Mercurial > hg > truffle
comparison src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp @ 3357:fc2b798ab316
6883834: ParNew: assert(!_g->to()->is_in_reserved(obj),"Scanning field twice?") with LargeObjects tests
Summary: Fixed process_chunk_boundaries(), used for parallel card scanning when using ParNew/CMS, so as to prevent double-scanning, or worse, non-scanning of imprecisely marked objects exceeding parallel chunk size. Made some sizing parameters for parallel card scanning diagnostic, disabled ParallelGCRetainPLAB, and elaborated and clarified some comments.
Reviewed-by: stefank, johnc
author | ysr |
---|---|
date | Tue, 10 May 2011 00:33:21 -0700 |
parents | c48ad6ab8bdf |
children | 7d64aa23eb96 |
comparison
equal
deleted
inserted
replaced
3325:54a56bbaf95b | 3357:fc2b798ab316 |
---|---|
27 #include "memory/cardTableModRefBS.hpp" | 27 #include "memory/cardTableModRefBS.hpp" |
28 #include "memory/cardTableRS.hpp" | 28 #include "memory/cardTableRS.hpp" |
29 #include "memory/sharedHeap.hpp" | 29 #include "memory/sharedHeap.hpp" |
30 #include "memory/space.inline.hpp" | 30 #include "memory/space.inline.hpp" |
31 #include "memory/universe.hpp" | 31 #include "memory/universe.hpp" |
32 #include "oops/oop.inline.hpp" | |
32 #include "runtime/java.hpp" | 33 #include "runtime/java.hpp" |
33 #include "runtime/mutexLocker.hpp" | 34 #include "runtime/mutexLocker.hpp" |
34 #include "runtime/virtualspace.hpp" | 35 #include "runtime/virtualspace.hpp" |
35 | 36 |
36 void CardTableModRefBS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, | 37 void CardTableModRefBS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, |
37 DirtyCardToOopClosure* dcto_cl, | 38 OopsInGenClosure* cl, |
38 ClearNoncleanCardWrapper* cl, | 39 CardTableRS* ct, |
39 int n_threads) { | 40 int n_threads) { |
40 assert(n_threads > 0, "Error: expected n_threads > 0"); | 41 assert(n_threads > 0, "Error: expected n_threads > 0"); |
41 assert((n_threads == 1 && ParallelGCThreads == 0) || | 42 assert((n_threads == 1 && ParallelGCThreads == 0) || |
42 n_threads <= (int)ParallelGCThreads, | 43 n_threads <= (int)ParallelGCThreads, |
43 "# worker threads != # requested!"); | 44 "# worker threads != # requested!"); |
47 size_t lowest_non_clean_chunk_size; | 48 size_t lowest_non_clean_chunk_size; |
48 get_LNC_array_for_space(sp, lowest_non_clean, | 49 get_LNC_array_for_space(sp, lowest_non_clean, |
49 lowest_non_clean_base_chunk_index, | 50 lowest_non_clean_base_chunk_index, |
50 lowest_non_clean_chunk_size); | 51 lowest_non_clean_chunk_size); |
51 | 52 |
52 int n_strides = n_threads * StridesPerThread; | 53 int n_strides = n_threads * ParGCStridesPerThread; |
53 SequentialSubTasksDone* pst = sp->par_seq_tasks(); | 54 SequentialSubTasksDone* pst = sp->par_seq_tasks(); |
54 pst->set_n_threads(n_threads); | 55 pst->set_n_threads(n_threads); |
55 pst->set_n_tasks(n_strides); | 56 pst->set_n_tasks(n_strides); |
56 | 57 |
57 int stride = 0; | 58 int stride = 0; |
58 while (!pst->is_task_claimed(/* reference */ stride)) { | 59 while (!pst->is_task_claimed(/* reference */ stride)) { |
59 process_stride(sp, mr, stride, n_strides, dcto_cl, cl, | 60 process_stride(sp, mr, stride, n_strides, cl, ct, |
60 lowest_non_clean, | 61 lowest_non_clean, |
61 lowest_non_clean_base_chunk_index, | 62 lowest_non_clean_base_chunk_index, |
62 lowest_non_clean_chunk_size); | 63 lowest_non_clean_chunk_size); |
63 } | 64 } |
64 if (pst->all_tasks_completed()) { | 65 if (pst->all_tasks_completed()) { |
77 void | 78 void |
78 CardTableModRefBS:: | 79 CardTableModRefBS:: |
79 process_stride(Space* sp, | 80 process_stride(Space* sp, |
80 MemRegion used, | 81 MemRegion used, |
81 jint stride, int n_strides, | 82 jint stride, int n_strides, |
82 DirtyCardToOopClosure* dcto_cl, | 83 OopsInGenClosure* cl, |
83 ClearNoncleanCardWrapper* cl, | 84 CardTableRS* ct, |
84 jbyte** lowest_non_clean, | 85 jbyte** lowest_non_clean, |
85 uintptr_t lowest_non_clean_base_chunk_index, | 86 uintptr_t lowest_non_clean_base_chunk_index, |
86 size_t lowest_non_clean_chunk_size) { | 87 size_t lowest_non_clean_chunk_size) { |
87 // We don't have to go downwards here; it wouldn't help anyway, | 88 // We go from higher to lower addresses here; it wouldn't help that much |
88 // because of parallelism. | 89 // because of the strided parallelism pattern used here. |
89 | 90 |
90 // Find the first card address of the first chunk in the stride that is | 91 // Find the first card address of the first chunk in the stride that is |
91 // at least "bottom" of the used region. | 92 // at least "bottom" of the used region. |
92 jbyte* start_card = byte_for(used.start()); | 93 jbyte* start_card = byte_for(used.start()); |
93 jbyte* end_card = byte_after(used.last()); | 94 jbyte* end_card = byte_after(used.last()); |
96 jbyte* chunk_card_start; | 97 jbyte* chunk_card_start; |
97 | 98 |
98 if ((uintptr_t)stride >= start_chunk_stride_num) { | 99 if ((uintptr_t)stride >= start_chunk_stride_num) { |
99 chunk_card_start = (jbyte*)(start_card + | 100 chunk_card_start = (jbyte*)(start_card + |
100 (stride - start_chunk_stride_num) * | 101 (stride - start_chunk_stride_num) * |
101 CardsPerStrideChunk); | 102 ParGCCardsPerStrideChunk); |
102 } else { | 103 } else { |
103 // Go ahead to the next chunk group boundary, then to the requested stride. | 104 // Go ahead to the next chunk group boundary, then to the requested stride. |
104 chunk_card_start = (jbyte*)(start_card + | 105 chunk_card_start = (jbyte*)(start_card + |
105 (n_strides - start_chunk_stride_num + stride) * | 106 (n_strides - start_chunk_stride_num + stride) * |
106 CardsPerStrideChunk); | 107 ParGCCardsPerStrideChunk); |
107 } | 108 } |
108 | 109 |
109 while (chunk_card_start < end_card) { | 110 while (chunk_card_start < end_card) { |
110 // We don't have to go downwards here; it wouldn't help anyway, | 111 // Even though we go from lower to higher addresses below, the |
111 // because of parallelism. (We take care with "min_done"; see below.) | 112 // strided parallelism can interleave the actual processing of the |
113 // dirty pages in various ways. For a specific chunk within this | |
114 // stride, we take care to avoid double scanning or missing a card | |
115 // by suitably initializing the "min_done" field in process_chunk_boundaries() | |
116 // below, together with the dirty region extension accomplished in | |
117 // DirtyCardToOopClosure::do_MemRegion(). | |
118 jbyte* chunk_card_end = chunk_card_start + ParGCCardsPerStrideChunk; | |
112 // Invariant: chunk_mr should be fully contained within the "used" region. | 119 // Invariant: chunk_mr should be fully contained within the "used" region. |
113 jbyte* chunk_card_end = chunk_card_start + CardsPerStrideChunk; | |
114 MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start), | 120 MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start), |
115 chunk_card_end >= end_card ? | 121 chunk_card_end >= end_card ? |
116 used.end() : addr_for(chunk_card_end)); | 122 used.end() : addr_for(chunk_card_end)); |
117 assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)"); | 123 assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)"); |
118 assert(used.contains(chunk_mr), "chunk_mr should be subset of used"); | 124 assert(used.contains(chunk_mr), "chunk_mr should be subset of used"); |
125 | |
126 DirtyCardToOopClosure* dcto_cl = sp->new_dcto_cl(cl, precision(), | |
127 cl->gen_boundary()); | |
128 ClearNoncleanCardWrapper clear_cl(dcto_cl, ct); | |
129 | |
119 | 130 |
120 // Process the chunk. | 131 // Process the chunk. |
121 process_chunk_boundaries(sp, | 132 process_chunk_boundaries(sp, |
122 dcto_cl, | 133 dcto_cl, |
123 chunk_mr, | 134 chunk_mr, |
124 used, | 135 used, |
125 lowest_non_clean, | 136 lowest_non_clean, |
126 lowest_non_clean_base_chunk_index, | 137 lowest_non_clean_base_chunk_index, |
127 lowest_non_clean_chunk_size); | 138 lowest_non_clean_chunk_size); |
128 | 139 |
140 // We want the LNC array updates above in process_chunk_boundaries | |
141 // to be visible before any of the card table value changes as a | |
142 // result of the dirty card iteration below. | |
143 OrderAccess::storestore(); | |
144 | |
129 // We do not call the non_clean_card_iterate_serial() version because | 145 // We do not call the non_clean_card_iterate_serial() version because |
130 // we want to clear the cards, and the ClearNoncleanCardWrapper closure | 146 // we want to clear the cards: clear_cl here does the work of finding |
131 // itself does the work of finding contiguous dirty ranges of cards to | 147 // contiguous dirty ranges of cards to process and clear. |
132 // process (and clear). | 148 clear_cl.do_MemRegion(chunk_mr); |
133 cl->do_MemRegion(chunk_mr); | |
134 | 149 |
135 // Find the next chunk of the stride. | 150 // Find the next chunk of the stride. |
136 chunk_card_start += CardsPerStrideChunk * n_strides; | 151 chunk_card_start += ParGCCardsPerStrideChunk * n_strides; |
137 } | 152 } |
138 } | 153 } |
154 | |
155 | |
156 // If you want a talkative process_chunk_boundaries, | |
157 // then #define NOISY(x) x | |
158 #ifdef NOISY | |
159 #error "Encountered a global preprocessor flag, NOISY, which might clash with local definition to follow" | |
160 #else | |
161 #define NOISY(x) | |
162 #endif | |
139 | 163 |
140 void | 164 void |
141 CardTableModRefBS:: | 165 CardTableModRefBS:: |
142 process_chunk_boundaries(Space* sp, | 166 process_chunk_boundaries(Space* sp, |
143 DirtyCardToOopClosure* dcto_cl, | 167 DirtyCardToOopClosure* dcto_cl, |
145 MemRegion used, | 169 MemRegion used, |
146 jbyte** lowest_non_clean, | 170 jbyte** lowest_non_clean, |
147 uintptr_t lowest_non_clean_base_chunk_index, | 171 uintptr_t lowest_non_clean_base_chunk_index, |
148 size_t lowest_non_clean_chunk_size) | 172 size_t lowest_non_clean_chunk_size) |
149 { | 173 { |
150 // We must worry about the chunk boundaries. | 174 // We must worry about non-array objects that cross chunk boundaries, |
151 | 175 // because such objects are both precisely and imprecisely marked: |
152 // First, set our max_to_do: | 176 // .. if the head of such an object is dirty, the entire object |
153 HeapWord* max_to_do = NULL; | 177 // needs to be scanned, under the interpretation that this |
178 // was an imprecise mark | |
179 // .. if the head of such an object is not dirty, we can assume | |
180 // precise marking and it's efficient to scan just the dirty | |
181 // cards. | |
182 // In either case, each scanned reference must be scanned precisely | |
183 // once so as to avoid cloning of a young referent. For efficiency, | |
184 // our closures depend on this property and do not protect against | |
185 // double scans. | |
186 | |
