459 460 461 462 463 464 465
// The easiest way to deal with the parsing of the NewSize /
// MaxNewSize / etc. parameteres is to re-use the code in the
// TwoGenerationCollectorPolicy class. This is similar to what
// ParallelScavenge does with its GenerationSizer class (see
// ParallelScavengeHeap::initialize()). We might change this in the
// future, but it's a good start.
class G1YoungGenSizer : public TwoGenerationCollectorPolicy {
466 467 468 469
private:
  size_t size_to_region_num(size_t byte_size) {
    return MAX2((size_t) 1, byte_size / HeapRegion::GrainBytes);
  }
470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486

public:
  G1YoungGenSizer() {
    initialize_flags();
    initialize_size_info();
  }
  size_t min_young_region_num() {
    return size_to_region_num(_min_gen0_size);
  }
  size_t initial_young_region_num() {
    return size_to_region_num(_initial_gen0_size);
  }
  size_t max_young_region_num() {
    return size_to_region_num(_max_gen0_size);
  }
};

487 488 489 490 491 492 493
void G1CollectorPolicy::update_young_list_size_using_newratio(size_t number_of_heap_regions) {
  assert(number_of_heap_regions > 0, "Heap must be initialized");
  size_t young_size = number_of_heap_regions / (NewRatio + 1);
  _min_desired_young_length = young_size;
  _max_desired_young_length = young_size;
}

494 495 496 497 498 499
void G1CollectorPolicy::init() {
  // Set aside an initial future to_space.
  _g1 = G1CollectedHeap::heap();

  assert(Heap_lock->owned_by_self(), "Locking discipline.");

500 501
  initialize_gc_policy_counters();

502 503
  G1YoungGenSizer sizer;
  size_t initial_region_num = sizer.initial_young_region_num();
504 505
  _min_desired_young_length = sizer.min_young_region_num();
  _max_desired_young_length = sizer.max_young_region_num();
506

507 508
  if (FLAG_IS_CMDLINE(NewRatio)) {
    if (FLAG_IS_CMDLINE(NewSize) || FLAG_IS_CMDLINE(MaxNewSize)) {
509
      warning("-XX:NewSize and -XX:MaxNewSize override -XX:NewRatio");
510 511
    } else {
      // Treat NewRatio as a fixed size that is only recalculated when the heap size changes
512
      update_young_list_size_using_newratio(_g1->n_regions());
513 514 515 516 517 518 519 520 521 522 523 524
      _using_new_ratio_calculations = true;
    }
  }

  // GenCollectorPolicy guarantees that min <= initial <= max.
  // Asserting here just to state that we rely on this property.
  assert(_min_desired_young_length <= _max_desired_young_length, "Invalid min/max young gen size values");
  assert(initial_region_num <= _max_desired_young_length, "Initial young gen size too large");
  assert(_min_desired_young_length <= initial_region_num, "Initial young gen size too small");

  set_adaptive_young_list_length(_min_desired_young_length < _max_desired_young_length);
  if (adaptive_young_list_length()) {
525
    _young_list_fixed_length = 0;
526
  } else {
527
    _young_list_fixed_length = initial_region_num;
528
  }
529
  _free_regions_at_end_of_collection = _g1->free_regions();
530
  update_young_list_target_length();
531
  _prev_eden_capacity = _young_list_target_length * HeapRegion::GrainBytes;
532 533 534 535

  // We may immediately start allocating regions and placing them on the
  // collection set list. Initialize the per-collection set info
  start_incremental_cset_building();
536 537
}

538
// Create the jstat counters for the policy.
539
void G1CollectorPolicy::initialize_gc_policy_counters() {
540
  _gc_policy_counters = new GCPolicyCounters("GarbageFirst", 1, 3);
541 542
}

543 544 545 546 547 548 549 550
bool G1CollectorPolicy::predict_will_fit(size_t young_length,
                                         double base_time_ms,
                                         size_t base_free_regions,
                                         double target_pause_time_ms) {
  if (young_length >= base_free_regions) {
    // end condition 1: not enough space for the young regions
    return false;
  }
551

552 553 554 555 556 557 558 559 560 561
  double accum_surv_rate = accum_yg_surv_rate_pred((int)(young_length - 1));
  size_t bytes_to_copy =
               (size_t) (accum_surv_rate * (double) HeapRegion::GrainBytes);
  double copy_time_ms = predict_object_copy_time_ms(bytes_to_copy);
  double young_other_time_ms = predict_young_other_time_ms(young_length);
  double pause_time_ms = base_time_ms + copy_time_ms + young_other_time_ms;
  if (pause_time_ms > target_pause_time_ms) {
    // end condition 2: prediction is over the target pause time
    return false;
  }
562

563 564 565 566 567
  size_t free_bytes =
                  (base_free_regions - young_length) * HeapRegion::GrainBytes;
  if ((2.0 * sigma()) * (double) bytes_to_copy > (double) free_bytes) {
    // end condition 3: out-of-space (conservatively!)
    return false;
568
  }
569 570 571 572 573

  // success!
  return true;
}

574 575 576
void G1CollectorPolicy::record_new_heap_size(size_t new_number_of_regions) {
  // re-calculate the necessary reserve
  double reserve_regions_d = (double) new_number_of_regions * _reserve_factor;
577 578 579
  // We use ceiling so that if reserve_regions_d is > 0.0 (but
  // smaller than 1.0) we'll get 1.
  _reserve_regions = (size_t) ceil(reserve_regions_d);
580 581 582 583 584 585

  if (_using_new_ratio_calculations) {
    // -XX:NewRatio was specified so we need to update the
    // young gen length when the heap size has changed.
    update_young_list_size_using_newratio(new_number_of_regions);
  }
586 587
}

588 589 590
size_t G1CollectorPolicy::calculate_young_list_desired_min_length(
                                                     size_t base_min_length) {
  size_t desired_min_length = 0;
591
  if (adaptive_young_list_length()) {
592 593 594 595 596 597 598
    if (_alloc_rate_ms_seq->num() > 3) {
      double now_sec = os::elapsedTime();
      double when_ms = _mmu_tracker->when_max_gc_sec(now_sec) * 1000.0;
      double alloc_rate_ms = predict_alloc_rate_ms();
      desired_min_length = (size_t) ceil(alloc_rate_ms * when_ms);
    } else {
      // otherwise we don't have enough info to make the prediction
599 600
    }
  }
601 602 603
  desired_min_length += base_min_length;
  // make sure we don't go below any user-defined minimum bound
  return MAX2(_min_desired_young_length, desired_min_length);
604 605
}

606 607 608 609
size_t G1CollectorPolicy::calculate_young_list_desired_max_length() {
  // Here, we might want to also take into account any additional
  // constraints (i.e., user-defined minimum bound). Currently, we
  // effectively don't set this bound.
610
  return _max_desired_young_length;
611
}
612

613 614 615 616 617 618
void G1CollectorPolicy::update_young_list_target_length(size_t rs_lengths) {
  if (rs_lengths == (size_t) -1) {
    // if it's set to the default value (-1), we should predict it;
    // otherwise, use the given value.
    rs_lengths = (size_t) get_new_prediction(_rs_lengths_seq);
  }
619

620
  // Calculate the absolute and desired min bounds.
621

622 623 624 625 626 627 628 629 630 631
  // This is how many young regions we already have (currently: the survivors).
  size_t base_min_length = recorded_survivor_regions();
  // This is the absolute minimum young length, which ensures that we
  // can allocate one eden region in the worst-case.
  size_t absolute_min_length = base_min_length + 1;
  size_t desired_min_length =
                     calculate_young_list_desired_min_length(base_min_length);
  if (desired_min_length < absolute_min_length) {
    desired_min_length = absolute_min_length;
  }
632

633
  // Calculate the absolute and desired max bounds.
634

635 636 637 638 639 640 641 642 643
  // We will try our best not to "eat" into the reserve.
  size_t absolute_max_length = 0;
  if (_free_regions_at_end_of_collection > _reserve_regions) {
    absolute_max_length = _free_regions_at_end_of_collection - _reserve_regions;
  }
  size_t desired_max_length = calculate_young_list_desired_max_length();
  if (desired_max_length > absolute_max_length) {
    desired_max_length = absolute_max_length;
  }
644

645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670
  size_t young_list_target_length = 0;
  if (adaptive_young_list_length()) {
    if (full_young_gcs()) {
      young_list_target_length =
                        calculate_young_list_target_length(rs_lengths,
                                                           base_min_length,
                                                           desired_min_length,
                                                           desired_max_length);
      _rs_lengths_prediction = rs_lengths;
    } else {
      // Don't calculate anything and let the code below bound it to
      // the desired_min_length, i.e., do the next GC as soon as
      // possible to maximize how many old regions we can add to it.
    }
  } else {
    if (full_young_gcs()) {
      young_list_target_length = _young_list_fixed_length;
    } else {
      // A bit arbitrary: during partially-young GCs we allocate half
      // the young regions to try to add old regions to the CSet.
      young_list_target_length = _young_list_fixed_length / 2;
      // We choose to accept that we might go under the desired min
      // length given that we intentionally ask for a smaller young gen.
      desired_min_length = absolute_min_length;
    }
  }
671

672 673 674 675 676 677 678 679 680
  // Make sure we don't go over the desired max length, nor under the
  // desired min length. In case they clash, desired_min_length wins
  // which is why that test is second.
  if (young_list_target_length > desired_max_length) {
    young_list_target_length = desired_max_length;
  }
  if (young_list_target_length < desired_min_length) {
    young_list_target_length = desired_min_length;
  }
681

682 683 684 685
  assert(young_list_target_length > recorded_survivor_regions(),
         "we should be able to allocate at least one eden region");
  assert(young_list_target_length >= absolute_min_length, "post-condition");
  _young_list_target_length = young_list_target_length;
686

687 688
  update_max_gc_locker_expansion();
}
689

690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791
size_t
G1CollectorPolicy::calculate_young_list_target_length(size_t rs_lengths,
                                                   size_t base_min_length,
                                                   size_t desired_min_length,
                                                   size_t desired_max_length) {
  assert(adaptive_young_list_length(), "pre-condition");
  assert(full_young_gcs(), "only call this for fully-young GCs");

  // In case some edge-condition makes the desired max length too small...
  if (desired_max_length <= desired_min_length) {
    return desired_min_length;
  }

  // We'll adjust min_young_length and max_young_length not to include
  // the already allocated young regions (i.e., so they reflect the
  // min and max eden regions we'll allocate). The base_min_length
  // will be reflected in the predictions by the
  // survivor_regions_evac_time prediction.
  assert(desired_min_length > base_min_length, "invariant");
  size_t min_young_length = desired_min_length - base_min_length;
  assert(desired_max_length > base_min_length, "invariant");
  size_t max_young_length = desired_max_length - base_min_length;

  double target_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0;
  double survivor_regions_evac_time = predict_survivor_regions_evac_time();
  size_t pending_cards = (size_t) get_new_prediction(_pending_cards_seq);
  size_t adj_rs_lengths = rs_lengths + predict_rs_length_diff();
  size_t scanned_cards = predict_young_card_num(adj_rs_lengths);
  double base_time_ms =
    predict_base_elapsed_time_ms(pending_cards, scanned_cards) +
    survivor_regions_evac_time;
  size_t available_free_regions = _free_regions_at_end_of_collection;
  size_t base_free_regions = 0;
  if (available_free_regions > _reserve_regions) {
    base_free_regions = available_free_regions - _reserve_regions;
  }

  // Here, we will make sure that the shortest young length that
  // makes sense fits within the target pause time.

  if (predict_will_fit(min_young_length, base_time_ms,
                       base_free_regions, target_pause_time_ms)) {
    // The shortest young length will fit into the target pause time;
    // we'll now check whether the absolute maximum number of young
    // regions will fit in the target pause time. If not, we'll do
    // a binary search between min_young_length and max_young_length.
    if (predict_will_fit(max_young_length, base_time_ms,
                         base_free_regions, target_pause_time_ms)) {
      // The maximum young length will fit into the target pause time.
      // We are done so set min young length to the maximum length (as
      // the result is assumed to be returned in min_young_length).
      min_young_length = max_young_length;
    } else {
      // The maximum possible number of young regions will not fit within
      // the target pause time so we'll search for the optimal
      // length. The loop invariants are:
      //
      // min_young_length < max_young_length
      // min_young_length is known to fit into the target pause time
      // max_young_length is known not to fit into the target pause time
      //
      // Going into the loop we know the above hold as we've just
      // checked them. Every time around the loop we check whether
      // the middle value between min_young_length and
      // max_young_length fits into the target pause time. If it
      // does, it becomes the new min. If it doesn't, it becomes
      // the new max. This way we maintain the loop invariants.

      assert(min_young_length < max_young_length, "invariant");
      size_t diff = (max_young_length - min_young_length) / 2;
      while (diff > 0) {
        size_t young_length = min_young_length + diff;
        if (predict_will_fit(young_length, base_time_ms,
                             base_free_regions, target_pause_time_ms)) {
          min_young_length = young_length;
        } else {
          max_young_length = young_length;
        }
        assert(min_young_length <  max_young_length, "invariant");
        diff = (max_young_length - min_young_length) / 2;
      }
      // The results is min_young_length which, according to the
      // loop invariants, should fit within the target pause time.

