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 G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
  34   : _g1h(g1h),
  35     _refs(g1h->task_queue(queue_num)),
  36     _dcq(&g1h->dirty_card_queue_set()),
  37     _ct_bs(g1h->g1_barrier_set()),
  38     _g1_rem(g1h->g1_rem_set()),
  39     _hash_seed(17), _queue_num(queue_num),
  40     _term_attempts(0),
  41     _surviving_alloc_buffer(g1h->desired_plab_sz(GCAllocForSurvived)),
  42     _tenured_alloc_buffer(g1h->desired_plab_sz(GCAllocForTenured)),
  43     _age_table(false), _scanner(g1h, rp),
  44     _strong_roots_time(0), _term_time(0),
  45     _alloc_buffer_waste(0), _undo_waste(0) {
  46   _scanner.set_par_scan_thread_state(this);
  47   // we allocate G1YoungSurvRateNumRegions plus one entries, since
  48   // we "sacrifice" entry 0 to keep track of surviving bytes for
  49   // non-young regions (where the age is -1)
  50   // We also add a few elements at the beginning and at the end in
  51   // an attempt to eliminate cache contention
  52   uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
  53   uint array_length = PADDING_ELEM_NUM +
  54                       real_length +
  55                       PADDING_ELEM_NUM;
  56   _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
  57   if (_surviving_young_words_base == NULL)
  58     vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
  59                           "Not enough space for young surv histo.");
  60   _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
  61   memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
  62 
  63   _alloc_buffers[GCAllocForSurvived] = &_surviving_alloc_buffer;
  64   _alloc_buffers[GCAllocForTenured]  = &_tenured_alloc_buffer;
  65 
  66   _start = os::elapsedTime();
  67 }
  68 
  69 G1ParScanThreadState::~G1ParScanThreadState() {
  70   retire_alloc_buffers();
  71   FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
  72 }
  73 
  74 void
  75 G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
  76 {
  77   st->print_raw_cr("GC Termination Stats");
  78   st->print_raw_cr("     elapsed  --strong roots-- -------termination-------"
  79                    " ------waste (KiB)------");
  80   st->print_raw_cr("thr     ms        ms      %        ms      %    attempts"
  81                    "  total   alloc    undo");
  82   st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
  83                    " ------- ------- -------");
  84 }
  85 
  86 void
  87 G1ParScanThreadState::print_termination_stats(int i,
  88                                               outputStream* const st) const
  89 {
  90   const double elapsed_ms = elapsed_time() * 1000.0;
  91   const double s_roots_ms = strong_roots_time() * 1000.0;
  92   const double term_ms    = term_time() * 1000.0;
  93   st->print_cr("%3d %9.2f %9.2f %6.2f "
  94                "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
  95                SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
  96                i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
  97                term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
  98                (alloc_buffer_waste() + undo_waste()) * HeapWordSize / K,
  99                alloc_buffer_waste() * HeapWordSize / K,
 100                undo_waste() * HeapWordSize / K);
 101 }
 102 
 103 #ifdef ASSERT
 104 bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
 105   assert(ref != NULL, "invariant");
 106   assert(UseCompressedOops, "sanity");
 107   assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, p2i(ref)));
 108   oop p = oopDesc::load_decode_heap_oop(ref);
 109   assert(_g1h->is_in_g1_reserved(p),
 110          err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 111   return true;
 112 }
 113 
 114 bool G1ParScanThreadState::verify_ref(oop* ref) const {
 115   assert(ref != NULL, "invariant");
 116   if (has_partial_array_mask(ref)) {
 117     // Must be in the collection set--it's already been copied.
 118     oop p = clear_partial_array_mask(ref);
 119     assert(_g1h->obj_in_cs(p),
 120            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 121   } else {
 122     oop p = oopDesc::load_decode_heap_oop(ref);
 123     assert(_g1h->is_in_g1_reserved(p),
 124            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
 125   }
 126   return true;
 127 }
 128 
 129 bool G1ParScanThreadState::verify_task(StarTask ref) const {
 130   if (ref.is_narrow()) {
 131     return verify_ref((narrowOop*) ref);
 132   } else {
 133     return verify_ref((oop*) ref);
 134   }
 135 }
 136 #endif // ASSERT
 137 
 138 void G1ParScanThreadState::trim_queue() {
 139   assert(_evac_failure_cl != NULL, "not set");
 140 
 141   StarTask ref;
 142   do {
 143     // Drain the overflow stack first, so other threads can steal.
 144     while (_refs->pop_overflow(ref)) {
 145       dispatch_reference(ref);
 146     }
 147 
 148     while (_refs->pop_local(ref)) {
 149       dispatch_reference(ref);
 150     }
 151   } while (!_refs->is_empty());
 152 }
 153 
 154 oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
 155   size_t word_sz = old->size();
 156   HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
 157   // +1 to make the -1 indexes valid...
 158   int       young_index = from_region->young_index_in_cset()+1;
 159   assert( (from_region->is_young() && young_index >  0) ||
 160          (!from_region->is_young() && young_index == 0), "invariant" );
 161   G1CollectorPolicy* g1p = _g1h->g1_policy();
 162   markOop m = old->mark();
 163   int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
 164                                            : m->age();
 165   GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
 166                                                              word_sz);
 167   HeapWord* obj_ptr = allocate(alloc_purpose, word_sz);
 168 #ifndef PRODUCT
 169   // Should this evacuation fail?
 170   if (_g1h->evacuation_should_fail()) {
 171     if (obj_ptr != NULL) {
 172       undo_allocation(alloc_purpose, obj_ptr, word_sz);
 173       obj_ptr = NULL;
 174     }
 175   }
 176 #endif // !PRODUCT
 177 
 178   if (obj_ptr == NULL) {
 179     // This will either forward-to-self, or detect that someone else has
 180     // installed a forwarding pointer.
 181     return _g1h->handle_evacuation_failure_par(this, old);
 182   }
 183 
 184   oop obj = oop(obj_ptr);
 185 
 186   // We're going to allocate linearly, so might as well prefetch ahead.
 187   Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
 188 
 189   oop forward_ptr = old->forward_to_atomic(obj);
 190   if (forward_ptr == NULL) {
 191     Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
 192 
 193     // alloc_purpose is just a hint to allocate() above, recheck the type of region
 194     // we actually allocated from and update alloc_purpose accordingly
 195     HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
 196     alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
 197 
 198     if (g1p->track_object_age(alloc_purpose)) {
 199       // We could simply do obj->incr_age(). However, this causes a
 200       // performance issue. obj->incr_age() will first check whether
 201       // the object has a displaced mark by checking its mark word;
 202       // getting the mark word from the new location of the object
 203       // stalls. So, given that we already have the mark word and we
 204       // are about to install it anyway, it's better to increase the
 205       // age on the mark word, when the object does not have a
 206       // displaced mark word. We're not expecting many objects to have
 207       // a displaced marked word, so that case is not optimized
 208       // further (it could be...) and we simply call obj->incr_age().
 209 
 210       if (m->has_displaced_mark_helper()) {
 211         // in this case, we have to install the mark word first,
 212         // otherwise obj looks to be forwarded (the old mark word,
 213         // which contains the forward pointer, was copied)
 214         obj->set_mark(m);
 215         obj->incr_age();
 216       } else {
 217         m = m->incr_age();
 218         obj->set_mark(m);
 219       }
 220       age_table()->add(obj, word_sz);
 221     } else {
 222       obj->set_mark(m);
 223     }
 224 
 225     if (G1StringDedup::is_enabled()) {
 226       G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
 227                                              to_region->is_young(),
 228                                              queue_num(),
 229                                              obj);
 230     }
 231 
 232     size_t* surv_young_words = surviving_young_words();
 233     surv_young_words[young_index] += word_sz;
 234 
 235     if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
 236       // We keep track of the next start index in the length field of
 237       // the to-space object. The actual length can be found in the
 238       // length field of the from-space object.
 239       arrayOop(obj)->set_length(0);
 240       oop* old_p = set_partial_array_mask(old);
 241       push_on_queue(old_p);
 242     } else {
 243       // No point in using the slower heap_region_containing() method,
 244       // given that we know obj is in the heap.
 245       _scanner.set_region(_g1h->heap_region_containing_raw(obj));
 246       obj->oop_iterate_backwards(&_scanner);
 247     }
 248   } else {
 249     undo_allocation(alloc_purpose, obj_ptr, word_sz);
 250     obj = forward_ptr;
 251   }
 252   return obj;
 253 }
 254 
 255 HeapWord* G1ParScanThreadState::allocate_slow(GCAllocPurpose purpose, size_t word_sz) {
 256   HeapWord* obj = NULL;
 257   size_t gclab_word_size = _g1h->desired_plab_sz(purpose);
 258   if (word_sz * 100 < gclab_word_size * ParallelGCBufferWastePct) {
 259     G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
 260     add_to_alloc_buffer_waste(alloc_buf->words_remaining());
 261     alloc_buf->retire(false /* end_of_gc */, false /* retain */);
 262 
 263     HeapWord* buf = _g1h->par_allocate_during_gc(purpose, gclab_word_size);
 264     if (buf == NULL) {
 265       return NULL; // Let caller handle allocation failure.
 266     }
 267     // Otherwise.
 268     alloc_buf->set_word_size(gclab_word_size);
 269     alloc_buf->set_buf(buf);
 270 
 271     obj = alloc_buf->allocate(word_sz);
 272     assert(obj != NULL, "buffer was definitely big enough...");
 273   } else {
 274     obj = _g1h->par_allocate_during_gc(purpose, word_sz);
 275   }
 276   return obj;
 277 }
 278 
 279 void G1ParScanThreadState::undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
 280   if (alloc_buffer(purpose)->contains(obj)) {
 281     assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
 282            "should contain whole object");
 283     alloc_buffer(purpose)->undo_allocation(obj, word_sz);
 284   } else {
 285     CollectedHeap::fill_with_object(obj, word_sz);
 286     add_to_undo_waste(word_sz);
 287   }
 288 }
 289 
 290 HeapWord* G1ParScanThreadState::allocate(GCAllocPurpose purpose, size_t word_sz) {
 291   HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
 292   if (obj != NULL) {
 293     return obj;
 294   }
 295   return allocate_slow(purpose, word_sz);
 296 }
 297 
 298 void G1ParScanThreadState::retire_alloc_buffers() {
 299   for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
 300     size_t waste = _alloc_buffers[ap]->words_remaining();
 301     add_to_alloc_buffer_waste(waste);
 302     _alloc_buffers[ap]->flush_stats_and_retire(_g1h->stats_for_purpose((GCAllocPurpose)ap),
 303                                                true /* end_of_gc */,
 304                                                false /* retain */);
 305   }
 306 }