154 uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start()); | 187 uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start()); |
155 cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index; | 188 cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index; |
156 | 189 |
190 NOISY(tty->print_cr("===========================================================================");) | |
191 NOISY(tty->print_cr(" process_chunk_boundary: Called with [" PTR_FORMAT "," PTR_FORMAT ")", | |
192 chunk_mr.start(), chunk_mr.end());) | |
193 | |
194 // First, set "our" lowest_non_clean entry, which would be | |
195 // used by the thread scanning an adjoining left chunk with | |
196 // a non-array object straddling the mutual boundary. | |
197 // Find the object that spans our boundary, if one exists. | |
198 // first_block is the block possibly straddling our left boundary. | |
199 HeapWord* first_block = sp->block_start(chunk_mr.start()); | |
200 assert((chunk_mr.start() != used.start()) || (first_block == chunk_mr.start()), | |
201 "First chunk should always have a co-initial block"); | |
202 // Does the block straddle the chunk's left boundary, and is it | |
203 // a non-array object? | |
204 if (first_block < chunk_mr.start() // first block straddles left bdry | |
205 && sp->block_is_obj(first_block) // first block is an object | |
206 && !(oop(first_block)->is_objArray() // first block is not an array (arrays are precisely dirtied) | |
207 || oop(first_block)->is_typeArray())) { | |
208 // Find our least non-clean card, so that a left neighbour | |
209 // does not scan an object straddling the mutual boundary | |
210 // too far to the right, and attempt to scan a portion of | |
211 // that object twice. | |
212 jbyte* first_dirty_card = NULL; | |
213 jbyte* last_card_of_first_obj = | |
214 byte_for(first_block + sp->block_size(first_block) - 1); | |
215 jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); | |
216 jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last()); | |
217 jbyte* last_card_to_check = | |
218 (jbyte*) MIN2((intptr_t) last_card_of_cur_chunk, | |
219 (intptr_t) last_card_of_first_obj); | |
220 // Note that this does not need to go beyond our last card | |
221 // if our first object completely straddles this chunk. | |
222 for (jbyte* cur = first_card_of_cur_chunk; | |
223 cur <= last_card_to_check; cur++) { | |
224 jbyte val = *cur; | |
225 if (card_will_be_scanned(val)) { | |
226 first_dirty_card = cur; break; | |
227 } else { | |
228 assert(!card_may_have_been_dirty(val), "Error"); | |
229 } | |
230 } | |
231 if (first_dirty_card != NULL) { | |
232 NOISY(tty->print_cr(" LNC: Found a dirty card at " PTR_FORMAT " in current chunk", | |
233 first_dirty_card);) | |
234 assert(0 <= cur_chunk_index && cur_chunk_index < lowest_non_clean_chunk_size, | |
235 "Bounds error."); | |
236 assert(lowest_non_clean[cur_chunk_index] == NULL, | |
237 "Write exactly once : value should be stable hereafter for this round"); | |
238 lowest_non_clean[cur_chunk_index] = first_dirty_card; | |
239 } NOISY(else { | |
240 tty->print_cr(" LNC: Found no dirty card in current chunk; leaving LNC entry NULL"); | |
241 // In the future, we could have this thread look for a non-NULL value to copy from its | |
242 // right neighbour (up to the end of the first object). | |
243 if (last_card_of_cur_chunk < last_card_of_first_obj) { | |
244 tty->print_cr(" LNC: BEWARE!!! first obj straddles past right end of chunk:\n" | |
245 " might be efficient to get value from right neighbour?"); | |
246 } | |
247 }) | |
248 } else { | |
249 // In this case we can help our neighbour by just asking them | |