      // These are the post-conditions of the binary search above:
      assert(min_young_length < max_young_length,
             "otherwise we should have discovered that max_young_length "
             "fits into the pause target and not done the binary search");
      assert(predict_will_fit(min_young_length, base_time_ms,
                              base_free_regions, target_pause_time_ms),
             "min_young_length, the result of the binary search, should "
             "fit into the pause target");
      assert(!predict_will_fit(min_young_length + 1, base_time_ms,
                               base_free_regions, target_pause_time_ms),
             "min_young_length, the result of the binary search, should be "
             "optimal, so no larger length should fit into the pause target");
    }
  } else {
    // Even the minimum length doesn't fit into the pause time
    // target, return it as the result nevertheless.
  }
  return base_min_length + min_young_length;
792 793
}

794 795 796 797 798 799 800 801 802 803
double G1CollectorPolicy::predict_survivor_regions_evac_time() {
  double survivor_regions_evac_time = 0.0;
  for (HeapRegion * r = _recorded_survivor_head;
       r != NULL && r != _recorded_survivor_tail->get_next_young_region();
       r = r->get_next_young_region()) {
    survivor_regions_evac_time += predict_region_elapsed_time_ms(r, true);
  }
  return survivor_regions_evac_time;
}

804
void G1CollectorPolicy::revise_young_list_target_length_if_necessary() {
805 806
  guarantee( adaptive_young_list_length(), "should not call this otherwise" );

807
  size_t rs_lengths = _g1->young_list()->sampled_rs_lengths();
808 809 810
  if (rs_lengths > _rs_lengths_prediction) {
    // add 10% to avoid having to recalculate often
    size_t rs_lengths_prediction = rs_lengths * 1100 / 1000;
811
    update_young_list_target_length(rs_lengths_prediction);
812 813 814
  }
}

815 816


817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834
HeapWord* G1CollectorPolicy::mem_allocate_work(size_t size,
                                               bool is_tlab,
                                               bool* gc_overhead_limit_was_exceeded) {
  guarantee(false, "Not using this policy feature yet.");
  return NULL;
}

// This method controls how a collector handles one or more
// of its generations being fully allocated.
HeapWord* G1CollectorPolicy::satisfy_failed_allocation(size_t size,
                                                       bool is_tlab) {
  guarantee(false, "Not using this policy feature yet.");
  return NULL;
}


#ifndef PRODUCT
bool G1CollectorPolicy::verify_young_ages() {
835
  HeapRegion* head = _g1->young_list()->first_region();
836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894
  return
    verify_young_ages(head, _short_lived_surv_rate_group);
  // also call verify_young_ages on any additional surv rate groups
}

bool
G1CollectorPolicy::verify_young_ages(HeapRegion* head,
                                     SurvRateGroup *surv_rate_group) {
  guarantee( surv_rate_group != NULL, "pre-condition" );

  const char* name = surv_rate_group->name();
  bool ret = true;
  int prev_age = -1;

  for (HeapRegion* curr = head;
       curr != NULL;
       curr = curr->get_next_young_region()) {
    SurvRateGroup* group = curr->surv_rate_group();
    if (group == NULL && !curr->is_survivor()) {
      gclog_or_tty->print_cr("## %s: encountered NULL surv_rate_group", name);
      ret = false;
    }

    if (surv_rate_group == group) {
      int age = curr->age_in_surv_rate_group();

      if (age < 0) {
        gclog_or_tty->print_cr("## %s: encountered negative age", name);
        ret = false;
      }

      if (age <= prev_age) {
        gclog_or_tty->print_cr("## %s: region ages are not strictly increasing "
                               "(%d, %d)", name, age, prev_age);
        ret = false;
      }
      prev_age = age;
    }
  }

  return ret;
}
#endif // PRODUCT

void G1CollectorPolicy::record_full_collection_start() {
  _cur_collection_start_sec = os::elapsedTime();
  // Release the future to-space so that it is available for compaction into.
  _g1->set_full_collection();
}

void G1CollectorPolicy::record_full_collection_end() {
  // Consider this like a collection pause for the purposes of allocation
  // since last pause.
  double end_sec = os::elapsedTime();
  double full_gc_time_sec = end_sec - _cur_collection_start_sec;
  double full_gc_time_ms = full_gc_time_sec * 1000.0;

  _all_full_gc_times_ms->add(full_gc_time_ms);

895
  update_recent_gc_times(end_sec, full_gc_time_ms);
896 897 898 899 900 901 902 903 904

  _g1->clear_full_collection();

  // "Nuke" the heuristics that control the fully/partially young GC
  // transitions and make sure we start with fully young GCs after the
  // Full GC.
  set_full_young_gcs(true);
  _last_full_young_gc = false;
  _should_revert_to_full_young_gcs = false;
905 906
  clear_initiate_conc_mark_if_possible();
  clear_during_initial_mark_pause();
907 908 909 910 911 912 913 914
  _known_garbage_bytes = 0;
  _known_garbage_ratio = 0.0;
  _in_marking_window = false;
  _in_marking_window_im = false;

  _short_lived_surv_rate_group->start_adding_regions();
  // also call this on any additional surv rate groups

915 916
  record_survivor_regions(0, NULL, NULL);

917 918 919 920
  _prev_region_num_young   = _region_num_young;
  _prev_region_num_tenured = _region_num_tenured;

  _free_regions_at_end_of_collection = _g1->free_regions();
921 922
  // Reset survivors SurvRateGroup.
  _survivor_surv_rate_group->reset();
923
  update_young_list_target_length();
924
}
925 926 927 928 929 930 931 932 933 934

void G1CollectorPolicy::record_stop_world_start() {
  _stop_world_start = os::elapsedTime();
}

void G1CollectorPolicy::record_collection_pause_start(double start_time_sec,
                                                      size_t start_used) {
  if (PrintGCDetails) {
    gclog_or_tty->stamp(PrintGCTimeStamps);
    gclog_or_tty->print("[GC pause");
935
    gclog_or_tty->print(" (%s)", full_young_gcs() ? "young" : "partial");
936 937
  }

938 939 940 941 942
  // We only need to do this here as the policy will only be applied
  // to the GC we're about to start. so, no point is calculating this
  // every time we calculate / recalculate the target young length.
  update_survivors_policy();

943 944 945
  assert(_g1->used() == _g1->recalculate_used(),
         err_msg("sanity, used: "SIZE_FORMAT" recalculate_used: "SIZE_FORMAT,
                 _g1->used(), _g1->recalculate_used()));
946 947 948 949 950 951 952 953 954 955 956 957

  double s_w_t_ms = (start_time_sec - _stop_world_start) * 1000.0;
  _all_stop_world_times_ms->add(s_w_t_ms);
  _stop_world_start = 0.0;

  _cur_collection_start_sec = start_time_sec;
  _cur_collection_pause_used_at_start_bytes = start_used;
  _cur_collection_pause_used_regions_at_start = _g1->used_regions();
  _pending_cards = _g1->pending_card_num();
  _max_pending_cards = _g1->max_pending_card_num();

  _bytes_in_collection_set_before_gc = 0;
958
  _bytes_copied_during_gc = 0;
959

960 961 962 963 964
  YoungList* young_list = _g1->young_list();
  _eden_bytes_before_gc = young_list->eden_used_bytes();
  _survivor_bytes_before_gc = young_list->survivor_used_bytes();
  _capacity_before_gc = _g1->capacity();

965 966 967 968 969
#ifdef DEBUG
  // initialise these to something well known so that we can spot
  // if they are not set properly

  for (int i = 0; i < _parallel_gc_threads; ++i) {
970 971 972 973 974 975 976 977 978 979
    _par_last_gc_worker_start_times_ms[i] = -1234.0;
    _par_last_ext_root_scan_times_ms[i] = -1234.0;
    _par_last_mark_stack_scan_times_ms[i] = -1234.0;
    _par_last_update_rs_times_ms[i] = -1234.0;
    _par_last_update_rs_processed_buffers[i] = -1234.0;
    _par_last_scan_rs_times_ms[i] = -1234.0;
    _par_last_obj_copy_times_ms[i] = -1234.0;
    _par_last_termination_times_ms[i] = -1234.0;
    _par_last_termination_attempts[i] = -1234.0;
    _par_last_gc_worker_end_times_ms[i] = -1234.0;
980
    _par_last_gc_worker_times_ms[i] = -1234.0;
981 982 983 984 985 986 987 988 989 990 991
  }
#endif

  for (int i = 0; i < _aux_num; ++i) {
    _cur_aux_times_ms[i] = 0.0;
    _cur_aux_times_set[i] = false;
  }

  _satb_drain_time_set = false;
  _last_satb_drain_processed_buffers = -1;

992
  _last_young_gc_full = false;
993 994 995

  // do that for any other surv rate groups
  _short_lived_surv_rate_group->stop_adding_regions();
996
  _survivors_age_table.clear();
997

998 999 1000 1001 1002 1003 1004
  assert( verify_young_ages(), "region age verification" );
}

void G1CollectorPolicy::record_mark_closure_time(double mark_closure_time_ms) {
  _mark_closure_time_ms = mark_closure_time_ms;
}

1005
void G1CollectorPolicy::record_concurrent_mark_init_end(double
1006 1007
                                                   mark_init_elapsed_time_ms) {
  _during_marking = true;
1008 1009
  assert(!initiate_conc_mark_if_possible(), "we should have cleared it by now");
  clear_during_initial_mark_pause();
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042
  _cur_mark_stop_world_time_ms = mark_init_elapsed_time_ms;
}

void G1CollectorPolicy::record_concurrent_mark_remark_start() {
  _mark_remark_start_sec = os::elapsedTime();
  _during_marking = false;
}

void G1CollectorPolicy::record_concurrent_mark_remark_end() {
  double end_time_sec = os::elapsedTime();
  double elapsed_time_ms = (end_time_sec - _mark_remark_start_sec)*1000.0;
  _concurrent_mark_remark_times_ms->add(elapsed_time_ms);
  _cur_mark_stop_world_time_ms += elapsed_time_ms;
  _prev_collection_pause_end_ms += elapsed_time_ms;

  _mmu_tracker->add_pause(_mark_remark_start_sec, end_time_sec, true);
}

void G1CollectorPolicy::record_concurrent_mark_cleanup_start() {
  _mark_cleanup_start_sec = os::elapsedTime();
}

void
G1CollectorPolicy::record_concurrent_mark_cleanup_end(size_t freed_bytes,
                                                      size_t max_live_bytes) {
  record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes);
  record_concurrent_mark_cleanup_end_work2();
}

void
G1CollectorPolicy::
record_concurrent_mark_cleanup_end_work1(size_t freed_bytes,
                                         size_t max_live_bytes) {
1043 1044 1045
  if (_n_marks < 2) {
    _n_marks++;
  }
1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064
}

// The important thing about this is that it includes "os::elapsedTime".
void G1CollectorPolicy::record_concurrent_mark_cleanup_end_work2() {
  double end_time_sec = os::elapsedTime();
  double elapsed_time_ms = (end_time_sec - _mark_cleanup_start_sec)*1000.0;
  _concurrent_mark_cleanup_times_ms->add(elapsed_time_ms);
  _cur_mark_stop_world_time_ms += elapsed_time_ms;
  _prev_collection_pause_end_ms += elapsed_time_ms;

  _mmu_tracker->add_pause(_mark_cleanup_start_sec, end_time_sec, true);

  _num_markings++;
  _n_pauses_at_mark_end = _n_pauses;
  _n_marks_since_last_pause++;
}

void
G1CollectorPolicy::record_concurrent_mark_cleanup_completed() {
1065 1066 1067
  _should_revert_to_full_young_gcs = false;
  _last_full_young_gc = true;
  _in_marking_window = false;
1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
}

void G1CollectorPolicy::record_concurrent_pause() {
  if (_stop_world_start > 0.0) {
    double yield_ms = (os::elapsedTime() - _stop_world_start) * 1000.0;
    _all_yield_times_ms->add(yield_ms);
  }
}

void G1CollectorPolicy::record_concurrent_pause_end() {
}

template<class T>
T sum_of(T* sum_arr, int start, int n, int N) {
  T sum = (T)0;
  for (int i = 0; i < n; i++) {
    int j = (start + i) % N;
    sum += sum_arr[j];
  }
  return sum;
}

1090 1091
void G1CollectorPolicy::print_par_stats(int level,
                                        const char* str,
1092
                                        double* data) {
1093 1094
  double min = data[0], max = data[0];
  double total = 0.0;
1095 1096
  LineBuffer buf(level);
  buf.append("[%s (ms):", str);
1097 1098 1099 1100 1101 1102 1103
  for (uint i = 0; i < ParallelGCThreads; ++i) {
    double val = data[i];
    if (val < min)
      min = val;
    if (val > max)
      max = val;
    total += val;
1104
    buf.append("  %3.1lf", val);
1105
  }
1106 1107 1108 1109
  buf.append_and_print_cr("");
  double avg = total / (double) ParallelGCThreads;
  buf.append_and_print_cr(" Avg: %5.1lf, Min: %5.1lf, Max: %5.1lf, Diff: %5.1lf]",
    avg, min, max, max - min);
1110 1111
}

1112 1113
void G1CollectorPolicy::print_par_sizes(int level,
                                        const char* str,
1114
                                        double* data) {
1115 1116
  double min = data[0], max = data[0];
  double total = 0.0;
1117 1118
  LineBuffer buf(level);
  buf.append("[%s :", str);
1119 1120 1121 1122 1123 1124 1125
  for (uint i = 0; i < ParallelGCThreads; ++i) {
    double val = data[i];
    if (val < min)
      min = val;
    if (val > max)
      max = val;
    total += val;
1126
    buf.append(" %d", (int) val);
1127
  }
1128 1129 1130 1131
  buf.append_and_print_cr("");
  double avg = total / (double) ParallelGCThreads;
  buf.append_and_print_cr(" Sum: %d, Avg: %d, Min: %d, Max: %d, Diff: %d]",
    (int)total, (int)avg, (int)min, (int)max, (int)max - (int)min);
1132 1133 1134 1135 1136
}

void G1CollectorPolicy::print_stats (int level,
                                     const char* str,
                                     double value) {
1137
  LineBuffer(level).append_and_print_cr("[%s: %5.1lf ms]", str, value);
1138 1139 1140 1141 1142
}

void G1CollectorPolicy::print_stats (int level,
                                     const char* str,
                                     int value) {
1143
  LineBuffer(level).append_and_print_cr("[%s: %d]", str, value);
1144 1145 1146
}

double G1CollectorPolicy::avg_value (double* data) {
1147
  if (G1CollectedHeap::use_parallel_gc_threads()) {
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157
    double ret = 0.0;
    for (uint i = 0; i < ParallelGCThreads; ++i)
      ret += data[i];
    return ret / (double) ParallelGCThreads;
  } else {
    return data[0];
  }
}

double G1CollectorPolicy::max_value (double* data) {
1158
  if (G1CollectedHeap::use_parallel_gc_threads()) {
1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169
    double ret = data[0];
    for (uint i = 1; i < ParallelGCThreads; ++i)
      if (data[i] > ret)
        ret = data[i];
    return ret;
  } else {
    return data[0];
  }
}

double G1CollectorPolicy::sum_of_values (double* data) {
1170
  if (G1CollectedHeap::use_parallel_gc_threads()) {
1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183
    double sum = 0.0;
    for (uint i = 0; i < ParallelGCThreads; i++)
      sum += data[i];
    return sum;
  } else {
    return data[0];
  }
}

double G1CollectorPolicy::max_sum (double* data1,
                                   double* data2) {
  double ret = data1[0] + data2[0];

1184
  if (G1CollectedHeap::use_parallel_gc_threads()) {
1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196
    for (uint i = 1; i < ParallelGCThreads; ++i) {
      double data = data1[i] + data2[i];
      if (data > ret)
        ret = data;
    }
  }
  return ret;
}

// Anything below that is considered to be zero
#define MIN_TIMER_GRANULARITY 0.0000001

1197
void G1CollectorPolicy::record_collection_pause_end() {
1198 1199
  double end_time_sec = os::elapsedTime();
  double elapsed_ms = _last_pause_time_ms;
1200
  bool parallel = G1CollectedHeap::use_parallel_gc_threads();
1201 1202 1203 1204 1205
  size_t rs_size =
    _cur_collection_pause_used_regions_at_start - collection_set_size();
  size_t cur_used_bytes = _g1->used();
  assert(cur_used_bytes == _g1->recalculate_used(), "It should!");
  bool last_pause_included_initial_mark = false;
1206
  bool update_stats = !_g1->evacuation_failed();
1207 1208 1209 1210 1211 1212 1213 1214 1215

#ifndef PRODUCT
  if (G1YoungSurvRateVerbose) {
    gclog_or_tty->print_cr("");
    _short_lived_surv_rate_group->print();
    // do that for any other surv rate groups too
  }
#endif // PRODUCT

1216 1217 1218
  last_pause_included_initial_mark = during_initial_mark_pause();
  if (last_pause_included_initial_mark)
    record_concurrent_mark_init_end(0.0);
1219

1220
  size_t marking_initiating_used_threshold =
1221
    (_g1->capacity() / 100) * InitiatingHeapOccupancyPercent;
1222

1223 1224
  if (!_g1->mark_in_progress() && !_last_full_young_gc) {
    assert(!last_pause_included_initial_mark, "invariant");
1225 1226
    if (cur_used_bytes > marking_initiating_used_threshold) {
      if (cur_used_bytes > _prev_collection_pause_used_at_end_bytes) {
1227 1228
        assert(!during_initial_mark_pause(), "we should not see this here");

1229 1230 1231 1232 1233 1234 1235 1236 1237
        ergo_verbose3(ErgoConcCycles,
                      "request concurrent cycle initiation",
                      ergo_format_reason("occupancy higher than threshold")
                      ergo_format_byte("occupancy")
                      ergo_format_byte_perc("threshold"),
                      cur_used_bytes,
                      marking_initiating_used_threshold,
                      (double) InitiatingHeapOccupancyPercent);

1238 1239 1240 1241
        // Note: this might have already been set, if during the last
        // pause we decided to start a cycle but at the beginning of
        // this pause we decided to postpone it. That's OK.
        set_initiate_conc_mark_if_possible();
1242 1243 1244 1245 1246 1247 1248 1249 1250
      } else {
        ergo_verbose2(ErgoConcCycles,
                  "do not request concurrent cycle initiation",
                  ergo_format_reason("occupancy lower than previous occupancy")
                  ergo_format_byte("occupancy")
                  ergo_format_byte("previous occupancy"),
                  cur_used_bytes,
                  _prev_collection_pause_used_at_end_bytes);
      }
1251 1252 1253
    }
  }

1254 1255
  _prev_collection_pause_used_at_end_bytes = cur_used_bytes;

1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
  _mmu_tracker->add_pause(end_time_sec - elapsed_ms/1000.0,
                          end_time_sec, false);

  guarantee(_cur_collection_pause_used_regions_at_start >=
            collection_set_size(),
            "Negative RS size?");

  // This assert is exempted when we're doing parallel collection pauses,
  // because the fragmentation caused by the parallel GC allocation buffers
  // can lead to more memory being used during collection than was used
  // before. Best leave this out until the fragmentation problem is fixed.
  // Pauses in which evacuation failed can also lead to negative
  // collections, since no space is reclaimed from a region containing an
  // object whose evacuation failed.
  // Further, we're now always doing parallel collection.  But I'm still
  // leaving this here as a placeholder for a more precise assertion later.
  // (DLD, 10/05.)
  assert((true || parallel) // Always using GC LABs now.
         || _g1->evacuation_failed()
         || _cur_collection_pause_used_at_start_bytes >= cur_used_bytes,
         "Negative collection");

  size_t freed_bytes =
    _cur_collection_pause_used_at_start_bytes - cur_used_bytes;
  size_t surviving_bytes = _collection_set_bytes_used_before - freed_bytes;
1281

1282 1283 1284 1285 1286 1287
  double survival_fraction =
    (double)surviving_bytes/
    (double)_collection_set_bytes_used_before;

  _n_pauses++;

1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307
  double ext_root_scan_time = avg_value(_par_last_ext_root_scan_times_ms);
  double mark_stack_scan_time = avg_value(_par_last_mark_stack_scan_times_ms);
  double update_rs_time = avg_value(_par_last_update_rs_times_ms);
  double update_rs_processed_buffers =
    sum_of_values(_par_last_update_rs_processed_buffers);
  double scan_rs_time = avg_value(_par_last_scan_rs_times_ms);
  double obj_copy_time = avg_value(_par_last_obj_copy_times_ms);
  double termination_time = avg_value(_par_last_termination_times_ms);

  double parallel_known_time = update_rs_time +
                               ext_root_scan_time +
                               mark_stack_scan_time +
                               scan_rs_time +
                               obj_copy_time +
                               termination_time;

  double parallel_other_time = _cur_collection_par_time_ms - parallel_known_time;

  PauseSummary* summary = _summary;

1308
  if (update_stats) {
1309
    _recent_rs_scan_times_ms->add(scan_rs_time);
1310 1311 1312
    _recent_pause_times_ms->add(elapsed_ms);
    _recent_rs_sizes->add(rs_size);

1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333
    MainBodySummary* body_summary = summary->main_body_summary();
    guarantee(body_summary != NULL, "should not be null!");

    if (_satb_drain_time_set)
      body_summary->record_satb_drain_time_ms(_cur_satb_drain_time_ms);
    else
      body_summary->record_satb_drain_time_ms(0.0);

    body_summary->record_ext_root_scan_time_ms(ext_root_scan_time);
    body_summary->record_mark_stack_scan_time_ms(mark_stack_scan_time);
    body_summary->record_update_rs_time_ms(update_rs_time);
    body_summary->record_scan_rs_time_ms(scan_rs_time);
    body_summary->record_obj_copy_time_ms(obj_copy_time);
    if (parallel) {
      body_summary->record_parallel_time_ms(_cur_collection_par_time_ms);
      body_summary->record_clear_ct_time_ms(_cur_clear_ct_time_ms);
      body_summary->record_termination_time_ms(termination_time);
      body_summary->record_parallel_other_time_ms(parallel_other_time);
    }
    body_summary->record_mark_closure_time_ms(_mark_closure_time_ms);

1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367
    // We exempt parallel collection from this check because Alloc Buffer
    // fragmentation can produce negative collections.  Same with evac
    // failure.
    // Further, we're now always doing parallel collection.  But I'm still
    // leaving this here as a placeholder for a more precise assertion later.
    // (DLD, 10/05.
    assert((true || parallel)
           || _g1->evacuation_failed()
           || surviving_bytes <= _collection_set_bytes_used_before,
           "Or else negative collection!");
    _recent_CS_bytes_used_before->add(_collection_set_bytes_used_before);
    _recent_CS_bytes_surviving->add(surviving_bytes);

    // this is where we update the allocation rate of the application
    double app_time_ms =
      (_cur_collection_start_sec * 1000.0 - _prev_collection_pause_end_ms);
    if (app_time_ms < MIN_TIMER_GRANULARITY) {
      // This usually happens due to the timer not having the required
      // granularity. Some Linuxes are the usual culprits.
      // We'll just set it to something (arbitrarily) small.
      app_time_ms = 1.0;
    }
    size_t regions_allocated =
      (_region_num_young - _prev_region_num_young) +
      (_region_num_tenured - _prev_region_num_tenured);
    double alloc_rate_ms = (double) regions_allocated / app_time_ms;
    _alloc_rate_ms_seq->add(alloc_rate_ms);
    _prev_region_num_young   = _region_num_young;
    _prev_region_num_tenured = _region_num_tenured;

    double interval_ms =
      (end_time_sec - _recent_prev_end_times_for_all_gcs_sec->oldest()) * 1000.0;
    update_recent_gc_times(end_time_sec, elapsed_ms);
    _recent_avg_pause_time_ratio = _recent_gc_times_ms->sum()/interval_ms;
1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379
    if (recent_avg_pause_time_ratio() < 0.0 ||
        (recent_avg_pause_time_ratio() - 1.0 > 0.0)) {
#ifndef PRODUCT
      // Dump info to allow post-facto debugging
      gclog_or_tty->print_cr("recent_avg_pause_time_ratio() out of bounds");
      gclog_or_tty->print_cr("-------------------------------------------");
      gclog_or_tty->print_cr("Recent GC Times (ms):");
      _recent_gc_times_ms->dump();
      gclog_or_tty->print_cr("(End Time=%3.3f) Recent GC End Times (s):", end_time_sec);
      _recent_prev_end_times_for_all_gcs_sec->dump();
      gclog_or_tty->print_cr("GC = %3.3f, Interval = %3.3f, Ratio = %3.3f",
                             _recent_gc_times_ms->sum(), interval_ms, recent_avg_pause_time_ratio());
1380 1381 1382 1383 1384
      // In debug mode, terminate the JVM if the user wants to debug at this point.
      assert(!G1FailOnFPError, "Debugging data for CR 6898948 has been dumped above");
#endif  // !PRODUCT
      // Clip ratio between 0.0 and 1.0, and continue. This will be fixed in
      // CR 6902692 by redoing the manner in which the ratio is incrementally computed.
1385 1386 1387 1388 1389 1390 1391
      if (_recent_avg_pause_time_ratio < 0.0) {
        _recent_avg_pause_time_ratio = 0.0;
      } else {
        assert(_recent_avg_pause_time_ratio - 1.0 > 0.0, "Ctl-point invariant");
        _recent_avg_pause_time_ratio = 1.0;
      }
    }
1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402
  }

  if (G1PolicyVerbose > 1) {
    gclog_or_tty->print_cr("   Recording collection pause(%d)", _n_pauses);
  }

  if (G1PolicyVerbose > 1) {
    gclog_or_tty->print_cr("      ET: %10.6f ms           (avg: %10.6f ms)\n"
                           "       ET-RS:  %10.6f ms      (avg: %10.6f ms)\n"
                           "      |RS|: " SIZE_FORMAT,
                           elapsed_ms, recent_avg_time_for_pauses_ms(),
1403
                           scan_rs_time, recent_avg_time_for_rs_scan_ms(),
1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
                           rs_size);

    gclog_or_tty->print_cr("       Used at start: " SIZE_FORMAT"K"
                           "       At end " SIZE_FORMAT "K\n"
                           "       garbage      : " SIZE_FORMAT "K"
                           "       of     " SIZE_FORMAT "K\n"
                           "       survival     : %6.2f%%  (%6.2f%% avg)",
                           _cur_collection_pause_used_at_start_bytes/K,
                           _g1->used()/K, freed_bytes/K,
                           _collection_set_bytes_used_before/K,
                           survival_fraction*100.0,
                           recent_avg_survival_fraction()*100.0);
    gclog_or_tty->print_cr("       Recent %% gc pause time: %6.2f",
                           recent_avg_pause_time_ratio() * 100.0);
  }

  double other_time_ms = elapsed_ms;

1422 1423 1424
  if (_satb_drain_time_set) {
    other_time_ms -= _cur_satb_drain_time_ms;
  }
1425

1426 1427 1428 1429 1430 1431 1432
  if (parallel) {
    other_time_ms -= _cur_collection_par_time_ms + _cur_clear_ct_time_ms;
  } else {
    other_time_ms -=
      update_rs_time +
      ext_root_scan_time + mark_stack_scan_time +
      scan_rs_time + obj_copy_time;
1433 1434 1435
  }

  if (PrintGCDetails) {
1436
    gclog_or_tty->print_cr("%s, %1.8lf secs]",
1437 1438 1439
                           (last_pause_included_initial_mark) ? " (initial-mark)" : "",
                           elapsed_ms / 1000.0);

1440 1441 1442 1443 1444 1445 1446 1447
    if (_satb_drain_time_set) {
      print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms);
    }
    if (_last_satb_drain_processed_buffers >= 0) {
      print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers);
    }
    if (parallel) {
      print_stats(1, "Parallel Time", _cur_collection_par_time_ms);
1448
      print_par_stats(2, "GC Worker Start Time", _par_last_gc_worker_start_times_ms);
1449
      print_par_stats(2, "Update RS", _par_last_update_rs_times_ms);
1450 1451 1452
      print_par_sizes(3, "Processed Buffers", _par_last_update_rs_processed_buffers);
      print_par_stats(2, "Ext Root Scanning", _par_last_ext_root_scan_times_ms);
      print_par_stats(2, "Mark Stack Scanning", _par_last_mark_stack_scan_times_ms);
1453 1454 1455
      print_par_stats(2, "Scan RS", _par_last_scan_rs_times_ms);
      print_par_stats(2, "Object Copy", _par_last_obj_copy_times_ms);
      print_par_stats(2, "Termination", _par_last_termination_times_ms);
1456 1457 1458 1459 1460 1461 1462 1463
      print_par_sizes(3, "Termination Attempts", _par_last_termination_attempts);
      print_par_stats(2, "GC Worker End Time", _par_last_gc_worker_end_times_ms);

      for (int i = 0; i < _parallel_gc_threads; i++) {
        _par_last_gc_worker_times_ms[i] = _par_last_gc_worker_end_times_ms[i] - _par_last_gc_worker_start_times_ms[i];
      }
      print_par_stats(2, "GC Worker Times", _par_last_gc_worker_times_ms);

1464
      print_stats(2, "Parallel Other", parallel_other_time);
1465 1466 1467 1468 1469 1470 1471 1472 1473
      print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
    } else {
      print_stats(1, "Update RS", update_rs_time);
      print_stats(2, "Processed Buffers",
                  (int)update_rs_processed_buffers);
      print_stats(1, "Ext Root Scanning", ext_root_scan_time);
      print_stats(1, "Mark Stack Scanning", mark_stack_scan_time);
      print_stats(1, "Scan RS", scan_rs_time);
      print_stats(1, "Object Copying", obj_copy_time);
1474
    }
1475 1476 1477 1478 1479 1480 1481 1482 1483
#ifndef PRODUCT
    print_stats(1, "Cur Clear CC", _cur_clear_cc_time_ms);
    print_stats(1, "Cum Clear CC", _cum_clear_cc_time_ms);
    print_stats(1, "Min Clear CC", _min_clear_cc_time_ms);
    print_stats(1, "Max Clear CC", _max_clear_cc_time_ms);
    if (_num_cc_clears > 0) {
      print_stats(1, "Avg Clear CC", _cum_clear_cc_time_ms / ((double)_num_cc_clears));
    }
#endif
1484
    print_stats(1, "Other", other_time_ms);
1485
    print_stats(2, "Choose CSet", _recorded_young_cset_choice_time_ms);
1486 1487
    print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
    print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
1488

1489 1490 1491 1492 1493 1494 1495 1496 1497 1498
    for (int i = 0; i < _aux_num; ++i) {
      if (_cur_aux_times_set[i]) {
        char buffer[96];
        sprintf(buffer, "Aux%d", i);
        print_stats(1, buffer, _cur_aux_times_ms[i]);
      }
    }
  }

  _all_pause_times_ms->add(elapsed_ms);
1499 1500 1501 1502
  if (update_stats) {
    summary->record_total_time_ms(elapsed_ms);
    summary->record_other_time_ms(other_time_ms);
  }
1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521
  for (int i = 0; i < _aux_num; ++i)
    if (_cur_aux_times_set[i])
      _all_aux_times_ms[i].add(_cur_aux_times_ms[i]);

  // Reset marks-between-pauses counter.
  _n_marks_since_last_pause = 0;

  // Update the efficiency-since-mark vars.
  double proc_ms = elapsed_ms * (double) _parallel_gc_threads;
  if (elapsed_ms < MIN_TIMER_GRANULARITY) {
    // This usually happens due to the timer not having the required
    // granularity. Some Linuxes are the usual culprits.
    // We'll just set it to something (arbitrarily) small.
    proc_ms = 1.0;
  }
  double cur_efficiency = (double) freed_bytes / proc_ms;

  bool new_in_marking_window = _in_marking_window;
  bool new_in_marking_window_im = false;
1522
  if (during_initial_mark_pause()) {
1523 1524 1525 1526
    new_in_marking_window = true;
    new_in_marking_window_im = true;
  }

1527
  if (_last_full_young_gc) {
1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538
    if (!last_pause_included_initial_mark) {
      ergo_verbose2(ErgoPartiallyYoungGCs,
                    "start partially-young GCs",
                    ergo_format_byte_perc("known garbage"),
                    _known_garbage_bytes, _known_garbage_ratio * 100.0);
      set_full_young_gcs(false);
    } else {
      ergo_verbose0(ErgoPartiallyYoungGCs,
                    "do not start partially-young GCs",
                    ergo_format_reason("concurrent cycle is about to start"));
    }
1539 1540
    _last_full_young_gc = false;
  }
1541

1542
  if ( !_last_young_gc_full ) {
1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572
    if (_should_revert_to_full_young_gcs) {
      ergo_verbose2(ErgoPartiallyYoungGCs,
                    "end partially-young GCs",
                    ergo_format_reason("partially-young GCs end requested")
                    ergo_format_byte_perc("known garbage"),
                    _known_garbage_bytes, _known_garbage_ratio * 100.0);
      set_full_young_gcs(true);
    } else if (_known_garbage_ratio < 0.05) {
      ergo_verbose3(ErgoPartiallyYoungGCs,
               "end partially-young GCs",
               ergo_format_reason("known garbage percent lower than threshold")
               ergo_format_byte_perc("known garbage")
               ergo_format_perc("threshold"),
               _known_garbage_bytes, _known_garbage_ratio * 100.0,
               0.05 * 100.0);
      set_full_young_gcs(true);
    } else if (adaptive_young_list_length() &&
              (get_gc_eff_factor() * cur_efficiency < predict_young_gc_eff())) {
      ergo_verbose5(ErgoPartiallyYoungGCs,
                    "end partially-young GCs",
                    ergo_format_reason("current GC efficiency lower than "
                                       "predicted fully-young GC efficiency")
                    ergo_format_double("GC efficiency factor")
                    ergo_format_double("current GC efficiency")
                    ergo_format_double("predicted fully-young GC efficiency")
                    ergo_format_byte_perc("known garbage"),
                    get_gc_eff_factor(), cur_efficiency,
                    predict_young_gc_eff(),
                    _known_garbage_bytes, _known_garbage_ratio * 100.0);
      set_full_young_gcs(true);
1573
    }
1574 1575
  }
  _should_revert_to_full_young_gcs = false;
1576

1577 1578
  if (_last_young_gc_full && !_during_marking) {
    _young_gc_eff_seq->add(cur_efficiency);
1579 1580 1581 1582 1583 1584 1585
  }

  _short_lived_surv_rate_group->start_adding_regions();
  // do that for any other surv rate groupsx

  // <NEW PREDICTION>

1586
  if (update_stats) {
1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634
    double pause_time_ms = elapsed_ms;

    size_t diff = 0;
    if (_max_pending_cards >= _pending_cards)
      diff = _max_pending_cards - _pending_cards;
    _pending_card_diff_seq->add((double) diff);

    double cost_per_card_ms = 0.0;
    if (_pending_cards > 0) {
      cost_per_card_ms = update_rs_time / (double) _pending_cards;
      _cost_per_card_ms_seq->add(cost_per_card_ms);
    }

    size_t cards_scanned = _g1->cards_scanned();

    double cost_per_entry_ms = 0.0;
    if (cards_scanned > 10) {
      cost_per_entry_ms = scan_rs_time / (double) cards_scanned;
      if (_last_young_gc_full)
        _cost_per_entry_ms_seq->add(cost_per_entry_ms);
      else
        _partially_young_cost_per_entry_ms_seq->add(cost_per_entry_ms);
    }

    if (_max_rs_lengths > 0) {
      double cards_per_entry_ratio =
        (double) cards_scanned / (double) _max_rs_lengths;
      if (_last_young_gc_full)
        _fully_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio);
      else
        _partially_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio);
    }

    size_t rs_length_diff = _max_rs_lengths - _recorded_rs_lengths;
    if (rs_length_diff >= 0)
      _rs_length_diff_seq->add((double) rs_length_diff);

    size_t copied_bytes = surviving_bytes;
    double cost_per_byte_ms = 0.0;
    if (copied_bytes > 0) {
      cost_per_byte_ms = obj_copy_time / (double) copied_bytes;
      if (_in_marking_window)
        _cost_per_byte_ms_during_cm_seq->add(cost_per_byte_ms);
      else
        _cost_per_byte_ms_seq->add(cost_per_byte_ms);
    }

    double all_other_time_ms = pause_time_ms -
1635
      (update_rs_time + scan_rs_time + obj_copy_time +
1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661
       _mark_closure_time_ms + termination_time);

    double young_other_time_ms = 0.0;
    if (_recorded_young_regions > 0) {
      young_other_time_ms =
        _recorded_young_cset_choice_time_ms +
        _recorded_young_free_cset_time_ms;
      _young_other_cost_per_region_ms_seq->add(young_other_time_ms /
                                             (double) _recorded_young_regions);
    }
    double non_young_other_time_ms = 0.0;
    if (_recorded_non_young_regions > 0) {
      non_young_other_time_ms =
        _recorded_non_young_cset_choice_time_ms +
        _recorded_non_young_free_cset_time_ms;

      _non_young_other_cost_per_region_ms_seq->add(non_young_other_time_ms /
                                         (double) _recorded_non_young_regions);
    }

    double constant_other_time_ms = all_other_time_ms -
      (young_other_time_ms + non_young_other_time_ms);
    _constant_other_time_ms_seq->add(constant_other_time_ms);

    double survival_ratio = 0.0;
    if (_bytes_in_collection_set_before_gc > 0) {
1662 1663
      survival_ratio = (double) _bytes_copied_during_gc /
                                   (double) _bytes_in_collection_set_before_gc;
1664 1665 1666 1667 1668 1669 1670
    }

    _pending_cards_seq->add((double) _pending_cards);
    _scanned_cards_seq->add((double) cards_scanned);
    _rs_lengths_seq->add((double) _max_rs_lengths);

    double expensive_region_limit_ms =
J
johnc 已提交
1671
      (double) MaxGCPauseMillis - predict_constant_other_time_ms();
1672 1673 1674
    if (expensive_region_limit_ms < 0.0) {
      // this means that the other time was predicted to be longer than
      // than the max pause time
J
johnc 已提交
1675
      expensive_region_limit_ms = (double) MaxGCPauseMillis;
1676 1677 1678 1679 1680 1681
    }
    _expensive_region_limit_ms = expensive_region_limit_ms;

    if (PREDICTIONS_VERBOSE) {
      gclog_or_tty->print_cr("");
      gclog_or_tty->print_cr("PREDICTIONS %1.4lf %d "
1682
                    "REGIONS %d %d %d "
1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725
                    "PENDING_CARDS %d %d "
                    "CARDS_SCANNED %d %d "
                    "RS_LENGTHS %d %d "
                    "RS_UPDATE %1.6lf %1.6lf RS_SCAN %1.6lf %1.6lf "
                    "SURVIVAL_RATIO %1.6lf %1.6lf "
                    "OBJECT_COPY %1.6lf %1.6lf OTHER_CONSTANT %1.6lf %1.6lf "
                    "OTHER_YOUNG %1.6lf %1.6lf "
                    "OTHER_NON_YOUNG %1.6lf %1.6lf "
                    "VTIME_DIFF %1.6lf TERMINATION %1.6lf "
                    "ELAPSED %1.6lf %1.6lf ",
                    _cur_collection_start_sec,
                    (!_last_young_gc_full) ? 2 :
                    (last_pause_included_initial_mark) ? 1 : 0,
                    _recorded_region_num,
                    _recorded_young_regions,
                    _recorded_non_young_regions,
                    _predicted_pending_cards, _pending_cards,
                    _predicted_cards_scanned, cards_scanned,
                    _predicted_rs_lengths, _max_rs_lengths,
                    _predicted_rs_update_time_ms, update_rs_time,
                    _predicted_rs_scan_time_ms, scan_rs_time,
                    _predicted_survival_ratio, survival_ratio,
                    _predicted_object_copy_time_ms, obj_copy_time,
                    _predicted_constant_other_time_ms, constant_other_time_ms,
                    _predicted_young_other_time_ms, young_other_time_ms,
                    _predicted_non_young_other_time_ms,
                    non_young_other_time_ms,
                    _vtime_diff_ms, termination_time,
                    _predicted_pause_time_ms, elapsed_ms);
    }

    if (G1PolicyVerbose > 0) {
      gclog_or_tty->print_cr("Pause Time, predicted: %1.4lfms (predicted %s), actual: %1.4lfms",
                    _predicted_pause_time_ms,
                    (_within_target) ? "within" : "outside",
                    elapsed_ms);
    }

  }

  _in_marking_window = new_in_marking_window;
  _in_marking_window_im = new_in_marking_window_im;
  _free_regions_at_end_of_collection = _g1->free_regions();
1726
  update_young_list_target_length();
1727

1728
  // Note that _mmu_tracker->max_gc_time() returns the time in seconds.
1729
  double update_rs_time_goal_ms = _mmu_tracker->max_gc_time() * MILLIUNITS * G1RSetUpdatingPauseTimePercent / 100.0;
1730
  adjust_concurrent_refinement(update_rs_time, update_rs_processed_buffers, update_rs_time_goal_ms);
1731 1732 1733
  // </NEW PREDICTION>
}

1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746
#define EXT_SIZE_FORMAT "%d%s"
#define EXT_SIZE_PARAMS(bytes)                                  \
  byte_size_in_proper_unit((bytes)),                            \
  proper_unit_for_byte_size((bytes))

void G1CollectorPolicy::print_heap_transition() {
  if (PrintGCDetails) {
    YoungList* young_list = _g1->young_list();
    size_t eden_bytes = young_list->eden_used_bytes();
    size_t survivor_bytes = young_list->survivor_used_bytes();
    size_t used_before_gc = _cur_collection_pause_used_at_start_bytes;
    size_t used = _g1->used();
    size_t capacity = _g1->capacity();
1747 1748
    size_t eden_capacity =
      (_young_list_target_length * HeapRegion::GrainBytes) - survivor_bytes;
1749 1750

    gclog_or_tty->print_cr(
1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766
      "   [Eden: "EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")->"EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT") "
      "Survivors: "EXT_SIZE_FORMAT"->"EXT_SIZE_FORMAT" "
      "Heap: "EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")->"
      EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")]",
      EXT_SIZE_PARAMS(_eden_bytes_before_gc),
      EXT_SIZE_PARAMS(_prev_eden_capacity),
      EXT_SIZE_PARAMS(eden_bytes),
      EXT_SIZE_PARAMS(eden_capacity),
      EXT_SIZE_PARAMS(_survivor_bytes_before_gc),
      EXT_SIZE_PARAMS(survivor_bytes),
      EXT_SIZE_PARAMS(used_before_gc),
      EXT_SIZE_PARAMS(_capacity_before_gc),
      EXT_SIZE_PARAMS(used),
      EXT_SIZE_PARAMS(capacity));

