Mon, 09 Aug 2010 05:41:05 -0700
6966222: G1: simplify TaskQueue overflow handling
Reviewed-by: tonyp, ysr
1 /*
2 * Copyright (c) 2001, 2010, 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 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_concurrentG1Refine.cpp.incl"
28 // Possible sizes for the card counts cache: odd primes that roughly double in size.
29 // (See jvmtiTagMap.cpp).
30 int ConcurrentG1Refine::_cc_cache_sizes[] = {
31 16381, 32771, 76831, 150001, 307261,
32 614563, 1228891, 2457733, 4915219, 9830479,
33 19660831, 39321619, 78643219, 157286461, -1
34 };
36 ConcurrentG1Refine::ConcurrentG1Refine() :
37 _card_counts(NULL), _card_epochs(NULL),
38 _n_card_counts(0), _max_n_card_counts(0),
39 _cache_size_index(0), _expand_card_counts(false),
40 _hot_cache(NULL),
41 _def_use_cache(false), _use_cache(false),
42 _n_periods(0),
43 _threads(NULL), _n_threads(0)
44 {
46 // Ergomonically select initial concurrent refinement parameters
47 if (FLAG_IS_DEFAULT(G1ConcRefinementGreenZone)) {
48 FLAG_SET_DEFAULT(G1ConcRefinementGreenZone, MAX2<int>(ParallelGCThreads, 1));
49 }
50 set_green_zone(G1ConcRefinementGreenZone);
52 if (FLAG_IS_DEFAULT(G1ConcRefinementYellowZone)) {
53 FLAG_SET_DEFAULT(G1ConcRefinementYellowZone, green_zone() * 3);
54 }
55 set_yellow_zone(MAX2<int>(G1ConcRefinementYellowZone, green_zone()));
57 if (FLAG_IS_DEFAULT(G1ConcRefinementRedZone)) {
58 FLAG_SET_DEFAULT(G1ConcRefinementRedZone, yellow_zone() * 2);
59 }
60 set_red_zone(MAX2<int>(G1ConcRefinementRedZone, yellow_zone()));
61 _n_worker_threads = thread_num();
62 // We need one extra thread to do the young gen rset size sampling.
63 _n_threads = _n_worker_threads + 1;
64 reset_threshold_step();
66 _threads = NEW_C_HEAP_ARRAY(ConcurrentG1RefineThread*, _n_threads);
67 int worker_id_offset = (int)DirtyCardQueueSet::num_par_ids();
68 ConcurrentG1RefineThread *next = NULL;
69 for (int i = _n_threads - 1; i >= 0; i--) {
70 ConcurrentG1RefineThread* t = new ConcurrentG1RefineThread(this, next, worker_id_offset, i);
71 assert(t != NULL, "Conc refine should have been created");
72 assert(t->cg1r() == this, "Conc refine thread should refer to this");
73 _threads[i] = t;
74 next = t;
75 }
76 }
78 void ConcurrentG1Refine::reset_threshold_step() {
79 if (FLAG_IS_DEFAULT(G1ConcRefinementThresholdStep)) {
80 _thread_threshold_step = (yellow_zone() - green_zone()) / (worker_thread_num() + 1);
81 } else {
82 _thread_threshold_step = G1ConcRefinementThresholdStep;
83 }
84 }
86 int ConcurrentG1Refine::thread_num() {
87 return MAX2<int>((G1ConcRefinementThreads > 0) ? G1ConcRefinementThreads : ParallelGCThreads, 1);
88 }
90 void ConcurrentG1Refine::init() {
91 if (G1ConcRSLogCacheSize > 0) {
92 _g1h = G1CollectedHeap::heap();
93 _max_n_card_counts =
94 (unsigned) (_g1h->g1_reserved_obj_bytes() >> CardTableModRefBS::card_shift);
96 size_t max_card_num = ((size_t)1 << (sizeof(unsigned)*BitsPerByte-1)) - 1;
97 guarantee(_max_n_card_counts < max_card_num, "card_num representation");
99 int desired = _max_n_card_counts / InitialCacheFraction;
100 for (_cache_size_index = 0;
101 _cc_cache_sizes[_cache_size_index] >= 0; _cache_size_index++) {
102 if (_cc_cache_sizes[_cache_size_index] >= desired) break;
103 }
104 _cache_size_index = MAX2(0, (_cache_size_index - 1));
106 int initial_size = _cc_cache_sizes[_cache_size_index];
107 if (initial_size < 0) initial_size = _max_n_card_counts;
109 // Make sure we don't go bigger than we will ever need
110 _n_card_counts = MIN2((unsigned) initial_size, _max_n_card_counts);
112 _card_counts = NEW_C_HEAP_ARRAY(CardCountCacheEntry, _n_card_counts);
113 _card_epochs = NEW_C_HEAP_ARRAY(CardEpochCacheEntry, _n_card_counts);
115 Copy::fill_to_bytes(&_card_counts[0],
116 _n_card_counts * sizeof(CardCountCacheEntry));
117 Copy::fill_to_bytes(&_card_epochs[0], _n_card_counts * sizeof(CardEpochCacheEntry));
119 ModRefBarrierSet* bs = _g1h->mr_bs();
120 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
121 _ct_bs = (CardTableModRefBS*)bs;
122 _ct_bot = _ct_bs->byte_for_const(_g1h->reserved_region().start());
124 _def_use_cache = true;
125 _use_cache = true;
126 _hot_cache_size = (1 << G1ConcRSLogCacheSize);
127 _hot_cache = NEW_C_HEAP_ARRAY(jbyte*, _hot_cache_size);
128 _n_hot = 0;
129 _hot_cache_idx = 0;
131 // For refining the cards in the hot cache in parallel
132 int n_workers = (ParallelGCThreads > 0 ?
133 _g1h->workers()->total_workers() : 1);
134 _hot_cache_par_chunk_size = MAX2(1, _hot_cache_size / n_workers);
135 _hot_cache_par_claimed_idx = 0;
136 }
137 }
139 void ConcurrentG1Refine::stop() {
140 if (_threads != NULL) {
141 for (int i = 0; i < _n_threads; i++) {
142 _threads[i]->stop();
143 }
144 }
145 }
147 void ConcurrentG1Refine::reinitialize_threads() {
148 reset_threshold_step();
149 if (_threads != NULL) {
150 for (int i = 0; i < _n_threads; i++) {
151 _threads[i]->initialize();
152 }
153 }
154 }
156 ConcurrentG1Refine::~ConcurrentG1Refine() {
157 if (G1ConcRSLogCacheSize > 0) {
158 assert(_card_counts != NULL, "Logic");
159 FREE_C_HEAP_ARRAY(CardCountCacheEntry, _card_counts);
160 assert(_card_epochs != NULL, "Logic");
161 FREE_C_HEAP_ARRAY(CardEpochCacheEntry, _card_epochs);
162 assert(_hot_cache != NULL, "Logic");
163 FREE_C_HEAP_ARRAY(jbyte*, _hot_cache);
164 }
165 if (_threads != NULL) {
166 for (int i = 0; i < _n_threads; i++) {
167 delete _threads[i];
168 }
169 FREE_C_HEAP_ARRAY(ConcurrentG1RefineThread*, _threads);
170 }
171 }
173 void ConcurrentG1Refine::threads_do(ThreadClosure *tc) {
174 if (_threads != NULL) {
175 for (int i = 0; i < _n_threads; i++) {
176 tc->do_thread(_threads[i]);
177 }
178 }
179 }
181 bool ConcurrentG1Refine::is_young_card(jbyte* card_ptr) {
182 HeapWord* start = _ct_bs->addr_for(card_ptr);
183 HeapRegion* r = _g1h->heap_region_containing(start);
184 if (r != NULL && r->is_young()) {
185 return true;
186 }
187 // This card is not associated with a heap region
188 // so can't be young.
189 return false;
190 }
192 jbyte* ConcurrentG1Refine::add_card_count(jbyte* card_ptr, int* count, bool* defer) {
193 unsigned new_card_num = ptr_2_card_num(card_ptr);
194 unsigned bucket = hash(new_card_num);
195 assert(0 <= bucket && bucket < _n_card_counts, "Bounds");
197 CardCountCacheEntry* count_ptr = &_card_counts[bucket];
198 CardEpochCacheEntry* epoch_ptr = &_card_epochs[bucket];
200 // We have to construct a new entry if we haven't updated the counts
201 // during the current period, or if the count was updated for a
202 // different card number.
203 unsigned int new_epoch = (unsigned int) _n_periods;
204 julong new_epoch_entry = make_epoch_entry(new_card_num, new_epoch);
206 while (true) {
207 // Fetch the previous epoch value
208 julong prev_epoch_entry = epoch_ptr->_value;
209 julong cas_res;
211 if (extract_epoch(prev_epoch_entry) != new_epoch) {
212 // This entry has not yet been updated during this period.
213 // Note: we update the epoch value atomically to ensure
214 // that there is only one winner that updates the cached
215 // card_ptr value even though all the refine threads share
216 // the same epoch value.
218 cas_res = (julong) Atomic::cmpxchg((jlong) new_epoch_entry,
219 (volatile jlong*)&epoch_ptr->_value,
220 (jlong) prev_epoch_entry);
222 if (cas_res == prev_epoch_entry) {
223 // We have successfully won the race to update the
224 // epoch and card_num value. Make it look like the
225 // count and eviction count were previously cleared.
226 count_ptr->_count = 1;
227 count_ptr->_evict_count = 0;
228 *count = 0;
229 // We can defer the processing of card_ptr
230 *defer = true;
231 return card_ptr;
232 }
233 // We did not win the race to update the epoch field, so some other
234 // thread must have done it. The value that gets returned by CAS
235 // should be the new epoch value.
236 assert(extract_epoch(cas_res) == new_epoch, "unexpected epoch");
237 // We could 'continue' here or just re-read the previous epoch value
238 prev_epoch_entry = epoch_ptr->_value;
239 }
241 // The epoch entry for card_ptr has been updated during this period.
242 unsigned old_card_num = extract_card_num(prev_epoch_entry);
244 // The card count that will be returned to caller
245 *count = count_ptr->_count;
247 // Are we updating the count for the same card?
248 if (new_card_num == old_card_num) {
249 // Same card - just update the count. We could have more than one
250 // thread racing to update count for the current card. It should be
251 // OK not to use a CAS as the only penalty should be some missed
252 // increments of the count which delays identifying the card as "hot".
254 if (*count < max_jubyte) count_ptr->_count++;
255 // We can defer the processing of card_ptr
256 *defer = true;
257 return card_ptr;
258 }
260 // Different card - evict old card info
261 if (count_ptr->_evict_count < max_jubyte) count_ptr->_evict_count++;
262 if (count_ptr->_evict_count > G1CardCountCacheExpandThreshold) {
263 // Trigger a resize the next time we clear
264 _expand_card_counts = true;
265 }
267 cas_res = (julong) Atomic::cmpxchg((jlong) new_epoch_entry,
268 (volatile jlong*)&epoch_ptr->_value,
269 (jlong) prev_epoch_entry);
271 if (cas_res == prev_epoch_entry) {
272 // We successfully updated the card num value in the epoch entry
273 count_ptr->_count = 0; // initialize counter for new card num
274 jbyte* old_card_ptr = card_num_2_ptr(old_card_num);
276 // Even though the region containg the card at old_card_num was not
277 // in the young list when old_card_num was recorded in the epoch
278 // cache it could have been added to the free list and subsequently
279 // added to the young list in the intervening time. See CR 6817995.
280 // We do not deal with this case here - it will be handled in
281 // HeapRegion::oops_on_card_seq_iterate_careful after it has been
282 // determined that the region containing the card has been allocated
283 // to, and it's safe to check the young type of the region.
285 // We do not want to defer processing of card_ptr in this case
286 // (we need to refine old_card_ptr and card_ptr)
287 *defer = false;
288 return old_card_ptr;
289 }
290 // Someone else beat us - try again.
291 }
292 }
294 jbyte* ConcurrentG1Refine::cache_insert(jbyte* card_ptr, bool* defer) {
295 int count;
296 jbyte* cached_ptr = add_card_count(card_ptr, &count, defer);
297 assert(cached_ptr != NULL, "bad cached card ptr");
299 // We've just inserted a card pointer into the card count cache
300 // and got back the card that we just inserted or (evicted) the
301 // previous contents of that count slot.
303 // The card we got back could be in a young region. When the
304 // returned card (if evicted) was originally inserted, we had
305 // determined that its containing region was not young. However
306 // it is possible for the region to be freed during a cleanup
307 // pause, then reallocated and tagged as young which will result
308 // in the returned card residing in a young region.
309 //
310 // We do not deal with this case here - the change from non-young
311 // to young could be observed at any time - it will be handled in
312 // HeapRegion::oops_on_card_seq_iterate_careful after it has been
313 // determined that the region containing the card has been allocated
314 // to.
316 // The card pointer we obtained from card count cache is not hot
317 // so do not store it in the cache; return it for immediate
318 // refining.
319 if (count < G1ConcRSHotCardLimit) {
320 return cached_ptr;
321 }
323 // Otherwise, the pointer we got from the _card_counts cache is hot.
324 jbyte* res = NULL;
325 MutexLockerEx x(HotCardCache_lock, Mutex::_no_safepoint_check_flag);
326 if (_n_hot == _hot_cache_size) {
327 res = _hot_cache[_hot_cache_idx];
328 _n_hot--;
329 }
330 // Now _n_hot < _hot_cache_size, and we can insert at _hot_cache_idx.
331 _hot_cache[_hot_cache_idx] = cached_ptr;
332 _hot_cache_idx++;
333 if (_hot_cache_idx == _hot_cache_size) _hot_cache_idx = 0;
334 _n_hot++;
336 // The card obtained from the hot card cache could be in a young
337 // region. See above on how this can happen.
339 return res;
340 }
342 void ConcurrentG1Refine::clean_up_cache(int worker_i,
343 G1RemSet* g1rs,
344 DirtyCardQueue* into_cset_dcq) {
345 assert(!use_cache(), "cache should be disabled");
346 int start_idx;
348 while ((start_idx = _hot_cache_par_claimed_idx) < _n_hot) { // read once
349 int end_idx = start_idx + _hot_cache_par_chunk_size;
351 if (start_idx ==
352 Atomic::cmpxchg(end_idx, &_hot_cache_par_claimed_idx, start_idx)) {
353 // The current worker has successfully claimed the chunk [start_idx..end_idx)
354 end_idx = MIN2(end_idx, _n_hot);
355 for (int i = start_idx; i < end_idx; i++) {
356 jbyte* entry = _hot_cache[i];
357 if (entry != NULL) {
358 if (g1rs->concurrentRefineOneCard(entry, worker_i, true)) {
359 // 'entry' contains references that point into the current
360 // collection set. We need to record 'entry' in the DCQS
361 // that's used for that purpose.
362 //
363 // The only time we care about recording cards that contain
364 // references that point into the collection set is during
365 // RSet updating while within an evacuation pause.
366 // In this case worker_i should be the id of a GC worker thread
367 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
368 assert(worker_i < (int) DirtyCardQueueSet::num_par_ids(), "incorrect worker id");
369 into_cset_dcq->enqueue(entry);
370 }
371 }
372 }
373 }
374 }
375 }
377 void ConcurrentG1Refine::expand_card_count_cache() {
378 if (_n_card_counts < _max_n_card_counts) {
379 int new_idx = _cache_size_index+1;
380 int new_size = _cc_cache_sizes[new_idx];
381 if (new_size < 0) new_size = _max_n_card_counts;
383 // Make sure we don't go bigger than we will ever need
384 new_size = MIN2((unsigned) new_size, _max_n_card_counts);
386 // Expand the card count and card epoch tables
387 if (new_size > (int)_n_card_counts) {
388 // We can just free and allocate a new array as we're
389 // not interested in preserving the contents
390 assert(_card_counts != NULL, "Logic!");
391 assert(_card_epochs != NULL, "Logic!");
392 FREE_C_HEAP_ARRAY(CardCountCacheEntry, _card_counts);
393 FREE_C_HEAP_ARRAY(CardEpochCacheEntry, _card_epochs);
394 _n_card_counts = new_size;
395 _card_counts = NEW_C_HEAP_ARRAY(CardCountCacheEntry, _n_card_counts);
396 _card_epochs = NEW_C_HEAP_ARRAY(CardEpochCacheEntry, _n_card_counts);
397 _cache_size_index = new_idx;
398 }
399 }
400 }
402 void ConcurrentG1Refine::clear_and_record_card_counts() {
403 if (G1ConcRSLogCacheSize == 0) return;
405 #ifndef PRODUCT
406 double start = os::elapsedTime();
407 #endif
409 if (_expand_card_counts) {
410 expand_card_count_cache();
411 _expand_card_counts = false;
412 // Only need to clear the epochs.
413 Copy::fill_to_bytes(&_card_epochs[0], _n_card_counts * sizeof(CardEpochCacheEntry));
414 }
416 int this_epoch = (int) _n_periods;
417 assert((this_epoch+1) <= max_jint, "to many periods");
418 // Update epoch
419 _n_periods++;
421 #ifndef PRODUCT
422 double elapsed = os::elapsedTime() - start;
423 _g1h->g1_policy()->record_cc_clear_time(elapsed * 1000.0);
424 #endif
425 }
427 void ConcurrentG1Refine::print_worker_threads_on(outputStream* st) const {
428 for (int i = 0; i < _n_threads; ++i) {
429 _threads[i]->print_on(st);
430 st->cr();
431 }
432 }