1 /* |
|
2 * Copyright (c) 2001, 2015, 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_G1_CONCURRENTMARK_INLINE_HPP |
|
26 #define SHARE_VM_GC_G1_CONCURRENTMARK_INLINE_HPP |
|
27 |
|
28 #include "gc/g1/concurrentMark.hpp" |
|
29 #include "gc/g1/g1CollectedHeap.inline.hpp" |
|
30 #include "gc/shared/taskqueue.inline.hpp" |
|
31 |
|
32 // Utility routine to set an exclusive range of cards on the given |
|
33 // card liveness bitmap |
|
34 inline void ConcurrentMark::set_card_bitmap_range(BitMap* card_bm, |
|
35 BitMap::idx_t start_idx, |
|
36 BitMap::idx_t end_idx, |
|
37 bool is_par) { |
|
38 |
|
39 // Set the exclusive bit range [start_idx, end_idx). |
|
40 assert((end_idx - start_idx) > 0, "at least one card"); |
|
41 assert(end_idx <= card_bm->size(), "sanity"); |
|
42 |
|
43 // Silently clip the end index |
|
44 end_idx = MIN2(end_idx, card_bm->size()); |
|
45 |
|
46 // For small ranges use a simple loop; otherwise use set_range or |
|
47 // use par_at_put_range (if parallel). The range is made up of the |
|
48 // cards that are spanned by an object/mem region so 8 cards will |
|
49 // allow up to object sizes up to 4K to be handled using the loop. |
|
50 if ((end_idx - start_idx) <= 8) { |
|
51 for (BitMap::idx_t i = start_idx; i < end_idx; i += 1) { |
|
52 if (is_par) { |
|
53 card_bm->par_set_bit(i); |
|
54 } else { |
|
55 card_bm->set_bit(i); |
|
56 } |
|
57 } |
|
58 } else { |
|
59 // Note BitMap::par_at_put_range() and BitMap::set_range() are exclusive. |
|
60 if (is_par) { |
|
61 card_bm->par_at_put_range(start_idx, end_idx, true); |
|
62 } else { |
|
63 card_bm->set_range(start_idx, end_idx); |
|
64 } |
|
65 } |
|
66 } |
|
67 |
|
68 // Returns the index in the liveness accounting card bitmap |
|
69 // for the given address |
|
70 inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) { |
|
71 // Below, the term "card num" means the result of shifting an address |
|
72 // by the card shift -- address 0 corresponds to card number 0. One |
|
73 // must subtract the card num of the bottom of the heap to obtain a |
|
74 // card table index. |
|
75 intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift); |
|
76 return card_num - heap_bottom_card_num(); |
|
77 } |
|
78 |
|
79 // Counts the given memory region in the given task/worker |
|
80 // counting data structures. |
|
81 inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr, |
|
82 size_t* marked_bytes_array, |
|
83 BitMap* task_card_bm) { |
|
84 G1CollectedHeap* g1h = _g1h; |
|
85 CardTableModRefBS* ct_bs = g1h->g1_barrier_set(); |
|
86 |
|
87 HeapWord* start = mr.start(); |
|
88 HeapWord* end = mr.end(); |
|
89 size_t region_size_bytes = mr.byte_size(); |
|
90 uint index = hr->hrm_index(); |
|
91 |
|
92 assert(hr == g1h->heap_region_containing(start), "sanity"); |
|
93 assert(marked_bytes_array != NULL, "pre-condition"); |
|
94 assert(task_card_bm != NULL, "pre-condition"); |
|
95 |
|
96 // Add to the task local marked bytes for this region. |
|
97 marked_bytes_array[index] += region_size_bytes; |
|
98 |
|
99 BitMap::idx_t start_idx = card_bitmap_index_for(start); |
|
100 BitMap::idx_t end_idx = card_bitmap_index_for(end); |
|
101 |
|
102 // Note: if we're looking at the last region in heap - end |
|
103 // could be actually just beyond the end of the heap; end_idx |
|
104 // will then correspond to a (non-existent) card that is also |
|
105 // just beyond the heap. |
|
106 if (g1h->is_in_g1_reserved(end) && !ct_bs->is_card_aligned(end)) { |
|
107 // end of region is not card aligned - increment to cover |
|
108 // all the cards spanned by the region. |
|
109 end_idx += 1; |
|
110 } |
|
111 // The card bitmap is task/worker specific => no need to use |
|
112 // the 'par' BitMap routines. |
|
113 // Set bits in the exclusive bit range [start_idx, end_idx). |
|
114 set_card_bitmap_range(task_card_bm, start_idx, end_idx, false /* is_par */); |
|
115 } |
|
116 |
|
117 // Counts the given object in the given task/worker counting data structures. |
|
118 inline void ConcurrentMark::count_object(oop obj, |
|
119 HeapRegion* hr, |
|
120 size_t* marked_bytes_array, |
|
121 BitMap* task_card_bm, |
|
122 size_t word_size) { |
|
123 assert(!hr->is_continues_humongous(), "Cannot enter count_object with continues humongous"); |
|
124 if (!hr->is_starts_humongous()) { |
|
125 MemRegion mr((HeapWord*)obj, word_size); |
|
126 count_region(mr, hr, marked_bytes_array, task_card_bm); |
|
127 } else { |
|
128 do { |
|
129 MemRegion mr(hr->bottom(), hr->top()); |
|
130 count_region(mr, hr, marked_bytes_array, task_card_bm); |
|
131 hr = _g1h->next_region_in_humongous(hr); |
|
132 } while (hr != NULL); |
|
133 } |
|
134 } |
|
135 |
|
136 // Attempts to mark the given object and, if successful, counts |
|
137 // the object in the given task/worker counting structures. |
|
138 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
139 HeapRegion* hr, |
|
140 size_t* marked_bytes_array, |
|
141 BitMap* task_card_bm) { |
|
142 if (_nextMarkBitMap->parMark((HeapWord*)obj)) { |
|
143 // Update the task specific count data for the object. |
|
144 count_object(obj, hr, marked_bytes_array, task_card_bm, obj->size()); |
|
145 return true; |
|
146 } |
|
147 return false; |
|
148 } |
|
149 |
|
150 // Attempts to mark the given object and, if successful, counts |
|
151 // the object in the task/worker counting structures for the |
|
152 // given worker id. |
|
153 inline bool ConcurrentMark::par_mark_and_count(oop obj, |
|
154 size_t word_size, |
|
155 HeapRegion* hr, |
|
156 uint worker_id) { |
|
157 if (_nextMarkBitMap->parMark((HeapWord*)obj)) { |
|
158 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); |
|
159 BitMap* task_card_bm = count_card_bitmap_for(worker_id); |
|
160 count_object(obj, hr, marked_bytes_array, task_card_bm, word_size); |
|
161 return true; |
|
162 } |
|
163 return false; |
|
164 } |
|
165 |
|
166 inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) { |
|
167 HeapWord* start_addr = MAX2(startWord(), mr.start()); |
|
168 HeapWord* end_addr = MIN2(endWord(), mr.end()); |
|
169 |
|
170 if (end_addr > start_addr) { |
|
171 // Right-open interval [start-offset, end-offset). |
|
172 BitMap::idx_t start_offset = heapWordToOffset(start_addr); |
|
173 BitMap::idx_t end_offset = heapWordToOffset(end_addr); |
|
174 |
|
175 start_offset = _bm.get_next_one_offset(start_offset, end_offset); |
|
176 while (start_offset < end_offset) { |
|
177 if (!cl->do_bit(start_offset)) { |
|
178 return false; |
|
179 } |
|
180 HeapWord* next_addr = MIN2(nextObject(offsetToHeapWord(start_offset)), end_addr); |
|
181 BitMap::idx_t next_offset = heapWordToOffset(next_addr); |
|
182 start_offset = _bm.get_next_one_offset(next_offset, end_offset); |
|
183 } |
|
184 } |
|
185 return true; |
|
186 } |
|
187 |
|
188 #define check_mark(addr) \ |
|
189 assert(_bmStartWord <= (addr) && (addr) < (_bmStartWord + _bmWordSize), \ |
|
190 "outside underlying space?"); \ |
|
191 assert(G1CollectedHeap::heap()->is_in_exact(addr), \ |
|
192 "Trying to access not available bitmap " PTR_FORMAT \ |
|
193 " corresponding to " PTR_FORMAT " (%u)", \ |
|
194 p2i(this), p2i(addr), G1CollectedHeap::heap()->addr_to_region(addr)); |
|
195 |
|
196 inline void CMBitMap::mark(HeapWord* addr) { |
|
197 check_mark(addr); |
|
198 _bm.set_bit(heapWordToOffset(addr)); |
|
199 } |
|
200 |
|
201 inline void CMBitMap::clear(HeapWord* addr) { |
|
202 check_mark(addr); |
|
203 _bm.clear_bit(heapWordToOffset(addr)); |
|
204 } |
|
205 |
|
206 inline bool CMBitMap::parMark(HeapWord* addr) { |
|
207 check_mark(addr); |
|
208 return _bm.par_set_bit(heapWordToOffset(addr)); |
|
209 } |
|
210 |
|
211 #undef check_mark |
|
212 |
|
213 template<typename Fn> |
|
214 inline void CMMarkStack::iterate(Fn fn) { |
|
215 assert(_saved_index == _index, "saved index: %d index: %d", _saved_index, _index); |
|
216 for (int i = 0; i < _index; ++i) { |
|
217 fn(_base[i]); |
|
218 } |
|
219 } |
|
220 |
|
221 // It scans an object and visits its children. |
|
222 inline void CMTask::scan_object(oop obj) { process_grey_object<true>(obj); } |
|
223 |
|
224 inline void CMTask::push(oop obj) { |
|
225 HeapWord* objAddr = (HeapWord*) obj; |
|
226 assert(_g1h->is_in_g1_reserved(objAddr), "invariant"); |
|
227 assert(!_g1h->is_on_master_free_list( |
|
228 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant"); |
|
229 assert(!_g1h->is_obj_ill(obj), "invariant"); |
|
230 assert(_nextMarkBitMap->isMarked(objAddr), "invariant"); |
|
231 |
|
232 if (!_task_queue->push(obj)) { |
|
233 // The local task queue looks full. We need to push some entries |
|
234 // to the global stack. |
|
235 move_entries_to_global_stack(); |
|
236 |
|
237 // this should succeed since, even if we overflow the global |
|
238 // stack, we should have definitely removed some entries from the |
|
239 // local queue. So, there must be space on it. |
|
240 bool success = _task_queue->push(obj); |
|
241 assert(success, "invariant"); |
|
242 } |
|
243 } |
|
244 |
|
245 inline bool CMTask::is_below_finger(oop obj, HeapWord* global_finger) const { |
|
246 // If obj is above the global finger, then the mark bitmap scan |
|
247 // will find it later, and no push is needed. Similarly, if we have |
|
248 // a current region and obj is between the local finger and the |
|
249 // end of the current region, then no push is needed. The tradeoff |
|
250 // of checking both vs only checking the global finger is that the |
|
251 // local check will be more accurate and so result in fewer pushes, |
|
252 // but may also be a little slower. |
|
253 HeapWord* objAddr = (HeapWord*)obj; |
|
254 if (_finger != NULL) { |
|
255 // We have a current region. |
|
256 |
|
257 // Finger and region values are all NULL or all non-NULL. We |
|
258 // use _finger to check since we immediately use its value. |
|
259 assert(_curr_region != NULL, "invariant"); |
|
260 assert(_region_limit != NULL, "invariant"); |
|
261 assert(_region_limit <= global_finger, "invariant"); |
|
262 |
|
263 // True if obj is less than the local finger, or is between |
|
264 // the region limit and the global finger. |
|
265 if (objAddr < _finger) { |
|
266 return true; |
|
267 } else if (objAddr < _region_limit) { |
|
268 return false; |
|
269 } // Else check global finger. |
|
270 } |
|
271 // Check global finger. |
|
272 return objAddr < global_finger; |
|
273 } |
|
274 |
|
275 template<bool scan> |
|
276 inline void CMTask::process_grey_object(oop obj) { |
|
277 assert(scan || obj->is_typeArray(), "Skipping scan of grey non-typeArray"); |
|
278 assert(_nextMarkBitMap->isMarked((HeapWord*) obj), "invariant"); |
|
279 |
|
280 size_t obj_size = obj->size(); |
|
281 _words_scanned += obj_size; |
|
282 |
|
283 if (scan) { |
|
284 obj->oop_iterate(_cm_oop_closure); |
|
285 } |
|
286 check_limits(); |
|
287 } |
|
288 |
|
289 |
|
290 |
|
291 inline void CMTask::make_reference_grey(oop obj, HeapRegion* hr) { |
|
292 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) { |
|
293 // No OrderAccess:store_load() is needed. It is implicit in the |
|
294 // CAS done in CMBitMap::parMark() call in the routine above. |
|
295 HeapWord* global_finger = _cm->finger(); |
|
296 |
|
297 // We only need to push a newly grey object on the mark |
|
298 // stack if it is in a section of memory the mark bitmap |
|
299 // scan has already examined. Mark bitmap scanning |
|
300 // maintains progress "fingers" for determining that. |
|
301 // |
|
302 // Notice that the global finger might be moving forward |
|
303 // concurrently. This is not a problem. In the worst case, we |
|
304 // mark the object while it is above the global finger and, by |
|
305 // the time we read the global finger, it has moved forward |
|
306 // past this object. In this case, the object will probably |
|
307 // be visited when a task is scanning the region and will also |
|
308 // be pushed on the stack. So, some duplicate work, but no |
|
309 // correctness problems. |
|
310 if (is_below_finger(obj, global_finger)) { |
|
311 if (obj->is_typeArray()) { |
|
312 // Immediately process arrays of primitive types, rather |
|
313 // than pushing on the mark stack. This keeps us from |
|
314 // adding humongous objects to the mark stack that might |
|
315 // be reclaimed before the entry is processed - see |
|
316 // selection of candidates for eager reclaim of humongous |
|
317 // objects. The cost of the additional type test is |
|
318 // mitigated by avoiding a trip through the mark stack, |
|
319 // by only doing a bookkeeping update and avoiding the |
|
320 // actual scan of the object - a typeArray contains no |
|
321 // references, and the metadata is built-in. |
|
322 process_grey_object<false>(obj); |
|
323 } else { |
|
324 push(obj); |
|
325 } |
|
326 } |
|
327 } |
|
328 } |
|
329 |
|
330 inline void CMTask::deal_with_reference(oop obj) { |
|
331 increment_refs_reached(); |
|
332 |
|
333 HeapWord* objAddr = (HeapWord*) obj; |
|
334 assert(obj->is_oop_or_null(true /* ignore mark word */), "Expected an oop or NULL at " PTR_FORMAT, p2i(obj)); |
|
335 if (_g1h->is_in_g1_reserved(objAddr)) { |
|
336 assert(obj != NULL, "null check is implicit"); |
|
337 if (!_nextMarkBitMap->isMarked(objAddr)) { |
|
338 // Only get the containing region if the object is not marked on the |
|
339 // bitmap (otherwise, it's a waste of time since we won't do |
|
340 // anything with it). |
|
341 HeapRegion* hr = _g1h->heap_region_containing(obj); |
|
342 if (!hr->obj_allocated_since_next_marking(obj)) { |
|
343 make_reference_grey(obj, hr); |
|
344 } |
|
345 } |
|
346 } |
|
347 } |
|
348 |
|
349 inline void ConcurrentMark::markPrev(oop p) { |
|
350 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity"); |
|
351 // Note we are overriding the read-only view of the prev map here, via |
|
352 // the cast. |
|
353 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p); |
|
354 } |
|
355 |
|
356 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, |
|
357 uint worker_id, HeapRegion* hr) { |
|
358 assert(obj != NULL, "pre-condition"); |
|
359 HeapWord* addr = (HeapWord*) obj; |
|
360 if (hr == NULL) { |
|
361 hr = _g1h->heap_region_containing(addr); |
|
362 } else { |
|
363 assert(hr->is_in(addr), "pre-condition"); |
|
364 } |
|
365 assert(hr != NULL, "sanity"); |
|
366 // Given that we're looking for a region that contains an object |
|
367 // header it's impossible to get back a HC region. |
|
368 assert(!hr->is_continues_humongous(), "sanity"); |
|
369 |
|
370 if (addr < hr->next_top_at_mark_start()) { |
|
371 if (!_nextMarkBitMap->isMarked(addr)) { |
|
372 par_mark_and_count(obj, word_size, hr, worker_id); |
|
373 } |
|
374 } |
|
375 } |
|
376 |
|
377 #endif // SHARE_VM_GC_G1_CONCURRENTMARK_INLINE_HPP |
|