Tue, 26 Jan 2010 16:52:29 -0800
6920090: G1: Disable ReduceInitialCardMarks at least until 6920109 is fixed
Summary: G1 now answers "no" to the query can_elide_initializing_store_barrier() in the product build. A debug flag allows alternate behaviour in debug builds.
Reviewed-by: iveresov, tonyp
ysr@777 | 1 | /* |
xdono@1014 | 2 | * Copyright 2001-2009 Sun Microsystems, Inc. All Rights Reserved. |
ysr@777 | 3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
ysr@777 | 4 | * |
ysr@777 | 5 | * This code is free software; you can redistribute it and/or modify it |
ysr@777 | 6 | * under the terms of the GNU General Public License version 2 only, as |
ysr@777 | 7 | * published by the Free Software Foundation. |
ysr@777 | 8 | * |
ysr@777 | 9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
ysr@777 | 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
ysr@777 | 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
ysr@777 | 12 | * version 2 for more details (a copy is included in the LICENSE file that |
ysr@777 | 13 | * accompanied this code). |
ysr@777 | 14 | * |
ysr@777 | 15 | * You should have received a copy of the GNU General Public License version |
ysr@777 | 16 | * 2 along with this work; if not, write to the Free Software Foundation, |
ysr@777 | 17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
ysr@777 | 18 | * |
ysr@777 | 19 | * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
ysr@777 | 20 | * CA 95054 USA or visit www.sun.com if you need additional information or |
ysr@777 | 21 | * have any questions. |
ysr@777 | 22 | * |
ysr@777 | 23 | */ |
ysr@777 | 24 | |
ysr@777 | 25 | #include "incls/_precompiled.incl" |
ysr@777 | 26 | #include "incls/_g1RemSet.cpp.incl" |
ysr@777 | 27 | |
ysr@777 | 28 | #define CARD_REPEAT_HISTO 0 |
ysr@777 | 29 | |
ysr@777 | 30 | #if CARD_REPEAT_HISTO |
ysr@777 | 31 | static size_t ct_freq_sz; |
ysr@777 | 32 | static jbyte* ct_freq = NULL; |
ysr@777 | 33 | |
ysr@777 | 34 | void init_ct_freq_table(size_t heap_sz_bytes) { |
ysr@777 | 35 | if (ct_freq == NULL) { |
ysr@777 | 36 | ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size; |
ysr@777 | 37 | ct_freq = new jbyte[ct_freq_sz]; |
ysr@777 | 38 | for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0; |
ysr@777 | 39 | } |
ysr@777 | 40 | } |
ysr@777 | 41 | |
ysr@777 | 42 | void ct_freq_note_card(size_t index) { |
ysr@777 | 43 | assert(0 <= index && index < ct_freq_sz, "Bounds error."); |
ysr@777 | 44 | if (ct_freq[index] < 100) { ct_freq[index]++; } |
ysr@777 | 45 | } |
ysr@777 | 46 | |
ysr@777 | 47 | static IntHistogram card_repeat_count(10, 10); |
ysr@777 | 48 | |
ysr@777 | 49 | void ct_freq_update_histo_and_reset() { |
ysr@777 | 50 | for (size_t j = 0; j < ct_freq_sz; j++) { |
ysr@777 | 51 | card_repeat_count.add_entry(ct_freq[j]); |
ysr@777 | 52 | ct_freq[j] = 0; |
ysr@777 | 53 | } |
ysr@777 | 54 | |
ysr@777 | 55 | } |
ysr@777 | 56 | #endif |
ysr@777 | 57 | |
ysr@777 | 58 | |
ysr@777 | 59 | class IntoCSOopClosure: public OopsInHeapRegionClosure { |
ysr@777 | 60 | OopsInHeapRegionClosure* _blk; |
ysr@777 | 61 | G1CollectedHeap* _g1; |
ysr@777 | 62 | public: |
ysr@777 | 63 | IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : |
ysr@777 | 64 | _g1(g1), _blk(blk) {} |
ysr@777 | 65 | void set_region(HeapRegion* from) { |
ysr@777 | 66 | _blk->set_region(from); |
ysr@777 | 67 | } |
ysr@1280 | 68 | virtual void do_oop(narrowOop* p) { do_oop_work(p); } |
ysr@1280 | 69 | virtual void do_oop( oop* p) { do_oop_work(p); } |
ysr@1280 | 70 | template <class T> void do_oop_work(T* p) { |
ysr@1280 | 71 | oop obj = oopDesc::load_decode_heap_oop(p); |
ysr@777 | 72 | if (_g1->obj_in_cs(obj)) _blk->do_oop(p); |
ysr@777 | 73 | } |
ysr@777 | 74 | bool apply_to_weak_ref_discovered_field() { return true; } |
ysr@777 | 75 | bool idempotent() { return true; } |
ysr@777 | 76 | }; |
ysr@777 | 77 | |
ysr@777 | 78 | class IntoCSRegionClosure: public HeapRegionClosure { |
ysr@777 | 79 | IntoCSOopClosure _blk; |
ysr@777 | 80 | G1CollectedHeap* _g1; |
ysr@777 | 81 | public: |
ysr@777 | 82 | IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : |
ysr@777 | 83 | _g1(g1), _blk(g1, blk) {} |
ysr@777 | 84 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 85 | if (!r->in_collection_set()) { |
ysr@777 | 86 | _blk.set_region(r); |
ysr@777 | 87 | if (r->isHumongous()) { |
ysr@777 | 88 | if (r->startsHumongous()) { |
ysr@777 | 89 | oop obj = oop(r->bottom()); |
ysr@777 | 90 | obj->oop_iterate(&_blk); |
ysr@777 | 91 | } |
ysr@777 | 92 | } else { |
ysr@777 | 93 | r->oop_before_save_marks_iterate(&_blk); |
ysr@777 | 94 | } |
ysr@777 | 95 | } |
ysr@777 | 96 | return false; |
ysr@777 | 97 | } |
ysr@777 | 98 | }; |
ysr@777 | 99 | |
ysr@777 | 100 | void |
ysr@777 | 101 | StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, |
ysr@777 | 102 | int worker_i) { |
ysr@777 | 103 | IntoCSRegionClosure rc(_g1, oc); |
ysr@777 | 104 | _g1->heap_region_iterate(&rc); |
ysr@777 | 105 | } |
ysr@777 | 106 | |
ysr@777 | 107 | class VerifyRSCleanCardOopClosure: public OopClosure { |
ysr@777 | 108 | G1CollectedHeap* _g1; |
ysr@777 | 109 | public: |
ysr@777 | 110 | VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {} |
ysr@777 | 111 | |
ysr@1280 | 112 | virtual void do_oop(narrowOop* p) { do_oop_work(p); } |
ysr@1280 | 113 | virtual void do_oop( oop* p) { do_oop_work(p); } |
ysr@1280 | 114 | template <class T> void do_oop_work(T* p) { |
ysr@1280 | 115 | oop obj = oopDesc::load_decode_heap_oop(p); |
ysr@777 | 116 | HeapRegion* to = _g1->heap_region_containing(obj); |
ysr@777 | 117 | guarantee(to == NULL || !to->in_collection_set(), |
ysr@777 | 118 | "Missed a rem set member."); |
ysr@777 | 119 | } |
ysr@777 | 120 | }; |
ysr@777 | 121 | |
ysr@777 | 122 | HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) |
ysr@777 | 123 | : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()), |
ysr@777 | 124 | _cg1r(g1->concurrent_g1_refine()), |
