Thu, 12 Jan 2012 00:06:47 -0800
6484965: G1: piggy-back liveness accounting phase on marking
Summary: Remove the separate counting phase of concurrent marking by tracking the amount of marked bytes and the cards spanned by marked objects in marking task/worker thread local data structures, which are updated as individual objects are marked.
Reviewed-by: brutisso, tonyp
1 /*
2 * Copyright (c) 2001, 2012, 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_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP
26 #define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP
28 #include "gc_implementation/g1/concurrentMark.hpp"
29 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
31 // Returns the index in the liveness accounting card bitmap
32 // for the given address
33 inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) {
34 // Below, the term "card num" means the result of shifting an address
35 // by the card shift -- address 0 corresponds to card number 0. One
36 // must subtract the card num of the bottom of the heap to obtain a
37 // card table index.
39 intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift);
40 return card_num - heap_bottom_card_num();
41 }
43 // Counts the given memory region in the given task/worker
44 // counting data structures.
45 inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr,
46 size_t* marked_bytes_array,
47 BitMap* task_card_bm) {
48 G1CollectedHeap* g1h = _g1h;
49 HeapWord* start = mr.start();
50 HeapWord* last = mr.last();
51 size_t region_size_bytes = mr.byte_size();
52 size_t index = hr->hrs_index();
54 assert(!hr->continuesHumongous(), "should not be HC region");
55 assert(hr == g1h->heap_region_containing(start), "sanity");
56 assert(hr == g1h->heap_region_containing(mr.last()), "sanity");
57 assert(marked_bytes_array != NULL, "pre-condition");
58 assert(task_card_bm != NULL, "pre-condition");
60 // Add to the task local marked bytes for this region.
61 marked_bytes_array[index] += region_size_bytes;
63 BitMap::idx_t start_idx = card_bitmap_index_for(start);
64 BitMap::idx_t last_idx = card_bitmap_index_for(last);
66 // The card bitmap is task/worker specific => no need to use 'par' routines.
67 // Set bits in the inclusive bit range [start_idx, last_idx].
68 //
69 // For small ranges use a simple loop; otherwise use set_range
70 // The range are the cards that are spanned by the object/region
71 // so 8 cards will allow objects/regions up to 4K to be handled
72 // using the loop.
73 if ((last_idx - start_idx) <= 8) {
74 for (BitMap::idx_t i = start_idx; i <= last_idx; i += 1) {
75 task_card_bm->set_bit(i);
76 }
77 } else {
78 assert(last_idx < task_card_bm->size(), "sanity");
79 // Note: BitMap::set_range() is exclusive.
80 task_card_bm->set_range(start_idx, last_idx+1);
81 }
82 }
84 // Counts the given memory region, which may be a single object, in the
85 // task/worker counting data structures for the given worker id.
86 inline void ConcurrentMark::count_region(MemRegion mr, uint worker_id) {
87 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id);
88 BitMap* task_card_bm = count_card_bitmap_for(worker_id);
89 HeapWord* addr = mr.start();
90 HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
91 count_region(mr, hr, marked_bytes_array, task_card_bm);
92 }
94 // Counts the given object in the given task/worker counting data structures.
95 inline void ConcurrentMark::count_object(oop obj,
96 HeapRegion* hr,
97 size_t* marked_bytes_array,
98 BitMap* task_card_bm) {
99 MemRegion mr((HeapWord*)obj, obj->size());
100 count_region(mr, hr, marked_bytes_array, task_card_bm);
101 }
103 // Counts the given object in the task/worker counting data
104 // structures for the given worker id.
105 inline void ConcurrentMark::count_object(oop obj, HeapRegion* hr, uint worker_id) {
106 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id);
107 BitMap* task_card_bm = count_card_bitmap_for(worker_id);
108 HeapWord* addr = (HeapWord*) obj;
109 count_object(obj, hr, marked_bytes_array, task_card_bm);
110 }
112 // Attempts to mark the given object and, if successful, counts
113 // the object in the given task/worker counting structures.
114 inline bool ConcurrentMark::par_mark_and_count(oop obj,
115 HeapRegion* hr,
116 size_t* marked_bytes_array,
117 BitMap* task_card_bm) {
118 HeapWord* addr = (HeapWord*)obj;
119 if (_nextMarkBitMap->parMark(addr)) {
120 // Update the task specific count data for the object.
121 count_object(obj, hr, marked_bytes_array, task_card_bm);
122 return true;
123 }
124 return false;
125 }
127 // Attempts to mark the given object and, if successful, counts
128 // the object in the task/worker counting structures for the
129 // given worker id.
130 inline bool ConcurrentMark::par_mark_and_count(oop obj,
131 HeapRegion* hr,
132 uint worker_id) {
133 HeapWord* addr = (HeapWord*)obj;
134 if (_nextMarkBitMap->parMark(addr)) {
135 // Update the task specific count data for the object.
136 count_object(obj, hr, worker_id);
137 return true;
138 }
139 return false;
140 }
142 // As above - but we don't know the heap region containing the
143 // object and so have to supply it.
144 inline bool ConcurrentMark::par_mark_and_count(oop obj, uint worker_id) {
145 HeapWord* addr = (HeapWord*)obj;
146 HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
147 return par_mark_and_count(obj, hr, worker_id);
148 }
150 // Similar to the above routine but we already know the size, in words, of
151 // the object that we wish to mark/count
152 inline bool ConcurrentMark::par_mark_and_count(oop obj,
153 size_t word_size,
154 uint worker_id) {
155 HeapWord* addr = (HeapWord*)obj;
156 if (_nextMarkBitMap->parMark(addr)) {
157 // Update the task specific count data for the object.
158 MemRegion mr(addr, word_size);
159 count_region(mr, worker_id);
160 return true;
161 }
162 return false;
163 }
165 // Unconditionally mark the given object, and unconditinally count
166 // the object in the counting structures for worker id 0.
167 // Should *not* be called from parallel code.
168 inline bool ConcurrentMark::mark_and_count(oop obj, HeapRegion* hr) {
169 HeapWord* addr = (HeapWord*)obj;
170 _nextMarkBitMap->mark(addr);
171 // Update the task specific count data for the object.
172 count_object(obj, hr, 0 /* worker_id */);
173 return true;
174 }
176 // As above - but we don't have the heap region containing the
177 // object, so we have to supply it.
178 inline bool ConcurrentMark::mark_and_count(oop obj) {
179 HeapWord* addr = (HeapWord*)obj;
180 HeapRegion* hr = _g1h->heap_region_containing_raw(addr);
181 return mark_and_count(obj, hr);
182 }
184 inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) {
185 HeapWord* start_addr = MAX2(startWord(), mr.start());
186 HeapWord* end_addr = MIN2(endWord(), mr.end());
188 if (end_addr > start_addr) {
189 // Right-open interval [start-offset, end-offset).
190 BitMap::idx_t start_offset = heapWordToOffset(start_addr);
191 BitMap::idx_t end_offset = heapWordToOffset(end_addr);
193 start_offset = _bm.get_next_one_offset(start_offset, end_offset);
194 while (start_offset < end_offset) {
195 HeapWord* obj_addr = offsetToHeapWord(start_offset);
196 oop obj = (oop) obj_addr;
197 if (!cl->do_bit(start_offset)) {
198 return false;
199 }
200 HeapWord* next_addr = MIN2(obj_addr + obj->size(), end_addr);
201 BitMap::idx_t next_offset = heapWordToOffset(next_addr);
202 start_offset = _bm.get_next_one_offset(next_offset, end_offset);
203 }
204 }
205 return true;
206 }
208 inline bool CMBitMapRO::iterate(BitMapClosure* cl) {
209 MemRegion mr(startWord(), sizeInWords());
210 return iterate(cl, mr);
211 }
213 inline void CMTask::push(oop obj) {
214 HeapWord* objAddr = (HeapWord*) obj;
215 assert(_g1h->is_in_g1_reserved(objAddr), "invariant");
216 assert(!_g1h->is_on_master_free_list(
217 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant");
218 assert(!_g1h->is_obj_ill(obj), "invariant");
219 assert(_nextMarkBitMap->isMarked(objAddr), "invariant");
221 if (_cm->verbose_high()) {
222 gclog_or_tty->print_cr("[%d] pushing "PTR_FORMAT, _task_id, (void*) obj);
223 }
225 if (!_task_queue->push(obj)) {
226 // The local task queue looks full. We need to push some entries
227 // to the global stack.
229 if (_cm->verbose_medium()) {
230 gclog_or_tty->print_cr("[%d] task queue overflow, "
231 "moving entries to the global stack",
232 _task_id);
233 }
234 move_entries_to_global_stack();
236 // this should succeed since, even if we overflow the global
237 // stack, we should have definitely removed some entries from the
238 // local queue. So, there must be space on it.
239 bool success = _task_queue->push(obj);
240 assert(success, "invariant");
241 }
243 statsOnly( int tmp_size = _task_queue->size();
244 if (tmp_size > _local_max_size) {
245 _local_max_size = tmp_size;
246 }
247 ++_local_pushes );
248 }
250 // This determines whether the method below will check both the local
251 // and global fingers when determining whether to push on the stack a
252 // gray object (value 1) or whether it will only check the global one
253 // (value 0). The tradeoffs are that the former will be a bit more
254 // accurate and possibly push less on the stack, but it might also be
255 // a little bit slower.
257 #define _CHECK_BOTH_FINGERS_ 1
259 inline void CMTask::deal_with_reference(oop obj) {
260 if (_cm->verbose_high()) {
261 gclog_or_tty->print_cr("[%d] we're dealing with reference = "PTR_FORMAT,
262 _task_id, (void*) obj);
263 }
265 ++_refs_reached;
267 HeapWord* objAddr = (HeapWord*) obj;
268 assert(obj->is_oop_or_null(true /* ignore mark word */), "Error");
269 if (_g1h->is_in_g1_reserved(objAddr)) {
270 assert(obj != NULL, "null check is implicit");
271 if (!_nextMarkBitMap->isMarked(objAddr)) {
272 // Only get the containing region if the object is not marked on the
273 // bitmap (otherwise, it's a waste of time since we won't do
274 // anything with it).
275 HeapRegion* hr = _g1h->heap_region_containing_raw(obj);
276 if (!hr->obj_allocated_since_next_marking(obj)) {
277 if (_cm->verbose_high()) {
278 gclog_or_tty->print_cr("[%d] "PTR_FORMAT" is not considered marked",
279 _task_id, (void*) obj);
280 }
282 // we need to mark it first
283 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) {
284 // No OrderAccess:store_load() is needed. It is implicit in the
285 // CAS done in CMBitMap::parMark() call in the routine above.
286 HeapWord* global_finger = _cm->finger();
288 #if _CHECK_BOTH_FINGERS_
289 // we will check both the local and global fingers
291 if (_finger != NULL && objAddr < _finger) {
292 if (_cm->verbose_high()) {
293 gclog_or_tty->print_cr("[%d] below the local finger ("PTR_FORMAT"), "
294 "pushing it", _task_id, _finger);
295 }
296 push(obj);
297 } else if (_curr_region != NULL && objAddr < _region_limit) {
298 // do nothing
299 } else if (objAddr < global_finger) {
300 // Notice that the global finger might be moving forward
301 // concurrently. This is not a problem. In the worst case, we
302 // mark the object while it is above the global finger and, by
303 // the time we read the global finger, it has moved forward
304 // passed this object. In this case, the object will probably
305 // be visited when a task is scanning the region and will also
306 // be pushed on the stack. So, some duplicate work, but no
307 // correctness problems.
309 if (_cm->verbose_high()) {
310 gclog_or_tty->print_cr("[%d] below the global finger "
311 "("PTR_FORMAT"), pushing it",
312 _task_id, global_finger);
313 }
314 push(obj);
315 } else {
316 // do nothing
317 }
318 #else // _CHECK_BOTH_FINGERS_
319 // we will only check the global finger
321 if (objAddr < global_finger) {
322 // see long comment above
324 if (_cm->verbose_high()) {
325 gclog_or_tty->print_cr("[%d] below the global finger "
326 "("PTR_FORMAT"), pushing it",
327 _task_id, global_finger);
328 }
329 push(obj);
330 }
331 #endif // _CHECK_BOTH_FINGERS_
332 }
333 }
334 }
335 }
336 }
338 inline void ConcurrentMark::markPrev(oop p) {
339 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity");
340 // Note we are overriding the read-only view of the prev map here, via
341 // the cast.
342 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p);
343 }
345 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, uint worker_id) {
346 HeapWord* addr = (HeapWord*) obj;
348 // Currently we don't do anything with word_size but we will use it
349 // in the very near future in the liveness calculation piggy-backing
350 // changes.
352 #ifdef ASSERT
353 HeapRegion* hr = _g1h->heap_region_containing(addr);
354 assert(hr != NULL, "sanity");
355 assert(!hr->is_survivor(), "should not allocate survivors during IM");
356 assert(addr < hr->next_top_at_mark_start(),
357 err_msg("addr: "PTR_FORMAT" hr: "HR_FORMAT" NTAMS: "PTR_FORMAT,
358 addr, HR_FORMAT_PARAMS(hr), hr->next_top_at_mark_start()));
359 // We cannot assert that word_size == obj->size() given that obj
360 // might not be in a consistent state (another thread might be in
361 // the process of copying it). So the best thing we can do is to
362 // assert that word_size is under an upper bound which is its
363 // containing region's capacity.
364 assert(word_size * HeapWordSize <= hr->capacity(),
365 err_msg("size: "SIZE_FORMAT" capacity: "SIZE_FORMAT" "HR_FORMAT,
366 word_size * HeapWordSize, hr->capacity(),
367 HR_FORMAT_PARAMS(hr)));
368 #endif // ASSERT
370 if (!_nextMarkBitMap->isMarked(addr)) {
371 par_mark_and_count(obj, word_size, worker_id);
372 }
373 }
375 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP