/*

 * Copyright 2001-2009 Sun Microsystems, Inc.  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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,

 * CA 95054 USA or visit www.sun.com if you need additional information or

 * have any questions.

 *

 */

#include "incls/_precompiled.incl"

#include "incls/_g1RemSet.cpp.incl"

#define CARD_REPEAT_HISTO 0

#if CARD_REPEAT_HISTO

static size_t ct_freq_sz;

static jbyte* ct_freq = NULL;

void init_ct_freq_table(size_t heap_sz_bytes) {

  if (ct_freq == NULL) {

    ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;

    ct_freq = new jbyte[ct_freq_sz];

    for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;

  }

}

void ct_freq_note_card(size_t index) {

  assert(0 <= index && index < ct_freq_sz, "Bounds error.");

  if (ct_freq[index] < 100) { ct_freq[index]++; }

}

static IntHistogram card_repeat_count(10, 10);

void ct_freq_update_histo_and_reset() {

  for (size_t j = 0; j < ct_freq_sz; j++) {

    card_repeat_count.add_entry(ct_freq[j]);

    ct_freq[j] = 0;

  }

}

#endif

class IntoCSOopClosure: public OopsInHeapRegionClosure {

  OopsInHeapRegionClosure* _blk;

  G1CollectedHeap* _g1;

public:

  IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :

    _g1(g1), _blk(blk) {}

  void set_region(HeapRegion* from) {

    _blk->set_region(from);

  }

  virtual void do_oop(narrowOop* p) { do_oop_work(p); }

  virtual void do_oop(      oop* p) { do_oop_work(p); }

  template <class T> void do_oop_work(T* p) {

    oop obj = oopDesc::load_decode_heap_oop(p);

    if (_g1->obj_in_cs(obj)) _blk->do_oop(p);

  }

  bool apply_to_weak_ref_discovered_field() { return true; }

  bool idempotent() { return true; }

};

class IntoCSRegionClosure: public HeapRegionClosure {

  IntoCSOopClosure _blk;

  G1CollectedHeap* _g1;

public:

  IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :

    _g1(g1), _blk(g1, blk) {}

  bool doHeapRegion(HeapRegion* r) {

    if (!r->in_collection_set()) {

      _blk.set_region(r);

      if (r->isHumongous()) {

        if (r->startsHumongous()) {

          oop obj = oop(r->bottom());

          obj->oop_iterate(&_blk);

        }

      } else {

        r->oop_before_save_marks_iterate(&_blk);

      }

    }

    return false;

  }

};

void

StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,

                                            int worker_i) {

  IntoCSRegionClosure rc(_g1, oc);

  _g1->heap_region_iterate(&rc);

}

class VerifyRSCleanCardOopClosure: public OopClosure {

  G1CollectedHeap* _g1;

public:

  VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}

  virtual void do_oop(narrowOop* p) { do_oop_work(p); }

  virtual void do_oop(      oop* p) { do_oop_work(p); }

  template <class T> void do_oop_work(T* p) {

    oop obj = oopDesc::load_decode_heap_oop(p);

    HeapRegion* to = _g1->heap_region_containing(obj);

    guarantee(to == NULL || !to->in_collection_set(),

              "Missed a rem set member.");

  }

};

HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)

  : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()),

    _cg1r(g1->concurrent_g1_refine()),

    _par_traversal_in_progress(false), _new_refs(NULL),

    _cards_scanned(NULL), _total_cards_scanned(0)

{

  _seq_task = new SubTasksDone(NumSeqTasks);

  guarantee(n_workers() > 0, "There should be some workers");

  _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, n_workers());

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

    _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<OopOrNarrowOopStar>(8192,true);

  }

}

HRInto_G1RemSet::~HRInto_G1RemSet() {

  delete _seq_task;

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

    delete _new_refs[i];

  }

  FREE_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, _new_refs);

}

void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {

  if (_g1->is_in_g1_reserved(mr.start())) {

    _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));

    if (_start_first == NULL) _start_first = mr.start();

  }

}

class ScanRSClosure : public HeapRegionClosure {

  size_t _cards_done, _cards;

  G1CollectedHeap* _g1h;

  OopsInHeapRegionClosure* _oc;

  G1BlockOffsetSharedArray* _bot_shared;

  CardTableModRefBS *_ct_bs;

  int _worker_i;

  bool _try_claimed;

  size_t _min_skip_distance, _max_skip_distance;

public:

  ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :

    _oc(oc),

    _cards(0),

    _cards_done(0),

    _worker_i(worker_i),

    _try_claimed(false)

  {

    _g1h = G1CollectedHeap::heap();

    _bot_shared = _g1h->bot_shared();

    _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());

    _min_skip_distance = 16;

    _max_skip_distance = 2 * _g1h->n_par_threads() * _min_skip_distance;

  }

  void set_try_claimed() { _try_claimed = true; }

  void scanCard(size_t index, HeapRegion *r) {

    _cards_done++;

    DirtyCardToOopClosure* cl =

      r->new_dcto_closure(_oc,

                         CardTableModRefBS::Precise,

                         HeapRegionDCTOC::IntoCSFilterKind);

    // Set the "from" region in the closure.

    _oc->set_region(r);

    HeapWord* card_start = _bot_shared->address_for_index(index);

    HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;

    Space *sp = SharedHeap::heap()->space_containing(card_start);

    MemRegion sm_region;

    if (ParallelGCThreads > 0) {

      // first find the used area

      sm_region = sp->used_region_at_save_marks();

    } else {

      // The closure is not idempotent.  We shouldn't look at objects

      // allocated during the GC.

      sm_region = sp->used_region_at_save_marks();

    }

    MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));

    if (!mr.is_empty()) {

      cl->do_MemRegion(mr);

    }

  }

  void printCard(HeapRegion* card_region, size_t card_index,

                 HeapWord* card_start) {

    gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "

                           "RS names card %p: "

                           "[" PTR_FORMAT ", " PTR_FORMAT ")",

                           _worker_i,

                           card_region->bottom(), card_region->end(),

                           card_index,

                           card_start, card_start + G1BlockOffsetSharedArray::N_words);

  }

  bool doHeapRegion(HeapRegion* r) {

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

    HeapRegionRemSet* hrrs = r->rem_set();

    if (hrrs->iter_is_complete()) return false; // All done.

    if (!_try_claimed && !hrrs->claim_iter()) return false;

    _g1h->push_dirty_cards_region(r);

    // If we didn't return above, then

    //   _try_claimed || r->claim_iter()

    // is true: either we're supposed to work on claimed-but-not-complete

    // regions, or we successfully claimed the region.

    HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);

    hrrs->init_iterator(iter);

    size_t card_index;

    size_t skip_distance = 0, current_card = 0, jump_to_card = 0;

    while (iter->has_next(card_index)) {

      if (current_card < jump_to_card) {

        ++current_card;

        continue;

      }

      HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);

#if 0

      gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",

                          card_start, card_start + CardTableModRefBS::card_size_in_words);

#endif

      HeapRegion* card_region = _g1h->heap_region_containing(card_start);

      assert(card_region != NULL, "Yielding cards not in the heap?");

      _cards++;

      if (!card_region->is_on_dirty_cards_region_list()) {

        _g1h->push_dirty_cards_region(card_region);

      }

       // If the card is dirty, then we will scan it during updateRS.

      if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {

          if (!_ct_bs->is_card_claimed(card_index) && _ct_bs->claim_card(card_index)) {

            scanCard(card_index, card_region);

          } else if (_try_claimed) {

            if (jump_to_card == 0 || jump_to_card != current_card) {

              // We did some useful work in the previous iteration.

              // Decrease the distance.

              skip_distance = MAX2(skip_distance >> 1, _min_skip_distance);

            } else {

              // Previous iteration resulted in a claim failure.

              // Increase the distance.

              skip_distance = MIN2(skip_distance << 1, _max_skip_distance);

            }

            jump_to_card = current_card + skip_distance;

          }

      }

      ++current_card;

    }

    if (!_try_claimed) {

      hrrs->set_iter_complete();

    }

    return false;

  }

  // Set all cards back to clean.

  void cleanup() {_g1h->cleanUpCardTable();}

  size_t cards_done() { return _cards_done;}

  size_t cards_looked_up() { return _cards;}

};

// We want the parallel threads to start their scanning at

// different collection set regions to avoid contention.

// If we have:

//          n collection set regions

//          p threads

// Then thread t will start at region t * floor (n/p)

HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) {

  HeapRegion* result = _g1p->collection_set();

  if (ParallelGCThreads > 0) {

    size_t cs_size = _g1p->collection_set_size();

    int n_workers = _g1->workers()->total_workers();

    size_t cs_spans = cs_size / n_workers;

    size_t ind      = cs_spans * worker_i;

    for (size_t i = 0; i < ind; i++)

      result = result->next_in_collection_set();

  }

  return result;

}

void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {

  double rs_time_start = os::elapsedTime();

  HeapRegion *startRegion = calculateStartRegion(worker_i);

  BufferingOopsInHeapRegionClosure boc(oc);

  ScanRSClosure scanRScl(&boc, worker_i);

  _g1->collection_set_iterate_from(startRegion, &scanRScl);

  scanRScl.set_try_claimed();

  _g1->collection_set_iterate_from(startRegion, &scanRScl);

  boc.done();

  double closure_app_time_sec = boc.closure_app_seconds();

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

    closure_app_time_sec;

  double closure_app_time_ms = closure_app_time_sec * 1000.0;

  assert( _cards_scanned != NULL, "invariant" );

  _cards_scanned[worker_i] = scanRScl.cards_done();

  _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0);

  _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);

  double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i);

  if (scan_new_refs_time_ms > 0.0) {

    closure_app_time_ms += scan_new_refs_time_ms;

  }

  _g1p->record_obj_copy_time(worker_i, closure_app_time_ms);

}

void HRInto_G1RemSet::updateRS(int worker_i) {

  ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();

  double start = os::elapsedTime();

  _g1p->record_update_rs_start_time(worker_i, start * 1000.0);

  // Apply the appropriate closure to all remaining log entries.

  _g1->iterate_dirty_card_closure(false, worker_i);

  // Now there should be no dirty cards.

  if (G1RSLogCheckCardTable) {

    CountNonCleanMemRegionClosure cl(_g1);

    _ct_bs->mod_card_iterate(&cl);

    // XXX This isn't true any more: keeping cards of young regions

    // marked dirty broke it.  Need some reasonable fix.

    guarantee(cl.n() == 0, "Card table should be clean.");

  }

  _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);

}

#ifndef PRODUCT

class PrintRSClosure : public HeapRegionClosure {

  int _count;

public:

  PrintRSClosure() : _count(0) {}

  bool doHeapRegion(HeapRegion* r) {

    HeapRegionRemSet* hrrs = r->rem_set();

    _count += (int) hrrs->occupied();

    if (hrrs->occupied() == 0) {

      gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "

                          "has no remset entries\n",

                          r->bottom(), r->end());

    } else {

      gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",

                          r->bottom(), r->end());

      r->print();

      hrrs->print();

      gclog_or_tty->print("\nDone printing rem set\n");

    }

    return false;

  }

  int occupied() {return _count;}

};

#endif

class CountRSSizeClosure: public HeapRegionClosure {

  size_t _n;

  size_t _tot;

  size_t _max;

  HeapRegion* _max_r;

  enum {

    N = 20,

    MIN = 6

  };

  int _histo[N];

public:

  CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {

    for (int i = 0; i < N; i++) _histo[i] = 0;

  }

  bool doHeapRegion(HeapRegion* r) {

    if (!r->continuesHumongous()) {

      size_t occ = r->rem_set()->occupied();

      _n++;

      _tot += occ;

      if (occ > _max) {

        _max = occ;

        _max_r = r;

      }

      // Fit it into a histo bin.

      int s = 1 << MIN;

      int i = 0;

      while (occ > (size_t) s && i < (N-1)) {

        s = s << 1;

        i++;

      }

      _histo[i]++;

    }

    return false;

  }

  size_t n() { return _n; }

  size_t tot() { return _tot; }

  size_t mx() { return _max; }

  HeapRegion* mxr() { return _max_r; }

  void print_histo() {

    int mx = N;

    while (mx >= 0) {

      if (_histo[mx-1] > 0) break;

      mx--;

    }

    gclog_or_tty->print_cr("Number of regions with given RS sizes:");

    gclog_or_tty->print_cr("           <= %8d   %8d", 1 << MIN, _histo[0]);

    for (int i = 1; i < mx-1; i++) {

      gclog_or_tty->print_cr("  %8d  - %8d   %8d",

                    (1 << (MIN + i - 1)) + 1,

                    1 << (MIN + i),

                    _histo[i]);

    }

    gclog_or_tty->print_cr("            > %8d   %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);

  }

};

template <class T> void

HRInto_G1RemSet::scanNewRefsRS_work(OopsInHeapRegionClosure* oc,

                                    int worker_i) {

  double scan_new_refs_start_sec = os::elapsedTime();

  G1CollectedHeap* g1h = G1CollectedHeap::heap();

  CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());

  for (int i = 0; i < _new_refs[worker_i]->length(); i++) {

    T* p = (T*) _new_refs[worker_i]->at(i);

    oop obj = oopDesc::load_decode_heap_oop(p);

    // *p was in the collection set when p was pushed on "_new_refs", but

    // another thread may have processed this location from an RS, so it

    // might not point into the CS any longer.  If so, it's obviously been

    // processed, and we don't need to do anything further.

    if (g1h->obj_in_cs(obj)) {

      HeapRegion* r = g1h->heap_region_containing(p);

      DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj));

      oc->set_region(r);

      // If "p" has already been processed concurrently, this is

      // idempotent.

      oc->do_oop(p);

    }

  }

  _g1p->record_scan_new_refs_time(worker_i,

                                  (os::elapsedTime() - scan_new_refs_start_sec)

                                  * 1000.0);

}

void HRInto_G1RemSet::cleanupHRRS() {

  HeapRegionRemSet::cleanup();

}

void

HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,

                                             int worker_i) {

#if CARD_REPEAT_HISTO

  ct_freq_update_histo_and_reset();

#endif

  if (worker_i == 0) {

    _cg1r->clear_and_record_card_counts();

  }

  // Make this into a command-line flag...

  if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {

    CountRSSizeClosure count_cl;

    _g1->heap_region_iterate(&count_cl);

    gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "

                  "max region is " PTR_FORMAT,

                  count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),

                  count_cl.mx(), count_cl.mxr());

    count_cl.print_histo();

  }

  if (ParallelGCThreads > 0) {

    // The two flags below were introduced temporarily to serialize

    // the updating and scanning of remembered sets. There are some

    // race conditions when these two operations are done in parallel

    // and they are causing failures. When we resolve said race

    // conditions, we'll revert back to parallel remembered set

    // updating and scanning. See CRs 6677707 and 6677708.

    if (G1ParallelRSetUpdatingEnabled || (worker_i == 0)) {

      updateRS(worker_i);

      scanNewRefsRS(oc, worker_i);

    } else {

      _g1p->record_update_rs_start_time(worker_i, os::elapsedTime() * 1000.0);

      _g1p->record_update_rs_processed_buffers(worker_i, 0.0);

      _g1p->record_update_rs_time(worker_i, 0.0);

      _g1p->record_scan_new_refs_time(worker_i, 0.0);

    }

    if (G1ParallelRSetScanningEnabled || (worker_i == 0)) {

      scanRS(oc, worker_i);

    } else {

      _g1p->record_scan_rs_start_time(worker_i, os::elapsedTime() * 1000.0);

      _g1p->record_scan_rs_time(worker_i, 0.0);

    }

  } else {

    assert(worker_i == 0, "invariant");

    updateRS(0);

    scanNewRefsRS(oc, 0);

    scanRS(oc, 0);

  }

}

void HRInto_G1RemSet::

prepare_for_oops_into_collection_set_do() {

#if G1_REM_SET_LOGGING

  PrintRSClosure cl;

  _g1->collection_set_iterate(&cl);

#endif

  cleanupHRRS();