tschatzl@6937: /*
tschatzl@6937:  * Copyright (c) 2014, Oracle and/or its affiliates. All rights reserved.
tschatzl@6937:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
tschatzl@6937:  *
tschatzl@6937:  * This code is free software; you can redistribute it and/or modify it
tschatzl@6937:  * under the terms of the GNU General Public License version 2 only, as
tschatzl@6937:  * published by the Free Software Foundation.
tschatzl@6937:  *
tschatzl@6937:  * This code is distributed in the hope that it will be useful, but WITHOUT
tschatzl@6937:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
tschatzl@6937:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
tschatzl@6937:  * version 2 for more details (a copy is included in the LICENSE file that
tschatzl@6937:  * accompanied this code).
tschatzl@6937:  *
tschatzl@6937:  * You should have received a copy of the GNU General Public License version
tschatzl@6937:  * 2 along with this work; if not, write to the Free Software Foundation,
tschatzl@6937:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
tschatzl@6937:  *
tschatzl@6937:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
tschatzl@6937:  * or visit www.oracle.com if you need additional information or have any
tschatzl@6937:  * questions.
tschatzl@6937:  *
tschatzl@6937:  */
tschatzl@6937: 
tschatzl@6937: #include "precompiled.hpp"
tschatzl@6937: #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
tschatzl@6937: #include "gc_implementation/g1/g1OopClosures.inline.hpp"
tschatzl@6937: #include "gc_implementation/g1/g1ParScanThreadState.inline.hpp"
tschatzl@6937: #include "oops/oop.inline.hpp"
tschatzl@6937: #include "oops/oop.pcgc.inline.hpp"
tschatzl@6937: #include "runtime/prefetch.inline.hpp"
tschatzl@6937: 
tschatzl@6937: G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num, ReferenceProcessor* rp)
tschatzl@6937:   : _g1h(g1h),
tschatzl@6937:     _refs(g1h->task_queue(queue_num)),
tschatzl@6937:     _dcq(&g1h->dirty_card_queue_set()),
tschatzl@6937:     _ct_bs(g1h->g1_barrier_set()),
tschatzl@6937:     _g1_rem(g1h->g1_rem_set()),
tschatzl@6937:     _hash_seed(17), _queue_num(queue_num),
tschatzl@6937:     _term_attempts(0),
tschatzl@6939:     _age_table(false), _scanner(g1h, rp),
sjohanss@7118:     _strong_roots_time(0), _term_time(0) {
tschatzl@6939:   _scanner.set_par_scan_thread_state(this);
tschatzl@6937:   // we allocate G1YoungSurvRateNumRegions plus one entries, since
tschatzl@6937:   // we "sacrifice" entry 0 to keep track of surviving bytes for
tschatzl@6937:   // non-young regions (where the age is -1)
tschatzl@6937:   // We also add a few elements at the beginning and at the end in
tschatzl@6937:   // an attempt to eliminate cache contention
tschatzl@6937:   uint real_length = 1 + _g1h->g1_policy()->young_cset_region_length();
tschatzl@6937:   uint array_length = PADDING_ELEM_NUM +
tschatzl@6937:                       real_length +
tschatzl@6937:                       PADDING_ELEM_NUM;
tschatzl@6937:   _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
tschatzl@6937:   if (_surviving_young_words_base == NULL)
tschatzl@6937:     vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
tschatzl@6937:                           "Not enough space for young surv histo.");
tschatzl@6937:   _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
tschatzl@6937:   memset(_surviving_young_words, 0, (size_t) real_length * sizeof(size_t));
tschatzl@6937: 
sjohanss@7118:   _g1_par_allocator = G1ParGCAllocator::create_allocator(_g1h);
tschatzl@6937: 
tschatzl@6937:   _start = os::elapsedTime();
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6938: G1ParScanThreadState::~G1ParScanThreadState() {
sjohanss@7118:   _g1_par_allocator->retire_alloc_buffers();
sjohanss@7118:   delete _g1_par_allocator;
tschatzl@6938:   FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base, mtGC);
tschatzl@6938: }
tschatzl@6938: 
tschatzl@6937: void
tschatzl@6937: G1ParScanThreadState::print_termination_stats_hdr(outputStream* const st)
tschatzl@6937: {
tschatzl@6937:   st->print_raw_cr("GC Termination Stats");
tschatzl@6937:   st->print_raw_cr("     elapsed  --strong roots-- -------termination-------"
tschatzl@6937:                    " ------waste (KiB)------");
tschatzl@6937:   st->print_raw_cr("thr     ms        ms      %        ms      %    attempts"
tschatzl@6937:                    "  total   alloc    undo");
tschatzl@6937:   st->print_raw_cr("--- --------- --------- ------ --------- ------ --------"
tschatzl@6937:                    " ------- ------- -------");
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6937: void
tschatzl@6937: G1ParScanThreadState::print_termination_stats(int i,
tschatzl@6937:                                               outputStream* const st) const
tschatzl@6937: {
tschatzl@6937:   const double elapsed_ms = elapsed_time() * 1000.0;
tschatzl@6937:   const double s_roots_ms = strong_roots_time() * 1000.0;
tschatzl@6937:   const double term_ms    = term_time() * 1000.0;
sjohanss@7118:   const size_t alloc_buffer_waste = _g1_par_allocator->alloc_buffer_waste();
sjohanss@7118:   const size_t undo_waste         = _g1_par_allocator->undo_waste();
tschatzl@6937:   st->print_cr("%3d %9.2f %9.2f %6.2f "
tschatzl@6937:                "%9.2f %6.2f " SIZE_FORMAT_W(8) " "
tschatzl@6937:                SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7) " " SIZE_FORMAT_W(7),
tschatzl@6937:                i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms,
tschatzl@6937:                term_ms, term_ms * 100 / elapsed_ms, term_attempts(),
sjohanss@7118:                (alloc_buffer_waste + undo_waste) * HeapWordSize / K,
sjohanss@7118:                alloc_buffer_waste * HeapWordSize / K,
sjohanss@7118:                undo_waste * HeapWordSize / K);
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6937: #ifdef ASSERT
tschatzl@6937: bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
tschatzl@6937:   assert(ref != NULL, "invariant");
tschatzl@6937:   assert(UseCompressedOops, "sanity");
tschatzl@6937:   assert(!has_partial_array_mask(ref), err_msg("ref=" PTR_FORMAT, p2i(ref)));
tschatzl@6937:   oop p = oopDesc::load_decode_heap_oop(ref);
tschatzl@6937:   assert(_g1h->is_in_g1_reserved(p),
tschatzl@6937:          err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
tschatzl@6937:   return true;
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6937: bool G1ParScanThreadState::verify_ref(oop* ref) const {
tschatzl@6937:   assert(ref != NULL, "invariant");
tschatzl@6937:   if (has_partial_array_mask(ref)) {
tschatzl@6937:     // Must be in the collection set--it's already been copied.
tschatzl@6937:     oop p = clear_partial_array_mask(ref);
tschatzl@6937:     assert(_g1h->obj_in_cs(p),
tschatzl@6937:            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
tschatzl@6937:   } else {
tschatzl@6937:     oop p = oopDesc::load_decode_heap_oop(ref);
tschatzl@6937:     assert(_g1h->is_in_g1_reserved(p),
tschatzl@6937:            err_msg("ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)));
tschatzl@6937:   }
tschatzl@6937:   return true;
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6937: bool G1ParScanThreadState::verify_task(StarTask ref) const {
tschatzl@6937:   if (ref.is_narrow()) {
tschatzl@6937:     return verify_ref((narrowOop*) ref);
tschatzl@6937:   } else {
tschatzl@6937:     return verify_ref((oop*) ref);
tschatzl@6937:   }
tschatzl@6937: }
tschatzl@6937: #endif // ASSERT
tschatzl@6937: 
tschatzl@6937: void G1ParScanThreadState::trim_queue() {
tschatzl@6937:   assert(_evac_failure_cl != NULL, "not set");
tschatzl@6937: 
tschatzl@6937:   StarTask ref;
tschatzl@6937:   do {
tschatzl@6937:     // Drain the overflow stack first, so other threads can steal.
tschatzl@6938:     while (_refs->pop_overflow(ref)) {
tschatzl@6938:       dispatch_reference(ref);
tschatzl@6937:     }
tschatzl@6937: 
tschatzl@6938:     while (_refs->pop_local(ref)) {
tschatzl@6938:       dispatch_reference(ref);
tschatzl@6937:     }
tschatzl@6938:   } while (!_refs->is_empty());
tschatzl@6937: }
tschatzl@6937: 
tschatzl@6937: oop G1ParScanThreadState::copy_to_survivor_space(oop const old) {
tschatzl@6937:   size_t word_sz = old->size();
tschatzl@6937:   HeapRegion* from_region = _g1h->heap_region_containing_raw(old);
tschatzl@6937:   // +1 to make the -1 indexes valid...
tschatzl@6937:   int       young_index = from_region->young_index_in_cset()+1;
tschatzl@6937:   assert( (from_region->is_young() && young_index >  0) ||
tschatzl@6937:          (!from_region->is_young() && young_index == 0), "invariant" );
tschatzl@6937:   G1CollectorPolicy* g1p = _g1h->g1_policy();
tschatzl@6937:   markOop m = old->mark();
tschatzl@6937:   int age = m->has_displaced_mark_helper() ? m->displaced_mark_helper()->age()
tschatzl@6937:                                            : m->age();
tschatzl@6937:   GCAllocPurpose alloc_purpose = g1p->evacuation_destination(from_region, age,
tschatzl@6937:                                                              word_sz);
sjohanss@7118:   AllocationContext_t context = from_region->allocation_context();
sjohanss@7118:   HeapWord* obj_ptr = _g1_par_allocator->allocate(alloc_purpose, word_sz, context);
tschatzl@6937: #ifndef PRODUCT
tschatzl@6937:   // Should this evacuation fail?
