rev 6626 : imported patch 8035400-move-g1parscanthreadstate-into-own-files
rev 6627 : imported patch 8035400-2-bengt-fixes
rev 6628 : imported patch 8035401-fix-visibility-of-g1parscanthreadstate
rev 6629 : imported patch 8035401-3-fix-constructor
rev 6630 : imported patch 8035401-2-another-inline-try
rev 6631 : imported patch 8040977-g1-crashes-with-G1DeferredRSUpdate-disabled

   1 /*
   2  * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
  27 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
  28 #include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
  29 #include "oops/oop.inline.hpp"
  30 #include "oops/oop.pcgc.inline.hpp"
  31 #include "runtime/prefetch.inline.hpp"
  32 
  33 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
  34 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list
  35 #endif // _MSC_VER
  36 
  37 G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
  38   : _g1h(g1h),
  39     _refs(g1h->task_queue(queue_num)),
  40     _dcq(&g1h->dirty_card_queue_set()),
  41     _ct_bs(g1h->g1_barrier_set()),
  42     _g1_rem(g1h->g1_rem_set()),
  43     _hash_seed(17), _queue_num(queue_num),
  44     _term_attempts(0),
  45     _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)),
  46     _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)),
  47     _age_table(false), _scanner(g1h, this, rp),
  48     _strong_roots_time(0), _term_time(0),
  49     _alloc_buffer_waste(0), _undo_waste(0) {

  50   // we allocate G1YoungSurvRateNumRegions plus one entries, since
  51   // we "sacrifice" entry 0 to keep track of surviving bytes for
  52   // non-young regions (where the age is -1)
  53   // We also add a few elements at the beginning and at the end in
  54   // an attempt to eliminate cache contention
  55   uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
  56   uint array_length = PADDING_ELEM_NUM +
  57                       real_length +
  58                       PADDING_ELEM_NUM;
  59   _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
  60   if (_surviving_young_words_base == NULL)
  61     vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
  62                           "Not enough space for young surv histo.");
  63   _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
  64   memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
  65 
  66   _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer;
  67   _alloc_buffers[GCAllocForTenured]  = &_tenured_alloc_buffer;
  68 
  69   _start = os::elapsedTime();
  70 }
  71 
  72 G1ParScanThreadState::~G1ParScanThreadState() {
  73   retire_alloc_buffers();
  74   FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
  75 }
  76 
  77 void
  78 G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
  79 {
  80   st->print_raw_cr("GC Termination Stats");
  81   st->print_raw_cr("     elapsed  --strong roots-- -------termination-------"
  82                    " ------waste (KiB)------");
  83   st->print_raw_cr("thr     ms        ms      %        ms      %    attempts"
  84                    "  total   alloc    undo");
  85   st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
  86                    " ------- ------- -------");
  87 }
  88 
  89 void
  90 G1ParScanThreadState::print_termination_stats(int i,
  91                                               outputStream* const st) const
  92 {
  93   const double elapsed_ms = elapsed_time() * 1000.0;
  94   const double s_roots_ms = strong_roots_time() * 1000.0;
  95   const double term_ms    = term_time() * 1000.0;
  96   st->print_cr("%3d %9.2f %9.2f %6.2f "
  97                "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
  98                SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
  99                i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
 100                term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
 101                (alloc_buffer_waste() + undo_waste()) * HeapWordSize / K,
 102                alloc_buffer_waste() * HeapWordSize / K,
 103                undo_waste() * HeapWordSize / K);
 104 }
 105 
 106 #ifdef ASSERT
 107 bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
 108   assert(ref != NULL, "invariant");
 109   assert(UseCompressedOops, "sanity");
 110   assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, p2i(ref)));
 111   oop p = oopDesc::load_decode_heap_oop(ref);
 112   assert(_g1h->is_in_g1_reserved(p),
 113          err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 114   return true;
 115 }
 116 
 117 bool G1ParScanThreadState::verify_ref(oop* ref) const {
 118   assert(ref != NULL, "invariant");
 119   if (has_partial_array_mask(ref)) {
 120     // Must be in the collection set--it's already been copied.
 121     oop p = clear_partial_array_mask(ref);
 122     assert(_g1h->obj_in_cs(p),
 123            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 124   } else {
 125     oop p = oopDesc::load_decode_heap_oop(ref);
 126     assert(_g1h->is_in_g1_reserved(p),
 127            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 128   }
 129   return true;
 130 }
 131 
 132 bool G1ParScanThreadState::verify_task(StarTask ref) const {
 133   if (ref.is_narrow()) {
 134     return verify_ref((narrowOop*) ref);
 135   } else {
 136     return verify_ref((oop*) ref);
 137   }
 138 }
 139 #endif // ASSERT
 140 
 141 void G1ParScanThreadState::trim_queue() {
 142   assert(_evac_failure_cl != NULL, "not set");
 143 
 144   StarTask ref;
 145   do {
 146     // Drain the overflow stack first, so other threads can steal.
 147     while (_refs->pop_overflow(ref)) {
 148       dispatch_reference(ref);
 149     }
 150 
 151     while (_refs->pop_local(ref)) {
 152       dispatch_reference(ref);
 153     }
 154   } while (!_refs->is_empty());
 155 }
 156 
 157 oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
 158   size_t word_sz = old->size();
 159   HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
 160   // +1 to make the -1 indexes valid...
 161   int       young_index = from_region->young_index_in_cset()+1;
 162   assert( (from_region->is_young() && young_index >  0) ||
 163          (!from_region->is_young() && young_index == 0), "invariant" );
 164   G1CollectorPolicy* g1p = _g1h->g1_policy();
 165   markOop m = old->mark();
 166   int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
 167                                            : m->age();
 168   GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
 169                                                              word_sz);
 170   HeapWord* obj_ptr = allocate(alloc_purpose, word_sz);
 171 #ifndef PRODUCT
 172   // Should this evacuation fail?
 173   if (_g1h->evacuation_should_fail()) {
 174     if (obj_ptr != NULL) {
 175       undo_allocation(alloc_purpose, obj_ptr, word_sz);
 176       obj_ptr = NULL;
 177     }
 178   }
 179 #endif // !PRODUCT
 180 
 181   if (obj_ptr == NULL) {
 182     // This will either forward-to-self, or detect that someone else has
 183     // installed a forwarding pointer.
 184     return _g1h->handle_evacuation_failure_par(this, old);
 185   }
 186 
 187   oop obj = oop(obj_ptr);
 188 
 189   // We're going to allocate linearly, so might as well prefetch ahead.
 190   Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
 191 
 192   oop forward_ptr = old->forward_to_atomic(obj);
 193   if (forward_ptr == NULL) {
 194     Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
 195 
 196     // alloc_purpose is just a hint to allocate() above, recheck the type of region
 197     // we actually allocated from and update alloc_purpose accordingly
 198     HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
 199     alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
 200 
 201     if (g1p->track_object_age(alloc_purpose)) {
 202       // We could simply do obj->incr_age(). However, this causes a
 203       // performance issue. obj->incr_age() will first check whether
 204       // the object has a displaced mark by checking its mark word;
 205       // getting the mark word from the new location of the object
 206       // stalls. So, given that we already have the mark word and we
 207       // are about to install it anyway, it's better to increase the
 208       // age on the mark word, when the object does not have a
 209       // displaced mark word. We're not expecting many objects to have
 210       // a displaced marked word, so that case is not optimized
 211       // further (it could be...) and we simply call obj->incr_age().
 212 
 213       if (m->has_displaced_mark_helper()) {
 214         // in this case, we have to install the mark word first,
 215         // otherwise obj looks to be forwarded (the old mark word,
 216         // which contains the forward pointer, was copied)
 217         obj->set_mark(m);
 218         obj->incr_age();
 219       } else {
 220         m = m->incr_age();
 221         obj->set_mark(m);
 222       }
 223       age_table()->add(obj, word_sz);
 224     } else {
 225       obj->set_mark(m);
 226     }
 227 
 228     if (G1StringDedup::is_enabled()) {
 229       G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
 230                                              to_region->is_young(),
 231                                              queue_num(),
 232                                              obj);
 233     }
 234 
 235     size_t* surv_young_words = surviving_young_words();
 236     surv_young_words[young_index] += word_sz;
 237 
 238     if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
 239       // We keep track of the next start index in the length field of
 240       // the to-space object. The actual length can be found in the
 241       // length field of the from-space object.
 242       arrayOop(obj)->set_length(0);
 243       oop* old_p = set_partial_array_mask(old);
 244       push_on_queue(old_p);
 245     } else {
 246       // No point in using the slower heap_region_containing() method,
 247       // given that we know obj is in the heap.
 248       _scanner.set_region(_g1h->heap_region_containing_raw(obj));
 249       obj->oop_iterate_backwards(&_scanner);
 250     }
 251   } else {
 252     undo_allocation(alloc_purpose, obj_ptr, word_sz);
 253     obj = forward_ptr;
 254   }
 255   return obj;
 256 }
 257 
 258 HeapWord* G1ParScanThreadState::allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
 259   HeapWord* obj = NULL;
 260   size_t gclab_word_size = _g1h->desired_plab_sz(purpose);
 261   if (word_sz * 100 < gclab_word_size * ParallelGCBufferWastePct) {
 262     G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
 263     add_to_alloc_buffer_waste(alloc_buf->words_remaining());
 264     alloc_buf->retire(false /* end_of_gc */, false /* retain */);
 265 
 266     HeapWord* buf = _g1h->par_allocate_during_gc(purpose, gclab_word_size);
 267     if (buf == NULL) {
 268       return NULL; // Let caller handle allocation failure.
 269     }
 270     // Otherwise.
 271     alloc_buf->set_word_size(gclab_word_size);
 272     alloc_buf->set_buf(buf);
 273 
 274     obj = alloc_buf->allocate(word_sz);
 275     assert(obj != NULL, "buffer was definitely big enough...");
 276   } else {
 277     obj = _g1h->par_allocate_during_gc(purpose, word_sz);
 278   }
 279   return obj;
 280 }
 281 
 282 void G1ParScanThreadState::undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
 283   if (alloc_buffer(purpose)->contains(obj)) {
 284     assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
 285            "should contain whole object");
 286     alloc_buffer(purpose)->undo_allocation(obj, word_sz);
 287   } else {
 288     CollectedHeap::fill_with_object(obj, word_sz);
 289     add_to_undo_waste(word_sz);
 290   }
 291 }
 292 
 293 HeapWord* G1ParScanThreadState::allocate(GCAllocPurpose purpose, size_t word_sz) {
 294   HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
 295   if (obj != NULL) {
 296     return obj;
 297   }
 298   return allocate_slow(purpose, word_sz);
 299 }
 300 
 301 void G1ParScanThreadState::retire_alloc_buffers() {
 302   for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
 303     size_t waste = _alloc_buffers[ap]->words_remaining();
 304     add_to_alloc_buffer_waste(waste);
 305     _alloc_buffers[ap]->flush_stats_and_retire(_g1h->stats_for_purpose((GCAllocPurpose)ap),
 306                                                true /* end_of_gc */,
 307                                                false /* retain */);
 308   }
 309 }
--- EOF ---