/*

 * Copyright (c) 2001, 2017, Oracle and/or its affiliates. All rights reserved.

 * 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.

 *

 * 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.

 *

 */

#include "precompiled.hpp"

#include "gc/g1/concurrentG1Refine.hpp"

#include "gc/g1/dirtyCardQueue.hpp"

#include "gc/g1/g1BlockOffsetTable.inline.hpp"

#include "gc/g1/g1CollectedHeap.inline.hpp"

#include "gc/g1/g1FromCardCache.hpp"

#include "gc/g1/g1GCPhaseTimes.hpp"

#include "gc/g1/g1HotCardCache.hpp"

#include "gc/g1/g1OopClosures.inline.hpp"

#include "gc/g1/g1RemSet.inline.hpp"

#include "gc/g1/g1SATBCardTableModRefBS.inline.hpp"

#include "gc/g1/heapRegion.inline.hpp"

#include "gc/g1/heapRegionManager.inline.hpp"

#include "gc/g1/heapRegionRemSet.hpp"

#include "gc/shared/gcTraceTime.inline.hpp"

#include "memory/iterator.hpp"

#include "memory/resourceArea.hpp"

#include "oops/oop.inline.hpp"

#include "utilities/align.hpp"

#include "utilities/globalDefinitions.hpp"

#include "utilities/intHisto.hpp"

#include "utilities/stack.inline.hpp"

// Collects information about the overall remembered set scan progress during an evacuation.

class G1RemSetScanState : public CHeapObj<mtGC> {

private:

  class G1ClearCardTableTask : public AbstractGangTask {

    G1CollectedHeap* _g1h;

    uint* _dirty_region_list;

    size_t _num_dirty_regions;

    size_t _chunk_length;

    size_t volatile _cur_dirty_regions;

  public:

    G1ClearCardTableTask(G1CollectedHeap* g1h,

                         uint* dirty_region_list,

                         size_t num_dirty_regions,

                         size_t chunk_length) :

      AbstractGangTask("G1 Clear Card Table Task"),

      _g1h(g1h),

      _dirty_region_list(dirty_region_list),

      _num_dirty_regions(num_dirty_regions),

      _chunk_length(chunk_length),

      _cur_dirty_regions(0) {

      assert(chunk_length > 0, "must be");

    }

    static size_t chunk_size() { return M; }

    void work(uint worker_id) {

      G1SATBCardTableModRefBS* ct_bs = _g1h->g1_barrier_set();

      while (_cur_dirty_regions < _num_dirty_regions) {

        size_t next = Atomic::add(_chunk_length, &_cur_dirty_regions) - _chunk_length;

        size_t max = MIN2(next + _chunk_length, _num_dirty_regions);

        for (size_t i = next; i < max; i++) {

          HeapRegion* r = _g1h->region_at(_dirty_region_list[i]);

          if (!r->is_survivor()) {

            ct_bs->clear(MemRegion(r->bottom(), r->end()));

          }

        }

      }

    }

  };

  size_t _max_regions;

  // Scan progress for the remembered set of a single region. Transitions from

  // Unclaimed -> Claimed -> Complete.

  // At each of the transitions the thread that does the transition needs to perform

  // some special action once. This is the reason for the extra "Claimed" state.

  typedef jint G1RemsetIterState;

  static const G1RemsetIterState Unclaimed = 0; // The remembered set has not been scanned yet.

  static const G1RemsetIterState Claimed = 1;   // The remembered set is currently being scanned.

  static const G1RemsetIterState Complete = 2;  // The remembered set has been completely scanned.

  G1RemsetIterState volatile* _iter_states;

  // The current location where the next thread should continue scanning in a region's

  // remembered set.

  size_t volatile* _iter_claims;

  // Temporary buffer holding the regions we used to store remembered set scan duplicate

  // information. These are also called "dirty". Valid entries are from [0.._cur_dirty_region)

  uint* _dirty_region_buffer;

  typedef jbyte IsDirtyRegionState;

  static const IsDirtyRegionState Clean = 0;

  static const IsDirtyRegionState Dirty = 1;

  // Holds a flag for every region whether it is in the _dirty_region_buffer already

  // to avoid duplicates. Uses jbyte since there are no atomic instructions for bools.

  IsDirtyRegionState* _in_dirty_region_buffer;

  size_t _cur_dirty_region;

  // Creates a snapshot of the current _top values at the start of collection to

  // filter out card marks that we do not want to scan.

  class G1ResetScanTopClosure : public HeapRegionClosure {

  private:

    HeapWord** _scan_top;

  public:

    G1ResetScanTopClosure(HeapWord** scan_top) : _scan_top(scan_top) { }

    virtual bool doHeapRegion(HeapRegion* r) {

      uint hrm_index = r->hrm_index();

      if (!r->in_collection_set() && r->is_old_or_humongous()) {

        _scan_top[hrm_index] = r->top();

      } else {

        _scan_top[hrm_index] = r->bottom();

      }

      return false;

    }

  };

  // For each region, contains the maximum top() value to be used during this garbage

  // collection. Subsumes common checks like filtering out everything but old and

  // humongous regions outside the collection set.

  // This is valid because we are not interested in scanning stray remembered set

  // entries from free or archive regions.

  HeapWord** _scan_top;

public:

  G1RemSetScanState() :

    _max_regions(0),

    _iter_states(NULL),

    _iter_claims(NULL),

    _dirty_region_buffer(NULL),

    _in_dirty_region_buffer(NULL),

    _cur_dirty_region(0),

    _scan_top(NULL) {

  }

  ~G1RemSetScanState() {

    if (_iter_states != NULL) {

      FREE_C_HEAP_ARRAY(G1RemsetIterState, _iter_states);

    }

    if (_iter_claims != NULL) {

      FREE_C_HEAP_ARRAY(size_t, _iter_claims);

    }

    if (_dirty_region_buffer != NULL) {

      FREE_C_HEAP_ARRAY(uint, _dirty_region_buffer);

    }

    if (_in_dirty_region_buffer != NULL) {

      FREE_C_HEAP_ARRAY(IsDirtyRegionState, _in_dirty_region_buffer);

    }

    if (_scan_top != NULL) {

      FREE_C_HEAP_ARRAY(HeapWord*, _scan_top);

    }

  }

  void initialize(uint max_regions) {

    assert(_iter_states == NULL, "Must not be initialized twice");

    assert(_iter_claims == NULL, "Must not be initialized twice");

    _max_regions = max_regions;

    _iter_states = NEW_C_HEAP_ARRAY(G1RemsetIterState, max_regions, mtGC);

    _iter_claims = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC);

    _dirty_region_buffer = NEW_C_HEAP_ARRAY(uint, max_regions, mtGC);

    _in_dirty_region_buffer = NEW_C_HEAP_ARRAY(IsDirtyRegionState, max_regions, mtGC);

    _scan_top = NEW_C_HEAP_ARRAY(HeapWord*, max_regions, mtGC);

  }

  void reset() {

    for (uint i = 0; i < _max_regions; i++) {

      _iter_states[i] = Unclaimed;

    }

    G1ResetScanTopClosure cl(_scan_top);

    G1CollectedHeap::heap()->heap_region_iterate(&cl);

    memset((void*)_iter_claims, 0, _max_regions * sizeof(size_t));

    memset(_in_dirty_region_buffer, Clean, _max_regions * sizeof(IsDirtyRegionState));

    _cur_dirty_region = 0;

  }

  // Attempt to claim the remembered set of the region for iteration. Returns true

  // if this call caused the transition from Unclaimed to Claimed.

  inline bool claim_iter(uint region) {

    assert(region < _max_regions, "Tried to access invalid region %u", region);

    if (_iter_states[region] != Unclaimed) {

      return false;

    }

    jint res = Atomic::cmpxchg(Claimed, (jint*)(&_iter_states[region]), Unclaimed);

    return (res == Unclaimed);

  }

  // Try to atomically sets the iteration state to "complete". Returns true for the

  // thread that caused the transition.

  inline bool set_iter_complete(uint region) {

    if (iter_is_complete(region)) {

      return false;

    }

    jint res = Atomic::cmpxchg(Complete, (jint*)(&_iter_states[region]), Claimed);

    return (res == Claimed);

  }

  // Returns true if the region's iteration is complete.

  inline bool iter_is_complete(uint region) const {

    assert(region < _max_regions, "Tried to access invalid region %u", region);

    return _iter_states[region] == Complete;

  }

  // The current position within the remembered set of the given region.

  inline size_t iter_claimed(uint region) const {

    assert(region < _max_regions, "Tried to access invalid region %u", region);

    return _iter_claims[region];

  }

  // Claim the next block of cards within the remembered set of the region with

  // step size.

  inline size_t iter_claimed_next(uint region, size_t step) {

    return Atomic::add(step, &_iter_claims[region]) - step;

  }

  void add_dirty_region(uint region) {

    if (_in_dirty_region_buffer[region] == Dirty) {

      return;

    }

    bool marked_as_dirty = Atomic::cmpxchg(Dirty, &_in_dirty_region_buffer[region], Clean) == Clean;

    if (marked_as_dirty) {

      size_t allocated = Atomic::add(1, &_cur_dirty_region) - 1;

      _dirty_region_buffer[allocated] = region;

    }

  }

  HeapWord* scan_top(uint region_idx) const {

    return _scan_top[region_idx];

  }

  // Clear the card table of "dirty" regions.

  void clear_card_table(WorkGang* workers) {

    if (_cur_dirty_region == 0) {

      return;

    }

    size_t const num_chunks = align_up(_cur_dirty_region * HeapRegion::CardsPerRegion, G1ClearCardTableTask::chunk_size()) / G1ClearCardTableTask::chunk_size();

    uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers());

    size_t const chunk_length = G1ClearCardTableTask::chunk_size() / HeapRegion::CardsPerRegion;

    // Iterate over the dirty cards region list.

    G1ClearCardTableTask cl(G1CollectedHeap::heap(), _dirty_region_buffer, _cur_dirty_region, chunk_length);

    log_debug(gc, ergo)("Running %s using %u workers for " SIZE_FORMAT " "

                        "units of work for " SIZE_FORMAT " regions.",

                        cl.name(), num_workers, num_chunks, _cur_dirty_region);

    workers->run_task(&cl, num_workers);

#ifndef PRODUCT

    // Need to synchronize with concurrent cleanup since it needs to

    // finish its card table clearing before we can verify.

    G1CollectedHeap::heap()->wait_while_free_regions_coming();

    G1CollectedHeap::heap()->verifier()->verify_card_table_cleanup();

#endif

  }

};

G1RemSet::G1RemSet(G1CollectedHeap* g1,

                   CardTableModRefBS* ct_bs,

                   G1HotCardCache* hot_card_cache) :

  _g1(g1),

  _scan_state(new G1RemSetScanState()),

  _num_conc_refined_cards(0),

  _ct_bs(ct_bs),

  _g1p(_g1->g1_policy()),

  _hot_card_cache(hot_card_cache),

  _prev_period_summary(),

  _into_cset_dirty_card_queue_set(false)

{

  if (log_is_enabled(Trace, gc, remset)) {

    _prev_period_summary.initialize(this);

  }

  // Initialize the card queue set used to hold cards containing

  // references into the collection set.

  _into_cset_dirty_card_queue_set.initialize(NULL, // Should never be called by the Java code

                                             DirtyCardQ_CBL_mon,

                                             DirtyCardQ_FL_lock,

                                             -1, // never trigger processing

                                             -1, // no limit on length

                                             Shared_DirtyCardQ_lock,

                                             &JavaThread::dirty_card_queue_set());

}

G1RemSet::~G1RemSet() {

  if (_scan_state != NULL) {

    delete _scan_state;

  }

}

uint G1RemSet::num_par_rem_sets() {

  return MAX2(DirtyCardQueueSet::num_par_ids() + ConcurrentG1Refine::thread_num(), ParallelGCThreads);

}

void G1RemSet::initialize(size_t capacity, uint max_regions) {

  G1FromCardCache::initialize(num_par_rem_sets(), max_regions);

  _scan_state->initialize(max_regions);

  {

    GCTraceTime(Debug, gc, marking)("Initialize Card Live Data");

    _card_live_data.initialize(capacity, max_regions);

  }

  if (G1PretouchAuxiliaryMemory) {

    GCTraceTime(Debug, gc, marking)("Pre-Touch Card Live Data");

    _card_live_data.pretouch();

  }

}

G1ScanRSForRegionClosure::G1ScanRSForRegionClosure(G1RemSetScanState* scan_state,

                                                   G1ScanObjsDuringScanRSClosure* scan_obj_on_card,

                                                   CodeBlobClosure* code_root_cl,

                                                   uint worker_i) :

  _scan_state(scan_state),

  _scan_objs_on_card_cl(scan_obj_on_card),

  _code_root_cl(code_root_cl),

  _strong_code_root_scan_time_sec(0.0),

  _cards_claimed(0),

  _cards_scanned(0),

  _cards_skipped(0),

  _worker_i(worker_i) {

  _g1h = G1CollectedHeap::heap();

  _bot = _g1h->bot();

  _ct_bs = _g1h->g1_barrier_set();

}

}

void G1ScanRSForRegionClosure::scan_strong_code_roots(HeapRegion* r) {

  double scan_start = os::elapsedTime();

  r->strong_code_roots_do(_code_root_cl);

  _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);

}

bool G1ScanRSForRegionClosure::doHeapRegion(HeapRegion* r) {

  assert(r->in_collection_set(), "should only be called on elements of CS.");

  uint region_idx = r->hrm_index();

  if (_scan_state->iter_is_complete(region_idx)) {

    return false;

  }

  if (_scan_state->claim_iter(region_idx)) {

    // If we ever free the collection set concurrently, we should also

    // clear the card table concurrently therefore we won't need to

    // add regions of the collection set to the dirty cards region.

    _scan_state->add_dirty_region(region_idx);

  }

  // We claim cards in blocks so as to reduce the contention.

  size_t const block_size = G1RSetScanBlockSize;

  HeapRegionRemSetIterator iter(r->rem_set());

  size_t card_index;

  size_t claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);

  for (size_t current_card = 0; iter.has_next(card_index); current_card++) {

    if (current_card >= claimed_card_block + block_size) {

      claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);

    }

    if (current_card < claimed_card_block) {

      _cards_skipped++;

      continue;

    }

    _cards_claimed++;

    }

  }

  if (_scan_state->set_iter_complete(region_idx)) {

    // Scan the strong code root list attached to the current region

    scan_strong_code_roots(r);

  }

  return false;

}

void G1RemSet::scan_rem_set(G1ParScanThreadState* pss,

                            CodeBlobClosure* heap_region_codeblobs,

                            uint worker_i) {

  double rs_time_start = os::elapsedTime();

  G1ScanObjsDuringScanRSClosure scan_cl(_g1, pss);

  G1ScanRSForRegionClosure cl(_scan_state, &scan_cl, heap_region_codeblobs, worker_i);

  _g1->collection_set_iterate_from(&cl, worker_i);

  double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -

                             cl.strong_code_root_scan_time_sec();

  G1GCPhaseTimes* p = _g1p->phase_times();

  p->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);

  p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_scanned(), G1GCPhaseTimes::ScannedCards);

  p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_claimed(), G1GCPhaseTimes::ClaimedCards);

  p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_skipped(), G1GCPhaseTimes::SkippedCards);

  p->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, cl.strong_code_root_scan_time_sec());

}

// Closure used for updating RSets and recording references that

// point into the collection set. Only called during an

// evacuation pause.

class G1RefineCardClosure: public CardTableEntryClosure {

  G1RemSet* _g1rs;

  DirtyCardQueue* _into_cset_dcq;

  G1ScanObjsDuringUpdateRSClosure* _update_rs_cl;

public:

  G1RefineCardClosure(G1CollectedHeap* g1h,

                      DirtyCardQueue* into_cset_dcq,

                      G1ScanObjsDuringUpdateRSClosure* update_rs_cl) :

    _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq), _update_rs_cl(update_rs_cl)

  {}

  bool do_card_ptr(jbyte* card_ptr, uint worker_i) {

    // The only time we care about recording cards that

    // contain references that point into the collection set

    // is during RSet updating within an evacuation pause.

    // In this case worker_i should be the id of a GC worker thread.

    assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");

    if (_g1rs->refine_card_during_gc(card_ptr, _update_rs_cl)) {

      // 'card_ptr' contains references that point into the collection

      // set. We need to record the card in the DCQS

      // (_into_cset_dirty_card_queue_set)

      // that's used for that purpose.

      //

      // Enqueue the card

      _into_cset_dcq->enqueue(card_ptr);

    }

    return true;

  }

};

void G1RemSet::update_rem_set(DirtyCardQueue* into_cset_dcq,

                              G1ParScanThreadState* pss,

                              uint worker_i) {