tschatzl@6937:   if (_g1h->evacuation_should_fail()) {
tschatzl@6937:     if (obj_ptr != NULL) {
sjohanss@7118:       _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
tschatzl@6937:       obj_ptr = NULL;
tschatzl@6937:     }
tschatzl@6937:   }
tschatzl@6937: #endif // !PRODUCT
tschatzl@6937: 
tschatzl@6937:   if (obj_ptr == NULL) {
tschatzl@6937:     // This will either forward-to-self, or detect that someone else has
tschatzl@6937:     // installed a forwarding pointer.
tschatzl@6937:     return _g1h->handle_evacuation_failure_par(this, old);
tschatzl@6937:   }
tschatzl@6937: 
tschatzl@6937:   oop obj = oop(obj_ptr);
tschatzl@6937: 
tschatzl@6937:   // We're going to allocate linearly, so might as well prefetch ahead.
tschatzl@6937:   Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
tschatzl@6937: 
tschatzl@6937:   oop forward_ptr = old->forward_to_atomic(obj);
tschatzl@6937:   if (forward_ptr == NULL) {
tschatzl@6937:     Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
tschatzl@6937: 
tschatzl@6937:     // alloc_purpose is just a hint to allocate() above, recheck the type of region
tschatzl@6937:     // we actually allocated from and update alloc_purpose accordingly
tschatzl@6937:     HeapRegion* to_region = _g1h->heap_region_containing_raw(obj_ptr);
tschatzl@6937:     alloc_purpose = to_region->is_young() ? GCAllocForSurvived : GCAllocForTenured;
tschatzl@6937: 
tschatzl@6937:     if (g1p->track_object_age(alloc_purpose)) {
tschatzl@6937:       // We could simply do obj->incr_age(). However, this causes a
tschatzl@6937:       // performance issue. obj->incr_age() will first check whether
tschatzl@6937:       // the object has a displaced mark by checking its mark word;
tschatzl@6937:       // getting the mark word from the new location of the object
tschatzl@6937:       // stalls. So, given that we already have the mark word and we
tschatzl@6937:       // are about to install it anyway, it's better to increase the
tschatzl@6937:       // age on the mark word, when the object does not have a
tschatzl@6937:       // displaced mark word. We're not expecting many objects to have
tschatzl@6937:       // a displaced marked word, so that case is not optimized
tschatzl@6937:       // further (it could be...) and we simply call obj->incr_age().
tschatzl@6937: 
tschatzl@6937:       if (m->has_displaced_mark_helper()) {
tschatzl@6937:         // in this case, we have to install the mark word first,
tschatzl@6937:         // otherwise obj looks to be forwarded (the old mark word,
tschatzl@6937:         // which contains the forward pointer, was copied)
tschatzl@6937:         obj->set_mark(m);
tschatzl@6937:         obj->incr_age();
tschatzl@6937:       } else {
tschatzl@6937:         m = m->incr_age();
tschatzl@6937:         obj->set_mark(m);
tschatzl@6937:       }
tschatzl@6937:       age_table()->add(obj, word_sz);
tschatzl@6937:     } else {
tschatzl@6937:       obj->set_mark(m);
tschatzl@6937:     }
tschatzl@6937: 
tschatzl@6937:     if (G1StringDedup::is_enabled()) {
tschatzl@6937:       G1StringDedup::enqueue_from_evacuation(from_region->is_young(),
tschatzl@6937:                                              to_region->is_young(),
tschatzl@6937:                                              queue_num(),
tschatzl@6937:                                              obj);
tschatzl@6937:     }
tschatzl@6937: 
tschatzl@6937:     size_t* surv_young_words = surviving_young_words();
tschatzl@6937:     surv_young_words[young_index] += word_sz;
tschatzl@6937: 
tschatzl@6937:     if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
tschatzl@6937:       // We keep track of the next start index in the length field of
tschatzl@6937:       // the to-space object. The actual length can be found in the
tschatzl@6937:       // length field of the from-space object.
tschatzl@6937:       arrayOop(obj)->set_length(0);
tschatzl@6937:       oop* old_p = set_partial_array_mask(old);
tschatzl@6937:       push_on_queue(old_p);
tschatzl@6937:     } else {
tschatzl@6937:       // No point in using the slower heap_region_containing() method,
tschatzl@6937:       // given that we know obj is in the heap.
tschatzl@6937:       _scanner.set_region(_g1h->heap_region_containing_raw(obj));
tschatzl@6937:       obj->oop_iterate_backwards(&_scanner);
tschatzl@6937:     }
tschatzl@6937:   } else {
sjohanss@7118:     _g1_par_allocator->undo_allocation(alloc_purpose, obj_ptr, word_sz, context);
tschatzl@6937:     obj = forward_ptr;
tschatzl@6937:   }
tschatzl@6937:   return obj;
tschatzl@6937: }