Tue, 10 May 2011 00:33:21 -0700
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
1 /*
2 * Copyright (c) 2000, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #ifndef SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP
26 #define SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP
28 #include "memory/modRefBarrierSet.hpp"
29 #include "oops/oop.hpp"
30 #include "oops/oop.inline2.hpp"
32 // This kind of "BarrierSet" allows a "CollectedHeap" to detect and
33 // enumerate ref fields that have been modified (since the last
34 // enumeration.)
36 // As it currently stands, this barrier is *imprecise*: when a ref field in
37 // an object "o" is modified, the card table entry for the card containing
38 // the head of "o" is dirtied, not necessarily the card containing the
39 // modified field itself. For object arrays, however, the barrier *is*
40 // precise; only the card containing the modified element is dirtied.
41 // Any MemRegionClosures used to scan dirty cards should take these
42 // considerations into account.
44 class Generation;
45 class OopsInGenClosure;
46 class DirtyCardToOopClosure;
47 class ClearNoncleanCardWrapper;
49 class CardTableModRefBS: public ModRefBarrierSet {
50 // Some classes get to look at some private stuff.
51 friend class BytecodeInterpreter;
52 friend class VMStructs;
53 friend class CardTableRS;
54 friend class CheckForUnmarkedOops; // Needs access to raw card bytes.
55 friend class SharkBuilder;
56 #ifndef PRODUCT
57 // For debugging.
58 friend class GuaranteeNotModClosure;
59 #endif
60 protected:
62 enum CardValues {
63 clean_card = -1,
64 // The mask contains zeros in places for all other values.
65 clean_card_mask = clean_card - 31,
67 dirty_card = 0,
68 precleaned_card = 1,
69 claimed_card = 2,
70 deferred_card = 4,
71 last_card = 8,
72 CT_MR_BS_last_reserved = 16
73 };
75 // dirty and precleaned are equivalent wrt younger_refs_iter.
76 static bool card_is_dirty_wrt_gen_iter(jbyte cv) {
77 return cv == dirty_card || cv == precleaned_card;
78 }
80 // Returns "true" iff the value "cv" will cause the card containing it
81 // to be scanned in the current traversal. May be overridden by
82 // subtypes.
83 virtual bool card_will_be_scanned(jbyte cv) {
84 return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv);
85 }
87 // Returns "true" iff the value "cv" may have represented a dirty card at
88 // some point.
89 virtual bool card_may_have_been_dirty(jbyte cv) {
90 return card_is_dirty_wrt_gen_iter(cv);
91 }
93 // The declaration order of these const fields is important; see the
94 // constructor before changing.
95 const MemRegion _whole_heap; // the region covered by the card table
96 const size_t _guard_index; // index of very last element in the card
97 // table; it is set to a guard value
98 // (last_card) and should never be modified
99 const size_t _last_valid_index; // index of the last valid element
100 const size_t _page_size; // page size used when mapping _byte_map
101 const size_t _byte_map_size; // in bytes
102 jbyte* _byte_map; // the card marking array
104 int _cur_covered_regions;
105 // The covered regions should be in address order.
106 MemRegion* _covered;
107 // The committed regions correspond one-to-one to the covered regions.
108 // They represent the card-table memory that has been committed to service
109 // the corresponding covered region. It may be that committed region for
110 // one covered region corresponds to a larger region because of page-size
111 // roundings. Thus, a committed region for one covered region may
112 // actually extend onto the card-table space for the next covered region.
113 MemRegion* _committed;
115 // The last card is a guard card, and we commit the page for it so
116 // we can use the card for verification purposes. We make sure we never
117 // uncommit the MemRegion for that page.
118 MemRegion _guard_region;
120 protected:
121 // Initialization utilities; covered_words is the size of the covered region
122 // in, um, words.
123 inline size_t cards_required(size_t covered_words);
124 inline size_t compute_byte_map_size();
126 // Finds and return the index of the region, if any, to which the given
127 // region would be contiguous. If none exists, assign a new region and
128 // returns its index. Requires that no more than the maximum number of
129 // covered regions defined in the constructor are ever in use.
130 int find_covering_region_by_base(HeapWord* base);
132 // Same as above, but finds the region containing the given address
133 // instead of starting at a given base address.
134 int find_covering_region_containing(HeapWord* addr);
136 // Resize one of the regions covered by the remembered set.
137 void resize_covered_region(MemRegion new_region);
139 // Returns the leftmost end of a committed region corresponding to a
140 // covered region before covered region "ind", or else "NULL" if "ind" is
141 // the first covered region.
142 HeapWord* largest_prev_committed_end(int ind) const;
144 // Returns the part of the region mr that doesn't intersect with
145 // any committed region other than self. Used to prevent uncommitting
146 // regions that are also committed by other regions. Also protects
147 // against uncommitting the guard region.
148 MemRegion committed_unique_to_self(int self, MemRegion mr) const;
150 // Mapping from address to card marking array entry
151 jbyte* byte_for(const void* p) const {
152 assert(_whole_heap.contains(p),
153 "out of bounds access to card marking array");
154 jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift];
155 assert(result >= _byte_map && result < _byte_map + _byte_map_size,
156 "out of bounds accessor for card marking array");
157 return result;
158 }
160 // The card table byte one after the card marking array
161 // entry for argument address. Typically used for higher bounds
162 // for loops iterating through the card table.
163 jbyte* byte_after(const void* p) const {
164 return byte_for(p) + 1;
165 }
167 // Iterate over the portion of the card-table which covers the given
168 // region mr in the given space and apply cl to any dirty sub-regions
169 // of mr. Dirty cards are _not_ cleared by the iterator method itself,
170 // but closures may arrange to do so on their own should they so wish.
171 void non_clean_card_iterate_serial(MemRegion mr, MemRegionClosure* cl);
173 // A variant of the above that will operate in a parallel mode if
174 // worker threads are available, and clear the dirty cards as it
175 // processes them.
176 // XXX ??? MemRegionClosure above vs OopsInGenClosure below XXX
177 // XXX some new_dcto_cl's take OopClosure's, plus as above there are
178 // some MemRegionClosures. Clean this up everywhere. XXX
179 void non_clean_card_iterate_possibly_parallel(Space* sp, MemRegion mr,
180 OopsInGenClosure* cl, CardTableRS* ct);
182 private:
183 // Work method used to implement non_clean_card_iterate_possibly_parallel()
184 // above in the parallel case.
185 void non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr,
186 OopsInGenClosure* cl, CardTableRS* ct,
187 int n_threads);
189 protected:
190 // Dirty the bytes corresponding to "mr" (not all of which must be
191 // covered.)
192 void dirty_MemRegion(MemRegion mr);
194 // Clear (to clean_card) the bytes entirely contained within "mr" (not
195 // all of which must be covered.)
196 void clear_MemRegion(MemRegion mr);
198 // *** Support for parallel card scanning.
200 // This is an array, one element per covered region of the card table.
201 // Each entry is itself an array, with one element per chunk in the
202 // covered region. Each entry of these arrays is the lowest non-clean
203 // card of the corresponding chunk containing part of an object from the
204 // previous chunk, or else NULL.
205 typedef jbyte* CardPtr;
206 typedef CardPtr* CardArr;
207 CardArr* _lowest_non_clean;
208 size_t* _lowest_non_clean_chunk_size;
209 uintptr_t* _lowest_non_clean_base_chunk_index;
210 int* _last_LNC_resizing_collection;
212 // Initializes "lowest_non_clean" to point to the array for the region
213 // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk
214 // index of the corresponding to the first element of that array.
215 // Ensures that these arrays are of sufficient size, allocating if necessary.
216 // May be called by several threads concurrently.
217 void get_LNC_array_for_space(Space* sp,
218 jbyte**& lowest_non_clean,
219 uintptr_t& lowest_non_clean_base_chunk_index,
220 size_t& lowest_non_clean_chunk_size);
222 // Returns the number of chunks necessary to cover "mr".
223 size_t chunks_to_cover(MemRegion mr) {
224 return (size_t)(addr_to_chunk_index(mr.last()) -
225 addr_to_chunk_index(mr.start()) + 1);
226 }
228 // Returns the index of the chunk in a stride which
229 // covers the given address.
230 uintptr_t addr_to_chunk_index(const void* addr) {
231 uintptr_t card = (uintptr_t) byte_for(addr);
232 return card / ParGCCardsPerStrideChunk;
233 }
235 // Apply cl, which must either itself apply dcto_cl or be dcto_cl,
236 // to the cards in the stride (of n_strides) within the given space.
237 void process_stride(Space* sp,
238 MemRegion used,
239 jint stride, int n_strides,
240 OopsInGenClosure* cl,
241 CardTableRS* ct,
242 jbyte** lowest_non_clean,
243 uintptr_t lowest_non_clean_base_chunk_index,
244 size_t lowest_non_clean_chunk_size);
246 // Makes sure that chunk boundaries are handled appropriately, by
247 // adjusting the min_done of dcto_cl, and by using a special card-table
248 // value to indicate how min_done should be set.
249 void process_chunk_boundaries(Space* sp,
250 DirtyCardToOopClosure* dcto_cl,
251 MemRegion chunk_mr,
252 MemRegion used,
253 jbyte** lowest_non_clean,
254 uintptr_t lowest_non_clean_base_chunk_index,
255 size_t lowest_non_clean_chunk_size);
257 public:
258 // Constants
259 enum SomePublicConstants {
260 card_shift = 9,
261 card_size = 1 << card_shift,
262 card_size_in_words = card_size / sizeof(HeapWord)
263 };
265 static int clean_card_val() { return clean_card; }
266 static int clean_card_mask_val() { return clean_card_mask; }
267 static int dirty_card_val() { return dirty_card; }
268 static int claimed_card_val() { return claimed_card; }
269 static int precleaned_card_val() { return precleaned_card; }
270 static int deferred_card_val() { return deferred_card; }
272 // For RTTI simulation.
273 bool is_a(BarrierSet::Name bsn) {
274 return bsn == BarrierSet::CardTableModRef || ModRefBarrierSet::is_a(bsn);
275 }
277 CardTableModRefBS(MemRegion whole_heap, int max_covered_regions);
279 // *** Barrier set functions.
281 bool has_write_ref_pre_barrier() { return false; }
283 inline bool write_ref_needs_barrier(void* field, oop new_val) {
284 // Note that this assumes the perm gen is the highest generation
285 // in the address space
286 return new_val != NULL && !new_val->is_perm();
287 }
289 // Record a reference update. Note that these versions are precise!
290 // The scanning code has to handle the fact that the write barrier may be
291 // either precise or imprecise. We make non-virtual inline variants of
292 // these functions here for performance.
293 protected:
294 void write_ref_field_work(oop obj, size_t offset, oop newVal);
295 virtual void write_ref_field_work(void* field, oop newVal);
296 public:
298 bool has_write_ref_array_opt() { return true; }
299 bool has_write_region_opt() { return true; }
301 inline void inline_write_region(MemRegion mr) {
302 dirty_MemRegion(mr);
303 }
304 protected:
305 void write_region_work(MemRegion mr) {
306 inline_write_region(mr);
307 }
308 public:
310 inline void inline_write_ref_array(MemRegion mr) {
311 dirty_MemRegion(mr);
312 }
313 protected:
314 void write_ref_array_work(MemRegion mr) {
315 inline_write_ref_array(mr);
316 }
317 public:
319 bool is_aligned(HeapWord* addr) {
320 return is_card_aligned(addr);
321 }
323 // *** Card-table-barrier-specific things.
325 template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {}
327 template <class T> inline void inline_write_ref_field(T* field, oop newVal) {
328 jbyte* byte = byte_for((void*)field);
329 *byte = dirty_card;
330 }
332 // These are used by G1, when it uses the card table as a temporary data
333 // structure for card claiming.
334 bool is_card_dirty(size_t card_index) {
335 return _byte_map[card_index] == dirty_card_val();
336 }
338 void mark_card_dirty(size_t card_index) {
339 _byte_map[card_index] = dirty_card_val();
340 }
342 bool is_card_claimed(size_t card_index) {
343 jbyte val = _byte_map[card_index];
344 return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val();
345 }
347 void set_card_claimed(size_t card_index) {
348 jbyte val = _byte_map[card_index];
349 if (val == clean_card_val()) {
350 val = (jbyte)claimed_card_val();
351 } else {
352 val |= (jbyte)claimed_card_val();
353 }
354 _byte_map[card_index] = val;
355 }
357 bool claim_card(size_t card_index);
359 bool is_card_clean(size_t card_index) {
360 return _byte_map[card_index] == clean_card_val();
361 }
363 bool is_card_deferred(size_t card_index) {
364 jbyte val = _byte_map[card_index];
365 return (val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val();
366 }
368 bool mark_card_deferred(size_t card_index);
370 // Card marking array base (adjusted for heap low boundary)
371 // This would be the 0th element of _byte_map, if the heap started at 0x0.
372 // But since the heap starts at some higher address, this points to somewhere
373 // before the beginning of the actual _byte_map.
374 jbyte* byte_map_base;
376 // Return true if "p" is at the start of a card.
377 bool is_card_aligned(HeapWord* p) {
378 jbyte* pcard = byte_for(p);
379 return (addr_for(pcard) == p);
380 }
382 HeapWord* align_to_card_boundary(HeapWord* p) {
383 jbyte* pcard = byte_for(p + card_size_in_words - 1);
384 return addr_for(pcard);
385 }
387 // The kinds of precision a CardTableModRefBS may offer.
388 enum PrecisionStyle {
389 Precise,
390 ObjHeadPreciseArray
391 };
393 // Tells what style of precision this card table offers.
394 PrecisionStyle precision() {
395 return ObjHeadPreciseArray; // Only one supported for now.
396 }
398 // ModRefBS functions.
399 virtual void invalidate(MemRegion mr, bool whole_heap = false);
400 void clear(MemRegion mr);
401 void dirty(MemRegion mr);
403 // *** Card-table-RemSet-specific things.
405 // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively
406 // includes all the modified cards (expressing each card as a
407 // MemRegion). Thus, several modified cards may be lumped into one
408 // region. The regions are non-overlapping, and are visited in
409 // *decreasing* address order. (This order aids with imprecise card
410 // marking, where a dirty card may cause scanning, and summarization
411 // marking, of objects that extend onto subsequent cards.)
412 void mod_card_iterate(MemRegionClosure* cl) {
413 non_clean_card_iterate_serial(_whole_heap, cl);
414 }
416 // Like the "mod_cards_iterate" above, except only invokes the closure
417 // for cards within the MemRegion "mr" (which is required to be
418 // card-aligned and sized.)
419 void mod_card_iterate(MemRegion mr, MemRegionClosure* cl) {
420 non_clean_card_iterate_serial(mr, cl);
421 }
423 static uintx ct_max_alignment_constraint();
425 // Apply closure "cl" to the dirty cards containing some part of
426 // MemRegion "mr".
427 void dirty_card_iterate(MemRegion mr, MemRegionClosure* cl);
429 // Return the MemRegion corresponding to the first maximal run
430 // of dirty cards lying completely within MemRegion mr.
431 // If reset is "true", then sets those card table entries to the given
432 // value.
433 MemRegion dirty_card_range_after_reset(MemRegion mr, bool reset,
434 int reset_val);
436 // Set all the dirty cards in the given region to precleaned state.
437 void preclean_dirty_cards(MemRegion mr);
439 // Provide read-only access to the card table array.
440 const jbyte* byte_for_const(const void* p) const {
441 return byte_for(p);
442 }
443 const jbyte* byte_after_const(const void* p) const {
444 return byte_after(p);
445 }
447 // Mapping from card marking array entry to address of first word
448 HeapWord* addr_for(const jbyte* p) const {
449 assert(p >= _byte_map && p < _byte_map + _byte_map_size,
450 "out of bounds access to card marking array");
451 size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte));
452 HeapWord* result = (HeapWord*) (delta << card_shift);
453 assert(_whole_heap.contains(result),
454 "out of bounds accessor from card marking array");
455 return result;
456 }
458 // Mapping from address to card marking array index.
459 size_t index_for(void* p) {
460 assert(_whole_heap.contains(p),
461 "out of bounds access to card marking array");
462 return byte_for(p) - _byte_map;
463 }
465 const jbyte* byte_for_index(const size_t card_index) const {
466 return _byte_map + card_index;
467 }
469 void verify();
470 void verify_guard();
472 // val_equals -> it will check that all cards covered by mr equal val
473 // !val_equals -> it will check that all cards covered by mr do not equal val
474 void verify_region(MemRegion mr, jbyte val, bool val_equals) PRODUCT_RETURN;
475 void verify_not_dirty_region(MemRegion mr) PRODUCT_RETURN;
476 void verify_dirty_region(MemRegion mr) PRODUCT_RETURN;
478 static size_t par_chunk_heapword_alignment() {
479 return ParGCCardsPerStrideChunk * card_size_in_words;
480 }
482 };
484 class CardTableRS;
486 // A specialization for the CardTableRS gen rem set.
487 class CardTableModRefBSForCTRS: public CardTableModRefBS {
488 CardTableRS* _rs;
489 protected:
490 bool card_will_be_scanned(jbyte cv);
491 bool card_may_have_been_dirty(jbyte cv);
492 public:
493 CardTableModRefBSForCTRS(MemRegion whole_heap,
494 int max_covered_regions) :
495 CardTableModRefBS(whole_heap, max_covered_regions) {}
497 void set_CTRS(CardTableRS* rs) { _rs = rs; }
498 };
501 #endif // SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP