--- /dev/null 2018-11-30 10:10:44.238550338 +0100 +++ new/src/hotspot/share/gc/shenandoah/shenandoahBarrierSet.cpp 2018-11-30 21:08:03.512420312 +0100 @@ -0,0 +1,376 @@ +/* + * Copyright (c) 2013, 2018, Red Hat, Inc. All rights reserved. + * + * 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/g1BarrierSet.hpp" +#include "gc/shenandoah/shenandoahAsserts.hpp" +#include "gc/shenandoah/shenandoahBarrierSet.hpp" +#include "gc/shenandoah/shenandoahBarrierSetAssembler.hpp" +#include "gc/shenandoah/shenandoahCollectorPolicy.hpp" +#include "gc/shenandoah/shenandoahHeap.inline.hpp" +#include "gc/shenandoah/shenandoahHeuristics.hpp" +#include "gc/shenandoah/shenandoahTraversalGC.hpp" +#include "memory/iterator.inline.hpp" +#include "runtime/interfaceSupport.inline.hpp" +#ifdef COMPILER1 +#include "gc/shenandoah/c1/shenandoahBarrierSetC1.hpp" +#endif +#ifdef COMPILER2 +#include "gc/shenandoah/c2/shenandoahBarrierSetC2.hpp" +#endif + +class ShenandoahBarrierSetC1; +class ShenandoahBarrierSetC2; + +template +class ShenandoahUpdateRefsForOopClosure: public BasicOopIterateClosure { +private: + ShenandoahHeap* _heap; + ShenandoahBarrierSet* _bs; + + template + inline void do_oop_work(T* p) { + oop o; + if (STOREVAL_WRITE_BARRIER) { + o = _heap->evac_update_with_forwarded(p); + if (!CompressedOops::is_null(o)) { + _bs->enqueue(o); + } + } else { + _heap->maybe_update_with_forwarded(p); + } + } +public: + ShenandoahUpdateRefsForOopClosure() : _heap(ShenandoahHeap::heap()), _bs(ShenandoahBarrierSet::barrier_set()) { + assert(UseShenandoahGC && ShenandoahCloneBarrier, "should be enabled"); + } + + virtual void do_oop(oop* p) { do_oop_work(p); } + virtual void do_oop(narrowOop* p) { do_oop_work(p); } +}; + +ShenandoahBarrierSet::ShenandoahBarrierSet(ShenandoahHeap* heap) : + BarrierSet(make_barrier_set_assembler(), + make_barrier_set_c1(), + make_barrier_set_c2(), + NULL /* barrier_set_nmethod */, + BarrierSet::FakeRtti(BarrierSet::ShenandoahBarrierSet)), + _heap(heap), + _satb_mark_queue_set() +{ +} + +ShenandoahBarrierSetAssembler* ShenandoahBarrierSet::assembler() { + BarrierSetAssembler* const bsa = BarrierSet::barrier_set()->barrier_set_assembler(); + return reinterpret_cast(bsa); +} + +void ShenandoahBarrierSet::print_on(outputStream* st) const { + st->print("ShenandoahBarrierSet"); +} + +bool ShenandoahBarrierSet::is_a(BarrierSet::Name bsn) { + return bsn == BarrierSet::ShenandoahBarrierSet; +} + +bool ShenandoahBarrierSet::is_aligned(HeapWord* hw) { + return true; +} + +template +void ShenandoahBarrierSet::write_ref_array_loop(HeapWord* start, size_t count) { + assert(UseShenandoahGC && ShenandoahCloneBarrier, "should be enabled"); + ShenandoahUpdateRefsForOopClosure cl; + T* dst = (T*) start; + for (size_t i = 0; i < count; i++) { + cl.do_oop(dst++); + } +} + +void ShenandoahBarrierSet::write_ref_array(HeapWord* start, size_t count) { + assert(UseShenandoahGC, "should be enabled"); + if (count == 0) return; + if (!ShenandoahCloneBarrier) return; + + if (!need_update_refs_barrier()) return; + + if (_heap->is_concurrent_traversal_in_progress()) { + ShenandoahEvacOOMScope oom_evac_scope; + if (UseCompressedOops) { + write_ref_array_loop(start, count); + } else { + write_ref_array_loop(start, count); + } + } else { + if (UseCompressedOops) { + write_ref_array_loop(start, count); + } else { + write_ref_array_loop(start, count); + } + } +} + +template +void ShenandoahBarrierSet::write_ref_array_pre_work(T* dst, size_t count) { + shenandoah_assert_not_in_cset_loc_except(dst, _heap->cancelled_gc()); + if (ShenandoahSATBBarrier && _heap->is_concurrent_mark_in_progress()) { + T* elem_ptr = dst; + for (size_t i = 0; i < count; i++, elem_ptr++) { + T heap_oop = RawAccess<>::oop_load(elem_ptr); + if (!CompressedOops::is_null(heap_oop)) { + enqueue(CompressedOops::decode_not_null(heap_oop)); + } + } + } +} + +void ShenandoahBarrierSet::write_ref_array_pre(oop* dst, size_t count, bool dest_uninitialized) { + if (! dest_uninitialized) { + write_ref_array_pre_work(dst, count); + } +} + +void ShenandoahBarrierSet::write_ref_array_pre(narrowOop* dst, size_t count, bool dest_uninitialized) { + if (! dest_uninitialized) { + write_ref_array_pre_work(dst, count); + } +} + +template +inline void ShenandoahBarrierSet::inline_write_ref_field_pre(T* field, oop new_val) { + shenandoah_assert_not_in_cset_loc_except(field, _heap->cancelled_gc()); + if (_heap->is_concurrent_mark_in_progress()) { + T heap_oop = RawAccess<>::oop_load(field); + if (!CompressedOops::is_null(heap_oop)) { + enqueue(CompressedOops::decode(heap_oop)); + } + } +} + +// These are the more general virtual versions. +void ShenandoahBarrierSet::write_ref_field_pre_work(oop* field, oop new_val) { + inline_write_ref_field_pre(field, new_val); +} + +void ShenandoahBarrierSet::write_ref_field_pre_work(narrowOop* field, oop new_val) { + inline_write_ref_field_pre(field, new_val); +} + +void ShenandoahBarrierSet::write_ref_field_pre_work(void* field, oop new_val) { + guarantee(false, "Not needed"); +} + +void ShenandoahBarrierSet::write_ref_field_work(void* v, oop o, bool release) { + shenandoah_assert_not_in_cset_loc_except(v, _heap->cancelled_gc()); + shenandoah_assert_not_forwarded_except (v, o, o == NULL || _heap->cancelled_gc() || !_heap->is_concurrent_mark_in_progress()); + shenandoah_assert_not_in_cset_except (v, o, o == NULL || _heap->cancelled_gc() || !_heap->is_concurrent_mark_in_progress()); +} + +void ShenandoahBarrierSet::write_region(MemRegion mr) { + assert(UseShenandoahGC, "should be enabled"); + if (!ShenandoahCloneBarrier) return; + if (! need_update_refs_barrier()) return; + + // This is called for cloning an object (see jvm.cpp) after the clone + // has been made. We are not interested in any 'previous value' because + // it would be NULL in any case. But we *are* interested in any oop* + // that potentially need to be updated. + + oop obj = oop(mr.start()); + shenandoah_assert_correct(NULL, obj); + if (_heap->is_concurrent_traversal_in_progress()) { + ShenandoahEvacOOMScope oom_evac_scope; + ShenandoahUpdateRefsForOopClosure cl; + obj->oop_iterate(&cl); + } else { + ShenandoahUpdateRefsForOopClosure cl; + obj->oop_iterate(&cl); + } +} + +oop ShenandoahBarrierSet::read_barrier(oop src) { + // Check for forwarded objects, because on Full GC path we might deal with + // non-trivial fwdptrs that contain Full GC specific metadata. We could check + // for is_full_gc_in_progress(), but this also covers the case of stable heap, + // which provides a bit of performance improvement. + if (ShenandoahReadBarrier && _heap->has_forwarded_objects()) { + return ShenandoahBarrierSet::resolve_forwarded(src); + } else { + return src; + } +} + +bool ShenandoahBarrierSet::obj_equals(oop obj1, oop obj2) { + bool eq = oopDesc::equals_raw(obj1, obj2); + if (! eq && ShenandoahAcmpBarrier) { + OrderAccess::loadload(); + obj1 = resolve_forwarded(obj1); + obj2 = resolve_forwarded(obj2); + eq = oopDesc::equals_raw(obj1, obj2); + } + return eq; +} + +oop ShenandoahBarrierSet::write_barrier_mutator(oop obj) { + assert(UseShenandoahGC && ShenandoahWriteBarrier, "should be enabled"); + assert(_heap->is_gc_in_progress_mask(ShenandoahHeap::EVACUATION | ShenandoahHeap::TRAVERSAL), "evac should be in progress"); + shenandoah_assert_in_cset(NULL, obj); + + oop fwd = resolve_forwarded_not_null(obj); + if (oopDesc::equals_raw(obj, fwd)) { + ShenandoahEvacOOMScope oom_evac_scope; + + Thread* thread = Thread::current(); + oop res_oop = _heap->evacuate_object(obj, thread); + + // Since we are already here and paid the price of getting through runtime call adapters + // and acquiring oom-scope, it makes sense to try and evacuate more adjacent objects, + // thus amortizing the overhead. For sparsely live heaps, scan costs easily dominate + // total assist costs, and can introduce a lot of evacuation latency. This is why we + // only scan for _nearest_ N objects, regardless if they are eligible for evac or not. + // The scan itself should also avoid touching the non-marked objects below TAMS, because + // their metadata (notably, klasses) may be incorrect already. + + size_t max = ShenandoahEvacAssist; + if (max > 0) { + // Traversal is special: it uses incomplete marking context, because it coalesces evac with mark. + // Other code uses complete marking context, because evac happens after the mark. + ShenandoahMarkingContext* ctx = _heap->is_concurrent_traversal_in_progress() ? + _heap->marking_context() : _heap->complete_marking_context(); + + ShenandoahHeapRegion* r = _heap->heap_region_containing(obj); + assert(r->is_cset(), "sanity"); + + HeapWord* cur = (HeapWord*)obj + obj->size() + ShenandoahBrooksPointer::word_size(); + + size_t count = 0; + while ((cur < r->top()) && ctx->is_marked(oop(cur)) && (count++ < max)) { + oop cur_oop = oop(cur); + if (oopDesc::equals_raw(cur_oop, resolve_forwarded_not_null(cur_oop))) { + _heap->evacuate_object(cur_oop, thread); + } + cur = cur + cur_oop->size() + ShenandoahBrooksPointer::word_size(); + } + } + + return res_oop; + } + return fwd; +} + +oop ShenandoahBarrierSet::write_barrier_impl(oop obj) { + assert(UseShenandoahGC && ShenandoahWriteBarrier, "should be enabled"); + if (!CompressedOops::is_null(obj)) { + bool evac_in_progress = _heap->is_gc_in_progress_mask(ShenandoahHeap::EVACUATION | ShenandoahHeap::TRAVERSAL); + oop fwd = resolve_forwarded_not_null(obj); + if (evac_in_progress && + _heap->in_collection_set(obj) && + oopDesc::equals_raw(obj, fwd)) { + Thread *t = Thread::current(); + if (t->is_GC_task_thread()) { + return _heap->evacuate_object(obj, t); + } else { + ShenandoahEvacOOMScope oom_evac_scope; + return _heap->evacuate_object(obj, t); + } + } else { + return fwd; + } + } else { + return obj; + } +} + +oop ShenandoahBarrierSet::write_barrier(oop obj) { + if (ShenandoahWriteBarrier && _heap->has_forwarded_objects()) { + return write_barrier_impl(obj); + } else { + return obj; + } +} + +oop ShenandoahBarrierSet::storeval_barrier(oop obj) { + if (ShenandoahStoreValEnqueueBarrier) { + if (!CompressedOops::is_null(obj)) { + obj = write_barrier(obj); + enqueue(obj); + } + } + if (ShenandoahStoreValReadBarrier) { + obj = resolve_forwarded(obj); + } + return obj; +} + +void ShenandoahBarrierSet::keep_alive_barrier(oop obj) { + if (ShenandoahKeepAliveBarrier && _heap->is_concurrent_mark_in_progress()) { + enqueue(obj); + } +} + +void ShenandoahBarrierSet::enqueue(oop obj) { + shenandoah_assert_not_forwarded_if(NULL, obj, _heap->is_concurrent_traversal_in_progress()); + if (!_satb_mark_queue_set.is_active()) return; + + // Filter marked objects before hitting the SATB queues. The same predicate would + // be used by SATBMQ::filter to eliminate already marked objects downstream, but + // filtering here helps to avoid wasteful SATB queueing work to begin with. + if (!_heap->requires_marking(obj)) return; + + Thread* thr = Thread::current(); + if (thr->is_Java_thread()) { + ShenandoahThreadLocalData::satb_mark_queue(thr).enqueue(obj); + } else { + MutexLockerEx x(Shared_SATB_Q_lock, Mutex::_no_safepoint_check_flag); + _satb_mark_queue_set.shared_satb_queue()->enqueue(obj); + } +} + +void ShenandoahBarrierSet::on_thread_create(Thread* thread) { + // Create thread local data + ShenandoahThreadLocalData::create(thread); +} + +void ShenandoahBarrierSet::on_thread_destroy(Thread* thread) { + // Destroy thread local data + ShenandoahThreadLocalData::destroy(thread); +} + +void ShenandoahBarrierSet::on_thread_attach(JavaThread* thread) { + assert(!SafepointSynchronize::is_at_safepoint(), "We should not be at a safepoint"); + assert(!ShenandoahThreadLocalData::satb_mark_queue(thread).is_active(), "SATB queue should not be active"); + assert(ShenandoahThreadLocalData::satb_mark_queue(thread).is_empty(), "SATB queue should be empty"); + if (ShenandoahBarrierSet::satb_mark_queue_set().is_active()) { + ShenandoahThreadLocalData::satb_mark_queue(thread).set_active(true); + } + ShenandoahThreadLocalData::set_gc_state(thread, _heap->gc_state()); + ShenandoahThreadLocalData::initialize_gclab(thread); +} + +void ShenandoahBarrierSet::on_thread_detach(JavaThread* thread) { + ShenandoahThreadLocalData::satb_mark_queue(thread).flush(); + PLAB* gclab = ShenandoahThreadLocalData::gclab(thread); + if (gclab != NULL) { + gclab->retire(); + } +}