| 1 | /* |
| 2 | * Copyright (c) 2014, 2019, 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/g1/g1Allocator.inline.hpp" |
| 27 | #include "gc/g1/g1CollectedHeap.inline.hpp" |
| 28 | #include "gc/g1/g1CollectionSet.hpp" |
| 29 | #include "gc/g1/g1OopClosures.inline.hpp" |
| 30 | #include "gc/g1/g1ParScanThreadState.inline.hpp" |
| 31 | #include "gc/g1/g1RootClosures.hpp" |
| 32 | #include "gc/g1/g1StringDedup.hpp" |
| 33 | #include "gc/shared/gcTrace.hpp" |
| 34 | #include "gc/shared/taskqueue.inline.hpp" |
| 35 | #include "memory/allocation.inline.hpp" |
| 36 | #include "oops/access.inline.hpp" |
| 37 | #include "oops/oop.inline.hpp" |
| 38 | #include "runtime/prefetch.inline.hpp" |
| 39 | |
| 40 | G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, |
| 41 | uint worker_id, |
| 42 | size_t young_cset_length, |
| 43 | size_t optional_cset_length) |
| 44 | : _g1h(g1h), |
| 45 | _refs(g1h->task_queue(worker_id)), |
| 46 | _dcq(&g1h->dirty_card_queue_set()), |
| 47 | _ct(g1h->card_table()), |
| 48 | _closures(NULL), |
| 49 | _plab_allocator(NULL), |
| 50 | _age_table(false), |
| 51 | _tenuring_threshold(g1h->policy()->tenuring_threshold()), |
| 52 | _scanner(g1h, this), |
| 53 | _worker_id(worker_id), |
| 54 | _stack_trim_upper_threshold(GCDrainStackTargetSize * 2 + 1), |
| 55 | _stack_trim_lower_threshold(GCDrainStackTargetSize), |
| 56 | _trim_ticks(), |
| 57 | _old_gen_is_full(false), |
| 58 | _num_optional_regions(optional_cset_length) |
| 59 | { |
| 60 | // we allocate G1YoungSurvRateNumRegions plus one entries, since |
| 61 | // we "sacrifice" entry 0 to keep track of surviving bytes for |
| 62 | // non-young regions (where the age is -1) |
| 63 | // We also add a few elements at the beginning and at the end in |
| 64 | // an attempt to eliminate cache contention |
| 65 | size_t real_length = 1 + young_cset_length; |
| 66 | size_t array_length = PADDING_ELEM_NUM + |
| 67 | real_length + |
| 68 | PADDING_ELEM_NUM; |
| 69 | _surviving_young_words_base = NEW_C_HEAP_ARRAY(size_t, array_length, mtGC); |
| 70 | if (_surviving_young_words_base == NULL) |
| 71 | vm_exit_out_of_memory(array_length * sizeof(size_t), OOM_MALLOC_ERROR, |
| 72 | "Not enough space for young surv histo." ); |
| 73 | _surviving_young_words = _surviving_young_words_base + PADDING_ELEM_NUM; |
| 74 | memset(_surviving_young_words, 0, real_length * sizeof(size_t)); |
| 75 | |
| 76 | _plab_allocator = new G1PLABAllocator(_g1h->allocator()); |
| 77 | |
| 78 | // The dest for Young is used when the objects are aged enough to |
| 79 | // need to be moved to the next space. |
| 80 | _dest[G1HeapRegionAttr::Young] = G1HeapRegionAttr::Old; |
| 81 | _dest[G1HeapRegionAttr::Old] = G1HeapRegionAttr::Old; |
| 82 | |
| 83 | _closures = G1EvacuationRootClosures::create_root_closures(this, _g1h); |
| 84 | |
| 85 | _oops_into_optional_regions = new G1OopStarChunkedList[_num_optional_regions]; |
| 86 | } |
| 87 | |
| 88 | // Pass locally gathered statistics to global state. |
| 89 | void G1ParScanThreadState::flush(size_t* surviving_young_words) { |
| 90 | _dcq.flush(); |
| 91 | // Update allocation statistics. |
| 92 | _plab_allocator->flush_and_retire_stats(); |
| 93 | _g1h->policy()->record_age_table(&_age_table); |
| 94 | |
| 95 | uint length = _g1h->collection_set()->young_region_length(); |
| 96 | for (uint region_index = 0; region_index < length; region_index++) { |
| 97 | surviving_young_words[region_index] += _surviving_young_words[region_index]; |
| 98 | } |
| 99 | } |
| 100 | |
| 101 | G1ParScanThreadState::~G1ParScanThreadState() { |
| 102 | delete _plab_allocator; |