ysr@777 | 125 | _par_traversal_in_progress(false), _new_refs(NULL), |
ysr@777 | 126 | _cards_scanned(NULL), _total_cards_scanned(0) |
ysr@777 | 127 | { |
ysr@777 | 128 | _seq_task = new SubTasksDone(NumSeqTasks); |
iveresov@1051 | 129 | guarantee(n_workers() > 0, "There should be some workers"); |
ysr@1280 | 130 | _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, n_workers()); |
iveresov@1051 | 131 | for (uint i = 0; i < n_workers(); i++) { |
ysr@1280 | 132 | _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<OopOrNarrowOopStar>(8192,true); |
iveresov@1051 | 133 | } |
ysr@777 | 134 | } |
ysr@777 | 135 | |
ysr@777 | 136 | HRInto_G1RemSet::~HRInto_G1RemSet() { |
ysr@777 | 137 | delete _seq_task; |
iveresov@1051 | 138 | for (uint i = 0; i < n_workers(); i++) { |
iveresov@1051 | 139 | delete _new_refs[i]; |
iveresov@1051 | 140 | } |
ysr@1280 | 141 | FREE_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, _new_refs); |
ysr@777 | 142 | } |
ysr@777 | 143 | |
ysr@777 | 144 | void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) { |
ysr@777 | 145 | if (_g1->is_in_g1_reserved(mr.start())) { |
ysr@777 | 146 | _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size)); |
ysr@777 | 147 | if (_start_first == NULL) _start_first = mr.start(); |
ysr@777 | 148 | } |
ysr@777 | 149 | } |
ysr@777 | 150 | |
ysr@777 | 151 | class ScanRSClosure : public HeapRegionClosure { |
ysr@777 | 152 | size_t _cards_done, _cards; |
ysr@777 | 153 | G1CollectedHeap* _g1h; |
ysr@777 | 154 | OopsInHeapRegionClosure* _oc; |
ysr@777 | 155 | G1BlockOffsetSharedArray* _bot_shared; |
ysr@777 | 156 | CardTableModRefBS *_ct_bs; |
ysr@777 | 157 | int _worker_i; |
ysr@777 | 158 | bool _try_claimed; |
iveresov@1182 | 159 | size_t _min_skip_distance, _max_skip_distance; |
ysr@777 | 160 | public: |
ysr@777 | 161 | ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) : |
ysr@777 | 162 | _oc(oc), |
ysr@777 | 163 | _cards(0), |
ysr@777 | 164 | _cards_done(0), |
ysr@777 | 165 | _worker_i(worker_i), |
ysr@777 | 166 | _try_claimed(false) |
ysr@777 | 167 | { |
ysr@777 | 168 | _g1h = G1CollectedHeap::heap(); |
ysr@777 | 169 | _bot_shared = _g1h->bot_shared(); |
ysr@777 | 170 | _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set()); |
iveresov@1182 | 171 | _min_skip_distance = 16; |
iveresov@1182 | 172 | _max_skip_distance = 2 * _g1h->n_par_threads() * _min_skip_distance; |
ysr@777 | 173 | } |
ysr@777 | 174 | |
ysr@777 | 175 | void set_try_claimed() { _try_claimed = true; } |
ysr@777 | 176 | |
ysr@777 | 177 | void scanCard(size_t index, HeapRegion *r) { |
ysr@777 | 178 | _cards_done++; |
ysr@777 | 179 | DirtyCardToOopClosure* cl = |
ysr@777 | 180 | r->new_dcto_closure(_oc, |
ysr@777 | 181 | CardTableModRefBS::Precise, |
ysr@777 | 182 | HeapRegionDCTOC::IntoCSFilterKind); |
ysr@777 | 183 | |
ysr@777 | 184 | // Set the "from" region in the closure. |
ysr@777 | 185 | _oc->set_region(r); |
ysr@777 | 186 | HeapWord* card_start = _bot_shared->address_for_index(index); |
ysr@777 | 187 | HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words; |
ysr@777 | 188 | Space *sp = SharedHeap::heap()->space_containing(card_start); |
ysr@777 | 189 | MemRegion sm_region; |
ysr@777 | 190 | if (ParallelGCThreads > 0) { |
ysr@777 | 191 | // first find the used area |
ysr@777 | 192 | sm_region = sp->used_region_at_save_marks(); |
ysr@777 | 193 | } else { |
ysr@777 | 194 | // The closure is not idempotent. We shouldn't look at objects |
ysr@777 | 195 | // allocated during the GC. |
ysr@777 | 196 | sm_region = sp->used_region_at_save_marks(); |
ysr@777 | 197 | } |
ysr@777 | 198 | MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end)); |
ysr@777 | 199 | if (!mr.is_empty()) { |
ysr@777 | 200 | cl->do_MemRegion(mr); |
ysr@777 | 201 | } |
ysr@777 | 202 | } |
ysr@777 | 203 | |
ysr@777 | 204 | void printCard(HeapRegion* card_region, size_t card_index, |
ysr@777 | 205 | HeapWord* card_start) { |
ysr@777 | 206 | gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") " |
ysr@777 | 207 | "RS names card %p: " |
ysr@777 | 208 | "[" PTR_FORMAT ", " PTR_FORMAT ")", |
ysr@777 | 209 | _worker_i, |
ysr@777 | 210 | card_region->bottom(), card_region->end(), |
ysr@777 | 211 | card_index, |
ysr@777 | 212 | card_start, card_start + G1BlockOffsetSharedArray::N_words); |
ysr@777 | 213 | } |
ysr@777 | 214 | |
ysr@777 | 215 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 216 | assert(r->in_collection_set(), "should only be called on elements of CS."); |
ysr@777 | 217 | HeapRegionRemSet* hrrs = r->rem_set(); |
ysr@777 | 218 | if (hrrs->iter_is_complete()) return false; // All done. |
ysr@777 | 219 | if (!_try_claimed && !hrrs->claim_iter()) return false; |
apetrusenko@1231 | 220 | _g1h->push_dirty_cards_region(r); |
ysr@777 | 221 | // If we didn't return above, then |
ysr@777 | 222 | // _try_claimed || r->claim_iter() |
ysr@777 | 223 | // is true: either we're supposed to work on claimed-but-not-complete |
ysr@777 | 224 | // regions, or we successfully claimed the region. |
ysr@777 | 225 | HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i); |
ysr@777 | 226 | hrrs->init_iterator(iter); |
ysr@777 | 227 | size_t card_index; |
iveresov@1182 | 228 | size_t skip_distance = 0, current_card = 0, jump_to_card = 0; |
ysr@777 | 229 | while (iter->has_next(card_index)) { |
iveresov@1182 | 230 | if (current_card < jump_to_card) { |
iveresov@1182 | 231 | ++current_card; |
iveresov@1182 | 232 | continue; |
iveresov@1182 | 233 | } |
ysr@777 | 234 | HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index); |
ysr@777 | 235 | #if 0 |
ysr@777 | 236 | gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n", |
ysr@777 | 237 | card_start, card_start + CardTableModRefBS::card_size_in_words); |
ysr@777 | 238 | #endif |
ysr@777 | 239 | |
ysr@777 | 240 | HeapRegion* card_region = _g1h->heap_region_containing(card_start); |
ysr@777 | 241 | assert(card_region != NULL, "Yielding cards not in the heap?"); |
ysr@777 | 242 | _cards++; |
ysr@777 | 243 | |
apetrusenko@1231 | 244 | if (!card_region->is_on_dirty_cards_region_list()) { |
apetrusenko@1231 | 245 | _g1h->push_dirty_cards_region(card_region); |
apetrusenko@1231 | 246 | } |
apetrusenko@1231 | 247 | |
iveresov@1182 | 248 | // If the card is dirty, then we will scan it during updateRS. |
iveresov@1182 | 249 | if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) { |
iveresov@1182 | 250 | if (!_ct_bs->is_card_claimed(card_index) && _ct_bs->claim_card(card_index)) { |
ysr@777 | 251 | scanCard(card_index, card_region); |
iveresov@1182 | 252 | } else if (_try_claimed) { |
iveresov@1182 | 253 | if (jump_to_card == 0 || jump_to_card != current_card) { |
iveresov@1182 | 254 | // We did some useful work in the previous iteration. |
iveresov@1182 | 255 | // Decrease the distance. |
iveresov@1182 | 256 | skip_distance = MAX2(skip_distance >> 1, _min_skip_distance); |
iveresov@1182 | 257 | } else { |
iveresov@1182 | 258 | // Previous iteration resulted in a claim failure. |
iveresov@1182 | 259 | // Increase the distance. |
iveresov@1182 | 260 | skip_distance = MIN2(skip_distance << 1, _max_skip_distance); |
iveresov@1182 | 261 | } |
iveresov@1182 | 262 | jump_to_card = current_card + skip_distance; |
iveresov@1182 | 263 | } |
ysr@777 | 264 | } |
iveresov@1182 | 265 | ++current_card; |
ysr@777 | 266 | } |
iveresov@1182 | 267 | if (!_try_claimed) { |
iveresov@1182 | 268 | hrrs->set_iter_complete(); |
iveresov@1182 | 269 | } |
ysr@777 | 270 | return false; |
ysr@777 | 271 | } |
ysr@777 | 272 | // Set all cards back to clean. |
ysr@777 | 273 | void cleanup() {_g1h->cleanUpCardTable();} |
ysr@777 | 274 | size_t cards_done() { return _cards_done;} |
ysr@777 | 275 | size_t cards_looked_up() { return _cards;} |
ysr@777 | 276 | }; |
ysr@777 | 277 | |
ysr@777 | 278 | // We want the parallel threads to start their scanning at |
ysr@777 | 279 | // different collection set regions to avoid contention. |
ysr@777 | 280 | // If we have: |
ysr@777 | 281 | // n collection set regions |
ysr@777 | 282 | // p threads |
ysr@777 | 283 | // Then thread t will start at region t * floor (n/p) |
ysr@777 | 284 | |
ysr@777 | 285 | HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) { |
ysr@777 | 286 | HeapRegion* result = _g1p->collection_set(); |
ysr@777 | 287 | if (ParallelGCThreads > 0) { |
ysr@777 | 288 | size_t cs_size = _g1p->collection_set_size(); |
ysr@777 | 289 | int n_workers = _g1->workers()->total_workers(); |
ysr@777 | 290 | size_t cs_spans = cs_size / n_workers; |
ysr@777 | 291 | size_t ind = cs_spans * worker_i; |
ysr@777 | 292 | for (size_t i = 0; i < ind; i++) |
ysr@777 | 293 | result = result->next_in_collection_set(); |
ysr@777 | 294 | } |
ysr@777 | 295 | return result; |
ysr@777 | 296 | } |
ysr@777 | 297 | |
ysr@777 | 298 | void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { |
ysr@777 | 299 | double rs_time_start = os::elapsedTime(); |
ysr@777 | 300 | HeapRegion *startRegion = calculateStartRegion(worker_i); |
ysr@777 | 301 | |
ysr@777 | 302 | BufferingOopsInHeapRegionClosure boc(oc); |
ysr@777 | 303 | ScanRSClosure scanRScl(&boc, worker_i); |
ysr@777 | 304 | _g1->collection_set_iterate_from(startRegion, &scanRScl); |
ysr@777 | 305 | scanRScl.set_try_claimed(); |
ysr@777 | 306 | _g1->collection_set_iterate_from(startRegion, &scanRScl); |
ysr@777 | 307 | |
ysr@777 | 308 | boc.done(); |
ysr@777 | 309 | double closure_app_time_sec = boc.closure_app_seconds(); |
ysr@777 | 310 | double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) - |
ysr@777 | 311 | closure_app_time_sec; |
ysr@777 | 312 | double closure_app_time_ms = closure_app_time_sec * 1000.0; |
ysr@777 | 313 | |
ysr@777 | 314 | assert( _cards_scanned != NULL, "invariant" ); |
ysr@777 | 315 | _cards_scanned[worker_i] = scanRScl.cards_done(); |
ysr@777 | 316 | |
ysr@777 | 317 | _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0); |
ysr@777 | 318 | _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); |
iveresov@1051 | 319 | |
iveresov@1051 | 320 | double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i); |
iveresov@1051 | 321 | if (scan_new_refs_time_ms > 0.0) { |
iveresov@1051 | 322 | closure_app_time_ms += scan_new_refs_time_ms; |
ysr@777 | 323 | } |
iveresov@1051 | 324 | |
ysr@777 | 325 | _g1p->record_obj_copy_time(worker_i, closure_app_time_ms); |
ysr@777 | 326 | } |
ysr@777 | 327 | |
ysr@777 | 328 | void HRInto_G1RemSet::updateRS(int worker_i) { |
ysr@777 | 329 | ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); |
ysr@777 | 330 | |
ysr@777 | 331 | double start = os::elapsedTime(); |
ysr@777 | 332 | _g1p->record_update_rs_start_time(worker_i, start * 1000.0); |
ysr@777 | 333 | |
iveresov@1229 | 334 | // Apply the appropriate closure to all remaining log entries. |
iveresov@1229 | 335 | _g1->iterate_dirty_card_closure(false, worker_i); |
iveresov@1229 | 336 | // Now there should be no dirty cards. |
iveresov@1229 | 337 | if (G1RSLogCheckCardTable) { |
iveresov@1229 | 338 | CountNonCleanMemRegionClosure cl(_g1); |
iveresov@1229 | 339 | _ct_bs->mod_card_iterate(&cl); |
iveresov@1229 | 340 | // XXX This isn't true any more: keeping cards of young regions |
iveresov@1229 | 341 | // marked dirty broke it. Need some reasonable fix. |
iveresov@1229 | 342 | guarantee(cl.n() == 0, "Card table should be clean."); |
ysr@777 | 343 | } |
iveresov@1229 | 344 | |
ysr@777 | 345 | _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0); |
ysr@777 | 346 | } |
ysr@777 | 347 | |
ysr@777 | 348 | #ifndef PRODUCT |
ysr@777 | 349 | class PrintRSClosure : public HeapRegionClosure { |
ysr@777 | 350 | int _count; |
ysr@777 | 351 | public: |
ysr@777 | 352 | PrintRSClosure() : _count(0) {} |
ysr@777 | 353 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 354 | HeapRegionRemSet* hrrs = r->rem_set(); |
ysr@777 | 355 | _count += (int) hrrs->occupied(); |
ysr@777 | 356 | if (hrrs->occupied() == 0) { |
ysr@777 | 357 | gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") " |
ysr@777 | 358 | "has no remset entries\n", |
ysr@777 | 359 | r->bottom(), r->end()); |
ysr@777 | 360 | } else { |
ysr@777 | 361 | gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n", |
ysr@777 | 362 | r->bottom(), r->end()); |
ysr@777 | 363 | r->print(); |
ysr@777 | 364 | hrrs->print(); |
ysr@777 | 365 | gclog_or_tty->print("\nDone printing rem set\n"); |
ysr@777 | 366 | } |
ysr@777 | 367 | return false; |
ysr@777 | 368 | } |
ysr@777 | 369 | int occupied() {return _count;} |
ysr@777 | 370 | }; |
ysr@777 | 371 | #endif |
ysr@777 | 372 | |
ysr@777 | 373 | class CountRSSizeClosure: public HeapRegionClosure { |
ysr@777 | 374 | size_t _n; |
ysr@777 | 375 | size_t _tot; |
ysr@777 | 376 | size_t _max; |
ysr@777 | 377 | HeapRegion* _max_r; |
ysr@777 | 378 | enum { |
ysr@777 | 379 | N = 20, |
ysr@777 | 380 | MIN = 6 |
ysr@777 | 381 | }; |
ysr@777 | 382 | int _histo[N]; |
ysr@777 | 383 | public: |
ysr@777 | 384 | CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) { |
ysr@777 | 385 | for (int i = 0; i < N; i++) _histo[i] = 0; |
ysr@777 | 386 | } |
ysr@777 | 387 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 388 | if (!r->continuesHumongous()) { |
ysr@777 | 389 | size_t occ = r->rem_set()->occupied(); |
ysr@777 | 390 | _n++; |
ysr@777 | 391 | _tot += occ; |
ysr@777 | 392 | if (occ > _max) { |
ysr@777 | 393 | _max = occ; |
ysr@777 | 394 | _max_r = r; |
ysr@777 | 395 | } |
ysr@777 | 396 | // Fit it into a histo bin. |
ysr@777 | 397 | int s = 1 << MIN; |
ysr@777 | 398 | int i = 0; |
ysr@777 | 399 | while (occ > (size_t) s && i < (N-1)) { |
ysr@777 | 400 | s = s << 1; |
ysr@777 | 401 | i++; |
ysr@777 | 402 | } |
ysr@777 | 403 | _histo[i]++; |
ysr@777 | 404 | } |
ysr@777 | 405 | return false; |
ysr@777 | 406 | } |
ysr@777 | 407 | size_t n() { return _n; } |
ysr@777 | 408 | size_t tot() { return _tot; } |
ysr@777 | 409 | size_t mx() { return _max; } |
ysr@777 | 410 | HeapRegion* mxr() { return _max_r; } |
ysr@777 | 411 | void print_histo() { |
ysr@777 | 412 | int mx = N; |
ysr@777 | 413 | while (mx >= 0) { |
ysr@777 | 414 | if (_histo[mx-1] > 0) break; |
ysr@777 | 415 | mx--; |
ysr@777 | 416 | } |
ysr@777 | 417 | gclog_or_tty->print_cr("Number of regions with given RS sizes:"); |
ysr@777 | 418 | gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]); |
ysr@777 | 419 | for (int i = 1; i < mx-1; i++) { |
ysr@777 | 420 | gclog_or_tty->print_cr(" %8d - %8d %8d", |
ysr@777 | 421 | (1 << (MIN + i - 1)) + 1, |
ysr@777 | 422 | 1 << (MIN + i), |
ysr@777 | 423 | _histo[i]); |
ysr@777 | 424 | } |
ysr@777 | 425 | gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]); |
ysr@777 | 426 | } |
ysr@777 | 427 | }; |
ysr@777 | 428 | |
ysr@1280 | 429 | template <class T> void |
ysr@1280 | 430 | HRInto_G1RemSet::scanNewRefsRS_work(OopsInHeapRegionClosure* oc, |
ysr@1280 | 431 | int worker_i) { |
ysr@777 | 432 | double scan_new_refs_start_sec = os::elapsedTime(); |
ysr@777 | 433 | G1CollectedHeap* g1h = G1CollectedHeap::heap(); |
ysr@777 | 434 | CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set()); |
iveresov@1051 | 435 | for (int i = 0; i < _new_refs[worker_i]->length(); i++) { |
ysr@1280 | 436 | T* p = (T*) _new_refs[worker_i]->at(i); |
ysr@1280 | 437 | oop obj = oopDesc::load_decode_heap_oop(p); |
ysr@777 | 438 | // *p was in the collection set when p was pushed on "_new_refs", but |
ysr@777 | 439 | // another thread may have processed this location from an RS, so it |
ysr@777 | 440 | // might not point into the CS any longer. If so, it's obviously been |
ysr@777 | 441 | // processed, and we don't need to do anything further. |
ysr@777 | 442 | if (g1h->obj_in_cs(obj)) { |
ysr@777 | 443 | HeapRegion* r = g1h->heap_region_containing(p); |
ysr@777 | 444 | |
ysr@777 | 445 | DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj)); |
ysr@777 | 446 | oc->set_region(r); |
ysr@777 | 447 | // If "p" has already been processed concurrently, this is |
ysr@777 | 448 | // idempotent. |
ysr@777 | 449 | oc->do_oop(p); |
ysr@777 | 450 | } |
ysr@777 | 451 | } |
ysr@777 | 452 | _g1p->record_scan_new_refs_time(worker_i, |
ysr@777 | 453 | (os::elapsedTime() - scan_new_refs_start_sec) |
ysr@777 | 454 | * 1000.0); |
ysr@777 | 455 | } |
ysr@777 | 456 | |
ysr@777 | 457 | void HRInto_G1RemSet::cleanupHRRS() { |
ysr@777 | 458 | HeapRegionRemSet::cleanup(); |
ysr@777 | 459 | } |
ysr@777 | 460 | |
ysr@777 | 461 | void |
ysr@777 | 462 | HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, |
ysr@777 | 463 | int worker_i) { |
ysr@777 | 464 | #if CARD_REPEAT_HISTO |
ysr@777 | 465 | ct_freq_update_histo_and_reset(); |
ysr@777 | 466 | #endif |
ysr@777 | 467 | if (worker_i == 0) { |
ysr@777 | 468 | _cg1r->clear_and_record_card_counts(); |
ysr@777 | 469 | } |
ysr@777 | 470 | |
ysr@777 | 471 | // Make this into a command-line flag... |
ysr@777 | 472 | if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) { |
ysr@777 | 473 | CountRSSizeClosure count_cl; |
ysr@777 | 474 | _g1->heap_region_iterate(&count_cl); |
ysr@777 | 475 | gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, " |
ysr@777 | 476 | "max region is " PTR_FORMAT, |
ysr@777 | 477 | count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(), |
ysr@777 | 478 | count_cl.mx(), count_cl.mxr()); |
ysr@777 | 479 | count_cl.print_histo(); |
ysr@777 | 480 | } |
ysr@777 | 481 | |
ysr@777 | 482 | if (ParallelGCThreads > 0) { |
tonyp@1073 | 483 | // The two flags below were introduced temporarily to serialize |
tonyp@1073 | 484 | // the updating and scanning of remembered sets. There are some |
tonyp@1073 | 485 | // race conditions when these two operations are done in parallel |
tonyp@1073 | 486 | // and they are causing failures. When we resolve said race |
tonyp@1073 | 487 | // conditions, we'll revert back to parallel remembered set |
tonyp@1073 | 488 | // updating and scanning. See CRs 6677707 and 6677708. |
johnc@1186 | 489 | if (G1ParallelRSetUpdatingEnabled || (worker_i == 0)) { |
ysr@777 | 490 | updateRS(worker_i); |
ysr@777 | 491 | scanNewRefsRS(oc, worker_i); |
tonyp@1083 | 492 | } else { |
iveresov@1229 | 493 | _g1p->record_update_rs_start_time(worker_i, os::elapsedTime() * 1000.0); |
tonyp@1083 | 494 | _g1p->record_update_rs_processed_buffers(worker_i, 0.0); |
tonyp@1083 | 495 | _g1p->record_update_rs_time(worker_i, 0.0); |
tonyp@1083 | 496 | _g1p->record_scan_new_refs_time(worker_i, 0.0); |
tonyp@1073 | 497 | } |
johnc@1186 | 498 | if (G1ParallelRSetScanningEnabled || (worker_i == 0)) { |
ysr@777 | 499 | scanRS(oc, worker_i); |
tonyp@1083 | 500 | } else { |
iveresov@1229 | 501 | _g1p->record_scan_rs_start_time(worker_i, os::elapsedTime() * 1000.0); |
tonyp@1083 | 502 | _g1p->record_scan_rs_time(worker_i, 0.0); |
ysr@777 | 503 | } |
ysr@777 | 504 | } else { |
ysr@777 | 505 | assert(worker_i == 0, "invariant"); |
ysr@777 | 506 | updateRS(0); |
iveresov@1051 | 507 | scanNewRefsRS(oc, 0); |
ysr@777 | 508 | scanRS(oc, 0); |
ysr@777 | 509 | } |
ysr@777 | 510 | } |
ysr@777 | 511 | |
ysr@777 | 512 | void HRInto_G1RemSet:: |
ysr@777 | 513 | prepare_for_oops_into_collection_set_do() { |
ysr@777 | 514 | #if G1_REM_SET_LOGGING |
ysr@777 | 515 | PrintRSClosure cl; |
ysr@777 | 516 | _g1->collection_set_iterate(&cl); |
ysr@777 | 517 | #endif |
ysr@777 | 518 | cleanupHRRS(); |
ysr@777 | 519 | ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); |
ysr@777 | 520 | _g1->set_refine_cte_cl_concurrency(false); |
ysr@777 | 521 | DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); |
ysr@777 | 522 | dcqs.concatenate_logs(); |
ysr@777 | 523 | |
ysr@777 | 524 | assert(!_par_traversal_in_progress, "Invariant between iterations."); |
ysr@777 | 525 | if (ParallelGCThreads > 0) { |
ysr@777 | 526 | set_par_traversal(true); |
iveresov@1051 | 527 | _seq_task->set_par_threads((int)n_workers()); |
ysr@777 | 528 | } |
ysr@777 | 529 | guarantee( _cards_scanned == NULL, "invariant" ); |
ysr@777 | 530 | _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); |
apetrusenko@980 | 531 | for (uint i = 0; i < n_workers(); ++i) { |
apetrusenko@980 | 532 | _cards_scanned[i] = 0; |
apetrusenko@980 | 533 | } |
ysr@777 | 534 | _total_cards_scanned = 0; |
ysr@777 | 535 | } |
ysr@777 | 536 | |
ysr@777 | 537 | |
ysr@777 | 538 | class cleanUpIteratorsClosure : public HeapRegionClosure { |
ysr@777 | 539 | bool doHeapRegion(HeapRegion *r) { |
ysr@777 | 540 | HeapRegionRemSet* hrrs = r->rem_set(); |
ysr@777 | 541 | hrrs->init_for_par_iteration(); |
ysr@777 | 542 | return false; |
ysr@777 | 543 | } |
ysr@777 | 544 | }; |
ysr@777 | 545 | |
iveresov@1051 | 546 | class UpdateRSetOopsIntoCSImmediate : public OopClosure { |
iveresov@1051 | 547 | G1CollectedHeap* _g1; |
iveresov@1051 | 548 | public: |
iveresov@1051 | 549 | UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { } |
ysr@1280 | 550 | virtual void do_oop(narrowOop* p) { do_oop_work(p); } |
ysr@1280 | 551 | virtual void do_oop( oop* p) { do_oop_work(p); } |
ysr@1280 | 552 | template <class T> void do_oop_work(T* p) { |
ysr@1280 | 553 | HeapRegion* to = _g1->heap_region_containing(oopDesc::load_decode_heap_oop(p)); |
iveresov@1051 | 554 | if (to->in_collection_set()) { |
apetrusenko@1112 | 555 | to->rem_set()->add_reference(p, 0); |
iveresov@1051 | 556 | } |
iveresov@1051 | 557 | } |
iveresov@1051 | 558 | }; |
iveresov@1051 | 559 | |
iveresov@1051 | 560 | class UpdateRSetOopsIntoCSDeferred : public OopClosure { |
iveresov@1051 | 561 | G1CollectedHeap* _g1; |
iveresov@1051 | 562 | CardTableModRefBS* _ct_bs; |
iveresov@1051 | 563 | DirtyCardQueue* _dcq; |
iveresov@1051 | 564 | public: |
iveresov@1051 | 565 | UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) : |
iveresov@1051 | 566 | _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { } |
ysr@1280 | 567 | virtual void do_oop(narrowOop* p) { do_oop_work(p); } |
ysr@1280 | 568 | virtual void do_oop( oop* p) { do_oop_work(p); } |
ysr@1280 | 569 | template <class T> void do_oop_work(T* p) { |
ysr@1280 | 570 | oop obj = oopDesc::load_decode_heap_oop(p); |
iveresov@1051 | 571 | if (_g1->obj_in_cs(obj)) { |
iveresov@1051 | 572 | size_t card_index = _ct_bs->index_for(p); |
iveresov@1051 | 573 | if (_ct_bs->mark_card_deferred(card_index)) { |
iveresov@1051 | 574 | _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index)); |
iveresov@1051 | 575 | } |
iveresov@1051 | 576 | } |
iveresov@1051 | 577 | } |
iveresov@1051 | 578 | }; |
iveresov@1051 | 579 | |
ysr@1280 | 580 | template <class T> void HRInto_G1RemSet::new_refs_iterate_work(OopClosure* cl) { |
iveresov@1051 | 581 | for (size_t i = 0; i < n_workers(); i++) { |
iveresov@1051 | 582 | for (int j = 0; j < _new_refs[i]->length(); j++) { |
ysr@1280 | 583 | T* p = (T*) _new_refs[i]->at(j); |
iveresov@1051 | 584 | cl->do_oop(p); |
iveresov@1051 | 585 | } |
iveresov@1051 | 586 | } |
iveresov@1051 | 587 | } |
iveresov@1051 | 588 | |
ysr@777 | 589 | void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() { |
ysr@777 | 590 | guarantee( _cards_scanned != NULL, "invariant" ); |
ysr@777 | 591 | _total_cards_scanned = 0; |
ysr@777 | 592 | for (uint i = 0; i < n_workers(); ++i) |
ysr@777 | 593 | _total_cards_scanned += _cards_scanned[i]; |
ysr@777 | 594 | FREE_C_HEAP_ARRAY(size_t, _cards_scanned); |
ysr@777 | 595 | _cards_scanned = NULL; |
ysr@777 | 596 | // Cleanup after copy |
ysr@777 | 597 | #if G1_REM_SET_LOGGING |
ysr@777 | 598 | PrintRSClosure cl; |
ysr@777 | 599 | _g1->heap_region_iterate(&cl); |
ysr@777 | 600 | #endif |
ysr@777 | 601 | _g1->set_refine_cte_cl_concurrency(true); |
ysr@777 | 602 | cleanUpIteratorsClosure iterClosure; |
ysr@777 | 603 | _g1->collection_set_iterate(&iterClosure); |
ysr@777 | 604 | // Set all cards back to clean. |
ysr@777 | 605 | _g1->cleanUpCardTable(); |
iveresov@1229 | 606 | |
ysr@777 | 607 | if (ParallelGCThreads > 0) { |
ysr@777 | 608 | set_par_traversal(false); |
ysr@777 | 609 | } |
iveresov@1051 | 610 | |
