1 | /* |
2 | * Copyright (c) 2001, 2018, 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 | #include "precompiled.hpp" |
26 | #include "classfile/classLoaderDataGraph.hpp" |
27 | #include "gc/shared/cardTableRS.hpp" |
28 | #include "gc/shared/genCollectedHeap.hpp" |
29 | #include "gc/shared/genOopClosures.hpp" |
30 | #include "gc/shared/generation.hpp" |
31 | #include "gc/shared/space.inline.hpp" |
32 | #include "memory/allocation.inline.hpp" |
33 | #include "memory/iterator.inline.hpp" |
34 | #include "oops/access.inline.hpp" |
35 | #include "oops/oop.inline.hpp" |
36 | #include "runtime/atomic.hpp" |
37 | #include "runtime/java.hpp" |
38 | #include "runtime/os.hpp" |
39 | #include "utilities/macros.hpp" |
40 | |
41 | class HasAccumulatedModifiedOopsClosure : public CLDClosure { |
42 | bool _found; |
43 | public: |
44 | HasAccumulatedModifiedOopsClosure() : _found(false) {} |
45 | void do_cld(ClassLoaderData* cld) { |
46 | if (_found) { |
47 | return; |
48 | } |
49 | |
50 | if (cld->has_accumulated_modified_oops()) { |
51 | _found = true; |
52 | } |
53 | } |
54 | bool found() { |
55 | return _found; |
56 | } |
57 | }; |
58 | |
59 | bool CLDRemSet::mod_union_is_clear() { |
60 | HasAccumulatedModifiedOopsClosure closure; |
61 | ClassLoaderDataGraph::cld_do(&closure); |
62 | |
63 | return !closure.found(); |
64 | } |
65 | |
66 | |
67 | class ClearCLDModUnionClosure : public CLDClosure { |
68 | public: |
69 | void do_cld(ClassLoaderData* cld) { |
70 | if (cld->has_accumulated_modified_oops()) { |
71 | cld->clear_accumulated_modified_oops(); |
72 | } |
73 | } |
74 | }; |
75 | |
76 | void CLDRemSet::clear_mod_union() { |
77 | ClearCLDModUnionClosure closure; |
78 | ClassLoaderDataGraph::cld_do(&closure); |
79 | } |
80 | |
81 | CardTable::CardValue CardTableRS::find_unused_youngergenP_card_value() { |
82 | for (CardValue v = youngergenP1_card; |
83 | v < cur_youngergen_and_prev_nonclean_card; |
84 | v++) { |
85 | bool seen = false; |
86 | for (int g = 0; g < _regions_to_iterate; g++) { |
87 | if (_last_cur_val_in_gen[g] == v) { |
88 | seen = true; |
89 | break; |
90 | } |
91 | } |
92 | if (!seen) { |
93 | return v; |
94 | } |
95 | } |
96 | ShouldNotReachHere(); |
97 | return 0; |
98 | } |
99 | |
100 | void CardTableRS::prepare_for_younger_refs_iterate(bool parallel) { |
101 | // Parallel or sequential, we must always set the prev to equal the |
102 | // last one written. |
103 | if (parallel) { |
104 | // Find a parallel value to be used next. |
105 | jbyte next_val = find_unused_youngergenP_card_value(); |
106 | set_cur_youngergen_card_val(next_val); |
107 | |
108 | } else { |
109 | // In an sequential traversal we will always write youngergen, so that |
110 | // the inline barrier is correct. |
111 | set_cur_youngergen_card_val(youngergen_card); |
112 | } |
113 | } |
114 | |
115 | void CardTableRS::younger_refs_iterate(Generation* g, |
116 | OopsInGenClosure* blk, |
117 | uint n_threads) { |
118 | // The indexing in this array is slightly odd. We want to access |
119 | // the old generation record here, which is at index 2. |
120 | _last_cur_val_in_gen[2] = cur_youngergen_card_val(); |
121 | g->younger_refs_iterate(blk, n_threads); |
122 | } |
123 | |
124 | inline bool ClearNoncleanCardWrapper::clear_card(CardValue* entry) { |
125 | if (_is_par) { |
126 | return clear_card_parallel(entry); |
127 | } else { |
128 | return clear_card_serial(entry); |
129 | } |
130 | } |
131 | |
132 | inline bool ClearNoncleanCardWrapper::clear_card_parallel(CardValue* entry) { |
133 | while (true) { |
134 | // In the parallel case, we may have to do this several times. |
135 | CardValue entry_val = *entry; |
136 | assert(entry_val != CardTableRS::clean_card_val(), |
137 | "We shouldn't be looking at clean cards, and this should " |
138 | "be the only place they get cleaned." ); |
