concurrentMarkSweepGeneration.cpp 372.5 KB
Newer Older
D
duke 已提交
1
/*
2
 * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved.
D
duke 已提交
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.
D
duke 已提交
22 23 24
 *
 */

25
#include "precompiled.hpp"
26
#include "classfile/classLoaderData.hpp"
27 28 29 30 31 32 33 34 35 36 37 38 39
#include "classfile/symbolTable.hpp"
#include "classfile/systemDictionary.hpp"
#include "code/codeCache.hpp"
#include "gc_implementation/concurrentMarkSweep/cmsAdaptiveSizePolicy.hpp"
#include "gc_implementation/concurrentMarkSweep/cmsCollectorPolicy.hpp"
#include "gc_implementation/concurrentMarkSweep/cmsGCAdaptivePolicyCounters.hpp"
#include "gc_implementation/concurrentMarkSweep/cmsOopClosures.inline.hpp"
#include "gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp"
#include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.inline.hpp"
#include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepThread.hpp"
#include "gc_implementation/concurrentMarkSweep/vmCMSOperations.hpp"
#include "gc_implementation/parNew/parNewGeneration.hpp"
#include "gc_implementation/shared/collectorCounters.hpp"
S
sla 已提交
40 41 42
#include "gc_implementation/shared/gcTimer.hpp"
#include "gc_implementation/shared/gcTrace.hpp"
#include "gc_implementation/shared/gcTraceTime.hpp"
43 44
#include "gc_implementation/shared/isGCActiveMark.hpp"
#include "gc_interface/collectedHeap.inline.hpp"
S
sla 已提交
45
#include "memory/allocation.hpp"
46 47 48 49 50 51 52 53 54
#include "memory/cardTableRS.hpp"
#include "memory/collectorPolicy.hpp"
#include "memory/gcLocker.inline.hpp"
#include "memory/genCollectedHeap.hpp"
#include "memory/genMarkSweep.hpp"
#include "memory/genOopClosures.inline.hpp"
#include "memory/iterator.hpp"
#include "memory/referencePolicy.hpp"
#include "memory/resourceArea.hpp"
55
#include "memory/tenuredGeneration.hpp"
56 57 58 59 60 61 62 63
#include "oops/oop.inline.hpp"
#include "prims/jvmtiExport.hpp"
#include "runtime/globals_extension.hpp"
#include "runtime/handles.inline.hpp"
#include "runtime/java.hpp"
#include "runtime/vmThread.hpp"
#include "services/memoryService.hpp"
#include "services/runtimeService.hpp"
D
duke 已提交
64 65 66

// statics
CMSCollector* ConcurrentMarkSweepGeneration::_collector = NULL;
S
sla 已提交
67 68
bool CMSCollector::_full_gc_requested = false;
GCCause::Cause CMSCollector::_full_gc_cause = GCCause::_no_gc;
D
duke 已提交
69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183

//////////////////////////////////////////////////////////////////
// In support of CMS/VM thread synchronization
//////////////////////////////////////////////////////////////////
// We split use of the CGC_lock into 2 "levels".
// The low-level locking is of the usual CGC_lock monitor. We introduce
// a higher level "token" (hereafter "CMS token") built on top of the
// low level monitor (hereafter "CGC lock").
// The token-passing protocol gives priority to the VM thread. The
// CMS-lock doesn't provide any fairness guarantees, but clients
// should ensure that it is only held for very short, bounded
// durations.
//
// When either of the CMS thread or the VM thread is involved in
// collection operations during which it does not want the other
// thread to interfere, it obtains the CMS token.
//
// If either thread tries to get the token while the other has
// it, that thread waits. However, if the VM thread and CMS thread
// both want the token, then the VM thread gets priority while the
// CMS thread waits. This ensures, for instance, that the "concurrent"
// phases of the CMS thread's work do not block out the VM thread
// for long periods of time as the CMS thread continues to hog
// the token. (See bug 4616232).
//
// The baton-passing functions are, however, controlled by the
// flags _foregroundGCShouldWait and _foregroundGCIsActive,
// and here the low-level CMS lock, not the high level token,
// ensures mutual exclusion.
//
// Two important conditions that we have to satisfy:
// 1. if a thread does a low-level wait on the CMS lock, then it
//    relinquishes the CMS token if it were holding that token
//    when it acquired the low-level CMS lock.
// 2. any low-level notifications on the low-level lock
//    should only be sent when a thread has relinquished the token.
//
// In the absence of either property, we'd have potential deadlock.
//
// We protect each of the CMS (concurrent and sequential) phases
// with the CMS _token_, not the CMS _lock_.
//
// The only code protected by CMS lock is the token acquisition code
// itself, see ConcurrentMarkSweepThread::[de]synchronize(), and the
// baton-passing code.
//
// Unfortunately, i couldn't come up with a good abstraction to factor and
// hide the naked CGC_lock manipulation in the baton-passing code
// further below. That's something we should try to do. Also, the proof
// of correctness of this 2-level locking scheme is far from obvious,
// and potentially quite slippery. We have an uneasy supsicion, for instance,
// that there may be a theoretical possibility of delay/starvation in the
// low-level lock/wait/notify scheme used for the baton-passing because of
// potential intereference with the priority scheme embodied in the
// CMS-token-passing protocol. See related comments at a CGC_lock->wait()
// invocation further below and marked with "XXX 20011219YSR".
// Indeed, as we note elsewhere, this may become yet more slippery
// in the presence of multiple CMS and/or multiple VM threads. XXX

class CMSTokenSync: public StackObj {
 private:
  bool _is_cms_thread;
 public:
  CMSTokenSync(bool is_cms_thread):
    _is_cms_thread(is_cms_thread) {
    assert(is_cms_thread == Thread::current()->is_ConcurrentGC_thread(),
           "Incorrect argument to constructor");
    ConcurrentMarkSweepThread::synchronize(_is_cms_thread);
  }

  ~CMSTokenSync() {
    assert(_is_cms_thread ?
             ConcurrentMarkSweepThread::cms_thread_has_cms_token() :
             ConcurrentMarkSweepThread::vm_thread_has_cms_token(),
          "Incorrect state");
    ConcurrentMarkSweepThread::desynchronize(_is_cms_thread);
  }
};

// Convenience class that does a CMSTokenSync, and then acquires
// upto three locks.
class CMSTokenSyncWithLocks: public CMSTokenSync {
 private:
  // Note: locks are acquired in textual declaration order
  // and released in the opposite order
  MutexLockerEx _locker1, _locker2, _locker3;
 public:
  CMSTokenSyncWithLocks(bool is_cms_thread, Mutex* mutex1,
                        Mutex* mutex2 = NULL, Mutex* mutex3 = NULL):
    CMSTokenSync(is_cms_thread),
    _locker1(mutex1, Mutex::_no_safepoint_check_flag),
    _locker2(mutex2, Mutex::_no_safepoint_check_flag),
    _locker3(mutex3, Mutex::_no_safepoint_check_flag)
  { }
};


// Wrapper class to temporarily disable icms during a foreground cms collection.
class ICMSDisabler: public StackObj {
 public:
  // The ctor disables icms and wakes up the thread so it notices the change;
  // the dtor re-enables icms.  Note that the CMSCollector methods will check
  // CMSIncrementalMode.
  ICMSDisabler()  { CMSCollector::disable_icms(); CMSCollector::start_icms(); }
  ~ICMSDisabler() { CMSCollector::enable_icms(); }
};

//////////////////////////////////////////////////////////////////
//  Concurrent Mark-Sweep Generation /////////////////////////////
//////////////////////////////////////////////////////////////////

NOT_PRODUCT(CompactibleFreeListSpace* debug_cms_space;)

// This struct contains per-thread things necessary to support parallel
// young-gen collection.
Z
zgu 已提交
184
class CMSParGCThreadState: public CHeapObj<mtGC> {
D
duke 已提交
185 186 187 188 189 190 191 192 193 194 195 196 197
 public:
  CFLS_LAB lab;
  PromotionInfo promo;

  // Constructor.
  CMSParGCThreadState(CompactibleFreeListSpace* cfls) : lab(cfls) {
    promo.setSpace(cfls);
  }
};

ConcurrentMarkSweepGeneration::ConcurrentMarkSweepGeneration(
     ReservedSpace rs, size_t initial_byte_size, int level,
     CardTableRS* ct, bool use_adaptive_freelists,
198
     FreeBlockDictionary<FreeChunk>::DictionaryChoice dictionaryChoice) :
D
duke 已提交
199
  CardGeneration(rs, initial_byte_size, level, ct),
200
  _dilatation_factor(((double)MinChunkSize)/((double)(CollectedHeap::min_fill_size()))),
201 202
  _debug_collection_type(Concurrent_collection_type),
  _did_compact(false)
D
duke 已提交
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231
{
  HeapWord* bottom = (HeapWord*) _virtual_space.low();
  HeapWord* end    = (HeapWord*) _virtual_space.high();

  _direct_allocated_words = 0;
  NOT_PRODUCT(
    _numObjectsPromoted = 0;
    _numWordsPromoted = 0;
    _numObjectsAllocated = 0;
    _numWordsAllocated = 0;
  )

  _cmsSpace = new CompactibleFreeListSpace(_bts, MemRegion(bottom, end),
                                           use_adaptive_freelists,
                                           dictionaryChoice);
  NOT_PRODUCT(debug_cms_space = _cmsSpace;)
  if (_cmsSpace == NULL) {
    vm_exit_during_initialization(
      "CompactibleFreeListSpace allocation failure");
  }
  _cmsSpace->_gen = this;

  _gc_stats = new CMSGCStats();

  // Verify the assumption that FreeChunk::_prev and OopDesc::_klass
  // offsets match. The ability to tell free chunks from objects
  // depends on this property.
  debug_only(
    FreeChunk* junk = NULL;
232
    assert(UseCompressedKlassPointers ||
233
           junk->prev_addr() == (void*)(oop(junk)->klass_addr()),
D
duke 已提交
234 235 236
           "Offset of FreeChunk::_prev within FreeChunk must match"
           "  that of OopDesc::_klass within OopDesc");
  )
237
  if (CollectedHeap::use_parallel_gc_threads()) {
D
duke 已提交
238 239
    typedef CMSParGCThreadState* CMSParGCThreadStatePtr;
    _par_gc_thread_states =
Z
zgu 已提交
240
      NEW_C_HEAP_ARRAY(CMSParGCThreadStatePtr, ParallelGCThreads, mtGC);
D
duke 已提交
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
    if (_par_gc_thread_states == NULL) {
      vm_exit_during_initialization("Could not allocate par gc structs");
    }
    for (uint i = 0; i < ParallelGCThreads; i++) {
      _par_gc_thread_states[i] = new CMSParGCThreadState(cmsSpace());
      if (_par_gc_thread_states[i] == NULL) {
        vm_exit_during_initialization("Could not allocate par gc structs");
      }
    }
  } else {
    _par_gc_thread_states = NULL;
  }
  _incremental_collection_failed = false;
  // The "dilatation_factor" is the expansion that can occur on
  // account of the fact that the minimum object size in the CMS
  // generation may be larger than that in, say, a contiguous young
  //  generation.
  // Ideally, in the calculation below, we'd compute the dilatation
  // factor as: MinChunkSize/(promoting_gen's min object size)
  // Since we do not have such a general query interface for the
  // promoting generation, we'll instead just use the mimimum
  // object size (which today is a header's worth of space);
  // note that all arithmetic is in units of HeapWords.
264
  assert(MinChunkSize >= CollectedHeap::min_fill_size(), "just checking");
D
duke 已提交
265 266 267
  assert(_dilatation_factor >= 1.0, "from previous assert");
}

268 269 270

// The field "_initiating_occupancy" represents the occupancy percentage
// at which we trigger a new collection cycle.  Unless explicitly specified
271
// via CMSInitiatingOccupancyFraction (argument "io" below), it
272 273 274 275 276
// is calculated by:
//
//   Let "f" be MinHeapFreeRatio in
//
//    _intiating_occupancy = 100-f +
277 278
//                           f * (CMSTriggerRatio/100)
//   where CMSTriggerRatio is the argument "tr" below.
279 280
//
// That is, if we assume the heap is at its desired maximum occupancy at the
281
// end of a collection, we let CMSTriggerRatio of the (purported) free
282 283
// space be allocated before initiating a new collection cycle.
//
284 285
void ConcurrentMarkSweepGeneration::init_initiating_occupancy(intx io, uintx tr) {
  assert(io <= 100 && tr <= 100, "Check the arguments");
286 287 288 289 290 291 292 293 294
  if (io >= 0) {
    _initiating_occupancy = (double)io / 100.0;
  } else {
    _initiating_occupancy = ((100 - MinHeapFreeRatio) +
                             (double)(tr * MinHeapFreeRatio) / 100.0)
                            / 100.0;
  }
}

D
duke 已提交
295 296 297 298 299 300 301 302
void ConcurrentMarkSweepGeneration::ref_processor_init() {
  assert(collector() != NULL, "no collector");
  collector()->ref_processor_init();
}

void CMSCollector::ref_processor_init() {
  if (_ref_processor == NULL) {
    // Allocate and initialize a reference processor
303 304 305 306 307 308 309 310 311
    _ref_processor =
      new ReferenceProcessor(_span,                               // span
                             (ParallelGCThreads > 1) && ParallelRefProcEnabled, // mt processing
                             (int) ParallelGCThreads,             // mt processing degree
                             _cmsGen->refs_discovery_is_mt(),     // mt discovery
                             (int) MAX2(ConcGCThreads, ParallelGCThreads), // mt discovery degree
                             _cmsGen->refs_discovery_is_atomic(), // discovery is not atomic
                             &_is_alive_closure,                  // closure for liveness info
                             false);                              // next field updates do not need write barrier
D
duke 已提交
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
    // Initialize the _ref_processor field of CMSGen
    _cmsGen->set_ref_processor(_ref_processor);

  }
}

CMSAdaptiveSizePolicy* CMSCollector::size_policy() {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  assert(gch->kind() == CollectedHeap::GenCollectedHeap,
    "Wrong type of heap");
  CMSAdaptiveSizePolicy* sp = (CMSAdaptiveSizePolicy*)
    gch->gen_policy()->size_policy();
  assert(sp->is_gc_cms_adaptive_size_policy(),
    "Wrong type of size policy");
  return sp;
}

CMSGCAdaptivePolicyCounters* CMSCollector::gc_adaptive_policy_counters() {
  CMSGCAdaptivePolicyCounters* results =
    (CMSGCAdaptivePolicyCounters*) collector_policy()->counters();
  assert(
    results->kind() == GCPolicyCounters::CMSGCAdaptivePolicyCountersKind,
    "Wrong gc policy counter kind");
  return results;
}


void ConcurrentMarkSweepGeneration::initialize_performance_counters() {

  const char* gen_name = "old";

  // Generation Counters - generation 1, 1 subspace
  _gen_counters = new GenerationCounters(gen_name, 1, 1, &_virtual_space);

  _space_counters = new GSpaceCounters(gen_name, 0,
                                       _virtual_space.reserved_size(),
                                       this, _gen_counters);
}

CMSStats::CMSStats(ConcurrentMarkSweepGeneration* cms_gen, unsigned int alpha):
  _cms_gen(cms_gen)
{
  assert(alpha <= 100, "bad value");
  _saved_alpha = alpha;

  // Initialize the alphas to the bootstrap value of 100.
  _gc0_alpha = _cms_alpha = 100;

  _cms_begin_time.update();
  _cms_end_time.update();

  _gc0_duration = 0.0;
  _gc0_period = 0.0;
  _gc0_promoted = 0;

  _cms_duration = 0.0;
  _cms_period = 0.0;
  _cms_allocated = 0;

  _cms_used_at_gc0_begin = 0;
  _cms_used_at_gc0_end = 0;
  _allow_duty_cycle_reduction = false;
  _valid_bits = 0;
  _icms_duty_cycle = CMSIncrementalDutyCycle;
}

378 379 380 381 382 383 384 385
double CMSStats::cms_free_adjustment_factor(size_t free) const {
  // TBD: CR 6909490
  return 1.0;
}

void CMSStats::adjust_cms_free_adjustment_factor(bool fail, size_t free) {
}

D
duke 已提交
386 387 388 389 390 391
// If promotion failure handling is on use
// the padded average size of the promotion for each
// young generation collection.
double CMSStats::time_until_cms_gen_full() const {
  size_t cms_free = _cms_gen->cmsSpace()->free();
  GenCollectedHeap* gch = GenCollectedHeap::heap();
392 393
  size_t expected_promotion = MIN2(gch->get_gen(0)->capacity(),
                                   (size_t) _cms_gen->gc_stats()->avg_promoted()->padded_average());
D
duke 已提交
394 395 396 397 398 399 400 401
  if (cms_free > expected_promotion) {
    // Start a cms collection if there isn't enough space to promote
    // for the next minor collection.  Use the padded average as
    // a safety factor.
    cms_free -= expected_promotion;

    // Adjust by the safety factor.
    double cms_free_dbl = (double)cms_free;
402 403 404 405 406
    double cms_adjustment = (100.0 - CMSIncrementalSafetyFactor)/100.0;
    // Apply a further correction factor which tries to adjust
    // for recent occurance of concurrent mode failures.
    cms_adjustment = cms_adjustment * cms_free_adjustment_factor(cms_free);
    cms_free_dbl = cms_free_dbl * cms_adjustment;
D
duke 已提交
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439

    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print_cr("CMSStats::time_until_cms_gen_full: cms_free "
        SIZE_FORMAT " expected_promotion " SIZE_FORMAT,
        cms_free, expected_promotion);
      gclog_or_tty->print_cr("  cms_free_dbl %f cms_consumption_rate %f",
        cms_free_dbl, cms_consumption_rate() + 1.0);
    }
    // Add 1 in case the consumption rate goes to zero.
    return cms_free_dbl / (cms_consumption_rate() + 1.0);
  }
  return 0.0;
}

// Compare the duration of the cms collection to the
// time remaining before the cms generation is empty.
// Note that the time from the start of the cms collection
// to the start of the cms sweep (less than the total
// duration of the cms collection) can be used.  This
// has been tried and some applications experienced
// promotion failures early in execution.  This was
// possibly because the averages were not accurate
// enough at the beginning.
double CMSStats::time_until_cms_start() const {
  // We add "gc0_period" to the "work" calculation
  // below because this query is done (mostly) at the
  // end of a scavenge, so we need to conservatively
  // account for that much possible delay
  // in the query so as to avoid concurrent mode failures
  // due to starting the collection just a wee bit too
  // late.
  double work = cms_duration() + gc0_period();
  double deadline = time_until_cms_gen_full();
440 441
  // If a concurrent mode failure occurred recently, we want to be
  // more conservative and halve our expected time_until_cms_gen_full()
D
duke 已提交
442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564
  if (work > deadline) {
    if (Verbose && PrintGCDetails) {
      gclog_or_tty->print(
        " CMSCollector: collect because of anticipated promotion "
        "before full %3.7f + %3.7f > %3.7f ", cms_duration(),
        gc0_period(), time_until_cms_gen_full());
    }
    return 0.0;
  }
  return work - deadline;
}

// Return a duty cycle based on old_duty_cycle and new_duty_cycle, limiting the
// amount of change to prevent wild oscillation.
unsigned int CMSStats::icms_damped_duty_cycle(unsigned int old_duty_cycle,
                                              unsigned int new_duty_cycle) {
  assert(old_duty_cycle <= 100, "bad input value");
  assert(new_duty_cycle <= 100, "bad input value");

  // Note:  use subtraction with caution since it may underflow (values are
  // unsigned).  Addition is safe since we're in the range 0-100.
  unsigned int damped_duty_cycle = new_duty_cycle;
  if (new_duty_cycle < old_duty_cycle) {
    const unsigned int largest_delta = MAX2(old_duty_cycle / 4, 5U);
    if (new_duty_cycle + largest_delta < old_duty_cycle) {
      damped_duty_cycle = old_duty_cycle - largest_delta;
    }
  } else if (new_duty_cycle > old_duty_cycle) {
    const unsigned int largest_delta = MAX2(old_duty_cycle / 4, 15U);
    if (new_duty_cycle > old_duty_cycle + largest_delta) {
      damped_duty_cycle = MIN2(old_duty_cycle + largest_delta, 100U);
    }
  }
  assert(damped_duty_cycle <= 100, "invalid duty cycle computed");

  if (CMSTraceIncrementalPacing) {
    gclog_or_tty->print(" [icms_damped_duty_cycle(%d,%d) = %d] ",
                           old_duty_cycle, new_duty_cycle, damped_duty_cycle);
  }
  return damped_duty_cycle;
}

unsigned int CMSStats::icms_update_duty_cycle_impl() {
  assert(CMSIncrementalPacing && valid(),
         "should be handled in icms_update_duty_cycle()");

  double cms_time_so_far = cms_timer().seconds();
  double scaled_duration = cms_duration_per_mb() * _cms_used_at_gc0_end / M;
  double scaled_duration_remaining = fabsd(scaled_duration - cms_time_so_far);

  // Avoid division by 0.
  double time_until_full = MAX2(time_until_cms_gen_full(), 0.01);
  double duty_cycle_dbl = 100.0 * scaled_duration_remaining / time_until_full;

  unsigned int new_duty_cycle = MIN2((unsigned int)duty_cycle_dbl, 100U);
  if (new_duty_cycle > _icms_duty_cycle) {
    // Avoid very small duty cycles (1 or 2); 0 is allowed.
    if (new_duty_cycle > 2) {
      _icms_duty_cycle = icms_damped_duty_cycle(_icms_duty_cycle,
                                                new_duty_cycle);
    }
  } else if (_allow_duty_cycle_reduction) {
    // The duty cycle is reduced only once per cms cycle (see record_cms_end()).
    new_duty_cycle = icms_damped_duty_cycle(_icms_duty_cycle, new_duty_cycle);
    // Respect the minimum duty cycle.
    unsigned int min_duty_cycle = (unsigned int)CMSIncrementalDutyCycleMin;
    _icms_duty_cycle = MAX2(new_duty_cycle, min_duty_cycle);
  }

  if (PrintGCDetails || CMSTraceIncrementalPacing) {
    gclog_or_tty->print(" icms_dc=%d ", _icms_duty_cycle);
  }

  _allow_duty_cycle_reduction = false;
  return _icms_duty_cycle;
}

#ifndef PRODUCT
void CMSStats::print_on(outputStream *st) const {
  st->print(" gc0_alpha=%d,cms_alpha=%d", _gc0_alpha, _cms_alpha);
  st->print(",gc0_dur=%g,gc0_per=%g,gc0_promo=" SIZE_FORMAT,
               gc0_duration(), gc0_period(), gc0_promoted());
  st->print(",cms_dur=%g,cms_dur_per_mb=%g,cms_per=%g,cms_alloc=" SIZE_FORMAT,
            cms_duration(), cms_duration_per_mb(),
            cms_period(), cms_allocated());
  st->print(",cms_since_beg=%g,cms_since_end=%g",
            cms_time_since_begin(), cms_time_since_end());
  st->print(",cms_used_beg=" SIZE_FORMAT ",cms_used_end=" SIZE_FORMAT,
            _cms_used_at_gc0_begin, _cms_used_at_gc0_end);
  if (CMSIncrementalMode) {
    st->print(",dc=%d", icms_duty_cycle());
  }

  if (valid()) {
    st->print(",promo_rate=%g,cms_alloc_rate=%g",
              promotion_rate(), cms_allocation_rate());
    st->print(",cms_consumption_rate=%g,time_until_full=%g",
              cms_consumption_rate(), time_until_cms_gen_full());
  }
  st->print(" ");
}
#endif // #ifndef PRODUCT

CMSCollector::CollectorState CMSCollector::_collectorState =
                             CMSCollector::Idling;
bool CMSCollector::_foregroundGCIsActive = false;
bool CMSCollector::_foregroundGCShouldWait = false;

CMSCollector::CMSCollector(ConcurrentMarkSweepGeneration* cmsGen,
                           CardTableRS*                   ct,
                           ConcurrentMarkSweepPolicy*     cp):
  _cmsGen(cmsGen),
  _ct(ct),
  _ref_processor(NULL),    // will be set later
  _conc_workers(NULL),     // may be set later
  _abort_preclean(false),
  _start_sampling(false),
  _between_prologue_and_epilogue(false),
  _markBitMap(0, Mutex::leaf + 1, "CMS_markBitMap_lock"),
  _modUnionTable((CardTableModRefBS::card_shift - LogHeapWordSize),
                 -1 /* lock-free */, "No_lock" /* dummy */),
  _modUnionClosure(&_modUnionTable),
  _modUnionClosurePar(&_modUnionTable),
565 566
  // Adjust my span to cover old (cms) gen
  _span(cmsGen->reserved()),
567 568
  // Construct the is_alive_closure with _span & markBitMap
  _is_alive_closure(_span, &_markBitMap),
D
duke 已提交
569 570 571
  _restart_addr(NULL),
  _overflow_list(NULL),
  _stats(cmsGen),
572
  _eden_chunk_lock(new Mutex(Mutex::leaf + 1, "CMS_eden_chunk_lock", true)),
D
duke 已提交
573 574 575 576 577 578 579 580
  _eden_chunk_array(NULL),     // may be set in ctor body
  _eden_chunk_capacity(0),     // -- ditto --
  _eden_chunk_index(0),        // -- ditto --
  _survivor_plab_array(NULL),  // -- ditto --
  _survivor_chunk_array(NULL), // -- ditto --
  _survivor_chunk_capacity(0), // -- ditto --
  _survivor_chunk_index(0),    // -- ditto --
  _ser_pmc_preclean_ovflw(0),
581
  _ser_kac_preclean_ovflw(0),
D
duke 已提交
582 583 584 585 586 587 588 589 590 591 592 593 594 595
  _ser_pmc_remark_ovflw(0),
  _par_pmc_remark_ovflw(0),
  _ser_kac_ovflw(0),
  _par_kac_ovflw(0),
#ifndef PRODUCT
  _num_par_pushes(0),
#endif
  _collection_count_start(0),
  _verifying(false),
  _icms_start_limit(NULL),
  _icms_stop_limit(NULL),
  _verification_mark_bm(0, Mutex::leaf + 1, "CMS_verification_mark_bm_lock"),
  _completed_initialization(false),
  _collector_policy(cp),
596 597
  _should_unload_classes(false),
  _concurrent_cycles_since_last_unload(0),
598
  _roots_scanning_options(0),
599
  _inter_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding),
S
sla 已提交
600 601 602 603
  _intra_sweep_estimate(CMS_SweepWeight, CMS_SweepPadding),
  _gc_tracer_cm(new (ResourceObj::C_HEAP, mtGC) CMSTracer()),
  _gc_timer_cm(new (ResourceObj::C_HEAP, mtGC) ConcurrentGCTimer()),
  _cms_start_registered(false)
D
duke 已提交
604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628
{
  if (ExplicitGCInvokesConcurrentAndUnloadsClasses) {
    ExplicitGCInvokesConcurrent = true;
  }
  // Now expand the span and allocate the collection support structures
  // (MUT, marking bit map etc.) to cover both generations subject to
  // collection.

  // For use by dirty card to oop closures.
  _cmsGen->cmsSpace()->set_collector(this);

  // Allocate MUT and marking bit map
  {
    MutexLockerEx x(_markBitMap.lock(), Mutex::_no_safepoint_check_flag);
    if (!_markBitMap.allocate(_span)) {
      warning("Failed to allocate CMS Bit Map");
      return;
    }
    assert(_markBitMap.covers(_span), "_markBitMap inconsistency?");
  }
  {
    _modUnionTable.allocate(_span);
    assert(_modUnionTable.covers(_span), "_modUnionTable inconsistency?");
  }

629
  if (!_markStack.allocate(MarkStackSize)) {
D
duke 已提交
630 631 632 633 634
    warning("Failed to allocate CMS Marking Stack");
    return;
  }

  // Support for multi-threaded concurrent phases
635
  if (CMSConcurrentMTEnabled) {
636
    if (FLAG_IS_DEFAULT(ConcGCThreads)) {
D
duke 已提交
637
      // just for now
638
      FLAG_SET_DEFAULT(ConcGCThreads, (ParallelGCThreads + 3)/4);
D
duke 已提交
639
    }
640
    if (ConcGCThreads > 1) {
D
duke 已提交
641
      _conc_workers = new YieldingFlexibleWorkGang("Parallel CMS Threads",
642
                                 ConcGCThreads, true);
D
duke 已提交
643 644 645 646
      if (_conc_workers == NULL) {
        warning("GC/CMS: _conc_workers allocation failure: "
              "forcing -CMSConcurrentMTEnabled");
        CMSConcurrentMTEnabled = false;
647 648
      } else {
        _conc_workers->initialize_workers();
D
duke 已提交
649 650 651 652 653 654
      }
    } else {
      CMSConcurrentMTEnabled = false;
    }
  }
  if (!CMSConcurrentMTEnabled) {
655
    ConcGCThreads = 0;
D
duke 已提交
656 657 658 659 660
  } else {
    // Turn off CMSCleanOnEnter optimization temporarily for
    // the MT case where it's not fixed yet; see 6178663.
    CMSCleanOnEnter = false;
  }
661
  assert((_conc_workers != NULL) == (ConcGCThreads > 1),
D
duke 已提交
662 663 664 665 666 667 668
         "Inconsistency");

  // Parallel task queues; these are shared for the
  // concurrent and stop-world phases of CMS, but
  // are not shared with parallel scavenge (ParNew).
  {
    uint i;
669
    uint num_queues = (uint) MAX2(ParallelGCThreads, ConcGCThreads);
D
duke 已提交
670 671 672 673 674 675 676 677 678

    if ((CMSParallelRemarkEnabled || CMSConcurrentMTEnabled
         || ParallelRefProcEnabled)
        && num_queues > 0) {
      _task_queues = new OopTaskQueueSet(num_queues);
      if (_task_queues == NULL) {
        warning("task_queues allocation failure.");
        return;
      }
Z
zgu 已提交
679
      _hash_seed = NEW_C_HEAP_ARRAY(int, num_queues, mtGC);
D
duke 已提交
680 681 682 683 684
      if (_hash_seed == NULL) {
        warning("_hash_seed array allocation failure");
        return;
      }

J
jcoomes 已提交
685
      typedef Padded<OopTaskQueue> PaddedOopTaskQueue;
D
duke 已提交
686
      for (i = 0; i < num_queues; i++) {
J
jcoomes 已提交
687 688
        PaddedOopTaskQueue *q = new PaddedOopTaskQueue();
        if (q == NULL) {
D
duke 已提交
689 690 691
          warning("work_queue allocation failure.");
          return;
        }
J
jcoomes 已提交
692
        _task_queues->register_queue(i, q);
D
duke 已提交
693 694 695 696 697 698 699 700
      }
      for (i = 0; i < num_queues; i++) {
        _task_queues->queue(i)->initialize();
        _hash_seed[i] = 17;  // copied from ParNew
      }
    }
  }

701 702
  _cmsGen ->init_initiating_occupancy(CMSInitiatingOccupancyFraction, CMSTriggerRatio);

D
duke 已提交
703
  // Clip CMSBootstrapOccupancy between 0 and 100.
704
  _bootstrap_occupancy = ((double)CMSBootstrapOccupancy)/(double)100;
D
duke 已提交
705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726

  _full_gcs_since_conc_gc = 0;

  // Now tell CMS generations the identity of their collector
  ConcurrentMarkSweepGeneration::set_collector(this);

  // Create & start a CMS thread for this CMS collector
  _cmsThread = ConcurrentMarkSweepThread::start(this);
  assert(cmsThread() != NULL, "CMS Thread should have been created");
  assert(cmsThread()->collector() == this,
         "CMS Thread should refer to this gen");
  assert(CGC_lock != NULL, "Where's the CGC_lock?");

  // Support for parallelizing young gen rescan
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  _young_gen = gch->prev_gen(_cmsGen);
  if (gch->supports_inline_contig_alloc()) {
    _top_addr = gch->top_addr();
    _end_addr = gch->end_addr();
    assert(_young_gen != NULL, "no _young_gen");
    _eden_chunk_index = 0;
    _eden_chunk_capacity = (_young_gen->max_capacity()+CMSSamplingGrain)/CMSSamplingGrain;
Z
zgu 已提交
727
    _eden_chunk_array = NEW_C_HEAP_ARRAY(HeapWord*, _eden_chunk_capacity, mtGC);
D
duke 已提交
728 729 730 731 732 733 734 735
    if (_eden_chunk_array == NULL) {
      _eden_chunk_capacity = 0;
      warning("GC/CMS: _eden_chunk_array allocation failure");
    }
  }
  assert(_eden_chunk_array != NULL || _eden_chunk_capacity == 0, "Error");

  // Support for parallelizing survivor space rescan
J
jmasa 已提交
736
  if ((CMSParallelRemarkEnabled && CMSParallelSurvivorRemarkEnabled) || CMSParallelInitialMarkEnabled) {
737 738 739
    const size_t max_plab_samples =
      ((DefNewGeneration*)_young_gen)->max_survivor_size()/MinTLABSize;

Z
zgu 已提交
740 741 742
    _survivor_plab_array  = NEW_C_HEAP_ARRAY(ChunkArray, ParallelGCThreads, mtGC);
    _survivor_chunk_array = NEW_C_HEAP_ARRAY(HeapWord*, 2*max_plab_samples, mtGC);
    _cursor               = NEW_C_HEAP_ARRAY(size_t, ParallelGCThreads, mtGC);
D
duke 已提交
743 744 745 746
    if (_survivor_plab_array == NULL || _survivor_chunk_array == NULL
        || _cursor == NULL) {
      warning("Failed to allocate survivor plab/chunk array");
      if (_survivor_plab_array  != NULL) {
Z
zgu 已提交
747
        FREE_C_HEAP_ARRAY(ChunkArray, _survivor_plab_array, mtGC);
D
duke 已提交
748 749 750
        _survivor_plab_array = NULL;
      }
      if (_survivor_chunk_array != NULL) {
Z
zgu 已提交
751
        FREE_C_HEAP_ARRAY(HeapWord*, _survivor_chunk_array, mtGC);
D
duke 已提交
752 753 754
        _survivor_chunk_array = NULL;
      }
      if (_cursor != NULL) {
Z
zgu 已提交
755
        FREE_C_HEAP_ARRAY(size_t, _cursor, mtGC);
D
duke 已提交
756 757 758 759 760
        _cursor = NULL;
      }
    } else {
      _survivor_chunk_capacity = 2*max_plab_samples;
      for (uint i = 0; i < ParallelGCThreads; i++) {
Z
zgu 已提交
761
        HeapWord** vec = NEW_C_HEAP_ARRAY(HeapWord*, max_plab_samples, mtGC);
D
duke 已提交
762 763 764
        if (vec == NULL) {
          warning("Failed to allocate survivor plab array");
          for (int j = i; j > 0; j--) {
Z
zgu 已提交
765
            FREE_C_HEAP_ARRAY(HeapWord*, _survivor_plab_array[j-1].array(), mtGC);
D
duke 已提交
766
          }
Z
zgu 已提交
767 768
          FREE_C_HEAP_ARRAY(ChunkArray, _survivor_plab_array, mtGC);
          FREE_C_HEAP_ARRAY(HeapWord*, _survivor_chunk_array, mtGC);
D
duke 已提交
769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800
          _survivor_plab_array = NULL;
          _survivor_chunk_array = NULL;
          _survivor_chunk_capacity = 0;
          break;
        } else {
          ChunkArray* cur =
            ::new (&_survivor_plab_array[i]) ChunkArray(vec,
                                                        max_plab_samples);
          assert(cur->end() == 0, "Should be 0");
          assert(cur->array() == vec, "Should be vec");
          assert(cur->capacity() == max_plab_samples, "Error");
        }
      }
    }
  }
  assert(   (   _survivor_plab_array  != NULL
             && _survivor_chunk_array != NULL)
         || (   _survivor_chunk_capacity == 0
             && _survivor_chunk_index == 0),
         "Error");

  // Choose what strong roots should be scanned depending on verification options
  if (!CMSClassUnloadingEnabled) {
    // If class unloading is disabled we want to include all classes into the root set.
    add_root_scanning_option(SharedHeap::SO_AllClasses);
  } else {
    add_root_scanning_option(SharedHeap::SO_SystemClasses);
  }

  NOT_PRODUCT(_overflow_counter = CMSMarkStackOverflowInterval;)
  _gc_counters = new CollectorCounters("CMS", 1);
  _completed_initialization = true;
801
  _inter_sweep_timer.start();  // start of time
D
duke 已提交
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839
}

const char* ConcurrentMarkSweepGeneration::name() const {
  return "concurrent mark-sweep generation";
}
void ConcurrentMarkSweepGeneration::update_counters() {
  if (UsePerfData) {
    _space_counters->update_all();
    _gen_counters->update_all();
  }
}

// this is an optimized version of update_counters(). it takes the
// used value as a parameter rather than computing it.
//
void ConcurrentMarkSweepGeneration::update_counters(size_t used) {
  if (UsePerfData) {
    _space_counters->update_used(used);
    _space_counters->update_capacity();
    _gen_counters->update_all();
  }
}

void ConcurrentMarkSweepGeneration::print() const {
  Generation::print();
  cmsSpace()->print();
}

#ifndef PRODUCT
void ConcurrentMarkSweepGeneration::print_statistics() {
  cmsSpace()->printFLCensus(0);
}
#endif

void ConcurrentMarkSweepGeneration::printOccupancy(const char *s) {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  if (PrintGCDetails) {
    if (Verbose) {
840
      gclog_or_tty->print("[%d %s-%s: "SIZE_FORMAT"("SIZE_FORMAT")]",
D
duke 已提交
841 842
        level(), short_name(), s, used(), capacity());
    } else {
843
      gclog_or_tty->print("[%d %s-%s: "SIZE_FORMAT"K("SIZE_FORMAT"K)]",
D
duke 已提交
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
        level(), short_name(), s, used() / K, capacity() / K);
    }
  }
  if (Verbose) {
    gclog_or_tty->print(" "SIZE_FORMAT"("SIZE_FORMAT")",
              gch->used(), gch->capacity());
  } else {
    gclog_or_tty->print(" "SIZE_FORMAT"K("SIZE_FORMAT"K)",
              gch->used() / K, gch->capacity() / K);
  }
}

size_t
ConcurrentMarkSweepGeneration::contiguous_available() const {
  // dld proposes an improvement in precision here. If the committed
  // part of the space ends in a free block we should add that to
  // uncommitted size in the calculation below. Will make this
  // change later, staying with the approximation below for the
  // time being. -- ysr.
  return MAX2(_virtual_space.uncommitted_size(), unsafe_max_alloc_nogc());
}

size_t
ConcurrentMarkSweepGeneration::unsafe_max_alloc_nogc() const {
  return _cmsSpace->max_alloc_in_words() * HeapWordSize;
}

size_t ConcurrentMarkSweepGeneration::max_available() const {
  return free() + _virtual_space.uncommitted_size();
}

875 876 877 878
bool ConcurrentMarkSweepGeneration::promotion_attempt_is_safe(size_t max_promotion_in_bytes) const {
  size_t available = max_available();
  size_t av_promo  = (size_t)gc_stats()->avg_promoted()->padded_average();
  bool   res = (available >= av_promo) || (available >= max_promotion_in_bytes);
879
  if (Verbose && PrintGCDetails) {
880 881 882 883 884
    gclog_or_tty->print_cr(
      "CMS: promo attempt is%s safe: available("SIZE_FORMAT") %s av_promo("SIZE_FORMAT"),"
      "max_promo("SIZE_FORMAT")",
      res? "":" not", available, res? ">=":"<",
      av_promo, max_promotion_in_bytes);
D
duke 已提交
885
  }
886
  return res;
D
duke 已提交
887 888
}

889 890 891 892 893 894 895 896
// At a promotion failure dump information on block layout in heap
// (cms old generation).
void ConcurrentMarkSweepGeneration::promotion_failure_occurred() {
  if (CMSDumpAtPromotionFailure) {
    cmsSpace()->dump_at_safepoint_with_locks(collector(), gclog_or_tty);
  }
}

D
duke 已提交
897 898 899 900 901 902 903 904 905 906
CompactibleSpace*
ConcurrentMarkSweepGeneration::first_compaction_space() const {
  return _cmsSpace;
}

void ConcurrentMarkSweepGeneration::reset_after_compaction() {
  // Clear the promotion information.  These pointers can be adjusted
  // along with all the other pointers into the heap but
  // compaction is expected to be a rare event with
  // a heap using cms so don't do it without seeing the need.
907
  if (CollectedHeap::use_parallel_gc_threads()) {
D
duke 已提交
908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928
    for (uint i = 0; i < ParallelGCThreads; i++) {
      _par_gc_thread_states[i]->promo.reset();
    }
  }
}

void ConcurrentMarkSweepGeneration::space_iterate(SpaceClosure* blk, bool usedOnly) {
  blk->do_space(_cmsSpace);
}

void ConcurrentMarkSweepGeneration::compute_new_size() {
  assert_locked_or_safepoint(Heap_lock);

  // If incremental collection failed, we just want to expand
  // to the limit.
  if (incremental_collection_failed()) {
    clear_incremental_collection_failed();
    grow_to_reserved();
    return;
  }

929 930
  // The heap has been compacted but not reset yet.
  // Any metric such as free() or used() will be incorrect.
931 932 933 934

  CardGeneration::compute_new_size();

  // Reset again after a possible resizing
935 936 937
  if (did_compact()) {
    cmsSpace()->reset_after_compaction();
  }
938 939 940 941 942 943 944 945 946 947 948 949 950
}

void ConcurrentMarkSweepGeneration::compute_new_size_free_list() {
  assert_locked_or_safepoint(Heap_lock);

  // If incremental collection failed, we just want to expand
  // to the limit.
  if (incremental_collection_failed()) {
    clear_incremental_collection_failed();
    grow_to_reserved();
    return;
  }

D
duke 已提交
951 952 953 954 955 956 957 958
  double free_percentage = ((double) free()) / capacity();
  double desired_free_percentage = (double) MinHeapFreeRatio / 100;
  double maximum_free_percentage = (double) MaxHeapFreeRatio / 100;

  // compute expansion delta needed for reaching desired free percentage
  if (free_percentage < desired_free_percentage) {
    size_t desired_capacity = (size_t)(used() / ((double) 1 - desired_free_percentage));
    assert(desired_capacity >= capacity(), "invalid expansion size");
959
    size_t expand_bytes = MAX2(desired_capacity - capacity(), MinHeapDeltaBytes);
D
duke 已提交
960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992
    if (PrintGCDetails && Verbose) {
      size_t desired_capacity = (size_t)(used() / ((double) 1 - desired_free_percentage));
      gclog_or_tty->print_cr("\nFrom compute_new_size: ");
      gclog_or_tty->print_cr("  Free fraction %f", free_percentage);
      gclog_or_tty->print_cr("  Desired free fraction %f",
        desired_free_percentage);
      gclog_or_tty->print_cr("  Maximum free fraction %f",
        maximum_free_percentage);
      gclog_or_tty->print_cr("  Capactiy "SIZE_FORMAT, capacity()/1000);
      gclog_or_tty->print_cr("  Desired capacity "SIZE_FORMAT,
        desired_capacity/1000);
      int prev_level = level() - 1;
      if (prev_level >= 0) {
        size_t prev_size = 0;
        GenCollectedHeap* gch = GenCollectedHeap::heap();
        Generation* prev_gen = gch->_gens[prev_level];
        prev_size = prev_gen->capacity();
          gclog_or_tty->print_cr("  Younger gen size "SIZE_FORMAT,
                                 prev_size/1000);
      }
      gclog_or_tty->print_cr("  unsafe_max_alloc_nogc "SIZE_FORMAT,
        unsafe_max_alloc_nogc()/1000);
      gclog_or_tty->print_cr("  contiguous available "SIZE_FORMAT,
        contiguous_available()/1000);
      gclog_or_tty->print_cr("  Expand by "SIZE_FORMAT" (bytes)",
        expand_bytes);
    }
    // safe if expansion fails
    expand(expand_bytes, 0, CMSExpansionCause::_satisfy_free_ratio);
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print_cr("  Expanded free fraction %f",
        ((double) free()) / capacity());
    }
993 994 995 996 997 998 999 1000
  } else {
    size_t desired_capacity = (size_t)(used() / ((double) 1 - desired_free_percentage));
    assert(desired_capacity <= capacity(), "invalid expansion size");
    size_t shrink_bytes = capacity() - desired_capacity;
    // Don't shrink unless the delta is greater than the minimum shrink we want
    if (shrink_bytes >= MinHeapDeltaBytes) {
      shrink_free_list_by(shrink_bytes);
    }
D
duke 已提交
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
  }
}

Mutex* ConcurrentMarkSweepGeneration::freelistLock() const {
  return cmsSpace()->freelistLock();
}

HeapWord* ConcurrentMarkSweepGeneration::allocate(size_t size,
                                                  bool   tlab) {
  CMSSynchronousYieldRequest yr;
  MutexLockerEx x(freelistLock(),
                  Mutex::_no_safepoint_check_flag);
  return have_lock_and_allocate(size, tlab);
}

HeapWord* ConcurrentMarkSweepGeneration::have_lock_and_allocate(size_t size,
1017
                                                  bool   tlab /* ignored */) {
D
duke 已提交
1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
  assert_lock_strong(freelistLock());
  size_t adjustedSize = CompactibleFreeListSpace::adjustObjectSize(size);
  HeapWord* res = cmsSpace()->allocate(adjustedSize);
  // Allocate the object live (grey) if the background collector has
  // started marking. This is necessary because the marker may
  // have passed this address and consequently this object will
  // not otherwise be greyed and would be incorrectly swept up.
  // Note that if this object contains references, the writing
  // of those references will dirty the card containing this object
  // allowing the object to be blackened (and its references scanned)
  // either during a preclean phase or at the final checkpoint.
  if (res != NULL) {
1030 1031 1032 1033
    // We may block here with an uninitialized object with
    // its mark-bit or P-bits not yet set. Such objects need
    // to be safely navigable by block_start().
    assert(oop(res)->klass_or_null() == NULL, "Object should be uninitialized here.");
1034
    assert(!((FreeChunk*)res)->is_free(), "Error, block will look free but show wrong size");
D
duke 已提交
1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
    collector()->direct_allocated(res, adjustedSize);
    _direct_allocated_words += adjustedSize;
    // allocation counters
    NOT_PRODUCT(
      _numObjectsAllocated++;
      _numWordsAllocated += (int)adjustedSize;
    )
  }
  return res;
}

// In the case of direct allocation by mutators in a generation that
// is being concurrently collected, the object must be allocated
// live (grey) if the background collector has started marking.
// This is necessary because the marker may
// have passed this address and consequently this object will
// not otherwise be greyed and would be incorrectly swept up.
// Note that if this object contains references, the writing
// of those references will dirty the card containing this object
// allowing the object to be blackened (and its references scanned)
// either during a preclean phase or at the final checkpoint.
void CMSCollector::direct_allocated(HeapWord* start, size_t size) {
  assert(_markBitMap.covers(start, size), "Out of bounds");
  if (_collectorState >= Marking) {
    MutexLockerEx y(_markBitMap.lock(),
                    Mutex::_no_safepoint_check_flag);
    // [see comments preceding SweepClosure::do_blk() below for details]
1062 1063 1064
    //
    // Can the P-bits be deleted now?  JJJ
    //
D
duke 已提交
1065 1066
    // 1. need to mark the object as live so it isn't collected
    // 2. need to mark the 2nd bit to indicate the object may be uninitialized
1067 1068 1069
    // 3. need to mark the end of the object so marking, precleaning or sweeping
    //    can skip over uninitialized or unparsable objects. An allocated
    //    object is considered uninitialized for our purposes as long as
1070
    //    its klass word is NULL.  All old gen objects are parsable
1071
    //    as soon as they are initialized.)
D
duke 已提交
1072 1073 1074 1075 1076 1077
    _markBitMap.mark(start);          // object is live
    _markBitMap.mark(start + 1);      // object is potentially uninitialized?
    _markBitMap.mark(start + size - 1);
                                      // mark end of object
  }
  // check that oop looks uninitialized
1078
  assert(oop(start)->klass_or_null() == NULL, "_klass should be NULL");
D
duke 已提交
1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096
}

void CMSCollector::promoted(bool par, HeapWord* start,
                            bool is_obj_array, size_t obj_size) {
  assert(_markBitMap.covers(start), "Out of bounds");
  // See comment in direct_allocated() about when objects should
  // be allocated live.
  if (_collectorState >= Marking) {
    // we already hold the marking bit map lock, taken in
    // the prologue
    if (par) {
      _markBitMap.par_mark(start);
    } else {
      _markBitMap.mark(start);
    }
    // We don't need to mark the object as uninitialized (as
    // in direct_allocated above) because this is being done with the
    // world stopped and the object will be initialized by the
1097 1098 1099 1100 1101 1102 1103
    // time the marking, precleaning or sweeping get to look at it.
    // But see the code for copying objects into the CMS generation,
    // where we need to ensure that concurrent readers of the
    // block offset table are able to safely navigate a block that
    // is in flux from being free to being allocated (and in
    // transition while being copied into) and subsequently
    // becoming a bona-fide object when the copy/promotion is complete.
D
duke 已提交
1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268
    assert(SafepointSynchronize::is_at_safepoint(),
           "expect promotion only at safepoints");

    if (_collectorState < Sweeping) {
      // Mark the appropriate cards in the modUnionTable, so that
      // this object gets scanned before the sweep. If this is
      // not done, CMS generation references in the object might
      // not get marked.
      // For the case of arrays, which are otherwise precisely
      // marked, we need to dirty the entire array, not just its head.
      if (is_obj_array) {
        // The [par_]mark_range() method expects mr.end() below to
        // be aligned to the granularity of a bit's representation
        // in the heap. In the case of the MUT below, that's a
        // card size.
        MemRegion mr(start,
                     (HeapWord*)round_to((intptr_t)(start + obj_size),
                        CardTableModRefBS::card_size /* bytes */));
        if (par) {
          _modUnionTable.par_mark_range(mr);
        } else {
          _modUnionTable.mark_range(mr);
        }
      } else {  // not an obj array; we can just mark the head
        if (par) {
          _modUnionTable.par_mark(start);
        } else {
          _modUnionTable.mark(start);
        }
      }
    }
  }
}

static inline size_t percent_of_space(Space* space, HeapWord* addr)
{
  size_t delta = pointer_delta(addr, space->bottom());
  return (size_t)(delta * 100.0 / (space->capacity() / HeapWordSize));
}

void CMSCollector::icms_update_allocation_limits()
{
  Generation* gen0 = GenCollectedHeap::heap()->get_gen(0);
  EdenSpace* eden = gen0->as_DefNewGeneration()->eden();

  const unsigned int duty_cycle = stats().icms_update_duty_cycle();
  if (CMSTraceIncrementalPacing) {
    stats().print();
  }

  assert(duty_cycle <= 100, "invalid duty cycle");
  if (duty_cycle != 0) {
    // The duty_cycle is a percentage between 0 and 100; convert to words and
    // then compute the offset from the endpoints of the space.
    size_t free_words = eden->free() / HeapWordSize;
    double free_words_dbl = (double)free_words;
    size_t duty_cycle_words = (size_t)(free_words_dbl * duty_cycle / 100.0);
    size_t offset_words = (free_words - duty_cycle_words) / 2;

    _icms_start_limit = eden->top() + offset_words;
    _icms_stop_limit = eden->end() - offset_words;

    // The limits may be adjusted (shifted to the right) by
    // CMSIncrementalOffset, to allow the application more mutator time after a
    // young gen gc (when all mutators were stopped) and before CMS starts and
    // takes away one or more cpus.
    if (CMSIncrementalOffset != 0) {
      double adjustment_dbl = free_words_dbl * CMSIncrementalOffset / 100.0;
      size_t adjustment = (size_t)adjustment_dbl;
      HeapWord* tmp_stop = _icms_stop_limit + adjustment;
      if (tmp_stop > _icms_stop_limit && tmp_stop < eden->end()) {
        _icms_start_limit += adjustment;
        _icms_stop_limit = tmp_stop;
      }
    }
  }
  if (duty_cycle == 0 || (_icms_start_limit == _icms_stop_limit)) {
    _icms_start_limit = _icms_stop_limit = eden->end();
  }

  // Install the new start limit.
  eden->set_soft_end(_icms_start_limit);

  if (CMSTraceIncrementalMode) {
    gclog_or_tty->print(" icms alloc limits:  "
                           PTR_FORMAT "," PTR_FORMAT
                           " (" SIZE_FORMAT "%%," SIZE_FORMAT "%%) ",
                           _icms_start_limit, _icms_stop_limit,
                           percent_of_space(eden, _icms_start_limit),
                           percent_of_space(eden, _icms_stop_limit));
    if (Verbose) {
      gclog_or_tty->print("eden:  ");
      eden->print_on(gclog_or_tty);
    }
  }
}

// Any changes here should try to maintain the invariant
// that if this method is called with _icms_start_limit
// and _icms_stop_limit both NULL, then it should return NULL
// and not notify the icms thread.
HeapWord*
CMSCollector::allocation_limit_reached(Space* space, HeapWord* top,
                                       size_t word_size)
{
  // A start_limit equal to end() means the duty cycle is 0, so treat that as a
  // nop.
  if (CMSIncrementalMode && _icms_start_limit != space->end()) {
    if (top <= _icms_start_limit) {
      if (CMSTraceIncrementalMode) {
        space->print_on(gclog_or_tty);
        gclog_or_tty->stamp();
        gclog_or_tty->print_cr(" start limit top=" PTR_FORMAT
                               ", new limit=" PTR_FORMAT
                               " (" SIZE_FORMAT "%%)",
                               top, _icms_stop_limit,
                               percent_of_space(space, _icms_stop_limit));
      }
      ConcurrentMarkSweepThread::start_icms();
      assert(top < _icms_stop_limit, "Tautology");
      if (word_size < pointer_delta(_icms_stop_limit, top)) {
        return _icms_stop_limit;
      }

      // The allocation will cross both the _start and _stop limits, so do the
      // stop notification also and return end().
      if (CMSTraceIncrementalMode) {
        space->print_on(gclog_or_tty);
        gclog_or_tty->stamp();
        gclog_or_tty->print_cr(" +stop limit top=" PTR_FORMAT
                               ", new limit=" PTR_FORMAT
                               " (" SIZE_FORMAT "%%)",
                               top, space->end(),
                               percent_of_space(space, space->end()));
      }
      ConcurrentMarkSweepThread::stop_icms();
      return space->end();
    }

    if (top <= _icms_stop_limit) {
      if (CMSTraceIncrementalMode) {
        space->print_on(gclog_or_tty);
        gclog_or_tty->stamp();
        gclog_or_tty->print_cr(" stop limit top=" PTR_FORMAT
                               ", new limit=" PTR_FORMAT
                               " (" SIZE_FORMAT "%%)",
                               top, space->end(),
                               percent_of_space(space, space->end()));
      }
      ConcurrentMarkSweepThread::stop_icms();
      return space->end();
    }

    if (CMSTraceIncrementalMode) {
      space->print_on(gclog_or_tty);
      gclog_or_tty->stamp();
      gclog_or_tty->print_cr(" end limit top=" PTR_FORMAT
                             ", new limit=" PTR_FORMAT,
                             top, NULL);
    }
  }

  return NULL;
}

1269
oop ConcurrentMarkSweepGeneration::promote(oop obj, size_t obj_size) {
D
duke 已提交
1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280
  assert(obj_size == (size_t)obj->size(), "bad obj_size passed in");
  // allocate, copy and if necessary update promoinfo --
  // delegate to underlying space.
  assert_lock_strong(freelistLock());

#ifndef PRODUCT
  if (Universe::heap()->promotion_should_fail()) {
    return NULL;
  }
#endif  // #ifndef PRODUCT

1281
  oop res = _cmsSpace->promote(obj, obj_size);
D
duke 已提交
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291
  if (res == NULL) {
    // expand and retry
    size_t s = _cmsSpace->expansionSpaceRequired(obj_size);  // HeapWords
    expand(s*HeapWordSize, MinHeapDeltaBytes,
      CMSExpansionCause::_satisfy_promotion);
    // Since there's currently no next generation, we don't try to promote
    // into a more senior generation.
    assert(next_gen() == NULL, "assumption, based upon which no attempt "
                               "is made to pass on a possibly failing "
                               "promotion to next generation");
1292
    res = _cmsSpace->promote(obj, obj_size);
D
duke 已提交
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
  }
  if (res != NULL) {
    // See comment in allocate() about when objects should
    // be allocated live.
    assert(obj->is_oop(), "Will dereference klass pointer below");
    collector()->promoted(false,           // Not parallel
                          (HeapWord*)res, obj->is_objArray(), obj_size);
    // promotion counters
    NOT_PRODUCT(
      _numObjectsPromoted++;
      _numWordsPromoted +=
        (int)(CompactibleFreeListSpace::adjustObjectSize(obj->size()));
    )
  }
  return res;
}


HeapWord*
ConcurrentMarkSweepGeneration::allocation_limit_reached(Space* space,
                                             HeapWord* top,
                                             size_t word_sz)
{
  return collector()->allocation_limit_reached(space, top, word_sz);
}

1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329
// IMPORTANT: Notes on object size recognition in CMS.
// ---------------------------------------------------
// A block of storage in the CMS generation is always in
// one of three states. A free block (FREE), an allocated
// object (OBJECT) whose size() method reports the correct size,
// and an intermediate state (TRANSIENT) in which its size cannot
// be accurately determined.
// STATE IDENTIFICATION:   (32 bit and 64 bit w/o COOPS)
// -----------------------------------------------------
// FREE:      klass_word & 1 == 1; mark_word holds block size
//
1330
// OBJECT:    klass_word installed; klass_word != 0 && klass_word & 1 == 0;
1331 1332 1333 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
//            obj->size() computes correct size
//
// TRANSIENT: klass_word == 0; size is indeterminate until we become an OBJECT
//
// STATE IDENTIFICATION: (64 bit+COOPS)
// ------------------------------------
// FREE:      mark_word & CMS_FREE_BIT == 1; mark_word & ~CMS_FREE_BIT gives block_size
//
// OBJECT:    klass_word installed; klass_word != 0;
//            obj->size() computes correct size
//
// TRANSIENT: klass_word == 0; size is indeterminate until we become an OBJECT
//
//
// STATE TRANSITION DIAGRAM
//
//        mut / parnew                     mut  /  parnew
// FREE --------------------> TRANSIENT ---------------------> OBJECT --|
//  ^                                                                   |
//  |------------------------ DEAD <------------------------------------|
//         sweep                            mut
//
// While a block is in TRANSIENT state its size cannot be determined
// so readers will either need to come back later or stall until
// the size can be determined. Note that for the case of direct
// allocation, P-bits, when available, may be used to determine the
// size of an object that may not yet have been initialized.

D
duke 已提交
1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385
// Things to support parallel young-gen collection.
oop
ConcurrentMarkSweepGeneration::par_promote(int thread_num,
                                           oop old, markOop m,
                                           size_t word_sz) {
#ifndef PRODUCT
  if (Universe::heap()->promotion_should_fail()) {
    return NULL;
  }
#endif  // #ifndef PRODUCT

  CMSParGCThreadState* ps = _par_gc_thread_states[thread_num];
  PromotionInfo* promoInfo = &ps->promo;
  // if we are tracking promotions, then first ensure space for
  // promotion (including spooling space for saving header if necessary).
  // then allocate and copy, then track promoted info if needed.
  // When tracking (see PromotionInfo::track()), the mark word may
  // be displaced and in this case restoration of the mark word
  // occurs in the (oop_since_save_marks_)iterate phase.
  if (promoInfo->tracking() && !promoInfo->ensure_spooling_space()) {
    // Out of space for allocating spooling buffers;
    // try expanding and allocating spooling buffers.
    if (!expand_and_ensure_spooling_space(promoInfo)) {
      return NULL;
    }
  }
  assert(promoInfo->has_spooling_space(), "Control point invariant");
1386 1387
  const size_t alloc_sz = CompactibleFreeListSpace::adjustObjectSize(word_sz);
  HeapWord* obj_ptr = ps->lab.alloc(alloc_sz);
D
duke 已提交
1388
  if (obj_ptr == NULL) {
1389
     obj_ptr = expand_and_par_lab_allocate(ps, alloc_sz);
D
duke 已提交
1390 1391 1392 1393 1394
     if (obj_ptr == NULL) {
       return NULL;
     }
  }
  oop obj = oop(obj_ptr);
1395
  OrderAccess::storestore();
1396
  assert(obj->klass_or_null() == NULL, "Object should be uninitialized here.");
1397
  assert(!((FreeChunk*)obj_ptr)->is_free(), "Error, block will look free but show wrong size");
1398
  // IMPORTANT: See note on object initialization for CMS above.
D
duke 已提交
1399 1400
  // Otherwise, copy the object.  Here we must be careful to insert the
  // klass pointer last, since this marks the block as an allocated object.
1401
  // Except with compressed oops it's the mark word.
D
duke 已提交
1402
  HeapWord* old_ptr = (HeapWord*)old;
1403 1404 1405
  // Restore the mark word copied above.
  obj->set_mark(m);
  assert(obj->klass_or_null() == NULL, "Object should be uninitialized here.");
1406
  assert(!((FreeChunk*)obj_ptr)->is_free(), "Error, block will look free but show wrong size");
1407 1408
  OrderAccess::storestore();

1409
  if (UseCompressedKlassPointers) {
1410 1411 1412
    // Copy gap missed by (aligned) header size calculation below
    obj->set_klass_gap(old->klass_gap());
  }
D
duke 已提交
1413 1414 1415 1416 1417
  if (word_sz > (size_t)oopDesc::header_size()) {
    Copy::aligned_disjoint_words(old_ptr + oopDesc::header_size(),
                                 obj_ptr + oopDesc::header_size(),
                                 word_sz - oopDesc::header_size());
  }
1418

D
duke 已提交
1419
  // Now we can track the promoted object, if necessary.  We take care
1420
  // to delay the transition from uninitialized to full object
D
duke 已提交
1421 1422 1423 1424 1425
  // (i.e., insertion of klass pointer) until after, so that it
  // atomically becomes a promoted object.
  if (promoInfo->tracking()) {
    promoInfo->track((PromotedObject*)obj, old->klass());
  }
1426
  assert(obj->klass_or_null() == NULL, "Object should be uninitialized here.");
1427
  assert(!((FreeChunk*)obj_ptr)->is_free(), "Error, block will look free but show wrong size");
1428
  assert(old->is_oop(), "Will use and dereference old klass ptr below");
1429 1430

  // Finally, install the klass pointer (this should be volatile).
1431
  OrderAccess::storestore();
D
duke 已提交
1432
  obj->set_klass(old->klass());
1433 1434
  // We should now be able to calculate the right size for this object
  assert(obj->is_oop() && obj->size() == (int)word_sz, "Error, incorrect size computed for promoted object");
D
duke 已提交
1435 1436 1437 1438 1439

  collector()->promoted(true,          // parallel
                        obj_ptr, old->is_objArray(), word_sz);

  NOT_PRODUCT(
1440 1441
    Atomic::inc_ptr(&_numObjectsPromoted);
    Atomic::add_ptr(alloc_sz, &_numWordsPromoted);
D
duke 已提交
1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458
  )

  return obj;
}

void
ConcurrentMarkSweepGeneration::
par_promote_alloc_undo(int thread_num,
                       HeapWord* obj, size_t word_sz) {
  // CMS does not support promotion undo.
  ShouldNotReachHere();
}

void
ConcurrentMarkSweepGeneration::
par_promote_alloc_done(int thread_num) {
  CMSParGCThreadState* ps = _par_gc_thread_states[thread_num];
1459
  ps->lab.retire(thread_num);
D
duke 已提交
1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484
}

void
ConcurrentMarkSweepGeneration::
par_oop_since_save_marks_iterate_done(int thread_num) {
  CMSParGCThreadState* ps = _par_gc_thread_states[thread_num];
  ParScanWithoutBarrierClosure* dummy_cl = NULL;
  ps->promo.promoted_oops_iterate_nv(dummy_cl);
}

bool ConcurrentMarkSweepGeneration::should_collect(bool   full,
                                                   size_t size,
                                                   bool   tlab)
{
  // We allow a STW collection only if a full
  // collection was requested.
  return full || should_allocate(size, tlab); // FIX ME !!!
  // This and promotion failure handling are connected at the
  // hip and should be fixed by untying them.
}

bool CMSCollector::shouldConcurrentCollect() {
  if (_full_gc_requested) {
    if (Verbose && PrintGCDetails) {
      gclog_or_tty->print_cr("CMSCollector: collect because of explicit "
1485
                             " gc request (or gc_locker)");
D
duke 已提交
1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520
    }
    return true;
  }

  // For debugging purposes, change the type of collection.
  // If the rotation is not on the concurrent collection
  // type, don't start a concurrent collection.
  NOT_PRODUCT(
    if (RotateCMSCollectionTypes &&
        (_cmsGen->debug_collection_type() !=
          ConcurrentMarkSweepGeneration::Concurrent_collection_type)) {
      assert(_cmsGen->debug_collection_type() !=
        ConcurrentMarkSweepGeneration::Unknown_collection_type,
        "Bad cms collection type");
      return false;
    }
  )

  FreelistLocker x(this);
  // ------------------------------------------------------------------
  // Print out lots of information which affects the initiation of
  // a collection.
  if (PrintCMSInitiationStatistics && stats().valid()) {
    gclog_or_tty->print("CMSCollector shouldConcurrentCollect: ");
    gclog_or_tty->stamp();
    gclog_or_tty->print_cr("");
    stats().print_on(gclog_or_tty);
    gclog_or_tty->print_cr("time_until_cms_gen_full %3.7f",
      stats().time_until_cms_gen_full());
    gclog_or_tty->print_cr("free="SIZE_FORMAT, _cmsGen->free());
    gclog_or_tty->print_cr("contiguous_available="SIZE_FORMAT,
                           _cmsGen->contiguous_available());
    gclog_or_tty->print_cr("promotion_rate=%g", stats().promotion_rate());
    gclog_or_tty->print_cr("cms_allocation_rate=%g", stats().cms_allocation_rate());
    gclog_or_tty->print_cr("occupancy=%3.7f", _cmsGen->occupancy());
1521
    gclog_or_tty->print_cr("initiatingOccupancy=%3.7f", _cmsGen->initiating_occupancy());
1522 1523
    gclog_or_tty->print_cr("metadata initialized %d",
      MetaspaceGC::should_concurrent_collect());
D
duke 已提交
1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551
  }
  // ------------------------------------------------------------------

  // If the estimated time to complete a cms collection (cms_duration())
  // is less than the estimated time remaining until the cms generation
  // is full, start a collection.
  if (!UseCMSInitiatingOccupancyOnly) {
    if (stats().valid()) {
      if (stats().time_until_cms_start() == 0.0) {
        return true;
      }
    } else {
      // We want to conservatively collect somewhat early in order
      // to try and "bootstrap" our CMS/promotion statistics;
      // this branch will not fire after the first successful CMS
      // collection because the stats should then be valid.
      if (_cmsGen->occupancy() >= _bootstrap_occupancy) {
        if (Verbose && PrintGCDetails) {
          gclog_or_tty->print_cr(
            " CMSCollector: collect for bootstrapping statistics:"
            " occupancy = %f, boot occupancy = %f", _cmsGen->occupancy(),
            _bootstrap_occupancy);
        }
        return true;
      }
    }
  }

1552
  // Otherwise, we start a collection cycle if
D
duke 已提交
1553 1554 1555
  // old gen want a collection cycle started. Each may use
  // an appropriate criterion for making this decision.
  // XXX We need to make sure that the gen expansion
1556 1557
  // criterion dovetails well with this. XXX NEED TO FIX THIS
  if (_cmsGen->should_concurrent_collect()) {
D
duke 已提交
1558 1559 1560 1561 1562 1563
    if (Verbose && PrintGCDetails) {
      gclog_or_tty->print_cr("CMS old gen initiated");
    }
    return true;
  }

1564 1565 1566 1567 1568 1569
  // We start a collection if we believe an incremental collection may fail;
  // this is not likely to be productive in practice because it's probably too
  // late anyway.
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  assert(gch->collector_policy()->is_two_generation_policy(),
         "You may want to check the correctness of the following");
1570 1571
  if (gch->incremental_collection_will_fail(true /* consult_young */)) {
    if (Verbose && PrintGCDetails) {
1572
      gclog_or_tty->print("CMSCollector: collect because incremental collection will fail ");
D
duke 已提交
1573 1574 1575 1576
    }
    return true;
  }

1577
  if (MetaspaceGC::should_concurrent_collect()) {
1578
      if (Verbose && PrintGCDetails) {
1579
      gclog_or_tty->print("CMSCollector: collect for metadata allocation ");
1580 1581 1582
      }
      return true;
    }
1583

D
duke 已提交
1584 1585 1586
  return false;
}

1587 1588
void CMSCollector::set_did_compact(bool v) { _cmsGen->set_did_compact(v); }

D
duke 已提交
1589 1590 1591 1592 1593
// Clear _expansion_cause fields of constituent generations
void CMSCollector::clear_expansion_cause() {
  _cmsGen->clear_expansion_cause();
}

1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
// We should be conservative in starting a collection cycle.  To
// start too eagerly runs the risk of collecting too often in the
// extreme.  To collect too rarely falls back on full collections,
// which works, even if not optimum in terms of concurrent work.
// As a work around for too eagerly collecting, use the flag
// UseCMSInitiatingOccupancyOnly.  This also has the advantage of
// giving the user an easily understandable way of controlling the
// collections.
// We want to start a new collection cycle if any of the following
// conditions hold:
// . our current occupancy exceeds the configured initiating occupancy
//   for this generation, or
// . we recently needed to expand this space and have not, since that
//   expansion, done a collection of this generation, or
// . the underlying space believes that it may be a good idea to initiate
//   a concurrent collection (this may be based on criteria such as the
//   following: the space uses linear allocation and linear allocation is
//   going to fail, or there is believed to be excessive fragmentation in
//   the generation, etc... or ...
// [.(currently done by CMSCollector::shouldConcurrentCollect() only for
1614
//   the case of the old generation; see CR 6543076):
1615 1616 1617
//   we may be approaching a point at which allocation requests may fail because
//   we will be out of sufficient free space given allocation rate estimates.]
bool ConcurrentMarkSweepGeneration::should_concurrent_collect() const {
D
duke 已提交
1618

1619 1620
  assert_lock_strong(freelistLock());
  if (occupancy() > initiating_occupancy()) {
D
duke 已提交
1621 1622
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print(" %s: collect because of occupancy %f / %f  ",
1623
        short_name(), occupancy(), initiating_occupancy());
D
duke 已提交
1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636
    }
    return true;
  }
  if (UseCMSInitiatingOccupancyOnly) {
    return false;
  }
  if (expansion_cause() == CMSExpansionCause::_satisfy_allocation) {
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print(" %s: collect because expanded for allocation ",
        short_name());
    }
    return true;
  }
1637
  if (_cmsSpace->should_concurrent_collect()) {
D
duke 已提交
1638
    if (PrintGCDetails && Verbose) {
1639
      gclog_or_tty->print(" %s: collect because cmsSpace says so ",
D
duke 已提交
1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687
        short_name());
    }
    return true;
  }
  return false;
}

void ConcurrentMarkSweepGeneration::collect(bool   full,
                                            bool   clear_all_soft_refs,
                                            size_t size,
                                            bool   tlab)
{
  collector()->collect(full, clear_all_soft_refs, size, tlab);
}

void CMSCollector::collect(bool   full,
                           bool   clear_all_soft_refs,
                           size_t size,
                           bool   tlab)
{
  if (!UseCMSCollectionPassing && _collectorState > Idling) {
    // For debugging purposes skip the collection if the state
    // is not currently idle
    if (TraceCMSState) {
      gclog_or_tty->print_cr("Thread " INTPTR_FORMAT " skipped full:%d CMS state %d",
        Thread::current(), full, _collectorState);
    }
    return;
  }

  // The following "if" branch is present for defensive reasons.
  // In the current uses of this interface, it can be replaced with:
  // assert(!GC_locker.is_active(), "Can't be called otherwise");
  // But I am not placing that assert here to allow future
  // generality in invoking this interface.
  if (GC_locker::is_active()) {
    // A consistency test for GC_locker
    assert(GC_locker::needs_gc(), "Should have been set already");
    // Skip this foreground collection, instead
    // expanding the heap if necessary.
    // Need the free list locks for the call to free() in compute_new_size()
    compute_new_size();
    return;
  }
  acquire_control_and_collect(full, clear_all_soft_refs);
  _full_gcs_since_conc_gc++;
}

S
sla 已提交
1688
void CMSCollector::request_full_gc(unsigned int full_gc_count, GCCause::Cause cause) {
D
duke 已提交
1689 1690 1691 1692 1693
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  unsigned int gc_count = gch->total_full_collections();
  if (gc_count == full_gc_count) {
    MutexLockerEx y(CGC_lock, Mutex::_no_safepoint_check_flag);
    _full_gc_requested = true;
S
sla 已提交
1694
    _full_gc_cause = cause;
D
duke 已提交
1695
    CGC_lock->notify();   // nudge CMS thread
1696 1697
  } else {
    assert(gc_count > full_gc_count, "Error: causal loop");
D
duke 已提交
1698 1699 1700
  }
}

S
sla 已提交
1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719
bool CMSCollector::is_external_interruption() {
  GCCause::Cause cause = GenCollectedHeap::heap()->gc_cause();
  return GCCause::is_user_requested_gc(cause) ||
         GCCause::is_serviceability_requested_gc(cause);
}

void CMSCollector::report_concurrent_mode_interruption() {
  if (is_external_interruption()) {
    if (PrintGCDetails) {
      gclog_or_tty->print(" (concurrent mode interrupted)");
    }
  } else {
    if (PrintGCDetails) {
      gclog_or_tty->print(" (concurrent mode failure)");
    }
    _gc_tracer_cm->report_concurrent_mode_failure();
  }
}

D
duke 已提交
1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 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 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 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

// The foreground and background collectors need to coordinate in order
// to make sure that they do not mutually interfere with CMS collections.
// When a background collection is active,
// the foreground collector may need to take over (preempt) and
// synchronously complete an ongoing collection. Depending on the
// frequency of the background collections and the heap usage
// of the application, this preemption can be seldom or frequent.
// There are only certain
// points in the background collection that the "collection-baton"
// can be passed to the foreground collector.
//
// The foreground collector will wait for the baton before
// starting any part of the collection.  The foreground collector
// will only wait at one location.
//
// The background collector will yield the baton before starting a new
// phase of the collection (e.g., before initial marking, marking from roots,
// precleaning, final re-mark, sweep etc.)  This is normally done at the head
// of the loop which switches the phases. The background collector does some
// of the phases (initial mark, final re-mark) with the world stopped.
// Because of locking involved in stopping the world,
// the foreground collector should not block waiting for the background
// collector when it is doing a stop-the-world phase.  The background
// collector will yield the baton at an additional point just before
// it enters a stop-the-world phase.  Once the world is stopped, the
// background collector checks the phase of the collection.  If the
// phase has not changed, it proceeds with the collection.  If the
// phase has changed, it skips that phase of the collection.  See
// the comments on the use of the Heap_lock in collect_in_background().
//
// Variable used in baton passing.
//   _foregroundGCIsActive - Set to true by the foreground collector when
//      it wants the baton.  The foreground clears it when it has finished
//      the collection.
//   _foregroundGCShouldWait - Set to true by the background collector
//        when it is running.  The foreground collector waits while
//      _foregroundGCShouldWait is true.
//  CGC_lock - monitor used to protect access to the above variables
//      and to notify the foreground and background collectors.
//  _collectorState - current state of the CMS collection.
//
// The foreground collector
//   acquires the CGC_lock
//   sets _foregroundGCIsActive
//   waits on the CGC_lock for _foregroundGCShouldWait to be false
//     various locks acquired in preparation for the collection
//     are released so as not to block the background collector
//     that is in the midst of a collection
//   proceeds with the collection
//   clears _foregroundGCIsActive
//   returns
//
// The background collector in a loop iterating on the phases of the
//      collection
//   acquires the CGC_lock
//   sets _foregroundGCShouldWait
//   if _foregroundGCIsActive is set
//     clears _foregroundGCShouldWait, notifies _CGC_lock
//     waits on _CGC_lock for _foregroundGCIsActive to become false
//     and exits the loop.
//   otherwise
//     proceed with that phase of the collection
//     if the phase is a stop-the-world phase,
//       yield the baton once more just before enqueueing
//       the stop-world CMS operation (executed by the VM thread).
//   returns after all phases of the collection are done
//

void CMSCollector::acquire_control_and_collect(bool full,
        bool clear_all_soft_refs) {
  assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint");
  assert(!Thread::current()->is_ConcurrentGC_thread(),
         "shouldn't try to acquire control from self!");

  // Start the protocol for acquiring control of the
  // collection from the background collector (aka CMS thread).
  assert(ConcurrentMarkSweepThread::vm_thread_has_cms_token(),
         "VM thread should have CMS token");
  // Remember the possibly interrupted state of an ongoing
  // concurrent collection
  CollectorState first_state = _collectorState;

  // Signal to a possibly ongoing concurrent collection that
  // we want to do a foreground collection.
  _foregroundGCIsActive = true;

  // Disable incremental mode during a foreground collection.
  ICMSDisabler icms_disabler;

  // release locks and wait for a notify from the background collector
  // releasing the locks in only necessary for phases which
  // do yields to improve the granularity of the collection.
  assert_lock_strong(bitMapLock());
  // We need to lock the Free list lock for the space that we are
  // currently collecting.
  assert(haveFreelistLocks(), "Must be holding free list locks");
  bitMapLock()->unlock();
  releaseFreelistLocks();
  {
    MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
    if (_foregroundGCShouldWait) {
      // We are going to be waiting for action for the CMS thread;
      // it had better not be gone (for instance at shutdown)!
      assert(ConcurrentMarkSweepThread::cmst() != NULL,
             "CMS thread must be running");
      // Wait here until the background collector gives us the go-ahead
      ConcurrentMarkSweepThread::clear_CMS_flag(
        ConcurrentMarkSweepThread::CMS_vm_has_token);  // release token
      // Get a possibly blocked CMS thread going:
      //   Note that we set _foregroundGCIsActive true above,
      //   without protection of the CGC_lock.
      CGC_lock->notify();
      assert(!ConcurrentMarkSweepThread::vm_thread_wants_cms_token(),
             "Possible deadlock");
      while (_foregroundGCShouldWait) {
        // wait for notification
        CGC_lock->wait(Mutex::_no_safepoint_check_flag);
        // Possibility of delay/starvation here, since CMS token does
        // not know to give priority to VM thread? Actually, i think
        // there wouldn't be any delay/starvation, but the proof of
        // that "fact" (?) appears non-trivial. XXX 20011219YSR
      }
      ConcurrentMarkSweepThread::set_CMS_flag(
        ConcurrentMarkSweepThread::CMS_vm_has_token);
    }
  }
  // The CMS_token is already held.  Get back the other locks.
  assert(ConcurrentMarkSweepThread::vm_thread_has_cms_token(),
         "VM thread should have CMS token");
  getFreelistLocks();
  bitMapLock()->lock_without_safepoint_check();
  if (TraceCMSState) {
    gclog_or_tty->print_cr("CMS foreground collector has asked for control "
      INTPTR_FORMAT " with first state %d", Thread::current(), first_state);
    gclog_or_tty->print_cr("    gets control with state %d", _collectorState);
  }

  // Check if we need to do a compaction, or if not, whether
  // we need to start the mark-sweep from scratch.
  bool should_compact    = false;
  bool should_start_over = false;
  decide_foreground_collection_type(clear_all_soft_refs,
    &should_compact, &should_start_over);

NOT_PRODUCT(
  if (RotateCMSCollectionTypes) {
    if (_cmsGen->debug_collection_type() ==
        ConcurrentMarkSweepGeneration::MSC_foreground_collection_type) {
      should_compact = true;
    } else if (_cmsGen->debug_collection_type() ==
               ConcurrentMarkSweepGeneration::MS_foreground_collection_type) {
      should_compact = false;
    }
  }
)

S
sla 已提交
1877 1878
  if (first_state > Idling) {
    report_concurrent_mode_interruption();
D
duke 已提交
1879 1880
  }

1881
  set_did_compact(should_compact);
D
duke 已提交
1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893
  if (should_compact) {
    // If the collection is being acquired from the background
    // collector, there may be references on the discovered
    // references lists that have NULL referents (being those
    // that were concurrently cleared by a mutator) or
    // that are no longer active (having been enqueued concurrently
    // by the mutator).
    // Scrub the list of those references because Mark-Sweep-Compact
    // code assumes referents are not NULL and that all discovered
    // Reference objects are active.
    ref_processor()->clean_up_discovered_references();

S
sla 已提交
1894 1895 1896 1897
    if (first_state > Idling) {
      save_heap_summary();
    }

D
duke 已提交
1898 1899 1900
    do_compaction_work(clear_all_soft_refs);

    // Has the GC time limit been exceeded?
1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913
    DefNewGeneration* young_gen = _young_gen->as_DefNewGeneration();
    size_t max_eden_size = young_gen->max_capacity() -
                           young_gen->to()->capacity() -
                           young_gen->from()->capacity();
    GenCollectedHeap* gch = GenCollectedHeap::heap();
    GCCause::Cause gc_cause = gch->gc_cause();
    size_policy()->check_gc_overhead_limit(_young_gen->used(),
                                           young_gen->eden()->used(),
                                           _cmsGen->max_capacity(),
                                           max_eden_size,
                                           full,
                                           gc_cause,
                                           gch->collector_policy());
D
duke 已提交
1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924
  } else {
    do_mark_sweep_work(clear_all_soft_refs, first_state,
      should_start_over);
  }
  // Reset the expansion cause, now that we just completed
  // a collection cycle.
  clear_expansion_cause();
  _foregroundGCIsActive = false;
  return;
}

1925
// Resize the tenured generation
D
duke 已提交
1926 1927 1928 1929 1930
// after obtaining the free list locks for the
// two generations.
void CMSCollector::compute_new_size() {
  assert_locked_or_safepoint(Heap_lock);
  FreelistLocker z(this);
1931
  MetaspaceGC::compute_new_size();
1932
  _cmsGen->compute_new_size_free_list();
D
duke 已提交
1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953
}

// A work method used by foreground collection to determine
// what type of collection (compacting or not, continuing or fresh)
// it should do.
// NOTE: the intent is to make UseCMSCompactAtFullCollection
// and CMSCompactWhenClearAllSoftRefs the default in the future
// and do away with the flags after a suitable period.
void CMSCollector::decide_foreground_collection_type(
  bool clear_all_soft_refs, bool* should_compact,
  bool* should_start_over) {
  // Normally, we'll compact only if the UseCMSCompactAtFullCollection
  // flag is set, and we have either requested a System.gc() or
  // the number of full gc's since the last concurrent cycle
  // has exceeded the threshold set by CMSFullGCsBeforeCompaction,
  // or if an incremental collection has failed
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  assert(gch->collector_policy()->is_two_generation_policy(),
         "You may want to check the correctness of the following");
  // Inform cms gen if this was due to partial collection failing.
  // The CMS gen may use this fact to determine its expansion policy.
1954
  if (gch->incremental_collection_will_fail(false /* don't consult_young */)) {
D
duke 已提交
1955 1956 1957 1958 1959 1960 1961 1962
    assert(!_cmsGen->incremental_collection_failed(),
           "Should have been noticed, reacted to and cleared");
    _cmsGen->set_incremental_collection_failed();
  }
  *should_compact =
    UseCMSCompactAtFullCollection &&
    ((_full_gcs_since_conc_gc >= CMSFullGCsBeforeCompaction) ||
     GCCause::is_user_requested_gc(gch->gc_cause()) ||
1963
     gch->incremental_collection_will_fail(true /* consult_young */));
D
duke 已提交
1964 1965 1966 1967 1968 1969 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
  *should_start_over = false;
  if (clear_all_soft_refs && !*should_compact) {
    // We are about to do a last ditch collection attempt
    // so it would normally make sense to do a compaction
    // to reclaim as much space as possible.
    if (CMSCompactWhenClearAllSoftRefs) {
      // Default: The rationale is that in this case either
      // we are past the final marking phase, in which case
      // we'd have to start over, or so little has been done
      // that there's little point in saving that work. Compaction
      // appears to be the sensible choice in either case.
      *should_compact = true;
    } else {
      // We have been asked to clear all soft refs, but not to
      // compact. Make sure that we aren't past the final checkpoint
      // phase, for that is where we process soft refs. If we are already
      // past that phase, we'll need to redo the refs discovery phase and
      // if necessary clear soft refs that weren't previously
      // cleared. We do so by remembering the phase in which
      // we came in, and if we are past the refs processing
      // phase, we'll choose to just redo the mark-sweep
      // collection from scratch.
      if (_collectorState > FinalMarking) {
        // We are past the refs processing phase;
        // start over and do a fresh synchronous CMS cycle
        _collectorState = Resetting; // skip to reset to start new cycle
        reset(false /* == !asynch */);
        *should_start_over = true;
      } // else we can continue a possibly ongoing current cycle
    }
  }
}

// A work method used by the foreground collector to do
// a mark-sweep-compact.
void CMSCollector::do_compaction_work(bool clear_all_soft_refs) {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
S
sla 已提交
2001 2002 2003 2004 2005 2006 2007 2008

  STWGCTimer* gc_timer = GenMarkSweep::gc_timer();
  gc_timer->register_gc_start(os::elapsed_counter());

  SerialOldTracer* gc_tracer = GenMarkSweep::gc_tracer();
  gc_tracer->report_gc_start(gch->gc_cause(), gc_timer->gc_start());

  GCTraceTime t("CMS:MSC ", PrintGCDetails && Verbose, true, NULL);
D
duke 已提交
2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
  if (PrintGC && Verbose && !(GCCause::is_user_requested_gc(gch->gc_cause()))) {
    gclog_or_tty->print_cr("Compact ConcurrentMarkSweepGeneration after %d "
      "collections passed to foreground collector", _full_gcs_since_conc_gc);
  }

  // Sample collection interval time and reset for collection pause.
  if (UseAdaptiveSizePolicy) {
    size_policy()->msc_collection_begin();
  }

  // Temporarily widen the span of the weak reference processing to
  // the entire heap.
  MemRegion new_span(GenCollectedHeap::heap()->reserved_region());
2022
  ReferenceProcessorSpanMutator rp_mut_span(ref_processor(), new_span);
D
duke 已提交
2023 2024
  // Temporarily, clear the "is_alive_non_header" field of the
  // reference processor.
2025
  ReferenceProcessorIsAliveMutator rp_mut_closure(ref_processor(), NULL);
D
duke 已提交
2026
  // Temporarily make reference _processing_ single threaded (non-MT).
2027
  ReferenceProcessorMTProcMutator rp_mut_mt_processing(ref_processor(), false);
D
duke 已提交
2028
  // Temporarily make refs discovery atomic
2029 2030 2031
  ReferenceProcessorAtomicMutator rp_mut_atomic(ref_processor(), true);
  // Temporarily make reference _discovery_ single threaded (non-MT)
  ReferenceProcessorMTDiscoveryMutator rp_mut_discovery(ref_processor(), false);
D
duke 已提交
2032 2033

  ref_processor()->set_enqueuing_is_done(false);
2034
  ref_processor()->enable_discovery(false /*verify_disabled*/, false /*check_no_refs*/);
2035
  ref_processor()->setup_policy(clear_all_soft_refs);
D
duke 已提交
2036 2037 2038 2039 2040 2041
  // If an asynchronous collection finishes, the _modUnionTable is
  // all clear.  If we are assuming the collection from an asynchronous
  // collection, clear the _modUnionTable.
  assert(_collectorState != Idling || _modUnionTable.isAllClear(),
    "_modUnionTable should be clear if the baton was not passed");
  _modUnionTable.clear_all();
2042 2043 2044
  assert(_collectorState != Idling || _ct->klass_rem_set()->mod_union_is_clear(),
    "mod union for klasses should be clear if the baton was passed");
  _ct->klass_rem_set()->clear_mod_union();
D
duke 已提交
2045 2046 2047 2048

  // We must adjust the allocation statistics being maintained
  // in the free list space. We do so by reading and clearing
  // the sweep timer and updating the block flux rate estimates below.
2049 2050 2051 2052 2053 2054 2055 2056
  assert(!_intra_sweep_timer.is_active(), "_intra_sweep_timer should be inactive");
  if (_inter_sweep_timer.is_active()) {
    _inter_sweep_timer.stop();
    // Note that we do not use this sample to update the _inter_sweep_estimate.
    _cmsGen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()),
                                            _inter_sweep_estimate.padded_average(),
                                            _intra_sweep_estimate.padded_average());
  }
D
duke 已提交
2057 2058 2059 2060 2061 2062 2063 2064 2065 2066

  GenMarkSweep::invoke_at_safepoint(_cmsGen->level(),
    ref_processor(), clear_all_soft_refs);
  #ifdef ASSERT
    CompactibleFreeListSpace* cms_space = _cmsGen->cmsSpace();
    size_t free_size = cms_space->free();
    assert(free_size ==
           pointer_delta(cms_space->end(), cms_space->compaction_top())
           * HeapWordSize,
      "All the free space should be compacted into one chunk at top");
2067
    assert(cms_space->dictionary()->total_chunk_size(
D
duke 已提交
2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080
                                      debug_only(cms_space->freelistLock())) == 0 ||
           cms_space->totalSizeInIndexedFreeLists() == 0,
      "All the free space should be in a single chunk");
    size_t num = cms_space->totalCount();
    assert((free_size == 0 && num == 0) ||
           (free_size > 0  && (num == 1 || num == 2)),
         "There should be at most 2 free chunks after compaction");
  #endif // ASSERT
  _collectorState = Resetting;
  assert(_restart_addr == NULL,
         "Should have been NULL'd before baton was passed");
  reset(false /* == !asynch */);
  _cmsGen->reset_after_compaction();
2081
  _concurrent_cycles_since_last_unload = 0;
D
duke 已提交
2082 2083 2084 2085 2086 2087 2088

  // Clear any data recorded in the PLAB chunk arrays.
  if (_survivor_plab_array != NULL) {
    reset_survivor_plab_arrays();
  }

  // Adjust the per-size allocation stats for the next epoch.
2089 2090 2091 2092
  _cmsGen->cmsSpace()->endSweepFLCensus(sweep_count() /* fake */);
  // Restart the "inter sweep timer" for the next epoch.
  _inter_sweep_timer.reset();
  _inter_sweep_timer.start();
D
duke 已提交
2093 2094 2095 2096 2097 2098

  // Sample collection pause time and reset for collection interval.
  if (UseAdaptiveSizePolicy) {
    size_policy()->msc_collection_end(gch->gc_cause());
  }

S
sla 已提交
2099 2100 2101 2102
  gc_timer->register_gc_end(os::elapsed_counter());

  gc_tracer->report_gc_end(gc_timer->gc_end(), gc_timer->time_partitions());

D
duke 已提交
2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133
  // For a mark-sweep-compact, compute_new_size() will be called
  // in the heap's do_collection() method.
}

// A work method used by the foreground collector to do
// a mark-sweep, after taking over from a possibly on-going
// concurrent mark-sweep collection.
void CMSCollector::do_mark_sweep_work(bool clear_all_soft_refs,
  CollectorState first_state, bool should_start_over) {
  if (PrintGC && Verbose) {
    gclog_or_tty->print_cr("Pass concurrent collection to foreground "
      "collector with count %d",
      _full_gcs_since_conc_gc);
  }
  switch (_collectorState) {
    case Idling:
      if (first_state == Idling || should_start_over) {
        // The background GC was not active, or should
        // restarted from scratch;  start the cycle.
        _collectorState = InitialMarking;
      }
      // If first_state was not Idling, then a background GC
      // was in progress and has now finished.  No need to do it
      // again.  Leave the state as Idling.
      break;
    case Precleaning:
      // In the foreground case don't do the precleaning since
      // it is not done concurrently and there is extra work
      // required.
      _collectorState = FinalMarking;
  }
S
sla 已提交
2134
  collect_in_foreground(clear_all_soft_refs, GenCollectedHeap::heap()->gc_cause());
D
duke 已提交
2135 2136 2137 2138 2139 2140

  // For a mark-sweep, compute_new_size() will be called
  // in the heap's do_collection() method.
}


2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173
void CMSCollector::print_eden_and_survivor_chunk_arrays() {
  DefNewGeneration* dng = _young_gen->as_DefNewGeneration();
  EdenSpace* eden_space = dng->eden();
  ContiguousSpace* from_space = dng->from();
  ContiguousSpace* to_space   = dng->to();
  // Eden
  if (_eden_chunk_array != NULL) {
    gclog_or_tty->print_cr("eden " PTR_FORMAT "-" PTR_FORMAT "-" PTR_FORMAT "(" SIZE_FORMAT ")",
                           eden_space->bottom(), eden_space->top(),
                           eden_space->end(), eden_space->capacity());
    gclog_or_tty->print_cr("_eden_chunk_index=" SIZE_FORMAT ", "
                           "_eden_chunk_capacity=" SIZE_FORMAT,
                           _eden_chunk_index, _eden_chunk_capacity);
    for (size_t i = 0; i < _eden_chunk_index; i++) {
      gclog_or_tty->print_cr("_eden_chunk_array[" SIZE_FORMAT "]=" PTR_FORMAT,
                             i, _eden_chunk_array[i]);
    }
  }
  // Survivor
  if (_survivor_chunk_array != NULL) {
    gclog_or_tty->print_cr("survivor " PTR_FORMAT "-" PTR_FORMAT "-" PTR_FORMAT "(" SIZE_FORMAT ")",
                           from_space->bottom(), from_space->top(),
                           from_space->end(), from_space->capacity());
    gclog_or_tty->print_cr("_survivor_chunk_index=" SIZE_FORMAT ", "
                           "_survivor_chunk_capacity=" SIZE_FORMAT,
                           _survivor_chunk_index, _survivor_chunk_capacity);
    for (size_t i = 0; i < _survivor_chunk_index; i++) {
      gclog_or_tty->print_cr("_survivor_chunk_array[" SIZE_FORMAT "]=" PTR_FORMAT,
                             i, _survivor_chunk_array[i]);
    }
  }
}

D
duke 已提交
2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226
void CMSCollector::getFreelistLocks() const {
  // Get locks for all free lists in all generations that this
  // collector is responsible for
  _cmsGen->freelistLock()->lock_without_safepoint_check();
}

void CMSCollector::releaseFreelistLocks() const {
  // Release locks for all free lists in all generations that this
  // collector is responsible for
  _cmsGen->freelistLock()->unlock();
}

bool CMSCollector::haveFreelistLocks() const {
  // Check locks for all free lists in all generations that this
  // collector is responsible for
  assert_lock_strong(_cmsGen->freelistLock());
  PRODUCT_ONLY(ShouldNotReachHere());
  return true;
}

// A utility class that is used by the CMS collector to
// temporarily "release" the foreground collector from its
// usual obligation to wait for the background collector to
// complete an ongoing phase before proceeding.
class ReleaseForegroundGC: public StackObj {
 private:
  CMSCollector* _c;
 public:
  ReleaseForegroundGC(CMSCollector* c) : _c(c) {
    assert(_c->_foregroundGCShouldWait, "Else should not need to call");
    MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
    // allow a potentially blocked foreground collector to proceed
    _c->_foregroundGCShouldWait = false;
    if (_c->_foregroundGCIsActive) {
      CGC_lock->notify();
    }
    assert(!ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
           "Possible deadlock");
  }

  ~ReleaseForegroundGC() {
    assert(!_c->_foregroundGCShouldWait, "Usage protocol violation?");
    MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
    _c->_foregroundGCShouldWait = true;
  }
};

// There are separate collect_in_background and collect_in_foreground because of
// the different locking requirements of the background collector and the
// foreground collector.  There was originally an attempt to share
// one "collect" method between the background collector and the foreground
// collector but the if-then-else required made it cleaner to have
// separate methods.
S
sla 已提交
2227
void CMSCollector::collect_in_background(bool clear_all_soft_refs, GCCause::Cause cause) {
D
duke 已提交
2228 2229 2230 2231 2232 2233 2234
  assert(Thread::current()->is_ConcurrentGC_thread(),
    "A CMS asynchronous collection is only allowed on a CMS thread.");

  GenCollectedHeap* gch = GenCollectedHeap::heap();
  {
    bool safepoint_check = Mutex::_no_safepoint_check_flag;
    MutexLockerEx hl(Heap_lock, safepoint_check);
2235
    FreelistLocker fll(this);
D
duke 已提交
2236 2237 2238 2239 2240 2241 2242 2243 2244 2245
    MutexLockerEx x(CGC_lock, safepoint_check);
    if (_foregroundGCIsActive || !UseAsyncConcMarkSweepGC) {
      // The foreground collector is active or we're
      // not using asynchronous collections.  Skip this
      // background collection.
      assert(!_foregroundGCShouldWait, "Should be clear");
      return;
    } else {
      assert(_collectorState == Idling, "Should be idling before start.");
      _collectorState = InitialMarking;
S
sla 已提交
2246
      register_gc_start(cause);
D
duke 已提交
2247 2248 2249
      // Reset the expansion cause, now that we are about to begin
      // a new cycle.
      clear_expansion_cause();
2250 2251 2252 2253

      // Clear the MetaspaceGC flag since a concurrent collection
      // is starting but also clear it after the collection.
      MetaspaceGC::set_should_concurrent_collect(false);
D
duke 已提交
2254
    }
2255 2256 2257
    // Decide if we want to enable class unloading as part of the
    // ensuing concurrent GC cycle.
    update_should_unload_classes();
D
duke 已提交
2258
    _full_gc_requested = false;           // acks all outstanding full gc requests
S
sla 已提交
2259
    _full_gc_cause = GCCause::_no_gc;
D
duke 已提交
2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 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 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 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 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384
    // Signal that we are about to start a collection
    gch->increment_total_full_collections();  // ... starting a collection cycle
    _collection_count_start = gch->total_full_collections();
  }

  // Used for PrintGC
  size_t prev_used;
  if (PrintGC && Verbose) {
    prev_used = _cmsGen->used(); // XXXPERM
  }

  // The change of the collection state is normally done at this level;
  // the exceptions are phases that are executed while the world is
  // stopped.  For those phases the change of state is done while the
  // world is stopped.  For baton passing purposes this allows the
  // background collector to finish the phase and change state atomically.
  // The foreground collector cannot wait on a phase that is done
  // while the world is stopped because the foreground collector already
  // has the world stopped and would deadlock.
  while (_collectorState != Idling) {
    if (TraceCMSState) {
      gclog_or_tty->print_cr("Thread " INTPTR_FORMAT " in CMS state %d",
        Thread::current(), _collectorState);
    }
    // The foreground collector
    //   holds the Heap_lock throughout its collection.
    //   holds the CMS token (but not the lock)
    //     except while it is waiting for the background collector to yield.
    //
    // The foreground collector should be blocked (not for long)
    //   if the background collector is about to start a phase
    //   executed with world stopped.  If the background
    //   collector has already started such a phase, the
    //   foreground collector is blocked waiting for the
    //   Heap_lock.  The stop-world phases (InitialMarking and FinalMarking)
    //   are executed in the VM thread.
    //
    // The locking order is
    //   PendingListLock (PLL)  -- if applicable (FinalMarking)
    //   Heap_lock  (both this & PLL locked in VM_CMS_Operation::prologue())
    //   CMS token  (claimed in
    //                stop_world_and_do() -->
    //                  safepoint_synchronize() -->
    //                    CMSThread::synchronize())

    {
      // Check if the FG collector wants us to yield.
      CMSTokenSync x(true); // is cms thread
      if (waitForForegroundGC()) {
        // We yielded to a foreground GC, nothing more to be
        // done this round.
        assert(_foregroundGCShouldWait == false, "We set it to false in "
               "waitForForegroundGC()");
        if (TraceCMSState) {
          gclog_or_tty->print_cr("CMS Thread " INTPTR_FORMAT
            " exiting collection CMS state %d",
            Thread::current(), _collectorState);
        }
        return;
      } else {
        // The background collector can run but check to see if the
        // foreground collector has done a collection while the
        // background collector was waiting to get the CGC_lock
        // above.  If yes, break so that _foregroundGCShouldWait
        // is cleared before returning.
        if (_collectorState == Idling) {
          break;
        }
      }
    }

    assert(_foregroundGCShouldWait, "Foreground collector, if active, "
      "should be waiting");

    switch (_collectorState) {
      case InitialMarking:
        {
          ReleaseForegroundGC x(this);
          stats().record_cms_begin();
          VM_CMS_Initial_Mark initial_mark_op(this);
          VMThread::execute(&initial_mark_op);
        }
        // The collector state may be any legal state at this point
        // since the background collector may have yielded to the
        // foreground collector.
        break;
      case Marking:
        // initial marking in checkpointRootsInitialWork has been completed
        if (markFromRoots(true)) { // we were successful
          assert(_collectorState == Precleaning, "Collector state should "
            "have changed");
        } else {
          assert(_foregroundGCIsActive, "Internal state inconsistency");
        }
        break;
      case Precleaning:
        if (UseAdaptiveSizePolicy) {
          size_policy()->concurrent_precleaning_begin();
        }
        // marking from roots in markFromRoots has been completed
        preclean();
        if (UseAdaptiveSizePolicy) {
          size_policy()->concurrent_precleaning_end();
        }
        assert(_collectorState == AbortablePreclean ||
               _collectorState == FinalMarking,
               "Collector state should have changed");
        break;
      case AbortablePreclean:
        if (UseAdaptiveSizePolicy) {
        size_policy()->concurrent_phases_resume();
        }
        abortable_preclean();
        if (UseAdaptiveSizePolicy) {
          size_policy()->concurrent_precleaning_end();
        }
        assert(_collectorState == FinalMarking, "Collector state should "
          "have changed");
        break;
      case FinalMarking:
        {
          ReleaseForegroundGC x(this);

          VM_CMS_Final_Remark final_remark_op(this);
          VMThread::execute(&final_remark_op);
2385
        }
D
duke 已提交
2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417
        assert(_foregroundGCShouldWait, "block post-condition");
        break;
      case Sweeping:
        if (UseAdaptiveSizePolicy) {
          size_policy()->concurrent_sweeping_begin();
        }
        // final marking in checkpointRootsFinal has been completed
        sweep(true);
        assert(_collectorState == Resizing, "Collector state change "
          "to Resizing must be done under the free_list_lock");
        _full_gcs_since_conc_gc = 0;

        // Stop the timers for adaptive size policy for the concurrent phases
        if (UseAdaptiveSizePolicy) {
          size_policy()->concurrent_sweeping_end();
          size_policy()->concurrent_phases_end(gch->gc_cause(),
                                             gch->prev_gen(_cmsGen)->capacity(),
                                             _cmsGen->free());
        }

      case Resizing: {
        // Sweeping has been completed...
        // At this point the background collection has completed.
        // Don't move the call to compute_new_size() down
        // into code that might be executed if the background
        // collection was preempted.
        {
          ReleaseForegroundGC x(this);   // unblock FG collection
          MutexLockerEx       y(Heap_lock, Mutex::_no_safepoint_check_flag);
          CMSTokenSync        z(true);   // not strictly needed.
          if (_collectorState == Resizing) {
            compute_new_size();
S
sla 已提交
2418
            save_heap_summary();
D
duke 已提交
2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431
            _collectorState = Resetting;
          } else {
            assert(_collectorState == Idling, "The state should only change"
                   " because the foreground collector has finished the collection");
          }
        }
        break;
      }
      case Resetting:
        // CMS heap resizing has been completed
        reset(true);
        assert(_collectorState == Idling, "Collector state should "
          "have changed");
2432 2433 2434

        MetaspaceGC::set_should_concurrent_collect(false);

D
duke 已提交
2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476
        stats().record_cms_end();
        // Don't move the concurrent_phases_end() and compute_new_size()
        // calls to here because a preempted background collection
        // has it's state set to "Resetting".
        break;
      case Idling:
      default:
        ShouldNotReachHere();
        break;
    }
    if (TraceCMSState) {
      gclog_or_tty->print_cr("  Thread " INTPTR_FORMAT " done - next CMS state %d",
        Thread::current(), _collectorState);
    }
    assert(_foregroundGCShouldWait, "block post-condition");
  }

  // Should this be in gc_epilogue?
  collector_policy()->counters()->update_counters();

  {
    // Clear _foregroundGCShouldWait and, in the event that the
    // foreground collector is waiting, notify it, before
    // returning.
    MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
    _foregroundGCShouldWait = false;
    if (_foregroundGCIsActive) {
      CGC_lock->notify();
    }
    assert(!ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
           "Possible deadlock");
  }
  if (TraceCMSState) {
    gclog_or_tty->print_cr("CMS Thread " INTPTR_FORMAT
      " exiting collection CMS state %d",
      Thread::current(), _collectorState);
  }
  if (PrintGC && Verbose) {
    _cmsGen->print_heap_change(prev_used);
  }
}

S
sla 已提交
2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509
void CMSCollector::register_foreground_gc_start(GCCause::Cause cause) {
  if (!_cms_start_registered) {
    register_gc_start(cause);
  }
}

void CMSCollector::register_gc_start(GCCause::Cause cause) {
  _cms_start_registered = true;
  _gc_timer_cm->register_gc_start(os::elapsed_counter());
  _gc_tracer_cm->report_gc_start(cause, _gc_timer_cm->gc_start());
}

void CMSCollector::register_gc_end() {
  if (_cms_start_registered) {
    report_heap_summary(GCWhen::AfterGC);

    _gc_timer_cm->register_gc_end(os::elapsed_counter());
    _gc_tracer_cm->report_gc_end(_gc_timer_cm->gc_end(), _gc_timer_cm->time_partitions());
    _cms_start_registered = false;
  }
}

void CMSCollector::save_heap_summary() {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  _last_heap_summary = gch->create_heap_summary();
  _last_metaspace_summary = gch->create_metaspace_summary();
}

void CMSCollector::report_heap_summary(GCWhen::Type when) {
  _gc_tracer_cm->report_gc_heap_summary(when, _last_heap_summary, _last_metaspace_summary);
}

void CMSCollector::collect_in_foreground(bool clear_all_soft_refs, GCCause::Cause cause) {
D
duke 已提交
2510 2511 2512 2513 2514 2515 2516
  assert(_foregroundGCIsActive && !_foregroundGCShouldWait,
         "Foreground collector should be waiting, not executing");
  assert(Thread::current()->is_VM_thread(), "A foreground collection"
    "may only be done by the VM Thread with the world stopped");
  assert(ConcurrentMarkSweepThread::vm_thread_has_cms_token(),
         "VM thread should have CMS token");

S
sla 已提交
2517 2518
  NOT_PRODUCT(GCTraceTime t("CMS:MS (foreground) ", PrintGCDetails && Verbose,
    true, NULL);)
D
duke 已提交
2519 2520 2521 2522 2523 2524 2525 2526 2527
  if (UseAdaptiveSizePolicy) {
    size_policy()->ms_collection_begin();
  }
  COMPILER2_PRESENT(DerivedPointerTableDeactivate dpt_deact);

  HandleMark hm;  // Discard invalid handles created during verification

  if (VerifyBeforeGC &&
      GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2528
    Universe::verify();
D
duke 已提交
2529 2530
  }

2531
  // Snapshot the soft reference policy to be used in this collection cycle.
2532
  ref_processor()->setup_policy(clear_all_soft_refs);
2533

D
duke 已提交
2534 2535 2536 2537 2538 2539 2540 2541
  bool init_mark_was_synchronous = false; // until proven otherwise
  while (_collectorState != Idling) {
    if (TraceCMSState) {
      gclog_or_tty->print_cr("Thread " INTPTR_FORMAT " in CMS state %d",
        Thread::current(), _collectorState);
    }
    switch (_collectorState) {
      case InitialMarking:
S
sla 已提交
2542
        register_foreground_gc_start(cause);
D
duke 已提交
2543 2544 2545 2546 2547 2548 2549 2550 2551
        init_mark_was_synchronous = true;  // fact to be exploited in re-mark
        checkpointRootsInitial(false);
        assert(_collectorState == Marking, "Collector state should have changed"
          " within checkpointRootsInitial()");
        break;
      case Marking:
        // initial marking in checkpointRootsInitialWork has been completed
        if (VerifyDuringGC &&
            GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2552
          Universe::verify("Verify before initial mark: ");
D
duke 已提交
2553 2554 2555 2556 2557 2558 2559 2560 2561 2562
        }
        {
          bool res = markFromRoots(false);
          assert(res && _collectorState == FinalMarking, "Collector state should "
            "have changed");
          break;
        }
      case FinalMarking:
        if (VerifyDuringGC &&
            GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2563
          Universe::verify("Verify before re-mark: ");
D
duke 已提交
2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
        }
        checkpointRootsFinal(false, clear_all_soft_refs,
                             init_mark_was_synchronous);
        assert(_collectorState == Sweeping, "Collector state should not "
          "have changed within checkpointRootsFinal()");
        break;
      case Sweeping:
        // final marking in checkpointRootsFinal has been completed
        if (VerifyDuringGC &&
            GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2574
          Universe::verify("Verify before sweep: ");
D
duke 已提交
2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588
        }
        sweep(false);
        assert(_collectorState == Resizing, "Incorrect state");
        break;
      case Resizing: {
        // Sweeping has been completed; the actual resize in this case
        // is done separately; nothing to be done in this state.
        _collectorState = Resetting;
        break;
      }
      case Resetting:
        // The heap has been resized.
        if (VerifyDuringGC &&
            GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2589
          Universe::verify("Verify before reset: ");
D
duke 已提交
2590
        }
S
sla 已提交
2591
        save_heap_summary();
D
duke 已提交
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
        reset(false);
        assert(_collectorState == Idling, "Collector state should "
          "have changed");
        break;
      case Precleaning:
      case AbortablePreclean:
        // Elide the preclean phase
        _collectorState = FinalMarking;
        break;
      default:
        ShouldNotReachHere();
    }
    if (TraceCMSState) {
      gclog_or_tty->print_cr("  Thread " INTPTR_FORMAT " done - next CMS state %d",
        Thread::current(), _collectorState);
    }
  }

  if (UseAdaptiveSizePolicy) {
    GenCollectedHeap* gch = GenCollectedHeap::heap();
    size_policy()->ms_collection_end(gch->gc_cause());
  }

  if (VerifyAfterGC &&
      GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
2617
    Universe::verify();
D
duke 已提交
2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680
  }
  if (TraceCMSState) {
    gclog_or_tty->print_cr("CMS Thread " INTPTR_FORMAT
      " exiting collection CMS state %d",
      Thread::current(), _collectorState);
  }
}

bool CMSCollector::waitForForegroundGC() {
  bool res = false;
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should have CMS token");
  // Block the foreground collector until the
  // background collectors decides whether to
  // yield.
  MutexLockerEx x(CGC_lock, Mutex::_no_safepoint_check_flag);
  _foregroundGCShouldWait = true;
  if (_foregroundGCIsActive) {
    // The background collector yields to the
    // foreground collector and returns a value
    // indicating that it has yielded.  The foreground
    // collector can proceed.
    res = true;
    _foregroundGCShouldWait = false;
    ConcurrentMarkSweepThread::clear_CMS_flag(
      ConcurrentMarkSweepThread::CMS_cms_has_token);
    ConcurrentMarkSweepThread::set_CMS_flag(
      ConcurrentMarkSweepThread::CMS_cms_wants_token);
    // Get a possibly blocked foreground thread going
    CGC_lock->notify();
    if (TraceCMSState) {
      gclog_or_tty->print_cr("CMS Thread " INTPTR_FORMAT " waiting at CMS state %d",
        Thread::current(), _collectorState);
    }
    while (_foregroundGCIsActive) {
      CGC_lock->wait(Mutex::_no_safepoint_check_flag);
    }
    ConcurrentMarkSweepThread::set_CMS_flag(
      ConcurrentMarkSweepThread::CMS_cms_has_token);
    ConcurrentMarkSweepThread::clear_CMS_flag(
      ConcurrentMarkSweepThread::CMS_cms_wants_token);
  }
  if (TraceCMSState) {
    gclog_or_tty->print_cr("CMS Thread " INTPTR_FORMAT " continuing at CMS state %d",
      Thread::current(), _collectorState);
  }
  return res;
}

// Because of the need to lock the free lists and other structures in
// the collector, common to all the generations that the collector is
// collecting, we need the gc_prologues of individual CMS generations
// delegate to their collector. It may have been simpler had the
// current infrastructure allowed one to call a prologue on a
// collector. In the absence of that we have the generation's
// prologue delegate to the collector, which delegates back
// some "local" work to a worker method in the individual generations
// that it's responsible for collecting, while itself doing any
// work common to all generations it's responsible for. A similar
// comment applies to the  gc_epilogue()'s.
// The role of the varaible _between_prologue_and_epilogue is to
// enforce the invocation protocol.
void CMSCollector::gc_prologue(bool full) {
2681
  // Call gc_prologue_work() for the CMSGen
D
duke 已提交
2682 2683 2684 2685 2686 2687 2688
  // we are responsible for.

  // The following locking discipline assumes that we are only called
  // when the world is stopped.
  assert(SafepointSynchronize::is_at_safepoint(), "world is stopped assumption");

  // The CMSCollector prologue must call the gc_prologues for the
2689
  // "generations" that it's responsible
D
duke 已提交
2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706
  // for.

  assert(   Thread::current()->is_VM_thread()
         || (   CMSScavengeBeforeRemark
             && Thread::current()->is_ConcurrentGC_thread()),
         "Incorrect thread type for prologue execution");

  if (_between_prologue_and_epilogue) {
    // We have already been invoked; this is a gc_prologue delegation
    // from yet another CMS generation that we are responsible for, just
    // ignore it since all relevant work has already been done.
    return;
  }

  // set a bit saying prologue has been called; cleared in epilogue
  _between_prologue_and_epilogue = true;
  // Claim locks for common data structures, then call gc_prologue_work()
2707
  // for each CMSGen.
D
duke 已提交
2708 2709 2710 2711 2712

  getFreelistLocks();   // gets free list locks on constituent spaces
  bitMapLock()->lock_without_safepoint_check();

  // Should call gc_prologue_work() for all cms gens we are responsible for
2713
  bool duringMarking =    _collectorState >= Marking
D
duke 已提交
2714
                         && _collectorState < Sweeping;
2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725

  // The young collections clear the modified oops state, which tells if
  // there are any modified oops in the class. The remark phase also needs
  // that information. Tell the young collection to save the union of all
  // modified klasses.
  if (duringMarking) {
    _ct->klass_rem_set()->set_accumulate_modified_oops(true);
  }

  bool registerClosure = duringMarking;

2726 2727
  ModUnionClosure* muc = CollectedHeap::use_parallel_gc_threads() ?
                                               &_modUnionClosurePar
D
duke 已提交
2728 2729 2730 2731 2732 2733 2734 2735 2736
                                               : &_modUnionClosure;
  _cmsGen->gc_prologue_work(full, registerClosure, muc);

  if (!full) {
    stats().record_gc0_begin();
  }
}

void ConcurrentMarkSweepGeneration::gc_prologue(bool full) {
2737 2738 2739 2740

  _capacity_at_prologue = capacity();
  _used_at_prologue = used();

D
duke 已提交
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
  // Delegate to CMScollector which knows how to coordinate between
  // this and any other CMS generations that it is responsible for
  // collecting.
  collector()->gc_prologue(full);
}

// This is a "private" interface for use by this generation's CMSCollector.
// Not to be called directly by any other entity (for instance,
// GenCollectedHeap, which calls the "public" gc_prologue method above).
void ConcurrentMarkSweepGeneration::gc_prologue_work(bool full,
  bool registerClosure, ModUnionClosure* modUnionClosure) {
  assert(!incremental_collection_failed(), "Shouldn't be set yet");
  assert(cmsSpace()->preconsumptionDirtyCardClosure() == NULL,
    "Should be NULL");
  if (registerClosure) {
    cmsSpace()->setPreconsumptionDirtyCardClosure(modUnionClosure);
  }
  cmsSpace()->gc_prologue();
  // Clear stat counters
  NOT_PRODUCT(
    assert(_numObjectsPromoted == 0, "check");
    assert(_numWordsPromoted   == 0, "check");
    if (Verbose && PrintGC) {
      gclog_or_tty->print("Allocated "SIZE_FORMAT" objects, "
                          SIZE_FORMAT" bytes concurrently",
      _numObjectsAllocated, _numWordsAllocated*sizeof(HeapWord));
    }
    _numObjectsAllocated = 0;
    _numWordsAllocated   = 0;
  )
}

void CMSCollector::gc_epilogue(bool full) {
  // The following locking discipline assumes that we are only called
  // when the world is stopped.
  assert(SafepointSynchronize::is_at_safepoint(),
         "world is stopped assumption");

  // Currently the CMS epilogue (see CompactibleFreeListSpace) merely checks
  // if linear allocation blocks need to be appropriately marked to allow the
  // the blocks to be parsable. We also check here whether we need to nudge the
  // CMS collector thread to start a new cycle (if it's not already active).
  assert(   Thread::current()->is_VM_thread()
         || (   CMSScavengeBeforeRemark
             && Thread::current()->is_ConcurrentGC_thread()),
         "Incorrect thread type for epilogue execution");

  if (!_between_prologue_and_epilogue) {
    // We have already been invoked; this is a gc_epilogue delegation
    // from yet another CMS generation that we are responsible for, just
    // ignore it since all relevant work has already been done.
    return;
  }
  assert(haveFreelistLocks(), "must have freelist locks");
  assert_lock_strong(bitMapLock());

2797 2798
  _ct->klass_rem_set()->set_accumulate_modified_oops(false);

D
duke 已提交
2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822
  _cmsGen->gc_epilogue_work(full);

  if (_collectorState == AbortablePreclean || _collectorState == Precleaning) {
    // in case sampling was not already enabled, enable it
    _start_sampling = true;
  }
  // reset _eden_chunk_array so sampling starts afresh
  _eden_chunk_index = 0;

  size_t cms_used   = _cmsGen->cmsSpace()->used();

  // update performance counters - this uses a special version of
  // update_counters() that allows the utilization to be passed as a
  // parameter, avoiding multiple calls to used().
  //
  _cmsGen->update_counters(cms_used);

  if (CMSIncrementalMode) {
    icms_update_allocation_limits();
  }

  bitMapLock()->unlock();
  releaseFreelistLocks();

2823 2824 2825 2826
  if (!CleanChunkPoolAsync) {
    Chunk::clean_chunk_pool();
  }

2827
  set_did_compact(false);
D
duke 已提交
2828 2829 2830 2831 2832 2833 2834
  _between_prologue_and_epilogue = false;  // ready for next cycle
}

void ConcurrentMarkSweepGeneration::gc_epilogue(bool full) {
  collector()->gc_epilogue(full);

  // Also reset promotion tracking in par gc thread states.
2835
  if (CollectedHeap::use_parallel_gc_threads()) {
D
duke 已提交
2836
    for (uint i = 0; i < ParallelGCThreads; i++) {
2837
      _par_gc_thread_states[i]->promo.stopTrackingPromotions(i);
D
duke 已提交
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
    }
  }
}

void ConcurrentMarkSweepGeneration::gc_epilogue_work(bool full) {
  assert(!incremental_collection_failed(), "Should have been cleared");
  cmsSpace()->setPreconsumptionDirtyCardClosure(NULL);
  cmsSpace()->gc_epilogue();
    // Print stat counters
  NOT_PRODUCT(
    assert(_numObjectsAllocated == 0, "check");
    assert(_numWordsAllocated == 0, "check");
    if (Verbose && PrintGC) {
      gclog_or_tty->print("Promoted "SIZE_FORMAT" objects, "
                          SIZE_FORMAT" bytes",
                 _numObjectsPromoted, _numWordsPromoted*sizeof(HeapWord));
    }
    _numObjectsPromoted = 0;
    _numWordsPromoted   = 0;
  )

  if (PrintGC && Verbose) {
    // Call down the chain in contiguous_available needs the freelistLock
    // so print this out before releasing the freeListLock.
    gclog_or_tty->print(" Contiguous available "SIZE_FORMAT" bytes ",
                        contiguous_available());
  }
}

#ifndef PRODUCT
bool CMSCollector::have_cms_token() {
  Thread* thr = Thread::current();
  if (thr->is_VM_thread()) {
    return ConcurrentMarkSweepThread::vm_thread_has_cms_token();
  } else if (thr->is_ConcurrentGC_thread()) {
    return ConcurrentMarkSweepThread::cms_thread_has_cms_token();
  } else if (thr->is_GC_task_thread()) {
    return ConcurrentMarkSweepThread::vm_thread_has_cms_token() &&
           ParGCRareEvent_lock->owned_by_self();
  }
  return false;
}
#endif

// Check reachability of the given heap address in CMS generation,
// treating all other generations as roots.
bool CMSCollector::is_cms_reachable(HeapWord* addr) {
  // We could "guarantee" below, rather than assert, but i'll
  // leave these as "asserts" so that an adventurous debugger
  // could try this in the product build provided some subset of
  // the conditions were met, provided they were intersted in the
  // results and knew that the computation below wouldn't interfere
  // with other concurrent computations mutating the structures
  // being read or written.
  assert(SafepointSynchronize::is_at_safepoint(),
         "Else mutations in object graph will make answer suspect");
  assert(have_cms_token(), "Should hold cms token");
  assert(haveFreelistLocks(), "must hold free list locks");
  assert_lock_strong(bitMapLock());

  // Clear the marking bit map array before starting, but, just
  // for kicks, first report if the given address is already marked
  gclog_or_tty->print_cr("Start: Address 0x%x is%s marked", addr,
                _markBitMap.isMarked(addr) ? "" : " not");

  if (verify_after_remark()) {
    MutexLockerEx x(verification_mark_bm()->lock(), Mutex::_no_safepoint_check_flag);
    bool result = verification_mark_bm()->isMarked(addr);
    gclog_or_tty->print_cr("TransitiveMark: Address 0x%x %s marked", addr,
                           result ? "IS" : "is NOT");
    return result;
  } else {
    gclog_or_tty->print_cr("Could not compute result");
    return false;
  }
}

2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931

void
CMSCollector::print_on_error(outputStream* st) {
  CMSCollector* collector = ConcurrentMarkSweepGeneration::_collector;
  if (collector != NULL) {
    CMSBitMap* bitmap = &collector->_markBitMap;
    st->print_cr("Marking Bits: (CMSBitMap*) " PTR_FORMAT, bitmap);
    bitmap->print_on_error(st, " Bits: ");

    st->cr();

    CMSBitMap* mut_bitmap = &collector->_modUnionTable;
    st->print_cr("Mod Union Table: (CMSBitMap*) " PTR_FORMAT, mut_bitmap);
    mut_bitmap->print_on_error(st, " Bits: ");
  }
}

D
duke 已提交
2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945
////////////////////////////////////////////////////////
// CMS Verification Support
////////////////////////////////////////////////////////
// Following the remark phase, the following invariant
// should hold -- each object in the CMS heap which is
// marked in markBitMap() should be marked in the verification_mark_bm().

class VerifyMarkedClosure: public BitMapClosure {
  CMSBitMap* _marks;
  bool       _failed;

 public:
  VerifyMarkedClosure(CMSBitMap* bm): _marks(bm), _failed(false) {}

2946
  bool do_bit(size_t offset) {
D
duke 已提交
2947 2948
    HeapWord* addr = _marks->offsetToHeapWord(offset);
    if (!_marks->isMarked(addr)) {
2949
      oop(addr)->print_on(gclog_or_tty);
D
duke 已提交
2950 2951 2952
      gclog_or_tty->print_cr(" ("INTPTR_FORMAT" should have been marked)", addr);
      _failed = true;
    }
2953
    return true;
D
duke 已提交
2954 2955 2956 2957 2958
  }

  bool failed() { return _failed; }
};

2959 2960
bool CMSCollector::verify_after_remark(bool silent) {
  if (!silent) gclog_or_tty->print(" [Verifying CMS Marking... ");
D
duke 已提交
2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997
  MutexLockerEx ml(verification_mark_bm()->lock(), Mutex::_no_safepoint_check_flag);
  static bool init = false;

  assert(SafepointSynchronize::is_at_safepoint(),
         "Else mutations in object graph will make answer suspect");
  assert(have_cms_token(),
         "Else there may be mutual interference in use of "
         " verification data structures");
  assert(_collectorState > Marking && _collectorState <= Sweeping,
         "Else marking info checked here may be obsolete");
  assert(haveFreelistLocks(), "must hold free list locks");
  assert_lock_strong(bitMapLock());


  // Allocate marking bit map if not already allocated
  if (!init) { // first time
    if (!verification_mark_bm()->allocate(_span)) {
      return false;
    }
    init = true;
  }

  assert(verification_mark_stack()->isEmpty(), "Should be empty");

  // Turn off refs discovery -- so we will be tracing through refs.
  // This is as intended, because by this time
  // GC must already have cleared any refs that need to be cleared,
  // and traced those that need to be marked; moreover,
  // the marking done here is not going to intefere in any
  // way with the marking information used by GC.
  NoRefDiscovery no_discovery(ref_processor());

  COMPILER2_PRESENT(DerivedPointerTableDeactivate dpt_deact;)

  // Clear any marks from a previous round
  verification_mark_bm()->clear_all();
  assert(verification_mark_stack()->isEmpty(), "markStack should be empty");
2998
  verify_work_stacks_empty();
D
duke 已提交
2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020

  GenCollectedHeap* gch = GenCollectedHeap::heap();
  gch->ensure_parsability(false);  // fill TLABs, but no need to retire them
  // Update the saved marks which may affect the root scans.
  gch->save_marks();

  if (CMSRemarkVerifyVariant == 1) {
    // In this first variant of verification, we complete
    // all marking, then check if the new marks-verctor is
    // a subset of the CMS marks-vector.
    verify_after_remark_work_1();
  } else if (CMSRemarkVerifyVariant == 2) {
    // In this second variant of verification, we flag an error
    // (i.e. an object reachable in the new marks-vector not reachable
    // in the CMS marks-vector) immediately, also indicating the
    // identify of an object (A) that references the unmarked object (B) --
    // presumably, a mutation to A failed to be picked up by preclean/remark?
    verify_after_remark_work_2();
  } else {
    warning("Unrecognized value %d for CMSRemarkVerifyVariant",
            CMSRemarkVerifyVariant);
  }
3021
  if (!silent) gclog_or_tty->print(" done] ");
D
duke 已提交
3022 3023 3024 3025 3026 3027 3028 3029
  return true;
}

void CMSCollector::verify_after_remark_work_1() {
  ResourceMark rm;
  HandleMark  hm;
  GenCollectedHeap* gch = GenCollectedHeap::heap();

3030 3031 3032
  // Get a clear set of claim bits for the strong roots processing to work with.
  ClassLoaderDataGraph::clear_claimed_marks();

D
duke 已提交
3033
  // Mark from roots one level into CMS
3034
  MarkRefsIntoClosure notOlder(_span, verification_mark_bm());
D
duke 已提交
3035 3036 3037 3038
  gch->rem_set()->prepare_for_younger_refs_iterate(false); // Not parallel.

  gch->gen_process_strong_roots(_cmsGen->level(),
                                true,   // younger gens are roots
3039
                                true,   // activate StrongRootsScope
3040
                                false,  // not scavenging
D
duke 已提交
3041
                                SharedHeap::ScanningOption(roots_scanning_options()),
3042 3043
                                &notOlder,
                                true,   // walk code active on stacks
3044 3045
                                NULL,
                                NULL); // SSS: Provide correct closure
D
duke 已提交
3046 3047 3048

  // Now mark from the roots
  MarkFromRootsClosure markFromRootsClosure(this, _span,
3049
    verification_mark_bm(), verification_mark_stack(),
D
duke 已提交
3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070
    false /* don't yield */, true /* verifying */);
  assert(_restart_addr == NULL, "Expected pre-condition");
  verification_mark_bm()->iterate(&markFromRootsClosure);
  while (_restart_addr != NULL) {
    // Deal with stack overflow: by restarting at the indicated
    // address.
    HeapWord* ra = _restart_addr;
    markFromRootsClosure.reset(ra);
    _restart_addr = NULL;
    verification_mark_bm()->iterate(&markFromRootsClosure, ra, _span.end());
  }
  assert(verification_mark_stack()->isEmpty(), "Should have been drained");
  verify_work_stacks_empty();

  // Marking completed -- now verify that each bit marked in
  // verification_mark_bm() is also marked in markBitMap(); flag all
  // errors by printing corresponding objects.
  VerifyMarkedClosure vcl(markBitMap());
  verification_mark_bm()->iterate(&vcl);
  if (vcl.failed()) {
    gclog_or_tty->print("Verification failed");
3071 3072
    Universe::heap()->print_on(gclog_or_tty);
    fatal("CMS: failed marking verification after remark");
D
duke 已提交
3073 3074 3075
  }
}

3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090
class VerifyKlassOopsKlassClosure : public KlassClosure {
  class VerifyKlassOopsClosure : public OopClosure {
    CMSBitMap* _bitmap;
   public:
    VerifyKlassOopsClosure(CMSBitMap* bitmap) : _bitmap(bitmap) { }
    void do_oop(oop* p)       { guarantee(*p == NULL || _bitmap->isMarked((HeapWord*) *p), "Should be marked"); }
    void do_oop(narrowOop* p) { ShouldNotReachHere(); }
  } _oop_closure;
 public:
  VerifyKlassOopsKlassClosure(CMSBitMap* bitmap) : _oop_closure(bitmap) {}
  void do_klass(Klass* k) {
    k->oops_do(&_oop_closure);
  }
};

D
duke 已提交
3091 3092 3093 3094 3095
void CMSCollector::verify_after_remark_work_2() {
  ResourceMark rm;
  HandleMark  hm;
  GenCollectedHeap* gch = GenCollectedHeap::heap();

3096 3097 3098
  // Get a clear set of claim bits for the strong roots processing to work with.
  ClassLoaderDataGraph::clear_claimed_marks();

D
duke 已提交
3099 3100
  // Mark from roots one level into CMS
  MarkRefsIntoVerifyClosure notOlder(_span, verification_mark_bm(),
3101
                                     markBitMap());
3102 3103
  CMKlassClosure klass_closure(&notOlder);

D
duke 已提交
3104 3105 3106
  gch->rem_set()->prepare_for_younger_refs_iterate(false); // Not parallel.
  gch->gen_process_strong_roots(_cmsGen->level(),
                                true,   // younger gens are roots
3107
                                true,   // activate StrongRootsScope
3108
                                false,  // not scavenging
D
duke 已提交
3109
                                SharedHeap::ScanningOption(roots_scanning_options()),
3110 3111
                                &notOlder,
                                true,   // walk code active on stacks
3112 3113
                                NULL,
                                &klass_closure);
D
duke 已提交
3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129

  // Now mark from the roots
  MarkFromRootsVerifyClosure markFromRootsClosure(this, _span,
    verification_mark_bm(), markBitMap(), verification_mark_stack());
  assert(_restart_addr == NULL, "Expected pre-condition");
  verification_mark_bm()->iterate(&markFromRootsClosure);
  while (_restart_addr != NULL) {
    // Deal with stack overflow: by restarting at the indicated
    // address.
    HeapWord* ra = _restart_addr;
    markFromRootsClosure.reset(ra);
    _restart_addr = NULL;
    verification_mark_bm()->iterate(&markFromRootsClosure, ra, _span.end());
  }
  assert(verification_mark_stack()->isEmpty(), "Should have been drained");
  verify_work_stacks_empty();
3130 3131 3132

  VerifyKlassOopsKlassClosure verify_klass_oops(verification_mark_bm());
  ClassLoaderDataGraph::classes_do(&verify_klass_oops);
D
duke 已提交
3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173

  // Marking completed -- now verify that each bit marked in
  // verification_mark_bm() is also marked in markBitMap(); flag all
  // errors by printing corresponding objects.
  VerifyMarkedClosure vcl(markBitMap());
  verification_mark_bm()->iterate(&vcl);
  assert(!vcl.failed(), "Else verification above should not have succeeded");
}

void ConcurrentMarkSweepGeneration::save_marks() {
  // delegate to CMS space
  cmsSpace()->save_marks();
  for (uint i = 0; i < ParallelGCThreads; i++) {
    _par_gc_thread_states[i]->promo.startTrackingPromotions();
  }
}

bool ConcurrentMarkSweepGeneration::no_allocs_since_save_marks() {
  return cmsSpace()->no_allocs_since_save_marks();
}

#define CMS_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix)    \
                                                                \
void ConcurrentMarkSweepGeneration::                            \
oop_since_save_marks_iterate##nv_suffix(OopClosureType* cl) {   \
  cl->set_generation(this);                                     \
  cmsSpace()->oop_since_save_marks_iterate##nv_suffix(cl);      \
  cl->reset_generation();                                       \
  save_marks();                                                 \
}

ALL_SINCE_SAVE_MARKS_CLOSURES(CMS_SINCE_SAVE_MARKS_DEFN)

void
ConcurrentMarkSweepGeneration::younger_refs_iterate(OopsInGenClosure* cl) {
  cl->set_generation(this);
  younger_refs_in_space_iterate(_cmsSpace, cl);
  cl->reset_generation();
}

void
3174
ConcurrentMarkSweepGeneration::oop_iterate(MemRegion mr, ExtendedOopClosure* cl) {
D
duke 已提交
3175 3176 3177 3178 3179 3180 3181 3182 3183
  if (freelistLock()->owned_by_self()) {
    Generation::oop_iterate(mr, cl);
  } else {
    MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
    Generation::oop_iterate(mr, cl);
  }
}

void
3184
ConcurrentMarkSweepGeneration::oop_iterate(ExtendedOopClosure* cl) {
D
duke 已提交
3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202
  if (freelistLock()->owned_by_self()) {
    Generation::oop_iterate(cl);
  } else {
    MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
    Generation::oop_iterate(cl);
  }
}

void
ConcurrentMarkSweepGeneration::object_iterate(ObjectClosure* cl) {
  if (freelistLock()->owned_by_self()) {
    Generation::object_iterate(cl);
  } else {
    MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
    Generation::object_iterate(cl);
  }
}

3203 3204 3205 3206 3207 3208 3209 3210 3211 3212
void
ConcurrentMarkSweepGeneration::safe_object_iterate(ObjectClosure* cl) {
  if (freelistLock()->owned_by_self()) {
    Generation::safe_object_iterate(cl);
  } else {
    MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
    Generation::safe_object_iterate(cl);
  }
}

D
duke 已提交
3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232
void
ConcurrentMarkSweepGeneration::post_compact() {
}

void
ConcurrentMarkSweepGeneration::prepare_for_verify() {
  // Fix the linear allocation blocks to look like free blocks.

  // Locks are normally acquired/released in gc_prologue/gc_epilogue, but those
  // are not called when the heap is verified during universe initialization and
  // at vm shutdown.
  if (freelistLock()->owned_by_self()) {
    cmsSpace()->prepare_for_verify();
  } else {
    MutexLockerEx fll(freelistLock(), Mutex::_no_safepoint_check_flag);
    cmsSpace()->prepare_for_verify();
  }
}

void
3233
ConcurrentMarkSweepGeneration::verify() {
D
duke 已提交
3234 3235 3236 3237
  // Locks are normally acquired/released in gc_prologue/gc_epilogue, but those
  // are not called when the heap is verified during universe initialization and
  // at vm shutdown.
  if (freelistLock()->owned_by_self()) {
3238
    cmsSpace()->verify();
D
duke 已提交
3239 3240
  } else {
    MutexLockerEx fll(freelistLock(), Mutex::_no_safepoint_check_flag);
3241
    cmsSpace()->verify();
D
duke 已提交
3242 3243 3244
  }
}

3245 3246
void CMSCollector::verify() {
  _cmsGen->verify();
D
duke 已提交
3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270
}

#ifndef PRODUCT
bool CMSCollector::overflow_list_is_empty() const {
  assert(_num_par_pushes >= 0, "Inconsistency");
  if (_overflow_list == NULL) {
    assert(_num_par_pushes == 0, "Inconsistency");
  }
  return _overflow_list == NULL;
}

// The methods verify_work_stacks_empty() and verify_overflow_empty()
// merely consolidate assertion checks that appear to occur together frequently.
void CMSCollector::verify_work_stacks_empty() const {
  assert(_markStack.isEmpty(), "Marking stack should be empty");
  assert(overflow_list_is_empty(), "Overflow list should be empty");
}

void CMSCollector::verify_overflow_empty() const {
  assert(overflow_list_is_empty(), "Overflow list should be empty");
  assert(no_preserved_marks(), "No preserved marks");
}
#endif // PRODUCT

3271
// Decide if we want to enable class unloading as part of the
3272
// ensuing concurrent GC cycle. We will collect and
3273 3274 3275 3276
// unload classes if it's the case that:
// (1) an explicit gc request has been made and the flag
//     ExplicitGCInvokesConcurrentAndUnloadsClasses is set, OR
// (2) (a) class unloading is enabled at the command line, and
3277
//     (b) old gen is getting really full
3278 3279 3280 3281 3282 3283 3284
// NOTE: Provided there is no change in the state of the heap between
// calls to this method, it should have idempotent results. Moreover,
// its results should be monotonically increasing (i.e. going from 0 to 1,
// but not 1 to 0) between successive calls between which the heap was
// not collected. For the implementation below, it must thus rely on
// the property that concurrent_cycles_since_last_unload()
// will not decrease unless a collection cycle happened and that
3285
// _cmsGen->is_too_full() are
3286 3287
// themselves also monotonic in that sense. See check_monotonicity()
// below.
3288
void CMSCollector::update_should_unload_classes() {
3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306
  _should_unload_classes = false;
  // Condition 1 above
  if (_full_gc_requested && ExplicitGCInvokesConcurrentAndUnloadsClasses) {
    _should_unload_classes = true;
  } else if (CMSClassUnloadingEnabled) { // Condition 2.a above
    // Disjuncts 2.b.(i,ii,iii) above
    _should_unload_classes = (concurrent_cycles_since_last_unload() >=
                              CMSClassUnloadingMaxInterval)
                           || _cmsGen->is_too_full();
  }
}

bool ConcurrentMarkSweepGeneration::is_too_full() const {
  bool res = should_concurrent_collect();
  res = res && (occupancy() > (double)CMSIsTooFullPercentage/100.0);
  return res;
}

D
duke 已提交
3307
void CMSCollector::setup_cms_unloading_and_verification_state() {
3308
  const  bool should_verify =   VerifyBeforeGC || VerifyAfterGC || VerifyDuringGC
D
duke 已提交
3309
                             || VerifyBeforeExit;
3310
  const  int  rso           =   SharedHeap::SO_Strings | SharedHeap::SO_CodeCache;
D
duke 已提交
3311

3312
  if (should_unload_classes()) {   // Should unload classes this cycle
D
duke 已提交
3313 3314 3315 3316 3317 3318
    remove_root_scanning_option(rso);  // Shrink the root set appropriately
    set_verifying(should_verify);    // Set verification state for this cycle
    return;                            // Nothing else needs to be done at this time
  }

  // Not unloading classes this cycle
3319 3320
  assert(!should_unload_classes(), "Inconsitency!");
  if ((!verifying() || unloaded_classes_last_cycle()) && should_verify) {
D
duke 已提交
3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349
    // Include symbols, strings and code cache elements to prevent their resurrection.
    add_root_scanning_option(rso);
    set_verifying(true);
  } else if (verifying() && !should_verify) {
    // We were verifying, but some verification flags got disabled.
    set_verifying(false);
    // Exclude symbols, strings and code cache elements from root scanning to
    // reduce IM and RM pauses.
    remove_root_scanning_option(rso);
  }
}


#ifndef PRODUCT
HeapWord* CMSCollector::block_start(const void* p) const {
  const HeapWord* addr = (HeapWord*)p;
  if (_span.contains(p)) {
    if (_cmsGen->cmsSpace()->is_in_reserved(addr)) {
      return _cmsGen->cmsSpace()->block_start(p);
    }
  }
  return NULL;
}
#endif

HeapWord*
ConcurrentMarkSweepGeneration::expand_and_allocate(size_t word_size,
                                                   bool   tlab,
                                                   bool   parallel) {
3350
  CMSSynchronousYieldRequest yr;
D
duke 已提交
3351 3352 3353 3354 3355 3356 3357
  assert(!tlab, "Can't deal with TLAB allocation");
  MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
  expand(word_size*HeapWordSize, MinHeapDeltaBytes,
    CMSExpansionCause::_satisfy_allocation);
  if (GCExpandToAllocateDelayMillis > 0) {
    os::sleep(Thread::current(), GCExpandToAllocateDelayMillis, false);
  }
3358
  return have_lock_and_allocate(word_size, tlab);
D
duke 已提交
3359 3360 3361 3362 3363
}

// YSR: All of this generation expansion/shrinking stuff is an exact copy of
// OneContigSpaceCardGeneration, which makes me wonder if we should move this
// to CardGeneration and share it...
3364 3365 3366 3367
bool ConcurrentMarkSweepGeneration::expand(size_t bytes, size_t expand_bytes) {
  return CardGeneration::expand(bytes, expand_bytes);
}

D
duke 已提交
3368 3369 3370 3371
void ConcurrentMarkSweepGeneration::expand(size_t bytes, size_t expand_bytes,
  CMSExpansionCause::Cause cause)
{

3372 3373
  bool success = expand(bytes, expand_bytes);

D
duke 已提交
3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437
  // remember why we expanded; this information is used
  // by shouldConcurrentCollect() when making decisions on whether to start
  // a new CMS cycle.
  if (success) {
    set_expansion_cause(cause);
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print_cr("Expanded CMS gen for %s",
        CMSExpansionCause::to_string(cause));
    }
  }
}

HeapWord* ConcurrentMarkSweepGeneration::expand_and_par_lab_allocate(CMSParGCThreadState* ps, size_t word_sz) {
  HeapWord* res = NULL;
  MutexLocker x(ParGCRareEvent_lock);
  while (true) {
    // Expansion by some other thread might make alloc OK now:
    res = ps->lab.alloc(word_sz);
    if (res != NULL) return res;
    // If there's not enough expansion space available, give up.
    if (_virtual_space.uncommitted_size() < (word_sz * HeapWordSize)) {
      return NULL;
    }
    // Otherwise, we try expansion.
    expand(word_sz*HeapWordSize, MinHeapDeltaBytes,
      CMSExpansionCause::_allocate_par_lab);
    // Now go around the loop and try alloc again;
    // A competing par_promote might beat us to the expansion space,
    // so we may go around the loop again if promotion fails agaion.
    if (GCExpandToAllocateDelayMillis > 0) {
      os::sleep(Thread::current(), GCExpandToAllocateDelayMillis, false);
    }
  }
}


bool ConcurrentMarkSweepGeneration::expand_and_ensure_spooling_space(
  PromotionInfo* promo) {
  MutexLocker x(ParGCRareEvent_lock);
  size_t refill_size_bytes = promo->refillSize() * HeapWordSize;
  while (true) {
    // Expansion by some other thread might make alloc OK now:
    if (promo->ensure_spooling_space()) {
      assert(promo->has_spooling_space(),
             "Post-condition of successful ensure_spooling_space()");
      return true;
    }
    // If there's not enough expansion space available, give up.
    if (_virtual_space.uncommitted_size() < refill_size_bytes) {
      return false;
    }
    // Otherwise, we try expansion.
    expand(refill_size_bytes, MinHeapDeltaBytes,
      CMSExpansionCause::_allocate_par_spooling_space);
    // Now go around the loop and try alloc again;
    // A competing allocation might beat us to the expansion space,
    // so we may go around the loop again if allocation fails again.
    if (GCExpandToAllocateDelayMillis > 0) {
      os::sleep(Thread::current(), GCExpandToAllocateDelayMillis, false);
    }
  }
}


3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457
void ConcurrentMarkSweepGeneration::shrink_by(size_t bytes) {
  assert_locked_or_safepoint(ExpandHeap_lock);
  // Shrink committed space
  _virtual_space.shrink_by(bytes);
  // Shrink space; this also shrinks the space's BOT
  _cmsSpace->set_end((HeapWord*) _virtual_space.high());
  size_t new_word_size = heap_word_size(_cmsSpace->capacity());
  // Shrink the shared block offset array
  _bts->resize(new_word_size);
  MemRegion mr(_cmsSpace->bottom(), new_word_size);
  // Shrink the card table
  Universe::heap()->barrier_set()->resize_covered_region(mr);

  if (Verbose && PrintGC) {
    size_t new_mem_size = _virtual_space.committed_size();
    size_t old_mem_size = new_mem_size + bytes;
    gclog_or_tty->print_cr("Shrinking %s from " SIZE_FORMAT "K to " SIZE_FORMAT "K",
                  name(), old_mem_size/K, new_mem_size/K);
  }
}
D
duke 已提交
3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477

void ConcurrentMarkSweepGeneration::shrink(size_t bytes) {
  assert_locked_or_safepoint(Heap_lock);
  size_t size = ReservedSpace::page_align_size_down(bytes);
  if (size > 0) {
    shrink_by(size);
  }
}

bool ConcurrentMarkSweepGeneration::grow_by(size_t bytes) {
  assert_locked_or_safepoint(Heap_lock);
  bool result = _virtual_space.expand_by(bytes);
  if (result) {
    size_t new_word_size =
      heap_word_size(_virtual_space.committed_size());
    MemRegion mr(_cmsSpace->bottom(), new_word_size);
    _bts->resize(new_word_size);  // resize the block offset shared array
    Universe::heap()->barrier_set()->resize_covered_region(mr);
    // Hmmmm... why doesn't CFLS::set_end verify locking?
    // This is quite ugly; FIX ME XXX
3478
    _cmsSpace->assert_locked(freelistLock());
D
duke 已提交
3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489
    _cmsSpace->set_end((HeapWord*)_virtual_space.high());

    // update the space and generation capacity counters
    if (UsePerfData) {
      _space_counters->update_capacity();
      _gen_counters->update_all();
    }

    if (Verbose && PrintGC) {
      size_t new_mem_size = _virtual_space.committed_size();
      size_t old_mem_size = new_mem_size - bytes;
3490
      gclog_or_tty->print_cr("Expanding %s from " SIZE_FORMAT "K by " SIZE_FORMAT "K to " SIZE_FORMAT "K",
D
duke 已提交
3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507
                    name(), old_mem_size/K, bytes/K, new_mem_size/K);
    }
  }
  return result;
}

bool ConcurrentMarkSweepGeneration::grow_to_reserved() {
  assert_locked_or_safepoint(Heap_lock);
  bool success = true;
  const size_t remaining_bytes = _virtual_space.uncommitted_size();
  if (remaining_bytes > 0) {
    success = grow_by(remaining_bytes);
    DEBUG_ONLY(if (!success) warning("grow to reserved failed");)
  }
  return success;
}

3508
void ConcurrentMarkSweepGeneration::shrink_free_list_by(size_t bytes) {
D
duke 已提交
3509 3510
  assert_locked_or_safepoint(Heap_lock);
  assert_lock_strong(freelistLock());
3511 3512 3513
  if (PrintGCDetails && Verbose) {
    warning("Shrinking of CMS not yet implemented");
  }
D
duke 已提交
3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552
  return;
}


// Simple ctor/dtor wrapper for accounting & timer chores around concurrent
// phases.
class CMSPhaseAccounting: public StackObj {
 public:
  CMSPhaseAccounting(CMSCollector *collector,
                     const char *phase,
                     bool print_cr = true);
  ~CMSPhaseAccounting();

 private:
  CMSCollector *_collector;
  const char *_phase;
  elapsedTimer _wallclock;
  bool _print_cr;

 public:
  // Not MT-safe; so do not pass around these StackObj's
  // where they may be accessed by other threads.
  jlong wallclock_millis() {
    assert(_wallclock.is_active(), "Wall clock should not stop");
    _wallclock.stop();  // to record time
    jlong ret = _wallclock.milliseconds();
    _wallclock.start(); // restart
    return ret;
  }
};

CMSPhaseAccounting::CMSPhaseAccounting(CMSCollector *collector,
                                       const char *phase,
                                       bool print_cr) :
  _collector(collector), _phase(phase), _print_cr(print_cr) {

  if (PrintCMSStatistics != 0) {
    _collector->resetYields();
  }
3553
  if (PrintGCDetails) {
D
duke 已提交
3554
    gclog_or_tty->date_stamp(PrintGCDateStamps);
3555 3556
    gclog_or_tty->stamp(PrintGCTimeStamps);
    gclog_or_tty->print_cr("[%s-concurrent-%s-start]",
D
duke 已提交
3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569
      _collector->cmsGen()->short_name(), _phase);
  }
  _collector->resetTimer();
  _wallclock.start();
  _collector->startTimer();
}

CMSPhaseAccounting::~CMSPhaseAccounting() {
  assert(_wallclock.is_active(), "Wall clock should not have stopped");
  _collector->stopTimer();
  _wallclock.stop();
  if (PrintGCDetails) {
    gclog_or_tty->date_stamp(PrintGCDateStamps);
3570
    gclog_or_tty->stamp(PrintGCTimeStamps);
D
duke 已提交
3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585
    gclog_or_tty->print("[%s-concurrent-%s: %3.3f/%3.3f secs]",
                 _collector->cmsGen()->short_name(),
                 _phase, _collector->timerValue(), _wallclock.seconds());
    if (_print_cr) {
      gclog_or_tty->print_cr("");
    }
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr(" (CMS-concurrent-%s yielded %d times)", _phase,
                    _collector->yields());
    }
  }
}

// CMS work

J
jmasa 已提交
3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610
// The common parts of CMSParInitialMarkTask and CMSParRemarkTask.
class CMSParMarkTask : public AbstractGangTask {
 protected:
  CMSCollector*     _collector;
  int               _n_workers;
  CMSParMarkTask(const char* name, CMSCollector* collector, int n_workers) :
      AbstractGangTask(name),
      _collector(collector),
      _n_workers(n_workers) {}
  // Work method in support of parallel rescan ... of young gen spaces
  void do_young_space_rescan(uint worker_id, OopsInGenClosure* cl,
                             ContiguousSpace* space,
                             HeapWord** chunk_array, size_t chunk_top);
  void work_on_young_gen_roots(uint worker_id, OopsInGenClosure* cl);
};

// Parallel initial mark task
class CMSParInitialMarkTask: public CMSParMarkTask {
 public:
  CMSParInitialMarkTask(CMSCollector* collector, int n_workers) :
      CMSParMarkTask("Scan roots and young gen for initial mark in parallel",
                     collector, n_workers) {}
  void work(uint worker_id);
};

D
duke 已提交
3611 3612 3613 3614 3615 3616
// Checkpoint the roots into this generation from outside
// this generation. [Note this initial checkpoint need only
// be approximate -- we'll do a catch up phase subsequently.]
void CMSCollector::checkpointRootsInitial(bool asynch) {
  assert(_collectorState == InitialMarking, "Wrong collector state");
  check_correct_thread_executing();
3617
  TraceCMSMemoryManagerStats tms(_collectorState,GenCollectedHeap::heap()->gc_cause());
3618

S
sla 已提交
3619 3620 3621
  save_heap_summary();
  report_heap_summary(GCWhen::BeforeGC);

D
duke 已提交
3622 3623 3624 3625 3626 3627 3628 3629
  ReferenceProcessor* rp = ref_processor();
  SpecializationStats::clear();
  assert(_restart_addr == NULL, "Control point invariant");
  if (asynch) {
    // acquire locks for subsequent manipulations
    MutexLockerEx x(bitMapLock(),
                    Mutex::_no_safepoint_check_flag);
    checkpointRootsInitialWork(asynch);
3630 3631
    // enable ("weak") refs discovery
    rp->enable_discovery(true /*verify_disabled*/, true /*check_no_refs*/);
D
duke 已提交
3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642
    _collectorState = Marking;
  } else {
    // (Weak) Refs discovery: this is controlled from genCollectedHeap::do_collection
    // which recognizes if we are a CMS generation, and doesn't try to turn on
    // discovery; verify that they aren't meddling.
    assert(!rp->discovery_is_atomic(),
           "incorrect setting of discovery predicate");
    assert(!rp->discovery_enabled(), "genCollectedHeap shouldn't control "
           "ref discovery for this generation kind");
    // already have locks
    checkpointRootsInitialWork(asynch);
3643 3644
    // now enable ("weak") refs discovery
    rp->enable_discovery(true /*verify_disabled*/, false /*verify_no_refs*/);
D
duke 已提交
3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666
    _collectorState = Marking;
  }
  SpecializationStats::print();
}

void CMSCollector::checkpointRootsInitialWork(bool asynch) {
  assert(SafepointSynchronize::is_at_safepoint(), "world should be stopped");
  assert(_collectorState == InitialMarking, "just checking");

  // If there has not been a GC[n-1] since last GC[n] cycle completed,
  // precede our marking with a collection of all
  // younger generations to keep floating garbage to a minimum.
  // XXX: we won't do this for now -- it's an optimization to be done later.

  // already have locks
  assert_lock_strong(bitMapLock());
  assert(_markBitMap.isAllClear(), "was reset at end of previous cycle");

  // Setup the verification and class unloading state for this
  // CMS collection cycle.
  setup_cms_unloading_and_verification_state();

S
sla 已提交
3667 3668
  NOT_PRODUCT(GCTraceTime t("\ncheckpointRootsInitialWork",
    PrintGCDetails && Verbose, true, _gc_timer_cm);)
D
duke 已提交
3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683
  if (UseAdaptiveSizePolicy) {
    size_policy()->checkpoint_roots_initial_begin();
  }

  // Reset all the PLAB chunk arrays if necessary.
  if (_survivor_plab_array != NULL && !CMSPLABRecordAlways) {
    reset_survivor_plab_arrays();
  }

  ResourceMark rm;
  HandleMark  hm;

  FalseClosure falseClosure;
  // In the case of a synchronous collection, we will elide the
  // remark step, so it's important to catch all the nmethod oops
3684 3685 3686 3687
  // in this step.
  // The final 'true' flag to gen_process_strong_roots will ensure this.
  // If 'async' is true, we can relax the nmethod tracing.
  MarkRefsIntoClosure notOlder(_span, &_markBitMap);
D
duke 已提交
3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699
  GenCollectedHeap* gch = GenCollectedHeap::heap();

  verify_work_stacks_empty();
  verify_overflow_empty();

  gch->ensure_parsability(false);  // fill TLABs, but no need to retire them
  // Update the saved marks which may affect the root scans.
  gch->save_marks();

  // weak reference processing has not started yet.
  ref_processor()->set_enqueuing_is_done(false);

3700 3701 3702 3703 3704 3705 3706 3707
  // Need to remember all newly created CLDs,
  // so that we can guarantee that the remark finds them.
  ClassLoaderDataGraph::remember_new_clds(true);

  // Whenever a CLD is found, it will be claimed before proceeding to mark
  // the klasses. The claimed marks need to be cleared before marking starts.
  ClassLoaderDataGraph::clear_claimed_marks();

3708 3709 3710 3711
  if (CMSPrintEdenSurvivorChunks) {
    print_eden_and_survivor_chunk_arrays();
  }

D
duke 已提交
3712 3713
  {
    COMPILER2_PRESENT(DerivedPointerTableDeactivate dpt_deact;)
J
jmasa 已提交
3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743
    if (CMSParallelInitialMarkEnabled && CollectedHeap::use_parallel_gc_threads()) {
      // The parallel version.
      FlexibleWorkGang* workers = gch->workers();
      assert(workers != NULL, "Need parallel worker threads.");
      int n_workers = workers->active_workers();
      CMSParInitialMarkTask tsk(this, n_workers);
      gch->set_par_threads(n_workers);
      initialize_sequential_subtasks_for_young_gen_rescan(n_workers);
      if (n_workers > 1) {
        GenCollectedHeap::StrongRootsScope srs(gch);
        workers->run_task(&tsk);
      } else {
        GenCollectedHeap::StrongRootsScope srs(gch);
        tsk.work(0);
      }
      gch->set_par_threads(0);
    } else {
      // The serial version.
      CMKlassClosure klass_closure(&notOlder);
      gch->rem_set()->prepare_for_younger_refs_iterate(false); // Not parallel.
      gch->gen_process_strong_roots(_cmsGen->level(),
                                    true,   // younger gens are roots
                                    true,   // activate StrongRootsScope
                                    false,  // not scavenging
                                    SharedHeap::ScanningOption(roots_scanning_options()),
                                    &notOlder,
                                    true,   // walk all of code cache if (so & SO_CodeCache)
                                    NULL,
                                    &klass_closure);
    }
D
duke 已提交
3744 3745 3746 3747 3748 3749 3750 3751 3752 3753
  }

  // Clear mod-union table; it will be dirtied in the prologue of
  // CMS generation per each younger generation collection.

  assert(_modUnionTable.isAllClear(),
       "Was cleared in most recent final checkpoint phase"
       " or no bits are set in the gc_prologue before the start of the next "
       "subsequent marking phase.");

3754 3755
  assert(_ct->klass_rem_set()->mod_union_is_clear(), "Must be");

D
duke 已提交
3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848
  // Save the end of the used_region of the constituent generations
  // to be used to limit the extent of sweep in each generation.
  save_sweep_limits();
  if (UseAdaptiveSizePolicy) {
    size_policy()->checkpoint_roots_initial_end(gch->gc_cause());
  }
  verify_overflow_empty();
}

bool CMSCollector::markFromRoots(bool asynch) {
  // we might be tempted to assert that:
  // assert(asynch == !SafepointSynchronize::is_at_safepoint(),
  //        "inconsistent argument?");
  // However that wouldn't be right, because it's possible that
  // a safepoint is indeed in progress as a younger generation
  // stop-the-world GC happens even as we mark in this generation.
  assert(_collectorState == Marking, "inconsistent state?");
  check_correct_thread_executing();
  verify_overflow_empty();

  bool res;
  if (asynch) {

    // Start the timers for adaptive size policy for the concurrent phases
    // Do it here so that the foreground MS can use the concurrent
    // timer since a foreground MS might has the sweep done concurrently
    // or STW.
    if (UseAdaptiveSizePolicy) {
      size_policy()->concurrent_marking_begin();
    }

    // Weak ref discovery note: We may be discovering weak
    // refs in this generation concurrent (but interleaved) with
    // weak ref discovery by a younger generation collector.

    CMSTokenSyncWithLocks ts(true, bitMapLock());
    TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
    CMSPhaseAccounting pa(this, "mark", !PrintGCDetails);
    res = markFromRootsWork(asynch);
    if (res) {
      _collectorState = Precleaning;
    } else { // We failed and a foreground collection wants to take over
      assert(_foregroundGCIsActive, "internal state inconsistency");
      assert(_restart_addr == NULL,  "foreground will restart from scratch");
      if (PrintGCDetails) {
        gclog_or_tty->print_cr("bailing out to foreground collection");
      }
    }
    if (UseAdaptiveSizePolicy) {
      size_policy()->concurrent_marking_end();
    }
  } else {
    assert(SafepointSynchronize::is_at_safepoint(),
           "inconsistent with asynch == false");
    if (UseAdaptiveSizePolicy) {
      size_policy()->ms_collection_marking_begin();
    }
    // already have locks
    res = markFromRootsWork(asynch);
    _collectorState = FinalMarking;
    if (UseAdaptiveSizePolicy) {
      GenCollectedHeap* gch = GenCollectedHeap::heap();
      size_policy()->ms_collection_marking_end(gch->gc_cause());
    }
  }
  verify_overflow_empty();
  return res;
}

bool CMSCollector::markFromRootsWork(bool asynch) {
  // iterate over marked bits in bit map, doing a full scan and mark
  // from these roots using the following algorithm:
  // . if oop is to the right of the current scan pointer,
  //   mark corresponding bit (we'll process it later)
  // . else (oop is to left of current scan pointer)
  //   push oop on marking stack
  // . drain the marking stack

  // Note that when we do a marking step we need to hold the
  // bit map lock -- recall that direct allocation (by mutators)
  // and promotion (by younger generation collectors) is also
  // marking the bit map. [the so-called allocate live policy.]
  // Because the implementation of bit map marking is not
  // robust wrt simultaneous marking of bits in the same word,
  // we need to make sure that there is no such interference
  // between concurrent such updates.

  // already have locks
  assert_lock_strong(bitMapLock());

  verify_work_stacks_empty();
  verify_overflow_empty();
  bool result = false;
3849
  if (CMSConcurrentMTEnabled && ConcGCThreads > 0) {
D
duke 已提交
3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863
    result = do_marking_mt(asynch);
  } else {
    result = do_marking_st(asynch);
  }
  return result;
}

// Forward decl
class CMSConcMarkingTask;

class CMSConcMarkingTerminator: public ParallelTaskTerminator {
  CMSCollector*       _collector;
  CMSConcMarkingTask* _task;
 public:
3864 3865
  virtual void yield();

D
duke 已提交
3866 3867 3868 3869
  // "n_threads" is the number of threads to be terminated.
  // "queue_set" is a set of work queues of other threads.
  // "collector" is the CMS collector associated with this task terminator.
  // "yield" indicates whether we need the gang as a whole to yield.
3870
  CMSConcMarkingTerminator(int n_threads, TaskQueueSetSuper* queue_set, CMSCollector* collector) :
D
duke 已提交
3871
    ParallelTaskTerminator(n_threads, queue_set),
3872 3873 3874 3875 3876 3877
    _collector(collector) { }

  void set_task(CMSConcMarkingTask* task) {
    _task = task;
  }
};
D
duke 已提交
3878

3879 3880 3881 3882
class CMSConcMarkingTerminatorTerminator: public TerminatorTerminator {
  CMSConcMarkingTask* _task;
 public:
  bool should_exit_termination();
D
duke 已提交
3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894
  void set_task(CMSConcMarkingTask* task) {
    _task = task;
  }
};

// MT Concurrent Marking Task
class CMSConcMarkingTask: public YieldingFlexibleGangTask {
  CMSCollector* _collector;
  int           _n_workers;                  // requested/desired # workers
  bool          _asynch;
  bool          _result;
  CompactibleFreeListSpace*  _cms_space;
3895 3896 3897
  char          _pad_front[64];   // padding to ...
  HeapWord*     _global_finger;   // ... avoid sharing cache line
  char          _pad_back[64];
3898
  HeapWord*     _restart_addr;
D
duke 已提交
3899 3900 3901 3902 3903 3904

  //  Exposed here for yielding support
  Mutex* const _bit_map_lock;

  // The per thread work queues, available here for stealing
  OopTaskQueueSet*  _task_queues;
3905 3906

  // Termination (and yielding) support
D
duke 已提交
3907
  CMSConcMarkingTerminator _term;
3908
  CMSConcMarkingTerminatorTerminator _term_term;
D
duke 已提交
3909 3910 3911 3912

 public:
  CMSConcMarkingTask(CMSCollector* collector,
                 CompactibleFreeListSpace* cms_space,
3913
                 bool asynch,
D
duke 已提交
3914 3915 3916 3917 3918
                 YieldingFlexibleWorkGang* workers,
                 OopTaskQueueSet* task_queues):
    YieldingFlexibleGangTask("Concurrent marking done multi-threaded"),
    _collector(collector),
    _cms_space(cms_space),
3919 3920
    _asynch(asynch), _n_workers(0), _result(true),
    _task_queues(task_queues),
3921
    _term(_n_workers, task_queues, _collector),
D
duke 已提交
3922 3923
    _bit_map_lock(collector->bitMapLock())
  {
3924
    _requested_size = _n_workers;
D
duke 已提交
3925
    _term.set_task(this);
3926
    _term_term.set_task(this);
3927
    _restart_addr = _global_finger = _cms_space->bottom();
D
duke 已提交
3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938
  }


  OopTaskQueueSet* task_queues()  { return _task_queues; }

  OopTaskQueue* work_queue(int i) { return task_queues()->queue(i); }

  HeapWord** global_finger_addr() { return &_global_finger; }

  CMSConcMarkingTerminator* terminator() { return &_term; }

3939 3940 3941 3942
  virtual void set_for_termination(int active_workers) {
    terminator()->reset_for_reuse(active_workers);
  }

3943
  void work(uint worker_id);
3944 3945 3946 3947 3948
  bool should_yield() {
    return    ConcurrentMarkSweepThread::should_yield()
           && !_collector->foregroundGCIsActive()
           && _asynch;
  }
D
duke 已提交
3949 3950 3951 3952 3953

  virtual void coordinator_yield();  // stuff done by coordinator
  bool result() { return _result; }

  void reset(HeapWord* ra) {
3954 3955
    assert(_global_finger >= _cms_space->end(),  "Postcondition of ::work(i)");
    _restart_addr = _global_finger = ra;
D
duke 已提交
3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967
    _term.reset_for_reuse();
  }

  static bool get_work_from_overflow_stack(CMSMarkStack* ovflw_stk,
                                           OopTaskQueue* work_q);

 private:
  void do_scan_and_mark(int i, CompactibleFreeListSpace* sp);
  void do_work_steal(int i);
  void bump_global_finger(HeapWord* f);
};

3968 3969 3970 3971 3972 3973 3974 3975 3976
bool CMSConcMarkingTerminatorTerminator::should_exit_termination() {
  assert(_task != NULL, "Error");
  return _task->yielding();
  // Note that we do not need the disjunct || _task->should_yield() above
  // because we want terminating threads to yield only if the task
  // is already in the midst of yielding, which happens only after at least one
  // thread has yielded.
}

D
duke 已提交
3977
void CMSConcMarkingTerminator::yield() {
3978
  if (_task->should_yield()) {
D
duke 已提交
3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013
    _task->yield();
  } else {
    ParallelTaskTerminator::yield();
  }
}

////////////////////////////////////////////////////////////////
// Concurrent Marking Algorithm Sketch
////////////////////////////////////////////////////////////////
// Until all tasks exhausted (both spaces):
// -- claim next available chunk
// -- bump global finger via CAS
// -- find first object that starts in this chunk
//    and start scanning bitmap from that position
// -- scan marked objects for oops
// -- CAS-mark target, and if successful:
//    . if target oop is above global finger (volatile read)
//      nothing to do
//    . if target oop is in chunk and above local finger
//        then nothing to do
//    . else push on work-queue
// -- Deal with possible overflow issues:
//    . local work-queue overflow causes stuff to be pushed on
//      global (common) overflow queue
//    . always first empty local work queue
//    . then get a batch of oops from global work queue if any
//    . then do work stealing
// -- When all tasks claimed (both spaces)
//    and local work queue empty,
//    then in a loop do:
//    . check global overflow stack; steal a batch of oops and trace
//    . try to steal from other threads oif GOS is empty
//    . if neither is available, offer termination
// -- Terminate and return result
//
4014
void CMSConcMarkingTask::work(uint worker_id) {
D
duke 已提交
4015 4016 4017 4018 4019 4020 4021
  elapsedTimer _timer;
  ResourceMark rm;
  HandleMark hm;

  DEBUG_ONLY(_collector->verify_overflow_empty();)

  // Before we begin work, our work queue should be empty
4022
  assert(work_queue(worker_id)->size() == 0, "Expected to be empty");
D
duke 已提交
4023 4024
  // Scan the bitmap covering _cms_space, tracing through grey objects.
  _timer.start();
4025
  do_scan_and_mark(worker_id, _cms_space);
D
duke 已提交
4026 4027 4028
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr("Finished cms space scanning in %dth thread: %3.3f sec",
4029 4030
      worker_id, _timer.seconds());
      // XXX: need xxx/xxx type of notation, two timers
D
duke 已提交
4031 4032 4033 4034 4035
  }

  // ... do work stealing
  _timer.reset();
  _timer.start();
4036
  do_work_steal(worker_id);
D
duke 已提交
4037 4038 4039
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr("Finished work stealing in %dth thread: %3.3f sec",
4040 4041
      worker_id, _timer.seconds());
      // XXX: need xxx/xxx type of notation, two timers
D
duke 已提交
4042 4043
  }
  assert(_collector->_markStack.isEmpty(), "Should have been emptied");
4044
  assert(work_queue(worker_id)->size() == 0, "Should have been emptied");
D
duke 已提交
4045 4046 4047 4048 4049
  // Note that under the current task protocol, the
  // following assertion is true even of the spaces
  // expanded since the completion of the concurrent
  // marking. XXX This will likely change under a strict
  // ABORT semantics.
4050 4051 4052 4053 4054 4055
  // After perm removal the comparison was changed to
  // greater than or equal to from strictly greater than.
  // Before perm removal the highest address sweep would
  // have been at the end of perm gen but now is at the
  // end of the tenured gen.
  assert(_global_finger >=  _cms_space->end(),
D
duke 已提交
4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086
         "All tasks have been completed");
  DEBUG_ONLY(_collector->verify_overflow_empty();)
}

void CMSConcMarkingTask::bump_global_finger(HeapWord* f) {
  HeapWord* read = _global_finger;
  HeapWord* cur  = read;
  while (f > read) {
    cur = read;
    read = (HeapWord*) Atomic::cmpxchg_ptr(f, &_global_finger, cur);
    if (cur == read) {
      // our cas succeeded
      assert(_global_finger >= f, "protocol consistency");
      break;
    }
  }
}

// This is really inefficient, and should be redone by
// using (not yet available) block-read and -write interfaces to the
// stack and the work_queue. XXX FIX ME !!!
bool CMSConcMarkingTask::get_work_from_overflow_stack(CMSMarkStack* ovflw_stk,
                                                      OopTaskQueue* work_q) {
  // Fast lock-free check
  if (ovflw_stk->length() == 0) {
    return false;
  }
  assert(work_q->size() == 0, "Shouldn't steal");
  MutexLockerEx ml(ovflw_stk->par_lock(),
                   Mutex::_no_safepoint_check_flag);
  // Grab up to 1/4 the size of the work queue
4087
  size_t num = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
D
duke 已提交
4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102
                    (size_t)ParGCDesiredObjsFromOverflowList);
  num = MIN2(num, ovflw_stk->length());
  for (int i = (int) num; i > 0; i--) {
    oop cur = ovflw_stk->pop();
    assert(cur != NULL, "Counted wrong?");
    work_q->push(cur);
  }
  return num > 0;
}

void CMSConcMarkingTask::do_scan_and_mark(int i, CompactibleFreeListSpace* sp) {
  SequentialSubTasksDone* pst = sp->conc_par_seq_tasks();
  int n_tasks = pst->n_tasks();
  // We allow that there may be no tasks to do here because
  // we are restarting after a stack overflow.
4103
  assert(pst->valid() || n_tasks == 0, "Uninitialized use?");
4104
  uint nth_task = 0;
D
duke 已提交
4105

4106 4107 4108 4109 4110 4111 4112 4113 4114
  HeapWord* aligned_start = sp->bottom();
  if (sp->used_region().contains(_restart_addr)) {
    // Align down to a card boundary for the start of 0th task
    // for this space.
    aligned_start =
      (HeapWord*)align_size_down((uintptr_t)_restart_addr,
                                 CardTableModRefBS::card_size);
  }

D
duke 已提交
4115 4116 4117 4118
  size_t chunk_size = sp->marking_task_size();
  while (!pst->is_task_claimed(/* reference */ nth_task)) {
    // Having claimed the nth task in this space,
    // compute the chunk that it corresponds to:
4119 4120
    MemRegion span = MemRegion(aligned_start + nth_task*chunk_size,
                               aligned_start + (nth_task+1)*chunk_size);
D
duke 已提交
4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134
    // Try and bump the global finger via a CAS;
    // note that we need to do the global finger bump
    // _before_ taking the intersection below, because
    // the task corresponding to that region will be
    // deemed done even if the used_region() expands
    // because of allocation -- as it almost certainly will
    // during start-up while the threads yield in the
    // closure below.
    HeapWord* finger = span.end();
    bump_global_finger(finger);   // atomically
    // There are null tasks here corresponding to chunks
    // beyond the "top" address of the space.
    span = span.intersection(sp->used_region());
    if (!span.is_empty()) {  // Non-null task
4135 4136 4137 4138 4139 4140 4141 4142 4143
      HeapWord* prev_obj;
      assert(!span.contains(_restart_addr) || nth_task == 0,
             "Inconsistency");
      if (nth_task == 0) {
        // For the 0th task, we'll not need to compute a block_start.
        if (span.contains(_restart_addr)) {
          // In the case of a restart because of stack overflow,
          // we might additionally skip a chunk prefix.
          prev_obj = _restart_addr;
D
duke 已提交
4144
        } else {
4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168
          prev_obj = span.start();
        }
      } else {
        // We want to skip the first object because
        // the protocol is to scan any object in its entirety
        // that _starts_ in this span; a fortiori, any
        // object starting in an earlier span is scanned
        // as part of an earlier claimed task.
        // Below we use the "careful" version of block_start
        // so we do not try to navigate uninitialized objects.
        prev_obj = sp->block_start_careful(span.start());
        // Below we use a variant of block_size that uses the
        // Printezis bits to avoid waiting for allocated
        // objects to become initialized/parsable.
        while (prev_obj < span.start()) {
          size_t sz = sp->block_size_no_stall(prev_obj, _collector);
          if (sz > 0) {
            prev_obj += sz;
          } else {
            // In this case we may end up doing a bit of redundant
            // scanning, but that appears unavoidable, short of
            // locking the free list locks; see bug 6324141.
            break;
          }
D
duke 已提交
4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195
        }
      }
      if (prev_obj < span.end()) {
        MemRegion my_span = MemRegion(prev_obj, span.end());
        // Do the marking work within a non-empty span --
        // the last argument to the constructor indicates whether the
        // iteration should be incremental with periodic yields.
        Par_MarkFromRootsClosure cl(this, _collector, my_span,
                                    &_collector->_markBitMap,
                                    work_queue(i),
                                    &_collector->_markStack,
                                    _asynch);
        _collector->_markBitMap.iterate(&cl, my_span.start(), my_span.end());
      } // else nothing to do for this task
    }   // else nothing to do for this task
  }
  // We'd be tempted to assert here that since there are no
  // more tasks left to claim in this space, the global_finger
  // must exceed space->top() and a fortiori space->end(). However,
  // that would not quite be correct because the bumping of
  // global_finger occurs strictly after the claiming of a task,
  // so by the time we reach here the global finger may not yet
  // have been bumped up by the thread that claimed the last
  // task.
  pst->all_tasks_completed();
}

4196
class Par_ConcMarkingClosure: public CMSOopClosure {
4197
 private:
4198
  CMSCollector* _collector;
4199
  CMSConcMarkingTask* _task;
D
duke 已提交
4200 4201 4202 4203
  MemRegion     _span;
  CMSBitMap*    _bit_map;
  CMSMarkStack* _overflow_stack;
  OopTaskQueue* _work_queue;
4204 4205
 protected:
  DO_OOP_WORK_DEFN
D
duke 已提交
4206
 public:
4207
  Par_ConcMarkingClosure(CMSCollector* collector, CMSConcMarkingTask* task, OopTaskQueue* work_queue,
4208 4209 4210
                         CMSBitMap* bit_map, CMSMarkStack* overflow_stack):
    CMSOopClosure(collector->ref_processor()),
    _collector(collector),
4211 4212
    _task(task),
    _span(collector->_span),
D
duke 已提交
4213 4214
    _work_queue(work_queue),
    _bit_map(bit_map),
4215 4216
    _overflow_stack(overflow_stack)
  { }
4217 4218
  virtual void do_oop(oop* p);
  virtual void do_oop(narrowOop* p);
4219

D
duke 已提交
4220 4221
  void trim_queue(size_t max);
  void handle_stack_overflow(HeapWord* lost);
4222 4223 4224 4225 4226
  void do_yield_check() {
    if (_task->should_yield()) {
      _task->yield();
    }
  }
D
duke 已提交
4227 4228
};

4229 4230 4231 4232 4233 4234
// Grey object scanning during work stealing phase --
// the salient assumption here is that any references
// that are in these stolen objects being scanned must
// already have been initialized (else they would not have
// been published), so we do not need to check for
// uninitialized objects before pushing here.
4235
void Par_ConcMarkingClosure::do_oop(oop obj) {
4236
  assert(obj->is_oop_or_null(true), "expected an oop or NULL");
4237
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255
  // Check if oop points into the CMS generation
  // and is not marked
  if (_span.contains(addr) && !_bit_map->isMarked(addr)) {
    // a white object ...
    // If we manage to "claim" the object, by being the
    // first thread to mark it, then we push it on our
    // marking stack
    if (_bit_map->par_mark(addr)) {     // ... now grey
      // push on work queue (grey set)
      bool simulate_overflow = false;
      NOT_PRODUCT(
        if (CMSMarkStackOverflowALot &&
            _collector->simulate_overflow()) {
          // simulate a stack overflow
          simulate_overflow = true;
        }
      )
      if (simulate_overflow ||
4256
          !(_work_queue->push(obj) || _overflow_stack->par_push(obj))) {
D
duke 已提交
4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269
        // stack overflow
        if (PrintCMSStatistics != 0) {
          gclog_or_tty->print_cr("CMS marking stack overflow (benign) at "
                                 SIZE_FORMAT, _overflow_stack->capacity());
        }
        // We cannot assert that the overflow stack is full because
        // it may have been emptied since.
        assert(simulate_overflow ||
               _work_queue->size() == _work_queue->max_elems(),
              "Else push should have succeeded");
        handle_stack_overflow(addr);
      }
    } // Else, some other thread got there first
4270
    do_yield_check();
D
duke 已提交
4271 4272 4273
  }
}

4274 4275 4276
void Par_ConcMarkingClosure::do_oop(oop* p)       { Par_ConcMarkingClosure::do_oop_work(p); }
void Par_ConcMarkingClosure::do_oop(narrowOop* p) { Par_ConcMarkingClosure::do_oop_work(p); }

D
duke 已提交
4277 4278 4279 4280 4281 4282 4283 4284
void Par_ConcMarkingClosure::trim_queue(size_t max) {
  while (_work_queue->size() > max) {
    oop new_oop;
    if (_work_queue->pop_local(new_oop)) {
      assert(new_oop->is_oop(), "Should be an oop");
      assert(_bit_map->isMarked((HeapWord*)new_oop), "Grey object");
      assert(_span.contains((HeapWord*)new_oop), "Not in span");
      new_oop->oop_iterate(this);  // do_oop() above
4285
      do_yield_check();
D
duke 已提交
4286 4287 4288 4289 4290 4291 4292 4293 4294
    }
  }
}

// Upon stack overflow, we discard (part of) the stack,
// remembering the least address amongst those discarded
// in CMSCollector's _restart_address.
void Par_ConcMarkingClosure::handle_stack_overflow(HeapWord* lost) {
  // We need to do this under a mutex to prevent other
4295
  // workers from interfering with the work done below.
D
duke 已提交
4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311
  MutexLockerEx ml(_overflow_stack->par_lock(),
                   Mutex::_no_safepoint_check_flag);
  // Remember the least grey address discarded
  HeapWord* ra = (HeapWord*)_overflow_stack->least_value(lost);
  _collector->lower_restart_addr(ra);
  _overflow_stack->reset();  // discard stack contents
  _overflow_stack->expand(); // expand the stack if possible
}


void CMSConcMarkingTask::do_work_steal(int i) {
  OopTaskQueue* work_q = work_queue(i);
  oop obj_to_scan;
  CMSBitMap* bm = &(_collector->_markBitMap);
  CMSMarkStack* ovflw = &(_collector->_markStack);
  int* seed = _collector->hash_seed(i);
4312
  Par_ConcMarkingClosure cl(_collector, this, work_q, bm, ovflw);
D
duke 已提交
4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324
  while (true) {
    cl.trim_queue(0);
    assert(work_q->size() == 0, "Should have been emptied above");
    if (get_work_from_overflow_stack(ovflw, work_q)) {
      // Can't assert below because the work obtained from the
      // overflow stack may already have been stolen from us.
      // assert(work_q->size() > 0, "Work from overflow stack");
      continue;
    } else if (task_queues()->steal(i, seed, /* reference */ obj_to_scan)) {
      assert(obj_to_scan->is_oop(), "Should be an oop");
      assert(bm->isMarked((HeapWord*)obj_to_scan), "Grey object");
      obj_to_scan->oop_iterate(&cl);
4325
    } else if (terminator()->offer_termination(&_term_term)) {
D
duke 已提交
4326 4327
      assert(work_q->size() == 0, "Impossible!");
      break;
4328 4329
    } else if (yielding() || should_yield()) {
      yield();
D
duke 已提交
4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376
    }
  }
}

// This is run by the CMS (coordinator) thread.
void CMSConcMarkingTask::coordinator_yield() {
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  // First give up the locks, then yield, then re-lock
  // We should probably use a constructor/destructor idiom to
  // do this unlock/lock or modify the MutexUnlocker class to
  // serve our purpose. XXX
  assert_lock_strong(_bit_map_lock);
  _bit_map_lock->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // It is possible for whichever thread initiated the yield request
  // not to get a chance to wake up and take the bitmap lock between
  // this thread releasing it and reacquiring it. So, while the
  // should_yield() flag is on, let's sleep for a bit to give the
  // other thread a chance to wake up. The limit imposed on the number
  // of iterations is defensive, to avoid any unforseen circumstances
  // putting us into an infinite loop. Since it's always been this
  // (coordinator_yield()) method that was observed to cause the
  // problem, we are using a parameter (CMSCoordinatorYieldSleepCount)
  // which is by default non-zero. For the other seven methods that
  // also perform the yield operation, as are using a different
  // parameter (CMSYieldSleepCount) which is by default zero. This way we
  // can enable the sleeping for those methods too, if necessary.
  // See 6442774.
  //
  // We really need to reconsider the synchronization between the GC
  // thread and the yield-requesting threads in the future and we
  // should really use wait/notify, which is the recommended
  // way of doing this type of interaction. Additionally, we should
  // consolidate the eight methods that do the yield operation and they
  // are almost identical into one for better maintenability and
  // readability. See 6445193.
  //
  // Tony 2006.06.29
  for (unsigned i = 0; i < CMSCoordinatorYieldSleepCount &&
4377 4378
                   ConcurrentMarkSweepThread::should_yield() &&
                   !CMSCollector::foregroundGCIsActive(); ++i) {
D
duke 已提交
4379 4380 4381 4382 4383 4384 4385 4386 4387 4388
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _bit_map_lock->lock_without_safepoint_check();
  _collector->startTimer();
}

bool CMSCollector::do_marking_mt(bool asynch) {
4389
  assert(ConcGCThreads > 0 && conc_workers() != NULL, "precondition");
4390 4391 4392 4393 4394
  int num_workers = AdaptiveSizePolicy::calc_active_conc_workers(
                                       conc_workers()->total_workers(),
                                       conc_workers()->active_workers(),
                                       Threads::number_of_non_daemon_threads());
  conc_workers()->set_active_workers(num_workers);
D
duke 已提交
4395 4396 4397

  CompactibleFreeListSpace* cms_space  = _cmsGen->cmsSpace();

4398 4399 4400 4401 4402
  CMSConcMarkingTask tsk(this,
                         cms_space,
                         asynch,
                         conc_workers(),
                         task_queues());
D
duke 已提交
4403 4404 4405 4406 4407 4408 4409 4410 4411

  // Since the actual number of workers we get may be different
  // from the number we requested above, do we need to do anything different
  // below? In particular, may be we need to subclass the SequantialSubTasksDone
  // class?? XXX
  cms_space ->initialize_sequential_subtasks_for_marking(num_workers);

  // Refs discovery is already non-atomic.
  assert(!ref_processor()->discovery_is_atomic(), "Should be non-atomic");
4412
  assert(ref_processor()->discovery_is_mt(), "Discovery should be MT");
D
duke 已提交
4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426
  conc_workers()->start_task(&tsk);
  while (tsk.yielded()) {
    tsk.coordinator_yield();
    conc_workers()->continue_task(&tsk);
  }
  // If the task was aborted, _restart_addr will be non-NULL
  assert(tsk.completed() || _restart_addr != NULL, "Inconsistency");
  while (_restart_addr != NULL) {
    // XXX For now we do not make use of ABORTED state and have not
    // yet implemented the right abort semantics (even in the original
    // single-threaded CMS case). That needs some more investigation
    // and is deferred for now; see CR# TBF. 07252005YSR. XXX
    assert(!CMSAbortSemantics || tsk.aborted(), "Inconsistency");
    // If _restart_addr is non-NULL, a marking stack overflow
T
twisti 已提交
4427
    // occurred; we need to do a fresh marking iteration from the
D
duke 已提交
4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461
    // indicated restart address.
    if (_foregroundGCIsActive && asynch) {
      // We may be running into repeated stack overflows, having
      // reached the limit of the stack size, while making very
      // slow forward progress. It may be best to bail out and
      // let the foreground collector do its job.
      // Clear _restart_addr, so that foreground GC
      // works from scratch. This avoids the headache of
      // a "rescan" which would otherwise be needed because
      // of the dirty mod union table & card table.
      _restart_addr = NULL;
      return false;
    }
    // Adjust the task to restart from _restart_addr
    tsk.reset(_restart_addr);
    cms_space ->initialize_sequential_subtasks_for_marking(num_workers,
                  _restart_addr);
    _restart_addr = NULL;
    // Get the workers going again
    conc_workers()->start_task(&tsk);
    while (tsk.yielded()) {
      tsk.coordinator_yield();
      conc_workers()->continue_task(&tsk);
    }
  }
  assert(tsk.completed(), "Inconsistency");
  assert(tsk.result() == true, "Inconsistency");
  return true;
}

bool CMSCollector::do_marking_st(bool asynch) {
  ResourceMark rm;
  HandleMark   hm;

4462 4463
  // Temporarily make refs discovery single threaded (non-MT)
  ReferenceProcessorMTDiscoveryMutator rp_mut_discovery(ref_processor(), false);
D
duke 已提交
4464
  MarkFromRootsClosure markFromRootsClosure(this, _span, &_markBitMap,
4465
    &_markStack, CMSYield && asynch);
D
duke 已提交
4466 4467 4468 4469
  // the last argument to iterate indicates whether the iteration
  // should be incremental with periodic yields.
  _markBitMap.iterate(&markFromRootsClosure);
  // If _restart_addr is non-NULL, a marking stack overflow
T
twisti 已提交
4470
  // occurred; we need to do a fresh iteration from the
D
duke 已提交
4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501
  // indicated restart address.
  while (_restart_addr != NULL) {
    if (_foregroundGCIsActive && asynch) {
      // We may be running into repeated stack overflows, having
      // reached the limit of the stack size, while making very
      // slow forward progress. It may be best to bail out and
      // let the foreground collector do its job.
      // Clear _restart_addr, so that foreground GC
      // works from scratch. This avoids the headache of
      // a "rescan" which would otherwise be needed because
      // of the dirty mod union table & card table.
      _restart_addr = NULL;
      return false;  // indicating failure to complete marking
    }
    // Deal with stack overflow:
    // we restart marking from _restart_addr
    HeapWord* ra = _restart_addr;
    markFromRootsClosure.reset(ra);
    _restart_addr = NULL;
    _markBitMap.iterate(&markFromRootsClosure, ra, _span.end());
  }
  return true;
}

void CMSCollector::preclean() {
  check_correct_thread_executing();
  assert(Thread::current()->is_ConcurrentGC_thread(), "Wrong thread");
  verify_work_stacks_empty();
  verify_overflow_empty();
  _abort_preclean = false;
  if (CMSPrecleaningEnabled) {
4502 4503 4504
    if (!CMSEdenChunksRecordAlways) {
      _eden_chunk_index = 0;
    }
D
duke 已提交
4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612
    size_t used = get_eden_used();
    size_t capacity = get_eden_capacity();
    // Don't start sampling unless we will get sufficiently
    // many samples.
    if (used < (capacity/(CMSScheduleRemarkSamplingRatio * 100)
                * CMSScheduleRemarkEdenPenetration)) {
      _start_sampling = true;
    } else {
      _start_sampling = false;
    }
    TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
    CMSPhaseAccounting pa(this, "preclean", !PrintGCDetails);
    preclean_work(CMSPrecleanRefLists1, CMSPrecleanSurvivors1);
  }
  CMSTokenSync x(true); // is cms thread
  if (CMSPrecleaningEnabled) {
    sample_eden();
    _collectorState = AbortablePreclean;
  } else {
    _collectorState = FinalMarking;
  }
  verify_work_stacks_empty();
  verify_overflow_empty();
}

// Try and schedule the remark such that young gen
// occupancy is CMSScheduleRemarkEdenPenetration %.
void CMSCollector::abortable_preclean() {
  check_correct_thread_executing();
  assert(CMSPrecleaningEnabled,  "Inconsistent control state");
  assert(_collectorState == AbortablePreclean, "Inconsistent control state");

  // If Eden's current occupancy is below this threshold,
  // immediately schedule the remark; else preclean
  // past the next scavenge in an effort to
  // schedule the pause as described avove. By choosing
  // CMSScheduleRemarkEdenSizeThreshold >= max eden size
  // we will never do an actual abortable preclean cycle.
  if (get_eden_used() > CMSScheduleRemarkEdenSizeThreshold) {
    TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
    CMSPhaseAccounting pa(this, "abortable-preclean", !PrintGCDetails);
    // We need more smarts in the abortable preclean
    // loop below to deal with cases where allocation
    // in young gen is very very slow, and our precleaning
    // is running a losing race against a horde of
    // mutators intent on flooding us with CMS updates
    // (dirty cards).
    // One, admittedly dumb, strategy is to give up
    // after a certain number of abortable precleaning loops
    // or after a certain maximum time. We want to make
    // this smarter in the next iteration.
    // XXX FIX ME!!! YSR
    size_t loops = 0, workdone = 0, cumworkdone = 0, waited = 0;
    while (!(should_abort_preclean() ||
             ConcurrentMarkSweepThread::should_terminate())) {
      workdone = preclean_work(CMSPrecleanRefLists2, CMSPrecleanSurvivors2);
      cumworkdone += workdone;
      loops++;
      // Voluntarily terminate abortable preclean phase if we have
      // been at it for too long.
      if ((CMSMaxAbortablePrecleanLoops != 0) &&
          loops >= CMSMaxAbortablePrecleanLoops) {
        if (PrintGCDetails) {
          gclog_or_tty->print(" CMS: abort preclean due to loops ");
        }
        break;
      }
      if (pa.wallclock_millis() > CMSMaxAbortablePrecleanTime) {
        if (PrintGCDetails) {
          gclog_or_tty->print(" CMS: abort preclean due to time ");
        }
        break;
      }
      // If we are doing little work each iteration, we should
      // take a short break.
      if (workdone < CMSAbortablePrecleanMinWorkPerIteration) {
        // Sleep for some time, waiting for work to accumulate
        stopTimer();
        cmsThread()->wait_on_cms_lock(CMSAbortablePrecleanWaitMillis);
        startTimer();
        waited++;
      }
    }
    if (PrintCMSStatistics > 0) {
      gclog_or_tty->print(" [%d iterations, %d waits, %d cards)] ",
                          loops, waited, cumworkdone);
    }
  }
  CMSTokenSync x(true); // is cms thread
  if (_collectorState != Idling) {
    assert(_collectorState == AbortablePreclean,
           "Spontaneous state transition?");
    _collectorState = FinalMarking;
  } // Else, a foreground collection completed this CMS cycle.
  return;
}

// Respond to an Eden sampling opportunity
void CMSCollector::sample_eden() {
  // Make sure a young gc cannot sneak in between our
  // reading and recording of a sample.
  assert(Thread::current()->is_ConcurrentGC_thread(),
         "Only the cms thread may collect Eden samples");
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "Should collect samples while holding CMS token");
  if (!_start_sampling) {
    return;
  }
4613 4614 4615
  // When CMSEdenChunksRecordAlways is true, the eden chunk array
  // is populated by the young generation.
  if (_eden_chunk_array != NULL && !CMSEdenChunksRecordAlways) {
D
duke 已提交
4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647
    if (_eden_chunk_index < _eden_chunk_capacity) {
      _eden_chunk_array[_eden_chunk_index] = *_top_addr;   // take sample
      assert(_eden_chunk_array[_eden_chunk_index] <= *_end_addr,
             "Unexpected state of Eden");
      // We'd like to check that what we just sampled is an oop-start address;
      // however, we cannot do that here since the object may not yet have been
      // initialized. So we'll instead do the check when we _use_ this sample
      // later.
      if (_eden_chunk_index == 0 ||
          (pointer_delta(_eden_chunk_array[_eden_chunk_index],
                         _eden_chunk_array[_eden_chunk_index-1])
           >= CMSSamplingGrain)) {
        _eden_chunk_index++;  // commit sample
      }
    }
  }
  if ((_collectorState == AbortablePreclean) && !_abort_preclean) {
    size_t used = get_eden_used();
    size_t capacity = get_eden_capacity();
    assert(used <= capacity, "Unexpected state of Eden");
    if (used >  (capacity/100 * CMSScheduleRemarkEdenPenetration)) {
      _abort_preclean = true;
    }
  }
}


size_t CMSCollector::preclean_work(bool clean_refs, bool clean_survivor) {
  assert(_collectorState == Precleaning ||
         _collectorState == AbortablePreclean, "incorrect state");
  ResourceMark rm;
  HandleMark   hm;
4648 4649 4650 4651 4652 4653

  // Precleaning is currently not MT but the reference processor
  // may be set for MT.  Disable it temporarily here.
  ReferenceProcessor* rp = ref_processor();
  ReferenceProcessorMTDiscoveryMutator rp_mut_discovery(rp, false);

D
duke 已提交
4654 4655 4656 4657 4658 4659 4660
  // Do one pass of scrubbing the discovered reference lists
  // to remove any reference objects with strongly-reachable
  // referents.
  if (clean_refs) {
    CMSPrecleanRefsYieldClosure yield_cl(this);
    assert(rp->span().equals(_span), "Spans should be equal");
    CMSKeepAliveClosure keep_alive(this, _span, &_markBitMap,
4661
                                   &_markStack, true /* preclean */);
D
duke 已提交
4662
    CMSDrainMarkingStackClosure complete_trace(this,
4663 4664
                                   _span, &_markBitMap, &_markStack,
                                   &keep_alive, true /* preclean */);
D
duke 已提交
4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680

    // We don't want this step to interfere with a young
    // collection because we don't want to take CPU
    // or memory bandwidth away from the young GC threads
    // (which may be as many as there are CPUs).
    // Note that we don't need to protect ourselves from
    // interference with mutators because they can't
    // manipulate the discovered reference lists nor affect
    // the computed reachability of the referents, the
    // only properties manipulated by the precleaning
    // of these reference lists.
    stopTimer();
    CMSTokenSyncWithLocks x(true /* is cms thread */,
                            bitMapLock());
    startTimer();
    sample_eden();
4681

D
duke 已提交
4682 4683 4684 4685 4686
    // The following will yield to allow foreground
    // collection to proceed promptly. XXX YSR:
    // The code in this method may need further
    // tweaking for better performance and some restructuring
    // for cleaner interfaces.
S
sla 已提交
4687
    GCTimer *gc_timer = NULL; // Currently not tracing concurrent phases
D
duke 已提交
4688
    rp->preclean_discovered_references(
S
sla 已提交
4689 4690
          rp->is_alive_non_header(), &keep_alive, &complete_trace, &yield_cl,
          gc_timer);
D
duke 已提交
4691 4692 4693 4694 4695 4696 4697 4698 4699 4700
  }

  if (clean_survivor) {  // preclean the active survivor space(s)
    assert(_young_gen->kind() == Generation::DefNew ||
           _young_gen->kind() == Generation::ParNew ||
           _young_gen->kind() == Generation::ASParNew,
         "incorrect type for cast");
    DefNewGeneration* dng = (DefNewGeneration*)_young_gen;
    PushAndMarkClosure pam_cl(this, _span, ref_processor(),
                             &_markBitMap, &_modUnionTable,
4701
                             &_markStack, true /* precleaning phase */);
D
duke 已提交
4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715
    stopTimer();
    CMSTokenSyncWithLocks ts(true /* is cms thread */,
                             bitMapLock());
    startTimer();
    unsigned int before_count =
      GenCollectedHeap::heap()->total_collections();
    SurvivorSpacePrecleanClosure
      sss_cl(this, _span, &_markBitMap, &_markStack,
             &pam_cl, before_count, CMSYield);
    dng->from()->object_iterate_careful(&sss_cl);
    dng->to()->object_iterate_careful(&sss_cl);
  }
  MarkRefsIntoAndScanClosure
    mrias_cl(_span, ref_processor(), &_markBitMap, &_modUnionTable,
4716
             &_markStack, this, CMSYield,
D
duke 已提交
4717 4718 4719 4720 4721 4722
             true /* precleaning phase */);
  // CAUTION: The following closure has persistent state that may need to
  // be reset upon a decrease in the sequence of addresses it
  // processes.
  ScanMarkedObjectsAgainCarefullyClosure
    smoac_cl(this, _span,
4723
      &_markBitMap, &_markStack, &mrias_cl, CMSYield);
D
duke 已提交
4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756

  // Preclean dirty cards in ModUnionTable and CardTable using
  // appropriate convergence criterion;
  // repeat CMSPrecleanIter times unless we find that
  // we are losing.
  assert(CMSPrecleanIter < 10, "CMSPrecleanIter is too large");
  assert(CMSPrecleanNumerator < CMSPrecleanDenominator,
         "Bad convergence multiplier");
  assert(CMSPrecleanThreshold >= 100,
         "Unreasonably low CMSPrecleanThreshold");

  size_t numIter, cumNumCards, lastNumCards, curNumCards;
  for (numIter = 0, cumNumCards = lastNumCards = curNumCards = 0;
       numIter < CMSPrecleanIter;
       numIter++, lastNumCards = curNumCards, cumNumCards += curNumCards) {
    curNumCards  = preclean_mod_union_table(_cmsGen, &smoac_cl);
    if (Verbose && PrintGCDetails) {
      gclog_or_tty->print(" (modUnionTable: %d cards)", curNumCards);
    }
    // Either there are very few dirty cards, so re-mark
    // pause will be small anyway, or our pre-cleaning isn't
    // that much faster than the rate at which cards are being
    // dirtied, so we might as well stop and re-mark since
    // precleaning won't improve our re-mark time by much.
    if (curNumCards <= CMSPrecleanThreshold ||
        (numIter > 0 &&
         (curNumCards * CMSPrecleanDenominator >
         lastNumCards * CMSPrecleanNumerator))) {
      numIter++;
      cumNumCards += curNumCards;
      break;
    }
  }
4757 4758 4759

  preclean_klasses(&mrias_cl, _cmsGen->freelistLock());

D
duke 已提交
4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836
  curNumCards = preclean_card_table(_cmsGen, &smoac_cl);
  cumNumCards += curNumCards;
  if (PrintGCDetails && PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr(" (cardTable: %d cards, re-scanned %d cards, %d iterations)",
                  curNumCards, cumNumCards, numIter);
  }
  return cumNumCards;   // as a measure of useful work done
}

// PRECLEANING NOTES:
// Precleaning involves:
// . reading the bits of the modUnionTable and clearing the set bits.
// . For the cards corresponding to the set bits, we scan the
//   objects on those cards. This means we need the free_list_lock
//   so that we can safely iterate over the CMS space when scanning
//   for oops.
// . When we scan the objects, we'll be both reading and setting
//   marks in the marking bit map, so we'll need the marking bit map.
// . For protecting _collector_state transitions, we take the CGC_lock.
//   Note that any races in the reading of of card table entries by the
//   CMS thread on the one hand and the clearing of those entries by the
//   VM thread or the setting of those entries by the mutator threads on the
//   other are quite benign. However, for efficiency it makes sense to keep
//   the VM thread from racing with the CMS thread while the latter is
//   dirty card info to the modUnionTable. We therefore also use the
//   CGC_lock to protect the reading of the card table and the mod union
//   table by the CM thread.
// . We run concurrently with mutator updates, so scanning
//   needs to be done carefully  -- we should not try to scan
//   potentially uninitialized objects.
//
// Locking strategy: While holding the CGC_lock, we scan over and
// reset a maximal dirty range of the mod union / card tables, then lock
// the free_list_lock and bitmap lock to do a full marking, then
// release these locks; and repeat the cycle. This allows for a
// certain amount of fairness in the sharing of these locks between
// the CMS collector on the one hand, and the VM thread and the
// mutators on the other.

// NOTE: preclean_mod_union_table() and preclean_card_table()
// further below are largely identical; if you need to modify
// one of these methods, please check the other method too.

size_t CMSCollector::preclean_mod_union_table(
  ConcurrentMarkSweepGeneration* gen,
  ScanMarkedObjectsAgainCarefullyClosure* cl) {
  verify_work_stacks_empty();
  verify_overflow_empty();

  // strategy: starting with the first card, accumulate contiguous
  // ranges of dirty cards; clear these cards, then scan the region
  // covered by these cards.

  // Since all of the MUT is committed ahead, we can just use
  // that, in case the generations expand while we are precleaning.
  // It might also be fine to just use the committed part of the
  // generation, but we might potentially miss cards when the
  // generation is rapidly expanding while we are in the midst
  // of precleaning.
  HeapWord* startAddr = gen->reserved().start();
  HeapWord* endAddr   = gen->reserved().end();

  cl->setFreelistLock(gen->freelistLock());   // needed for yielding

  size_t numDirtyCards, cumNumDirtyCards;
  HeapWord *nextAddr, *lastAddr;
  for (cumNumDirtyCards = numDirtyCards = 0,
       nextAddr = lastAddr = startAddr;
       nextAddr < endAddr;
       nextAddr = lastAddr, cumNumDirtyCards += numDirtyCards) {

    ResourceMark rm;
    HandleMark   hm;

    MemRegion dirtyRegion;
    {
      stopTimer();
4837
      // Potential yield point
D
duke 已提交
4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861
      CMSTokenSync ts(true);
      startTimer();
      sample_eden();
      // Get dirty region starting at nextOffset (inclusive),
      // simultaneously clearing it.
      dirtyRegion =
        _modUnionTable.getAndClearMarkedRegion(nextAddr, endAddr);
      assert(dirtyRegion.start() >= nextAddr,
             "returned region inconsistent?");
    }
    // Remember where the next search should begin.
    // The returned region (if non-empty) is a right open interval,
    // so lastOffset is obtained from the right end of that
    // interval.
    lastAddr = dirtyRegion.end();
    // Should do something more transparent and less hacky XXX
    numDirtyCards =
      _modUnionTable.heapWordDiffToOffsetDiff(dirtyRegion.word_size());

    // We'll scan the cards in the dirty region (with periodic
    // yields for foreground GC as needed).
    if (!dirtyRegion.is_empty()) {
      assert(numDirtyCards > 0, "consistency check");
      HeapWord* stop_point = NULL;
4862
      stopTimer();
4863
      // Potential yield point
4864 4865 4866
      CMSTokenSyncWithLocks ts(true, gen->freelistLock(),
                               bitMapLock());
      startTimer();
D
duke 已提交
4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880
      {
        verify_work_stacks_empty();
        verify_overflow_empty();
        sample_eden();
        stop_point =
          gen->cmsSpace()->object_iterate_careful_m(dirtyRegion, cl);
      }
      if (stop_point != NULL) {
        // The careful iteration stopped early either because it found an
        // uninitialized object, or because we were in the midst of an
        // "abortable preclean", which should now be aborted. Redirty
        // the bits corresponding to the partially-scanned or unscanned
        // cards. We'll either restart at the next block boundary or
        // abort the preclean.
4881 4882
        assert((_collectorState == AbortablePreclean && should_abort_preclean()),
               "Should only be AbortablePreclean.");
D
duke 已提交
4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937
        _modUnionTable.mark_range(MemRegion(stop_point, dirtyRegion.end()));
        if (should_abort_preclean()) {
          break; // out of preclean loop
        } else {
          // Compute the next address at which preclean should pick up;
          // might need bitMapLock in order to read P-bits.
          lastAddr = next_card_start_after_block(stop_point);
        }
      }
    } else {
      assert(lastAddr == endAddr, "consistency check");
      assert(numDirtyCards == 0, "consistency check");
      break;
    }
  }
  verify_work_stacks_empty();
  verify_overflow_empty();
  return cumNumDirtyCards;
}

// NOTE: preclean_mod_union_table() above and preclean_card_table()
// below are largely identical; if you need to modify
// one of these methods, please check the other method too.

size_t CMSCollector::preclean_card_table(ConcurrentMarkSweepGeneration* gen,
  ScanMarkedObjectsAgainCarefullyClosure* cl) {
  // strategy: it's similar to precleamModUnionTable above, in that
  // we accumulate contiguous ranges of dirty cards, mark these cards
  // precleaned, then scan the region covered by these cards.
  HeapWord* endAddr   = (HeapWord*)(gen->_virtual_space.high());
  HeapWord* startAddr = (HeapWord*)(gen->_virtual_space.low());

  cl->setFreelistLock(gen->freelistLock());   // needed for yielding

  size_t numDirtyCards, cumNumDirtyCards;
  HeapWord *lastAddr, *nextAddr;

  for (cumNumDirtyCards = numDirtyCards = 0,
       nextAddr = lastAddr = startAddr;
       nextAddr < endAddr;
       nextAddr = lastAddr, cumNumDirtyCards += numDirtyCards) {

    ResourceMark rm;
    HandleMark   hm;

    MemRegion dirtyRegion;
    {
      // See comments in "Precleaning notes" above on why we
      // do this locking. XXX Could the locking overheads be
      // too high when dirty cards are sparse? [I don't think so.]
      stopTimer();
      CMSTokenSync x(true); // is cms thread
      startTimer();
      sample_eden();
      // Get and clear dirty region from card table
4938 4939 4940 4941 4942
      dirtyRegion = _ct->ct_bs()->dirty_card_range_after_reset(
                                    MemRegion(nextAddr, endAddr),
                                    true,
                                    CardTableModRefBS::precleaned_card_val());

D
duke 已提交
4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959
      assert(dirtyRegion.start() >= nextAddr,
             "returned region inconsistent?");
    }
    lastAddr = dirtyRegion.end();
    numDirtyCards =
      dirtyRegion.word_size()/CardTableModRefBS::card_size_in_words;

    if (!dirtyRegion.is_empty()) {
      stopTimer();
      CMSTokenSyncWithLocks ts(true, gen->freelistLock(), bitMapLock());
      startTimer();
      sample_eden();
      verify_work_stacks_empty();
      verify_overflow_empty();
      HeapWord* stop_point =
        gen->cmsSpace()->object_iterate_careful_m(dirtyRegion, cl);
      if (stop_point != NULL) {
4960 4961
        assert((_collectorState == AbortablePreclean && should_abort_preclean()),
               "Should only be AbortablePreclean.");
D
duke 已提交
4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978
        _ct->ct_bs()->invalidate(MemRegion(stop_point, dirtyRegion.end()));
        if (should_abort_preclean()) {
          break; // out of preclean loop
        } else {
          // Compute the next address at which preclean should pick up.
          lastAddr = next_card_start_after_block(stop_point);
        }
      }
    } else {
      break;
    }
  }
  verify_work_stacks_empty();
  verify_overflow_empty();
  return cumNumDirtyCards;
}

4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007
class PrecleanKlassClosure : public KlassClosure {
  CMKlassClosure _cm_klass_closure;
 public:
  PrecleanKlassClosure(OopClosure* oop_closure) : _cm_klass_closure(oop_closure) {}
  void do_klass(Klass* k) {
    if (k->has_accumulated_modified_oops()) {
      k->clear_accumulated_modified_oops();

      _cm_klass_closure.do_klass(k);
    }
  }
};

// The freelist lock is needed to prevent asserts, is it really needed?
void CMSCollector::preclean_klasses(MarkRefsIntoAndScanClosure* cl, Mutex* freelistLock) {

  cl->set_freelistLock(freelistLock);

  CMSTokenSyncWithLocks ts(true, freelistLock, bitMapLock());

  // SSS: Add equivalent to ScanMarkedObjectsAgainCarefullyClosure::do_yield_check and should_abort_preclean?
  // SSS: We should probably check if precleaning should be aborted, at suitable intervals?
  PrecleanKlassClosure preclean_klass_closure(cl);
  ClassLoaderDataGraph::classes_do(&preclean_klass_closure);

  verify_work_stacks_empty();
  verify_overflow_empty();
}

D
duke 已提交
5008 5009 5010 5011 5012 5013 5014
void CMSCollector::checkpointRootsFinal(bool asynch,
  bool clear_all_soft_refs, bool init_mark_was_synchronous) {
  assert(_collectorState == FinalMarking, "incorrect state transition?");
  check_correct_thread_executing();
  // world is stopped at this checkpoint
  assert(SafepointSynchronize::is_at_safepoint(),
         "world should be stopped");
5015 5016
  TraceCMSMemoryManagerStats tms(_collectorState,GenCollectedHeap::heap()->gc_cause());

D
duke 已提交
5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031
  verify_work_stacks_empty();
  verify_overflow_empty();

  SpecializationStats::clear();
  if (PrintGCDetails) {
    gclog_or_tty->print("[YG occupancy: "SIZE_FORMAT" K ("SIZE_FORMAT" K)]",
                        _young_gen->used() / K,
                        _young_gen->capacity() / K);
  }
  if (asynch) {
    if (CMSScavengeBeforeRemark) {
      GenCollectedHeap* gch = GenCollectedHeap::heap();
      // Temporarily set flag to false, GCH->do_collection will
      // expect it to be false and set to true
      FlagSetting fl(gch->_is_gc_active, false);
S
sla 已提交
5032 5033
      NOT_PRODUCT(GCTraceTime t("Scavenge-Before-Remark",
        PrintGCDetails && Verbose, true, _gc_timer_cm);)
D
duke 已提交
5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061
      int level = _cmsGen->level() - 1;
      if (level >= 0) {
        gch->do_collection(true,        // full (i.e. force, see below)
                           false,       // !clear_all_soft_refs
                           0,           // size
                           false,       // is_tlab
                           level        // max_level
                          );
      }
    }
    FreelistLocker x(this);
    MutexLockerEx y(bitMapLock(),
                    Mutex::_no_safepoint_check_flag);
    assert(!init_mark_was_synchronous, "but that's impossible!");
    checkpointRootsFinalWork(asynch, clear_all_soft_refs, false);
  } else {
    // already have all the locks
    checkpointRootsFinalWork(asynch, clear_all_soft_refs,
                             init_mark_was_synchronous);
  }
  verify_work_stacks_empty();
  verify_overflow_empty();
  SpecializationStats::print();
}

void CMSCollector::checkpointRootsFinalWork(bool asynch,
  bool clear_all_soft_refs, bool init_mark_was_synchronous) {

S
sla 已提交
5062
  NOT_PRODUCT(GCTraceTime tr("checkpointRootsFinalWork", PrintGCDetails, false, _gc_timer_cm);)
D
duke 已提交
5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075

  assert(haveFreelistLocks(), "must have free list locks");
  assert_lock_strong(bitMapLock());

  if (UseAdaptiveSizePolicy) {
    size_policy()->checkpoint_roots_final_begin();
  }

  ResourceMark rm;
  HandleMark   hm;

  GenCollectedHeap* gch = GenCollectedHeap::heap();

5076
  if (should_unload_classes()) {
D
duke 已提交
5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098
    CodeCache::gc_prologue();
  }
  assert(haveFreelistLocks(), "must have free list locks");
  assert_lock_strong(bitMapLock());

  if (!init_mark_was_synchronous) {
    // We might assume that we need not fill TLAB's when
    // CMSScavengeBeforeRemark is set, because we may have just done
    // a scavenge which would have filled all TLAB's -- and besides
    // Eden would be empty. This however may not always be the case --
    // for instance although we asked for a scavenge, it may not have
    // happened because of a JNI critical section. We probably need
    // a policy for deciding whether we can in that case wait until
    // the critical section releases and then do the remark following
    // the scavenge, and skip it here. In the absence of that policy,
    // or of an indication of whether the scavenge did indeed occur,
    // we cannot rely on TLAB's having been filled and must do
    // so here just in case a scavenge did not happen.
    gch->ensure_parsability(false);  // fill TLAB's, but no need to retire them
    // Update the saved marks which may affect the root scans.
    gch->save_marks();

5099 5100 5101 5102
    if (CMSPrintEdenSurvivorChunks) {
      print_eden_and_survivor_chunk_arrays();
    }

D
duke 已提交
5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115
    {
      COMPILER2_PRESENT(DerivedPointerTableDeactivate dpt_deact;)

      // Note on the role of the mod union table:
      // Since the marker in "markFromRoots" marks concurrently with
      // mutators, it is possible for some reachable objects not to have been
      // scanned. For instance, an only reference to an object A was
      // placed in object B after the marker scanned B. Unless B is rescanned,
      // A would be collected. Such updates to references in marked objects
      // are detected via the mod union table which is the set of all cards
      // dirtied since the first checkpoint in this GC cycle and prior to
      // the most recent young generation GC, minus those cleaned up by the
      // concurrent precleaning.
5116
      if (CMSParallelRemarkEnabled && CollectedHeap::use_parallel_gc_threads()) {
S
sla 已提交
5117
        GCTraceTime t("Rescan (parallel) ", PrintGCDetails, false, _gc_timer_cm);
D
duke 已提交
5118 5119
        do_remark_parallel();
      } else {
S
sla 已提交
5120 5121
        GCTraceTime t("Rescan (non-parallel) ", PrintGCDetails, false,
                    _gc_timer_cm);
D
duke 已提交
5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133
        do_remark_non_parallel();
      }
    }
  } else {
    assert(!asynch, "Can't have init_mark_was_synchronous in asynch mode");
    // The initial mark was stop-world, so there's no rescanning to
    // do; go straight on to the next step below.
  }
  verify_work_stacks_empty();
  verify_overflow_empty();

  {
S
sla 已提交
5134
    NOT_PRODUCT(GCTraceTime ts("refProcessingWork", PrintGCDetails, false, _gc_timer_cm);)
D
duke 已提交
5135 5136 5137 5138 5139
    refProcessingWork(asynch, clear_all_soft_refs);
  }
  verify_work_stacks_empty();
  verify_overflow_empty();

5140
  if (should_unload_classes()) {
D
duke 已提交
5141 5142
    CodeCache::gc_epilogue();
  }
5143
  JvmtiExport::gc_epilogue();
D
duke 已提交
5144 5145 5146 5147 5148 5149 5150

  // If we encountered any (marking stack / work queue) overflow
  // events during the current CMS cycle, take appropriate
  // remedial measures, where possible, so as to try and avoid
  // recurrence of that condition.
  assert(_markStack.isEmpty(), "No grey objects");
  size_t ser_ovflw = _ser_pmc_remark_ovflw + _ser_pmc_preclean_ovflw +
5151
                     _ser_kac_ovflw        + _ser_kac_preclean_ovflw;
D
duke 已提交
5152 5153 5154
  if (ser_ovflw > 0) {
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr("Marking stack overflow (benign) "
5155 5156
        "(pmc_pc="SIZE_FORMAT", pmc_rm="SIZE_FORMAT", kac="SIZE_FORMAT
        ", kac_preclean="SIZE_FORMAT")",
D
duke 已提交
5157
        _ser_pmc_preclean_ovflw, _ser_pmc_remark_ovflw,
5158
        _ser_kac_ovflw, _ser_kac_preclean_ovflw);
D
duke 已提交
5159 5160 5161 5162
    }
    _markStack.expand();
    _ser_pmc_remark_ovflw = 0;
    _ser_pmc_preclean_ovflw = 0;
5163
    _ser_kac_preclean_ovflw = 0;
D
duke 已提交
5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194
    _ser_kac_ovflw = 0;
  }
  if (_par_pmc_remark_ovflw > 0 || _par_kac_ovflw > 0) {
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr("Work queue overflow (benign) "
        "(pmc_rm="SIZE_FORMAT", kac="SIZE_FORMAT")",
        _par_pmc_remark_ovflw, _par_kac_ovflw);
    }
    _par_pmc_remark_ovflw = 0;
    _par_kac_ovflw = 0;
  }
  if (PrintCMSStatistics != 0) {
     if (_markStack._hit_limit > 0) {
       gclog_or_tty->print_cr(" (benign) Hit max stack size limit ("SIZE_FORMAT")",
                              _markStack._hit_limit);
     }
     if (_markStack._failed_double > 0) {
       gclog_or_tty->print_cr(" (benign) Failed stack doubling ("SIZE_FORMAT"),"
                              " current capacity "SIZE_FORMAT,
                              _markStack._failed_double,
                              _markStack.capacity());
     }
  }
  _markStack._hit_limit = 0;
  _markStack._failed_double = 0;

  if ((VerifyAfterGC || VerifyDuringGC) &&
      GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
    verify_after_remark();
  }

S
sla 已提交
5195 5196
  _gc_tracer_cm->report_object_count_after_gc(&_is_alive_closure);

D
duke 已提交
5197 5198 5199
  // Change under the freelistLocks.
  _collectorState = Sweeping;
  // Call isAllClear() under bitMapLock
5200 5201 5202 5203
  assert(_modUnionTable.isAllClear(),
      "Should be clear by end of the final marking");
  assert(_ct->klass_rem_set()->mod_union_is_clear(),
      "Should be clear by end of the final marking");
D
duke 已提交
5204 5205 5206 5207 5208
  if (UseAdaptiveSizePolicy) {
    size_policy()->checkpoint_roots_final_end(gch->gc_cause());
  }
}

J
jmasa 已提交
5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239 5240 5241 5242 5243 5244 5245 5246 5247 5248 5249 5250 5251 5252 5253
void CMSParInitialMarkTask::work(uint worker_id) {
  elapsedTimer _timer;
  ResourceMark rm;
  HandleMark   hm;

  // ---------- scan from roots --------------
  _timer.start();
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  Par_MarkRefsIntoClosure par_mri_cl(_collector->_span, &(_collector->_markBitMap));
  CMKlassClosure klass_closure(&par_mri_cl);

  // ---------- young gen roots --------------
  {
    work_on_young_gen_roots(worker_id, &par_mri_cl);
    _timer.stop();
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr(
        "Finished young gen initial mark scan work in %dth thread: %3.3f sec",
        worker_id, _timer.seconds());
    }
  }

  // ---------- remaining roots --------------
  _timer.reset();
  _timer.start();
  gch->gen_process_strong_roots(_collector->_cmsGen->level(),
                                false,     // yg was scanned above
                                false,     // this is parallel code
                                false,     // not scavenging
                                SharedHeap::ScanningOption(_collector->CMSCollector::roots_scanning_options()),
                                &par_mri_cl,
                                true,   // walk all of code cache if (so & SO_CodeCache)
                                NULL,
                                &klass_closure);
  assert(_collector->should_unload_classes()
         || (_collector->CMSCollector::roots_scanning_options() & SharedHeap::SO_CodeCache),
         "if we didn't scan the code cache, we have to be ready to drop nmethods with expired weak oops");
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr(
      "Finished remaining root initial mark scan work in %dth thread: %3.3f sec",
      worker_id, _timer.seconds());
  }
}

D
duke 已提交
5254
// Parallel remark task
J
jmasa 已提交
5255
class CMSParRemarkTask: public CMSParMarkTask {
D
duke 已提交
5256 5257 5258 5259 5260 5261 5262
  CompactibleFreeListSpace* _cms_space;

  // The per-thread work queues, available here for stealing.
  OopTaskQueueSet*       _task_queues;
  ParallelTaskTerminator _term;

 public:
5263 5264
  // A value of 0 passed to n_workers will cause the number of
  // workers to be taken from the active workers in the work gang.
D
duke 已提交
5265 5266
  CMSParRemarkTask(CMSCollector* collector,
                   CompactibleFreeListSpace* cms_space,
5267
                   int n_workers, FlexibleWorkGang* workers,
D
duke 已提交
5268
                   OopTaskQueueSet* task_queues):
J
jmasa 已提交
5269 5270
    CMSParMarkTask("Rescan roots and grey objects in parallel",
                   collector, n_workers),
5271
    _cms_space(cms_space),
D
duke 已提交
5272
    _task_queues(task_queues),
5273
    _term(n_workers, task_queues) { }
D
duke 已提交
5274 5275 5276 5277 5278 5279

  OopTaskQueueSet* task_queues() { return _task_queues; }

  OopTaskQueue* work_queue(int i) { return task_queues()->queue(i); }

  ParallelTaskTerminator* terminator() { return &_term; }
5280
  int n_workers() { return _n_workers; }
D
duke 已提交
5281

5282
  void work(uint worker_id);
D
duke 已提交
5283 5284 5285 5286 5287 5288 5289 5290 5291 5292

 private:
  // ... of  dirty cards in old space
  void do_dirty_card_rescan_tasks(CompactibleFreeListSpace* sp, int i,
                                  Par_MarkRefsIntoAndScanClosure* cl);

  // ... work stealing for the above
  void do_work_steal(int i, Par_MarkRefsIntoAndScanClosure* cl, int* seed);
};

5293 5294 5295 5296 5297 5298 5299 5300 5301 5302 5303 5304 5305 5306 5307 5308 5309 5310 5311 5312 5313 5314 5315
class RemarkKlassClosure : public KlassClosure {
  CMKlassClosure _cm_klass_closure;
 public:
  RemarkKlassClosure(OopClosure* oop_closure) : _cm_klass_closure(oop_closure) {}
  void do_klass(Klass* k) {
    // Check if we have modified any oops in the Klass during the concurrent marking.
    if (k->has_accumulated_modified_oops()) {
      k->clear_accumulated_modified_oops();

      // We could have transfered the current modified marks to the accumulated marks,
      // like we do with the Card Table to Mod Union Table. But it's not really necessary.
    } else if (k->has_modified_oops()) {
      // Don't clear anything, this info is needed by the next young collection.
    } else {
      // No modified oops in the Klass.
      return;
    }

    // The klass has modified fields, need to scan the klass.
    _cm_klass_closure.do_klass(k);
  }
};

J
jmasa 已提交
5316 5317 5318 5319 5320 5321 5322 5323 5324 5325 5326 5327 5328 5329 5330 5331 5332 5333 5334
void CMSParMarkTask::work_on_young_gen_roots(uint worker_id, OopsInGenClosure* cl) {
  DefNewGeneration* dng = _collector->_young_gen->as_DefNewGeneration();
  EdenSpace* eden_space = dng->eden();
  ContiguousSpace* from_space = dng->from();
  ContiguousSpace* to_space   = dng->to();

  HeapWord** eca = _collector->_eden_chunk_array;
  size_t     ect = _collector->_eden_chunk_index;
  HeapWord** sca = _collector->_survivor_chunk_array;
  size_t     sct = _collector->_survivor_chunk_index;

  assert(ect <= _collector->_eden_chunk_capacity, "out of bounds");
  assert(sct <= _collector->_survivor_chunk_capacity, "out of bounds");

  do_young_space_rescan(worker_id, cl, to_space, NULL, 0);
  do_young_space_rescan(worker_id, cl, from_space, sca, sct);
  do_young_space_rescan(worker_id, cl, eden_space, eca, ect);
}

5335 5336 5337 5338 5339
// work_queue(i) is passed to the closure
// Par_MarkRefsIntoAndScanClosure.  The "i" parameter
// also is passed to do_dirty_card_rescan_tasks() and to
// do_work_steal() to select the i-th task_queue.

5340
void CMSParRemarkTask::work(uint worker_id) {
D
duke 已提交
5341 5342 5343 5344 5345 5346 5347 5348 5349 5350
  elapsedTimer _timer;
  ResourceMark rm;
  HandleMark   hm;

  // ---------- rescan from roots --------------
  _timer.start();
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  Par_MarkRefsIntoAndScanClosure par_mrias_cl(_collector,
    _collector->_span, _collector->ref_processor(),
    &(_collector->_markBitMap),
5351
    work_queue(worker_id));
D
duke 已提交
5352 5353 5354 5355 5356 5357 5358

  // Rescan young gen roots first since these are likely
  // coarsely partitioned and may, on that account, constitute
  // the critical path; thus, it's best to start off that
  // work first.
  // ---------- young gen roots --------------
  {
J
jmasa 已提交
5359
    work_on_young_gen_roots(worker_id, &par_mrias_cl);
D
duke 已提交
5360 5361 5362 5363
    _timer.stop();
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr(
        "Finished young gen rescan work in %dth thread: %3.3f sec",
5364
        worker_id, _timer.seconds());
D
duke 已提交
5365 5366 5367 5368 5369 5370 5371 5372
    }
  }

  // ---------- remaining roots --------------
  _timer.reset();
  _timer.start();
  gch->gen_process_strong_roots(_collector->_cmsGen->level(),
                                false,     // yg was scanned above
5373
                                false,     // this is parallel code
5374
                                false,     // not scavenging
D
duke 已提交
5375
                                SharedHeap::ScanningOption(_collector->CMSCollector::roots_scanning_options()),
5376 5377
                                &par_mrias_cl,
                                true,   // walk all of code cache if (so & SO_CodeCache)
5378 5379
                                NULL,
                                NULL);     // The dirty klasses will be handled below
5380 5381 5382
  assert(_collector->should_unload_classes()
         || (_collector->CMSCollector::roots_scanning_options() & SharedHeap::SO_CodeCache),
         "if we didn't scan the code cache, we have to be ready to drop nmethods with expired weak oops");
D
duke 已提交
5383 5384 5385 5386
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr(
      "Finished remaining root rescan work in %dth thread: %3.3f sec",
5387
      worker_id, _timer.seconds());
D
duke 已提交
5388 5389
  }

5390 5391 5392 5393 5394 5395 5396 5397 5398 5399 5400 5401 5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413 5414 5415 5416 5417 5418 5419 5420 5421 5422 5423 5424 5425 5426 5427 5428 5429 5430 5431 5432 5433 5434 5435 5436
  // ---------- unhandled CLD scanning ----------
  if (worker_id == 0) { // Single threaded at the moment.
    _timer.reset();
    _timer.start();

    // Scan all new class loader data objects and new dependencies that were
    // introduced during concurrent marking.
    ResourceMark rm;
    GrowableArray<ClassLoaderData*>* array = ClassLoaderDataGraph::new_clds();
    for (int i = 0; i < array->length(); i++) {
      par_mrias_cl.do_class_loader_data(array->at(i));
    }

    // We don't need to keep track of new CLDs anymore.
    ClassLoaderDataGraph::remember_new_clds(false);

    _timer.stop();
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr(
          "Finished unhandled CLD scanning work in %dth thread: %3.3f sec",
          worker_id, _timer.seconds());
    }
  }

  // ---------- dirty klass scanning ----------
  if (worker_id == 0) { // Single threaded at the moment.
    _timer.reset();
    _timer.start();

    // Scan all classes that was dirtied during the concurrent marking phase.
    RemarkKlassClosure remark_klass_closure(&par_mrias_cl);
    ClassLoaderDataGraph::classes_do(&remark_klass_closure);

    _timer.stop();
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print_cr(
          "Finished dirty klass scanning work in %dth thread: %3.3f sec",
          worker_id, _timer.seconds());
    }
  }

  // We might have added oops to ClassLoaderData::_handles during the
  // concurrent marking phase. These oops point to newly allocated objects
  // that are guaranteed to be kept alive. Either by the direct allocation
  // code, or when the young collector processes the strong roots. Hence,
  // we don't have to revisit the _handles block during the remark phase.

D
duke 已提交
5437 5438 5439 5440 5441
  // ---------- rescan dirty cards ------------
  _timer.reset();
  _timer.start();

  // Do the rescan tasks for each of the two spaces
5442
  // (cms_space) in turn.
5443 5444
  // "worker_id" is passed to select the task_queue for "worker_id"
  do_dirty_card_rescan_tasks(_cms_space, worker_id, &par_mrias_cl);
D
duke 已提交
5445 5446 5447 5448
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr(
      "Finished dirty card rescan work in %dth thread: %3.3f sec",
5449
      worker_id, _timer.seconds());
D
duke 已提交
5450 5451 5452 5453 5454 5455
  }

  // ---------- steal work from other threads ...
  // ---------- ... and drain overflow list.
  _timer.reset();
  _timer.start();
5456
  do_work_steal(worker_id, &par_mrias_cl, _collector->hash_seed(worker_id));
D
duke 已提交
5457 5458 5459 5460
  _timer.stop();
  if (PrintCMSStatistics != 0) {
    gclog_or_tty->print_cr(
      "Finished work stealing in %dth thread: %3.3f sec",
5461
      worker_id, _timer.seconds());
D
duke 已提交
5462 5463 5464
  }
}

5465
// Note that parameter "i" is not used.
D
duke 已提交
5466
void
J
jmasa 已提交
5467 5468
CMSParMarkTask::do_young_space_rescan(uint worker_id,
  OopsInGenClosure* cl, ContiguousSpace* space,
D
duke 已提交
5469 5470 5471 5472 5473 5474 5475 5476 5477 5478 5479 5480 5481
  HeapWord** chunk_array, size_t chunk_top) {
  // Until all tasks completed:
  // . claim an unclaimed task
  // . compute region boundaries corresponding to task claimed
  //   using chunk_array
  // . par_oop_iterate(cl) over that region

  ResourceMark rm;
  HandleMark   hm;

  SequentialSubTasksDone* pst = space->par_seq_tasks();
  assert(pst->valid(), "Uninitialized use?");

5482 5483
  uint nth_task = 0;
  uint n_tasks  = pst->n_tasks();
D
duke 已提交
5484 5485 5486 5487 5488 5489 5490 5491 5492 5493 5494

  HeapWord *start, *end;
  while (!pst->is_task_claimed(/* reference */ nth_task)) {
    // We claimed task # nth_task; compute its boundaries.
    if (chunk_top == 0) {  // no samples were taken
      assert(nth_task == 0 && n_tasks == 1, "Can have only 1 EdenSpace task");
      start = space->bottom();
      end   = space->top();
    } else if (nth_task == 0) {
      start = space->bottom();
      end   = chunk_array[nth_task];
5495
    } else if (nth_task < (uint)chunk_top) {
D
duke 已提交
5496 5497 5498 5499
      assert(nth_task >= 1, "Control point invariant");
      start = chunk_array[nth_task - 1];
      end   = chunk_array[nth_task];
    } else {
5500
      assert(nth_task == (uint)chunk_top, "Control point invariant");
D
duke 已提交
5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525 5526 5527 5528 5529 5530 5531 5532 5533 5534 5535 5536 5537 5538 5539 5540 5541 5542 5543 5544 5545 5546 5547 5548 5549 5550 5551 5552 5553 5554 5555 5556 5557
      start = chunk_array[chunk_top - 1];
      end   = space->top();
    }
    MemRegion mr(start, end);
    // Verify that mr is in space
    assert(mr.is_empty() || space->used_region().contains(mr),
           "Should be in space");
    // Verify that "start" is an object boundary
    assert(mr.is_empty() || oop(mr.start())->is_oop(),
           "Should be an oop");
    space->par_oop_iterate(mr, cl);
  }
  pst->all_tasks_completed();
}

void
CMSParRemarkTask::do_dirty_card_rescan_tasks(
  CompactibleFreeListSpace* sp, int i,
  Par_MarkRefsIntoAndScanClosure* cl) {
  // Until all tasks completed:
  // . claim an unclaimed task
  // . compute region boundaries corresponding to task claimed
  // . transfer dirty bits ct->mut for that region
  // . apply rescanclosure to dirty mut bits for that region

  ResourceMark rm;
  HandleMark   hm;

  OopTaskQueue* work_q = work_queue(i);
  ModUnionClosure modUnionClosure(&(_collector->_modUnionTable));
  // CAUTION! CAUTION! CAUTION! CAUTION! CAUTION! CAUTION! CAUTION!
  // CAUTION: This closure has state that persists across calls to
  // the work method dirty_range_iterate_clear() in that it has
  // imbedded in it a (subtype of) UpwardsObjectClosure. The
  // use of that state in the imbedded UpwardsObjectClosure instance
  // assumes that the cards are always iterated (even if in parallel
  // by several threads) in monotonically increasing order per each
  // thread. This is true of the implementation below which picks
  // card ranges (chunks) in monotonically increasing order globally
  // and, a-fortiori, in monotonically increasing order per thread
  // (the latter order being a subsequence of the former).
  // If the work code below is ever reorganized into a more chaotic
  // work-partitioning form than the current "sequential tasks"
  // paradigm, the use of that persistent state will have to be
  // revisited and modified appropriately. See also related
  // bug 4756801 work on which should examine this code to make
  // sure that the changes there do not run counter to the
  // assumptions made here and necessary for correctness and
  // efficiency. Note also that this code might yield inefficient
  // behaviour in the case of very large objects that span one or
  // more work chunks. Such objects would potentially be scanned
  // several times redundantly. Work on 4756801 should try and
  // address that performance anomaly if at all possible. XXX
  MemRegion  full_span  = _collector->_span;
  CMSBitMap* bm    = &(_collector->_markBitMap);     // shared
  MarkFromDirtyCardsClosure
    greyRescanClosure(_collector, full_span, // entire span of interest
5558
                      sp, bm, work_q, cl);
D
duke 已提交
5559 5560 5561

  SequentialSubTasksDone* pst = sp->conc_par_seq_tasks();
  assert(pst->valid(), "Uninitialized use?");
5562
  uint nth_task = 0;
D
duke 已提交
5563 5564 5565 5566 5567 5568 5569 5570 5571 5572 5573 5574 5575 5576 5577 5578 5579 5580 5581 5582 5583 5584 5585 5586 5587 5588 5589 5590 5591 5592 5593 5594 5595 5596 5597 5598 5599 5600 5601 5602 5603 5604 5605 5606 5607 5608 5609 5610 5611 5612 5613 5614 5615 5616 5617 5618 5619 5620
  const int alignment = CardTableModRefBS::card_size * BitsPerWord;
  MemRegion span = sp->used_region();
  HeapWord* start_addr = span.start();
  HeapWord* end_addr = (HeapWord*)round_to((intptr_t)span.end(),
                                           alignment);
  const size_t chunk_size = sp->rescan_task_size(); // in HeapWord units
  assert((HeapWord*)round_to((intptr_t)start_addr, alignment) ==
         start_addr, "Check alignment");
  assert((size_t)round_to((intptr_t)chunk_size, alignment) ==
         chunk_size, "Check alignment");

  while (!pst->is_task_claimed(/* reference */ nth_task)) {
    // Having claimed the nth_task, compute corresponding mem-region,
    // which is a-fortiori aligned correctly (i.e. at a MUT bopundary).
    // The alignment restriction ensures that we do not need any
    // synchronization with other gang-workers while setting or
    // clearing bits in thus chunk of the MUT.
    MemRegion this_span = MemRegion(start_addr + nth_task*chunk_size,
                                    start_addr + (nth_task+1)*chunk_size);
    // The last chunk's end might be way beyond end of the
    // used region. In that case pull back appropriately.
    if (this_span.end() > end_addr) {
      this_span.set_end(end_addr);
      assert(!this_span.is_empty(), "Program logic (calculation of n_tasks)");
    }
    // Iterate over the dirty cards covering this chunk, marking them
    // precleaned, and setting the corresponding bits in the mod union
    // table. Since we have been careful to partition at Card and MUT-word
    // boundaries no synchronization is needed between parallel threads.
    _collector->_ct->ct_bs()->dirty_card_iterate(this_span,
                                                 &modUnionClosure);

    // Having transferred these marks into the modUnionTable,
    // rescan the marked objects on the dirty cards in the modUnionTable.
    // Even if this is at a synchronous collection, the initial marking
    // may have been done during an asynchronous collection so there
    // may be dirty bits in the mod-union table.
    _collector->_modUnionTable.dirty_range_iterate_clear(
                  this_span, &greyRescanClosure);
    _collector->_modUnionTable.verifyNoOneBitsInRange(
                                 this_span.start(),
                                 this_span.end());
  }
  pst->all_tasks_completed();  // declare that i am done
}

// . see if we can share work_queues with ParNew? XXX
void
CMSParRemarkTask::do_work_steal(int i, Par_MarkRefsIntoAndScanClosure* cl,
                                int* seed) {
  OopTaskQueue* work_q = work_queue(i);
  NOT_PRODUCT(int num_steals = 0;)
  oop obj_to_scan;
  CMSBitMap* bm = &(_collector->_markBitMap);

  while (true) {
    // Completely finish any left over work from (an) earlier round(s)
    cl->trim_queue(0);
5621 5622
    size_t num_from_overflow_list = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
                                         (size_t)ParGCDesiredObjsFromOverflowList);
D
duke 已提交
5623
    // Now check if there's any work in the overflow list
5624 5625 5626 5627
    // Passing ParallelGCThreads as the third parameter, no_of_gc_threads,
    // only affects the number of attempts made to get work from the
    // overflow list and does not affect the number of workers.  Just
    // pass ParallelGCThreads so this behavior is unchanged.
D
duke 已提交
5628
    if (_collector->par_take_from_overflow_list(num_from_overflow_list,
5629 5630
                                                work_q,
                                                ParallelGCThreads)) {
D
duke 已提交
5631 5632 5633 5634 5635 5636 5637 5638 5639 5640 5641 5642 5643 5644 5645 5646 5647 5648 5649 5650 5651 5652 5653 5654 5655 5656 5657 5658 5659 5660 5661 5662
      // found something in global overflow list;
      // not yet ready to go stealing work from others.
      // We'd like to assert(work_q->size() != 0, ...)
      // because we just took work from the overflow list,
      // but of course we can't since all of that could have
      // been already stolen from us.
      // "He giveth and He taketh away."
      continue;
    }
    // Verify that we have no work before we resort to stealing
    assert(work_q->size() == 0, "Have work, shouldn't steal");
    // Try to steal from other queues that have work
    if (task_queues()->steal(i, seed, /* reference */ obj_to_scan)) {
      NOT_PRODUCT(num_steals++;)
      assert(obj_to_scan->is_oop(), "Oops, not an oop!");
      assert(bm->isMarked((HeapWord*)obj_to_scan), "Stole an unmarked oop?");
      // Do scanning work
      obj_to_scan->oop_iterate(cl);
      // Loop around, finish this work, and try to steal some more
    } else if (terminator()->offer_termination()) {
        break;  // nirvana from the infinite cycle
    }
  }
  NOT_PRODUCT(
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print("\n\t(%d: stole %d oops)", i, num_steals);
    }
  )
  assert(work_q->size() == 0 && _collector->overflow_list_is_empty(),
         "Else our work is not yet done");
}

5663 5664 5665 5666 5667 5668 5669 5670 5671 5672 5673 5674 5675 5676 5677 5678 5679 5680 5681 5682 5683 5684 5685 5686 5687 5688
// Record object boundaries in _eden_chunk_array by sampling the eden
// top in the slow-path eden object allocation code path and record
// the boundaries, if CMSEdenChunksRecordAlways is true. If
// CMSEdenChunksRecordAlways is false, we use the other asynchronous
// sampling in sample_eden() that activates during the part of the
// preclean phase.
void CMSCollector::sample_eden_chunk() {
  if (CMSEdenChunksRecordAlways && _eden_chunk_array != NULL) {
    if (_eden_chunk_lock->try_lock()) {
      // Record a sample. This is the critical section. The contents
      // of the _eden_chunk_array have to be non-decreasing in the
      // address order.
      _eden_chunk_array[_eden_chunk_index] = *_top_addr;
      assert(_eden_chunk_array[_eden_chunk_index] <= *_end_addr,
             "Unexpected state of Eden");
      if (_eden_chunk_index == 0 ||
          ((_eden_chunk_array[_eden_chunk_index] > _eden_chunk_array[_eden_chunk_index-1]) &&
           (pointer_delta(_eden_chunk_array[_eden_chunk_index],
                          _eden_chunk_array[_eden_chunk_index-1]) >= CMSSamplingGrain))) {
        _eden_chunk_index++;  // commit sample
      }
      _eden_chunk_lock->unlock();
    }
  }
}

D
duke 已提交
5689 5690 5691 5692 5693 5694 5695 5696 5697 5698 5699 5700 5701 5702 5703 5704 5705 5706 5707 5708 5709 5710 5711
// Return a thread-local PLAB recording array, as appropriate.
void* CMSCollector::get_data_recorder(int thr_num) {
  if (_survivor_plab_array != NULL &&
      (CMSPLABRecordAlways ||
       (_collectorState > Marking && _collectorState < FinalMarking))) {
    assert(thr_num < (int)ParallelGCThreads, "thr_num is out of bounds");
    ChunkArray* ca = &_survivor_plab_array[thr_num];
    ca->reset();   // clear it so that fresh data is recorded
    return (void*) ca;
  } else {
    return NULL;
  }
}

// Reset all the thread-local PLAB recording arrays
void CMSCollector::reset_survivor_plab_arrays() {
  for (uint i = 0; i < ParallelGCThreads; i++) {
    _survivor_plab_array[i].reset();
  }
}

// Merge the per-thread plab arrays into the global survivor chunk
// array which will provide the partitioning of the survivor space
J
jmasa 已提交
5712
// for CMS initial scan and rescan.
5713 5714
void CMSCollector::merge_survivor_plab_arrays(ContiguousSpace* surv,
                                              int no_of_gc_threads) {
D
duke 已提交
5715 5716
  assert(_survivor_plab_array  != NULL, "Error");
  assert(_survivor_chunk_array != NULL, "Error");
J
jmasa 已提交
5717 5718
  assert(_collectorState == FinalMarking ||
         (CMSParallelInitialMarkEnabled && _collectorState == InitialMarking), "Error");
5719
  for (int j = 0; j < no_of_gc_threads; j++) {
D
duke 已提交
5720 5721 5722 5723 5724 5725 5726
    _cursor[j] = 0;
  }
  HeapWord* top = surv->top();
  size_t i;
  for (i = 0; i < _survivor_chunk_capacity; i++) {  // all sca entries
    HeapWord* min_val = top;          // Higher than any PLAB address
    uint      min_tid = 0;            // position of min_val this round
5727
    for (int j = 0; j < no_of_gc_threads; j++) {
D
duke 已提交
5728 5729 5730 5731 5732 5733 5734 5735 5736 5737 5738 5739 5740 5741 5742 5743 5744 5745 5746 5747 5748 5749 5750 5751 5752 5753 5754 5755 5756 5757 5758 5759 5760
      ChunkArray* cur_sca = &_survivor_plab_array[j];
      if (_cursor[j] == cur_sca->end()) {
        continue;
      }
      assert(_cursor[j] < cur_sca->end(), "ctl pt invariant");
      HeapWord* cur_val = cur_sca->nth(_cursor[j]);
      assert(surv->used_region().contains(cur_val), "Out of bounds value");
      if (cur_val < min_val) {
        min_tid = j;
        min_val = cur_val;
      } else {
        assert(cur_val < top, "All recorded addresses should be less");
      }
    }
    // At this point min_val and min_tid are respectively
    // the least address in _survivor_plab_array[j]->nth(_cursor[j])
    // and the thread (j) that witnesses that address.
    // We record this address in the _survivor_chunk_array[i]
    // and increment _cursor[min_tid] prior to the next round i.
    if (min_val == top) {
      break;
    }
    _survivor_chunk_array[i] = min_val;
    _cursor[min_tid]++;
  }
  // We are all done; record the size of the _survivor_chunk_array
  _survivor_chunk_index = i; // exclusive: [0, i)
  if (PrintCMSStatistics > 0) {
    gclog_or_tty->print(" (Survivor:" SIZE_FORMAT "chunks) ", i);
  }
  // Verify that we used up all the recorded entries
  #ifdef ASSERT
    size_t total = 0;
5761
    for (int j = 0; j < no_of_gc_threads; j++) {
D
duke 已提交
5762 5763 5764 5765 5766 5767 5768 5769 5770 5771 5772 5773 5774 5775 5776 5777 5778 5779 5780
      assert(_cursor[j] == _survivor_plab_array[j].end(), "Ctl pt invariant");
      total += _cursor[j];
    }
    assert(total == _survivor_chunk_index, "Ctl Pt Invariant");
    // Check that the merged array is in sorted order
    if (total > 0) {
      for (size_t i = 0; i < total - 1; i++) {
        if (PrintCMSStatistics > 0) {
          gclog_or_tty->print(" (chunk" SIZE_FORMAT ":" INTPTR_FORMAT ") ",
                              i, _survivor_chunk_array[i]);
        }
        assert(_survivor_chunk_array[i] < _survivor_chunk_array[i+1],
               "Not sorted");
      }
    }
  #endif // ASSERT
}

// Set up the space's par_seq_tasks structure for work claiming
J
jmasa 已提交
5781
// for parallel initial scan and rescan of young gen.
D
duke 已提交
5782 5783 5784 5785 5786 5787 5788 5789 5790 5791 5792 5793 5794 5795
// See ParRescanTask where this is currently used.
void
CMSCollector::
initialize_sequential_subtasks_for_young_gen_rescan(int n_threads) {
  assert(n_threads > 0, "Unexpected n_threads argument");
  DefNewGeneration* dng = (DefNewGeneration*)_young_gen;

  // Eden space
  {
    SequentialSubTasksDone* pst = dng->eden()->par_seq_tasks();
    assert(!pst->valid(), "Clobbering existing data?");
    // Each valid entry in [0, _eden_chunk_index) represents a task.
    size_t n_tasks = _eden_chunk_index + 1;
    assert(n_tasks == 1 || _eden_chunk_array != NULL, "Error");
5796 5797 5798
    // Sets the condition for completion of the subtask (how many threads
    // need to finish in order to be done).
    pst->set_n_threads(n_threads);
D
duke 已提交
5799 5800 5801 5802 5803
    pst->set_n_tasks((int)n_tasks);
  }

  // Merge the survivor plab arrays into _survivor_chunk_array
  if (_survivor_plab_array != NULL) {
5804
    merge_survivor_plab_arrays(dng->from(), n_threads);
D
duke 已提交
5805 5806 5807 5808 5809 5810 5811 5812
  } else {
    assert(_survivor_chunk_index == 0, "Error");
  }

  // To space
  {
    SequentialSubTasksDone* pst = dng->to()->par_seq_tasks();
    assert(!pst->valid(), "Clobbering existing data?");
5813 5814 5815
    // Sets the condition for completion of the subtask (how many threads
    // need to finish in order to be done).
    pst->set_n_threads(n_threads);
D
duke 已提交
5816 5817 5818 5819 5820 5821 5822 5823 5824 5825
    pst->set_n_tasks(1);
    assert(pst->valid(), "Error");
  }

  // From space
  {
    SequentialSubTasksDone* pst = dng->from()->par_seq_tasks();
    assert(!pst->valid(), "Clobbering existing data?");
    size_t n_tasks = _survivor_chunk_index + 1;
    assert(n_tasks == 1 || _survivor_chunk_array != NULL, "Error");
5826 5827 5828
    // Sets the condition for completion of the subtask (how many threads
    // need to finish in order to be done).
    pst->set_n_threads(n_threads);
D
duke 已提交
5829 5830 5831 5832 5833 5834 5835 5836
    pst->set_n_tasks((int)n_tasks);
    assert(pst->valid(), "Error");
  }
}

// Parallel version of remark
void CMSCollector::do_remark_parallel() {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
5837
  FlexibleWorkGang* workers = gch->workers();
D
duke 已提交
5838
  assert(workers != NULL, "Need parallel worker threads.");
5839 5840 5841 5842 5843 5844 5845 5846 5847
  // Choose to use the number of GC workers most recently set
  // into "active_workers".  If active_workers is not set, set it
  // to ParallelGCThreads.
  int n_workers = workers->active_workers();
  if (n_workers == 0) {
    assert(n_workers > 0, "Should have been set during scavenge");
    n_workers = ParallelGCThreads;
    workers->set_active_workers(n_workers);
  }
D
duke 已提交
5848 5849 5850
  CompactibleFreeListSpace* cms_space  = _cmsGen->cmsSpace();

  CMSParRemarkTask tsk(this,
5851
    cms_space,
D
duke 已提交
5852 5853 5854 5855 5856 5857 5858 5859 5860 5861 5862 5863 5864 5865 5866 5867 5868 5869 5870 5871 5872 5873 5874 5875 5876 5877 5878 5879 5880 5881 5882
    n_workers, workers, task_queues());

  // Set up for parallel process_strong_roots work.
  gch->set_par_threads(n_workers);
  // We won't be iterating over the cards in the card table updating
  // the younger_gen cards, so we shouldn't call the following else
  // the verification code as well as subsequent younger_refs_iterate
  // code would get confused. XXX
  // gch->rem_set()->prepare_for_younger_refs_iterate(true); // parallel

  // The young gen rescan work will not be done as part of
  // process_strong_roots (which currently doesn't knw how to
  // parallelize such a scan), but rather will be broken up into
  // a set of parallel tasks (via the sampling that the [abortable]
  // preclean phase did of EdenSpace, plus the [two] tasks of
  // scanning the [two] survivor spaces. Further fine-grain
  // parallelization of the scanning of the survivor spaces
  // themselves, and of precleaning of the younger gen itself
  // is deferred to the future.
  initialize_sequential_subtasks_for_young_gen_rescan(n_workers);

  // The dirty card rescan work is broken up into a "sequence"
  // of parallel tasks (per constituent space) that are dynamically
  // claimed by the parallel threads.
  cms_space->initialize_sequential_subtasks_for_rescan(n_workers);

  // It turns out that even when we're using 1 thread, doing the work in a
  // separate thread causes wide variance in run times.  We can't help this
  // in the multi-threaded case, but we special-case n=1 here to get
  // repeatable measurements of the 1-thread overhead of the parallel code.
  if (n_workers > 1) {
5883 5884 5885 5886
    // Make refs discovery MT-safe, if it isn't already: it may not
    // necessarily be so, since it's possible that we are doing
    // ST marking.
    ReferenceProcessorMTDiscoveryMutator mt(ref_processor(), true);
5887
    GenCollectedHeap::StrongRootsScope srs(gch);
D
duke 已提交
5888 5889
    workers->run_task(&tsk);
  } else {
5890
    ReferenceProcessorMTDiscoveryMutator mt(ref_processor(), false);
5891
    GenCollectedHeap::StrongRootsScope srs(gch);
D
duke 已提交
5892 5893
    tsk.work(0);
  }
5894

D
duke 已提交
5895 5896 5897 5898 5899 5900 5901 5902 5903 5904 5905
  gch->set_par_threads(0);  // 0 ==> non-parallel.
  // restore, single-threaded for now, any preserved marks
  // as a result of work_q overflow
  restore_preserved_marks_if_any();
}

// Non-parallel version of remark
void CMSCollector::do_remark_non_parallel() {
  ResourceMark rm;
  HandleMark   hm;
  GenCollectedHeap* gch = GenCollectedHeap::heap();
5906 5907
  ReferenceProcessorMTDiscoveryMutator mt(ref_processor(), false);

D
duke 已提交
5908
  MarkRefsIntoAndScanClosure
5909 5910
    mrias_cl(_span, ref_processor(), &_markBitMap, NULL /* not precleaning */,
             &_markStack, this,
D
duke 已提交
5911 5912 5913 5914
             false /* should_yield */, false /* not precleaning */);
  MarkFromDirtyCardsClosure
    markFromDirtyCardsClosure(this, _span,
                              NULL,  // space is set further below
5915
                              &_markBitMap, &_markStack, &mrias_cl);
D
duke 已提交
5916
  {
S
sla 已提交
5917
    GCTraceTime t("grey object rescan", PrintGCDetails, false, _gc_timer_cm);
5918 5919
    // Iterate over the dirty cards, setting the corresponding bits in the
    // mod union table.
D
duke 已提交
5920 5921 5922 5923 5924 5925 5926 5927 5928 5929 5930 5931 5932 5933 5934 5935 5936 5937 5938 5939 5940 5941 5942 5943 5944 5945 5946 5947 5948 5949 5950
    {
      ModUnionClosure modUnionClosure(&_modUnionTable);
      _ct->ct_bs()->dirty_card_iterate(
                      _cmsGen->used_region(),
                      &modUnionClosure);
    }
    // Having transferred these marks into the modUnionTable, we just need
    // to rescan the marked objects on the dirty cards in the modUnionTable.
    // The initial marking may have been done during an asynchronous
    // collection so there may be dirty bits in the mod-union table.
    const int alignment =
      CardTableModRefBS::card_size * BitsPerWord;
    {
      // ... First handle dirty cards in CMS gen
      markFromDirtyCardsClosure.set_space(_cmsGen->cmsSpace());
      MemRegion ur = _cmsGen->used_region();
      HeapWord* lb = ur.start();
      HeapWord* ub = (HeapWord*)round_to((intptr_t)ur.end(), alignment);
      MemRegion cms_span(lb, ub);
      _modUnionTable.dirty_range_iterate_clear(cms_span,
                                               &markFromDirtyCardsClosure);
      verify_work_stacks_empty();
      if (PrintCMSStatistics != 0) {
        gclog_or_tty->print(" (re-scanned "SIZE_FORMAT" dirty cards in cms gen) ",
          markFromDirtyCardsClosure.num_dirty_cards());
      }
    }
  }
  if (VerifyDuringGC &&
      GenCollectedHeap::heap()->total_collections() >= VerifyGCStartAt) {
    HandleMark hm;  // Discard invalid handles created during verification
5951
    Universe::verify();
D
duke 已提交
5952 5953
  }
  {
S
sla 已提交
5954
    GCTraceTime t("root rescan", PrintGCDetails, false, _gc_timer_cm);
D
duke 已提交
5955 5956 5957 5958

    verify_work_stacks_empty();

    gch->rem_set()->prepare_for_younger_refs_iterate(false); // Not parallel.
5959
    GenCollectedHeap::StrongRootsScope srs(gch);
D
duke 已提交
5960 5961
    gch->gen_process_strong_roots(_cmsGen->level(),
                                  true,  // younger gens as roots
5962
                                  false, // use the local StrongRootsScope
5963
                                  false, // not scavenging
D
duke 已提交
5964
                                  SharedHeap::ScanningOption(roots_scanning_options()),
5965 5966
                                  &mrias_cl,
                                  true,   // walk code active on stacks
5967 5968 5969
                                  NULL,
                                  NULL);  // The dirty klasses will be handled below

5970 5971 5972
    assert(should_unload_classes()
           || (roots_scanning_options() & SharedHeap::SO_CodeCache),
           "if we didn't scan the code cache, we have to be ready to drop nmethods with expired weak oops");
D
duke 已提交
5973
  }
5974 5975

  {
S
sla 已提交
5976
    GCTraceTime t("visit unhandled CLDs", PrintGCDetails, false, _gc_timer_cm);
5977 5978 5979 5980 5981 5982 5983 5984 5985 5986 5987 5988 5989 5990 5991 5992 5993 5994

    verify_work_stacks_empty();

    // Scan all class loader data objects that might have been introduced
    // during concurrent marking.
    ResourceMark rm;
    GrowableArray<ClassLoaderData*>* array = ClassLoaderDataGraph::new_clds();
    for (int i = 0; i < array->length(); i++) {
      mrias_cl.do_class_loader_data(array->at(i));
    }

    // We don't need to keep track of new CLDs anymore.
    ClassLoaderDataGraph::remember_new_clds(false);

    verify_work_stacks_empty();
  }

  {
S
sla 已提交
5995
    GCTraceTime t("dirty klass scan", PrintGCDetails, false, _gc_timer_cm);
5996 5997 5998 5999 6000 6001 6002 6003 6004 6005 6006 6007 6008 6009 6010

    verify_work_stacks_empty();

    RemarkKlassClosure remark_klass_closure(&mrias_cl);
    ClassLoaderDataGraph::classes_do(&remark_klass_closure);

    verify_work_stacks_empty();
  }

  // We might have added oops to ClassLoaderData::_handles during the
  // concurrent marking phase. These oops point to newly allocated objects
  // that are guaranteed to be kept alive. Either by the direct allocation
  // code, or when the young collector processes the strong roots. Hence,
  // we don't have to revisit the _handles block during the remark phase.

D
duke 已提交
6011 6012 6013 6014 6015 6016 6017 6018 6019 6020 6021
  verify_work_stacks_empty();
  // Restore evacuated mark words, if any, used for overflow list links
  if (!CMSOverflowEarlyRestoration) {
    restore_preserved_marks_if_any();
  }
  verify_overflow_empty();
}

////////////////////////////////////////////////////////
// Parallel Reference Processing Task Proxy Class
////////////////////////////////////////////////////////
6022
class CMSRefProcTaskProxy: public AbstractGangTaskWOopQueues {
D
duke 已提交
6023 6024 6025
  typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask;
  CMSCollector*          _collector;
  CMSBitMap*             _mark_bit_map;
6026
  const MemRegion        _span;
D
duke 已提交
6027 6028 6029 6030 6031 6032 6033
  ProcessTask&           _task;

public:
  CMSRefProcTaskProxy(ProcessTask&     task,
                      CMSCollector*    collector,
                      const MemRegion& span,
                      CMSBitMap*       mark_bit_map,
6034
                      AbstractWorkGang* workers,
D
duke 已提交
6035
                      OopTaskQueueSet* task_queues):
6036 6037 6038 6039
    // XXX Should superclass AGTWOQ also know about AWG since it knows
    // about the task_queues used by the AWG? Then it could initialize
    // the terminator() object. See 6984287. The set_for_termination()
    // below is a temporary band-aid for the regression in 6984287.
6040 6041
    AbstractGangTaskWOopQueues("Process referents by policy in parallel",
      task_queues),
D
duke 已提交
6042
    _task(task),
6043
    _collector(collector), _span(span), _mark_bit_map(mark_bit_map)
6044 6045 6046 6047 6048
  {
    assert(_collector->_span.equals(_span) && !_span.is_empty(),
           "Inconsistency in _span");
    set_for_termination(workers->active_workers());
  }
D
duke 已提交
6049

6050
  OopTaskQueueSet* task_queues() { return queues(); }
D
duke 已提交
6051 6052 6053 6054 6055 6056 6057 6058

  OopTaskQueue* work_queue(int i) { return task_queues()->queue(i); }

  void do_work_steal(int i,
                     CMSParDrainMarkingStackClosure* drain,
                     CMSParKeepAliveClosure* keep_alive,
                     int* seed);

6059
  virtual void work(uint worker_id);
D
duke 已提交
6060 6061
};

6062
void CMSRefProcTaskProxy::work(uint worker_id) {
6063
  assert(_collector->_span.equals(_span), "Inconsistency in _span");
D
duke 已提交
6064
  CMSParKeepAliveClosure par_keep_alive(_collector, _span,
6065
                                        _mark_bit_map,
6066
                                        work_queue(worker_id));
D
duke 已提交
6067
  CMSParDrainMarkingStackClosure par_drain_stack(_collector, _span,
6068
                                                 _mark_bit_map,
6069
                                                 work_queue(worker_id));
6070
  CMSIsAliveClosure is_alive_closure(_span, _mark_bit_map);
6071
  _task.work(worker_id, is_alive_closure, par_keep_alive, par_drain_stack);
D
duke 已提交
6072
  if (_task.marks_oops_alive()) {
6073 6074
    do_work_steal(worker_id, &par_drain_stack, &par_keep_alive,
                  _collector->hash_seed(worker_id));
D
duke 已提交
6075
  }
6076
  assert(work_queue(worker_id)->size() == 0, "work_queue should be empty");
D
duke 已提交
6077 6078 6079 6080 6081 6082 6083 6084 6085 6086 6087 6088 6089
  assert(_collector->_overflow_list == NULL, "non-empty _overflow_list");
}

class CMSRefEnqueueTaskProxy: public AbstractGangTask {
  typedef AbstractRefProcTaskExecutor::EnqueueTask EnqueueTask;
  EnqueueTask& _task;

public:
  CMSRefEnqueueTaskProxy(EnqueueTask& task)
    : AbstractGangTask("Enqueue reference objects in parallel"),
      _task(task)
  { }

6090
  virtual void work(uint worker_id)
D
duke 已提交
6091
  {
6092
    _task.work(worker_id);
D
duke 已提交
6093 6094 6095 6096
  }
};

CMSParKeepAliveClosure::CMSParKeepAliveClosure(CMSCollector* collector,
6097
  MemRegion span, CMSBitMap* bit_map, OopTaskQueue* work_queue):
D
duke 已提交
6098 6099 6100
   _span(span),
   _bit_map(bit_map),
   _work_queue(work_queue),
6101
   _mark_and_push(collector, span, bit_map, work_queue),
D
duke 已提交
6102 6103 6104 6105 6106 6107 6108 6109 6110 6111 6112 6113 6114 6115 6116 6117
   _low_water_mark(MIN2((uint)(work_queue->max_elems()/4),
                        (uint)(CMSWorkQueueDrainThreshold * ParallelGCThreads)))
{ }

// . see if we can share work_queues with ParNew? XXX
void CMSRefProcTaskProxy::do_work_steal(int i,
  CMSParDrainMarkingStackClosure* drain,
  CMSParKeepAliveClosure* keep_alive,
  int* seed) {
  OopTaskQueue* work_q = work_queue(i);
  NOT_PRODUCT(int num_steals = 0;)
  oop obj_to_scan;

  while (true) {
    // Completely finish any left over work from (an) earlier round(s)
    drain->trim_queue(0);
6118 6119
    size_t num_from_overflow_list = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
                                         (size_t)ParGCDesiredObjsFromOverflowList);
D
duke 已提交
6120
    // Now check if there's any work in the overflow list
6121 6122 6123 6124
    // Passing ParallelGCThreads as the third parameter, no_of_gc_threads,
    // only affects the number of attempts made to get work from the
    // overflow list and does not affect the number of workers.  Just
    // pass ParallelGCThreads so this behavior is unchanged.
D
duke 已提交
6125
    if (_collector->par_take_from_overflow_list(num_from_overflow_list,
6126 6127
                                                work_q,
                                                ParallelGCThreads)) {
D
duke 已提交
6128 6129 6130 6131 6132 6133 6134 6135 6136 6137 6138 6139 6140 6141 6142 6143 6144 6145 6146 6147 6148 6149 6150 6151 6152 6153 6154 6155 6156 6157 6158 6159
      // Found something in global overflow list;
      // not yet ready to go stealing work from others.
      // We'd like to assert(work_q->size() != 0, ...)
      // because we just took work from the overflow list,
      // but of course we can't, since all of that might have
      // been already stolen from us.
      continue;
    }
    // Verify that we have no work before we resort to stealing
    assert(work_q->size() == 0, "Have work, shouldn't steal");
    // Try to steal from other queues that have work
    if (task_queues()->steal(i, seed, /* reference */ obj_to_scan)) {
      NOT_PRODUCT(num_steals++;)
      assert(obj_to_scan->is_oop(), "Oops, not an oop!");
      assert(_mark_bit_map->isMarked((HeapWord*)obj_to_scan), "Stole an unmarked oop?");
      // Do scanning work
      obj_to_scan->oop_iterate(keep_alive);
      // Loop around, finish this work, and try to steal some more
    } else if (terminator()->offer_termination()) {
      break;  // nirvana from the infinite cycle
    }
  }
  NOT_PRODUCT(
    if (PrintCMSStatistics != 0) {
      gclog_or_tty->print("\n\t(%d: stole %d oops)", i, num_steals);
    }
  )
}

void CMSRefProcTaskExecutor::execute(ProcessTask& task)
{
  GenCollectedHeap* gch = GenCollectedHeap::heap();
6160
  FlexibleWorkGang* workers = gch->workers();
D
duke 已提交
6161 6162 6163 6164
  assert(workers != NULL, "Need parallel worker threads.");
  CMSRefProcTaskProxy rp_task(task, &_collector,
                              _collector.ref_processor()->span(),
                              _collector.markBitMap(),
6165
                              workers, _collector.task_queues());
D
duke 已提交
6166 6167 6168 6169 6170 6171 6172
  workers->run_task(&rp_task);
}

void CMSRefProcTaskExecutor::execute(EnqueueTask& task)
{

  GenCollectedHeap* gch = GenCollectedHeap::heap();
6173
  FlexibleWorkGang* workers = gch->workers();
D
duke 已提交
6174 6175 6176 6177 6178 6179 6180 6181 6182 6183
  assert(workers != NULL, "Need parallel worker threads.");
  CMSRefEnqueueTaskProxy enq_task(task);
  workers->run_task(&enq_task);
}

void CMSCollector::refProcessingWork(bool asynch, bool clear_all_soft_refs) {

  ResourceMark rm;
  HandleMark   hm;

6184 6185 6186
  ReferenceProcessor* rp = ref_processor();
  assert(rp->span().equals(_span), "Spans should be equal");
  assert(!rp->enqueuing_is_done(), "Enqueuing should not be complete");
D
duke 已提交
6187
  // Process weak references.
6188
  rp->setup_policy(clear_all_soft_refs);
D
duke 已提交
6189 6190 6191
  verify_work_stacks_empty();

  CMSKeepAliveClosure cmsKeepAliveClosure(this, _span, &_markBitMap,
6192
                                          &_markStack, false /* !preclean */);
D
duke 已提交
6193 6194
  CMSDrainMarkingStackClosure cmsDrainMarkingStackClosure(this,
                                _span, &_markBitMap, &_markStack,
6195
                                &cmsKeepAliveClosure, false /* !preclean */);
D
duke 已提交
6196
  {
S
sla 已提交
6197 6198 6199
    GCTraceTime t("weak refs processing", PrintGCDetails, false, _gc_timer_cm);

    ReferenceProcessorStats stats;
D
duke 已提交
6200
    if (rp->processing_is_mt()) {
6201 6202 6203 6204 6205
      // Set the degree of MT here.  If the discovery is done MT, there
      // may have been a different number of threads doing the discovery
      // and a different number of discovered lists may have Ref objects.
      // That is OK as long as the Reference lists are balanced (see
      // balance_all_queues() and balance_queues()).
6206 6207 6208 6209 6210 6211 6212 6213 6214 6215 6216
      GenCollectedHeap* gch = GenCollectedHeap::heap();
      int active_workers = ParallelGCThreads;
      FlexibleWorkGang* workers = gch->workers();
      if (workers != NULL) {
        active_workers = workers->active_workers();
        // The expectation is that active_workers will have already
        // been set to a reasonable value.  If it has not been set,
        // investigate.
        assert(active_workers > 0, "Should have been set during scavenge");
      }
      rp->set_active_mt_degree(active_workers);
D
duke 已提交
6217
      CMSRefProcTaskExecutor task_executor(*this);
S
sla 已提交
6218
      stats = rp->process_discovered_references(&_is_alive_closure,
D
duke 已提交
6219 6220
                                        &cmsKeepAliveClosure,
                                        &cmsDrainMarkingStackClosure,
S
sla 已提交
6221 6222
                                        &task_executor,
                                        _gc_timer_cm);
D
duke 已提交
6223
    } else {
S
sla 已提交
6224
      stats = rp->process_discovered_references(&_is_alive_closure,
D
duke 已提交
6225 6226
                                        &cmsKeepAliveClosure,
                                        &cmsDrainMarkingStackClosure,
S
sla 已提交
6227 6228
                                        NULL,
                                        _gc_timer_cm);
D
duke 已提交
6229
    }
S
sla 已提交
6230 6231
    _gc_tracer_cm->report_gc_reference_stats(stats);

D
duke 已提交
6232 6233
  }

6234 6235 6236
  // This is the point where the entire marking should have completed.
  verify_work_stacks_empty();

6237
  if (should_unload_classes()) {
D
duke 已提交
6238
    {
S
sla 已提交
6239
      GCTraceTime t("class unloading", PrintGCDetails, false, _gc_timer_cm);
D
duke 已提交
6240

6241
      // Unload classes and purge the SystemDictionary.
D
duke 已提交
6242 6243
      bool purged_class = SystemDictionary::do_unloading(&_is_alive_closure);

6244
      // Unload nmethods.
6245
      CodeCache::do_unloading(&_is_alive_closure, purged_class);
D
duke 已提交
6246

6247
      // Prune dead klasses from subklass/sibling/implementor lists.
6248
      Klass::clean_weak_klass_links(&_is_alive_closure);
D
duke 已提交
6249 6250 6251
    }

    {
S
sla 已提交
6252
      GCTraceTime t("scrub symbol table", PrintGCDetails, false, _gc_timer_cm);
6253 6254
      // Clean up unreferenced symbols in symbol table.
      SymbolTable::unlink();
D
duke 已提交
6255 6256 6257
    }
  }

6258 6259 6260
  // CMS doesn't use the StringTable as hard roots when class unloading is turned off.
  // Need to check if we really scanned the StringTable.
  if ((roots_scanning_options() & SharedHeap::SO_Strings) == 0) {
S
sla 已提交
6261
    GCTraceTime t("scrub string table", PrintGCDetails, false, _gc_timer_cm);
6262
    // Delete entries for dead interned strings.
6263 6264 6265
    StringTable::unlink(&_is_alive_closure);
  }

D
duke 已提交
6266 6267 6268 6269 6270 6271
  // Restore any preserved marks as a result of mark stack or
  // work queue overflow
  restore_preserved_marks_if_any();  // done single-threaded for now

  rp->set_enqueuing_is_done(true);
  if (rp->processing_is_mt()) {
6272
    rp->balance_all_queues();
D
duke 已提交
6273 6274 6275 6276 6277 6278 6279 6280 6281 6282 6283 6284 6285 6286 6287 6288 6289 6290 6291 6292 6293 6294 6295 6296 6297 6298 6299 6300 6301 6302 6303 6304 6305 6306 6307 6308 6309 6310 6311 6312 6313 6314 6315
    CMSRefProcTaskExecutor task_executor(*this);
    rp->enqueue_discovered_references(&task_executor);
  } else {
    rp->enqueue_discovered_references(NULL);
  }
  rp->verify_no_references_recorded();
  assert(!rp->discovery_enabled(), "should have been disabled");
}

#ifndef PRODUCT
void CMSCollector::check_correct_thread_executing() {
  Thread* t = Thread::current();
  // Only the VM thread or the CMS thread should be here.
  assert(t->is_ConcurrentGC_thread() || t->is_VM_thread(),
         "Unexpected thread type");
  // If this is the vm thread, the foreground process
  // should not be waiting.  Note that _foregroundGCIsActive is
  // true while the foreground collector is waiting.
  if (_foregroundGCShouldWait) {
    // We cannot be the VM thread
    assert(t->is_ConcurrentGC_thread(),
           "Should be CMS thread");
  } else {
    // We can be the CMS thread only if we are in a stop-world
    // phase of CMS collection.
    if (t->is_ConcurrentGC_thread()) {
      assert(_collectorState == InitialMarking ||
             _collectorState == FinalMarking,
             "Should be a stop-world phase");
      // The CMS thread should be holding the CMS_token.
      assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
             "Potential interference with concurrently "
             "executing VM thread");
    }
  }
}
#endif

void CMSCollector::sweep(bool asynch) {
  assert(_collectorState == Sweeping, "just checking");
  check_correct_thread_executing();
  verify_work_stacks_empty();
  verify_overflow_empty();
6316
  increment_sweep_count();
6317
  TraceCMSMemoryManagerStats tms(_collectorState,GenCollectedHeap::heap()->gc_cause());
6318

6319 6320
  _inter_sweep_timer.stop();
  _inter_sweep_estimate.sample(_inter_sweep_timer.seconds());
D
duke 已提交
6321 6322
  size_policy()->avg_cms_free_at_sweep()->sample(_cmsGen->free());

6323 6324 6325
  assert(!_intra_sweep_timer.is_active(), "Should not be active");
  _intra_sweep_timer.reset();
  _intra_sweep_timer.start();
D
duke 已提交
6326 6327 6328
  if (asynch) {
    TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
    CMSPhaseAccounting pa(this, "sweep", !PrintGCDetails);
6329
    // First sweep the old gen
D
duke 已提交
6330 6331 6332 6333 6334 6335 6336 6337 6338 6339 6340
    {
      CMSTokenSyncWithLocks ts(true, _cmsGen->freelistLock(),
                               bitMapLock());
      sweepWork(_cmsGen, asynch);
    }

    // Update Universe::_heap_*_at_gc figures.
    // We need all the free list locks to make the abstract state
    // transition from Sweeping to Resetting. See detailed note
    // further below.
    {
6341
      CMSTokenSyncWithLocks ts(true, _cmsGen->freelistLock());
D
duke 已提交
6342 6343 6344 6345 6346 6347 6348 6349 6350 6351 6352 6353 6354 6355 6356 6357
      // Update heap occupancy information which is used as
      // input to soft ref clearing policy at the next gc.
      Universe::update_heap_info_at_gc();
      _collectorState = Resizing;
    }
  } else {
    // already have needed locks
    sweepWork(_cmsGen,  asynch);
    // Update heap occupancy information which is used as
    // input to soft ref clearing policy at the next gc.
    Universe::update_heap_info_at_gc();
    _collectorState = Resizing;
  }
  verify_work_stacks_empty();
  verify_overflow_empty();

6358 6359 6360 6361
  if (should_unload_classes()) {
    ClassLoaderDataGraph::purge();
  }

6362 6363 6364 6365 6366
  _intra_sweep_timer.stop();
  _intra_sweep_estimate.sample(_intra_sweep_timer.seconds());

  _inter_sweep_timer.reset();
  _inter_sweep_timer.start();
D
duke 已提交
6367

6368 6369 6370 6371 6372
  // We need to use a monotonically non-deccreasing time in ms
  // or we will see time-warp warnings and os::javaTimeMillis()
  // does not guarantee monotonicity.
  jlong now = os::javaTimeNanos() / NANOSECS_PER_MILLISEC;
  update_time_of_last_gc(now);
D
duke 已提交
6373 6374 6375 6376 6377 6378 6379 6380 6381 6382 6383 6384 6385 6386 6387 6388 6389

  // NOTE on abstract state transitions:
  // Mutators allocate-live and/or mark the mod-union table dirty
  // based on the state of the collection.  The former is done in
  // the interval [Marking, Sweeping] and the latter in the interval
  // [Marking, Sweeping).  Thus the transitions into the Marking state
  // and out of the Sweeping state must be synchronously visible
  // globally to the mutators.
  // The transition into the Marking state happens with the world
  // stopped so the mutators will globally see it.  Sweeping is
  // done asynchronously by the background collector so the transition
  // from the Sweeping state to the Resizing state must be done
  // under the freelistLock (as is the check for whether to
  // allocate-live and whether to dirty the mod-union table).
  assert(_collectorState == Resizing, "Change of collector state to"
    " Resizing must be done under the freelistLocks (plural)");

6390 6391
  // Now that sweeping has been completed, we clear
  // the incremental_collection_failed flag,
D
duke 已提交
6392 6393 6394 6395 6396
  // thus inviting a younger gen collection to promote into
  // this generation. If such a promotion may still fail,
  // the flag will be set again when a young collection is
  // attempted.
  GenCollectedHeap* gch = GenCollectedHeap::heap();
6397
  gch->clear_incremental_collection_failed();  // Worth retrying as fresh space may have been freed up
D
duke 已提交
6398 6399 6400 6401 6402 6403
  gch->update_full_collections_completed(_collection_count_start);
}

// FIX ME!!! Looks like this belongs in CFLSpace, with
// CMSGen merely delegating to it.
void ConcurrentMarkSweepGeneration::setNearLargestChunk() {
6404
  double nearLargestPercent = FLSLargestBlockCoalesceProximity;
D
duke 已提交
6405 6406
  HeapWord*  minAddr        = _cmsSpace->bottom();
  HeapWord*  largestAddr    =
6407
    (HeapWord*) _cmsSpace->dictionary()->find_largest_dict();
6408
  if (largestAddr == NULL) {
D
duke 已提交
6409 6410 6411 6412 6413 6414 6415
    // The dictionary appears to be empty.  In this case
    // try to coalesce at the end of the heap.
    largestAddr = _cmsSpace->end();
  }
  size_t largestOffset     = pointer_delta(largestAddr, minAddr);
  size_t nearLargestOffset =
    (size_t)((double)largestOffset * nearLargestPercent) - MinChunkSize;
6416 6417 6418 6419 6420 6421 6422
  if (PrintFLSStatistics != 0) {
    gclog_or_tty->print_cr(
      "CMS: Large Block: " PTR_FORMAT ";"
      " Proximity: " PTR_FORMAT " -> " PTR_FORMAT,
      largestAddr,
      _cmsSpace->nearLargestChunk(), minAddr + nearLargestOffset);
  }
D
duke 已提交
6423 6424 6425 6426 6427 6428 6429 6430 6431 6432 6433 6434 6435 6436 6437 6438 6439 6440 6441 6442 6443 6444 6445 6446 6447 6448 6449 6450 6451 6452 6453 6454 6455 6456 6457 6458 6459 6460 6461 6462 6463 6464 6465 6466 6467 6468 6469 6470 6471 6472 6473 6474 6475 6476 6477 6478 6479 6480 6481 6482 6483 6484 6485 6486 6487 6488 6489 6490 6491 6492 6493 6494 6495 6496 6497 6498 6499 6500 6501
  _cmsSpace->set_nearLargestChunk(minAddr + nearLargestOffset);
}

bool ConcurrentMarkSweepGeneration::isNearLargestChunk(HeapWord* addr) {
  return addr >= _cmsSpace->nearLargestChunk();
}

FreeChunk* ConcurrentMarkSweepGeneration::find_chunk_at_end() {
  return _cmsSpace->find_chunk_at_end();
}

void ConcurrentMarkSweepGeneration::update_gc_stats(int current_level,
                                                    bool full) {
  // The next lower level has been collected.  Gather any statistics
  // that are of interest at this point.
  if (!full && (current_level + 1) == level()) {
    // Gather statistics on the young generation collection.
    collector()->stats().record_gc0_end(used());
  }
}

CMSAdaptiveSizePolicy* ConcurrentMarkSweepGeneration::size_policy() {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  assert(gch->kind() == CollectedHeap::GenCollectedHeap,
    "Wrong type of heap");
  CMSAdaptiveSizePolicy* sp = (CMSAdaptiveSizePolicy*)
    gch->gen_policy()->size_policy();
  assert(sp->is_gc_cms_adaptive_size_policy(),
    "Wrong type of size policy");
  return sp;
}

void ConcurrentMarkSweepGeneration::rotate_debug_collection_type() {
  if (PrintGCDetails && Verbose) {
    gclog_or_tty->print("Rotate from %d ", _debug_collection_type);
  }
  _debug_collection_type = (CollectionTypes) (_debug_collection_type + 1);
  _debug_collection_type =
    (CollectionTypes) (_debug_collection_type % Unknown_collection_type);
  if (PrintGCDetails && Verbose) {
    gclog_or_tty->print_cr("to %d ", _debug_collection_type);
  }
}

void CMSCollector::sweepWork(ConcurrentMarkSweepGeneration* gen,
  bool asynch) {
  // We iterate over the space(s) underlying this generation,
  // checking the mark bit map to see if the bits corresponding
  // to specific blocks are marked or not. Blocks that are
  // marked are live and are not swept up. All remaining blocks
  // are swept up, with coalescing on-the-fly as we sweep up
  // contiguous free and/or garbage blocks:
  // We need to ensure that the sweeper synchronizes with allocators
  // and stop-the-world collectors. In particular, the following
  // locks are used:
  // . CMS token: if this is held, a stop the world collection cannot occur
  // . freelistLock: if this is held no allocation can occur from this
  //                 generation by another thread
  // . bitMapLock: if this is held, no other thread can access or update
  //

  // Note that we need to hold the freelistLock if we use
  // block iterate below; else the iterator might go awry if
  // a mutator (or promotion) causes block contents to change
  // (for instance if the allocator divvies up a block).
  // If we hold the free list lock, for all practical purposes
  // young generation GC's can't occur (they'll usually need to
  // promote), so we might as well prevent all young generation
  // GC's while we do a sweeping step. For the same reason, we might
  // as well take the bit map lock for the entire duration

  // check that we hold the requisite locks
  assert(have_cms_token(), "Should hold cms token");
  assert(   (asynch && ConcurrentMarkSweepThread::cms_thread_has_cms_token())
         || (!asynch && ConcurrentMarkSweepThread::vm_thread_has_cms_token()),
        "Should possess CMS token to sweep");
  assert_lock_strong(gen->freelistLock());
  assert_lock_strong(bitMapLock());

6502 6503 6504 6505 6506
  assert(!_inter_sweep_timer.is_active(), "Was switched off in an outer context");
  assert(_intra_sweep_timer.is_active(),  "Was switched on  in an outer context");
  gen->cmsSpace()->beginSweepFLCensus((float)(_inter_sweep_timer.seconds()),
                                      _inter_sweep_estimate.padded_average(),
                                      _intra_sweep_estimate.padded_average());
D
duke 已提交
6507 6508 6509 6510 6511 6512 6513 6514 6515 6516 6517 6518
  gen->setNearLargestChunk();

  {
    SweepClosure sweepClosure(this, gen, &_markBitMap,
                            CMSYield && asynch);
    gen->cmsSpace()->blk_iterate_careful(&sweepClosure);
    // We need to free-up/coalesce garbage/blocks from a
    // co-terminal free run. This is done in the SweepClosure
    // destructor; so, do not remove this scope, else the
    // end-of-sweep-census below will be off by a little bit.
  }
  gen->cmsSpace()->sweep_completed();
6519
  gen->cmsSpace()->endSweepFLCensus(sweep_count());
6520 6521 6522 6523 6524
  if (should_unload_classes()) {                // unloaded classes this cycle,
    _concurrent_cycles_since_last_unload = 0;   // ... reset count
  } else {                                      // did not unload classes,
    _concurrent_cycles_since_last_unload++;     // ... increment count
  }
D
duke 已提交
6525 6526 6527 6528 6529 6530 6531 6532 6533 6534 6535 6536 6537 6538 6539 6540 6541 6542 6543 6544 6545 6546 6547 6548 6549 6550 6551 6552 6553 6554 6555 6556 6557 6558 6559 6560 6561 6562 6563 6564 6565 6566 6567 6568 6569 6570
}

// Reset CMS data structures (for now just the marking bit map)
// preparatory for the next cycle.
void CMSCollector::reset(bool asynch) {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  CMSAdaptiveSizePolicy* sp = size_policy();
  AdaptiveSizePolicyOutput(sp, gch->total_collections());
  if (asynch) {
    CMSTokenSyncWithLocks ts(true, bitMapLock());

    // If the state is not "Resetting", the foreground  thread
    // has done a collection and the resetting.
    if (_collectorState != Resetting) {
      assert(_collectorState == Idling, "The state should only change"
        " because the foreground collector has finished the collection");
      return;
    }

    // Clear the mark bitmap (no grey objects to start with)
    // for the next cycle.
    TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
    CMSPhaseAccounting cmspa(this, "reset", !PrintGCDetails);

    HeapWord* curAddr = _markBitMap.startWord();
    while (curAddr < _markBitMap.endWord()) {
      size_t remaining  = pointer_delta(_markBitMap.endWord(), curAddr);
      MemRegion chunk(curAddr, MIN2(CMSBitMapYieldQuantum, remaining));
      _markBitMap.clear_large_range(chunk);
      if (ConcurrentMarkSweepThread::should_yield() &&
          !foregroundGCIsActive() &&
          CMSYield) {
        assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
               "CMS thread should hold CMS token");
        assert_lock_strong(bitMapLock());
        bitMapLock()->unlock();
        ConcurrentMarkSweepThread::desynchronize(true);
        ConcurrentMarkSweepThread::acknowledge_yield_request();
        stopTimer();
        if (PrintCMSStatistics != 0) {
          incrementYields();
        }
        icms_wait();

        // See the comment in coordinator_yield()
        for (unsigned i = 0; i < CMSYieldSleepCount &&
6571 6572
                         ConcurrentMarkSweepThread::should_yield() &&
                         !CMSCollector::foregroundGCIsActive(); ++i) {
D
duke 已提交
6573 6574 6575 6576 6577 6578 6579 6580 6581 6582
          os::sleep(Thread::current(), 1, false);
          ConcurrentMarkSweepThread::acknowledge_yield_request();
        }

        ConcurrentMarkSweepThread::synchronize(true);
        bitMapLock()->lock_without_safepoint_check();
        startTimer();
      }
      curAddr = chunk.end();
    }
6583 6584 6585 6586 6587
    // A successful mostly concurrent collection has been done.
    // Because only the full (i.e., concurrent mode failure) collections
    // are being measured for gc overhead limits, clean the "near" flag
    // and count.
    sp->reset_gc_overhead_limit_count();
D
duke 已提交
6588 6589 6590 6591 6592 6593 6594 6595 6596 6597 6598 6599 6600 6601 6602 6603 6604 6605
    _collectorState = Idling;
  } else {
    // already have the lock
    assert(_collectorState == Resetting, "just checking");
    assert_lock_strong(bitMapLock());
    _markBitMap.clear_all();
    _collectorState = Idling;
  }

  // Stop incremental mode after a cycle completes, so that any future cycles
  // are triggered by allocation.
  stop_icms();

  NOT_PRODUCT(
    if (RotateCMSCollectionTypes) {
      _cmsGen->rotate_debug_collection_type();
    }
  )
S
sla 已提交
6606 6607

  register_gc_end();
D
duke 已提交
6608 6609
}

6610
void CMSCollector::do_CMS_operation(CMS_op_type op, GCCause::Cause gc_cause) {
D
duke 已提交
6611 6612
  gclog_or_tty->date_stamp(PrintGC && PrintGCDateStamps);
  TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
S
sla 已提交
6613
  GCTraceTime t(GCCauseString("GC", gc_cause), PrintGC, !PrintGCDetails, NULL);
D
duke 已提交
6614 6615 6616 6617
  TraceCollectorStats tcs(counters());

  switch (op) {
    case CMS_op_checkpointRootsInitial: {
6618
      SvcGCMarker sgcm(SvcGCMarker::OTHER);
D
duke 已提交
6619 6620 6621 6622 6623 6624 6625
      checkpointRootsInitial(true);       // asynch
      if (PrintGC) {
        _cmsGen->printOccupancy("initial-mark");
      }
      break;
    }
    case CMS_op_checkpointRootsFinal: {
6626
      SvcGCMarker sgcm(SvcGCMarker::OTHER);
D
duke 已提交
6627 6628 6629 6630 6631 6632 6633 6634 6635 6636 6637 6638 6639 6640 6641 6642 6643 6644 6645 6646 6647 6648 6649 6650 6651 6652 6653 6654 6655 6656 6657 6658 6659 6660 6661
      checkpointRootsFinal(true,    // asynch
                           false,   // !clear_all_soft_refs
                           false);  // !init_mark_was_synchronous
      if (PrintGC) {
        _cmsGen->printOccupancy("remark");
      }
      break;
    }
    default:
      fatal("No such CMS_op");
  }
}

#ifndef PRODUCT
size_t const CMSCollector::skip_header_HeapWords() {
  return FreeChunk::header_size();
}

// Try and collect here conditions that should hold when
// CMS thread is exiting. The idea is that the foreground GC
// thread should not be blocked if it wants to terminate
// the CMS thread and yet continue to run the VM for a while
// after that.
void CMSCollector::verify_ok_to_terminate() const {
  assert(Thread::current()->is_ConcurrentGC_thread(),
         "should be called by CMS thread");
  assert(!_foregroundGCShouldWait, "should be false");
  // We could check here that all the various low-level locks
  // are not held by the CMS thread, but that is overkill; see
  // also CMSThread::verify_ok_to_terminate() where the CGC_lock
  // is checked.
}
#endif

size_t CMSCollector::block_size_using_printezis_bits(HeapWord* addr) const {
6662 6663
   assert(_markBitMap.isMarked(addr) && _markBitMap.isMarked(addr + 1),
          "missing Printezis mark?");
D
duke 已提交
6664 6665 6666 6667 6668 6669 6670 6671 6672 6673 6674
  HeapWord* nextOneAddr = _markBitMap.getNextMarkedWordAddress(addr + 2);
  size_t size = pointer_delta(nextOneAddr + 1, addr);
  assert(size == CompactibleFreeListSpace::adjustObjectSize(size),
         "alignment problem");
  assert(size >= 3, "Necessary for Printezis marks to work");
  return size;
}

// A variant of the above (block_size_using_printezis_bits()) except
// that we return 0 if the P-bits are not yet set.
size_t CMSCollector::block_size_if_printezis_bits(HeapWord* addr) const {
6675 6676
  if (_markBitMap.isMarked(addr + 1)) {
    assert(_markBitMap.isMarked(addr), "P-bit can be set only for marked objects");
D
duke 已提交
6677 6678 6679 6680 6681 6682 6683
    HeapWord* nextOneAddr = _markBitMap.getNextMarkedWordAddress(addr + 2);
    size_t size = pointer_delta(nextOneAddr + 1, addr);
    assert(size == CompactibleFreeListSpace::adjustObjectSize(size),
           "alignment problem");
    assert(size >= 3, "Necessary for Printezis marks to work");
    return size;
  }
6684
  return 0;
D
duke 已提交
6685 6686 6687 6688 6689
}

HeapWord* CMSCollector::next_card_start_after_block(HeapWord* addr) const {
  size_t sz = 0;
  oop p = (oop)addr;
6690
  if (p->klass_or_null() != NULL) {
D
duke 已提交
6691 6692 6693 6694 6695 6696 6697 6698 6699 6700 6701 6702 6703 6704 6705 6706 6707 6708 6709 6710 6711
    sz = CompactibleFreeListSpace::adjustObjectSize(p->size());
  } else {
    sz = block_size_using_printezis_bits(addr);
  }
  assert(sz > 0, "size must be nonzero");
  HeapWord* next_block = addr + sz;
  HeapWord* next_card  = (HeapWord*)round_to((uintptr_t)next_block,
                                             CardTableModRefBS::card_size);
  assert(round_down((uintptr_t)addr,      CardTableModRefBS::card_size) <
         round_down((uintptr_t)next_card, CardTableModRefBS::card_size),
         "must be different cards");
  return next_card;
}


// CMS Bit Map Wrapper /////////////////////////////////////////

// Construct a CMS bit map infrastructure, but don't create the
// bit vector itself. That is done by a separate call CMSBitMap::allocate()
// further below.
CMSBitMap::CMSBitMap(int shifter, int mutex_rank, const char* mutex_name):
6712
  _bm(),
D
duke 已提交
6713 6714 6715 6716 6717 6718 6719 6720 6721 6722 6723 6724 6725 6726 6727 6728 6729 6730 6731 6732 6733 6734 6735 6736
  _shifter(shifter),
  _lock(mutex_rank >= 0 ? new Mutex(mutex_rank, mutex_name, true) : NULL)
{
  _bmStartWord = 0;
  _bmWordSize  = 0;
}

bool CMSBitMap::allocate(MemRegion mr) {
  _bmStartWord = mr.start();
  _bmWordSize  = mr.word_size();
  ReservedSpace brs(ReservedSpace::allocation_align_size_up(
                     (_bmWordSize >> (_shifter + LogBitsPerByte)) + 1));
  if (!brs.is_reserved()) {
    warning("CMS bit map allocation failure");
    return false;
  }
  // For now we'll just commit all of the bit map up fromt.
  // Later on we'll try to be more parsimonious with swap.
  if (!_virtual_space.initialize(brs, brs.size())) {
    warning("CMS bit map backing store failure");
    return false;
  }
  assert(_virtual_space.committed_size() == brs.size(),
         "didn't reserve backing store for all of CMS bit map?");
6737
  _bm.set_map((BitMap::bm_word_t*)_virtual_space.low());
D
duke 已提交
6738 6739 6740 6741 6742 6743 6744 6745 6746 6747 6748 6749 6750 6751 6752 6753 6754 6755 6756 6757 6758 6759 6760 6761 6762 6763 6764 6765 6766 6767
  assert(_virtual_space.committed_size() << (_shifter + LogBitsPerByte) >=
         _bmWordSize, "inconsistency in bit map sizing");
  _bm.set_size(_bmWordSize >> _shifter);

  // bm.clear(); // can we rely on getting zero'd memory? verify below
  assert(isAllClear(),
         "Expected zero'd memory from ReservedSpace constructor");
  assert(_bm.size() == heapWordDiffToOffsetDiff(sizeInWords()),
         "consistency check");
  return true;
}

void CMSBitMap::dirty_range_iterate_clear(MemRegion mr, MemRegionClosure* cl) {
  HeapWord *next_addr, *end_addr, *last_addr;
  assert_locked();
  assert(covers(mr), "out-of-range error");
  // XXX assert that start and end are appropriately aligned
  for (next_addr = mr.start(), end_addr = mr.end();
       next_addr < end_addr; next_addr = last_addr) {
    MemRegion dirty_region = getAndClearMarkedRegion(next_addr, end_addr);
    last_addr = dirty_region.end();
    if (!dirty_region.is_empty()) {
      cl->do_MemRegion(dirty_region);
    } else {
      assert(last_addr == end_addr, "program logic");
      return;
    }
  }
}

6768 6769 6770 6771
void CMSBitMap::print_on_error(outputStream* st, const char* prefix) const {
  _bm.print_on_error(st, prefix);
}

D
duke 已提交
6772 6773 6774 6775 6776 6777 6778 6779 6780 6781 6782 6783 6784 6785 6786 6787 6788 6789 6790 6791 6792 6793 6794 6795 6796 6797 6798 6799 6800 6801 6802 6803 6804 6805 6806 6807 6808 6809 6810 6811 6812 6813 6814 6815 6816 6817 6818 6819 6820 6821 6822 6823 6824 6825 6826 6827 6828 6829 6830 6831 6832 6833 6834 6835 6836 6837 6838 6839 6840 6841
#ifndef PRODUCT
void CMSBitMap::assert_locked() const {
  CMSLockVerifier::assert_locked(lock());
}

bool CMSBitMap::covers(MemRegion mr) const {
  // assert(_bm.map() == _virtual_space.low(), "map inconsistency");
  assert((size_t)_bm.size() == (_bmWordSize >> _shifter),
         "size inconsistency");
  return (mr.start() >= _bmStartWord) &&
         (mr.end()   <= endWord());
}

bool CMSBitMap::covers(HeapWord* start, size_t size) const {
    return (start >= _bmStartWord && (start + size) <= endWord());
}

void CMSBitMap::verifyNoOneBitsInRange(HeapWord* left, HeapWord* right) {
  // verify that there are no 1 bits in the interval [left, right)
  FalseBitMapClosure falseBitMapClosure;
  iterate(&falseBitMapClosure, left, right);
}

void CMSBitMap::region_invariant(MemRegion mr)
{
  assert_locked();
  // mr = mr.intersection(MemRegion(_bmStartWord, _bmWordSize));
  assert(!mr.is_empty(), "unexpected empty region");
  assert(covers(mr), "mr should be covered by bit map");
  // convert address range into offset range
  size_t start_ofs = heapWordToOffset(mr.start());
  // Make sure that end() is appropriately aligned
  assert(mr.end() == (HeapWord*)round_to((intptr_t)mr.end(),
                        (1 << (_shifter+LogHeapWordSize))),
         "Misaligned mr.end()");
  size_t end_ofs   = heapWordToOffset(mr.end());
  assert(end_ofs > start_ofs, "Should mark at least one bit");
}

#endif

bool CMSMarkStack::allocate(size_t size) {
  // allocate a stack of the requisite depth
  ReservedSpace rs(ReservedSpace::allocation_align_size_up(
                   size * sizeof(oop)));
  if (!rs.is_reserved()) {
    warning("CMSMarkStack allocation failure");
    return false;
  }
  if (!_virtual_space.initialize(rs, rs.size())) {
    warning("CMSMarkStack backing store failure");
    return false;
  }
  assert(_virtual_space.committed_size() == rs.size(),
         "didn't reserve backing store for all of CMS stack?");
  _base = (oop*)(_virtual_space.low());
  _index = 0;
  _capacity = size;
  NOT_PRODUCT(_max_depth = 0);
  return true;
}

// XXX FIX ME !!! In the MT case we come in here holding a
// leaf lock. For printing we need to take a further lock
// which has lower rank. We need to recallibrate the two
// lock-ranks involved in order to be able to rpint the
// messages below. (Or defer the printing to the caller.
// For now we take the expedient path of just disabling the
// messages for the problematic case.)
void CMSMarkStack::expand() {
6842 6843
  assert(_capacity <= MarkStackSizeMax, "stack bigger than permitted");
  if (_capacity == MarkStackSizeMax) {
D
duke 已提交
6844 6845 6846 6847 6848 6849 6850
    if (_hit_limit++ == 0 && !CMSConcurrentMTEnabled && PrintGCDetails) {
      // We print a warning message only once per CMS cycle.
      gclog_or_tty->print_cr(" (benign) Hit CMSMarkStack max size limit");
    }
    return;
  }
  // Double capacity if possible
6851
  size_t new_capacity = MIN2(_capacity*2, MarkStackSizeMax);
D
duke 已提交
6852 6853 6854 6855 6856 6857 6858 6859 6860 6861 6862 6863 6864 6865 6866 6867 6868 6869 6870 6871 6872 6873 6874 6875 6876 6877 6878 6879 6880 6881 6882
  // Do not give up existing stack until we have managed to
  // get the double capacity that we desired.
  ReservedSpace rs(ReservedSpace::allocation_align_size_up(
                   new_capacity * sizeof(oop)));
  if (rs.is_reserved()) {
    // Release the backing store associated with old stack
    _virtual_space.release();
    // Reinitialize virtual space for new stack
    if (!_virtual_space.initialize(rs, rs.size())) {
      fatal("Not enough swap for expanded marking stack");
    }
    _base = (oop*)(_virtual_space.low());
    _index = 0;
    _capacity = new_capacity;
  } else if (_failed_double++ == 0 && !CMSConcurrentMTEnabled && PrintGCDetails) {
    // Failed to double capacity, continue;
    // we print a detail message only once per CMS cycle.
    gclog_or_tty->print(" (benign) Failed to expand marking stack from "SIZE_FORMAT"K to "
            SIZE_FORMAT"K",
            _capacity / K, new_capacity / K);
  }
}


// Closures
// XXX: there seems to be a lot of code  duplication here;
// should refactor and consolidate common code.

// This closure is used to mark refs into the CMS generation in
// the CMS bit map. Called at the first checkpoint. This closure
// assumes that we do not need to re-mark dirty cards; if the CMS
6883
// generation on which this is used is not an oldest
D
duke 已提交
6884 6885 6886
// generation then this will lose younger_gen cards!

MarkRefsIntoClosure::MarkRefsIntoClosure(
6887
  MemRegion span, CMSBitMap* bitMap):
D
duke 已提交
6888
    _span(span),
6889
    _bitMap(bitMap)
D
duke 已提交
6890 6891 6892 6893 6894
{
    assert(_ref_processor == NULL, "deliberately left NULL");
    assert(_bitMap->covers(_span), "_bitMap/_span mismatch");
}

6895
void MarkRefsIntoClosure::do_oop(oop obj) {
D
duke 已提交
6896
  // if p points into _span, then mark corresponding bit in _markBitMap
6897 6898 6899 6900 6901
  assert(obj->is_oop(), "expected an oop");
  HeapWord* addr = (HeapWord*)obj;
  if (_span.contains(addr)) {
    // this should be made more efficient
    _bitMap->mark(addr);
D
duke 已提交
6902 6903 6904
  }
}

6905 6906 6907
void MarkRefsIntoClosure::do_oop(oop* p)       { MarkRefsIntoClosure::do_oop_work(p); }
void MarkRefsIntoClosure::do_oop(narrowOop* p) { MarkRefsIntoClosure::do_oop_work(p); }

J
jmasa 已提交
6908 6909 6910 6911 6912 6913 6914 6915 6916 6917 6918 6919 6920 6921 6922 6923 6924 6925 6926 6927 6928 6929
Par_MarkRefsIntoClosure::Par_MarkRefsIntoClosure(
  MemRegion span, CMSBitMap* bitMap):
    _span(span),
    _bitMap(bitMap)
{
    assert(_ref_processor == NULL, "deliberately left NULL");
    assert(_bitMap->covers(_span), "_bitMap/_span mismatch");
}

void Par_MarkRefsIntoClosure::do_oop(oop obj) {
  // if p points into _span, then mark corresponding bit in _markBitMap
  assert(obj->is_oop(), "expected an oop");
  HeapWord* addr = (HeapWord*)obj;
  if (_span.contains(addr)) {
    // this should be made more efficient
    _bitMap->par_mark(addr);
  }
}

void Par_MarkRefsIntoClosure::do_oop(oop* p)       { Par_MarkRefsIntoClosure::do_oop_work(p); }
void Par_MarkRefsIntoClosure::do_oop(narrowOop* p) { Par_MarkRefsIntoClosure::do_oop_work(p); }

D
duke 已提交
6930 6931
// A variant of the above, used for CMS marking verification.
MarkRefsIntoVerifyClosure::MarkRefsIntoVerifyClosure(
6932
  MemRegion span, CMSBitMap* verification_bm, CMSBitMap* cms_bm):
D
duke 已提交
6933 6934
    _span(span),
    _verification_bm(verification_bm),
6935 6936
    _cms_bm(cms_bm)
{
D
duke 已提交
6937 6938 6939 6940
    assert(_ref_processor == NULL, "deliberately left NULL");
    assert(_verification_bm->covers(_span), "_verification_bm/_span mismatch");
}

6941
void MarkRefsIntoVerifyClosure::do_oop(oop obj) {
D
duke 已提交
6942
  // if p points into _span, then mark corresponding bit in _markBitMap
6943 6944 6945 6946 6947 6948 6949 6950
  assert(obj->is_oop(), "expected an oop");
  HeapWord* addr = (HeapWord*)obj;
  if (_span.contains(addr)) {
    _verification_bm->mark(addr);
    if (!_cms_bm->isMarked(addr)) {
      oop(addr)->print();
      gclog_or_tty->print_cr(" (" INTPTR_FORMAT " should have been marked)", addr);
      fatal("... aborting");
D
duke 已提交
6951 6952 6953 6954
    }
  }
}

6955 6956 6957
void MarkRefsIntoVerifyClosure::do_oop(oop* p)       { MarkRefsIntoVerifyClosure::do_oop_work(p); }
void MarkRefsIntoVerifyClosure::do_oop(narrowOop* p) { MarkRefsIntoVerifyClosure::do_oop_work(p); }

D
duke 已提交
6958 6959 6960 6961 6962 6963 6964 6965 6966 6967 6968 6969 6970 6971 6972 6973 6974
//////////////////////////////////////////////////
// MarkRefsIntoAndScanClosure
//////////////////////////////////////////////////

MarkRefsIntoAndScanClosure::MarkRefsIntoAndScanClosure(MemRegion span,
                                                       ReferenceProcessor* rp,
                                                       CMSBitMap* bit_map,
                                                       CMSBitMap* mod_union_table,
                                                       CMSMarkStack*  mark_stack,
                                                       CMSCollector* collector,
                                                       bool should_yield,
                                                       bool concurrent_precleaning):
  _collector(collector),
  _span(span),
  _bit_map(bit_map),
  _mark_stack(mark_stack),
  _pushAndMarkClosure(collector, span, rp, bit_map, mod_union_table,
6975
                      mark_stack, concurrent_precleaning),
D
duke 已提交
6976 6977 6978 6979 6980 6981 6982 6983 6984 6985 6986 6987 6988 6989 6990
  _yield(should_yield),
  _concurrent_precleaning(concurrent_precleaning),
  _freelistLock(NULL)
{
  _ref_processor = rp;
  assert(_ref_processor != NULL, "_ref_processor shouldn't be NULL");
}

// This closure is used to mark refs into the CMS generation at the
// second (final) checkpoint, and to scan and transitively follow
// the unmarked oops. It is also used during the concurrent precleaning
// phase while scanning objects on dirty cards in the CMS generation.
// The marks are made in the marking bit map and the marking stack is
// used for keeping the (newly) grey objects during the scan.
// The parallel version (Par_...) appears further below.
6991 6992 6993 6994 6995 6996 6997
void MarkRefsIntoAndScanClosure::do_oop(oop obj) {
  if (obj != NULL) {
    assert(obj->is_oop(), "expected an oop");
    HeapWord* addr = (HeapWord*)obj;
    assert(_mark_stack->isEmpty(), "pre-condition (eager drainage)");
    assert(_collector->overflow_list_is_empty(),
           "overflow list should be empty");
D
duke 已提交
6998 6999 7000 7001 7002 7003 7004
    if (_span.contains(addr) &&
        !_bit_map->isMarked(addr)) {
      // mark bit map (object is now grey)
      _bit_map->mark(addr);
      // push on marking stack (stack should be empty), and drain the
      // stack by applying this closure to the oops in the oops popped
      // from the stack (i.e. blacken the grey objects)
7005
      bool res = _mark_stack->push(obj);
D
duke 已提交
7006 7007 7008 7009 7010 7011 7012 7013 7014 7015 7016 7017 7018 7019 7020 7021 7022 7023 7024 7025 7026 7027 7028 7029 7030 7031 7032 7033 7034 7035 7036 7037 7038 7039
      assert(res, "Should have space to push on empty stack");
      do {
        oop new_oop = _mark_stack->pop();
        assert(new_oop != NULL && new_oop->is_oop(), "Expected an oop");
        assert(_bit_map->isMarked((HeapWord*)new_oop),
               "only grey objects on this stack");
        // iterate over the oops in this oop, marking and pushing
        // the ones in CMS heap (i.e. in _span).
        new_oop->oop_iterate(&_pushAndMarkClosure);
        // check if it's time to yield
        do_yield_check();
      } while (!_mark_stack->isEmpty() ||
               (!_concurrent_precleaning && take_from_overflow_list()));
        // if marking stack is empty, and we are not doing this
        // during precleaning, then check the overflow list
    }
    assert(_mark_stack->isEmpty(), "post-condition (eager drainage)");
    assert(_collector->overflow_list_is_empty(),
           "overflow list was drained above");
    // We could restore evacuated mark words, if any, used for
    // overflow list links here because the overflow list is
    // provably empty here. That would reduce the maximum
    // size requirements for preserved_{oop,mark}_stack.
    // But we'll just postpone it until we are all done
    // so we can just stream through.
    if (!_concurrent_precleaning && CMSOverflowEarlyRestoration) {
      _collector->restore_preserved_marks_if_any();
      assert(_collector->no_preserved_marks(), "No preserved marks");
    }
    assert(!CMSOverflowEarlyRestoration || _collector->no_preserved_marks(),
           "All preserved marks should have been restored above");
  }
}

7040 7041 7042
void MarkRefsIntoAndScanClosure::do_oop(oop* p)       { MarkRefsIntoAndScanClosure::do_oop_work(p); }
void MarkRefsIntoAndScanClosure::do_oop(narrowOop* p) { MarkRefsIntoAndScanClosure::do_oop_work(p); }

D
duke 已提交
7043 7044 7045 7046 7047 7048 7049 7050 7051 7052 7053 7054 7055 7056 7057 7058 7059 7060
void MarkRefsIntoAndScanClosure::do_yield_work() {
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  assert_lock_strong(_freelistLock);
  assert_lock_strong(_bit_map->lock());
  // relinquish the free_list_lock and bitMaplock()
  _bit_map->lock()->unlock();
  _freelistLock->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
7061 7062 7063 7064 7065
  for (unsigned i = 0;
       i < CMSYieldSleepCount &&
       ConcurrentMarkSweepThread::should_yield() &&
       !CMSCollector::foregroundGCIsActive();
       ++i) {
D
duke 已提交
7066 7067 7068 7069 7070 7071 7072 7073 7074 7075 7076 7077 7078 7079 7080 7081
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _freelistLock->lock_without_safepoint_check();
  _bit_map->lock()->lock_without_safepoint_check();
  _collector->startTimer();
}

///////////////////////////////////////////////////////////
// Par_MarkRefsIntoAndScanClosure: a parallel version of
//                                 MarkRefsIntoAndScanClosure
///////////////////////////////////////////////////////////
Par_MarkRefsIntoAndScanClosure::Par_MarkRefsIntoAndScanClosure(
  CMSCollector* collector, MemRegion span, ReferenceProcessor* rp,
7082
  CMSBitMap* bit_map, OopTaskQueue* work_queue):
D
duke 已提交
7083 7084 7085 7086 7087
  _span(span),
  _bit_map(bit_map),
  _work_queue(work_queue),
  _low_water_mark(MIN2((uint)(work_queue->max_elems()/4),
                       (uint)(CMSWorkQueueDrainThreshold * ParallelGCThreads))),
7088
  _par_pushAndMarkClosure(collector, span, rp, bit_map, work_queue)
D
duke 已提交
7089 7090 7091 7092 7093 7094 7095 7096 7097 7098 7099 7100
{
  _ref_processor = rp;
  assert(_ref_processor != NULL, "_ref_processor shouldn't be NULL");
}

// This closure is used to mark refs into the CMS generation at the
// second (final) checkpoint, and to scan and transitively follow
// the unmarked oops. The marks are made in the marking bit map and
// the work_queue is used for keeping the (newly) grey objects during
// the scan phase whence they are also available for stealing by parallel
// threads. Since the marking bit map is shared, updates are
// synchronized (via CAS).
7101 7102
void Par_MarkRefsIntoAndScanClosure::do_oop(oop obj) {
  if (obj != NULL) {
D
duke 已提交
7103 7104
    // Ignore mark word because this could be an already marked oop
    // that may be chained at the end of the overflow list.
7105
    assert(obj->is_oop(true), "expected an oop");
7106
    HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
7107 7108 7109 7110 7111 7112 7113 7114 7115 7116 7117 7118 7119
    if (_span.contains(addr) &&
        !_bit_map->isMarked(addr)) {
      // mark bit map (object will become grey):
      // It is possible for several threads to be
      // trying to "claim" this object concurrently;
      // the unique thread that succeeds in marking the
      // object first will do the subsequent push on
      // to the work queue (or overflow list).
      if (_bit_map->par_mark(addr)) {
        // push on work_queue (which may not be empty), and trim the
        // queue to an appropriate length by applying this closure to
        // the oops in the oops popped from the stack (i.e. blacken the
        // grey objects)
7120
        bool res = _work_queue->push(obj);
D
duke 已提交
7121 7122 7123 7124 7125 7126 7127
        assert(res, "Low water mark should be less than capacity?");
        trim_queue(_low_water_mark);
      } // Else, another thread claimed the object
    }
  }
}

7128 7129 7130
void Par_MarkRefsIntoAndScanClosure::do_oop(oop* p)       { Par_MarkRefsIntoAndScanClosure::do_oop_work(p); }
void Par_MarkRefsIntoAndScanClosure::do_oop(narrowOop* p) { Par_MarkRefsIntoAndScanClosure::do_oop_work(p); }

D
duke 已提交
7131 7132 7133 7134 7135 7136 7137 7138 7139 7140 7141 7142 7143 7144 7145 7146 7147
// This closure is used to rescan the marked objects on the dirty cards
// in the mod union table and the card table proper.
size_t ScanMarkedObjectsAgainCarefullyClosure::do_object_careful_m(
  oop p, MemRegion mr) {

  size_t size = 0;
  HeapWord* addr = (HeapWord*)p;
  DEBUG_ONLY(_collector->verify_work_stacks_empty();)
  assert(_span.contains(addr), "we are scanning the CMS generation");
  // check if it's time to yield
  if (do_yield_check()) {
    // We yielded for some foreground stop-world work,
    // and we have been asked to abort this ongoing preclean cycle.
    return 0;
  }
  if (_bitMap->isMarked(addr)) {
    // it's marked; is it potentially uninitialized?
7148
    if (p->klass_or_null() != NULL) {
D
duke 已提交
7149 7150 7151 7152 7153 7154
        // an initialized object; ignore mark word in verification below
        // since we are running concurrent with mutators
        assert(p->is_oop(true), "should be an oop");
        if (p->is_objArray()) {
          // objArrays are precisely marked; restrict scanning
          // to dirty cards only.
7155 7156
          size = CompactibleFreeListSpace::adjustObjectSize(
                   p->oop_iterate(_scanningClosure, mr));
D
duke 已提交
7157 7158 7159 7160 7161 7162
        } else {
          // A non-array may have been imprecisely marked; we need
          // to scan object in its entirety.
          size = CompactibleFreeListSpace::adjustObjectSize(
                   p->oop_iterate(_scanningClosure));
        }
D
drchase 已提交
7163
        #ifdef ASSERT
D
duke 已提交
7164 7165 7166 7167 7168 7169 7170 7171 7172 7173 7174
          size_t direct_size =
            CompactibleFreeListSpace::adjustObjectSize(p->size());
          assert(size == direct_size, "Inconsistency in size");
          assert(size >= 3, "Necessary for Printezis marks to work");
          if (!_bitMap->isMarked(addr+1)) {
            _bitMap->verifyNoOneBitsInRange(addr+2, addr+size);
          } else {
            _bitMap->verifyNoOneBitsInRange(addr+2, addr+size-1);
            assert(_bitMap->isMarked(addr+size-1),
                   "inconsistent Printezis mark");
          }
D
drchase 已提交
7175
        #endif // ASSERT
D
duke 已提交
7176 7177 7178 7179 7180 7181 7182 7183 7184 7185 7186 7187 7188
    } else {
      // an unitialized object
      assert(_bitMap->isMarked(addr+1), "missing Printezis mark?");
      HeapWord* nextOneAddr = _bitMap->getNextMarkedWordAddress(addr + 2);
      size = pointer_delta(nextOneAddr + 1, addr);
      assert(size == CompactibleFreeListSpace::adjustObjectSize(size),
             "alignment problem");
      // Note that pre-cleaning needn't redirty the card. OopDesc::set_klass()
      // will dirty the card when the klass pointer is installed in the
      // object (signalling the completion of initialization).
    }
  } else {
    // Either a not yet marked object or an uninitialized object
7189
    if (p->klass_or_null() == NULL) {
D
duke 已提交
7190 7191 7192 7193 7194 7195 7196 7197 7198 7199 7200 7201 7202 7203 7204 7205 7206 7207 7208 7209 7210 7211 7212 7213 7214 7215 7216 7217 7218 7219 7220 7221 7222
      // An uninitialized object, skip to the next card, since
      // we may not be able to read its P-bits yet.
      assert(size == 0, "Initial value");
    } else {
      // An object not (yet) reached by marking: we merely need to
      // compute its size so as to go look at the next block.
      assert(p->is_oop(true), "should be an oop");
      size = CompactibleFreeListSpace::adjustObjectSize(p->size());
    }
  }
  DEBUG_ONLY(_collector->verify_work_stacks_empty();)
  return size;
}

void ScanMarkedObjectsAgainCarefullyClosure::do_yield_work() {
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  assert_lock_strong(_freelistLock);
  assert_lock_strong(_bitMap->lock());
  // relinquish the free_list_lock and bitMaplock()
  _bitMap->lock()->unlock();
  _freelistLock->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
  for (unsigned i = 0; i < CMSYieldSleepCount &&
7223 7224
                   ConcurrentMarkSweepThread::should_yield() &&
                   !CMSCollector::foregroundGCIsActive(); ++i) {
D
duke 已提交
7225 7226 7227 7228 7229 7230 7231 7232 7233 7234 7235 7236 7237 7238 7239 7240 7241 7242 7243 7244 7245
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _freelistLock->lock_without_safepoint_check();
  _bitMap->lock()->lock_without_safepoint_check();
  _collector->startTimer();
}


//////////////////////////////////////////////////////////////////
// SurvivorSpacePrecleanClosure
//////////////////////////////////////////////////////////////////
// This (single-threaded) closure is used to preclean the oops in
// the survivor spaces.
size_t SurvivorSpacePrecleanClosure::do_object_careful(oop p) {

  HeapWord* addr = (HeapWord*)p;
  DEBUG_ONLY(_collector->verify_work_stacks_empty();)
  assert(!_span.contains(addr), "we are scanning the survivor spaces");
7246
  assert(p->klass_or_null() != NULL, "object should be initializd");
D
duke 已提交
7247 7248 7249 7250 7251 7252 7253 7254 7255 7256 7257 7258 7259 7260 7261 7262 7263 7264 7265 7266 7267 7268 7269 7270 7271 7272 7273 7274 7275 7276 7277 7278 7279 7280 7281 7282 7283 7284 7285 7286 7287 7288 7289 7290 7291 7292 7293 7294 7295 7296 7297 7298 7299 7300 7301 7302 7303 7304 7305 7306 7307 7308 7309 7310 7311 7312 7313 7314 7315 7316
  // an initialized object; ignore mark word in verification below
  // since we are running concurrent with mutators
  assert(p->is_oop(true), "should be an oop");
  // Note that we do not yield while we iterate over
  // the interior oops of p, pushing the relevant ones
  // on our marking stack.
  size_t size = p->oop_iterate(_scanning_closure);
  do_yield_check();
  // Observe that below, we do not abandon the preclean
  // phase as soon as we should; rather we empty the
  // marking stack before returning. This is to satisfy
  // some existing assertions. In general, it may be a
  // good idea to abort immediately and complete the marking
  // from the grey objects at a later time.
  while (!_mark_stack->isEmpty()) {
    oop new_oop = _mark_stack->pop();
    assert(new_oop != NULL && new_oop->is_oop(), "Expected an oop");
    assert(_bit_map->isMarked((HeapWord*)new_oop),
           "only grey objects on this stack");
    // iterate over the oops in this oop, marking and pushing
    // the ones in CMS heap (i.e. in _span).
    new_oop->oop_iterate(_scanning_closure);
    // check if it's time to yield
    do_yield_check();
  }
  unsigned int after_count =
    GenCollectedHeap::heap()->total_collections();
  bool abort = (_before_count != after_count) ||
               _collector->should_abort_preclean();
  return abort ? 0 : size;
}

void SurvivorSpacePrecleanClosure::do_yield_work() {
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  assert_lock_strong(_bit_map->lock());
  // Relinquish the bit map lock
  _bit_map->lock()->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
  for (unsigned i = 0; i < CMSYieldSleepCount &&
                       ConcurrentMarkSweepThread::should_yield() &&
                       !CMSCollector::foregroundGCIsActive(); ++i) {
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _bit_map->lock()->lock_without_safepoint_check();
  _collector->startTimer();
}

// This closure is used to rescan the marked objects on the dirty cards
// in the mod union table and the card table proper. In the parallel
// case, although the bitMap is shared, we do a single read so the
// isMarked() query is "safe".
bool ScanMarkedObjectsAgainClosure::do_object_bm(oop p, MemRegion mr) {
  // Ignore mark word because we are running concurrent with mutators
  assert(p->is_oop_or_null(true), "expected an oop or null");
  HeapWord* addr = (HeapWord*)p;
  assert(_span.contains(addr), "we are scanning the CMS generation");
  bool is_obj_array = false;
D
drchase 已提交
7317
  #ifdef ASSERT
D
duke 已提交
7318 7319 7320 7321 7322 7323
    if (!_parallel) {
      assert(_mark_stack->isEmpty(), "pre-condition (eager drainage)");
      assert(_collector->overflow_list_is_empty(),
             "overflow list should be empty");

    }
D
drchase 已提交
7324
  #endif // ASSERT
D
duke 已提交
7325 7326 7327 7328 7329 7330 7331 7332 7333 7334 7335 7336 7337 7338 7339 7340 7341 7342 7343
  if (_bit_map->isMarked(addr)) {
    // Obj arrays are precisely marked, non-arrays are not;
    // so we scan objArrays precisely and non-arrays in their
    // entirety.
    if (p->is_objArray()) {
      is_obj_array = true;
      if (_parallel) {
        p->oop_iterate(_par_scan_closure, mr);
      } else {
        p->oop_iterate(_scan_closure, mr);
      }
    } else {
      if (_parallel) {
        p->oop_iterate(_par_scan_closure);
      } else {
        p->oop_iterate(_scan_closure);
      }
    }
  }
D
drchase 已提交
7344
  #ifdef ASSERT
D
duke 已提交
7345 7346 7347 7348 7349 7350
    if (!_parallel) {
      assert(_mark_stack->isEmpty(), "post-condition (eager drainage)");
      assert(_collector->overflow_list_is_empty(),
             "overflow list should be empty");

    }
D
drchase 已提交
7351
  #endif // ASSERT
D
duke 已提交
7352 7353 7354 7355 7356 7357 7358 7359 7360 7361 7362 7363 7364 7365 7366 7367 7368 7369 7370 7371 7372 7373 7374 7375 7376 7377 7378 7379 7380 7381 7382 7383 7384
  return is_obj_array;
}

MarkFromRootsClosure::MarkFromRootsClosure(CMSCollector* collector,
                        MemRegion span,
                        CMSBitMap* bitMap, CMSMarkStack*  markStack,
                        bool should_yield, bool verifying):
  _collector(collector),
  _span(span),
  _bitMap(bitMap),
  _mut(&collector->_modUnionTable),
  _markStack(markStack),
  _yield(should_yield),
  _skipBits(0)
{
  assert(_markStack->isEmpty(), "stack should be empty");
  _finger = _bitMap->startWord();
  _threshold = _finger;
  assert(_collector->_restart_addr == NULL, "Sanity check");
  assert(_span.contains(_finger), "Out of bounds _finger?");
  DEBUG_ONLY(_verifying = verifying;)
}

void MarkFromRootsClosure::reset(HeapWord* addr) {
  assert(_markStack->isEmpty(), "would cause duplicates on stack");
  assert(_span.contains(addr), "Out of bounds _finger?");
  _finger = addr;
  _threshold = (HeapWord*)round_to(
                 (intptr_t)_finger, CardTableModRefBS::card_size);
}

// Should revisit to see if this should be restructured for
// greater efficiency.
7385
bool MarkFromRootsClosure::do_bit(size_t offset) {
D
duke 已提交
7386 7387
  if (_skipBits > 0) {
    _skipBits--;
7388
    return true;
D
duke 已提交
7389 7390 7391 7392 7393 7394 7395 7396 7397 7398 7399
  }
  // convert offset into a HeapWord*
  HeapWord* addr = _bitMap->startWord() + offset;
  assert(_bitMap->endWord() && addr < _bitMap->endWord(),
         "address out of range");
  assert(_bitMap->isMarked(addr), "tautology");
  if (_bitMap->isMarked(addr+1)) {
    // this is an allocated but not yet initialized object
    assert(_skipBits == 0, "tautology");
    _skipBits = 2;  // skip next two marked bits ("Printezis-marks")
    oop p = oop(addr);
7400
    if (p->klass_or_null() == NULL) {
D
duke 已提交
7401 7402 7403 7404 7405 7406 7407 7408
      DEBUG_ONLY(if (!_verifying) {)
        // We re-dirty the cards on which this object lies and increase
        // the _threshold so that we'll come back to scan this object
        // during the preclean or remark phase. (CMSCleanOnEnter)
        if (CMSCleanOnEnter) {
          size_t sz = _collector->block_size_using_printezis_bits(addr);
          HeapWord* end_card_addr   = (HeapWord*)round_to(
                                         (intptr_t)(addr+sz), CardTableModRefBS::card_size);
7409
          MemRegion redirty_range = MemRegion(addr, end_card_addr);
D
duke 已提交
7410 7411 7412 7413 7414 7415 7416 7417 7418 7419
          assert(!redirty_range.is_empty(), "Arithmetical tautology");
          // Bump _threshold to end_card_addr; note that
          // _threshold cannot possibly exceed end_card_addr, anyhow.
          // This prevents future clearing of the card as the scan proceeds
          // to the right.
          assert(_threshold <= end_card_addr,
                 "Because we are just scanning into this object");
          if (_threshold < end_card_addr) {
            _threshold = end_card_addr;
          }
7420
          if (p->klass_or_null() != NULL) {
D
duke 已提交
7421 7422 7423 7424 7425
            // Redirty the range of cards...
            _mut->mark_range(redirty_range);
          } // ...else the setting of klass will dirty the card anyway.
        }
      DEBUG_ONLY(})
7426
      return true;
D
duke 已提交
7427 7428 7429
    }
  }
  scanOopsInOop(addr);
7430
  return true;
D
duke 已提交
7431 7432 7433 7434 7435 7436 7437 7438 7439 7440 7441 7442 7443 7444 7445 7446 7447 7448 7449 7450 7451 7452 7453 7454 7455 7456 7457 7458 7459 7460 7461 7462 7463 7464 7465 7466 7467 7468 7469 7470
}

// We take a break if we've been at this for a while,
// so as to avoid monopolizing the locks involved.
void MarkFromRootsClosure::do_yield_work() {
  // First give up the locks, then yield, then re-lock
  // We should probably use a constructor/destructor idiom to
  // do this unlock/lock or modify the MutexUnlocker class to
  // serve our purpose. XXX
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  assert_lock_strong(_bitMap->lock());
  _bitMap->lock()->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
  for (unsigned i = 0; i < CMSYieldSleepCount &&
                       ConcurrentMarkSweepThread::should_yield() &&
                       !CMSCollector::foregroundGCIsActive(); ++i) {
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _bitMap->lock()->lock_without_safepoint_check();
  _collector->startTimer();
}

void MarkFromRootsClosure::scanOopsInOop(HeapWord* ptr) {
  assert(_bitMap->isMarked(ptr), "expected bit to be set");
  assert(_markStack->isEmpty(),
         "should drain stack to limit stack usage");
  // convert ptr to an oop preparatory to scanning
7471
  oop obj = oop(ptr);
D
duke 已提交
7472 7473
  // Ignore mark word in verification below, since we
  // may be running concurrent with mutators.
7474
  assert(obj->is_oop(true), "should be an oop");
D
duke 已提交
7475 7476
  assert(_finger <= ptr, "_finger runneth ahead");
  // advance the finger to right end of this object
7477
  _finger = ptr + obj->size();
D
duke 已提交
7478 7479 7480 7481 7482 7483 7484 7485 7486 7487 7488 7489 7490 7491 7492 7493 7494 7495 7496 7497 7498 7499 7500 7501 7502 7503 7504 7505 7506 7507 7508 7509 7510 7511 7512 7513 7514 7515 7516
  assert(_finger > ptr, "we just incremented it above");
  // On large heaps, it may take us some time to get through
  // the marking phase (especially if running iCMS). During
  // this time it's possible that a lot of mutations have
  // accumulated in the card table and the mod union table --
  // these mutation records are redundant until we have
  // actually traced into the corresponding card.
  // Here, we check whether advancing the finger would make
  // us cross into a new card, and if so clear corresponding
  // cards in the MUT (preclean them in the card-table in the
  // future).

  DEBUG_ONLY(if (!_verifying) {)
    // The clean-on-enter optimization is disabled by default,
    // until we fix 6178663.
    if (CMSCleanOnEnter && (_finger > _threshold)) {
      // [_threshold, _finger) represents the interval
      // of cards to be cleared  in MUT (or precleaned in card table).
      // The set of cards to be cleared is all those that overlap
      // with the interval [_threshold, _finger); note that
      // _threshold is always kept card-aligned but _finger isn't
      // always card-aligned.
      HeapWord* old_threshold = _threshold;
      assert(old_threshold == (HeapWord*)round_to(
              (intptr_t)old_threshold, CardTableModRefBS::card_size),
             "_threshold should always be card-aligned");
      _threshold = (HeapWord*)round_to(
                     (intptr_t)_finger, CardTableModRefBS::card_size);
      MemRegion mr(old_threshold, _threshold);
      assert(!mr.is_empty(), "Control point invariant");
      assert(_span.contains(mr), "Should clear within span");
      _mut->clear_range(mr);
    }
  DEBUG_ONLY(})
  // Note: the finger doesn't advance while we drain
  // the stack below.
  PushOrMarkClosure pushOrMarkClosure(_collector,
                                      _span, _bitMap, _markStack,
                                      _finger, this);
7517
  bool res = _markStack->push(obj);
D
duke 已提交
7518 7519 7520 7521 7522 7523 7524 7525 7526 7527 7528 7529 7530 7531 7532 7533 7534 7535 7536 7537 7538 7539 7540 7541 7542 7543 7544 7545 7546 7547 7548 7549 7550 7551 7552 7553 7554 7555
  assert(res, "Empty non-zero size stack should have space for single push");
  while (!_markStack->isEmpty()) {
    oop new_oop = _markStack->pop();
    // Skip verifying header mark word below because we are
    // running concurrent with mutators.
    assert(new_oop->is_oop(true), "Oops! expected to pop an oop");
    // now scan this oop's oops
    new_oop->oop_iterate(&pushOrMarkClosure);
    do_yield_check();
  }
  assert(_markStack->isEmpty(), "tautology, emphasizing post-condition");
}

Par_MarkFromRootsClosure::Par_MarkFromRootsClosure(CMSConcMarkingTask* task,
                       CMSCollector* collector, MemRegion span,
                       CMSBitMap* bit_map,
                       OopTaskQueue* work_queue,
                       CMSMarkStack*  overflow_stack,
                       bool should_yield):
  _collector(collector),
  _whole_span(collector->_span),
  _span(span),
  _bit_map(bit_map),
  _mut(&collector->_modUnionTable),
  _work_queue(work_queue),
  _overflow_stack(overflow_stack),
  _yield(should_yield),
  _skip_bits(0),
  _task(task)
{
  assert(_work_queue->size() == 0, "work_queue should be empty");
  _finger = span.start();
  _threshold = _finger;     // XXX Defer clear-on-enter optimization for now
  assert(_span.contains(_finger), "Out of bounds _finger?");
}

// Should revisit to see if this should be restructured for
// greater efficiency.
7556
bool Par_MarkFromRootsClosure::do_bit(size_t offset) {
D
duke 已提交
7557 7558
  if (_skip_bits > 0) {
    _skip_bits--;
7559
    return true;
D
duke 已提交
7560 7561 7562 7563 7564 7565 7566 7567 7568 7569 7570
  }
  // convert offset into a HeapWord*
  HeapWord* addr = _bit_map->startWord() + offset;
  assert(_bit_map->endWord() && addr < _bit_map->endWord(),
         "address out of range");
  assert(_bit_map->isMarked(addr), "tautology");
  if (_bit_map->isMarked(addr+1)) {
    // this is an allocated object that might not yet be initialized
    assert(_skip_bits == 0, "tautology");
    _skip_bits = 2;  // skip next two marked bits ("Printezis-marks")
    oop p = oop(addr);
7571
    if (p->klass_or_null() == NULL) {
D
duke 已提交
7572 7573
      // in the case of Clean-on-Enter optimization, redirty card
      // and avoid clearing card by increasing  the threshold.
7574
      return true;
D
duke 已提交
7575 7576 7577
    }
  }
  scan_oops_in_oop(addr);
7578
  return true;
D
duke 已提交
7579 7580 7581 7582 7583 7584 7585 7586 7587
}

void Par_MarkFromRootsClosure::scan_oops_in_oop(HeapWord* ptr) {
  assert(_bit_map->isMarked(ptr), "expected bit to be set");
  // Should we assert that our work queue is empty or
  // below some drain limit?
  assert(_work_queue->size() == 0,
         "should drain stack to limit stack usage");
  // convert ptr to an oop preparatory to scanning
7588
  oop obj = oop(ptr);
D
duke 已提交
7589 7590
  // Ignore mark word in verification below, since we
  // may be running concurrent with mutators.
7591
  assert(obj->is_oop(true), "should be an oop");
D
duke 已提交
7592 7593
  assert(_finger <= ptr, "_finger runneth ahead");
  // advance the finger to right end of this object
7594
  _finger = ptr + obj->size();
D
duke 已提交
7595 7596 7597 7598 7599 7600 7601 7602 7603 7604 7605 7606 7607 7608 7609 7610 7611 7612 7613 7614 7615 7616 7617 7618 7619 7620 7621 7622 7623 7624 7625 7626 7627 7628 7629 7630 7631 7632 7633 7634 7635 7636
  assert(_finger > ptr, "we just incremented it above");
  // On large heaps, it may take us some time to get through
  // the marking phase (especially if running iCMS). During
  // this time it's possible that a lot of mutations have
  // accumulated in the card table and the mod union table --
  // these mutation records are redundant until we have
  // actually traced into the corresponding card.
  // Here, we check whether advancing the finger would make
  // us cross into a new card, and if so clear corresponding
  // cards in the MUT (preclean them in the card-table in the
  // future).

  // The clean-on-enter optimization is disabled by default,
  // until we fix 6178663.
  if (CMSCleanOnEnter && (_finger > _threshold)) {
    // [_threshold, _finger) represents the interval
    // of cards to be cleared  in MUT (or precleaned in card table).
    // The set of cards to be cleared is all those that overlap
    // with the interval [_threshold, _finger); note that
    // _threshold is always kept card-aligned but _finger isn't
    // always card-aligned.
    HeapWord* old_threshold = _threshold;
    assert(old_threshold == (HeapWord*)round_to(
            (intptr_t)old_threshold, CardTableModRefBS::card_size),
           "_threshold should always be card-aligned");
    _threshold = (HeapWord*)round_to(
                   (intptr_t)_finger, CardTableModRefBS::card_size);
    MemRegion mr(old_threshold, _threshold);
    assert(!mr.is_empty(), "Control point invariant");
    assert(_span.contains(mr), "Should clear within span"); // _whole_span ??
    _mut->clear_range(mr);
  }

  // Note: the local finger doesn't advance while we drain
  // the stack below, but the global finger sure can and will.
  HeapWord** gfa = _task->global_finger_addr();
  Par_PushOrMarkClosure pushOrMarkClosure(_collector,
                                      _span, _bit_map,
                                      _work_queue,
                                      _overflow_stack,
                                      _finger,
                                      gfa, this);
7637
  bool res = _work_queue->push(obj);   // overflow could occur here
D
duke 已提交
7638 7639 7640 7641 7642 7643 7644 7645 7646 7647 7648 7649 7650 7651 7652 7653 7654 7655 7656 7657 7658 7659 7660 7661 7662 7663 7664 7665 7666 7667 7668 7669 7670 7671 7672 7673 7674 7675 7676 7677 7678 7679 7680 7681 7682 7683 7684 7685 7686 7687 7688 7689 7690 7691 7692 7693 7694 7695
  assert(res, "Will hold once we use workqueues");
  while (true) {
    oop new_oop;
    if (!_work_queue->pop_local(new_oop)) {
      // We emptied our work_queue; check if there's stuff that can
      // be gotten from the overflow stack.
      if (CMSConcMarkingTask::get_work_from_overflow_stack(
            _overflow_stack, _work_queue)) {
        do_yield_check();
        continue;
      } else {  // done
        break;
      }
    }
    // Skip verifying header mark word below because we are
    // running concurrent with mutators.
    assert(new_oop->is_oop(true), "Oops! expected to pop an oop");
    // now scan this oop's oops
    new_oop->oop_iterate(&pushOrMarkClosure);
    do_yield_check();
  }
  assert(_work_queue->size() == 0, "tautology, emphasizing post-condition");
}

// Yield in response to a request from VM Thread or
// from mutators.
void Par_MarkFromRootsClosure::do_yield_work() {
  assert(_task != NULL, "sanity");
  _task->yield();
}

// A variant of the above used for verifying CMS marking work.
MarkFromRootsVerifyClosure::MarkFromRootsVerifyClosure(CMSCollector* collector,
                        MemRegion span,
                        CMSBitMap* verification_bm, CMSBitMap* cms_bm,
                        CMSMarkStack*  mark_stack):
  _collector(collector),
  _span(span),
  _verification_bm(verification_bm),
  _cms_bm(cms_bm),
  _mark_stack(mark_stack),
  _pam_verify_closure(collector, span, verification_bm, cms_bm,
                      mark_stack)
{
  assert(_mark_stack->isEmpty(), "stack should be empty");
  _finger = _verification_bm->startWord();
  assert(_collector->_restart_addr == NULL, "Sanity check");
  assert(_span.contains(_finger), "Out of bounds _finger?");
}

void MarkFromRootsVerifyClosure::reset(HeapWord* addr) {
  assert(_mark_stack->isEmpty(), "would cause duplicates on stack");
  assert(_span.contains(addr), "Out of bounds _finger?");
  _finger = addr;
}

// Should revisit to see if this should be restructured for
// greater efficiency.
7696
bool MarkFromRootsVerifyClosure::do_bit(size_t offset) {
D
duke 已提交
7697 7698 7699 7700 7701 7702 7703 7704 7705 7706
  // convert offset into a HeapWord*
  HeapWord* addr = _verification_bm->startWord() + offset;
  assert(_verification_bm->endWord() && addr < _verification_bm->endWord(),
         "address out of range");
  assert(_verification_bm->isMarked(addr), "tautology");
  assert(_cms_bm->isMarked(addr), "tautology");

  assert(_mark_stack->isEmpty(),
         "should drain stack to limit stack usage");
  // convert addr to an oop preparatory to scanning
7707 7708
  oop obj = oop(addr);
  assert(obj->is_oop(), "should be an oop");
D
duke 已提交
7709 7710
  assert(_finger <= addr, "_finger runneth ahead");
  // advance the finger to right end of this object
7711
  _finger = addr + obj->size();
D
duke 已提交
7712 7713 7714
  assert(_finger > addr, "we just incremented it above");
  // Note: the finger doesn't advance while we drain
  // the stack below.
7715
  bool res = _mark_stack->push(obj);
D
duke 已提交
7716 7717 7718 7719 7720 7721 7722 7723
  assert(res, "Empty non-zero size stack should have space for single push");
  while (!_mark_stack->isEmpty()) {
    oop new_oop = _mark_stack->pop();
    assert(new_oop->is_oop(), "Oops! expected to pop an oop");
    // now scan this oop's oops
    new_oop->oop_iterate(&_pam_verify_closure);
  }
  assert(_mark_stack->isEmpty(), "tautology, emphasizing post-condition");
7724
  return true;
D
duke 已提交
7725 7726 7727 7728 7729 7730
}

PushAndMarkVerifyClosure::PushAndMarkVerifyClosure(
  CMSCollector* collector, MemRegion span,
  CMSBitMap* verification_bm, CMSBitMap* cms_bm,
  CMSMarkStack*  mark_stack):
7731
  CMSOopClosure(collector->ref_processor()),
D
duke 已提交
7732 7733 7734 7735 7736 7737 7738
  _collector(collector),
  _span(span),
  _verification_bm(verification_bm),
  _cms_bm(cms_bm),
  _mark_stack(mark_stack)
{ }

7739 7740
void PushAndMarkVerifyClosure::do_oop(oop* p)       { PushAndMarkVerifyClosure::do_oop_work(p); }
void PushAndMarkVerifyClosure::do_oop(narrowOop* p) { PushAndMarkVerifyClosure::do_oop_work(p); }
D
duke 已提交
7741 7742 7743 7744 7745 7746 7747 7748 7749 7750 7751 7752

// Upon stack overflow, we discard (part of) the stack,
// remembering the least address amongst those discarded
// in CMSCollector's _restart_address.
void PushAndMarkVerifyClosure::handle_stack_overflow(HeapWord* lost) {
  // Remember the least grey address discarded
  HeapWord* ra = (HeapWord*)_mark_stack->least_value(lost);
  _collector->lower_restart_addr(ra);
  _mark_stack->reset();  // discard stack contents
  _mark_stack->expand(); // expand the stack if possible
}

7753 7754 7755
void PushAndMarkVerifyClosure::do_oop(oop obj) {
  assert(obj->is_oop_or_null(), "expected an oop or NULL");
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
7756 7757 7758 7759 7760
  if (_span.contains(addr) && !_verification_bm->isMarked(addr)) {
    // Oop lies in _span and isn't yet grey or black
    _verification_bm->mark(addr);            // now grey
    if (!_cms_bm->isMarked(addr)) {
      oop(addr)->print();
7761 7762
      gclog_or_tty->print_cr(" (" INTPTR_FORMAT " should have been marked)",
                             addr);
D
duke 已提交
7763 7764 7765
      fatal("... aborting");
    }

7766
    if (!_mark_stack->push(obj)) { // stack overflow
D
duke 已提交
7767 7768 7769 7770 7771 7772 7773 7774 7775 7776 7777 7778 7779 7780 7781 7782 7783
      if (PrintCMSStatistics != 0) {
        gclog_or_tty->print_cr("CMS marking stack overflow (benign) at "
                               SIZE_FORMAT, _mark_stack->capacity());
      }
      assert(_mark_stack->isFull(), "Else push should have succeeded");
      handle_stack_overflow(addr);
    }
    // anything including and to the right of _finger
    // will be scanned as we iterate over the remainder of the
    // bit map
  }
}

PushOrMarkClosure::PushOrMarkClosure(CMSCollector* collector,
                     MemRegion span,
                     CMSBitMap* bitMap, CMSMarkStack*  markStack,
                     HeapWord* finger, MarkFromRootsClosure* parent) :
7784 7785
  CMSOopClosure(collector->ref_processor()),
  _collector(collector),
D
duke 已提交
7786 7787 7788 7789
  _span(span),
  _bitMap(bitMap),
  _markStack(markStack),
  _finger(finger),
7790
  _parent(parent)
D
duke 已提交
7791 7792 7793 7794 7795 7796 7797 7798 7799 7800
{ }

Par_PushOrMarkClosure::Par_PushOrMarkClosure(CMSCollector* collector,
                     MemRegion span,
                     CMSBitMap* bit_map,
                     OopTaskQueue* work_queue,
                     CMSMarkStack*  overflow_stack,
                     HeapWord* finger,
                     HeapWord** global_finger_addr,
                     Par_MarkFromRootsClosure* parent) :
7801 7802
  CMSOopClosure(collector->ref_processor()),
  _collector(collector),
D
duke 已提交
7803 7804 7805 7806 7807 7808 7809
  _whole_span(collector->_span),
  _span(span),
  _bit_map(bit_map),
  _work_queue(work_queue),
  _overflow_stack(overflow_stack),
  _finger(finger),
  _global_finger_addr(global_finger_addr),
7810
  _parent(parent)
D
duke 已提交
7811 7812
{ }

7813 7814
// Assumes thread-safe access by callers, who are
// responsible for mutual exclusion.
D
duke 已提交
7815 7816 7817 7818 7819 7820 7821 7822 7823 7824 7825 7826 7827 7828 7829 7830 7831 7832 7833 7834 7835 7836 7837 7838 7839
void CMSCollector::lower_restart_addr(HeapWord* low) {
  assert(_span.contains(low), "Out of bounds addr");
  if (_restart_addr == NULL) {
    _restart_addr = low;
  } else {
    _restart_addr = MIN2(_restart_addr, low);
  }
}

// Upon stack overflow, we discard (part of) the stack,
// remembering the least address amongst those discarded
// in CMSCollector's _restart_address.
void PushOrMarkClosure::handle_stack_overflow(HeapWord* lost) {
  // Remember the least grey address discarded
  HeapWord* ra = (HeapWord*)_markStack->least_value(lost);
  _collector->lower_restart_addr(ra);
  _markStack->reset();  // discard stack contents
  _markStack->expand(); // expand the stack if possible
}

// Upon stack overflow, we discard (part of) the stack,
// remembering the least address amongst those discarded
// in CMSCollector's _restart_address.
void Par_PushOrMarkClosure::handle_stack_overflow(HeapWord* lost) {
  // We need to do this under a mutex to prevent other
7840
  // workers from interfering with the work done below.
D
duke 已提交
7841 7842 7843 7844 7845 7846 7847 7848 7849
  MutexLockerEx ml(_overflow_stack->par_lock(),
                   Mutex::_no_safepoint_check_flag);
  // Remember the least grey address discarded
  HeapWord* ra = (HeapWord*)_overflow_stack->least_value(lost);
  _collector->lower_restart_addr(ra);
  _overflow_stack->reset();  // discard stack contents
  _overflow_stack->expand(); // expand the stack if possible
}

7850 7851 7852 7853 7854
void CMKlassClosure::do_klass(Klass* k) {
  assert(_oop_closure != NULL, "Not initialized?");
  k->oops_do(_oop_closure);
}

7855
void PushOrMarkClosure::do_oop(oop obj) {
D
duke 已提交
7856
  // Ignore mark word because we are running concurrent with mutators.
7857 7858
  assert(obj->is_oop_or_null(true), "expected an oop or NULL");
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
7859 7860 7861 7862 7863 7864 7865 7866 7867 7868 7869 7870 7871 7872 7873
  if (_span.contains(addr) && !_bitMap->isMarked(addr)) {
    // Oop lies in _span and isn't yet grey or black
    _bitMap->mark(addr);            // now grey
    if (addr < _finger) {
      // the bit map iteration has already either passed, or
      // sampled, this bit in the bit map; we'll need to
      // use the marking stack to scan this oop's oops.
      bool simulate_overflow = false;
      NOT_PRODUCT(
        if (CMSMarkStackOverflowALot &&
            _collector->simulate_overflow()) {
          // simulate a stack overflow
          simulate_overflow = true;
        }
      )
7874
      if (simulate_overflow || !_markStack->push(obj)) { // stack overflow
D
duke 已提交
7875 7876 7877 7878 7879 7880 7881 7882 7883 7884 7885 7886 7887 7888 7889
        if (PrintCMSStatistics != 0) {
          gclog_or_tty->print_cr("CMS marking stack overflow (benign) at "
                                 SIZE_FORMAT, _markStack->capacity());
        }
        assert(simulate_overflow || _markStack->isFull(), "Else push should have succeeded");
        handle_stack_overflow(addr);
      }
    }
    // anything including and to the right of _finger
    // will be scanned as we iterate over the remainder of the
    // bit map
    do_yield_check();
  }
}

7890 7891 7892 7893
void PushOrMarkClosure::do_oop(oop* p)       { PushOrMarkClosure::do_oop_work(p); }
void PushOrMarkClosure::do_oop(narrowOop* p) { PushOrMarkClosure::do_oop_work(p); }

void Par_PushOrMarkClosure::do_oop(oop obj) {
D
duke 已提交
7894
  // Ignore mark word because we are running concurrent with mutators.
7895 7896
  assert(obj->is_oop_or_null(true), "expected an oop or NULL");
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
7897 7898 7899 7900 7901 7902 7903 7904 7905 7906 7907 7908 7909 7910 7911 7912 7913 7914 7915 7916 7917 7918 7919 7920 7921 7922 7923 7924
  if (_whole_span.contains(addr) && !_bit_map->isMarked(addr)) {
    // Oop lies in _span and isn't yet grey or black
    // We read the global_finger (volatile read) strictly after marking oop
    bool res = _bit_map->par_mark(addr);    // now grey
    volatile HeapWord** gfa = (volatile HeapWord**)_global_finger_addr;
    // Should we push this marked oop on our stack?
    // -- if someone else marked it, nothing to do
    // -- if target oop is above global finger nothing to do
    // -- if target oop is in chunk and above local finger
    //      then nothing to do
    // -- else push on work queue
    if (   !res       // someone else marked it, they will deal with it
        || (addr >= *gfa)  // will be scanned in a later task
        || (_span.contains(addr) && addr >= _finger)) { // later in this chunk
      return;
    }
    // the bit map iteration has already either passed, or
    // sampled, this bit in the bit map; we'll need to
    // use the marking stack to scan this oop's oops.
    bool simulate_overflow = false;
    NOT_PRODUCT(
      if (CMSMarkStackOverflowALot &&
          _collector->simulate_overflow()) {
        // simulate a stack overflow
        simulate_overflow = true;
      }
    )
    if (simulate_overflow ||
7925
        !(_work_queue->push(obj) || _overflow_stack->par_push(obj))) {
D
duke 已提交
7926 7927 7928 7929 7930 7931 7932 7933 7934 7935 7936 7937 7938 7939 7940 7941
      // stack overflow
      if (PrintCMSStatistics != 0) {
        gclog_or_tty->print_cr("CMS marking stack overflow (benign) at "
                               SIZE_FORMAT, _overflow_stack->capacity());
      }
      // We cannot assert that the overflow stack is full because
      // it may have been emptied since.
      assert(simulate_overflow ||
             _work_queue->size() == _work_queue->max_elems(),
            "Else push should have succeeded");
      handle_stack_overflow(addr);
    }
    do_yield_check();
  }
}

7942 7943
void Par_PushOrMarkClosure::do_oop(oop* p)       { Par_PushOrMarkClosure::do_oop_work(p); }
void Par_PushOrMarkClosure::do_oop(narrowOop* p) { Par_PushOrMarkClosure::do_oop_work(p); }
D
duke 已提交
7944 7945 7946 7947 7948 7949 7950 7951

PushAndMarkClosure::PushAndMarkClosure(CMSCollector* collector,
                                       MemRegion span,
                                       ReferenceProcessor* rp,
                                       CMSBitMap* bit_map,
                                       CMSBitMap* mod_union_table,
                                       CMSMarkStack*  mark_stack,
                                       bool           concurrent_precleaning):
7952 7953
  CMSOopClosure(rp),
  _collector(collector),
D
duke 已提交
7954 7955 7956 7957
  _span(span),
  _bit_map(bit_map),
  _mod_union_table(mod_union_table),
  _mark_stack(mark_stack),
7958
  _concurrent_precleaning(concurrent_precleaning)
D
duke 已提交
7959 7960 7961 7962 7963 7964
{
  assert(_ref_processor != NULL, "_ref_processor shouldn't be NULL");
}

// Grey object rescan during pre-cleaning and second checkpoint phases --
// the non-parallel version (the parallel version appears further below.)
7965
void PushAndMarkClosure::do_oop(oop obj) {
7966 7967 7968 7969 7970 7971
  // Ignore mark word verification. If during concurrent precleaning,
  // the object monitor may be locked. If during the checkpoint
  // phases, the object may already have been reached by a  different
  // path and may be at the end of the global overflow list (so
  // the mark word may be NULL).
  assert(obj->is_oop_or_null(true /* ignore mark word */),
D
duke 已提交
7972
         "expected an oop or NULL");
7973
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
7974 7975 7976 7977 7978 7979 7980 7981 7982 7983 7984 7985 7986 7987
  // Check if oop points into the CMS generation
  // and is not marked
  if (_span.contains(addr) && !_bit_map->isMarked(addr)) {
    // a white object ...
    _bit_map->mark(addr);         // ... now grey
    // push on the marking stack (grey set)
    bool simulate_overflow = false;
    NOT_PRODUCT(
      if (CMSMarkStackOverflowALot &&
          _collector->simulate_overflow()) {
        // simulate a stack overflow
        simulate_overflow = true;
      }
    )
7988
    if (simulate_overflow || !_mark_stack->push(obj)) {
D
duke 已提交
7989
      if (_concurrent_precleaning) {
7990
         // During precleaning we can just dirty the appropriate card(s)
D
duke 已提交
7991
         // in the mod union table, thus ensuring that the object remains
7992 7993 7994 7995 7996 7997 7998 7999 8000 8001 8002 8003 8004 8005 8006 8007 8008
         // in the grey set  and continue. In the case of object arrays
         // we need to dirty all of the cards that the object spans,
         // since the rescan of object arrays will be limited to the
         // dirty cards.
         // Note that no one can be intefering with us in this action
         // of dirtying the mod union table, so no locking or atomics
         // are required.
         if (obj->is_objArray()) {
           size_t sz = obj->size();
           HeapWord* end_card_addr = (HeapWord*)round_to(
                                        (intptr_t)(addr+sz), CardTableModRefBS::card_size);
           MemRegion redirty_range = MemRegion(addr, end_card_addr);
           assert(!redirty_range.is_empty(), "Arithmetical tautology");
           _mod_union_table->mark_range(redirty_range);
         } else {
           _mod_union_table->mark(addr);
         }
D
duke 已提交
8009 8010 8011 8012
         _collector->_ser_pmc_preclean_ovflw++;
      } else {
         // During the remark phase, we need to remember this oop
         // in the overflow list.
8013
         _collector->push_on_overflow_list(obj);
D
duke 已提交
8014 8015 8016 8017 8018 8019 8020 8021 8022 8023
         _collector->_ser_pmc_remark_ovflw++;
      }
    }
  }
}

Par_PushAndMarkClosure::Par_PushAndMarkClosure(CMSCollector* collector,
                                               MemRegion span,
                                               ReferenceProcessor* rp,
                                               CMSBitMap* bit_map,
8024 8025 8026
                                               OopTaskQueue* work_queue):
  CMSOopClosure(rp),
  _collector(collector),
D
duke 已提交
8027 8028
  _span(span),
  _bit_map(bit_map),
8029
  _work_queue(work_queue)
D
duke 已提交
8030 8031 8032 8033
{
  assert(_ref_processor != NULL, "_ref_processor shouldn't be NULL");
}

8034 8035 8036
void PushAndMarkClosure::do_oop(oop* p)       { PushAndMarkClosure::do_oop_work(p); }
void PushAndMarkClosure::do_oop(narrowOop* p) { PushAndMarkClosure::do_oop_work(p); }

D
duke 已提交
8037 8038
// Grey object rescan during second checkpoint phase --
// the parallel version.
8039
void Par_PushAndMarkClosure::do_oop(oop obj) {
D
duke 已提交
8040 8041 8042 8043 8044 8045 8046 8047 8048 8049 8050
  // In the assert below, we ignore the mark word because
  // this oop may point to an already visited object that is
  // on the overflow stack (in which case the mark word has
  // been hijacked for chaining into the overflow stack --
  // if this is the last object in the overflow stack then
  // its mark word will be NULL). Because this object may
  // have been subsequently popped off the global overflow
  // stack, and the mark word possibly restored to the prototypical
  // value, by the time we get to examined this failing assert in
  // the debugger, is_oop_or_null(false) may subsequently start
  // to hold.
8051
  assert(obj->is_oop_or_null(true),
D
duke 已提交
8052
         "expected an oop or NULL");
8053
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
8054 8055 8056 8057 8058 8059 8060 8061 8062 8063 8064 8065 8066 8067 8068 8069 8070
  // Check if oop points into the CMS generation
  // and is not marked
  if (_span.contains(addr) && !_bit_map->isMarked(addr)) {
    // a white object ...
    // If we manage to "claim" the object, by being the
    // first thread to mark it, then we push it on our
    // marking stack
    if (_bit_map->par_mark(addr)) {     // ... now grey
      // push on work queue (grey set)
      bool simulate_overflow = false;
      NOT_PRODUCT(
        if (CMSMarkStackOverflowALot &&
            _collector->par_simulate_overflow()) {
          // simulate a stack overflow
          simulate_overflow = true;
        }
      )
8071 8072
      if (simulate_overflow || !_work_queue->push(obj)) {
        _collector->par_push_on_overflow_list(obj);
D
duke 已提交
8073 8074 8075 8076 8077 8078
        _collector->_par_pmc_remark_ovflw++; //  imprecise OK: no need to CAS
      }
    } // Else, some other thread got there first
  }
}

8079 8080 8081
void Par_PushAndMarkClosure::do_oop(oop* p)       { Par_PushAndMarkClosure::do_oop_work(p); }
void Par_PushAndMarkClosure::do_oop(narrowOop* p) { Par_PushAndMarkClosure::do_oop_work(p); }

D
duke 已提交
8082 8083 8084 8085 8086 8087 8088 8089 8090 8091 8092 8093 8094 8095 8096 8097 8098 8099 8100 8101 8102 8103 8104 8105 8106 8107 8108 8109 8110 8111 8112 8113 8114 8115 8116 8117 8118 8119 8120 8121 8122 8123 8124 8125 8126 8127 8128 8129 8130 8131 8132 8133 8134 8135 8136 8137 8138 8139 8140 8141 8142 8143 8144 8145 8146 8147 8148 8149 8150 8151 8152 8153 8154 8155 8156 8157 8158 8159 8160 8161
void CMSPrecleanRefsYieldClosure::do_yield_work() {
  Mutex* bml = _collector->bitMapLock();
  assert_lock_strong(bml);
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");

  bml->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);

  ConcurrentMarkSweepThread::acknowledge_yield_request();

  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
  for (unsigned i = 0; i < CMSYieldSleepCount &&
                       ConcurrentMarkSweepThread::should_yield() &&
                       !CMSCollector::foregroundGCIsActive(); ++i) {
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  bml->lock();

  _collector->startTimer();
}

bool CMSPrecleanRefsYieldClosure::should_return() {
  if (ConcurrentMarkSweepThread::should_yield()) {
    do_yield_work();
  }
  return _collector->foregroundGCIsActive();
}

void MarkFromDirtyCardsClosure::do_MemRegion(MemRegion mr) {
  assert(((size_t)mr.start())%CardTableModRefBS::card_size_in_words == 0,
         "mr should be aligned to start at a card boundary");
  // We'd like to assert:
  // assert(mr.word_size()%CardTableModRefBS::card_size_in_words == 0,
  //        "mr should be a range of cards");
  // However, that would be too strong in one case -- the last
  // partition ends at _unallocated_block which, in general, can be
  // an arbitrary boundary, not necessarily card aligned.
  if (PrintCMSStatistics != 0) {
    _num_dirty_cards +=
         mr.word_size()/CardTableModRefBS::card_size_in_words;
  }
  _space->object_iterate_mem(mr, &_scan_cl);
}

SweepClosure::SweepClosure(CMSCollector* collector,
                           ConcurrentMarkSweepGeneration* g,
                           CMSBitMap* bitMap, bool should_yield) :
  _collector(collector),
  _g(g),
  _sp(g->cmsSpace()),
  _limit(_sp->sweep_limit()),
  _freelistLock(_sp->freelistLock()),
  _bitMap(bitMap),
  _yield(should_yield),
  _inFreeRange(false),           // No free range at beginning of sweep
  _freeRangeInFreeLists(false),  // No free range at beginning of sweep
  _lastFreeRangeCoalesced(false),
  _freeFinger(g->used_region().start())
{
  NOT_PRODUCT(
    _numObjectsFreed = 0;
    _numWordsFreed   = 0;
    _numObjectsLive = 0;
    _numWordsLive = 0;
    _numObjectsAlreadyFree = 0;
    _numWordsAlreadyFree = 0;
    _last_fc = NULL;

    _sp->initializeIndexedFreeListArrayReturnedBytes();
8162
    _sp->dictionary()->initialize_dict_returned_bytes();
D
duke 已提交
8163 8164 8165 8166
  )
  assert(_limit >= _sp->bottom() && _limit <= _sp->end(),
         "sweep _limit out of bounds");
  if (CMSTraceSweeper) {
8167 8168
    gclog_or_tty->print_cr("\n====================\nStarting new sweep with limit " PTR_FORMAT,
                        _limit);
D
duke 已提交
8169 8170 8171
  }
}

8172 8173 8174 8175 8176 8177 8178 8179 8180 8181 8182 8183 8184 8185 8186
void SweepClosure::print_on(outputStream* st) const {
  tty->print_cr("_sp = [" PTR_FORMAT "," PTR_FORMAT ")",
                _sp->bottom(), _sp->end());
  tty->print_cr("_limit = " PTR_FORMAT, _limit);
  tty->print_cr("_freeFinger = " PTR_FORMAT, _freeFinger);
  NOT_PRODUCT(tty->print_cr("_last_fc = " PTR_FORMAT, _last_fc);)
  tty->print_cr("_inFreeRange = %d, _freeRangeInFreeLists = %d, _lastFreeRangeCoalesced = %d",
                _inFreeRange, _freeRangeInFreeLists, _lastFreeRangeCoalesced);
}

#ifndef PRODUCT
// Assertion checking only:  no useful work in product mode --
// however, if any of the flags below become product flags,
// you may need to review this code to see if it needs to be
// enabled in product mode.
D
duke 已提交
8187 8188
SweepClosure::~SweepClosure() {
  assert_lock_strong(_freelistLock);
8189 8190
  assert(_limit >= _sp->bottom() && _limit <= _sp->end(),
         "sweep _limit out of bounds");
D
duke 已提交
8191
  if (inFreeRange()) {
8192 8193 8194 8195 8196 8197 8198 8199 8200 8201 8202 8203 8204 8205 8206 8207 8208 8209
    warning("inFreeRange() should have been reset; dumping state of SweepClosure");
    print();
    ShouldNotReachHere();
  }
  if (Verbose && PrintGC) {
    gclog_or_tty->print("Collected "SIZE_FORMAT" objects, " SIZE_FORMAT " bytes",
                        _numObjectsFreed, _numWordsFreed*sizeof(HeapWord));
    gclog_or_tty->print_cr("\nLive "SIZE_FORMAT" objects,  "
                           SIZE_FORMAT" bytes  "
      "Already free "SIZE_FORMAT" objects, "SIZE_FORMAT" bytes",
      _numObjectsLive, _numWordsLive*sizeof(HeapWord),
      _numObjectsAlreadyFree, _numWordsAlreadyFree*sizeof(HeapWord));
    size_t totalBytes = (_numWordsFreed + _numWordsLive + _numWordsAlreadyFree)
                        * sizeof(HeapWord);
    gclog_or_tty->print_cr("Total sweep: "SIZE_FORMAT" bytes", totalBytes);

    if (PrintCMSStatistics && CMSVerifyReturnedBytes) {
      size_t indexListReturnedBytes = _sp->sumIndexedFreeListArrayReturnedBytes();
8210 8211 8212
      size_t dict_returned_bytes = _sp->dictionary()->sum_dict_returned_bytes();
      size_t returned_bytes = indexListReturnedBytes + dict_returned_bytes;
      gclog_or_tty->print("Returned "SIZE_FORMAT" bytes", returned_bytes);
8213 8214 8215
      gclog_or_tty->print("   Indexed List Returned "SIZE_FORMAT" bytes",
        indexListReturnedBytes);
      gclog_or_tty->print_cr("        Dictionary Returned "SIZE_FORMAT" bytes",
8216
        dict_returned_bytes);
D
duke 已提交
8217
    }
8218
  }
D
duke 已提交
8219
  if (CMSTraceSweeper) {
8220 8221
    gclog_or_tty->print_cr("end of sweep with _limit = " PTR_FORMAT "\n================",
                           _limit);
D
duke 已提交
8222 8223
  }
}
8224
#endif  // PRODUCT
D
duke 已提交
8225 8226 8227 8228

void SweepClosure::initialize_free_range(HeapWord* freeFinger,
    bool freeRangeInFreeLists) {
  if (CMSTraceSweeper) {
8229 8230
    gclog_or_tty->print("---- Start free range at 0x%x with free block (%d)\n",
               freeFinger, freeRangeInFreeLists);
D
duke 已提交
8231 8232 8233 8234 8235 8236 8237 8238 8239 8240
  }
  assert(!inFreeRange(), "Trampling existing free range");
  set_inFreeRange(true);
  set_lastFreeRangeCoalesced(false);

  set_freeFinger(freeFinger);
  set_freeRangeInFreeLists(freeRangeInFreeLists);
  if (CMSTestInFreeList) {
    if (freeRangeInFreeLists) {
      FreeChunk* fc = (FreeChunk*) freeFinger;
8241
      assert(fc->is_free(), "A chunk on the free list should be free.");
D
duke 已提交
8242
      assert(fc->size() > 0, "Free range should have a size");
8243
      assert(_sp->verify_chunk_in_free_list(fc), "Chunk is not in free lists");
D
duke 已提交
8244 8245 8246 8247 8248 8249 8250 8251 8252 8253 8254 8255 8256 8257 8258 8259 8260 8261 8262 8263 8264 8265 8266 8267 8268 8269 8270 8271 8272 8273 8274 8275 8276 8277 8278
    }
  }
}

// Note that the sweeper runs concurrently with mutators. Thus,
// it is possible for direct allocation in this generation to happen
// in the middle of the sweep. Note that the sweeper also coalesces
// contiguous free blocks. Thus, unless the sweeper and the allocator
// synchronize appropriately freshly allocated blocks may get swept up.
// This is accomplished by the sweeper locking the free lists while
// it is sweeping. Thus blocks that are determined to be free are
// indeed free. There is however one additional complication:
// blocks that have been allocated since the final checkpoint and
// mark, will not have been marked and so would be treated as
// unreachable and swept up. To prevent this, the allocator marks
// the bit map when allocating during the sweep phase. This leads,
// however, to a further complication -- objects may have been allocated
// but not yet initialized -- in the sense that the header isn't yet
// installed. The sweeper can not then determine the size of the block
// in order to skip over it. To deal with this case, we use a technique
// (due to Printezis) to encode such uninitialized block sizes in the
// bit map. Since the bit map uses a bit per every HeapWord, but the
// CMS generation has a minimum object size of 3 HeapWords, it follows
// that "normal marks" won't be adjacent in the bit map (there will
// always be at least two 0 bits between successive 1 bits). We make use
// of these "unused" bits to represent uninitialized blocks -- the bit
// corresponding to the start of the uninitialized object and the next
// bit are both set. Finally, a 1 bit marks the end of the object that
// started with the two consecutive 1 bits to indicate its potentially
// uninitialized state.

size_t SweepClosure::do_blk_careful(HeapWord* addr) {
  FreeChunk* fc = (FreeChunk*)addr;
  size_t res;

8279 8280 8281 8282 8283
  // Check if we are done sweeping. Below we check "addr >= _limit" rather
  // than "addr == _limit" because although _limit was a block boundary when
  // we started the sweep, it may no longer be one because heap expansion
  // may have caused us to coalesce the block ending at the address _limit
  // with a newly expanded chunk (this happens when _limit was set to the
8284 8285
  // previous _end of the space), so we may have stepped past _limit:
  // see the following Zeno-like trail of CRs 6977970, 7008136, 7042740.
8286
  if (addr >= _limit) { // we have swept up to or past the limit: finish up
D
duke 已提交
8287 8288
    assert(_limit >= _sp->bottom() && _limit <= _sp->end(),
           "sweep _limit out of bounds");
8289
    assert(addr < _sp->end(), "addr out of bounds");
8290
    // Flush any free range we might be holding as a single
8291 8292
    // coalesced chunk to the appropriate free list.
    if (inFreeRange()) {
8293 8294
      assert(freeFinger() >= _sp->bottom() && freeFinger() < _limit,
             err_msg("freeFinger() " PTR_FORMAT" is out-of-bounds", freeFinger()));
8295 8296 8297 8298 8299 8300 8301 8302 8303 8304 8305 8306
      flush_cur_free_chunk(freeFinger(),
                           pointer_delta(addr, freeFinger()));
      if (CMSTraceSweeper) {
        gclog_or_tty->print("Sweep: last chunk: ");
        gclog_or_tty->print("put_free_blk 0x%x ("SIZE_FORMAT") "
                   "[coalesced:"SIZE_FORMAT"]\n",
                   freeFinger(), pointer_delta(addr, freeFinger()),
                   lastFreeRangeCoalesced());
      }
    }

    // help the iterator loop finish
8307
    return pointer_delta(_sp->end(), addr);
D
duke 已提交
8308 8309
  }

8310
  assert(addr < _limit, "sweep invariant");
D
duke 已提交
8311 8312
  // check if we should yield
  do_yield_check(addr);
8313
  if (fc->is_free()) {
D
duke 已提交
8314 8315
    // Chunk that is already free
    res = fc->size();
8316
    do_already_free_chunk(fc);
D
duke 已提交
8317
    debug_only(_sp->verifyFreeLists());
8318 8319 8320 8321 8322 8323 8324 8325 8326 8327
    // If we flush the chunk at hand in lookahead_and_flush()
    // and it's coalesced with a preceding chunk, then the
    // process of "mangling" the payload of the coalesced block
    // will cause erasure of the size information from the
    // (erstwhile) header of all the coalesced blocks but the
    // first, so the first disjunct in the assert will not hold
    // in that specific case (in which case the second disjunct
    // will hold).
    assert(res == fc->size() || ((HeapWord*)fc) + res >= _limit,
           "Otherwise the size info doesn't change at this step");
D
duke 已提交
8328 8329 8330 8331 8332 8333 8334
    NOT_PRODUCT(
      _numObjectsAlreadyFree++;
      _numWordsAlreadyFree += res;
    )
    NOT_PRODUCT(_last_fc = fc;)
  } else if (!_bitMap->isMarked(addr)) {
    // Chunk is fresh garbage
8335
    res = do_garbage_chunk(fc);
D
duke 已提交
8336 8337 8338 8339 8340 8341 8342
    debug_only(_sp->verifyFreeLists());
    NOT_PRODUCT(
      _numObjectsFreed++;
      _numWordsFreed += res;
    )
  } else {
    // Chunk that is alive.
8343
    res = do_live_chunk(fc);
D
duke 已提交
8344 8345 8346 8347 8348 8349 8350 8351 8352 8353 8354 8355 8356 8357 8358 8359 8360 8361 8362 8363 8364 8365 8366 8367 8368 8369 8370 8371 8372 8373 8374 8375 8376 8377 8378 8379 8380 8381 8382 8383 8384 8385 8386 8387 8388 8389 8390 8391 8392 8393 8394 8395
    debug_only(_sp->verifyFreeLists());
    NOT_PRODUCT(
        _numObjectsLive++;
        _numWordsLive += res;
    )
  }
  return res;
}

// For the smart allocation, record following
//  split deaths - a free chunk is removed from its free list because
//      it is being split into two or more chunks.
//  split birth - a free chunk is being added to its free list because
//      a larger free chunk has been split and resulted in this free chunk.
//  coal death - a free chunk is being removed from its free list because
//      it is being coalesced into a large free chunk.
//  coal birth - a free chunk is being added to its free list because
//      it was created when two or more free chunks where coalesced into
//      this free chunk.
//
// These statistics are used to determine the desired number of free
// chunks of a given size.  The desired number is chosen to be relative
// to the end of a CMS sweep.  The desired number at the end of a sweep
// is the
//      count-at-end-of-previous-sweep (an amount that was enough)
//              - count-at-beginning-of-current-sweep  (the excess)
//              + split-births  (gains in this size during interval)
//              - split-deaths  (demands on this size during interval)
// where the interval is from the end of one sweep to the end of the
// next.
//
// When sweeping the sweeper maintains an accumulated chunk which is
// the chunk that is made up of chunks that have been coalesced.  That
// will be termed the left-hand chunk.  A new chunk of garbage that
// is being considered for coalescing will be referred to as the
// right-hand chunk.
//
// When making a decision on whether to coalesce a right-hand chunk with
// the current left-hand chunk, the current count vs. the desired count
// of the left-hand chunk is considered.  Also if the right-hand chunk
// is near the large chunk at the end of the heap (see
// ConcurrentMarkSweepGeneration::isNearLargestChunk()), then the
// left-hand chunk is coalesced.
//
// When making a decision about whether to split a chunk, the desired count
// vs. the current count of the candidate to be split is also considered.
// If the candidate is underpopulated (currently fewer chunks than desired)
// a chunk of an overpopulated (currently more chunks than desired) size may
// be chosen.  The "hint" associated with a free list, if non-null, points
// to a free list which may be overpopulated.
//

8396
void SweepClosure::do_already_free_chunk(FreeChunk* fc) {
8397
  const size_t size = fc->size();
D
duke 已提交
8398 8399 8400
  // Chunks that cannot be coalesced are not in the
  // free lists.
  if (CMSTestInFreeList && !fc->cantCoalesce()) {
8401
    assert(_sp->verify_chunk_in_free_list(fc),
D
duke 已提交
8402 8403 8404 8405
      "free chunk should be in free lists");
  }
  // a chunk that is already free, should not have been
  // marked in the bit map
8406
  HeapWord* const addr = (HeapWord*) fc;
D
duke 已提交
8407 8408 8409 8410 8411
  assert(!_bitMap->isMarked(addr), "free chunk should be unmarked");
  // Verify that the bit map has no bits marked between
  // addr and purported end of this block.
  _bitMap->verifyNoOneBitsInRange(addr + 1, addr + size);

8412
  // Some chunks cannot be coalesced under any circumstances.
D
duke 已提交
8413 8414 8415 8416 8417
  // See the definition of cantCoalesce().
  if (!fc->cantCoalesce()) {
    // This chunk can potentially be coalesced.
    if (_sp->adaptive_freelists()) {
      // All the work is done in
8418
      do_post_free_or_garbage_chunk(fc, size);
D
duke 已提交
8419 8420 8421 8422 8423 8424
    } else {  // Not adaptive free lists
      // this is a free chunk that can potentially be coalesced by the sweeper;
      if (!inFreeRange()) {
        // if the next chunk is a free block that can't be coalesced
        // it doesn't make sense to remove this chunk from the free lists
        FreeChunk* nextChunk = (FreeChunk*)(addr + size);
8425 8426
        assert((HeapWord*)nextChunk <= _sp->end(), "Chunk size out of bounds?");
        if ((HeapWord*)nextChunk < _sp->end() &&     // There is another free chunk to the right ...
8427
            nextChunk->is_free()               &&     // ... which is free...
8428
            nextChunk->cantCoalesce()) {             // ... but can't be coalesced
D
duke 已提交
8429 8430 8431 8432 8433 8434 8435 8436 8437 8438 8439 8440 8441 8442
          // nothing to do
        } else {
          // Potentially the start of a new free range:
          // Don't eagerly remove it from the free lists.
          // No need to remove it if it will just be put
          // back again.  (Also from a pragmatic point of view
          // if it is a free block in a region that is beyond
          // any allocated blocks, an assertion will fail)
          // Remember the start of a free run.
          initialize_free_range(addr, true);
          // end - can coalesce with next chunk
        }
      } else {
        // the midst of a free range, we are coalescing
8443
        print_free_block_coalesced(fc);
D
duke 已提交
8444 8445 8446 8447 8448 8449 8450 8451 8452 8453 8454 8455 8456 8457 8458
        if (CMSTraceSweeper) {
          gclog_or_tty->print("  -- pick up free block 0x%x (%d)\n", fc, size);
        }
        // remove it from the free lists
        _sp->removeFreeChunkFromFreeLists(fc);
        set_lastFreeRangeCoalesced(true);
        // If the chunk is being coalesced and the current free range is
        // in the free lists, remove the current free range so that it
        // will be returned to the free lists in its entirety - all
        // the coalesced pieces included.
        if (freeRangeInFreeLists()) {
          FreeChunk* ffc = (FreeChunk*) freeFinger();
          assert(ffc->size() == pointer_delta(addr, freeFinger()),
            "Size of free range is inconsistent with chunk size.");
          if (CMSTestInFreeList) {
8459
            assert(_sp->verify_chunk_in_free_list(ffc),
D
duke 已提交
8460 8461 8462 8463 8464 8465 8466
              "free range is not in free lists");
          }
          _sp->removeFreeChunkFromFreeLists(ffc);
          set_freeRangeInFreeLists(false);
        }
      }
    }
8467 8468 8469 8470
    // Note that if the chunk is not coalescable (the else arm
    // below), we unconditionally flush, without needing to do
    // a "lookahead," as we do below.
    if (inFreeRange()) lookahead_and_flush(fc, size);
D
duke 已提交
8471 8472 8473 8474 8475 8476 8477
  } else {
    // Code path common to both original and adaptive free lists.