iveresov@1051 | 611 | if (_g1->evacuation_failed()) { |
iveresov@1051 | 612 | // Restore remembered sets for the regions pointing into |
iveresov@1051 | 613 | // the collection set. |
iveresov@1051 | 614 | if (G1DeferredRSUpdate) { |
iveresov@1051 | 615 | DirtyCardQueue dcq(&_g1->dirty_card_queue_set()); |
iveresov@1051 | 616 | UpdateRSetOopsIntoCSDeferred deferred_update(_g1, &dcq); |
iveresov@1051 | 617 | new_refs_iterate(&deferred_update); |
iveresov@1051 | 618 | } else { |
iveresov@1051 | 619 | UpdateRSetOopsIntoCSImmediate immediate_update(_g1); |
iveresov@1051 | 620 | new_refs_iterate(&immediate_update); |
iveresov@1051 | 621 | } |
iveresov@1051 | 622 | } |
iveresov@1051 | 623 | for (uint i = 0; i < n_workers(); i++) { |
iveresov@1051 | 624 | _new_refs[i]->clear(); |
iveresov@1051 | 625 | } |
iveresov@1051 | 626 | |
ysr@777 | 627 | assert(!_par_traversal_in_progress, "Invariant between iterations."); |
ysr@777 | 628 | } |
ysr@777 | 629 | |
ysr@777 | 630 | class UpdateRSObjectClosure: public ObjectClosure { |
ysr@777 | 631 | UpdateRSOopClosure* _update_rs_oop_cl; |
ysr@777 | 632 | public: |
ysr@777 | 633 | UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) : |
ysr@777 | 634 | _update_rs_oop_cl(update_rs_oop_cl) {} |
ysr@777 | 635 | void do_object(oop obj) { |
ysr@777 | 636 | obj->oop_iterate(_update_rs_oop_cl); |
ysr@777 | 637 | } |
ysr@777 | 638 | |
ysr@777 | 639 | }; |
ysr@777 | 640 | |
ysr@777 | 641 | class ScrubRSClosure: public HeapRegionClosure { |
ysr@777 | 642 | G1CollectedHeap* _g1h; |
ysr@777 | 643 | BitMap* _region_bm; |
ysr@777 | 644 | BitMap* _card_bm; |
ysr@777 | 645 | CardTableModRefBS* _ctbs; |
ysr@777 | 646 | public: |
ysr@777 | 647 | ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) : |
ysr@777 | 648 | _g1h(G1CollectedHeap::heap()), |
ysr@777 | 649 | _region_bm(region_bm), _card_bm(card_bm), |
ysr@777 | 650 | _ctbs(NULL) |
ysr@777 | 651 | { |
ysr@777 | 652 | ModRefBarrierSet* bs = _g1h->mr_bs(); |
ysr@777 | 653 | guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition"); |
ysr@777 | 654 | _ctbs = (CardTableModRefBS*)bs; |
ysr@777 | 655 | } |
ysr@777 | 656 | |
ysr@777 | 657 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 658 | if (!r->continuesHumongous()) { |
ysr@777 | 659 | r->rem_set()->scrub(_ctbs, _region_bm, _card_bm); |
ysr@777 | 660 | } |
ysr@777 | 661 | return false; |
ysr@777 | 662 | } |
ysr@777 | 663 | }; |
ysr@777 | 664 | |
ysr@777 | 665 | void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { |
ysr@777 | 666 | ScrubRSClosure scrub_cl(region_bm, card_bm); |
ysr@777 | 667 | _g1->heap_region_iterate(&scrub_cl); |
ysr@777 | 668 | } |
ysr@777 | 669 | |
ysr@777 | 670 | void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, |
ysr@777 | 671 | int worker_num, int claim_val) { |
ysr@777 | 672 | ScrubRSClosure scrub_cl(region_bm, card_bm); |
ysr@777 | 673 | _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val); |
ysr@777 | 674 | } |
ysr@777 | 675 | |
ysr@777 | 676 | |
ysr@777 | 677 | static IntHistogram out_of_histo(50, 50); |
ysr@777 | 678 | |
johnc@1325 | 679 | void HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i) { |
johnc@1325 | 680 | // Construct the region representing the card. |
johnc@1325 | 681 | HeapWord* start = _ct_bs->addr_for(card_ptr); |
johnc@1325 | 682 | // And find the region containing it. |
johnc@1325 | 683 | HeapRegion* r = _g1->heap_region_containing(start); |
johnc@1325 | 684 | assert(r != NULL, "unexpected null"); |
johnc@1325 | 685 | |
johnc@1325 | 686 | HeapWord* end = _ct_bs->addr_for(card_ptr + 1); |
johnc@1325 | 687 | MemRegion dirtyRegion(start, end); |
johnc@1325 | 688 | |
johnc@1325 | 689 | #if CARD_REPEAT_HISTO |
johnc@1325 | 690 | init_ct_freq_table(_g1->g1_reserved_obj_bytes()); |
johnc@1325 | 691 | ct_freq_note_card(_ct_bs->index_for(start)); |
johnc@1325 | 692 | #endif |
johnc@1325 | 693 | |
johnc@1325 | 694 | UpdateRSOopClosure update_rs_oop_cl(this, worker_i); |
johnc@1325 | 695 | update_rs_oop_cl.set_from(r); |
johnc@1325 | 696 | FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl); |
johnc@1325 | 697 | |
johnc@1325 | 698 | // Undirty the card. |
johnc@1325 | 699 | *card_ptr = CardTableModRefBS::clean_card_val(); |
johnc@1325 | 700 | // We must complete this write before we do any of the reads below. |
johnc@1325 | 701 | OrderAccess::storeload(); |
johnc@1325 | 702 | // And process it, being careful of unallocated portions of TLAB's. |
johnc@1325 | 703 | HeapWord* stop_point = |
johnc@1325 | 704 | r->oops_on_card_seq_iterate_careful(dirtyRegion, |
johnc@1325 | 705 | &filter_then_update_rs_oop_cl); |
johnc@1325 | 706 | // If stop_point is non-null, then we encountered an unallocated region |
johnc@1325 | 707 | // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the |
johnc@1325 | 708 | // card and re-enqueue: if we put off the card until a GC pause, then the |
johnc@1325 | 709 | // unallocated portion will be filled in. Alternatively, we might try |
johnc@1325 | 710 | // the full complexity of the technique used in "regular" precleaning. |
johnc@1325 | 711 | if (stop_point != NULL) { |
johnc@1325 | 712 | // The card might have gotten re-dirtied and re-enqueued while we |
johnc@1325 | 713 | // worked. (In fact, it's pretty likely.) |
johnc@1325 | 714 | if (*card_ptr != CardTableModRefBS::dirty_card_val()) { |
johnc@1325 | 715 | *card_ptr = CardTableModRefBS::dirty_card_val(); |
johnc@1325 | 716 | MutexLockerEx x(Shared_DirtyCardQ_lock, |
johnc@1325 | 717 | Mutex::_no_safepoint_check_flag); |
johnc@1325 | 718 | DirtyCardQueue* sdcq = |
johnc@1325 | 719 | JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); |
johnc@1325 | 720 | sdcq->enqueue(card_ptr); |
johnc@1325 | 721 | } |
johnc@1325 | 722 | } else { |
johnc@1325 | 723 | out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region()); |
johnc@1325 | 724 | _conc_refine_cards++; |
johnc@1325 | 725 | } |
johnc@1325 | 726 | } |
johnc@1325 | 727 | |