139 | if (CardTableRS::card_is_dirty_wrt_gen_iter(entry_val) |
140 | || _ct->is_prev_youngergen_card_val(entry_val)) { |
141 | CardValue res = |
142 | Atomic::cmpxchg(CardTableRS::clean_card_val(), entry, entry_val); |
143 | if (res == entry_val) { |
144 | break; |
145 | } else { |
146 | assert(res == CardTableRS::cur_youngergen_and_prev_nonclean_card, |
147 | "The CAS above should only fail if another thread did " |
148 | "a GC write barrier." ); |
149 | } |
150 | } else if (entry_val == |
151 | CardTableRS::cur_youngergen_and_prev_nonclean_card) { |
152 | // Parallelism shouldn't matter in this case. Only the thread |
153 | // assigned to scan the card should change this value. |
154 | *entry = _ct->cur_youngergen_card_val(); |
155 | break; |
156 | } else { |
157 | assert(entry_val == _ct->cur_youngergen_card_val(), |
158 | "Should be the only possibility." ); |
159 | // In this case, the card was clean before, and become |
160 | // cur_youngergen only because of processing of a promoted object. |
161 | // We don't have to look at the card. |
162 | return false; |
163 | } |
164 | } |
165 | return true; |
166 | } |
167 | |
168 | |
169 | inline bool ClearNoncleanCardWrapper::clear_card_serial(CardValue* entry) { |
170 | CardValue entry_val = *entry; |
171 | assert(entry_val != CardTableRS::clean_card_val(), |
172 | "We shouldn't be looking at clean cards, and this should " |
173 | "be the only place they get cleaned." ); |
174 | assert(entry_val != CardTableRS::cur_youngergen_and_prev_nonclean_card, |
175 | "This should be possible in the sequential case." ); |
176 | *entry = CardTableRS::clean_card_val(); |
177 | return true; |
178 | } |
179 | |
180 | ClearNoncleanCardWrapper::ClearNoncleanCardWrapper( |
181 | DirtyCardToOopClosure* dirty_card_closure, CardTableRS* ct, bool is_par) : |
182 | _dirty_card_closure(dirty_card_closure), _ct(ct), _is_par(is_par) { |
183 | } |
184 | |
185 | bool ClearNoncleanCardWrapper::is_word_aligned(CardTable::CardValue* entry) { |
186 | return (((intptr_t)entry) & (BytesPerWord-1)) == 0; |
187 | } |
188 | |
189 | // The regions are visited in *decreasing* address order. |
190 | // This order aids with imprecise card marking, where a dirty |
191 | // card may cause scanning, and summarization marking, of objects |
192 | // that extend onto subsequent cards. |
193 | void ClearNoncleanCardWrapper::do_MemRegion(MemRegion mr) { |
194 | assert(mr.word_size() > 0, "Error" ); |
195 | assert(_ct->is_aligned(mr.start()), "mr.start() should be card aligned" ); |
196 | // mr.end() may not necessarily be card aligned. |
197 | CardValue* cur_entry = _ct->byte_for(mr.last()); |
198 | const CardValue* limit = _ct->byte_for(mr.start()); |
199 | HeapWord* end_of_non_clean = mr.end(); |
200 | HeapWord* start_of_non_clean = end_of_non_clean; |
201 | while (cur_entry >= limit) { |
202 | HeapWord* cur_hw = _ct->addr_for(cur_entry); |
203 | if ((*cur_entry != CardTableRS::clean_card_val()) && clear_card(cur_entry)) { |
204 | // Continue the dirty range by opening the |
205 | // dirty window one card to the left. |
206 | start_of_non_clean = cur_hw; |
207 | } else { |
208 | // We hit a "clean" card; process any non-empty |
209 | // "dirty" range accumulated so far. |
210 | if (start_of_non_clean < end_of_non_clean) { |
211 | const MemRegion mrd(start_of_non_clean, end_of_non_clean); |
212 | _dirty_card_closure->do_MemRegion(mrd); |
213 | } |
214 | |
215 | // fast forward through potential continuous whole-word range of clean cards beginning at a word-boundary |
216 | if (is_word_aligned(cur_entry)) { |
217 | CardValue* cur_row = cur_entry - BytesPerWord; |
218 | while (cur_row >= limit && *((intptr_t*)cur_row) == CardTableRS::clean_card_row_val()) { |
219 | cur_row -= BytesPerWord; |
220 | } |
221 | cur_entry = cur_row + BytesPerWord; |
222 | cur_hw = _ct->addr_for(cur_entry); |
223 | } |
224 | |
225 | // Reset the dirty window, while continuing to look |
226 | // for the next dirty card that will start a |
227 | // new dirty window. |
228 | end_of_non_clean = cur_hw; |
229 | start_of_non_clean = cur_hw; |
230 | } |
231 | // Note that "cur_entry" leads "start_of_non_clean" in |
232 | // its leftward excursion after this point |
233 | // in the loop and, when we hit the left end of "mr", |
234 | // will point off of the left end of the card-table |
235 | // for "mr". |
236 | cur_entry--; |
237 | } |
238 | // If the first card of "mr" was dirty, we will have |
239 | // been left with a dirty window, co-initial with "mr", |
240 | // which we now process. |
241 | if (start_of_non_clean < end_of_non_clean) { |
242 | const MemRegion mrd(start_of_non_clean, end_of_non_clean); |
243 | _dirty_card_closure->do_MemRegion(mrd); |
244 | } |
245 | } |
246 | |
247 | // clean (by dirty->clean before) ==> cur_younger_gen |
248 | // dirty ==> cur_youngergen_and_prev_nonclean_card |
249 | // precleaned ==> cur_youngergen_and_prev_nonclean_card |
250 | // prev-younger-gen ==> cur_youngergen_and_prev_nonclean_card |
251 | // cur-younger-gen ==> cur_younger_gen |
252 | // cur_youngergen_and_prev_nonclean_card ==> no change. |
253 | void CardTableRS::write_ref_field_gc_par(void* field, oop new_val) { |
254 | volatile CardValue* entry = byte_for(field); |
255 | do { |
256 | CardValue entry_val = *entry; |
257 | // We put this first because it's probably the most common case. |
258 | if (entry_val == clean_card_val()) { |
259 | // No threat of contention with cleaning threads. |
260 | *entry = cur_youngergen_card_val(); |
261 | return; |
262 | } else if (card_is_dirty_wrt_gen_iter(entry_val) |
263 | || is_prev_youngergen_card_val(entry_val)) { |
264 | // Mark it as both cur and prev youngergen; card cleaning thread will |
265 | // eventually remove the previous stuff. |
266 | CardValue new_val = cur_youngergen_and_prev_nonclean_card; |
267 | CardValue res = Atomic::cmpxchg(new_val, entry, entry_val); |
268 | // Did the CAS succeed? |
269 | if (res == entry_val) return; |
270 | // Otherwise, retry, to see the new value. |
271 | continue; |
272 | } else { |
273 | assert(entry_val == cur_youngergen_and_prev_nonclean_card |
274 | || entry_val == cur_youngergen_card_val(), |
275 | "should be only possibilities." ); |
276 | return; |
277 | } |
278 | } while (true); |
279 | } |
280 | |
281 | void CardTableRS::younger_refs_in_space_iterate(Space* sp, |
282 | OopsInGenClosure* cl, |
283 | uint n_threads) { |
284 | verify_used_region_at_save_marks(sp); |
285 | |
286 | const MemRegion urasm = sp->used_region_at_save_marks(); |
287 | non_clean_card_iterate_possibly_parallel(sp, urasm, cl, this, n_threads); |
288 | } |
289 | |
290 | #ifdef ASSERT |
291 | void CardTableRS::verify_used_region_at_save_marks(Space* sp) const { |
292 | MemRegion ur = sp->used_region(); |
293 | MemRegion urasm = sp->used_region_at_save_marks(); |
294 | |
295 | assert(ur.contains(urasm), |
296 | "Did you forget to call save_marks()? " |
297 | "[" PTR_FORMAT ", " PTR_FORMAT ") is not contained in " |
298 | "[" PTR_FORMAT ", " PTR_FORMAT ")" , |
299 | p2i(urasm.start()), p2i(urasm.end()), p2i(ur.start()), p2i(ur.end())); |
300 | } |
301 | #endif |
302 | |
303 | void CardTableRS::clear_into_younger(Generation* old_gen) { |
304 | assert(GenCollectedHeap::heap()->is_old_gen(old_gen), |
305 | "Should only be called for the old generation" ); |
306 | // The card tables for the youngest gen need never be cleared. |
307 | // There's a bit of subtlety in the clear() and invalidate() |
308 | // methods that we exploit here and in invalidate_or_clear() |
309 | // below to avoid missing cards at the fringes. If clear() or |
310 | // invalidate() are changed in the future, this code should |
311 | // be revisited. 20040107.ysr |