250 // to stop at our first card (even though it may not be dirty). | |
251 NOISY(tty->print_cr(" LNC: first block is not a non-array object; setting LNC to first card of current chunk");) | |
252 assert(lowest_non_clean[cur_chunk_index] == NULL, "Write once : value should be stable hereafter"); | |
253 jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); | |
254 lowest_non_clean[cur_chunk_index] = first_card_of_cur_chunk; | |
255 } | |
256 NOISY(tty->print_cr(" process_chunk_boundary: lowest_non_clean[" INTPTR_FORMAT "] = " PTR_FORMAT | |
257 " which corresponds to the heap address " PTR_FORMAT, | |
258 cur_chunk_index, lowest_non_clean[cur_chunk_index], | |
259 (lowest_non_clean[cur_chunk_index] != NULL) | |
260 ? addr_for(lowest_non_clean[cur_chunk_index]) | |
261 : NULL);) | |
262 NOISY(tty->print_cr("---------------------------------------------------------------------------");) | |
263 | |
264 // Next, set our own max_to_do, which will strictly/exclusively bound | |
265 // the highest address that we will scan past the right end of our chunk. | |
266 HeapWord* max_to_do = NULL; | |
157 if (chunk_mr.end() < used.end()) { | 267 if (chunk_mr.end() < used.end()) { |
158 // This is not the last chunk in the used region. What is the last | 268 // This is not the last chunk in the used region. |
159 // object? | 269 // What is our last block? We check the first block of |
160 HeapWord* last_block = sp->block_start(chunk_mr.end()); | 270 // the next (right) chunk rather than strictly check our last block |
271 // because it's potentially more efficient to do so. | |
272 HeapWord* const last_block = sp->block_start(chunk_mr.end()); | |
161 assert(last_block <= chunk_mr.end(), "In case this property changes."); | 273 assert(last_block <= chunk_mr.end(), "In case this property changes."); |
162 if (last_block == chunk_mr.end() | 274 if ((last_block == chunk_mr.end()) // our last block does not straddle boundary |
163 || !sp->block_is_obj(last_block)) { | 275 || !sp->block_is_obj(last_block) // last_block isn't an object |
276 || oop(last_block)->is_objArray() // last_block is an array (precisely marked) | |
277 || oop(last_block)->is_typeArray()) { | |
164 max_to_do = chunk_mr.end(); | 278 max_to_do = chunk_mr.end(); |
165 | 279 NOISY(tty->print_cr(" process_chunk_boundary: Last block on this card is not a non-array object;\n" |
280 " max_to_do left at " PTR_FORMAT, max_to_do);) | |
166 } else { | 281 } else { |
167 // It is an object and starts before the end of the current chunk. | 282 assert(last_block < chunk_mr.end(), "Tautology"); |
283 // It is a non-array object that straddles the right boundary of this chunk. | |
168 // last_obj_card is the card corresponding to the start of the last object | 284 // last_obj_card is the card corresponding to the start of the last object |
169 // in the chunk. Note that the last object may not start in | 285 // in the chunk. Note that the last object may not start in |
170 // the chunk. | 286 // the chunk. |
171 jbyte* last_obj_card = byte_for(last_block); | 287 jbyte* const last_obj_card = byte_for(last_block); |
172 if (!card_may_have_been_dirty(*last_obj_card)) { | 288 const jbyte val = *last_obj_card; |
173 // The card containing the head is not dirty. Any marks in | 289 if (!card_will_be_scanned(val)) { |
290 assert(!card_may_have_been_dirty(val), "Error"); | |
291 // The card containing the head is not dirty. Any marks on | |
174 // subsequent cards still in this chunk must have been made | 292 // subsequent cards still in this chunk must have been made |
175 // precisely; we can cap processing at the end. | 293 // precisely; we can cap processing at the end of our chunk. |
176 max_to_do = chunk_mr.end(); | 294 max_to_do = chunk_mr.end(); |
295 NOISY(tty->print_cr(" process_chunk_boundary: Head of last object on this card is not dirty;\n" | |
296 " max_to_do left at " PTR_FORMAT, | |
297 max_to_do);) | |
177 } else { | 298 } else { |
178 // The last object must be considered dirty, and extends onto the | 299 // The last object must be considered dirty, and extends onto the |
179 // following chunk. Look for a dirty card in that chunk that will | 300 // following chunk. Look for a dirty card in that chunk that will |
180 // bound our processing. | 301 // bound our processing. |
181 jbyte* limit_card = NULL; | 302 jbyte* limit_card = NULL; |
182 size_t last_block_size = sp->block_size(last_block); | 303 const size_t last_block_size = sp->block_size(last_block); |
183 jbyte* last_card_of_last_obj = | 304 jbyte* const last_card_of_last_obj = |
184 byte_for(last_block + last_block_size - 1); | 305 byte_for(last_block + last_block_size - 1); |
185 jbyte* first_card_of_next_chunk = byte_for(chunk_mr.end()); | 306 jbyte* const first_card_of_next_chunk = byte_for(chunk_mr.end()); |
186 // This search potentially goes a long distance looking | 307 // This search potentially goes a long distance looking |
187 // for the next card that will be scanned. For example, | 308 // for the next card that will be scanned, terminating |
188 // an object that is an array of primitives will not | 309 // at the end of the last_block, if no earlier dirty card |
189 // have any cards covering regions interior to the array | 310 // is found. |
190 // that will need to be scanned. The scan can be terminated | 311 assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start()) == ParGCCardsPerStrideChunk, |
191 // at the last card of the next chunk. That would leave | 312 "last card of next chunk may be wrong"); |
192 // limit_card as NULL and would result in "max_to_do" | |
193 // being set with the LNC value or with the end | |
194 // of the last block. | |
195 jbyte* last_card_of_next_chunk = first_card_of_next_chunk + | |
196 CardsPerStrideChunk; | |
197 assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start()) | |
198 == CardsPerStrideChunk, "last card of next chunk may be wrong"); | |
199 jbyte* last_card_to_check = (jbyte*) MIN2(last_card_of_last_obj, | |
200 last_card_of_next_chunk); | |
201 for (jbyte* cur = first_card_of_next_chunk; | 313 for (jbyte* cur = first_card_of_next_chunk; |
202 cur <= last_card_to_check; cur++) { | 314 cur <= last_card_of_last_obj; cur++) { |
203 if (card_will_be_scanned(*cur)) { | 315 const jbyte val = *cur; |
316 if (card_will_be_scanned(val)) { | |
317 NOISY(tty->print_cr(" Found a non-clean card " PTR_FORMAT " with value 0x%x", | |
318 cur, (int)val);) | |
204 limit_card = cur; break; | 319 limit_card = cur; break; |
320 } else { | |
321 assert(!card_may_have_been_dirty(val), "Error: card can't be skipped"); | |
205 } | 322 } |
206 } | 323 } |
207 assert(0 <= cur_chunk_index+1 && | 324 if (limit_card != NULL) { |
208 cur_chunk_index+1 < lowest_non_clean_chunk_size, | 325 max_to_do = addr_for(limit_card); |
326 assert(limit_card != NULL && max_to_do != NULL, "Error"); | |
327 NOISY(tty->print_cr(" process_chunk_boundary: Found a dirty card at " PTR_FORMAT | |
328 " max_to_do set at " PTR_FORMAT " which is before end of last block in chunk: " | |
329 PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, | |
330 limit_card, max_to_do, last_block, last_block_size, (last_block+last_block_size));) | |
331 } else { | |
332 // The following is a pessimistic value, because it's possible | |
333 // that a dirty card on a subsequent chunk has been cleared by | |
334 // the time we get to look at it; we'll correct for that further below, | |
335 // using the LNC array which records the least non-clean card | |
336 // before cards were cleared in a particular chunk. | |
337 limit_card = last_card_of_last_obj; | |
338 max_to_do = last_block + last_block_size; | |
339 assert(limit_card != NULL && max_to_do != NULL, "Error"); | |
340 NOISY(tty->print_cr(" process_chunk_boundary: Found no dirty card before end of last block in chunk\n" | |
341 " Setting limit_card to " PTR_FORMAT | |
342 " and max_to_do " PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, | |
343 limit_card, last_block, last_block_size, max_to_do);) | |
344 } | |
345 assert(0 < cur_chunk_index+1 && cur_chunk_index+1 < lowest_non_clean_chunk_size, | |
209 "Bounds error."); | 346 "Bounds error."); |
210 // LNC for the next chunk | 347 // It is possible that a dirty card for the last object may have been |
211 jbyte* lnc_card = lowest_non_clean[cur_chunk_index+1]; | 348 // cleared before we had a chance to examine it. In that case, the value |
212 if (limit_card == NULL) { | 349 // will have been logged in the LNC for that chunk. |
213 limit_card = lnc_card; | 350 // We need to examine as many chunks to the right as this object |
351 // covers. | |
352 const uintptr_t last_chunk_index_to_check = addr_to_chunk_index(last_block + last_block_size - 1) | |
353 - lowest_non_clean_base_chunk_index; | |
354 DEBUG_ONLY(const uintptr_t last_chunk_index = addr_to_chunk_index(used.end()) | |
355 - lowest_non_clean_base_chunk_index;) | |
356 assert(last_chunk_index_to_check <= last_chunk_index, | |
357 err_msg("Out of bounds: last_chunk_index_to_check " INTPTR_FORMAT | |
358 " exceeds last_chunk_index " INTPTR_FORMAT, | |
359 last_chunk_index_to_check, last_chunk_index)); | |
360 for (uintptr_t lnc_index = cur_chunk_index + 1; | |
361 lnc_index <= last_chunk_index_to_check; | |
362 lnc_index++) { | |
363 jbyte* lnc_card = lowest_non_clean[lnc_index]; | |
364 if (lnc_card != NULL) { | |
365 // we can stop at the first non-NULL entry we find | |
366 if (lnc_card <= limit_card) { | |
367 NOISY(tty->print_cr(" process_chunk_boundary: LNC card " PTR_FORMAT " is lower than limit_card " PTR_FORMAT, | |
368 " max_to_do will be lowered to " PTR_FORMAT " from " PTR_FORMAT, | |
369 lnc_card, limit_card, addr_for(lnc_card), max_to_do);) | |
370 limit_card = lnc_card; | |
371 max_to_do = addr_for(limit_card); | |
372 assert(limit_card != NULL && max_to_do != NULL, "Error"); | |
373 } | |
374 // In any case, we break now | |
375 break; | |
376 } // else continue to look for a non-NULL entry if any | |
214 } | 377 } |
215 if (limit_card != NULL) { | 378 assert(limit_card != NULL && max_to_do != NULL, "Error"); |
216 if (lnc_card != NULL) { | |
217 limit_card = (jbyte*)MIN2((intptr_t)limit_card, | |
218 (intptr_t)lnc_card); | |
219 } | |
220 max_to_do = addr_for(limit_card); | |
221 } else { | |
222 max_to_do = last_block + last_block_size; | |
223 } | |
224 } | 379 } |
380 assert(max_to_do != NULL, "OOPS 1 !"); | |
225 } | 381 } |
226 assert(max_to_do != NULL, "OOPS!"); | 382 assert(max_to_do != NULL, "OOPS 2!"); |
227 } else { | 383 } else { |
228 max_to_do = used.end(); | 384 max_to_do = used.end(); |
229 } | 385 NOISY(tty->print_cr(" process_chunk_boundary: Last chunk of this space;\n" |
386 " max_to_do left at " PTR_FORMAT, | |
387 max_to_do);) | |
388 } | |
389 assert(max_to_do != NULL, "OOPS 3!"); | |