ysr@777 | 728 | void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { |
ysr@777 | 729 | // If the card is no longer dirty, nothing to do. |
ysr@777 | 730 | if (*card_ptr != CardTableModRefBS::dirty_card_val()) return; |
ysr@777 | 731 | |
ysr@777 | 732 | // Construct the region representing the card. |
ysr@777 | 733 | HeapWord* start = _ct_bs->addr_for(card_ptr); |
ysr@777 | 734 | // And find the region containing it. |
ysr@777 | 735 | HeapRegion* r = _g1->heap_region_containing(start); |
ysr@777 | 736 | if (r == NULL) { |
ysr@777 | 737 | guarantee(_g1->is_in_permanent(start), "Or else where?"); |
ysr@777 | 738 | return; // Not in the G1 heap (might be in perm, for example.) |
ysr@777 | 739 | } |
ysr@777 | 740 | // Why do we have to check here whether a card is on a young region, |
ysr@777 | 741 | // given that we dirty young regions and, as a result, the |
ysr@777 | 742 | // post-barrier is supposed to filter them out and never to enqueue |
ysr@777 | 743 | // them? When we allocate a new region as the "allocation region" we |
ysr@777 | 744 | // actually dirty its cards after we release the lock, since card |
ysr@777 | 745 | // dirtying while holding the lock was a performance bottleneck. So, |
ysr@777 | 746 | // as a result, it is possible for other threads to actually |
ysr@777 | 747 | // allocate objects in the region (after the acquire the lock) |
ysr@777 | 748 | // before all the cards on the region are dirtied. This is unlikely, |
ysr@777 | 749 | // and it doesn't happen often, but it can happen. So, the extra |
ysr@777 | 750 | // check below filters out those cards. |
iveresov@1072 | 751 | if (r->is_young()) { |
ysr@777 | 752 | return; |
ysr@777 | 753 | } |
ysr@777 | 754 | // While we are processing RSet buffers during the collection, we |
ysr@777 | 755 | // actually don't want to scan any cards on the collection set, |
ysr@777 | 756 | // since we don't want to update remebered sets with entries that |
ysr@777 | 757 | // point into the collection set, given that live objects from the |
ysr@777 | 758 | // collection set are about to move and such entries will be stale |
ysr@777 | 759 | // very soon. This change also deals with a reliability issue which |
ysr@777 | 760 | // involves scanning a card in the collection set and coming across |
ysr@777 | 761 | // an array that was being chunked and looking malformed. Note, |
ysr@777 | 762 | // however, that if evacuation fails, we have to scan any objects |
ysr@777 | 763 | // that were not moved and create any missing entries. |
ysr@777 | 764 | if (r->in_collection_set()) { |
ysr@777 | 765 | return; |
ysr@777 | 766 | } |
ysr@777 | 767 | |
johnc@1325 | 768 | // Should we defer processing the card? |
johnc@1325 | 769 | // |
johnc@1325 | 770 | // Previously the result from the insert_cache call would be |
johnc@1325 | 771 | // either card_ptr (implying that card_ptr was currently "cold"), |
johnc@1325 | 772 | // null (meaning we had inserted the card ptr into the "hot" |
johnc@1325 | 773 | // cache, which had some headroom), or a "hot" card ptr |
johnc@1325 | 774 | // extracted from the "hot" cache. |
johnc@1325 | 775 | // |
johnc@1325 | 776 | // Now that the _card_counts cache in the ConcurrentG1Refine |
johnc@1325 | 777 | // instance is an evicting hash table, the result we get back |
johnc@1325 | 778 | // could be from evicting the card ptr in an already occupied |
johnc@1325 | 779 | // bucket (in which case we have replaced the card ptr in the |
johnc@1325 | 780 | // bucket with card_ptr and "defer" is set to false). To avoid |
johnc@1325 | 781 | // having a data structure (updates to which would need a lock) |
johnc@1325 | 782 | // to hold these unprocessed dirty cards, we need to immediately |
johnc@1325 | 783 | // process card_ptr. The actions needed to be taken on return |
johnc@1325 | 784 | // from cache_insert are summarized in the following table: |
johnc@1325 | 785 | // |
johnc@1325 | 786 | // res defer action |
johnc@1325 | 787 | // -------------------------------------------------------------- |
johnc@1325 | 788 | // null false card evicted from _card_counts & replaced with |
johnc@1325 | 789 | // card_ptr; evicted ptr added to hot cache. |
johnc@1325 | 790 | // No need to process res; immediately process card_ptr |
johnc@1325 | 791 | // |
johnc@1325 | 792 | // null true card not evicted from _card_counts; card_ptr added |
johnc@1325 | 793 | // to hot cache. |
johnc@1325 | 794 | // Nothing to do. |
johnc@1325 | 795 | // |
johnc@1325 | 796 | // non-null false card evicted from _card_counts & replaced with |
johnc@1325 | 797 | // card_ptr; evicted ptr is currently "cold" or |
johnc@1325 | 798 | // caused an eviction from the hot cache. |
johnc@1325 | 799 | // Immediately process res; process card_ptr. |
johnc@1325 | 800 | // |
johnc@1325 | 801 | // non-null true card not evicted from _card_counts; card_ptr is |
johnc@1325 | 802 | // currently cold, or caused an eviction from hot |
johnc@1325 | 803 | // cache. |
johnc@1325 | 804 | // Immediately process res; no need to process card_ptr. |
johnc@1325 | 805 | |
johnc@1325 | 806 | jbyte* res = card_ptr; |
johnc@1325 | 807 | bool defer = false; |
ysr@777 | 808 | if (_cg1r->use_cache()) { |
johnc@1325 | 809 | jbyte* res = _cg1r->cache_insert(card_ptr, &defer); |
johnc@1325 | 810 | if (res != NULL && (res != card_ptr || defer)) { |
johnc@1325 | 811 | start = _ct_bs->addr_for(res); |
johnc@1325 | 812 | r = _g1->heap_region_containing(start); |
johnc@1325 | 813 | if (r == NULL) { |
johnc@1325 | 814 | assert(_g1->is_in_permanent(start), "Or else where?"); |
johnc@1325 | 815 | } else { |
johnc@1325 | 816 | guarantee(!r->is_young(), "It was evicted in the current minor cycle."); |
johnc@1325 | 817 | // Process card pointer we get back from the hot card cache |
johnc@1325 | 818 | concurrentRefineOneCard_impl(res, worker_i); |
johnc@1325 | 819 | } |
ysr@777 | 820 | } |
ysr@777 | 821 | } |
ysr@777 | 822 | |
