8150629: Initializing all ParScanThreadStates causes significant unaccounted "Other" times
Summary: Lazily allocate ParScanThreadStates within the worker threads instead of doing this work upfront serially.
Reviewed-by: mgerdin, jmasa
/*
* Copyright (c) 2014, 2016, Oracle and/or its affiliates. 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 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/g1Allocator.inline.hpp"
#include "gc/g1/g1CollectedHeap.inline.hpp"
#include "gc/g1/g1OopClosures.inline.hpp"
#include "gc/g1/g1ParScanThreadState.inline.hpp"
#include "gc/g1/g1RootClosures.hpp"
#include "gc/g1/g1StringDedup.hpp"
#include "gc/shared/taskqueue.inline.hpp"
#include "oops/oop.inline.hpp"
#include "runtime/prefetch.inline.hpp"
G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint worker_id, size_t young_cset_length)
: _g1h(g1h),
_refs(g1h->task_queue(worker_id)),
_dcq(&g1h->dirty_card_queue_set()),
_ct_bs(g1h->g1_barrier_set()),
_closures(NULL),
_hash_seed(17),
_worker_id(worker_id),
_tenuring_threshold(g1h->g1_policy()->tenuring_threshold()),
_age_table(false),
_scanner(g1h, this),
_old_gen_is_full(false)
{
// we allocate G1YoungSurvRateNumRegions plus one entries, since
// we "sacrifice" entry 0 to keep track of surviving bytes for
// non-young regions (where the age is -1)
// We also add a few elements at the beginning and at the end in
// an attempt to eliminate cache contention
size_t real_length = 1 + young_cset_length;
size_t array_length = PADDING_ELEM_NUM +
real_length +
PADDING_ELEM_NUM;
_surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC);
if (_surviving_young_words_base == NULL)
vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR,
"Not enough space for young surv histo.");
_surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM;
memset(_surviving_young_words, 0, real_length * sizeof(size_t));
_plab_allocator = G1PLABAllocator::create_allocator(_g1h->allocator());
_dest[InCSetState::NotInCSet] = InCSetState::NotInCSet;
// The dest for Young is used when the objects are aged enough to
// need to be moved to the next space.
_dest[InCSetState::Young] = InCSetState::Old;
_dest[InCSetState::Old] = InCSetState::Old;
_closures = G1EvacuationRootClosures::create_root_closures(this, _g1h);
}
// Pass locally gathered statistics to global state.
void G1ParScanThreadState::flush(size_t* surviving_young_words) {
_dcq.flush();
// Update allocation statistics.
_plab_allocator->flush_and_retire_stats();
_g1h->g1_policy()->record_age_table(&_age_table);
uint length = _g1h->g1_policy()->young_cset_region_length();
for (uint region_index = 0; region_index < length; region_index++) {
surviving_young_words[region_index] += _surviving_young_words[region_index];
}
}
G1ParScanThreadState::~G1ParScanThreadState() {
delete _plab_allocator;
delete _closures;
FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base);
}
void G1ParScanThreadState::waste(size_t& wasted, size_t& undo_wasted) {
_plab_allocator->waste(wasted, undo_wasted);
}
#ifdef ASSERT
bool G1ParScanThreadState::verify_ref(narrowOop* ref) const {
assert(ref != NULL, "invariant");
assert(UseCompressedOops, "sanity");
assert(!has_partial_array_mask(ref), "ref=" PTR_FORMAT, p2i(ref));
oop p = oopDesc::load_decode_heap_oop(ref);
assert(_g1h->is_in_g1_reserved(p),
"ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p));
return true;
}
bool G1ParScanThreadState::verify_ref(oop* ref) const {
assert(ref != NULL, "invariant");
if (has_partial_array_mask(ref)) {
// Must be in the collection set--it's already been copied.
oop p = clear_partial_array_mask(ref);
assert(_g1h->obj_in_cs(p),
"ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p));
} else {
oop p = oopDesc::load_decode_heap_oop(ref);
assert(_g1h->is_in_g1_reserved(p),
"ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p));
}
return true;
}
bool G1ParScanThreadState::verify_task(StarTask ref) const {
if (ref.is_narrow()) {
return verify_ref((narrowOop*) ref);
} else {
return verify_ref((oop*) ref);
}
}
#endif // ASSERT
void G1ParScanThreadState::trim_queue() {
StarTask ref;
do {
// Drain the overflow stack first, so other threads can steal.
while (_refs->pop_overflow(ref)) {
dispatch_reference(ref);
}
while (_refs->pop_local(ref)) {
dispatch_reference(ref);
}
} while (!_refs->is_empty());
}
HeapWord* G1ParScanThreadState::allocate_in_next_plab(InCSetState const state,
InCSetState* dest,
size_t word_sz,
AllocationContext_t const context,
bool previous_plab_refill_failed) {
assert(state.is_in_cset_or_humongous(), "Unexpected state: " CSETSTATE_FORMAT, state.value());
assert(dest->is_in_cset_or_humongous(), "Unexpected dest: " CSETSTATE_FORMAT, dest->value());
// Right now we only have two types of regions (young / old) so
// let's keep the logic here simple. We can generalize it when necessary.
if (dest->is_young()) {
bool plab_refill_in_old_failed = false;
HeapWord* const obj_ptr = _plab_allocator->allocate(InCSetState::Old,
word_sz,
context,
&plab_refill_in_old_failed);
// Make sure that we won't attempt to copy any other objects out
// of a survivor region (given that apparently we cannot allocate
// any new ones) to avoid coming into this slow path again and again.
// Only consider failed PLAB refill here: failed inline allocations are
// typically large, so not indicative of remaining space.
if (previous_plab_refill_failed) {
_tenuring_threshold = 0;
}
if (obj_ptr != NULL) {
dest->set_old();
} else {
// We just failed to allocate in old gen. The same idea as explained above
// for making survivor gen unavailable for allocation applies for old gen.
_old_gen_is_full = plab_refill_in_old_failed;
}
return obj_ptr;
} else {
_old_gen_is_full = previous_plab_refill_failed;
assert(dest->is_old(), "Unexpected dest: " CSETSTATE_FORMAT, dest->value());
// no other space to try.
return NULL;
}
}
InCSetState G1ParScanThreadState::next_state(InCSetState const state, markOop const m, uint& age) {
if (state.is_young()) {
age = !m->has_displaced_mark_helper() ? m->age()
: m->displaced_mark_helper()->age();
if (age < _tenuring_threshold) {
return state;
}
}
return dest(state);
}
void G1ParScanThreadState::report_promotion_event(InCSetState const dest_state,
oop const old, size_t word_sz, uint age,
HeapWord * const obj_ptr,
const AllocationContext_t context) const {
G1PLAB* alloc_buf = _plab_allocator->alloc_buffer(dest_state, context);
if (alloc_buf->contains(obj_ptr)) {
_g1h->_gc_tracer_stw->report_promotion_in_new_plab_event(old->klass(), word_sz, age,
dest_state.value() == InCSetState::Old,
alloc_buf->word_sz());
} else {
_g1h->_gc_tracer_stw->report_promotion_outside_plab_event(old->klass(), word_sz, age,
dest_state.value() == InCSetState::Old);
}
}
oop G1ParScanThreadState::copy_to_survivor_space(InCSetState const state,
oop const old,
markOop const old_mark) {
const size_t word_sz = old->size();
HeapRegion* const from_region = _g1h->heap_region_containing(old);
// +1 to make the -1 indexes valid...
const int young_index = from_region->young_index_in_cset()+1;
assert( (from_region->is_young() && young_index > 0) ||
(!from_region->is_young() && young_index == 0), "invariant" );
const AllocationContext_t context = from_region->allocation_context();
uint age = 0;
InCSetState dest_state = next_state(state, old_mark, age);
// The second clause is to prevent premature evacuation failure in case there
// is still space in survivor, but old gen is full.
if (_old_gen_is_full && dest_state.is_old()) {
return handle_evacuation_failure_par(old, old_mark);
}
HeapWord* obj_ptr = _plab_allocator->plab_allocate(dest_state, word_sz, context);
// PLAB allocations should succeed most of the time, so we'll
// normally check against NULL once and that's it.
if (obj_ptr == NULL) {
bool plab_refill_failed = false;
obj_ptr = _plab_allocator->allocate_direct_or_new_plab(dest_state, word_sz, context, &plab_refill_failed);
if (obj_ptr == NULL) {
obj_ptr = allocate_in_next_plab(state, &dest_state, word_sz, context, plab_refill_failed);
if (obj_ptr == NULL) {
// This will either forward-to-self, or detect that someone else has
// installed a forwarding pointer.
return handle_evacuation_failure_par(old, old_mark);
}
}
if (_g1h->_gc_tracer_stw->should_report_promotion_events()) {
// The events are checked individually as part of the actual commit
report_promotion_event(dest_state, old, word_sz, age, obj_ptr, context);
}
}
assert(obj_ptr != NULL, "when we get here, allocation should have succeeded");
assert(_g1h->is_in_reserved(obj_ptr), "Allocated memory should be in the heap");
#ifndef PRODUCT
// Should this evacuation fail?
if (_g1h->evacuation_should_fail()) {
// Doing this after all the allocation attempts also tests the
// undo_allocation() method too.
_plab_allocator->undo_allocation(dest_state, obj_ptr, word_sz, context);
return handle_evacuation_failure_par(old, old_mark);
}
#endif // !PRODUCT
// We're going to allocate linearly, so might as well prefetch ahead.
Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes);
const oop obj = oop(obj_ptr);
const oop forward_ptr = old->forward_to_atomic(obj);
if (forward_ptr == NULL) {
Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz);
if (dest_state.is_young()) {
if (age < markOopDesc::max_age) {
age++;
}
if (old_mark->has_displaced_mark_helper()) {
// In this case, we have to install the mark word first,
// otherwise obj looks to be forwarded (the old mark word,
// which contains the forward pointer, was copied)
obj->set_mark(old_mark);
markOop new_mark = old_mark->displaced_mark_helper()->set_age(age);
old_mark->set_displaced_mark_helper(new_mark);
} else {
obj->set_mark(old_mark->set_age(age));
}
_age_table.add(age, word_sz);
} else {
obj->set_mark(old_mark);
}
if (G1StringDedup::is_enabled()) {
const bool is_from_young = state.is_young();
const bool is_to_young = dest_state.is_young();
assert(is_from_young == _g1h->heap_region_containing(old)->is_young(),
"sanity");
assert(is_to_young == _g1h->heap_region_containing(obj)->is_young(),
"sanity");
G1StringDedup::enqueue_from_evacuation(is_from_young,
is_to_young,
_worker_id,
obj);
}
_surviving_young_words[young_index] += word_sz;
if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) {
// We keep track of the next start index in the length field of
// the to-space object. The actual length can be found in the
// length field of the from-space object.
arrayOop(obj)->set_length(0);
oop* old_p = set_partial_array_mask(old);
push_on_queue(old_p);
} else {
HeapRegion* const to_region = _g1h->heap_region_containing(obj_ptr);
_scanner.set_region(to_region);
obj->oop_iterate_backwards(&_scanner);
}
return obj;
} else {
_plab_allocator->undo_allocation(dest_state, obj_ptr, word_sz, context);
return forward_ptr;
}
}
G1ParScanThreadState* G1ParScanThreadStateSet::state_for_worker(uint worker_id) {
assert(worker_id < _n_workers, "out of bounds access");
if (_states[worker_id] == NULL) {
_states[worker_id] = new_par_scan_state(worker_id, _young_cset_length);
}
return _states[worker_id];
}
void G1ParScanThreadStateSet::add_cards_scanned(uint worker_id, size_t cards_scanned) {
assert(worker_id < _n_workers, "out of bounds access");
_cards_scanned[worker_id] += cards_scanned;
}
size_t G1ParScanThreadStateSet::total_cards_scanned() const {
assert(_flushed, "thread local state from the per thread states should have been flushed");
return _total_cards_scanned;
}
const size_t* G1ParScanThreadStateSet::surviving_young_words() const {
assert(_flushed, "thread local state from the per thread states should have been flushed");
return _surviving_young_words_total;
}
void G1ParScanThreadStateSet::flush() {
assert(!_flushed, "thread local state from the per thread states should be flushed once");
assert(_total_cards_scanned == 0, "should have been cleared");
for (uint worker_index = 0; worker_index < _n_workers; ++worker_index) {
G1ParScanThreadState* pss = _states[worker_index];
if (pss == NULL) {
continue;
}
_total_cards_scanned += _cards_scanned[worker_index];
pss->flush(_surviving_young_words_total);
delete pss;
_states[worker_index] = NULL;
}
_flushed = true;
}
oop G1ParScanThreadState::handle_evacuation_failure_par(oop old, markOop m) {
assert(_g1h->obj_in_cs(old), "Object " PTR_FORMAT " should be in the CSet", p2i(old));
oop forward_ptr = old->forward_to_atomic(old);
if (forward_ptr == NULL) {
// Forward-to-self succeeded. We are the "owner" of the object.
HeapRegion* r = _g1h->heap_region_containing(old);
if (!r->evacuation_failed()) {
r->set_evacuation_failed(true);
_g1h->hr_printer()->evac_failure(r);
}
_g1h->preserve_mark_during_evac_failure(_worker_id, old, m);
_scanner.set_region(r);
old->oop_iterate_backwards(&_scanner);
return old;
} else {
// Forward-to-self failed. Either someone else managed to allocate
// space for this object (old != forward_ptr) or they beat us in
// self-forwarding it (old == forward_ptr).
assert(old == forward_ptr || !_g1h->obj_in_cs(forward_ptr),
"Object " PTR_FORMAT " forwarded to: " PTR_FORMAT " "
"should not be in the CSet",
p2i(old), p2i(forward_ptr));
return forward_ptr;
}
}