230 // Now we can set the closure we're using so it doesn't to beyond | 390 // Now we can set the closure we're using so it doesn't to beyond |
231 // max_to_do. | 391 // max_to_do. |
232 dcto_cl->set_min_done(max_to_do); | 392 dcto_cl->set_min_done(max_to_do); |
233 #ifndef PRODUCT | 393 #ifndef PRODUCT |
234 dcto_cl->set_last_bottom(max_to_do); | 394 dcto_cl->set_last_bottom(max_to_do); |
235 #endif | 395 #endif |
236 | 396 NOISY(tty->print_cr("===========================================================================\n");) |
237 // Now we set *our" lowest_non_clean entry. | |
238 // Find the object that spans our boundary, if one exists. | |
239 // Nothing to do on the first chunk. | |
240 if (chunk_mr.start() > used.start()) { | |
241 // first_block is the block possibly spanning the chunk start | |
242 HeapWord* first_block = sp->block_start(chunk_mr.start()); | |
243 // Does the block span the start of the chunk and is it | |
244 // an object? | |
245 if (first_block < chunk_mr.start() && | |
246 sp->block_is_obj(first_block)) { | |
247 jbyte* first_dirty_card = NULL; | |
248 jbyte* last_card_of_first_obj = | |
249 byte_for(first_block + sp->block_size(first_block) - 1); | |
250 jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); | |
251 jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last()); | |
252 jbyte* last_card_to_check = | |
253 (jbyte*) MIN2((intptr_t) last_card_of_cur_chunk, | |
254 (intptr_t) last_card_of_first_obj); | |
255 for (jbyte* cur = first_card_of_cur_chunk; | |
256 cur <= last_card_to_check; cur++) { | |
257 if (card_will_be_scanned(*cur)) { | |
258 first_dirty_card = cur; break; | |
259 } | |
260 } | |
261 if (first_dirty_card != NULL) { | |
262 assert(0 <= cur_chunk_index && | |
263 cur_chunk_index < lowest_non_clean_chunk_size, | |
264 "Bounds error."); | |
265 lowest_non_clean[cur_chunk_index] = first_dirty_card; | |
266 } | |
267 } | |
268 } | |
269 } | 397 } |
398 | |
399 #undef NOISY | |
270 | 400 |
271 void | 401 void |
272 CardTableModRefBS:: | 402 CardTableModRefBS:: |
273 get_LNC_array_for_space(Space* sp, | 403 get_LNC_array_for_space(Space* sp, |
274 jbyte**& lowest_non_clean, | 404 jbyte**& lowest_non_clean, |
281 | 411 |
282 // Only the first thread to obtain the lock will resize the | 412 // Only the first thread to obtain the lock will resize the |
283 // LNC array for the covered region. Any later expansion can't affect | 413 // LNC array for the covered region. Any later expansion can't affect |
284 // the used_at_save_marks region. | 414 // the used_at_save_marks region. |
285 // (I observed a bug in which the first thread to execute this would | 415 // (I observed a bug in which the first thread to execute this would |
286 // resize, and then it would cause "expand_and_allocates" that would | 416 // resize, and then it would cause "expand_and_allocate" that would |
287 // Increase the number of chunks in the covered region. Then a second | 417 // increase the number of chunks in the covered region. Then a second |
288 // thread would come and execute this, see that the size didn't match, | 418 // thread would come and execute this, see that the size didn't match, |
289 // and free and allocate again. So the first thread would be using a | 419 // and free and allocate again. So the first thread would be using a |
290 // freed "_lowest_non_clean" array.) | 420 // freed "_lowest_non_clean" array.) |
291 | 421 |
292 // Do a dirty read here. If we pass the conditional then take the rare | 422 // Do a dirty read here. If we pass the conditional then take the rare |