|
1 /* |
|
2 * Copyright (c) 2001, 2014, 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 #ifndef SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP |
|
26 #define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP |
|
27 |
|
28 #include "gc_implementation/g1/concurrentMark.hpp" |
|
29 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp" |
|
30 |
|
31 // Utility routine to set an exclusive range of cards on the given |
|
32 // card liveness bitmap |
|
33 inline void ConcurrentMark::set_card_bitmap_range(BitMap* card_bm, |
|
34 BitMap::idx_t start_idx, |
|
35 BitMap::idx_t end_idx, |
|
36 bool is_par) { |
|
37 |
|
38 // Set the exclusive bit range [start_idx, end_idx). |
|
39 assert((end_idx - start_idx) > 0, "at least one card"); |
|
40 assert(end_idx <= card_bm->size(), "sanity"); |
|
41 |
|
42 // Silently clip the end index |
|
43 end_idx = MIN2(end_idx, card_bm->size()); |
|
44 |
|
45 // For small ranges use a simple loop; otherwise use set_range or |
|
46 // use par_at_put_range (if parallel). The range is made up of the |
|
47 // cards that are spanned by an object/mem region so 8 cards will |
|
48 // allow up to object sizes up to 4K to be handled using the loop. |
|
49 if ((end_idx - start_idx) <= 8) { |
|
50 for (BitMap::idx_t i = start_idx; i < end_idx; i += 1) { |
|
51 if (is_par) { |
|
52 card_bm->par_set_bit(i); |
|
53 } else { |
|
54 card_bm->set_bit(i); |
|
55 } |
|
56 } |
|
57 } else { |
|
58 // Note BitMap::par_at_put_range() and BitMap::set_range() are exclusive. |
|
59 if (is_par) { |
|
60 card_bm->par_at_put_range(start_idx, end_idx, true); |
|
61 } else { |
|
62 card_bm->set_range(start_idx, end_idx); |
|
63 } |
|
64 } |
|
65 } |
|
66 |
|
67 // Returns the index in the liveness accounting card bitmap |
|
68 // for the given address |
|
69 inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) { |
|
70 // Below, the term "card num" means the result of shifting an address |
|
71 // by the card shift -- address 0 corresponds to card number 0. One |
|
72 // must subtract the card num of the bottom of the heap to obtain a |
|
73 // card table index. |
|
74 intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift); |
|
75 return card_num - heap_bottom_card_num(); |
|
76 } |
|
77 |
|
78 // Counts the given memory region in the given task/worker |
|
79 // counting data structures. |
|
80 inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr, |
|
81 size_t* marked_bytes_array, |
|
82 BitMap* task_card_bm) { |
|
83 G1CollectedHeap* g1h = _g1h; |
|
84 CardTableModRefBS* ct_bs = g1h->g1_barrier_set(); |
|
85 |
|
86 HeapWord* start = mr.start(); |
|
87 HeapWord* end = mr.end(); |
|
88 size_t region_size_bytes = mr.byte_size(); |
|
89 uint index = hr->hrs_index(); |
|
90 |
|
91 assert(!hr->continuesHumongous(), "should not be HC region"); |
|
92 assert(hr == g1h->heap_region_containing(start), "sanity"); |
|
93 assert(hr == g1h->heap_region_containing(mr.last()), "sanity"); |
|
94 assert(marked_bytes_array != NULL, "pre-condition"); |
|
95 assert(task_card_bm != NULL, "pre-condition"); |
|
96 |
|
97 // Add to the task local marked bytes for this region. |
|
98 marked_bytes_array[index] += region_size_bytes; |
|
99 |
|
100 BitMap::idx_t start_idx = card_bitmap_index_for(start); |
|
101 BitMap::idx_t end_idx = card_bitmap_index_for(end); |
|
102 |
|
103 // Note: if we're looking at the last region in heap - end |
|
104 // could be actually just beyond the end of the heap; end_idx |
|
105 // will then correspond to a (non-existent) card that is also |
|
106 // just beyond the heap. |
|
107 if (g1h->is_in_g1_reserved(end) && !ct_bs->is_card_aligned(end)) { |
|
108 // end of region is not card aligned - incremement to cover |
|
109 // all the cards spanned by the region. |
|
110 end_idx += 1; |
|
111 } |
|
112 // The card bitmap is task/worker specific => no need to use |
|
113 // the 'par' BitMap routines. |
|
114 // Set bits in the exclusive bit range [start_idx, end_idx). |
|
115 set_card_bitmap_range(task_card_bm, start_idx, end_idx, false /* is_par */); |
|
116 } |
|
117 |
|
118 // Counts the given memory region in the task/worker counting |
|
119 // data structures for the given worker id. |
|