johnc@1325 | 823 | if (!defer) { |
johnc@1325 | 824 | concurrentRefineOneCard_impl(card_ptr, worker_i); |
ysr@777 | 825 | } |
ysr@777 | 826 | } |
ysr@777 | 827 | |
ysr@777 | 828 | class HRRSStatsIter: public HeapRegionClosure { |
ysr@777 | 829 | size_t _occupied; |
ysr@777 | 830 | size_t _total_mem_sz; |
ysr@777 | 831 | size_t _max_mem_sz; |
ysr@777 | 832 | HeapRegion* _max_mem_sz_region; |
ysr@777 | 833 | public: |
ysr@777 | 834 | HRRSStatsIter() : |
ysr@777 | 835 | _occupied(0), |
ysr@777 | 836 | _total_mem_sz(0), |
ysr@777 | 837 | _max_mem_sz(0), |
ysr@777 | 838 | _max_mem_sz_region(NULL) |
ysr@777 | 839 | {} |
ysr@777 | 840 | |
ysr@777 | 841 | bool doHeapRegion(HeapRegion* r) { |
ysr@777 | 842 | if (r->continuesHumongous()) return false; |
ysr@777 | 843 | size_t mem_sz = r->rem_set()->mem_size(); |
ysr@777 | 844 | if (mem_sz > _max_mem_sz) { |
ysr@777 | 845 | _max_mem_sz = mem_sz; |
ysr@777 | 846 | _max_mem_sz_region = r; |
ysr@777 | 847 | } |
ysr@777 | 848 | _total_mem_sz += mem_sz; |
ysr@777 | 849 | size_t occ = r->rem_set()->occupied(); |
ysr@777 | 850 | _occupied += occ; |
ysr@777 | 851 | return false; |
ysr@777 | 852 | } |
ysr@777 | 853 | size_t total_mem_sz() { return _total_mem_sz; } |
ysr@777 | 854 | size_t max_mem_sz() { return _max_mem_sz; } |
ysr@777 | 855 | size_t occupied() { return _occupied; } |
ysr@777 | 856 | HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; } |
ysr@777 | 857 | }; |
ysr@777 | 858 | |
iveresov@1229 | 859 | class PrintRSThreadVTimeClosure : public ThreadClosure { |
iveresov@1229 | 860 | public: |
iveresov@1229 | 861 | virtual void do_thread(Thread *t) { |
iveresov@1229 | 862 | ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t; |
iveresov@1229 | 863 | gclog_or_tty->print(" %5.2f", crt->vtime_accum()); |
iveresov@1229 | 864 | } |
iveresov@1229 | 865 | }; |
iveresov@1229 | 866 | |
ysr@777 | 867 | void HRInto_G1RemSet::print_summary_info() { |
ysr@777 | 868 | G1CollectedHeap* g1 = G1CollectedHeap::heap(); |
ysr@777 | 869 | |
ysr@777 | 870 | #if CARD_REPEAT_HISTO |
ysr@777 | 871 | gclog_or_tty->print_cr("\nG1 card_repeat count histogram: "); |
ysr@777 | 872 | gclog_or_tty->print_cr(" # of repeats --> # of cards with that number."); |
ysr@777 | 873 | card_repeat_count.print_on(gclog_or_tty); |
ysr@777 | 874 | #endif |
ysr@777 | 875 | |
ysr@777 | 876 | if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) { |
ysr@777 | 877 | gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: "); |
ysr@777 | 878 | gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number."); |
ysr@777 | 879 | out_of_histo.print_on(gclog_or_tty); |
ysr@777 | 880 | } |
iveresov@1229 | 881 | gclog_or_tty->print_cr("\n Concurrent RS processed %d cards", |
iveresov@1229 | 882 | _conc_refine_cards); |
ysr@777 | 883 | DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); |
ysr@777 | 884 | jint tot_processed_buffers = |
ysr@777 | 885 | dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread(); |
ysr@777 | 886 | gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers); |
iveresov@1229 | 887 | gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.", |
ysr@777 | 888 | dcqs.processed_buffers_rs_thread(), |
ysr@777 | 889 | 100.0*(float)dcqs.processed_buffers_rs_thread()/ |
ysr@777 | 890 | (float)tot_processed_buffers); |
ysr@777 | 891 | gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.", |
ysr@777 | 892 | dcqs.processed_buffers_mut(), |
ysr@777 | 893 | 100.0*(float)dcqs.processed_buffers_mut()/ |
ysr@777 | 894 | (float)tot_processed_buffers); |
iveresov@1229 | 895 | gclog_or_tty->print_cr(" Conc RS threads times(s)"); |
iveresov@1229 | 896 | PrintRSThreadVTimeClosure p; |
iveresov@1229 | 897 | gclog_or_tty->print(" "); |
iveresov@1229 | 898 | g1->concurrent_g1_refine()->threads_do(&p); |
ysr@777 | 899 | gclog_or_tty->print_cr(""); |
iveresov@1229 | 900 | |
ysr@777 | 901 | if (G1UseHRIntoRS) { |
ysr@777 | 902 | HRRSStatsIter blk; |
ysr@777 | 903 | g1->heap_region_iterate(&blk); |
ysr@777 | 904 | gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." |
ysr@777 | 905 | " Max = " SIZE_FORMAT "K.", |
ysr@777 | 906 | blk.total_mem_sz()/K, blk.max_mem_sz()/K); |
ysr@777 | 907 | gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," |
ysr@777 | 908 | " free_lists = " SIZE_FORMAT "K.", |
ysr@777 | 909 | HeapRegionRemSet::static_mem_size()/K, |
ysr@777 | 910 | HeapRegionRemSet::fl_mem_size()/K); |
ysr@777 | 911 | gclog_or_tty->print_cr(" %d occupied cards represented.", |
ysr@777 | 912 | blk.occupied()); |
ysr@777 | 913 | gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" |
apetrusenko@1112 | 914 | ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", |
ysr@777 | 915 | blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), |
ysr@777 | 916 | (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, |
ysr@777 | 917 | (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); |
ysr@777 | 918 | gclog_or_tty->print_cr(" Did %d coarsenings.", |
ysr@777 | 919 | HeapRegionRemSet::n_coarsenings()); |
ysr@777 | 920 | |
ysr@777 | 921 | } |
ysr@777 | 922 | } |
ysr@777 | 923 | void HRInto_G1RemSet::prepare_for_verify() { |
iveresov@1072 | 924 | if (G1HRRSFlushLogBuffersOnVerify && |
iveresov@1072 | 925 | (VerifyBeforeGC || VerifyAfterGC) |
iveresov@1072 | 926 | && !_g1->full_collection()) { |
ysr@777 | 927 | cleanupHRRS(); |
ysr@777 | 928 | _g1->set_refine_cte_cl_concurrency(false); |
ysr@777 | 929 | if (SafepointSynchronize::is_at_safepoint()) { |
ysr@777 | 930 | DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); |
ysr@777 | 931 | dcqs.concatenate_logs(); |
ysr@777 | 932 | } |
ysr@777 | 933 | bool cg1r_use_cache = _cg1r->use_cache(); |
ysr@777 | 934 | _cg1r->set_use_cache(false); |
ysr@777 | 935 | updateRS(0); |
ysr@777 | 936 | _cg1r->set_use_cache(cg1r_use_cache); |
iveresov@1072 | 937 | |
iveresov@1072 | 938 | assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed"); |
ysr@777 | 939 | } |
ysr@777 | 940 | } |