    // cant coalesce with previous block; this should be treated
    // as the end of a free run if any
    if (inFreeRange()) {
      // we kicked some butt; time to pick up the garbage
8478 8479
      assert(freeFinger() < addr, "freeFinger points too high");
      flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
D
duke 已提交
8480 8481 8482 8483 8484
    }
    // else, nothing to do, just continue
  }
}

8485
size_t SweepClosure::do_garbage_chunk(FreeChunk* fc) {
D
duke 已提交
8486 8487 8488
  // This is a chunk of garbage.  It is not in any free list.
  // Add it to a free list or let it possibly be coalesced into
  // a larger chunk.
8489 8490
  HeapWord* const addr = (HeapWord*) fc;
  const size_t size = CompactibleFreeListSpace::adjustObjectSize(oop(addr)->size());
D
duke 已提交
8491 8492 8493 8494 8495 8496

  if (_sp->adaptive_freelists()) {
    // Verify that the bit map has no bits marked between
    // addr and purported end of just dead object.
    _bitMap->verifyNoOneBitsInRange(addr + 1, addr + size);

8497
    do_post_free_or_garbage_chunk(fc, size);
D
duke 已提交
8498 8499 8500 8501 8502 8503 8504 8505 8506 8507 8508 8509 8510 8511 8512 8513 8514 8515 8516 8517
  } else {
    if (!inFreeRange()) {
      // start of a new free range
      assert(size > 0, "A free range should have a size");
      initialize_free_range(addr, false);
    } else {
      // this will be swept up when we hit the end of the
      // free range
      if (CMSTraceSweeper) {
        gclog_or_tty->print("  -- pick up garbage 0x%x (%d) \n", fc, size);
      }
      // If the chunk is being coalesced and the current free range is
      // in the free lists, remove the current free range so that it
      // will be returned to the free lists in its entirety - all
      // the coalesced pieces included.
      if (freeRangeInFreeLists()) {
        FreeChunk* ffc = (FreeChunk*)freeFinger();
        assert(ffc->size() == pointer_delta(addr, freeFinger()),
          "Size of free range is inconsistent with chunk size.");
        if (CMSTestInFreeList) {
8518
          assert(_sp->verify_chunk_in_free_list(ffc),
D
duke 已提交
8519 8520 8521 8522 8523 8524 8525 8526 8527 8528 8529 8530 8531
            "free range is not in free lists");
        }
        _sp->removeFreeChunkFromFreeLists(ffc);
        set_freeRangeInFreeLists(false);
      }
      set_lastFreeRangeCoalesced(true);
    }
    // this will be swept up when we hit the end of the free range

    // Verify that the bit map has no bits marked between
    // addr and purported end of just dead object.
    _bitMap->verifyNoOneBitsInRange(addr + 1, addr + size);
  }
8532 8533 8534
  assert(_limit >= addr + size,
         "A freshly garbage chunk can't possibly straddle over _limit");
  if (inFreeRange()) lookahead_and_flush(fc, size);
D
duke 已提交
8535 8536 8537
  return size;
}

8538
size_t SweepClosure::do_live_chunk(FreeChunk* fc) {
D
duke 已提交
8539 8540 8541 8542
  HeapWord* addr = (HeapWord*) fc;
  // The sweeper has just found a live object. Return any accumulated
  // left hand chunk to the free lists.
  if (inFreeRange()) {
8543 8544
    assert(freeFinger() < addr, "freeFinger points too high");
    flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
D
duke 已提交
8545 8546
  }

8547
  // This object is live: we'd normally expect this to be
D
duke 已提交
8548 8549 8550 8551 8552 8553 8554 8555 8556 8557 8558 8559 8560 8561
  // an oop, and like to assert the following:
  // assert(oop(addr)->is_oop(), "live block should be an oop");
  // However, as we commented above, this may be an object whose
  // header hasn't yet been initialized.
  size_t size;
  assert(_bitMap->isMarked(addr), "Tautology for this control point");
  if (_bitMap->isMarked(addr + 1)) {
    // Determine the size from the bit map, rather than trying to
    // compute it from the object header.
    HeapWord* nextOneAddr = _bitMap->getNextMarkedWordAddress(addr + 2);
    size = pointer_delta(nextOneAddr + 1, addr);
    assert(size == CompactibleFreeListSpace::adjustObjectSize(size),
           "alignment problem");

D
drchase 已提交
8562
#ifdef ASSERT
8563
      if (oop(addr)->klass_or_null() != NULL) {
D
duke 已提交
8564 8565 8566 8567 8568 8569
        // Ignore mark word because we are running concurrent with mutators
        assert(oop(addr)->is_oop(true), "live block should be an oop");
        assert(size ==
               CompactibleFreeListSpace::adjustObjectSize(oop(addr)->size()),
               "P-mark and computed size do not agree");
      }
8570
#endif
D
duke 已提交
8571 8572 8573

  } else {
    // This should be an initialized object that's alive.
8574
    assert(oop(addr)->klass_or_null() != NULL,
D
duke 已提交
8575 8576 8577 8578 8579 8580 8581 8582 8583 8584 8585 8586 8587
           "Should be an initialized object");
    // Ignore mark word because we are running concurrent with mutators
    assert(oop(addr)->is_oop(true), "live block should be an oop");
    // Verify that the bit map has no bits marked between
    // addr and purported end of this block.
    size = CompactibleFreeListSpace::adjustObjectSize(oop(addr)->size());
    assert(size >= 3, "Necessary for Printezis marks to work");
    assert(!_bitMap->isMarked(addr+1), "Tautology for this control point");
    DEBUG_ONLY(_bitMap->verifyNoOneBitsInRange(addr+2, addr+size);)
  }
  return size;
}

8588 8589 8590 8591
void SweepClosure::do_post_free_or_garbage_chunk(FreeChunk* fc,
                                                 size_t chunkSize) {
  // do_post_free_or_garbage_chunk() should only be called in the case
  // of the adaptive free list allocator.
8592
  const bool fcInFreeLists = fc->is_free();
D
duke 已提交
8593 8594 8595
  assert(_sp->adaptive_freelists(), "Should only be used in this case.");
  assert((HeapWord*)fc <= _limit, "sweep invariant");
  if (CMSTestInFreeList && fcInFreeLists) {
8596
    assert(_sp->verify_chunk_in_free_list(fc), "free chunk is not in free lists");
D
duke 已提交
8597 8598 8599 8600 8601 8602
  }

  if (CMSTraceSweeper) {
    gclog_or_tty->print_cr("  -- pick up another chunk at 0x%x (%d)", fc, chunkSize);
  }

8603
  HeapWord* const fc_addr = (HeapWord*) fc;
D
duke 已提交
8604 8605

  bool coalesce;
8606 8607
  const size_t left  = pointer_delta(fc_addr, freeFinger());
  const size_t right = chunkSize;
D
duke 已提交
8608 8609 8610 8611 8612 8613 8614 8615 8616 8617 8618 8619 8620 8621 8622 8623 8624 8625 8626 8627 8628 8629 8630 8631 8632 8633 8634 8635 8636 8637 8638 8639
  switch (FLSCoalescePolicy) {
    // numeric value forms a coalition aggressiveness metric
    case 0:  { // never coalesce
      coalesce = false;
      break;
    }
    case 1: { // coalesce if left & right chunks on overpopulated lists
      coalesce = _sp->coalOverPopulated(left) &&
                 _sp->coalOverPopulated(right);
      break;
    }
    case 2: { // coalesce if left chunk on overpopulated list (default)
      coalesce = _sp->coalOverPopulated(left);
      break;
    }
    case 3: { // coalesce if left OR right chunk on overpopulated list
      coalesce = _sp->coalOverPopulated(left) ||
                 _sp->coalOverPopulated(right);
      break;
    }
    case 4: { // always coalesce
      coalesce = true;
      break;
    }
    default:
     ShouldNotReachHere();
  }

  // Should the current free range be coalesced?
  // If the chunk is in a free range and either we decided to coalesce above
  // or the chunk is near the large block at the end of the heap
  // (isNearLargestChunk() returns true), then coalesce this chunk.
8640 8641
  const bool doCoalesce = inFreeRange()
                          && (coalesce || _g->isNearLargestChunk(fc_addr));
D
duke 已提交
8642 8643 8644 8645 8646
  if (doCoalesce) {
    // Coalesce the current free range on the left with the new
    // chunk on the right.  If either is on a free list,
    // it must be removed from the list and stashed in the closure.
    if (freeRangeInFreeLists()) {
8647 8648
      FreeChunk* const ffc = (FreeChunk*)freeFinger();
      assert(ffc->size() == pointer_delta(fc_addr, freeFinger()),
D
duke 已提交
8649 8650
        "Size of free range is inconsistent with chunk size.");
      if (CMSTestInFreeList) {
8651
        assert(_sp->verify_chunk_in_free_list(ffc),
D
duke 已提交
8652 8653 8654 8655 8656 8657 8658 8659 8660 8661 8662 8663 8664
          "Chunk is not in free lists");
      }
      _sp->coalDeath(ffc->size());
      _sp->removeFreeChunkFromFreeLists(ffc);
      set_freeRangeInFreeLists(false);
    }
    if (fcInFreeLists) {
      _sp->coalDeath(chunkSize);
      assert(fc->size() == chunkSize,
        "The chunk has the wrong size or is not in the free lists");
      _sp->removeFreeChunkFromFreeLists(fc);
    }
    set_lastFreeRangeCoalesced(true);
8665
    print_free_block_coalesced(fc);
D
duke 已提交
8666 8667 8668 8669 8670
  } else {  // not in a free range and/or should not coalesce
    // Return the current free range and start a new one.
    if (inFreeRange()) {
      // In a free range but cannot coalesce with the right hand chunk.
      // Put the current free range into the free lists.
8671
      flush_cur_free_chunk(freeFinger(),
8672
                           pointer_delta(fc_addr, freeFinger()));
D
duke 已提交
8673 8674 8675 8676 8677 8678
    }
    // Set up for new free range.  Pass along whether the right hand
    // chunk is in the free lists.
    initialize_free_range((HeapWord*)fc, fcInFreeLists);
  }
}
8679

8680 8681 8682 8683 8684 8685 8686 8687 8688 8689 8690 8691 8692 8693 8694 8695 8696 8697 8698 8699
// Lookahead flush:
// If we are tracking a free range, and this is the last chunk that
// we'll look at because its end crosses past _limit, we'll preemptively
// flush it along with any free range we may be holding on to. Note that
// this can be the case only for an already free or freshly garbage
// chunk. If this block is an object, it can never straddle
// over _limit. The "straddling" occurs when _limit is set at
// the previous end of the space when this cycle started, and
// a subsequent heap expansion caused the previously co-terminal
// free block to be coalesced with the newly expanded portion,
// thus rendering _limit a non-block-boundary making it dangerous
// for the sweeper to step over and examine.
void SweepClosure::lookahead_and_flush(FreeChunk* fc, size_t chunk_size) {
  assert(inFreeRange(), "Should only be called if currently in a free range.");
  HeapWord* const eob = ((HeapWord*)fc) + chunk_size;
  assert(_sp->used_region().contains(eob - 1),
         err_msg("eob = " PTR_FORMAT " out of bounds wrt _sp = [" PTR_FORMAT "," PTR_FORMAT ")"
                 " when examining fc = " PTR_FORMAT "(" SIZE_FORMAT ")",
                 _limit, _sp->bottom(), _sp->end(), fc, chunk_size));
  if (eob >= _limit) {
8700
    assert(eob == _limit || fc->is_free(), "Only a free chunk should allow us to cross over the limit");
8701 8702 8703 8704 8705 8706 8707 8708 8709 8710 8711 8712 8713 8714 8715
    if (CMSTraceSweeper) {
      gclog_or_tty->print_cr("_limit " PTR_FORMAT " reached or crossed by block "
                             "[" PTR_FORMAT "," PTR_FORMAT ") in space "
                             "[" PTR_FORMAT "," PTR_FORMAT ")",
                             _limit, fc, eob, _sp->bottom(), _sp->end());
    }
    // Return the storage we are tracking back into the free lists.
    if (CMSTraceSweeper) {
      gclog_or_tty->print_cr("Flushing ... ");
    }
    assert(freeFinger() < eob, "Error");
    flush_cur_free_chunk( freeFinger(), pointer_delta(eob, freeFinger()));
  }
}

8716
void SweepClosure::flush_cur_free_chunk(HeapWord* chunk, size_t size) {
D
duke 已提交
8717 8718 8719 8720
  assert(inFreeRange(), "Should only be called if currently in a free range.");
  assert(size > 0,
    "A zero sized chunk cannot be added to the free lists.");
  if (!freeRangeInFreeLists()) {
8721
    if (CMSTestInFreeList) {
D
duke 已提交
8722
      FreeChunk* fc = (FreeChunk*) chunk;
8723 8724
      fc->set_size(size);
      assert(!_sp->verify_chunk_in_free_list(fc),
D
duke 已提交
8725 8726 8727 8728 8729 8730 8731 8732 8733 8734 8735 8736 8737 8738 8739 8740
        "chunk should not be in free lists yet");
    }
    if (CMSTraceSweeper) {
      gclog_or_tty->print_cr(" -- add free block 0x%x (%d) to free lists",
                    chunk, size);
    }
    // A new free range is going to be starting.  The current
    // free range has not been added to the free lists yet or
    // was removed so add it back.
    // If the current free range was coalesced, then the death
    // of the free range was recorded.  Record a birth now.
    if (lastFreeRangeCoalesced()) {
      _sp->coalBirth(size);
    }
    _sp->addChunkAndRepairOffsetTable(chunk, size,
            lastFreeRangeCoalesced());
8741 8742
  } else if (CMSTraceSweeper) {
    gclog_or_tty->print_cr("Already in free list: nothing to flush");
D
duke 已提交
8743 8744 8745 8746 8747 8748 8749 8750 8751 8752 8753 8754 8755 8756 8757
  }
  set_inFreeRange(false);
  set_freeRangeInFreeLists(false);
}

// We take a break if we've been at this for a while,
// so as to avoid monopolizing the locks involved.
void SweepClosure::do_yield_work(HeapWord* addr) {
  // Return current free chunk being used for coalescing (if any)
  // to the appropriate freelist.  After yielding, the next
  // free block encountered will start a coalescing range of
  // free blocks.  If the next free block is adjacent to the
  // chunk just flushed, they will need to wait for the next
  // sweep to be coalesced.
  if (inFreeRange()) {
8758
    flush_cur_free_chunk(freeFinger(), pointer_delta(addr, freeFinger()));
D
duke 已提交
8759 8760 8761 8762 8763 8764 8765 8766 8767 8768 8769 8770 8771 8772 8773 8774 8775 8776 8777 8778 8779 8780 8781 8782 8783 8784 8785 8786 8787 8788 8789 8790 8791 8792 8793 8794 8795 8796 8797
  }

  // First give up the locks, then yield, then re-lock.
  // We should probably use a constructor/destructor idiom to
  // do this unlock/lock or modify the MutexUnlocker class to
  // serve our purpose. XXX
  assert_lock_strong(_bitMap->lock());
  assert_lock_strong(_freelistLock);
  assert(ConcurrentMarkSweepThread::cms_thread_has_cms_token(),
         "CMS thread should hold CMS token");
  _bitMap->lock()->unlock();
  _freelistLock->unlock();
  ConcurrentMarkSweepThread::desynchronize(true);
  ConcurrentMarkSweepThread::acknowledge_yield_request();
  _collector->stopTimer();
  GCPauseTimer p(_collector->size_policy()->concurrent_timer_ptr());
  if (PrintCMSStatistics != 0) {
    _collector->incrementYields();
  }
  _collector->icms_wait();

  // See the comment in coordinator_yield()
  for (unsigned i = 0; i < CMSYieldSleepCount &&
                       ConcurrentMarkSweepThread::should_yield() &&
                       !CMSCollector::foregroundGCIsActive(); ++i) {
    os::sleep(Thread::current(), 1, false);
    ConcurrentMarkSweepThread::acknowledge_yield_request();
  }

  ConcurrentMarkSweepThread::synchronize(true);
  _freelistLock->lock();
  _bitMap->lock()->lock_without_safepoint_check();
  _collector->startTimer();
}

#ifndef PRODUCT
// This is actually very useful in a product build if it can
// be called from the debugger.  Compile it into the product
// as needed.
8798 8799
bool debug_verify_chunk_in_free_list(FreeChunk* fc) {
  return debug_cms_space->verify_chunk_in_free_list(fc);
D
duke 已提交
8800
}
8801
#endif
D
duke 已提交
8802

8803
void SweepClosure::print_free_block_coalesced(FreeChunk* fc) const {
D
duke 已提交
8804
  if (CMSTraceSweeper) {
8805 8806
    gclog_or_tty->print_cr("Sweep:coal_free_blk " PTR_FORMAT " (" SIZE_FORMAT ")",
                           fc, fc->size());
D
duke 已提交
8807 8808 8809 8810 8811 8812 8813 8814 8815 8816
  }
}

// CMSIsAliveClosure
bool CMSIsAliveClosure::do_object_b(oop obj) {
  HeapWord* addr = (HeapWord*)obj;
  return addr != NULL &&
         (!_span.contains(addr) || _bit_map->isMarked(addr));
}

8817

8818 8819 8820
CMSKeepAliveClosure::CMSKeepAliveClosure( CMSCollector* collector,
                      MemRegion span,
                      CMSBitMap* bit_map, CMSMarkStack* mark_stack,
8821 8822
                      bool cpc):
  _collector(collector),
8823 8824 8825 8826 8827 8828 8829 8830
  _span(span),
  _bit_map(bit_map),
  _mark_stack(mark_stack),
  _concurrent_precleaning(cpc) {
  assert(!_span.is_empty(), "Empty span could spell trouble");
}


D
duke 已提交
8831
// CMSKeepAliveClosure: the serial version
8832 8833
void CMSKeepAliveClosure::do_oop(oop obj) {
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
8834 8835 8836 8837 8838 8839 8840 8841 8842 8843 8844
  if (_span.contains(addr) &&
      !_bit_map->isMarked(addr)) {
    _bit_map->mark(addr);
    bool simulate_overflow = false;
    NOT_PRODUCT(
      if (CMSMarkStackOverflowALot &&
          _collector->simulate_overflow()) {
        // simulate a stack overflow
        simulate_overflow = true;
      }
    )
8845
    if (simulate_overflow || !_mark_stack->push(obj)) {
8846 8847 8848 8849 8850 8851 8852 8853 8854 8855 8856 8857 8858 8859 8860 8861 8862 8863 8864 8865 8866 8867 8868
      if (_concurrent_precleaning) {
        // We dirty the overflown object and let the remark
        // phase deal with it.
        assert(_collector->overflow_list_is_empty(), "Error");
        // In the case of object arrays, we need to dirty all of
        // the cards that the object spans. No locking or atomics
        // are needed since no one else can be mutating the mod union
        // table.
        if (obj->is_objArray()) {
          size_t sz = obj->size();
          HeapWord* end_card_addr =
            (HeapWord*)round_to((intptr_t)(addr+sz), CardTableModRefBS::card_size);
          MemRegion redirty_range = MemRegion(addr, end_card_addr);
          assert(!redirty_range.is_empty(), "Arithmetical tautology");
          _collector->_modUnionTable.mark_range(redirty_range);
        } else {
          _collector->_modUnionTable.mark(addr);
        }
        _collector->_ser_kac_preclean_ovflw++;
      } else {
        _collector->push_on_overflow_list(obj);
        _collector->_ser_kac_ovflw++;
      }
D
duke 已提交
8869 8870 8871 8872
    }
  }
}

8873 8874 8875
void CMSKeepAliveClosure::do_oop(oop* p)       { CMSKeepAliveClosure::do_oop_work(p); }
void CMSKeepAliveClosure::do_oop(narrowOop* p) { CMSKeepAliveClosure::do_oop_work(p); }

D
duke 已提交
8876 8877 8878
// CMSParKeepAliveClosure: a parallel version of the above.
// The work queues are private to each closure (thread),
// but (may be) available for stealing by other threads.
8879 8880
void CMSParKeepAliveClosure::do_oop(oop obj) {
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
8881 8882 8883 8884 8885 8886
  if (_span.contains(addr) &&
      !_bit_map->isMarked(addr)) {
    // In general, during recursive tracing, several threads
    // may be concurrently getting here; the first one to
    // "tag" it, claims it.
    if (_bit_map->par_mark(addr)) {
8887
      bool res = _work_queue->push(obj);
D
duke 已提交
8888 8889 8890 8891 8892 8893 8894 8895
      assert(res, "Low water mark should be much less than capacity");
      // Do a recursive trim in the hope that this will keep
      // stack usage lower, but leave some oops for potential stealers
      trim_queue(_low_water_mark);
    } // Else, another thread got there first
  }
}

8896 8897 8898
void CMSParKeepAliveClosure::do_oop(oop* p)       { CMSParKeepAliveClosure::do_oop_work(p); }
void CMSParKeepAliveClosure::do_oop(narrowOop* p) { CMSParKeepAliveClosure::do_oop_work(p); }

D
duke 已提交
8899 8900 8901 8902 8903 8904 8905 8906 8907 8908 8909 8910 8911 8912 8913
void CMSParKeepAliveClosure::trim_queue(uint max) {
  while (_work_queue->size() > max) {
    oop new_oop;
    if (_work_queue->pop_local(new_oop)) {
      assert(new_oop != NULL && new_oop->is_oop(), "Expected an oop");
      assert(_bit_map->isMarked((HeapWord*)new_oop),
             "no white objects on this stack!");
      assert(_span.contains((HeapWord*)new_oop), "Out of bounds oop");
      // iterate over the oops in this oop, marking and pushing
      // the ones in CMS heap (i.e. in _span).
      new_oop->oop_iterate(&_mark_and_push);
    }
  }
}

8914 8915 8916 8917
CMSInnerParMarkAndPushClosure::CMSInnerParMarkAndPushClosure(
                                CMSCollector* collector,
                                MemRegion span, CMSBitMap* bit_map,
                                OopTaskQueue* work_queue):
8918
  _collector(collector),
8919 8920 8921 8922
  _span(span),
  _bit_map(bit_map),
  _work_queue(work_queue) { }

8923 8924
void CMSInnerParMarkAndPushClosure::do_oop(oop obj) {
  HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
8925 8926 8927 8928 8929 8930 8931 8932 8933 8934 8935
  if (_span.contains(addr) &&
      !_bit_map->isMarked(addr)) {
    if (_bit_map->par_mark(addr)) {
      bool simulate_overflow = false;
      NOT_PRODUCT(
        if (CMSMarkStackOverflowALot &&
            _collector->par_simulate_overflow()) {
          // simulate a stack overflow
          simulate_overflow = true;
        }
      )
8936 8937
      if (simulate_overflow || !_work_queue->push(obj)) {
        _collector->par_push_on_overflow_list(obj);
D
duke 已提交
8938 8939 8940 8941 8942 8943
        _collector->_par_kac_ovflw++;
      }
    } // Else another thread got there already
  }
}

8944 8945 8946
void CMSInnerParMarkAndPushClosure::do_oop(oop* p)       { CMSInnerParMarkAndPushClosure::do_oop_work(p); }
void CMSInnerParMarkAndPushClosure::do_oop(narrowOop* p) { CMSInnerParMarkAndPushClosure::do_oop_work(p); }

D
duke 已提交
8947 8948 8949 8950 8951 8952 8953 8954 8955 8956 8957 8958 8959 8960 8961 8962 8963 8964 8965 8966 8967 8968 8969 8970 8971 8972 8973
//////////////////////////////////////////////////////////////////
//  CMSExpansionCause                /////////////////////////////
//////////////////////////////////////////////////////////////////
const char* CMSExpansionCause::to_string(CMSExpansionCause::Cause cause) {
  switch (cause) {
    case _no_expansion:
      return "No expansion";
    case _satisfy_free_ratio:
      return "Free ratio";
    case _satisfy_promotion:
      return "Satisfy promotion";
    case _satisfy_allocation:
      return "allocation";
    case _allocate_par_lab:
      return "Par LAB";
    case _allocate_par_spooling_space:
      return "Par Spooling Space";
    case _adaptive_size_policy:
      return "Ergonomics";
    default:
      return "unknown";
  }
}

void CMSDrainMarkingStackClosure::do_void() {
  // the max number to take from overflow list at a time
  const size_t num = _mark_stack->capacity()/4;
8974 8975
  assert(!_concurrent_precleaning || _collector->overflow_list_is_empty(),
         "Overflow list should be NULL during concurrent phases");
D
duke 已提交
8976 8977 8978
  while (!_mark_stack->isEmpty() ||
         // if stack is empty, check the overflow list
         _collector->take_from_overflow_list(num, _mark_stack)) {
8979 8980
    oop obj = _mark_stack->pop();
    HeapWord* addr = (HeapWord*)obj;
D
duke 已提交
8981 8982
    assert(_span.contains(addr), "Should be within span");
    assert(_bit_map->isMarked(addr), "Should be marked");
8983 8984
    assert(obj->is_oop(), "Should be an oop");
    obj->oop_iterate(_keep_alive);
D
duke 已提交
8985 8986 8987 8988 8989 8990 8991 8992 8993 8994 8995 8996 8997 8998 8999 9000 9001 9002 9003 9004 9005 9006 9007 9008 9009 9010 9011 9012 9013 9014 9015 9016 9017 9018 9019 9020 9021 9022 9023 9024 9025 9026 9027 9028 9029 9030 9031 9032 9033 9034 9035 9036 9037 9038 9039 9040 9041 9042 9043 9044
  }
}

void CMSParDrainMarkingStackClosure::do_void() {
  // drain queue
  trim_queue(0);
}

// Trim our work_queue so its length is below max at return
void CMSParDrainMarkingStackClosure::trim_queue(uint max) {
  while (_work_queue->size() > max) {
    oop new_oop;
    if (_work_queue->pop_local(new_oop)) {
      assert(new_oop->is_oop(), "Expected an oop");
      assert(_bit_map->isMarked((HeapWord*)new_oop),
             "no white objects on this stack!");
      assert(_span.contains((HeapWord*)new_oop), "Out of bounds oop");
      // iterate over the oops in this oop, marking and pushing
      // the ones in CMS heap (i.e. in _span).
      new_oop->oop_iterate(&_mark_and_push);
    }
  }
}

////////////////////////////////////////////////////////////////////
// Support for Marking Stack Overflow list handling and related code
////////////////////////////////////////////////////////////////////
// Much of the following code is similar in shape and spirit to the
// code used in ParNewGC. We should try and share that code
// as much as possible in the future.

#ifndef PRODUCT
// Debugging support for CMSStackOverflowALot

// It's OK to call this multi-threaded;  the worst thing
// that can happen is that we'll get a bunch of closely
// spaced simulated oveflows, but that's OK, in fact
// probably good as it would exercise the overflow code
// under contention.
bool CMSCollector::simulate_overflow() {
  if (_overflow_counter-- <= 0) { // just being defensive
    _overflow_counter = CMSMarkStackOverflowInterval;
    return true;
  } else {
    return false;
  }
}

bool CMSCollector::par_simulate_overflow() {
  return simulate_overflow();
}
#endif

// Single-threaded
bool CMSCollector::take_from_overflow_list(size_t num, CMSMarkStack* stack) {
  assert(stack->isEmpty(), "Expected precondition");
  assert(stack->capacity() > num, "Shouldn't bite more than can chew");
  size_t i = num;
  oop  cur = _overflow_list;
  const markOop proto = markOopDesc::prototype();
9045
  NOT_PRODUCT(ssize_t n = 0;)
D
duke 已提交
9046 9047 9048 9049 9050 9051 9052 9053 9054 9055 9056 9057 9058 9059 9060 9061
  for (oop next; i > 0 && cur != NULL; cur = next, i--) {
    next = oop(cur->mark());
    cur->set_mark(proto);   // until proven otherwise
    assert(cur->is_oop(), "Should be an oop");
    bool res = stack->push(cur);
    assert(res, "Bit off more than can chew?");
    NOT_PRODUCT(n++;)
  }
  _overflow_list = cur;
#ifndef PRODUCT
  assert(_num_par_pushes >= n, "Too many pops?");
  _num_par_pushes -=n;
#endif
  return !stack->isEmpty();
}

9062 9063 9064 9065 9066 9067 9068 9069 9070 9071 9072 9073 9074 9075 9076 9077 9078 9079 9080 9081 9082 9083 9084 9085
#define BUSY  (oop(0x1aff1aff))
// (MT-safe) Get a prefix of at most "num" from the list.
// The overflow list is chained through the mark word of
// each object in the list. We fetch the entire list,
// break off a prefix of the right size and return the
// remainder. If other threads try to take objects from
// the overflow list at that time, they will wait for
// some time to see if data becomes available. If (and
// only if) another thread places one or more object(s)
// on the global list before we have returned the suffix
// to the global list, we will walk down our local list
// to find its end and append the global list to
// our suffix before returning it. This suffix walk can
// prove to be expensive (quadratic in the amount of traffic)
// when there are many objects in the overflow list and
// there is much producer-consumer contention on the list.
// *NOTE*: The overflow list manipulation code here and
// in ParNewGeneration:: are very similar in shape,
// except that in the ParNew case we use the old (from/eden)
// copy of the object to thread the list via its klass word.
// Because of the common code, if you make any changes in
// the code below, please check the ParNew version to see if
// similar changes might be needed.
// CR 6797058 has been filed to consolidate the common code.
D
duke 已提交
9086
bool CMSCollector::par_take_from_overflow_list(size_t num,
9087 9088
                                               OopTaskQueue* work_q,
                                               int no_of_gc_threads) {
9089
  assert(work_q->size() == 0, "First empty local work queue");
D
duke 已提交
9090 9091 9092 9093 9094
  assert(num < work_q->max_elems(), "Can't bite more than we can chew");
  if (_overflow_list == NULL) {
    return false;
  }
  // Grab the entire list; we'll put back a suffix
9095 9096
  oop prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list);
  Thread* tid = Thread::current();
9097 9098 9099
  // Before "no_of_gc_threads" was introduced CMSOverflowSpinCount was
  // set to ParallelGCThreads.
  size_t CMSOverflowSpinCount = (size_t) no_of_gc_threads; // was ParallelGCThreads;
9100 9101 9102 9103 9104 9105 9106 9107 9108 9109 9110 9111 9112 9113 9114 9115 9116 9117 9118 9119 9120 9121 9122 9123 9124 9125
  size_t sleep_time_millis = MAX2((size_t)1, num/100);
  // If the list is busy, we spin for a short while,
  // sleeping between attempts to get the list.
  for (size_t spin = 0; prefix == BUSY && spin < CMSOverflowSpinCount; spin++) {
    os::sleep(tid, sleep_time_millis, false);
    if (_overflow_list == NULL) {
      // Nothing left to take
      return false;
    } else if (_overflow_list != BUSY) {
      // Try and grab the prefix
      prefix = (oop)Atomic::xchg_ptr(BUSY, &_overflow_list);
    }
  }
  // If the list was found to be empty, or we spun long
  // enough, we give up and return empty-handed. If we leave
  // the list in the BUSY state below, it must be the case that
  // some other thread holds the overflow list and will set it
  // to a non-BUSY state in the future.
  if (prefix == NULL || prefix == BUSY) {
     // Nothing to take or waited long enough
     if (prefix == NULL) {
       // Write back the NULL in case we overwrote it with BUSY above
       // and it is still the same value.
       (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY);
     }
     return false;
D
duke 已提交
9126
  }
9127
  assert(prefix != NULL && prefix != BUSY, "Error");
D
duke 已提交
9128 9129
  size_t i = num;
  oop cur = prefix;
9130
  // Walk down the first "num" objects, unless we reach the end.
D
duke 已提交
9131
  for (; i > 1 && cur->mark() != NULL; cur = oop(cur->mark()), i--);
9132 9133 9134 9135 9136 9137 9138 9139 9140 9141 9142
  if (cur->mark() == NULL) {
    // We have "num" or fewer elements in the list, so there
    // is nothing to return to the global list.
    // Write back the NULL in lieu of the BUSY we wrote
    // above, if it is still the same value.
    if (_overflow_list == BUSY) {
      (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY);
    }
  } else {
    // Chop off the suffix and rerturn it to the global list.
    assert(cur->mark() != BUSY, "Error");
D
duke 已提交
9143 9144
    oop suffix_head = cur->mark(); // suffix will be put back on global list
    cur->set_mark(NULL);           // break off suffix
9145 9146 9147 9148
    // It's possible that the list is still in the empty(busy) state
    // we left it in a short while ago; in that case we may be
    // able to place back the suffix without incurring the cost
    // of a walk down the list.
D
duke 已提交
9149
    oop observed_overflow_list = _overflow_list;
9150 9151 9152
    oop cur_overflow_list = observed_overflow_list;
    bool attached = false;
    while (observed_overflow_list == BUSY || observed_overflow_list == NULL) {
D
duke 已提交
9153
      observed_overflow_list =
9154 9155 9156 9157 9158 9159 9160 9161 9162 9163 9164 9165 9166 9167 9168 9169 9170 9171 9172 9173 9174 9175 9176 9177 9178 9179 9180 9181 9182
        (oop) Atomic::cmpxchg_ptr(suffix_head, &_overflow_list, cur_overflow_list);
      if (cur_overflow_list == observed_overflow_list) {
        attached = true;
        break;
      } else cur_overflow_list = observed_overflow_list;
    }
    if (!attached) {
      // Too bad, someone else sneaked in (at least) an element; we'll need
      // to do a splice. Find tail of suffix so we can prepend suffix to global
      // list.
      for (cur = suffix_head; cur->mark() != NULL; cur = (oop)(cur->mark()));
      oop suffix_tail = cur;
      assert(suffix_tail != NULL && suffix_tail->mark() == NULL,
             "Tautology");
      observed_overflow_list = _overflow_list;
      do {
        cur_overflow_list = observed_overflow_list;
        if (cur_overflow_list != BUSY) {
          // Do the splice ...
          suffix_tail->set_mark(markOop(cur_overflow_list));
        } else { // cur_overflow_list == BUSY
          suffix_tail->set_mark(NULL);
        }
        // ... and try to place spliced list back on overflow_list ...
        observed_overflow_list =
          (oop) Atomic::cmpxchg_ptr(suffix_head, &_overflow_list, cur_overflow_list);
      } while (cur_overflow_list != observed_overflow_list);
      // ... until we have succeeded in doing so.
    }
D
duke 已提交
9183 9184 9185 9186 9187 9188
  }

  // Push the prefix elements on work_q
  assert(prefix != NULL, "control point invariant");
  const markOop proto = markOopDesc::prototype();
  oop next;
9189
  NOT_PRODUCT(ssize_t n = 0;)
D
duke 已提交
9190 9191 9192 9193 9194 9195 9196 9197 9198 9199 9200 9201 9202 9203 9204 9205 9206 9207 9208 9209 9210 9211 9212 9213 9214 9215 9216 9217 9218 9219 9220 9221 9222
  for (cur = prefix; cur != NULL; cur = next) {
    next = oop(cur->mark());
    cur->set_mark(proto);   // until proven otherwise
    assert(cur->is_oop(), "Should be an oop");
    bool res = work_q->push(cur);
    assert(res, "Bit off more than we can chew?");
    NOT_PRODUCT(n++;)
  }
#ifndef PRODUCT
  assert(_num_par_pushes >= n, "Too many pops?");
  Atomic::add_ptr(-(intptr_t)n, &_num_par_pushes);
#endif
  return true;
}

// Single-threaded
void CMSCollector::push_on_overflow_list(oop p) {
  NOT_PRODUCT(_num_par_pushes++;)
  assert(p->is_oop(), "Not an oop");
  preserve_mark_if_necessary(p);
  p->set_mark((markOop)_overflow_list);
  _overflow_list = p;
}

// Multi-threaded; use CAS to prepend to overflow list
void CMSCollector::par_push_on_overflow_list(oop p) {
  NOT_PRODUCT(Atomic::inc_ptr(&_num_par_pushes);)
  assert(p->is_oop(), "Not an oop");
  par_preserve_mark_if_necessary(p);
  oop observed_overflow_list = _overflow_list;
  oop cur_overflow_list;
  do {
    cur_overflow_list = observed_overflow_list;
9223 9224 9225 9226 9227
    if (cur_overflow_list != BUSY) {
      p->set_mark(markOop(cur_overflow_list));
    } else {
      p->set_mark(NULL);
    }
D
duke 已提交
9228 9229 9230 9231
    observed_overflow_list =
      (oop) Atomic::cmpxchg_ptr(p, &_overflow_list, cur_overflow_list);
  } while (cur_overflow_list != observed_overflow_list);
}
9232
#undef BUSY
D
duke 已提交
9233 9234 9235 9236 9237 9238 9239 9240

// Single threaded
// General Note on GrowableArray: pushes may silently fail
// because we are (temporarily) out of C-heap for expanding
// the stack. The problem is quite ubiquitous and affects
// a lot of code in the JVM. The prudent thing for GrowableArray
// to do (for now) is to exit with an error. However, that may
// be too draconian in some cases because the caller may be
9241
// able to recover without much harm. For such cases, we
D
duke 已提交
9242 9243 9244 9245 9246 9247 9248
// should probably introduce a "soft_push" method which returns
// an indication of success or failure with the assumption that
// the caller may be able to recover from a failure; code in
// the VM can then be changed, incrementally, to deal with such
// failures where possible, thus, incrementally hardening the VM
// in such low resource situations.
void CMSCollector::preserve_mark_work(oop p, markOop m) {
9249 9250
  _preserved_oop_stack.push(p);
  _preserved_mark_stack.push(m);
D
duke 已提交
9251
  assert(m == p->mark(), "Mark word changed");
9252
  assert(_preserved_oop_stack.size() == _preserved_mark_stack.size(),
D
duke 已提交
9253 9254 9255 9256 9257 9258 9259 9260 9261 9262 9263 9264 9265 9266 9267 9268 9269 9270 9271 9272 9273 9274 9275 9276 9277 9278 9279 9280 9281 9282 9283 9284 9285 9286 9287 9288 9289 9290 9291 9292 9293 9294 9295 9296 9297 9298
         "bijection");
}

// Single threaded
void CMSCollector::preserve_mark_if_necessary(oop p) {
  markOop m = p->mark();
  if (m->must_be_preserved(p)) {
    preserve_mark_work(p, m);
  }
}

void CMSCollector::par_preserve_mark_if_necessary(oop p) {
  markOop m = p->mark();
  if (m->must_be_preserved(p)) {
    MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
    // Even though we read the mark word without holding
    // the lock, we are assured that it will not change
    // because we "own" this oop, so no other thread can
    // be trying to push it on the overflow list; see
    // the assertion in preserve_mark_work() that checks
    // that m == p->mark().
    preserve_mark_work(p, m);
  }
}

// We should be able to do this multi-threaded,
// a chunk of stack being a task (this is
// correct because each oop only ever appears
// once in the overflow list. However, it's
// not very easy to completely overlap this with
// other operations, so will generally not be done
// until all work's been completed. Because we
// expect the preserved oop stack (set) to be small,
// it's probably fine to do this single-threaded.
// We can explore cleverer concurrent/overlapped/parallel
// processing of preserved marks if we feel the
// need for this in the future. Stack overflow should
// be so rare in practice and, when it happens, its
// effect on performance so great that this will
// likely just be in the noise anyway.
void CMSCollector::restore_preserved_marks_if_any() {
  assert(SafepointSynchronize::is_at_safepoint(),
         "world should be stopped");
  assert(Thread::current()->is_ConcurrentGC_thread() ||
         Thread::current()->is_VM_thread(),
         "should be single-threaded");
9299 9300
  assert(_preserved_oop_stack.size() == _preserved_mark_stack.size(),
         "bijection");
D
duke 已提交
9301

9302 9303
  while (!_preserved_oop_stack.is_empty()) {
    oop p = _preserved_oop_stack.pop();
D
duke 已提交
9304 9305 9306 9307
    assert(p->is_oop(), "Should be an oop");
    assert(_span.contains(p), "oop should be in _span");
    assert(p->mark() == markOopDesc::prototype(),
           "Set when taken from overflow list");
9308
    markOop m = _preserved_mark_stack.pop();
D
duke 已提交
9309 9310
    p->set_mark(m);
  }
9311
  assert(_preserved_mark_stack.is_empty() && _preserved_oop_stack.is_empty(),
D
duke 已提交
9312 9313 9314 9315 9316
         "stacks were cleared above");
}

#ifndef PRODUCT
bool CMSCollector::no_preserved_marks() const {
9317
  return _preserved_mark_stack.is_empty() && _preserved_oop_stack.is_empty();
D
duke 已提交
9318 9319 9320 9321 9322 9323 9324 9325 9326 9327 9328 9329 9330 9331 9332 9333 9334 9335 9336 9337 9338 9339 9340 9341 9342 9343 9344 9345 9346 9347 9348 9349 9350 9351 9352 9353 9354 9355 9356 9357 9358 9359 9360 9361 9362 9363 9364 9365 9366 9367 9368 9369 9370 9371 9372 9373 9374 9375 9376 9377 9378 9379 9380 9381 9382 9383 9384 9385 9386 9387 9388 9389 9390 9391 9392 9393 9394 9395 9396 9397 9398 9399 9400 9401 9402 9403 9404 9405 9406 9407 9408 9409 9410 9411 9412 9413 9414
}
#endif

CMSAdaptiveSizePolicy* ASConcurrentMarkSweepGeneration::cms_size_policy() const
{
  GenCollectedHeap* gch = (GenCollectedHeap*) GenCollectedHeap::heap();
  CMSAdaptiveSizePolicy* size_policy =
    (CMSAdaptiveSizePolicy*) gch->gen_policy()->size_policy();
  assert(size_policy->is_gc_cms_adaptive_size_policy(),
    "Wrong type for size policy");
  return size_policy;
}

void ASConcurrentMarkSweepGeneration::resize(size_t cur_promo_size,
                                           size_t desired_promo_size) {
  if (cur_promo_size < desired_promo_size) {
    size_t expand_bytes = desired_promo_size - cur_promo_size;
    if (PrintAdaptiveSizePolicy && Verbose) {
      gclog_or_tty->print_cr(" ASConcurrentMarkSweepGeneration::resize "
        "Expanding tenured generation by " SIZE_FORMAT " (bytes)",
        expand_bytes);
    }
    expand(expand_bytes,
           MinHeapDeltaBytes,
           CMSExpansionCause::_adaptive_size_policy);
  } else if (desired_promo_size < cur_promo_size) {
    size_t shrink_bytes = cur_promo_size - desired_promo_size;
    if (PrintAdaptiveSizePolicy && Verbose) {
      gclog_or_tty->print_cr(" ASConcurrentMarkSweepGeneration::resize "
        "Shrinking tenured generation by " SIZE_FORMAT " (bytes)",
        shrink_bytes);
    }
    shrink(shrink_bytes);
  }
}

CMSGCAdaptivePolicyCounters* ASConcurrentMarkSweepGeneration::gc_adaptive_policy_counters() {
  GenCollectedHeap* gch = GenCollectedHeap::heap();
  CMSGCAdaptivePolicyCounters* counters =
    (CMSGCAdaptivePolicyCounters*) gch->collector_policy()->counters();
  assert(counters->kind() == GCPolicyCounters::CMSGCAdaptivePolicyCountersKind,
    "Wrong kind of counters");
  return counters;
}


void ASConcurrentMarkSweepGeneration::update_counters() {
  if (UsePerfData) {
    _space_counters->update_all();
    _gen_counters->update_all();
    CMSGCAdaptivePolicyCounters* counters = gc_adaptive_policy_counters();
    GenCollectedHeap* gch = GenCollectedHeap::heap();
    CMSGCStats* gc_stats_l = (CMSGCStats*) gc_stats();
    assert(gc_stats_l->kind() == GCStats::CMSGCStatsKind,
      "Wrong gc statistics type");
    counters->update_counters(gc_stats_l);
  }
}

void ASConcurrentMarkSweepGeneration::update_counters(size_t used) {
  if (UsePerfData) {
    _space_counters->update_used(used);
    _space_counters->update_capacity();
    _gen_counters->update_all();

    CMSGCAdaptivePolicyCounters* counters = gc_adaptive_policy_counters();
    GenCollectedHeap* gch = GenCollectedHeap::heap();
    CMSGCStats* gc_stats_l = (CMSGCStats*) gc_stats();
    assert(gc_stats_l->kind() == GCStats::CMSGCStatsKind,
      "Wrong gc statistics type");
    counters->update_counters(gc_stats_l);
  }
}

void ASConcurrentMarkSweepGeneration::shrink_by(size_t desired_bytes) {
  assert_locked_or_safepoint(Heap_lock);
  assert_lock_strong(freelistLock());
  HeapWord* old_end = _cmsSpace->end();
  HeapWord* unallocated_start = _cmsSpace->unallocated_block();
  assert(old_end >= unallocated_start, "Miscalculation of unallocated_start");
  FreeChunk* chunk_at_end = find_chunk_at_end();
  if (chunk_at_end == NULL) {
    // No room to shrink
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print_cr("No room to shrink: old_end  "
        PTR_FORMAT "  unallocated_start  " PTR_FORMAT
        " chunk_at_end  " PTR_FORMAT,
        old_end, unallocated_start, chunk_at_end);
    }
    return;
  } else {

    // Find the chunk at the end of the space and determine
    // how much it can be shrunk.
    size_t shrinkable_size_in_bytes = chunk_at_end->size();
    size_t aligned_shrinkable_size_in_bytes =
      align_size_down(shrinkable_size_in_bytes, os::vm_page_size());
9415
    assert(unallocated_start <= (HeapWord*) chunk_at_end->end(),
D
duke 已提交
9416 9417 9418 9419 9420 9421 9422 9423 9424 9425 9426 9427 9428 9429 9430 9431 9432 9433 9434 9435 9436 9437 9438 9439 9440 9441 9442 9443 9444 9445 9446 9447 9448 9449 9450
      "Inconsistent chunk at end of space");
    size_t bytes = MIN2(desired_bytes, aligned_shrinkable_size_in_bytes);
    size_t word_size_before = heap_word_size(_virtual_space.committed_size());

    // Shrink the underlying space
    _virtual_space.shrink_by(bytes);
    if (PrintGCDetails && Verbose) {
      gclog_or_tty->print_cr("ConcurrentMarkSweepGeneration::shrink_by:"
        " desired_bytes " SIZE_FORMAT
        " shrinkable_size_in_bytes " SIZE_FORMAT
        " aligned_shrinkable_size_in_bytes " SIZE_FORMAT
        "  bytes  " SIZE_FORMAT,
        desired_bytes, shrinkable_size_in_bytes,
        aligned_shrinkable_size_in_bytes, bytes);
      gclog_or_tty->print_cr("          old_end  " SIZE_FORMAT
        "  unallocated_start  " SIZE_FORMAT,
        old_end, unallocated_start);
    }

    // If the space did shrink (shrinking is not guaranteed),
    // shrink the chunk at the end by the appropriate amount.
    if (((HeapWord*)_virtual_space.high()) < old_end) {
      size_t new_word_size =
        heap_word_size(_virtual_space.committed_size());

      // Have to remove the chunk from the dictionary because it is changing
      // size and might be someplace elsewhere in the dictionary.

      // Get the chunk at end, shrink it, and put it
      // back.
      _cmsSpace->removeChunkFromDictionary(chunk_at_end);
      size_t word_size_change = word_size_before - new_word_size;
      size_t chunk_at_end_old_size = chunk_at_end->size();
      assert(chunk_at_end_old_size >= word_size_change,
        "Shrink is too large");
9451
      chunk_at_end->set_size(chunk_at_end_old_size -
D
duke 已提交
9452 9453 9454 9455 9456 9457 9458 9459 9460 9461 9462 9463 9464 9465 9466 9467 9468 9469 9470 9471 9472 9473 9474
                          word_size_change);
      _cmsSpace->freed((HeapWord*) chunk_at_end->end(),
        word_size_change);

      _cmsSpace->returnChunkToDictionary(chunk_at_end);

      MemRegion mr(_cmsSpace->bottom(), new_word_size);
      _bts->resize(new_word_size);  // resize the block offset shared array
      Universe::heap()->barrier_set()->resize_covered_region(mr);
      _cmsSpace->assert_locked();
      _cmsSpace->set_end((HeapWord*)_virtual_space.high());

      NOT_PRODUCT(_cmsSpace->dictionary()->verify());

      // update the space and generation capacity counters
      if (UsePerfData) {
        _space_counters->update_capacity();
        _gen_counters->update_all();
      }

      if (Verbose && PrintGCDetails) {
        size_t new_mem_size = _virtual_space.committed_size();
        size_t old_mem_size = new_mem_size + bytes;
9475
        gclog_or_tty->print_cr("Shrinking %s from " SIZE_FORMAT "K by " SIZE_FORMAT "K to " SIZE_FORMAT "K",
D
duke 已提交
9476 9477 9478 9479 9480 9481
                      name(), old_mem_size/K, bytes/K, new_mem_size/K);
      }
    }

    assert(_cmsSpace->unallocated_block() <= _cmsSpace->end(),
      "Inconsistency at end of space");
9482
    assert(chunk_at_end->end() == (uintptr_t*) _cmsSpace->end(),
D
duke 已提交
9483 9484 9485 9486 9487 9488 9489
      "Shrinking is inconsistent");
    return;
  }
}
// Transfer some number of overflown objects to usual marking
// stack. Return true if some objects were transferred.
bool MarkRefsIntoAndScanClosure::take_from_overflow_list() {
9490
  size_t num = MIN2((size_t)(_mark_stack->capacity() - _mark_stack->length())/4,
D
duke 已提交
9491 9492 9493 9494 9495 9496 9497 9498 9499 9500 9501 9502 9503 9504 9505 9506 9507 9508 9509 9510
                    (size_t)ParGCDesiredObjsFromOverflowList);

  bool res = _collector->take_from_overflow_list(num, _mark_stack);
  assert(_collector->overflow_list_is_empty() || res,
         "If list is not empty, we should have taken something");
  assert(!res || !_mark_stack->isEmpty(),
         "If we took something, it should now be on our stack");
  return res;
}

size_t MarkDeadObjectsClosure::do_blk(HeapWord* addr) {
  size_t res = _sp->block_size_no_stall(addr, _collector);
  if (_sp->block_is_obj(addr)) {
    if (_live_bit_map->isMarked(addr)) {
      // It can't have been dead in a previous cycle
      guarantee(!_dead_bit_map->isMarked(addr), "No resurrection!");
    } else {
      _dead_bit_map->mark(addr);      // mark the dead object
    }
  }
9511
  // Could be 0, if the block size could not be computed without stalling.
D
duke 已提交
9512 9513
  return res;
}
9514

9515
TraceCMSMemoryManagerStats::TraceCMSMemoryManagerStats(CMSCollector::CollectorState phase, GCCause::Cause cause): TraceMemoryManagerStats() {
9516 9517 9518 9519

  switch (phase) {
    case CMSCollector::InitialMarking:
      initialize(true  /* fullGC */ ,
9520
                 cause /* cause of the GC */,
9521 9522 9523 9524 9525 9526 9527 9528 9529 9530 9531
                 true  /* recordGCBeginTime */,
                 true  /* recordPreGCUsage */,
                 false /* recordPeakUsage */,
                 false /* recordPostGCusage */,
                 true  /* recordAccumulatedGCTime */,
                 false /* recordGCEndTime */,
                 false /* countCollection */  );
      break;

    case CMSCollector::FinalMarking:
      initialize(true  /* fullGC */ ,
9532
                 cause /* cause of the GC */,
9533 9534 9535 9536 9537 9538 9539 9540 9541 9542 9543
                 false /* recordGCBeginTime */,
                 false /* recordPreGCUsage */,
                 false /* recordPeakUsage */,
                 false /* recordPostGCusage */,
                 true  /* recordAccumulatedGCTime */,
                 false /* recordGCEndTime */,
                 false /* countCollection */  );
      break;

    case CMSCollector::Sweeping:
      initialize(true  /* fullGC */ ,
9544
                 cause /* cause of the GC */,
9545 9546 9547 9548 9549 9550 9551 9552 9553 9554 9555 9556 9557
                 false /* recordGCBeginTime */,
                 false /* recordPreGCUsage */,
                 true  /* recordPeakUsage */,
                 true  /* recordPostGCusage */,
                 false /* recordAccumulatedGCTime */,
                 true  /* recordGCEndTime */,
                 true  /* countCollection */  );
      break;

    default:
      ShouldNotReachHere();
  }
}