312 | clear(old_gen->prev_used_region()); |
313 | } |
314 | |
315 | void CardTableRS::invalidate_or_clear(Generation* old_gen) { |
316 | assert(GenCollectedHeap::heap()->is_old_gen(old_gen), |
317 | "Should only be called for the old generation" ); |
318 | // Invalidate the cards for the currently occupied part of |
319 | // the old generation and clear the cards for the |
320 | // unoccupied part of the generation (if any, making use |
321 | // of that generation's prev_used_region to determine that |
322 | // region). No need to do anything for the youngest |
323 | // generation. Also see note#20040107.ysr above. |
324 | MemRegion used_mr = old_gen->used_region(); |
325 | MemRegion to_be_cleared_mr = old_gen->prev_used_region().minus(used_mr); |
326 | if (!to_be_cleared_mr.is_empty()) { |
327 | clear(to_be_cleared_mr); |
328 | } |
329 | invalidate(used_mr); |
330 | } |
331 | |
332 | |
333 | class VerifyCleanCardClosure: public BasicOopIterateClosure { |
334 | private: |
335 | HeapWord* _boundary; |
336 | HeapWord* _begin; |
337 | HeapWord* _end; |
338 | protected: |
339 | template <class T> void do_oop_work(T* p) { |
340 | HeapWord* jp = (HeapWord*)p; |
341 | assert(jp >= _begin && jp < _end, |
342 | "Error: jp " PTR_FORMAT " should be within " |
343 | "[_begin, _end) = [" PTR_FORMAT "," PTR_FORMAT ")" , |
344 | p2i(jp), p2i(_begin), p2i(_end)); |
345 | oop obj = RawAccess<>::oop_load(p); |
346 | guarantee(obj == NULL || (HeapWord*)obj >= _boundary, |
347 | "pointer " PTR_FORMAT " at " PTR_FORMAT " on " |
348 | "clean card crosses boundary" PTR_FORMAT, |
349 | p2i(obj), p2i(jp), p2i(_boundary)); |
350 | } |
351 | |
352 | public: |
353 | VerifyCleanCardClosure(HeapWord* b, HeapWord* begin, HeapWord* end) : |
354 | _boundary(b), _begin(begin), _end(end) { |
355 | assert(b <= begin, |
356 | "Error: boundary " PTR_FORMAT " should be at or below begin " PTR_FORMAT, |
357 | p2i(b), p2i(begin)); |
358 | assert(begin <= end, |
359 | "Error: begin " PTR_FORMAT " should be strictly below end " PTR_FORMAT, |
360 | p2i(begin), p2i(end)); |
361 | } |
362 | |
363 | virtual void do_oop(oop* p) { VerifyCleanCardClosure::do_oop_work(p); } |
364 | virtual void do_oop(narrowOop* p) { VerifyCleanCardClosure::do_oop_work(p); } |
365 | }; |
366 | |
367 | class VerifyCTSpaceClosure: public SpaceClosure { |
368 | private: |
369 | CardTableRS* _ct; |
370 | HeapWord* _boundary; |
371 | public: |
372 | VerifyCTSpaceClosure(CardTableRS* ct, HeapWord* boundary) : |
373 | _ct(ct), _boundary(boundary) {} |
374 | virtual void do_space(Space* s) { _ct->verify_space(s, _boundary); } |
375 | }; |
376 | |
377 | class VerifyCTGenClosure: public GenCollectedHeap::GenClosure { |
378 | CardTableRS* _ct; |
379 | public: |
380 | VerifyCTGenClosure(CardTableRS* ct) : _ct(ct) {} |
381 | void do_generation(Generation* gen) { |
382 | // Skip the youngest generation. |
383 | if (GenCollectedHeap::heap()->is_young_gen(gen)) { |
384 | return; |
385 | } |
386 | // Normally, we're interested in pointers to younger generations. |
387 | VerifyCTSpaceClosure blk(_ct, gen->reserved().start()); |
388 | gen->space_iterate(&blk, true); |
389 | } |
390 | }; |
391 | |
392 | void CardTableRS::verify_space(Space* s, HeapWord* gen_boundary) { |
393 | // We don't need to do young-gen spaces. |
394 | if (s->end() <= gen_boundary) return; |
395 | MemRegion used = s->used_region(); |
396 | |
397 | CardValue* cur_entry = byte_for(used.start()); |
398 | CardValue* limit = byte_after(used.last()); |
399 | while (cur_entry < limit) { |
400 | if (*cur_entry == clean_card_val()) { |
401 | CardValue* first_dirty = cur_entry+1; |
402 | while (first_dirty < limit && |
403 | *first_dirty == clean_card_val()) { |
404 | first_dirty++; |
405 | } |
406 | // If the first object is a regular object, and it has a |
407 | // young-to-old field, that would mark the previous card. |
408 | HeapWord* boundary = addr_for(cur_entry); |
409 | HeapWord* end = (first_dirty >= limit) ? used.end() : addr_for(first_dirty); |
410 | HeapWord* boundary_block = s->block_start(boundary); |
411 | HeapWord* begin = boundary; // Until proven otherwise. |
412 | HeapWord* start_block = boundary_block; // Until proven otherwise. |
413 | if (boundary_block < boundary) { |
414 | if (s->block_is_obj(boundary_block) && s->obj_is_alive(boundary_block)) { |
415 | oop boundary_obj = oop(boundary_block); |
416 | if (!boundary_obj->is_objArray() && |
417 | !boundary_obj->is_typeArray()) { |
418 | guarantee(cur_entry > byte_for(used.start()), |
419 | "else boundary would be boundary_block" ); |
420 | if (*byte_for(boundary_block) != clean_card_val()) { |
421 | begin = boundary_block + s->block_size(boundary_block); |
422 | start_block = begin; |
423 | } |
424 | } |
425 | } |
426 | } |
427 | // Now traverse objects until end. |
428 | if (begin < end) { |
429 | MemRegion mr(begin, end); |
430 | VerifyCleanCardClosure verify_blk(gen_boundary, begin, end); |
431 | for (HeapWord* cur = start_block; cur < end; cur += s->block_size(cur)) { |
432 | if (s->block_is_obj(cur) && s->obj_is_alive(cur)) { |
433 | oop(cur)->oop_iterate(&verify_blk, mr); |
434 | } |
435 | } |
436 | } |
437 | cur_entry = first_dirty; |
438 | } else { |
439 | // We'd normally expect that cur_youngergen_and_prev_nonclean_card |
440 | // is a transient value, that cannot be in the card table |
441 | // except during GC, and thus assert that: |
442 | // guarantee(*cur_entry != cur_youngergen_and_prev_nonclean_card, |
443 | // "Illegal CT value"); |
444 | // That however, need not hold, as will become clear in the |
445 | // following... |
446 | |
447 | // We'd normally expect that if we are in the parallel case, |
448 | // we can't have left a prev value (which would be different |
449 | // from the current value) in the card table, and so we'd like to |
450 | // assert that: |
451 | // guarantee(cur_youngergen_card_val() == youngergen_card |
452 | // || !is_prev_youngergen_card_val(*cur_entry), |
453 | // "Illegal CT value"); |
454 | // That, however, may not hold occasionally, because of |
455 | // CMS or MSC in the old gen. To wit, consider the |
456 | // following two simple illustrative scenarios: |
457 | // (a) CMS: Consider the case where a large object L |
458 | // spanning several cards is allocated in the old |
459 | // gen, and has a young gen reference stored in it, dirtying |
460 | // some interior cards. A young collection scans the card, |
461 | // finds a young ref and installs a youngergenP_n value. |
462 | // L then goes dead. Now a CMS collection starts, |
463 | // finds L dead and sweeps it up. Assume that L is |
464 | // abutting _unallocated_blk, so _unallocated_blk is |
465 | // adjusted down to (below) L. Assume further that |
466 | // no young collection intervenes during this CMS cycle. |
467 | // The next young gen cycle will not get to look at this |
468 | // youngergenP_n card since it lies in the unoccupied |
469 | // part of the space. |
470 | // Some young collections later the blocks on this |
471 | // card can be re-allocated either due to direct allocation |
472 | // or due to absorbing promotions. At this time, the |
473 | // before-gc verification will fail the above assert. |
474 | // (b) MSC: In this case, an object L with a young reference |
475 | // is on a card that (therefore) holds a youngergen_n value. |
476 | // Suppose also that L lies towards the end of the used |
477 | // the used space before GC. An MSC collection |
478 | // occurs that compacts to such an extent that this |
479 | // card is no longer in the occupied part of the space. |
480 | // Since current code in MSC does not always clear cards |
481 | // in the unused part of old gen, this stale youngergen_n |
482 | // value is left behind and can later be covered by |
483 | // an object when promotion or direct allocation |
484 | // re-allocates that part of the heap. |
485 | // |
486 | // Fortunately, the presence of such stale card values is |
487 | // "only" a minor annoyance in that subsequent young collections |
488 | // might needlessly scan such cards, but would still never corrupt |
489 | // the heap as a result. However, it's likely not to be a significant |
490 | // performance inhibitor in practice. For instance, |
491 | // some recent measurements with unoccupied cards eagerly cleared |
492 | // out to maintain this invariant, showed next to no |
493 | // change in young collection times; of course one can construct |
494 | // degenerate examples where the cost can be significant.) |
495 | // Note, in particular, that if the "stale" card is modified |
496 | // after re-allocation, it would be dirty, not "stale". Thus, |
497 | // we can never have a younger ref in such a card and it is |
498 | // safe not to scan that card in any collection. [As we see |
499 | // below, we do some unnecessary scanning |
500 | // in some cases in the current parallel scanning algorithm.] |
501 | // |
502 | // The main point below is that the parallel card scanning code |
503 | // deals correctly with these stale card values. There are two main |
504 | // cases to consider where we have a stale "young gen" value and a |
505 | // "derivative" case to consider, where we have a stale |
506 | // "cur_younger_gen_and_prev_non_clean" value, as will become |
507 | // apparent in the case analysis below. |
508 | // o Case 1. If the stale value corresponds to a younger_gen_n |
509 | // value other than the cur_younger_gen value then the code |
510 | // treats this as being tantamount to a prev_younger_gen |
511 | // card. This means that the card may be unnecessarily scanned. |
512 | // There are two sub-cases to consider: |
513 | // o Case 1a. Let us say that the card is in the occupied part |
514 | // of the generation at the time the collection begins. In |
515 | // that case the card will be either cleared when it is scanned |
516 | // for young pointers, or will be set to cur_younger_gen as a |
517 | // result of promotion. (We have elided the normal case where |
518 | // the scanning thread and the promoting thread interleave |
519 | // possibly resulting in a transient |
520 | // cur_younger_gen_and_prev_non_clean value before settling |
521 | // to cur_younger_gen. [End Case 1a.] |
522 | // o Case 1b. Consider now the case when the card is in the unoccupied |
523 | // part of the space which becomes occupied because of promotions |
524 | // into it during the current young GC. In this case the card |
525 | // will never be scanned for young references. The current |
526 | // code will set the card value to either |
527 | // cur_younger_gen_and_prev_non_clean or leave |
528 | // it with its stale value -- because the promotions didn't |
529 | // result in any younger refs on that card. Of these two |
530 | // cases, the latter will be covered in Case 1a during |
531 | // a subsequent scan. To deal with the former case, we need |
532 | // to further consider how we deal with a stale value of |
533 | // cur_younger_gen_and_prev_non_clean in our case analysis |
534 | // below. This we do in Case 3 below. [End Case 1b] |
535 | // [End Case 1] |
536 | // o Case 2. If the stale value corresponds to cur_younger_gen being |
537 | // a value not necessarily written by a current promotion, the |
538 | // card will not be scanned by the younger refs scanning code. |
539 | // (This is OK since as we argued above such cards cannot contain |
540 | // any younger refs.) The result is that this value will be |
541 | // treated as a prev_younger_gen value in a subsequent collection, |
542 | // which is addressed in Case 1 above. [End Case 2] |
543 | // o Case 3. We here consider the "derivative" case from Case 1b. above |
544 | // because of which we may find a stale |
545 | // cur_younger_gen_and_prev_non_clean card value in the table. |
546 | // Once again, as in Case 1, we consider two subcases, depending |
547 | // on whether the card lies in the occupied or unoccupied part |
548 | // of the space at the start of the young collection. |
549 | // o Case 3a. Let us say the card is in the occupied part of |
550 | // the old gen at the start of the young collection. In that |
551 | // case, the card will be scanned by the younger refs scanning |
552 | // code which will set it to cur_younger_gen. In a subsequent |
553 | // scan, the card will be considered again and get its final |
554 | // correct value. [End Case 3a] |
555 | // o Case 3b. Now consider the case where the card is in the |
556 | // unoccupied part of the old gen, and is occupied as a result |
557 | // of promotions during thus young gc. In that case, |
558 | // the card will not be scanned for younger refs. The presence |
559 | // of newly promoted objects on the card will then result in |
560 | // its keeping the value cur_younger_gen_and_prev_non_clean |
561 | // value, which we have dealt with in Case 3 here. [End Case 3b] |
562 | // [End Case 3] |
563 | // |
564 | // (Please refer to the code in the helper class |
565 | // ClearNonCleanCardWrapper and in CardTable for details.) |
566 | // |
567 | // The informal arguments above can be tightened into a formal |
568 | // correctness proof and it behooves us to write up such a proof, |
569 | // or to use model checking to prove that there are no lingering |
570 | // concerns. |
571 | // |
572 | // Clearly because of Case 3b one cannot bound the time for |
573 | // which a card will retain what we have called a "stale" value. |
574 | // However, one can obtain a Loose upper bound on the redundant |
575 | // work as a result of such stale values. Note first that any |
576 | // time a stale card lies in the occupied part of the space at |
577 | // the start of the collection, it is scanned by younger refs |
578 | // code and we can define a rank function on card values that |
579 | // declines when this is so. Note also that when a card does not |
580 | // lie in the occupied part of the space at the beginning of a |
581 | // young collection, its rank can either decline or stay unchanged. |
582 | // In this case, no extra work is done in terms of redundant |
583 | // younger refs scanning of that card. |
584 | // Then, the case analysis above reveals that, in the worst case, |
585 | // any such stale card will be scanned unnecessarily at most twice. |
586 | // |
587 | // It is nonetheless advisable to try and get rid of some of this |
588 | // redundant work in a subsequent (low priority) re-design of |
589 | // the card-scanning code, if only to simplify the underlying |
590 | // state machine analysis/proof. ysr 1/28/2002. XXX |
591 | cur_entry++; |
592 | } |
593 | } |
594 | } |
595 | |
596 | void CardTableRS::verify() { |
597 | // At present, we only know how to verify the card table RS for |
598 | // generational heaps. |
599 | VerifyCTGenClosure blk(this); |
600 | GenCollectedHeap::heap()->generation_iterate(&blk, false); |
601 | CardTable::verify(); |
602 | } |
603 | |
604 | CardTableRS::CardTableRS(MemRegion whole_heap, bool scanned_concurrently) : |
605 | CardTable(whole_heap, scanned_concurrently), |
606 | _cur_youngergen_card_val(youngergenP1_card), |
607 | // LNC functionality |
608 | _lowest_non_clean(NULL), |
609 | _lowest_non_clean_chunk_size(NULL), |
610 | _lowest_non_clean_base_chunk_index(NULL), |
611 | _last_LNC_resizing_collection(NULL) |
612 | { |
613 | // max_gens is really GenCollectedHeap::heap()->gen_policy()->number_of_generations() |
614 | // (which is always 2, young & old), but GenCollectedHeap has not been initialized yet. |
615 | uint max_gens = 2; |
616 | _last_cur_val_in_gen = NEW_C_HEAP_ARRAY3(CardValue, max_gens + 1, |
617 | mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL); |
618 | if (_last_cur_val_in_gen == NULL) { |
619 | vm_exit_during_initialization("Could not create last_cur_val_in_gen array." ); |
620 | } |
621 | for (uint i = 0; i < max_gens + 1; i++) { |
622 | _last_cur_val_in_gen[i] = clean_card_val(); |
623 | } |
624 | } |
625 | |
626 | CardTableRS::~CardTableRS() { |
627 | if (_last_cur_val_in_gen) { |
628 | FREE_C_HEAP_ARRAY(CardValue, _last_cur_val_in_gen); |
629 | _last_cur_val_in_gen = NULL; |
630 | } |
631 | if (_lowest_non_clean) { |
632 | FREE_C_HEAP_ARRAY(CardArr, _lowest_non_clean); |
633 | _lowest_non_clean = NULL; |
634 | } |
635 | if (_lowest_non_clean_chunk_size) { |
636 | FREE_C_HEAP_ARRAY(size_t, _lowest_non_clean_chunk_size); |
637 | _lowest_non_clean_chunk_size = NULL; |
638 | } |
639 | if (_lowest_non_clean_base_chunk_index) { |
640 | FREE_C_HEAP_ARRAY(uintptr_t, _lowest_non_clean_base_chunk_index); |
641 | _lowest_non_clean_base_chunk_index = NULL; |
642 | } |
643 | if (_last_LNC_resizing_collection) { |
644 | FREE_C_HEAP_ARRAY(int, _last_LNC_resizing_collection); |
645 | _last_LNC_resizing_collection = NULL; |
646 | } |
647 | } |
648 | |
649 | void CardTableRS::initialize() { |
650 | CardTable::initialize(); |
651 | _lowest_non_clean = |
652 | NEW_C_HEAP_ARRAY(CardArr, _max_covered_regions, mtGC); |
653 | _lowest_non_clean_chunk_size = |
654 | NEW_C_HEAP_ARRAY(size_t, _max_covered_regions, mtGC); |
655 | _lowest_non_clean_base_chunk_index = |
656 | NEW_C_HEAP_ARRAY(uintptr_t, _max_covered_regions, mtGC); |
657 | _last_LNC_resizing_collection = |
658 | NEW_C_HEAP_ARRAY(int, _max_covered_regions, mtGC); |
659 | if (_lowest_non_clean == NULL |
660 | || _lowest_non_clean_chunk_size == NULL |
661 | || _lowest_non_clean_base_chunk_index == NULL |
662 | || _last_LNC_resizing_collection == NULL) |
663 | vm_exit_during_initialization("couldn't allocate an LNC array." ); |
664 | for (int i = 0; i < _max_covered_regions; i++) { |
665 | _lowest_non_clean[i] = NULL; |
666 | _lowest_non_clean_chunk_size[i] = 0; |
667 | _last_LNC_resizing_collection[i] = -1; |
668 | } |
669 | } |
670 | |
671 | bool CardTableRS::card_will_be_scanned(CardValue cv) { |
672 | return card_is_dirty_wrt_gen_iter(cv) || is_prev_nonclean_card_val(cv); |
673 | } |
674 | |
675 | bool CardTableRS::card_may_have_been_dirty(CardValue cv) { |
676 | return |
677 | cv != clean_card && |
678 | (card_is_dirty_wrt_gen_iter(cv) || |
679 | CardTableRS::youngergen_may_have_been_dirty(cv)); |
680 | } |
681 | |
682 | void CardTableRS::non_clean_card_iterate_possibly_parallel( |
683 | Space* sp, |
684 | MemRegion mr, |
685 | OopsInGenClosure* cl, |
686 | CardTableRS* ct, |
687 | uint n_threads) |
688 | { |
689 | if (!mr.is_empty()) { |
690 | if (n_threads > 0) { |
691 | non_clean_card_iterate_parallel_work(sp, mr, cl, ct, n_threads); |
692 | } else { |
693 | // clear_cl finds contiguous dirty ranges of cards to process and clear. |
694 | |
695 | // This is the single-threaded version used by DefNew. |
696 | const bool parallel = false; |
697 | |
698 | DirtyCardToOopClosure* dcto_cl = sp->new_dcto_cl(cl, precision(), cl->gen_boundary(), parallel); |
699 | ClearNoncleanCardWrapper clear_cl(dcto_cl, ct, parallel); |
700 | |
701 | clear_cl.do_MemRegion(mr); |
702 | } |
703 | } |
704 | } |
705 | |
706 | void CardTableRS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, |
707 | OopsInGenClosure* cl, CardTableRS* ct, |
708 | uint n_threads) { |
709 | fatal("Parallel gc not supported here." ); |
710 | } |
711 | |
712 | bool CardTableRS::is_in_young(oop obj) const { |
713 | return GenCollectedHeap::heap()->is_in_young(obj); |
714 | } |
715 | |