    _prev_eden_capacity = eden_capacity;
1767 1768 1769 1770 1771 1772 1773
  } else if (PrintGC) {
    _g1->print_size_transition(gclog_or_tty,
                               _cur_collection_pause_used_at_start_bytes,
                               _g1->used(), _g1->capacity());
  }
}

1774 1775
// <NEW PREDICTION>

1776 1777 1778 1779 1780 1781
void G1CollectorPolicy::adjust_concurrent_refinement(double update_rs_time,
                                                     double update_rs_processed_buffers,
                                                     double goal_ms) {
  DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
  ConcurrentG1Refine *cg1r = G1CollectedHeap::heap()->concurrent_g1_refine();

1782
  if (G1UseAdaptiveConcRefinement) {
1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816
    const int k_gy = 3, k_gr = 6;
    const double inc_k = 1.1, dec_k = 0.9;

    int g = cg1r->green_zone();
    if (update_rs_time > goal_ms) {
      g = (int)(g * dec_k);  // Can become 0, that's OK. That would mean a mutator-only processing.
    } else {
      if (update_rs_time < goal_ms && update_rs_processed_buffers > g) {
        g = (int)MAX2(g * inc_k, g + 1.0);
      }
    }
    // Change the refinement threads params
    cg1r->set_green_zone(g);
    cg1r->set_yellow_zone(g * k_gy);
    cg1r->set_red_zone(g * k_gr);
    cg1r->reinitialize_threads();

    int processing_threshold_delta = MAX2((int)(cg1r->green_zone() * sigma()), 1);
    int processing_threshold = MIN2(cg1r->green_zone() + processing_threshold_delta,
                                    cg1r->yellow_zone());
    // Change the barrier params
    dcqs.set_process_completed_threshold(processing_threshold);
    dcqs.set_max_completed_queue(cg1r->red_zone());
  }

  int curr_queue_size = dcqs.completed_buffers_num();
  if (curr_queue_size >= cg1r->yellow_zone()) {
    dcqs.set_completed_queue_padding(curr_queue_size);
  } else {
    dcqs.set_completed_queue_padding(0);
  }
  dcqs.notify_if_necessary();
}

1817 1818 1819 1820 1821 1822
double
G1CollectorPolicy::
predict_young_collection_elapsed_time_ms(size_t adjustment) {
  guarantee( adjustment == 0 || adjustment == 1, "invariant" );

  G1CollectedHeap* g1h = G1CollectedHeap::heap();
1823
  size_t young_num = g1h->young_list()->length();
1824 1825 1826 1827 1828
  if (young_num == 0)
    return 0.0;

  young_num += adjustment;
  size_t pending_cards = predict_pending_cards();
1829
  size_t rs_lengths = g1h->young_list()->sampled_rs_lengths() +
1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902
                      predict_rs_length_diff();
  size_t card_num;
  if (full_young_gcs())
    card_num = predict_young_card_num(rs_lengths);
  else
    card_num = predict_non_young_card_num(rs_lengths);
  size_t young_byte_size = young_num * HeapRegion::GrainBytes;
  double accum_yg_surv_rate =
    _short_lived_surv_rate_group->accum_surv_rate(adjustment);

  size_t bytes_to_copy =
    (size_t) (accum_yg_surv_rate * (double) HeapRegion::GrainBytes);

  return
    predict_rs_update_time_ms(pending_cards) +
    predict_rs_scan_time_ms(card_num) +
    predict_object_copy_time_ms(bytes_to_copy) +
    predict_young_other_time_ms(young_num) +
    predict_constant_other_time_ms();
}

double
G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards) {
  size_t rs_length = predict_rs_length_diff();
  size_t card_num;
  if (full_young_gcs())
    card_num = predict_young_card_num(rs_length);
  else
    card_num = predict_non_young_card_num(rs_length);
  return predict_base_elapsed_time_ms(pending_cards, card_num);
}

double
G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards,
                                                size_t scanned_cards) {
  return
    predict_rs_update_time_ms(pending_cards) +
    predict_rs_scan_time_ms(scanned_cards) +
    predict_constant_other_time_ms();
}

double
G1CollectorPolicy::predict_region_elapsed_time_ms(HeapRegion* hr,
                                                  bool young) {
  size_t rs_length = hr->rem_set()->occupied();
  size_t card_num;
  if (full_young_gcs())
    card_num = predict_young_card_num(rs_length);
  else
    card_num = predict_non_young_card_num(rs_length);
  size_t bytes_to_copy = predict_bytes_to_copy(hr);

  double region_elapsed_time_ms =
    predict_rs_scan_time_ms(card_num) +
    predict_object_copy_time_ms(bytes_to_copy);

  if (young)
    region_elapsed_time_ms += predict_young_other_time_ms(1);
  else
    region_elapsed_time_ms += predict_non_young_other_time_ms(1);

  return region_elapsed_time_ms;
}

size_t
G1CollectorPolicy::predict_bytes_to_copy(HeapRegion* hr) {
  size_t bytes_to_copy;
  if (hr->is_marked())
    bytes_to_copy = hr->max_live_bytes();
  else {
    guarantee( hr->is_young() && hr->age_in_surv_rate_group() != -1,
               "invariant" );
    int age = hr->age_in_surv_rate_group();
1903
    double yg_surv_rate = predict_yg_surv_rate(age, hr->surv_rate_group());
1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919
    bytes_to_copy = (size_t) ((double) hr->used() * yg_surv_rate);
  }

  return bytes_to_copy;
}

void
G1CollectorPolicy::start_recording_regions() {
  _recorded_rs_lengths            = 0;
  _recorded_young_regions         = 0;
  _recorded_non_young_regions     = 0;

#if PREDICTIONS_VERBOSE
  _recorded_marked_bytes          = 0;
  _recorded_young_bytes           = 0;
  _predicted_bytes_to_copy        = 0;
1920 1921
  _predicted_rs_lengths           = 0;
  _predicted_cards_scanned        = 0;
1922 1923 1924 1925
#endif // PREDICTIONS_VERBOSE
}

void
1926
G1CollectorPolicy::record_cset_region_info(HeapRegion* hr, bool young) {
1927
#if PREDICTIONS_VERBOSE
1928
  if (!young) {
1929 1930 1931 1932 1933 1934 1935 1936 1937 1938
    _recorded_marked_bytes += hr->max_live_bytes();
  }
  _predicted_bytes_to_copy += predict_bytes_to_copy(hr);
#endif // PREDICTIONS_VERBOSE

  size_t rs_length = hr->rem_set()->occupied();
  _recorded_rs_lengths += rs_length;
}

void
1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961
G1CollectorPolicy::record_non_young_cset_region(HeapRegion* hr) {
  assert(!hr->is_young(), "should not call this");
  ++_recorded_non_young_regions;
  record_cset_region_info(hr, false);
}

void
G1CollectorPolicy::set_recorded_young_regions(size_t n_regions) {
  _recorded_young_regions = n_regions;
}

void G1CollectorPolicy::set_recorded_young_bytes(size_t bytes) {
#if PREDICTIONS_VERBOSE
  _recorded_young_bytes = bytes;
#endif // PREDICTIONS_VERBOSE
}

void G1CollectorPolicy::set_recorded_rs_lengths(size_t rs_lengths) {
  _recorded_rs_lengths = rs_lengths;
}

void G1CollectorPolicy::set_predicted_bytes_to_copy(size_t bytes) {
  _predicted_bytes_to_copy = bytes;
1962 1963 1964 1965
}

void
G1CollectorPolicy::end_recording_regions() {
1966 1967 1968 1969
  // The _predicted_pause_time_ms field is referenced in code
  // not under PREDICTIONS_VERBOSE. Let's initialize it.
  _predicted_pause_time_ms = -1.0;

1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
#if PREDICTIONS_VERBOSE
  _predicted_pending_cards = predict_pending_cards();
  _predicted_rs_lengths = _recorded_rs_lengths + predict_rs_length_diff();
  if (full_young_gcs())
    _predicted_cards_scanned += predict_young_card_num(_predicted_rs_lengths);
  else
    _predicted_cards_scanned +=
      predict_non_young_card_num(_predicted_rs_lengths);
  _recorded_region_num = _recorded_young_regions + _recorded_non_young_regions;

  _predicted_rs_update_time_ms =
    predict_rs_update_time_ms(_g1->pending_card_num());
  _predicted_rs_scan_time_ms =
    predict_rs_scan_time_ms(_predicted_cards_scanned);
  _predicted_object_copy_time_ms =
    predict_object_copy_time_ms(_predicted_bytes_to_copy);
  _predicted_constant_other_time_ms =
    predict_constant_other_time_ms();
  _predicted_young_other_time_ms =
    predict_young_other_time_ms(_recorded_young_regions);
  _predicted_non_young_other_time_ms =
    predict_non_young_other_time_ms(_recorded_non_young_regions);

  _predicted_pause_time_ms =
    _predicted_rs_update_time_ms +
    _predicted_rs_scan_time_ms +
    _predicted_object_copy_time_ms +
    _predicted_constant_other_time_ms +
    _predicted_young_other_time_ms +
    _predicted_non_young_other_time_ms;
#endif // PREDICTIONS_VERBOSE
}

void G1CollectorPolicy::check_if_region_is_too_expensive(double
                                                           predicted_time_ms) {
  // I don't think we need to do this when in young GC mode since
  // marking will be initiated next time we hit the soft limit anyway...
  if (predicted_time_ms > _expensive_region_limit_ms) {
2008 2009 2010 2011 2012 2013
    ergo_verbose2(ErgoPartiallyYoungGCs,
              "request partially-young GCs end",
              ergo_format_reason("predicted region time higher than threshold")
              ergo_format_ms("predicted region time")
              ergo_format_ms("threshold"),
              predicted_time_ms, _expensive_region_limit_ms);
2014 2015
    // no point in doing another partial one
    _should_revert_to_full_young_gcs = true;
2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029
  }
}

// </NEW PREDICTION>


void G1CollectorPolicy::update_recent_gc_times(double end_time_sec,
                                               double elapsed_ms) {
  _recent_gc_times_ms->add(elapsed_ms);
  _recent_prev_end_times_for_all_gcs_sec->add(end_time_sec);
  _prev_collection_pause_end_ms = end_time_sec * 1000.0;
}

double G1CollectorPolicy::recent_avg_time_for_pauses_ms() {
2030 2031 2032 2033
  if (_recent_pause_times_ms->num() == 0) {
    return (double) MaxGCPauseMillis;
  }
  return _recent_pause_times_ms->avg();
2034 2035
}

2036 2037
double G1CollectorPolicy::recent_avg_time_for_rs_scan_ms() {
  if (_recent_rs_scan_times_ms->num() == 0) {
J
johnc 已提交
2038
    return (double)MaxGCPauseMillis/3.0;
2039 2040
  }
  return _recent_rs_scan_times_ms->avg();
2041 2042 2043
}

int G1CollectorPolicy::number_of_recent_gcs() {
2044
  assert(_recent_rs_scan_times_ms->num() ==
2045 2046 2047 2048 2049
         _recent_pause_times_ms->num(), "Sequence out of sync");
  assert(_recent_pause_times_ms->num() ==
         _recent_CS_bytes_used_before->num(), "Sequence out of sync");
  assert(_recent_CS_bytes_used_before->num() ==
         _recent_CS_bytes_surviving->num(), "Sequence out of sync");
2050

2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074
  return _recent_pause_times_ms->num();
}

double G1CollectorPolicy::recent_avg_survival_fraction() {
  return recent_avg_survival_fraction_work(_recent_CS_bytes_surviving,
                                           _recent_CS_bytes_used_before);
}

double G1CollectorPolicy::last_survival_fraction() {
  return last_survival_fraction_work(_recent_CS_bytes_surviving,
                                     _recent_CS_bytes_used_before);
}

double
G1CollectorPolicy::recent_avg_survival_fraction_work(TruncatedSeq* surviving,
                                                     TruncatedSeq* before) {
  assert(surviving->num() == before->num(), "Sequence out of sync");
  if (before->sum() > 0.0) {
      double recent_survival_rate = surviving->sum() / before->sum();
      // We exempt parallel collection from this check because Alloc Buffer
      // fragmentation can produce negative collections.
      // Further, we're now always doing parallel collection.  But I'm still
      // leaving this here as a placeholder for a more precise assertion later.
      // (DLD, 10/05.)
2075
      assert((true || G1CollectedHeap::use_parallel_gc_threads()) ||
2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094
             _g1->evacuation_failed() ||
             recent_survival_rate <= 1.0, "Or bad frac");
      return recent_survival_rate;
  } else {
    return 1.0; // Be conservative.
  }
}

double
G1CollectorPolicy::last_survival_fraction_work(TruncatedSeq* surviving,
                                               TruncatedSeq* before) {
  assert(surviving->num() == before->num(), "Sequence out of sync");
  if (surviving->num() > 0 && before->last() > 0.0) {
    double last_survival_rate = surviving->last() / before->last();
    // We exempt parallel collection from this check because Alloc Buffer
    // fragmentation can produce negative collections.
    // Further, we're now always doing parallel collection.  But I'm still
    // leaving this here as a placeholder for a more precise assertion later.
    // (DLD, 10/05.)
2095
    assert((true || G1CollectedHeap::use_parallel_gc_threads()) ||
2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122
           last_survival_rate <= 1.0, "Or bad frac");
    return last_survival_rate;
  } else {
    return 1.0;
  }
}

static const int survival_min_obs = 5;
static double survival_min_obs_limits[] = { 0.9, 0.7, 0.5, 0.3, 0.1 };
static const double min_survival_rate = 0.1;

double
G1CollectorPolicy::conservative_avg_survival_fraction_work(double avg,
                                                           double latest) {
  double res = avg;
  if (number_of_recent_gcs() < survival_min_obs) {
    res = MAX2(res, survival_min_obs_limits[number_of_recent_gcs()]);
  }
  res = MAX2(res, latest);
  res = MAX2(res, min_survival_rate);
  // In the parallel case, LAB fragmentation can produce "negative
  // collections"; so can evac failure.  Cap at 1.0
  res = MIN2(res, 1.0);
  return res;
}

size_t G1CollectorPolicy::expansion_amount() {
2123 2124 2125
  double recent_gc_overhead = recent_avg_pause_time_ratio() * 100.0;
  double threshold = _gc_overhead_perc;
  if (recent_gc_overhead > threshold) {
J
johnc 已提交
2126 2127 2128 2129
    // We will double the existing space, or take
    // G1ExpandByPercentOfAvailable % of the available expansion
    // space, whichever is smaller, bounded below by a minimum
    // expansion (unless that's all that's left.)
2130
    const size_t min_expand_bytes = 1*M;
2131
    size_t reserved_bytes = _g1->max_capacity();
2132 2133 2134 2135
    size_t committed_bytes = _g1->capacity();
    size_t uncommitted_bytes = reserved_bytes - committed_bytes;
    size_t expand_bytes;
    size_t expand_bytes_via_pct =
J
johnc 已提交
2136
      uncommitted_bytes * G1ExpandByPercentOfAvailable / 100;
2137 2138 2139
    expand_bytes = MIN2(expand_bytes_via_pct, committed_bytes);
    expand_bytes = MAX2(expand_bytes, min_expand_bytes);
    expand_bytes = MIN2(expand_bytes, uncommitted_bytes);
2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152

    ergo_verbose5(ErgoHeapSizing,
                  "attempt heap expansion",
                  ergo_format_reason("recent GC overhead higher than "
                                     "threshold after GC")
                  ergo_format_perc("recent GC overhead")
                  ergo_format_perc("threshold")
                  ergo_format_byte("uncommitted")
                  ergo_format_byte_perc("calculated expansion amount"),
                  recent_gc_overhead, threshold,
                  uncommitted_bytes,
                  expand_bytes_via_pct, (double) G1ExpandByPercentOfAvailable);

2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182
    return expand_bytes;
  } else {
    return 0;
  }
}

void G1CollectorPolicy::note_start_of_mark_thread() {
  _mark_thread_startup_sec = os::elapsedTime();
}

class CountCSClosure: public HeapRegionClosure {
  G1CollectorPolicy* _g1_policy;
public:
  CountCSClosure(G1CollectorPolicy* g1_policy) :
    _g1_policy(g1_policy) {}
  bool doHeapRegion(HeapRegion* r) {
    _g1_policy->_bytes_in_collection_set_before_gc += r->used();
    return false;
  }
};

void G1CollectorPolicy::count_CS_bytes_used() {
  CountCSClosure cs_closure(this);
  _g1->collection_set_iterate(&cs_closure);
}

void G1CollectorPolicy::print_summary (int level,
                                       const char* str,
                                       NumberSeq* seq) const {
  double sum = seq->sum();
2183
  LineBuffer(level + 1).append_and_print_cr("%-24s = %8.2lf s (avg = %8.2lf ms)",
2184 2185 2186 2187 2188 2189 2190
                str, sum / 1000.0, seq->avg());
}

void G1CollectorPolicy::print_summary_sd (int level,
                                          const char* str,
                                          NumberSeq* seq) const {
  print_summary(level, str, seq);
2191
  LineBuffer(level + 6).append_and_print_cr("(num = %5d, std dev = %8.2lf ms, max = %8.2lf ms)",
2192 2193 2194 2195 2196 2197 2198
                seq->num(), seq->sd(), seq->maximum());
}

void G1CollectorPolicy::check_other_times(int level,
                                        NumberSeq* other_times_ms,
                                        NumberSeq* calc_other_times_ms) const {
  bool should_print = false;
2199
  LineBuffer buf(level + 2);
2200 2201 2202 2203 2204 2205 2206 2207

  double max_sum = MAX2(fabs(other_times_ms->sum()),
                        fabs(calc_other_times_ms->sum()));
  double min_sum = MIN2(fabs(other_times_ms->sum()),
                        fabs(calc_other_times_ms->sum()));
  double sum_ratio = max_sum / min_sum;
  if (sum_ratio > 1.1) {
    should_print = true;
2208
    buf.append_and_print_cr("## CALCULATED OTHER SUM DOESN'T MATCH RECORDED ###");
2209 2210 2211 2212 2213 2214 2215 2216 2217
  }

  double max_avg = MAX2(fabs(other_times_ms->avg()),
                        fabs(calc_other_times_ms->avg()));
  double min_avg = MIN2(fabs(other_times_ms->avg()),
                        fabs(calc_other_times_ms->avg()));
  double avg_ratio = max_avg / min_avg;
  if (avg_ratio > 1.1) {
    should_print = true;
2218
    buf.append_and_print_cr("## CALCULATED OTHER AVG DOESN'T MATCH RECORDED ###");
2219 2220 2221
  }

  if (other_times_ms->sum() < -0.01) {
2222
    buf.append_and_print_cr("## RECORDED OTHER SUM IS NEGATIVE ###");
2223 2224 2225
  }

  if (other_times_ms->avg() < -0.01) {
2226
    buf.append_and_print_cr("## RECORDED OTHER AVG IS NEGATIVE ###");
2227 2228 2229 2230
  }

  if (calc_other_times_ms->sum() < -0.01) {
    should_print = true;
2231
    buf.append_and_print_cr("## CALCULATED OTHER SUM IS NEGATIVE ###");
2232 2233 2234 2235
  }

  if (calc_other_times_ms->avg() < -0.01) {
    should_print = true;
2236
    buf.append_and_print_cr("## CALCULATED OTHER AVG IS NEGATIVE ###");
2237 2238 2239 2240 2241 2242 2243
  }

  if (should_print)
    print_summary(level, "Other(Calc)", calc_other_times_ms);
}

void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
2244
  bool parallel = G1CollectedHeap::use_parallel_gc_threads();
2245 2246
  MainBodySummary*    body_summary = summary->main_body_summary();
  if (summary->get_total_seq()->num() > 0) {
2247
    print_summary_sd(0, "Evacuation Pauses", summary->get_total_seq());
2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270
    if (body_summary != NULL) {
      print_summary(1, "SATB Drain", body_summary->get_satb_drain_seq());
      if (parallel) {
        print_summary(1, "Parallel Time", body_summary->get_parallel_seq());
        print_summary(2, "Update RS", body_summary->get_update_rs_seq());
        print_summary(2, "Ext Root Scanning",
                      body_summary->get_ext_root_scan_seq());
        print_summary(2, "Mark Stack Scanning",
                      body_summary->get_mark_stack_scan_seq());
        print_summary(2, "Scan RS", body_summary->get_scan_rs_seq());
        print_summary(2, "Object Copy", body_summary->get_obj_copy_seq());
        print_summary(2, "Termination", body_summary->get_termination_seq());
        print_summary(2, "Other", body_summary->get_parallel_other_seq());
        {
          NumberSeq* other_parts[] = {
            body_summary->get_update_rs_seq(),
            body_summary->get_ext_root_scan_seq(),
            body_summary->get_mark_stack_scan_seq(),
            body_summary->get_scan_rs_seq(),
            body_summary->get_obj_copy_seq(),
            body_summary->get_termination_seq()
          };
          NumberSeq calc_other_times_ms(body_summary->get_parallel_seq(),
2271
                                        6, other_parts);
2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288
          check_other_times(2, body_summary->get_parallel_other_seq(),
                            &calc_other_times_ms);
        }
        print_summary(1, "Mark Closure", body_summary->get_mark_closure_seq());
        print_summary(1, "Clear CT", body_summary->get_clear_ct_seq());
      } else {
        print_summary(1, "Update RS", body_summary->get_update_rs_seq());
        print_summary(1, "Ext Root Scanning",
                      body_summary->get_ext_root_scan_seq());
        print_summary(1, "Mark Stack Scanning",
                      body_summary->get_mark_stack_scan_seq());
        print_summary(1, "Scan RS", body_summary->get_scan_rs_seq());
        print_summary(1, "Object Copy", body_summary->get_obj_copy_seq());
      }
    }
    print_summary(1, "Other", summary->get_other_seq());
    {
2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313
      if (body_summary != NULL) {
        NumberSeq calc_other_times_ms;
        if (parallel) {
          // parallel
          NumberSeq* other_parts[] = {
            body_summary->get_satb_drain_seq(),
            body_summary->get_parallel_seq(),
            body_summary->get_clear_ct_seq()
          };
          calc_other_times_ms = NumberSeq(summary->get_total_seq(),
                                                3, other_parts);
        } else {
          // serial
          NumberSeq* other_parts[] = {
            body_summary->get_satb_drain_seq(),
            body_summary->get_update_rs_seq(),
            body_summary->get_ext_root_scan_seq(),
            body_summary->get_mark_stack_scan_seq(),
            body_summary->get_scan_rs_seq(),
            body_summary->get_obj_copy_seq()
          };
          calc_other_times_ms = NumberSeq(summary->get_total_seq(),
                                                6, other_parts);
        }
        check_other_times(1,  summary->get_other_seq(), &calc_other_times_ms);
2314 2315 2316
      }
    }
  } else {
2317
    LineBuffer(1).append_and_print_cr("none");
2318
  }
2319
  LineBuffer(0).append_and_print_cr("");
2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331
}

void G1CollectorPolicy::print_tracing_info() const {
  if (TraceGen0Time) {
    gclog_or_tty->print_cr("ALL PAUSES");
    print_summary_sd(0, "Total", _all_pause_times_ms);
    gclog_or_tty->print_cr("");
    gclog_or_tty->print_cr("");
    gclog_or_tty->print_cr("   Full Young GC Pauses:    %8d", _full_young_pause_num);
    gclog_or_tty->print_cr("   Partial Young GC Pauses: %8d", _partial_young_pause_num);
    gclog_or_tty->print_cr("");

2332 2333
    gclog_or_tty->print_cr("EVACUATION PAUSES");
    print_summary(_summary);
2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374

    gclog_or_tty->print_cr("MISC");
    print_summary_sd(0, "Stop World", _all_stop_world_times_ms);
    print_summary_sd(0, "Yields", _all_yield_times_ms);
    for (int i = 0; i < _aux_num; ++i) {
      if (_all_aux_times_ms[i].num() > 0) {
        char buffer[96];
        sprintf(buffer, "Aux%d", i);
        print_summary_sd(0, buffer, &_all_aux_times_ms[i]);
      }
    }

    size_t all_region_num = _region_num_young + _region_num_tenured;
    gclog_or_tty->print_cr("   New Regions %8d, Young %8d (%6.2lf%%), "
               "Tenured %8d (%6.2lf%%)",
               all_region_num,
               _region_num_young,
               (double) _region_num_young / (double) all_region_num * 100.0,
               _region_num_tenured,
               (double) _region_num_tenured / (double) all_region_num * 100.0);
  }
  if (TraceGen1Time) {
    if (_all_full_gc_times_ms->num() > 0) {
      gclog_or_tty->print("\n%4d full_gcs: total time = %8.2f s",
                 _all_full_gc_times_ms->num(),
                 _all_full_gc_times_ms->sum() / 1000.0);
      gclog_or_tty->print_cr(" (avg = %8.2fms).", _all_full_gc_times_ms->avg());
      gclog_or_tty->print_cr("                     [std. dev = %8.2f ms, max = %8.2f ms]",
                    _all_full_gc_times_ms->sd(),
                    _all_full_gc_times_ms->maximum());
    }
  }
}

void G1CollectorPolicy::print_yg_surv_rate_info() const {
#ifndef PRODUCT
  _short_lived_surv_rate_group->print_surv_rate_summary();
  // add this call for any other surv rate groups
#endif // PRODUCT
}

2375
void G1CollectorPolicy::update_region_num(bool young) {
2376
  if (young) {
2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394
    ++_region_num_young;
  } else {
    ++_region_num_tenured;
  }
}

#ifndef PRODUCT
// for debugging, bit of a hack...
static char*
region_num_to_mbs(int length) {
  static char buffer[64];
  double bytes = (double) (length * HeapRegion::GrainBytes);
  double mbs = bytes / (double) (1024 * 1024);
  sprintf(buffer, "%7.2lfMB", mbs);
  return buffer;
}
#endif // PRODUCT

2395
size_t G1CollectorPolicy::max_regions(int purpose) {
2396 2397
  switch (purpose) {
    case GCAllocForSurvived:
2398
      return _max_survivor_regions;
2399
    case GCAllocForTenured:
2400
      return REGIONS_UNLIMITED;
2401
    default:
2402 2403
      ShouldNotReachHere();
      return REGIONS_UNLIMITED;
2404 2405 2406
  };
}

2407
void G1CollectorPolicy::update_max_gc_locker_expansion() {
2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421
  size_t expansion_region_num = 0;
  if (GCLockerEdenExpansionPercent > 0) {
    double perc = (double) GCLockerEdenExpansionPercent / 100.0;
    double expansion_region_num_d = perc * (double) _young_list_target_length;
    // We use ceiling so that if expansion_region_num_d is > 0.0 (but
    // less than 1.0) we'll get 1.
    expansion_region_num = (size_t) ceil(expansion_region_num_d);
  } else {
    assert(expansion_region_num == 0, "sanity");
  }
  _young_list_max_length = _young_list_target_length + expansion_region_num;
  assert(_young_list_target_length <= _young_list_max_length, "post-condition");
}

2422
// Calculates survivor space parameters.
2423 2424 2425 2426 2427 2428 2429
void G1CollectorPolicy::update_survivors_policy() {
  double max_survivor_regions_d =
                 (double) _young_list_target_length / (double) SurvivorRatio;
  // We use ceiling so that if max_survivor_regions_d is > 0.0 (but
  // smaller than 1.0) we'll get 1.
  _max_survivor_regions = (size_t) ceil(max_survivor_regions_d);

2430
  _tenuring_threshold = _survivors_age_table.compute_tenuring_threshold(
2431 2432 2433
        HeapRegion::GrainWords * _max_survivor_regions);
}

2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455
#ifndef PRODUCT
class HRSortIndexIsOKClosure: public HeapRegionClosure {
  CollectionSetChooser* _chooser;
public:
  HRSortIndexIsOKClosure(CollectionSetChooser* chooser) :
    _chooser(chooser) {}

  bool doHeapRegion(HeapRegion* r) {
    if (!r->continuesHumongous()) {
      assert(_chooser->regionProperlyOrdered(r), "Ought to be.");
    }
    return false;
  }
};

bool G1CollectorPolicy_BestRegionsFirst::assertMarkedBytesDataOK() {
  HRSortIndexIsOKClosure cl(_collectionSetChooser);
  _g1->heap_region_iterate(&cl);
  return true;
}
#endif

2456 2457
bool G1CollectorPolicy::force_initial_mark_if_outside_cycle(
                                                     GCCause::Cause gc_cause) {
2458 2459
  bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle();
  if (!during_cycle) {
2460 2461 2462 2463 2464
    ergo_verbose1(ErgoConcCycles,
                  "request concurrent cycle initiation",
                  ergo_format_reason("requested by GC cause")
                  ergo_format_str("GC cause"),
                  GCCause::to_string(gc_cause));
2465 2466 2467
    set_initiate_conc_mark_if_possible();
    return true;
  } else {
2468 2469 2470 2471 2472
    ergo_verbose1(ErgoConcCycles,
                  "do not request concurrent cycle initiation",
                  ergo_format_reason("concurrent cycle already in progress")
                  ergo_format_str("GC cause"),
                  GCCause::to_string(gc_cause));
2473 2474 2475 2476
    return false;
  }
}

2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499
void
G1CollectorPolicy::decide_on_conc_mark_initiation() {
  // We are about to decide on whether this pause will be an
  // initial-mark pause.

  // First, during_initial_mark_pause() should not be already set. We
  // will set it here if we have to. However, it should be cleared by
  // the end of the pause (it's only set for the duration of an
  // initial-mark pause).
  assert(!during_initial_mark_pause(), "pre-condition");

  if (initiate_conc_mark_if_possible()) {
    // We had noticed on a previous pause that the heap occupancy has
    // gone over the initiating threshold and we should start a
    // concurrent marking cycle. So we might initiate one.

    bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle();
    if (!during_cycle) {
      // The concurrent marking thread is not "during a cycle", i.e.,
      // it has completed the last one. So we can go ahead and
      // initiate a new cycle.

      set_during_initial_mark_pause();
2500 2501 2502 2503 2504 2505 2506
      // We do not allow non-full young GCs during marking.
      if (!full_young_gcs()) {
        set_full_young_gcs(true);
        ergo_verbose0(ErgoPartiallyYoungGCs,
                      "end partially-young GCs",
                      ergo_format_reason("concurrent cycle is about to start"));
      }
2507 2508 2509 2510

      // And we can now clear initiate_conc_mark_if_possible() as
      // we've already acted on it.
      clear_initiate_conc_mark_if_possible();
2511 2512 2513 2514

      ergo_verbose0(ErgoConcCycles,
                  "initiate concurrent cycle",
                  ergo_format_reason("concurrent cycle initiation requested"));
2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527
    } else {
      // The concurrent marking thread is still finishing up the
      // previous cycle. If we start one right now the two cycles
      // overlap. In particular, the concurrent marking thread might
      // be in the process of clearing the next marking bitmap (which
      // we will use for the next cycle if we start one). Starting a
      // cycle now will be bad given that parts of the marking
      // information might get cleared by the marking thread. And we
      // cannot wait for the marking thread to finish the cycle as it
      // periodically yields while clearing the next marking bitmap
      // and, if it's in a yield point, it's waiting for us to
      // finish. So, at this point we will not start a cycle and we'll
      // let the concurrent marking thread complete the last one.
2528 2529 2530
      ergo_verbose0(ErgoConcCycles,
                    "do not initiate concurrent cycle",
                    ergo_format_reason("concurrent cycle already in progress"));
2531 2532 2533 2534
    }
  }
}

2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635
void
G1CollectorPolicy_BestRegionsFirst::
record_collection_pause_start(double start_time_sec, size_t start_used) {
  G1CollectorPolicy::record_collection_pause_start(start_time_sec, start_used);
}

class KnownGarbageClosure: public HeapRegionClosure {
  CollectionSetChooser* _hrSorted;

public:
  KnownGarbageClosure(CollectionSetChooser* hrSorted) :
    _hrSorted(hrSorted)
  {}

  bool doHeapRegion(HeapRegion* r) {
    // We only include humongous regions in collection
    // sets when concurrent mark shows that their contained object is
    // unreachable.

    // Do we have any marking information for this region?
    if (r->is_marked()) {
      // We don't include humongous regions in collection
      // sets because we collect them immediately at the end of a marking
      // cycle.  We also don't include young regions because we *must*
      // include them in the next collection pause.
      if (!r->isHumongous() && !r->is_young()) {
        _hrSorted->addMarkedHeapRegion(r);
      }
    }
    return false;
  }
};

class ParKnownGarbageHRClosure: public HeapRegionClosure {
  CollectionSetChooser* _hrSorted;
  jint _marked_regions_added;
  jint _chunk_size;
  jint _cur_chunk_idx;
  jint _cur_chunk_end; // Cur chunk [_cur_chunk_idx, _cur_chunk_end)
  int _worker;
  int _invokes;

  void get_new_chunk() {
    _cur_chunk_idx = _hrSorted->getParMarkedHeapRegionChunk(_chunk_size);
    _cur_chunk_end = _cur_chunk_idx + _chunk_size;
  }
  void add_region(HeapRegion* r) {
    if (_cur_chunk_idx == _cur_chunk_end) {
      get_new_chunk();
    }
    assert(_cur_chunk_idx < _cur_chunk_end, "postcondition");
    _hrSorted->setMarkedHeapRegion(_cur_chunk_idx, r);
    _marked_regions_added++;
    _cur_chunk_idx++;
  }

public:
  ParKnownGarbageHRClosure(CollectionSetChooser* hrSorted,
                           jint chunk_size,
                           int worker) :
    _hrSorted(hrSorted), _chunk_size(chunk_size), _worker(worker),
    _marked_regions_added(0), _cur_chunk_idx(0), _cur_chunk_end(0),
    _invokes(0)
  {}

  bool doHeapRegion(HeapRegion* r) {
    // We only include humongous regions in collection
    // sets when concurrent mark shows that their contained object is
    // unreachable.
    _invokes++;

    // Do we have any marking information for this region?
    if (r->is_marked()) {
      // We don't include humongous regions in collection
      // sets because we collect them immediately at the end of a marking
      // cycle.
      // We also do not include young regions in collection sets
      if (!r->isHumongous() && !r->is_young()) {
        add_region(r);
      }
    }
    return false;
  }
  jint marked_regions_added() { return _marked_regions_added; }
  int invokes() { return _invokes; }
};

class ParKnownGarbageTask: public AbstractGangTask {
  CollectionSetChooser* _hrSorted;
  jint _chunk_size;
  G1CollectedHeap* _g1;
public:
  ParKnownGarbageTask(CollectionSetChooser* hrSorted, jint chunk_size) :
    AbstractGangTask("ParKnownGarbageTask"),
    _hrSorted(hrSorted), _chunk_size(chunk_size),
    _g1(G1CollectedHeap::heap())
  {}

  void work(int i) {
    ParKnownGarbageHRClosure parKnownGarbageCl(_hrSorted, _chunk_size, i);
    // Back to zero for the claim value.
2636 2637
    _g1->heap_region_par_iterate_chunked(&parKnownGarbageCl, i,
                                         HeapRegion::InitialClaimValue);
2638 2639 2640
    jint regions_added = parKnownGarbageCl.marked_regions_added();
    _hrSorted->incNumMarkedHeapRegions(regions_added);
    if (G1PrintParCleanupStats) {
2641
      gclog_or_tty->print_cr("     Thread %d called %d times, added %d regions to list.",
2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661
                 i, parKnownGarbageCl.invokes(), regions_added);
    }
  }
};

void
G1CollectorPolicy_BestRegionsFirst::
record_concurrent_mark_cleanup_end(size_t freed_bytes,
                                   size_t max_live_bytes) {
  double start;
  if (G1PrintParCleanupStats) start = os::elapsedTime();
  record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes);

  _collectionSetChooser->clearMarkedHeapRegions();
  double clear_marked_end;
  if (G1PrintParCleanupStats) {
    clear_marked_end = os::elapsedTime();
    gclog_or_tty->print_cr("  clear marked regions + work1: %8.3f ms.",
                  (clear_marked_end - start)*1000.0);
  }
2662
  if (G1CollectedHeap::use_parallel_gc_threads()) {
2663
    const size_t OverpartitionFactor = 4;
2664 2665
    const size_t MinWorkUnit = 8;
    const size_t WorkUnit =
2666
      MAX2(_g1->n_regions() / (ParallelGCThreads * OverpartitionFactor),
2667
           MinWorkUnit);
2668
    _collectionSetChooser->prepareForAddMarkedHeapRegionsPar(_g1->n_regions(),
2669
                                                             WorkUnit);
2670
    ParKnownGarbageTask parKnownGarbageTask(_collectionSetChooser,
2671
                                            (int) WorkUnit);
2672
    _g1->workers()->run_task(&parKnownGarbageTask);
2673 2674 2675

    assert(_g1->check_heap_region_claim_values(HeapRegion::InitialClaimValue),
           "sanity check");
2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702
  } else {
    KnownGarbageClosure knownGarbagecl(_collectionSetChooser);
    _g1->heap_region_iterate(&knownGarbagecl);
  }
  double known_garbage_end;
  if (G1PrintParCleanupStats) {
    known_garbage_end = os::elapsedTime();
    gclog_or_tty->print_cr("  compute known garbage: %8.3f ms.",
                  (known_garbage_end - clear_marked_end)*1000.0);
  }
  _collectionSetChooser->sortMarkedHeapRegions();
  double sort_end;
  if (G1PrintParCleanupStats) {
    sort_end = os::elapsedTime();
    gclog_or_tty->print_cr("  sorting: %8.3f ms.",
                  (sort_end - known_garbage_end)*1000.0);
  }

  record_concurrent_mark_cleanup_end_work2();
  double work2_end;
  if (G1PrintParCleanupStats) {
    work2_end = os::elapsedTime();
    gclog_or_tty->print_cr("  work2: %8.3f ms.",
                  (work2_end - sort_end)*1000.0);
  }
}

2703
// Add the heap region at the head of the non-incremental collection set
2704 2705
void G1CollectorPolicy::
add_to_collection_set(HeapRegion* hr) {
2706 2707 2708
  assert(_inc_cset_build_state == Active, "Precondition");
  assert(!hr->is_young(), "non-incremental add of young region");

2709 2710 2711
  if (_g1->mark_in_progress())
    _g1->concurrent_mark()->registerCSetRegion(hr);

2712
  assert(!hr->in_collection_set(), "should not already be in the CSet");
2713 2714 2715 2716 2717
  hr->set_in_collection_set(true);
  hr->set_next_in_collection_set(_collection_set);
  _collection_set = hr;
  _collection_set_size++;
  _collection_set_bytes_used_before += hr->used();
2718
  _g1->register_region_with_in_cset_fast_test(hr);
2719 2720
}

2721 2722 2723 2724 2725 2726 2727 2728 2729
// Initialize the per-collection-set information
void G1CollectorPolicy::start_incremental_cset_building() {
  assert(_inc_cset_build_state == Inactive, "Precondition");

  _inc_cset_head = NULL;
  _inc_cset_tail = NULL;
  _inc_cset_size = 0;
  _inc_cset_bytes_used_before = 0;

2730
  _inc_cset_young_index = 0;
2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914

  _inc_cset_max_finger = 0;
  _inc_cset_recorded_young_bytes = 0;
  _inc_cset_recorded_rs_lengths = 0;
  _inc_cset_predicted_elapsed_time_ms = 0;
  _inc_cset_predicted_bytes_to_copy = 0;
  _inc_cset_build_state = Active;
}

void G1CollectorPolicy::add_to_incremental_cset_info(HeapRegion* hr, size_t rs_length) {
  // This routine is used when:
  // * adding survivor regions to the incremental cset at the end of an
  //   evacuation pause,
  // * adding the current allocation region to the incremental cset
  //   when it is retired, and
  // * updating existing policy information for a region in the
  //   incremental cset via young list RSet sampling.
  // Therefore this routine may be called at a safepoint by the
  // VM thread, or in-between safepoints by mutator threads (when
  // retiring the current allocation region) or a concurrent
  // refine thread (RSet sampling).

  double region_elapsed_time_ms = predict_region_elapsed_time_ms(hr, true);
  size_t used_bytes = hr->used();

  _inc_cset_recorded_rs_lengths += rs_length;
  _inc_cset_predicted_elapsed_time_ms += region_elapsed_time_ms;

  _inc_cset_bytes_used_before += used_bytes;

  // Cache the values we have added to the aggregated informtion
  // in the heap region in case we have to remove this region from
  // the incremental collection set, or it is updated by the
  // rset sampling code
  hr->set_recorded_rs_length(rs_length);
  hr->set_predicted_elapsed_time_ms(region_elapsed_time_ms);

#if PREDICTIONS_VERBOSE
  size_t bytes_to_copy = predict_bytes_to_copy(hr);
  _inc_cset_predicted_bytes_to_copy += bytes_to_copy;

  // Record the number of bytes used in this region
  _inc_cset_recorded_young_bytes += used_bytes;

  // Cache the values we have added to the aggregated informtion
  // in the heap region in case we have to remove this region from
  // the incremental collection set, or it is updated by the
  // rset sampling code
  hr->set_predicted_bytes_to_copy(bytes_to_copy);
#endif // PREDICTIONS_VERBOSE
}

void G1CollectorPolicy::remove_from_incremental_cset_info(HeapRegion* hr) {
  // This routine is currently only called as part of the updating of
  // existing policy information for regions in the incremental cset that
  // is performed by the concurrent refine thread(s) as part of young list
  // RSet sampling. Therefore we should not be at a safepoint.

  assert(!SafepointSynchronize::is_at_safepoint(), "should not be at safepoint");
  assert(hr->is_young(), "it should be");

  size_t used_bytes = hr->used();
  size_t old_rs_length = hr->recorded_rs_length();
  double old_elapsed_time_ms = hr->predicted_elapsed_time_ms();

  // Subtract the old recorded/predicted policy information for
  // the given heap region from the collection set info.
  _inc_cset_recorded_rs_lengths -= old_rs_length;
  _inc_cset_predicted_elapsed_time_ms -= old_elapsed_time_ms;

  _inc_cset_bytes_used_before -= used_bytes;

  // Clear the values cached in the heap region
  hr->set_recorded_rs_length(0);
  hr->set_predicted_elapsed_time_ms(0);

#if PREDICTIONS_VERBOSE
  size_t old_predicted_bytes_to_copy = hr->predicted_bytes_to_copy();
  _inc_cset_predicted_bytes_to_copy -= old_predicted_bytes_to_copy;

  // Subtract the number of bytes used in this region
  _inc_cset_recorded_young_bytes -= used_bytes;

  // Clear the values cached in the heap region
  hr->set_predicted_bytes_to_copy(0);
#endif // PREDICTIONS_VERBOSE
}

void G1CollectorPolicy::update_incremental_cset_info(HeapRegion* hr, size_t new_rs_length) {
  // Update the collection set information that is dependent on the new RS length
  assert(hr->is_young(), "Precondition");

  remove_from_incremental_cset_info(hr);
  add_to_incremental_cset_info(hr, new_rs_length);
}

void G1CollectorPolicy::add_region_to_incremental_cset_common(HeapRegion* hr) {
  assert( hr->is_young(), "invariant");
  assert( hr->young_index_in_cset() == -1, "invariant" );
  assert(_inc_cset_build_state == Active, "Precondition");

  // We need to clear and set the cached recorded/cached collection set
  // information in the heap region here (before the region gets added
  // to the collection set). An individual heap region's cached values
  // are calculated, aggregated with the policy collection set info,
  // and cached in the heap region here (initially) and (subsequently)
  // by the Young List sampling code.

  size_t rs_length = hr->rem_set()->occupied();
  add_to_incremental_cset_info(hr, rs_length);

  HeapWord* hr_end = hr->end();
  _inc_cset_max_finger = MAX2(_inc_cset_max_finger, hr_end);

  assert(!hr->in_collection_set(), "invariant");
  hr->set_in_collection_set(true);
  assert( hr->next_in_collection_set() == NULL, "invariant");

  _inc_cset_size++;
  _g1->register_region_with_in_cset_fast_test(hr);

  hr->set_young_index_in_cset((int) _inc_cset_young_index);
  ++_inc_cset_young_index;
}

// Add the region at the RHS of the incremental cset
void G1CollectorPolicy::add_region_to_incremental_cset_rhs(HeapRegion* hr) {
  // We should only ever be appending survivors at the end of a pause
  assert( hr->is_survivor(), "Logic");

  // Do the 'common' stuff
  add_region_to_incremental_cset_common(hr);

  // Now add the region at the right hand side
  if (_inc_cset_tail == NULL) {
    assert(_inc_cset_head == NULL, "invariant");
    _inc_cset_head = hr;
  } else {
    _inc_cset_tail->set_next_in_collection_set(hr);
  }
  _inc_cset_tail = hr;
}

// Add the region to the LHS of the incremental cset
void G1CollectorPolicy::add_region_to_incremental_cset_lhs(HeapRegion* hr) {
  // Survivors should be added to the RHS at the end of a pause
  assert(!hr->is_survivor(), "Logic");

  // Do the 'common' stuff
  add_region_to_incremental_cset_common(hr);

  // Add the region at the left hand side
  hr->set_next_in_collection_set(_inc_cset_head);
  if (_inc_cset_head == NULL) {
    assert(_inc_cset_tail == NULL, "Invariant");
    _inc_cset_tail = hr;
  }
  _inc_cset_head = hr;
}

#ifndef PRODUCT
void G1CollectorPolicy::print_collection_set(HeapRegion* list_head, outputStream* st) {
  assert(list_head == inc_cset_head() || list_head == collection_set(), "must be");

  st->print_cr("\nCollection_set:");
  HeapRegion* csr = list_head;
  while (csr != NULL) {
    HeapRegion* next = csr->next_in_collection_set();
    assert(csr->in_collection_set(), "bad CS");
    st->print_cr("  [%08x-%08x], t: %08x, P: %08x, N: %08x, C: %08x, "
                 "age: %4d, y: %d, surv: %d",
                        csr->bottom(), csr->end(),
                        csr->top(),
                        csr->prev_top_at_mark_start(),
                        csr->next_top_at_mark_start(),
                        csr->top_at_conc_mark_count(),
                        csr->age_in_surv_rate_group_cond(),
                        csr->is_young(),
                        csr->is_survivor());
    csr = next;
  }
}
#endif // !PRODUCT

2915
void
2916 2917
G1CollectorPolicy_BestRegionsFirst::choose_collection_set(
                                                  double target_pause_time_ms) {
2918 2919 2920
  // Set this here - in case we're not doing young collections.
  double non_young_start_time_sec = os::elapsedTime();

2921 2922
  YoungList* young_list = _g1->young_list();

2923 2924
  start_recording_regions();

2925 2926 2927 2928
  guarantee(target_pause_time_ms > 0.0,
            err_msg("target_pause_time_ms = %1.6lf should be positive",
                    target_pause_time_ms));
  guarantee(_collection_set == NULL, "Precondition");
2929 2930 2931 2932

  double base_time_ms = predict_base_elapsed_time_ms(_pending_cards);
  double predicted_pause_time_ms = base_time_ms;

2933
  double time_remaining_ms = target_pause_time_ms - base_time_ms;
2934

2935 2936 2937 2938 2939 2940 2941
  ergo_verbose3(ErgoCSetConstruction | ErgoHigh,
                "start choosing CSet",
                ergo_format_ms("predicted base time")
                ergo_format_ms("remaining time")
                ergo_format_ms("target pause time"),
                base_time_ms, time_remaining_ms, target_pause_time_ms);

2942
  // the 10% and 50% values are arbitrary...
2943 2944 2945
  double threshold = 0.10 * target_pause_time_ms;
  if (time_remaining_ms < threshold) {
    double prev_time_remaining_ms = time_remaining_ms;
2946
    time_remaining_ms = 0.50 * target_pause_time_ms;
2947
    _within_target = false;
2948 2949 2950 2951 2952 2953 2954
    ergo_verbose3(ErgoCSetConstruction,
                  "adjust remaining time",
                  ergo_format_reason("remaining time lower than threshold")
                  ergo_format_ms("remaining time")
                  ergo_format_ms("threshold")
                  ergo_format_ms("adjusted remaining time"),
                  prev_time_remaining_ms, threshold, time_remaining_ms);
2955 2956 2957 2958
  } else {
    _within_target = true;
  }

2959
  size_t expansion_bytes = _g1->expansion_regions() * HeapRegion::GrainBytes;
2960 2961

  HeapRegion* hr;
2962
  double young_start_time_sec = os::elapsedTime();
2963

2964 2965
  _collection_set_bytes_used_before = 0;
  _collection_set_size = 0;
2966 2967
  _young_cset_length  = 0;
  _last_young_gc_full = full_young_gcs() ? true : false;
2968

2969
  if (_last_young_gc_full) {
2970
    ++_full_young_pause_num;
2971
  } else {
2972
    ++_partial_young_pause_num;
2973
  }
2974

2975 2976 2977
  // The young list is laid with the survivor regions from the previous
  // pause are appended to the RHS of the young list, i.e.
  //   [Newly Young Regions ++ Survivors from last pause].
2978

2979 2980 2981 2982
  size_t survivor_region_num = young_list->survivor_length();
  size_t eden_region_num = young_list->length() - survivor_region_num;
  size_t old_region_num = 0;
  hr = young_list->first_survivor_region();
2983 2984 2985 2986 2987
  while (hr != NULL) {
    assert(hr->is_survivor(), "badly formed young list");
    hr->set_young();
    hr = hr->get_next_young_region();
  }
2988

2989 2990
  // Clear the fields that point to the survivor list - they are all young now.
  young_list->clear_survivors();
2991

2992 2993
  if (_g1->mark_in_progress())
    _g1->concurrent_mark()->register_collection_set_finger(_inc_cset_max_finger);
2994

2995 2996 2997 2998 2999 3000
  _young_cset_length = _inc_cset_young_index;
  _collection_set = _inc_cset_head;
  _collection_set_size = _inc_cset_size;
  _collection_set_bytes_used_before = _inc_cset_bytes_used_before;
  time_remaining_ms -= _inc_cset_predicted_elapsed_time_ms;
  predicted_pause_time_ms += _inc_cset_predicted_elapsed_time_ms;
3001

3002 3003 3004 3005 3006 3007 3008 3009
  ergo_verbose3(ErgoCSetConstruction | ErgoHigh,
                "add young regions to CSet",
                ergo_format_region("eden")
                ergo_format_region("survivors")
                ergo_format_ms("predicted young region time"),
                eden_region_num, survivor_region_num,
                _inc_cset_predicted_elapsed_time_ms);

3010 3011 3012 3013 3014
  // The number of recorded young regions is the incremental
  // collection set's current size
  set_recorded_young_regions(_inc_cset_size);
  set_recorded_rs_lengths(_inc_cset_recorded_rs_lengths);
  set_recorded_young_bytes(_inc_cset_recorded_young_bytes);
3015
#if PREDICTIONS_VERBOSE
3016
  set_predicted_bytes_to_copy(_inc_cset_predicted_bytes_to_copy);
3017 3018
#endif // PREDICTIONS_VERBOSE

3019
  assert(_inc_cset_size == young_list->length(), "Invariant");
3020

3021 3022 3023
  double young_end_time_sec = os::elapsedTime();
  _recorded_young_cset_choice_time_ms =
    (young_end_time_sec - young_start_time_sec) * 1000.0;
3024

3025 3026
  // We are doing young collections so reset this.
  non_young_start_time_sec = young_end_time_sec;
3027

3028
  if (!full_young_gcs()) {
3029 3030 3031
    bool should_continue = true;
    NumberSeq seq;
    double avg_prediction = 100000000000000000.0; // something very large
3032

3033 3034
    size_t prev_collection_set_size = _collection_set_size;
    double prev_predicted_pause_time_ms = predicted_pause_time_ms;
3035 3036 3037
    do {
      hr = _collectionSetChooser->getNextMarkedRegion(time_remaining_ms,
                                                      avg_prediction);
3038
      if (hr != NULL) {
3039 3040 3041 3042
        double predicted_time_ms = predict_region_elapsed_time_ms(hr, false);
        time_remaining_ms -= predicted_time_ms;
        predicted_pause_time_ms += predicted_time_ms;
        add_to_collection_set(hr);
3043
        record_non_young_cset_region(hr);
3044 3045 3046
        seq.add(predicted_time_ms);
        avg_prediction = seq.avg() + seq.sd();
      }
3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063

      should_continue = true;
      if (hr == NULL) {
        // No need for an ergo verbose message here,
        // getNextMarkRegion() does this when it returns NULL.
        should_continue = false;
      } else {
        if (adaptive_young_list_length()) {
          if (time_remaining_ms < 0.0) {
            ergo_verbose1(ErgoCSetConstruction,
                          "stop adding old regions to CSet",
                          ergo_format_reason("remaining time is lower than 0")
                          ergo_format_ms("remaining time"),
                          time_remaining_ms);
            should_continue = false;
          }
        } else {
3064
          if (_collection_set_size >= _young_list_fixed_length) {
3065 3066
            ergo_verbose2(ErgoCSetConstruction,
                          "stop adding old regions to CSet",
3067
                          ergo_format_reason("CSet length reached target")
3068 3069 3070 3071 3072 3073 3074
                          ergo_format_region("CSet")
                          ergo_format_region("young target"),
                          _collection_set_size, _young_list_fixed_length);
            should_continue = false;
          }
        }
      }
3075 3076 3077
    } while (should_continue);

    if (!adaptive_young_list_length() &&
3078 3079 3080 3081 3082 3083 3084
        _collection_set_size < _young_list_fixed_length) {
      ergo_verbose2(ErgoCSetConstruction,
                    "request partially-young GCs end",
                    ergo_format_reason("CSet length lower than target")
                    ergo_format_region("CSet")
                    ergo_format_region("young target"),
                    _collection_set_size, _young_list_fixed_length);
3085
      _should_revert_to_full_young_gcs  = true;
3086 3087 3088 3089 3090 3091 3092 3093 3094 3095
    }

    old_region_num = _collection_set_size - prev_collection_set_size;

    ergo_verbose2(ErgoCSetConstruction | ErgoHigh,
                  "add old regions to CSet",
                  ergo_format_region("old")
                  ergo_format_ms("predicted old region time"),
                  old_region_num,
                  predicted_pause_time_ms - prev_predicted_pause_time_ms);
3096 3097
  }

3098 3099
  stop_incremental_cset_building();

3100 3101 3102 3103
  count_CS_bytes_used();

  end_recording_regions();

3104 3105 3106 3107 3108 3109 3110 3111 3112 3113
  ergo_verbose5(ErgoCSetConstruction,
                "finish choosing CSet",
                ergo_format_region("eden")
                ergo_format_region("survivors")
                ergo_format_region("old")
                ergo_format_ms("predicted pause time")
                ergo_format_ms("target pause time"),
                eden_region_num, survivor_region_num, old_region_num,
                predicted_pause_time_ms, target_pause_time_ms);

3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124
  double non_young_end_time_sec = os::elapsedTime();
  _recorded_non_young_cset_choice_time_ms =
    (non_young_end_time_sec - non_young_start_time_sec) * 1000.0;
}

void G1CollectorPolicy_BestRegionsFirst::record_full_collection_end() {
  G1CollectorPolicy::record_full_collection_end();
  _collectionSetChooser->updateAfterFullCollection();
}

void G1CollectorPolicy_BestRegionsFirst::
3125 3126
record_collection_pause_end() {
  G1CollectorPolicy::record_collection_pause_end();
3127 3128
  assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end.");
}