120 inline void ConcurrentMark::count_region(MemRegion mr, |
|
121 HeapRegion* hr, |
|
122 uint worker_id) { |
|
123 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); |
|
124 BitMap* task_card_bm = count_card_bitmap_for(worker_id); |
|
125 count_region(mr, hr, marked_bytes_array, task_card_bm); |
|
126 } |
|
127 |
|
128 // Counts the given memory region, which may be a single object, in the |
|
129 // task/worker counting data structures for the given worker id. |
|
130 inline void ConcurrentMark::count_region(MemRegion mr, uint worker_id) { |
|
131 HeapWord* addr = mr.start(); |
|
132 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); |
|
133 count_region(mr, hr, worker_id); |
|
134 } |
|
135 |
|
136 // Counts the given object in the given task/worker counting data structures. |
|
137 inline void ConcurrentMark::count_object(oop obj, |
|
138 HeapRegion* hr, |
|
139 size_t* marked_bytes_array, |
|
140 BitMap* task_card_bm) { |
|
141 MemRegion mr((HeapWord*)obj, obj->size()); |
|
142 count_region(mr, hr, marked_bytes_array, task_card_bm); |
|
143 } |
|
144 |
|
145 // Counts the given object in the task/worker counting data |
|
146 // structures for the given worker id. |
|
147 inline void ConcurrentMark::count_object(oop obj, |
|
148 HeapRegion* hr, |
|
149 uint worker_id) { |
|
150 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); |
|
151 BitMap* task_card_bm = count_card_bitmap_for(worker_id); |
|
152 HeapWord* addr = (HeapWord*) obj; |
|
153 count_object(obj, hr, marked_bytes_array, task_card_bm); |
|
154 } |
|
155 |
|
156 // Attempts to mark the given object and, if successful, counts |
|
157 // the object in the given task/worker counting structures. |
|
158 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
159 HeapRegion* hr, |
|
160 size_t* marked_bytes_array, |
|
161 BitMap* task_card_bm) { |
|
162 HeapWord* addr = (HeapWord*)obj; |
|
163 if (_nextMarkBitMap->parMark(addr)) { |
|
164 // Update the task specific count data for the object. |
|
165 count_object(obj, hr, marked_bytes_array, task_card_bm); |
|
166 return true; |
|
167 } |
|
168 return false; |
|
169 } |
|
170 |
|
171 // Attempts to mark the given object and, if successful, counts |
|
172 // the object in the task/worker counting structures for the |
|
173 // given worker id. |
|
174 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
175 size_t word_size, |
|
176 HeapRegion* hr, |
|
177 uint worker_id) { |
|
178 HeapWord* addr = (HeapWord*)obj; |
|
179 if (_nextMarkBitMap->parMark(addr)) { |
|
180 MemRegion mr(addr, word_size); |
|
181 count_region(mr, hr, worker_id); |
|
182 return true; |
|
183 } |
|
184 return false; |
|
185 } |
|
186 |
|
187 // Attempts to mark the given object and, if successful, counts |
|
188 // the object in the task/worker counting structures for the |
|
189 // given worker id. |
|
190 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
191 HeapRegion* hr, |
|
192 uint worker_id) { |
|
193 HeapWord* addr = (HeapWord*)obj; |
|
194 if (_nextMarkBitMap->parMark(addr)) { |
|
195 // Update the task specific count data for the object. |
|
196 count_object(obj, hr, worker_id); |
|
197 return true; |
|
198 } |
|
199 return false; |
|
200 } |
|
201 |
|
202 // As above - but we don't know the heap region containing the |
|
203 // object and so have to supply it. |
|
204 inline bool ConcurrentMark::par_mark_and_count(oop obj, uint worker_id) { |
|
205 HeapWord* addr = (HeapWord*)obj; |
|
206 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); |
|
207 return par_mark_and_count(obj, hr, worker_id); |
|
208 } |
|
209 |
|
210 // Similar to the above routine but we already know the size, in words, of |
|
211 // the object that we wish to mark/count |
|
212 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
213 size_t word_size, |
|
214 uint worker_id) { |
|
215 HeapWord* addr = (HeapWord*)obj; |
|
216 if (_nextMarkBitMap->parMark(addr)) { |
|
217 // Update the task specific count data for the object. |
|
218 MemRegion mr(addr, word_size); |
|
219 count_region(mr, worker_id); |
|
220 return true; |
|
221 } |
|
222 return false; |
|
223 } |
|
224 |
|
225 // Unconditionally mark the given object, and unconditinally count |
|