| 103 | delete _closures; |
| 104 | FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base); |
| 105 | delete[] _oops_into_optional_regions; |
| 106 | } |
| 107 | |
| 108 | size_t G1ParScanThreadState::lab_waste_words() const { |
| 109 | return _plab_allocator->waste(); |
| 110 | } |
| 111 | |
| 112 | size_t G1ParScanThreadState::lab_undo_waste_words() const { |
| 113 | return _plab_allocator->undo_waste(); |
| 114 | } |
| 115 | |
| 116 | #ifdef ASSERT |
| 117 | bool G1ParScanThreadState::verify_ref(narrowOop* ref) const { |
| 118 | assert(ref != NULL, "invariant" ); |
| 119 | assert(UseCompressedOops, "sanity" ); |
| 120 | assert(!has_partial_array_mask(ref), "ref=" PTR_FORMAT, p2i(ref)); |
| 121 | oop p = RawAccess<>::oop_load(ref); |
| 122 | assert(_g1h->is_in_g1_reserved(p), |
| 123 | "ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)); |
| 124 | return true; |
| 125 | } |
| 126 | |
| 127 | bool G1ParScanThreadState::verify_ref(oop* ref) const { |
| 128 | assert(ref != NULL, "invariant" ); |
| 129 | if (has_partial_array_mask(ref)) { |
| 130 | // Must be in the collection set--it's already been copied. |
| 131 | oop p = clear_partial_array_mask(ref); |
| 132 | assert(_g1h->is_in_cset(p), |
| 133 | "ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)); |
| 134 | } else { |
| 135 | oop p = RawAccess<>::oop_load(ref); |
| 136 | assert(_g1h->is_in_g1_reserved(p), |
| 137 | "ref=" PTR_FORMAT " p=" PTR_FORMAT, p2i(ref), p2i(p)); |
| 138 | } |
| 139 | return true; |
| 140 | } |
| 141 | |
| 142 | bool G1ParScanThreadState::verify_task(StarTask ref) const { |
| 143 | if (ref.is_narrow()) { |
| 144 | return verify_ref((narrowOop*) ref); |
| 145 | } else { |
| 146 | return verify_ref((oop*) ref); |
| 147 | } |
| 148 | } |
| 149 | #endif // ASSERT |
| 150 | |
| 151 | void G1ParScanThreadState::trim_queue() { |
| 152 | StarTask ref; |
| 153 | do { |
| 154 | // Fully drain the queue. |
| 155 | trim_queue_to_threshold(0); |
| 156 | } while (!_refs->is_empty()); |
| 157 | } |
| 158 | |
| 159 | HeapWord* G1ParScanThreadState::allocate_in_next_plab(G1HeapRegionAttr const region_attr, |
| 160 | G1HeapRegionAttr* dest, |
| 161 | size_t word_sz, |
| 162 | bool previous_plab_refill_failed) { |
| 163 | assert(region_attr.is_in_cset_or_humongous(), "Unexpected region attr type: %s" , region_attr.get_type_str()); |
| 164 | assert(dest->is_in_cset_or_humongous(), "Unexpected dest: %s region attr" , dest->get_type_str()); |
| 165 | |
| 166 | // Right now we only have two types of regions (young / old) so |
| 167 | // let's keep the logic here simple. We can generalize it when necessary. |
| 168 | if (dest->is_young()) { |
| 169 | bool plab_refill_in_old_failed = false; |
| 170 | HeapWord* const obj_ptr = _plab_allocator->allocate(G1HeapRegionAttr::Old, |
| 171 | word_sz, |
| 172 | &plab_refill_in_old_failed); |
| 173 | // Make sure that we won't attempt to copy any other objects out |
| 174 | // of a survivor region (given that apparently we cannot allocate |
| 175 | // any new ones) to avoid coming into this slow path again and again. |
| 176 | // Only consider failed PLAB refill here: failed inline allocations are |
| 177 | // typically large, so not indicative of remaining space. |
| 178 | if (previous_plab_refill_failed) { |
| 179 | _tenuring_threshold = 0; |
| 180 | } |
| 181 | |
| 182 | if (obj_ptr != NULL) { |
| 183 | dest->set_old(); |
| 184 | } else { |
| 185 | // We just failed to allocate in old gen. The same idea as explained above |
| 186 | // for making survivor gen unavailable for allocation applies for old gen. |
| 187 | _old_gen_is_full = plab_refill_in_old_failed; |
| 188 | } |
| 189 | return obj_ptr; |
| 190 | } else { |
| 191 | _old_gen_is_full = previous_plab_refill_failed; |
| 192 | assert(dest->is_old(), "Unexpected dest region attr: %s" , dest->get_type_str()); |
| 193 | // no other space to try. |
| 194 | return NULL; |
| 195 | } |
| 196 | } |
| 197 | |
| 198 | G1HeapRegionAttr G1ParScanThreadState::next_region_attr(G1HeapRegionAttr const region_attr, markOop const m, uint& age) { |
| 199 | if (region_attr.is_young()) { |
| 200 | age = !m->has_displaced_mark_helper() ? m->age() |
| 201 | : m->displaced_mark_helper()->age(); |
| 202 | if (age < _tenuring_threshold) { |
| 203 | return region_attr; |
| 204 | } |
| 205 | } |
| 206 | return dest(region_attr); |
| 207 | } |
| 208 | |
| 209 | void G1ParScanThreadState::report_promotion_event(G1HeapRegionAttr const dest_attr, |
| 210 | oop const old, size_t word_sz, uint age, |
| 211 | HeapWord * const obj_ptr) const { |
| 212 | PLAB* alloc_buf = _plab_allocator->alloc_buffer(dest_attr); |
| 213 | if (alloc_buf->contains(obj_ptr)) { |
| 214 | _g1h->_gc_tracer_stw->report_promotion_in_new_plab_event(old->klass(), word_sz * HeapWordSize, age, |
| 215 | dest_attr.type() == G1HeapRegionAttr::Old, |
| 216 | alloc_buf->word_sz() * HeapWordSize); |
| 217 | } else { |
| 218 | _g1h->_gc_tracer_stw->report_promotion_outside_plab_event(old->klass(), word_sz * HeapWordSize, age, |
| 219 | dest_attr.type() == G1HeapRegionAttr::Old); |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | oop G1ParScanThreadState::copy_to_survivor_space(G1HeapRegionAttr const region_attr, |
| 224 | oop const old, |
| 225 | markOop const old_mark) { |
| 226 | const size_t word_sz = old->size(); |
| 227 | HeapRegion* const from_region = _g1h->heap_region_containing(old); |
| 228 | // +1 to make the -1 indexes valid... |
| 229 | const int young_index = from_region->young_index_in_cset()+1; |
| 230 | assert( (from_region->is_young() && young_index > 0) || |
| 231 | (!from_region->is_young() && young_index == 0), "invariant" ); |
| 232 | |
| 233 | uint age = 0; |
| 234 | G1HeapRegionAttr dest_attr = next_region_attr(region_attr, old_mark, age); |
| 235 | // The second clause is to prevent premature evacuation failure in case there |
| 236 | // is still space in survivor, but old gen is full. |
| 237 | if (_old_gen_is_full && dest_attr.is_old()) { |
| 238 | return handle_evacuation_failure_par(old, old_mark); |
| 239 | } |
| 240 | HeapWord* obj_ptr = _plab_allocator->plab_allocate(dest_attr, word_sz); |
| 241 | |
| 242 | // PLAB allocations should succeed most of the time, so we'll |
| 243 | // normally check against NULL once and that's it. |
| 244 | if (obj_ptr == NULL) { |
| 245 | bool plab_refill_failed = false; |
| 246 | obj_ptr = _plab_allocator->allocate_direct_or_new_plab(dest_attr, word_sz, &plab_refill_failed); |
| 247 | if (obj_ptr == NULL) { |
| 248 | obj_ptr = allocate_in_next_plab(region_attr, &dest_attr, word_sz, plab_refill_failed); |
| 249 | if (obj_ptr == NULL) { |
| 250 | // This will either forward-to-self, or detect that someone else has |
| 251 | // installed a forwarding pointer. |
| 252 | return handle_evacuation_failure_par(old, old_mark); |
| 253 | } |
| 254 | } |
| 255 | if (_g1h->_gc_tracer_stw->should_report_promotion_events()) { |
| 256 | // The events are checked individually as part of the actual commit |
| 257 | report_promotion_event(dest_attr, old, word_sz, age, obj_ptr); |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | assert(obj_ptr != NULL, "when we get here, allocation should have succeeded" ); |
| 262 | assert(_g1h->is_in_reserved(obj_ptr), "Allocated memory should be in the heap" ); |
| 263 | |
| 264 | #ifndef PRODUCT |
| 265 | // Should this evacuation fail? |
| 266 | if (_g1h->evacuation_should_fail()) { |
| 267 | // Doing this after all the allocation attempts also tests the |
| 268 | // undo_allocation() method too. |
| 269 | _plab_allocator->undo_allocation(dest_attr, obj_ptr, word_sz); |
| 270 | return handle_evacuation_failure_par(old, old_mark); |
| 271 | } |
| 272 | #endif // !PRODUCT |
| 273 | |
| 274 | // We're going to allocate linearly, so might as well prefetch ahead. |
| 275 | Prefetch::write(obj_ptr, PrefetchCopyIntervalInBytes); |
| 276 | |
| 277 | const oop obj = oop(obj_ptr); |
| 278 | const oop forward_ptr = old->forward_to_atomic(obj, old_mark, memory_order_relaxed); |
| 279 | if (forward_ptr == NULL) { |
| 280 | Copy::aligned_disjoint_words((HeapWord*) old, obj_ptr, word_sz); |
| 281 | |
| 282 | if (dest_attr.is_young()) { |
| 283 | if (age < markOopDesc::max_age) { |
| 284 | age++; |
| 285 | } |
| 286 | if (old_mark->has_displaced_mark_helper()) { |
| 287 | // In this case, we have to install the mark word first, |
| 288 | // otherwise obj looks to be forwarded (the old mark word, |
| 289 | // which contains the forward pointer, was copied) |
| 290 | obj->set_mark_raw(old_mark); |
| 291 | markOop new_mark = old_mark->displaced_mark_helper()->set_age(age); |
| 292 | old_mark->set_displaced_mark_helper(new_mark); |
| 293 | } else { |
| 294 | obj->set_mark_raw(old_mark->set_age(age)); |
| 295 | } |
| 296 | _age_table.add(age, word_sz); |
| 297 | } else { |
| 298 | obj->set_mark_raw(old_mark); |
| 299 | } |
| 300 | |
| 301 | if (G1StringDedup::is_enabled()) { |
| 302 | const bool is_from_young = region_attr.is_young(); |
| 303 | const bool is_to_young = dest_attr.is_young(); |
| 304 | assert(is_from_young == _g1h->heap_region_containing(old)->is_young(), |
| 305 | "sanity" ); |
| 306 | assert(is_to_young == _g1h->heap_region_containing(obj)->is_young(), |
| 307 | "sanity" ); |
| 308 | G1StringDedup::enqueue_from_evacuation(is_from_young, |
| 309 | is_to_young, |
| 310 | _worker_id, |
| 311 | obj); |
| 312 | } |
| 313 | |
| 314 | _surviving_young_words[young_index] += word_sz; |
| 315 | |
| 316 | if (obj->is_objArray() && arrayOop(obj)->length() >= ParGCArrayScanChunk) { |
| 317 | // We keep track of the next start index in the length field of |
| 318 | // the to-space object. The actual length can be found in the |
| 319 | // length field of the from-space object. |
| 320 | arrayOop(obj)->set_length(0); |
| 321 | oop* old_p = set_partial_array_mask(old); |
| 322 | do_oop_partial_array(old_p); |
| 323 | } else { |
| 324 | G1ScanInYoungSetter x(&_scanner, dest_attr.is_young()); |
| 325 | obj->oop_iterate_backwards(&_scanner); |
| 326 | } |
| 327 | return obj; |
| 328 | } else { |
| 329 | _plab_allocator->undo_allocation(dest_attr, obj_ptr, word_sz); |
| 330 | return forward_ptr; |
| 331 | } |
| 332 | } |
| 333 | |
| 334 | G1ParScanThreadState* G1ParScanThreadStateSet::state_for_worker(uint worker_id) { |
| 335 | assert(worker_id < _n_workers, "out of bounds access" ); |
| 336 | if (_states[worker_id] == NULL) { |
| 337 | _states[worker_id] = |
| 338 | new G1ParScanThreadState(_g1h, worker_id, _young_cset_length, _optional_cset_length); |
| 339 | } |
| 340 | return _states[worker_id]; |
| 341 | } |
| 342 | |
| 343 | const size_t* G1ParScanThreadStateSet::surviving_young_words() const { |
| 344 | assert(_flushed, "thread local state from the per thread states should have been flushed" ); |
| 345 | return _surviving_young_words_total; |
| 346 | } |
| 347 | |
| 348 | void G1ParScanThreadStateSet::flush() { |
| 349 | assert(!_flushed, "thread local state from the per thread states should be flushed once" ); |
| 350 | |
| 351 | for (uint worker_index = 0; worker_index < _n_workers; ++worker_index) { |
| 352 | G1ParScanThreadState* pss = _states[worker_index]; |
| 353 | |
| 354 | if (pss == NULL) { |
| 355 | continue; |
| 356 | } |
| 357 | |
| 358 | pss->flush(_surviving_young_words_total); |
| 359 | delete pss; |
| 360 | _states[worker_index] = NULL; |
| 361 | } |
| 362 | _flushed = true; |
| 363 | } |
| 364 | |
| 365 | void G1ParScanThreadStateSet::record_unused_optional_region(HeapRegion* hr) { |
| 366 | for (uint worker_index = 0; worker_index < _n_workers; ++worker_index) { |
| 367 | G1ParScanThreadState* pss = _states[worker_index]; |
| 368 | |
| 369 | if (pss == NULL) { |
| 370 | continue; |
| 371 | } |
| 372 | |
| 373 | size_t used_memory = pss->oops_into_optional_region(hr)->used_memory(); |
| 374 | _g1h->phase_times()->record_or_add_thread_work_item(G1GCPhaseTimes::OptScanRS, worker_index, used_memory, G1GCPhaseTimes::ScanRSUsedMemory); |
| 375 | } |
| 376 | } |
| 377 | |
| 378 | oop G1ParScanThreadState::handle_evacuation_failure_par(oop old, markOop m) { |
| 379 | assert(_g1h->is_in_cset(old), "Object " PTR_FORMAT " should be in the CSet" , p2i(old)); |
| 380 | |
| 381 | oop forward_ptr = old->forward_to_atomic(old, m, memory_order_relaxed); |
| 382 | if (forward_ptr == NULL) { |
| 383 | // Forward-to-self succeeded. We are the "owner" of the object. |
| 384 | HeapRegion* r = _g1h->heap_region_containing(old); |
| 385 | |
| 386 | if (!r->evacuation_failed()) { |
| 387 | r->set_evacuation_failed(true); |
| 388 | _g1h->hr_printer()->evac_failure(r); |
| 389 | } |
| 390 | |
| 391 | _g1h->preserve_mark_during_evac_failure(_worker_id, old, m); |
| 392 | |
| 393 | G1ScanInYoungSetter x(&_scanner, r->is_young()); |
| 394 | old->oop_iterate_backwards(&_scanner); |
| 395 | |
| 396 | return old; |
| 397 | } else { |
| 398 | // Forward-to-self failed. Either someone else managed to allocate |
| 399 | // space for this object (old != forward_ptr) or they beat us in |
| 400 | // self-forwarding it (old == forward_ptr). |
| 401 | assert(old == forward_ptr || !_g1h->is_in_cset(forward_ptr), |
| 402 | "Object " PTR_FORMAT " forwarded to: " PTR_FORMAT " " |
| 403 | "should not be in the CSet" , |
| 404 | p2i(old), p2i(forward_ptr)); |
| 405 | return forward_ptr; |
| 406 | } |
| 407 | } |
| 408 | G1ParScanThreadStateSet::G1ParScanThreadStateSet(G1CollectedHeap* g1h, |
| 409 | uint n_workers, |
| 410 | size_t young_cset_length, |
| 411 | size_t optional_cset_length) : |
| 412 | _g1h(g1h), |
| 413 | _states(NEW_C_HEAP_ARRAY(G1ParScanThreadState*, n_workers, mtGC)), |
| 414 | _surviving_young_words_total(NEW_C_HEAP_ARRAY(size_t, young_cset_length, mtGC)), |
| 415 | _young_cset_length(young_cset_length), |
| 416 | _optional_cset_length(optional_cset_length), |
| 417 | _n_workers(n_workers), |
| 418 | _flushed(false) { |
| 419 | for (uint i = 0; i < n_workers; ++i) { |
| 420 | _states[i] = NULL; |
| 421 | } |
| 422 | memset(_surviving_young_words_total, 0, young_cset_length * sizeof(size_t)); |
| 423 | } |
| 424 | |
| 425 | G1ParScanThreadStateSet::~G1ParScanThreadStateSet() { |
| 426 | assert(_flushed, "thread local state from the per thread states should have been flushed" ); |
| 427 | FREE_C_HEAP_ARRAY(G1ParScanThreadState*, _states); |
| 428 | FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_total); |
| 429 | } |
| 430 | |