226 // the object in the counting structures for worker id 0. |
|
227 // Should *not* be called from parallel code. |
|
228 inline bool ConcurrentMark::mark_and_count(oop obj, HeapRegion* hr) { |
|
229 HeapWord* addr = (HeapWord*)obj; |
|
230 _nextMarkBitMap->mark(addr); |
|
231 // Update the task specific count data for the object. |
|
232 count_object(obj, hr, 0 /* worker_id */); |
|
233 return true; |
|
234 } |
|
235 |
|
236 // As above - but we don't have the heap region containing the |
|
237 // object, so we have to supply it. |
|
238 inline bool ConcurrentMark::mark_and_count(oop obj) { |
|
239 HeapWord* addr = (HeapWord*)obj; |
|
240 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); |
|
241 return mark_and_count(obj, hr); |
|
242 } |
|
243 |
|
244 inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) { |
|
245 HeapWord* start_addr = MAX2(startWord(), mr.start()); |
|
246 HeapWord* end_addr = MIN2(endWord(), mr.end()); |
|
247 |
|
248 if (end_addr > start_addr) { |
|
249 // Right-open interval [start-offset, end-offset). |
|
250 BitMap::idx_t start_offset = heapWordToOffset(start_addr); |
|
251 BitMap::idx_t end_offset = heapWordToOffset(end_addr); |
|
252 |
|
253 start_offset = _bm.get_next_one_offset(start_offset, end_offset); |
|
254 while (start_offset < end_offset) { |
|
255 if (!cl->do_bit(start_offset)) { |
|
256 return false; |
|
257 } |
|
258 HeapWord* next_addr = MIN2(nextObject(offsetToHeapWord(start_offset)), end_addr); |
|
259 BitMap::idx_t next_offset = heapWordToOffset(next_addr); |
|
260 start_offset = _bm.get_next_one_offset(next_offset, end_offset); |
|
261 } |
|
262 } |
|
263 return true; |
|
264 } |
|
265 |
|
266 inline bool CMBitMapRO::iterate(BitMapClosure* cl) { |
|
267 MemRegion mr(startWord(), sizeInWords()); |
|
268 return iterate(cl, mr); |
|
269 } |
|
270 |
|
271 inline void CMTask::push(oop obj) { |
|
272 HeapWord* objAddr = (HeapWord*) obj; |
|
273 assert(_g1h->is_in_g1_reserved(objAddr), "invariant"); |
|
274 assert(!_g1h->is_on_master_free_list( |
|
275 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant"); |
|
276 assert(!_g1h->is_obj_ill(obj), "invariant"); |
|
277 assert(_nextMarkBitMap->isMarked(objAddr), "invariant"); |
|
278 |
|
279 if (_cm->verbose_high()) { |
|
280 gclog_or_tty->print_cr("[%u] pushing " PTR_FORMAT, _worker_id, p2i((void*) obj)); |
|
281 } |
|
282 |
|
283 if (!_task_queue->push(obj)) { |
|
284 // The local task queue looks full. We need to push some entries |
|
285 // to the global stack. |
|
286 |
|
287 if (_cm->verbose_medium()) { |
|
288 gclog_or_tty->print_cr("[%u] task queue overflow, " |
|
289 "moving entries to the global stack", |
|
290 _worker_id); |
|
291 } |
|
292 move_entries_to_global_stack(); |
|
293 |
|
294 // this should succeed since, even if we overflow the global |
|
295 // stack, we should have definitely removed some entries from the |
|
296 // local queue. So, there must be space on it. |
|
297 bool success = _task_queue->push(obj); |
|
298 assert(success, "invariant"); |
|
299 } |
|
300 |
|
301 statsOnly( int tmp_size = _task_queue->size(); |
|
302 if (tmp_size > _local_max_size) { |
|
303 _local_max_size = tmp_size; |
|
304 } |
|
305 ++_local_pushes ); |
|
306 } |
|
307 |
|
308 // This determines whether the method below will check both the local |
|
309 // and global fingers when determining whether to push on the stack a |
|
310 // gray object (value 1) or whether it will only check the global one |
|
311 // (value 0). The tradeoffs are that the former will be a bit more |
|
312 // accurate and possibly push less on the stack, but it might also be |
|
313 // a little bit slower. |
|
314 |
|
315 #define _CHECK_BOTH_FINGERS_ 1 |
|
316 |
|
317 inline void CMTask::deal_with_reference(oop obj) { |
|
318 if (_cm->verbose_high()) { |
|
319 gclog_or_tty->print_cr("[%u] we're dealing with reference = "PTR_FORMAT, |
|
320 _worker_id, p2i((void*) obj)); |
|
321 } |
|
322 |
|
323 ++_refs_reached; |
|
324 |
|
325 HeapWord* objAddr = (HeapWord*) obj; |
|
326 assert(obj->is_oop_or_null(true /* ignore mark word */), "Error"); |
|
327 if (_g1h->is_in_g1_reserved(objAddr)) { |
|
328 assert(obj != NULL, "null check is implicit"); |
|
329 if (!_nextMarkBitMap->isMarked(objAddr)) { |
|
330 // Only get the containing region if the object is not marked on the |
|
331 // bitmap (otherwise, it's a waste of time since we won't do |
|
332 // anything with it). |
|
333 HeapRegion* hr = _g1h->heap_region_containing_raw(obj); |
|
334 if (!hr->obj_allocated_since_next_marking(obj)) { |
|
335 if (_cm->verbose_high()) { |
|
336 gclog_or_tty->print_cr("[%u] "PTR_FORMAT" is not considered marked", |
|
337 _worker_id, p2i((void*) obj)); |
|
338 } |
|
339 |
|
340 // we need to mark it first |
|
341 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) { |
|
342 // No OrderAccess:store_load() is needed. It is implicit in the |
|
343 // CAS done in CMBitMap::parMark() call in the routine above. |
|
344 HeapWord* global_finger = _cm->finger(); |
|
345 |
|
346 #if _CHECK_BOTH_FINGERS_ |
|
347 // we will check both the local and global fingers |
|
348 |
|
349 if (_finger != NULL && objAddr < _finger) { |
|
350 if (_cm->verbose_high()) { |
|
351 gclog_or_tty->print_cr("[%u] below the local finger ("PTR_FORMAT"), " |
|
352 "pushing it", _worker_id, p2i(_finger)); |
|
353 } |
|
354 push(obj); |
|
355 } else if (_curr_region != NULL && objAddr < _region_limit) { |
|
356 // do nothing |
|
357 } else if (objAddr < global_finger) { |
|
358 // Notice that the global finger might be moving forward |
|
359 // concurrently. This is not a problem. In the worst case, we |
|
360 // mark the object while it is above the global finger and, by |
|
361 // the time we read the global finger, it has moved forward |
|
362 // passed this object. In this case, the object will probably |
|
363 // be visited when a task is scanning the region and will also |
|
364 // be pushed on the stack. So, some duplicate work, but no |
|
365 // correctness problems. |
|
366 |
|
367 if (_cm->verbose_high()) { |
|
368 gclog_or_tty->print_cr("[%u] below the global finger " |
|
369 "("PTR_FORMAT"), pushing it", |
|
370 _worker_id, p2i(global_finger)); |
|
371 } |
|
372 push(obj); |
|
373 } else { |
|
374 // do nothing |
|
375 } |
|
376 #else // _CHECK_BOTH_FINGERS_ |
|
377 // we will only check the global finger |
|
378 |
|
379 if (objAddr < global_finger) { |
|
380 // see long comment above |
|
381 |
|
382 if (_cm->verbose_high()) { |
|
383 gclog_or_tty->print_cr("[%u] below the global finger " |
|
384 "("PTR_FORMAT"), pushing it", |
|
385 _worker_id, p2i(global_finger)); |
|
386 } |
|
387 push(obj); |
|
388 } |
|
389 #endif // _CHECK_BOTH_FINGERS_ |
|
390 } |
|
391 } |
|
392 } |
|
393 } |
|
394 } |
|
395 |
|
396 inline void ConcurrentMark::markPrev(oop p) { |
|
397 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity"); |
|
398 // Note we are overriding the read-only view of the prev map here, via |
|
399 // the cast. |
|
400 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p); |
|
401 } |
|
402 |
|
403 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, |
|
404 uint worker_id, HeapRegion* hr) { |
|
405 assert(obj != NULL, "pre-condition"); |
|
406 HeapWord* addr = (HeapWord*) obj; |
|
407 if (hr == NULL) { |
|
408 hr = _g1h->heap_region_containing_raw(addr); |
|
409 } else { |
|
410 assert(hr->is_in(addr), "pre-condition"); |
|
411 } |
|
412 assert(hr != NULL, "sanity"); |
|
413 // Given that we're looking for a region that contains an object |
|
414 // header it's impossible to get back a HC region. |
|
415 assert(!hr->continuesHumongous(), "sanity"); |
|
416 |
|
417 // We cannot assert that word_size == obj->size() given that obj |
|
418 // might not be in a consistent state (another thread might be in |
|
419 // the process of copying it). So the best thing we can do is to |
|
420 // assert that word_size is under an upper bound which is its |
|
421 // containing region's capacity. |
|
422 assert(word_size * HeapWordSize <= hr->capacity(), |
|
423 err_msg("size: "SIZE_FORMAT" capacity: "SIZE_FORMAT" "HR_FORMAT, |
|
424 word_size * HeapWordSize, hr->capacity(), |
|
425 HR_FORMAT_PARAMS(hr))); |
|
426 |
|
427 if (addr < hr->next_top_at_mark_start()) { |
|
428 if (!_nextMarkBitMap->isMarked(addr)) { |
|
429 par_mark_and_count(obj, word_size, hr, worker_id); |
|
430 } |
|
431 } |
|
432 } |
|
433 |
|
434 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP |