|
1 /* |
|
2 * Copyright (c) 2005, 2013, Oracle and/or its affiliates. All rights reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. |
|
8 * |
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
12 * version 2 for more details (a copy is included in the LICENSE file that |
|
13 * accompanied this code). |
|
14 * |
|
15 * You should have received a copy of the GNU General Public License version |
|
16 * 2 along with this work; if not, write to the Free Software Foundation, |
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
18 * |
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
20 * or visit www.oracle.com if you need additional information or have any |
|
21 * questions. |
|
22 * |
|
23 */ |
|
24 |
|
25 #ifndef SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP |
|
26 #define SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP |
|
27 |
|
28 #include "gc_implementation/parallelScavenge/objectStartArray.hpp" |
|
29 #include "gc_implementation/parallelScavenge/parMarkBitMap.hpp" |
|
30 #include "gc_implementation/parallelScavenge/psCompactionManager.hpp" |
|
31 #include "gc_implementation/shared/collectorCounters.hpp" |
|
32 #include "gc_implementation/shared/markSweep.hpp" |
|
33 #include "gc_implementation/shared/mutableSpace.hpp" |
|
34 #include "memory/sharedHeap.hpp" |
|
35 #include "oops/oop.hpp" |
|
36 |
|
37 class ParallelScavengeHeap; |
|
38 class PSAdaptiveSizePolicy; |
|
39 class PSYoungGen; |
|
40 class PSOldGen; |
|
41 class ParCompactionManager; |
|
42 class ParallelTaskTerminator; |
|
43 class PSParallelCompact; |
|
44 class GCTaskManager; |
|
45 class GCTaskQueue; |
|
46 class PreGCValues; |
|
47 class MoveAndUpdateClosure; |
|
48 class RefProcTaskExecutor; |
|
49 class ParallelOldTracer; |
|
50 class STWGCTimer; |
|
51 |
|
52 // The SplitInfo class holds the information needed to 'split' a source region |
|
53 // so that the live data can be copied to two destination *spaces*. Normally, |
|
54 // all the live data in a region is copied to a single destination space (e.g., |
|
55 // everything live in a region in eden is copied entirely into the old gen). |
|
56 // However, when the heap is nearly full, all the live data in eden may not fit |
|
57 // into the old gen. Copying only some of the regions from eden to old gen |
|
58 // requires finding a region that does not contain a partial object (i.e., no |
|
59 // live object crosses the region boundary) somewhere near the last object that |
|
60 // does fit into the old gen. Since it's not always possible to find such a |
|
61 // region, splitting is necessary for predictable behavior. |
|
62 // |
|
63 // A region is always split at the end of the partial object. This avoids |
|
64 // additional tests when calculating the new location of a pointer, which is a |
|
65 // very hot code path. The partial object and everything to its left will be |
|
66 // copied to another space (call it dest_space_1). The live data to the right |
|
67 // of the partial object will be copied either within the space itself, or to a |
|
68 // different destination space (distinct from dest_space_1). |
|
69 // |
|
70 // Split points are identified during the summary phase, when region |
|
71 // destinations are computed: data about the split, including the |
|
72 // partial_object_size, is recorded in a SplitInfo record and the |
|
73 // partial_object_size field in the summary data is set to zero. The zeroing is |
|
74 // possible (and necessary) since the partial object will move to a different |
|
75 // destination space than anything to its right, thus the partial object should |
|
76 // not affect the locations of any objects to its right. |
|
77 // |
|
78 // The recorded data is used during the compaction phase, but only rarely: when |
|
79 // the partial object on the split region will be copied across a destination |
|
80 // region boundary. This test is made once each time a region is filled, and is |
|
81 // a simple address comparison, so the overhead is negligible (see |
|
82 // PSParallelCompact::first_src_addr()). |
|
83 // |
|
84 // Notes: |
|
85 // |
|
86 // Only regions with partial objects are split; a region without a partial |
|
87 // object does not need any extra bookkeeping. |
|
88 // |
|
89 // At most one region is split per space, so the amount of data required is |
|
90 // constant. |
|
91 // |
|
92 // A region is split only when the destination space would overflow. Once that |
|
93 // happens, the destination space is abandoned and no other data (even from |
|
94 // other source spaces) is targeted to that destination space. Abandoning the |
|
95 // destination space may leave a somewhat large unused area at the end, if a |
|
96 // large object caused the overflow. |
|
97 // |
|
98 // Future work: |
|
99 // |
|
100 // More bookkeeping would be required to continue to use the destination space. |
|
101 // The most general solution would allow data from regions in two different |
|
102 // source spaces to be "joined" in a single destination region. At the very |
|
103 // least, additional code would be required in next_src_region() to detect the |
|
104 // join and skip to an out-of-order source region. If the join region was also |
|
105 // the last destination region to which a split region was copied (the most |
|
106 // likely case), then additional work would be needed to get fill_region() to |
|
107 // stop iteration and switch to a new source region at the right point. Basic |
|
108 // idea would be to use a fake value for the top of the source space. It is |
|
109 // doable, if a bit tricky. |
|
110 // |
|
111 // A simpler (but less general) solution would fill the remainder of the |
|
112 // destination region with a dummy object and continue filling the next |
|
113 // destination region. |
|
114 |
|
115 class SplitInfo |
|
116 { |
|
117 public: |
|
118 // Return true if this split info is valid (i.e., if a split has been |
|
119 // recorded). The very first region cannot have a partial object and thus is |
|
120 // never split, so 0 is the 'invalid' value. |
|
121 bool is_valid() const { return _src_region_idx > 0; } |
|
122 |
|
123 // Return true if this split holds data for the specified source region. |
|
124 inline bool is_split(size_t source_region) const; |
|
125 |
|
126 // The index of the split region, the size of the partial object on that |
|
127 // region and the destination of the partial object. |
|
128 size_t src_region_idx() const { return _src_region_idx; } |
|
129 size_t partial_obj_size() const { return _partial_obj_size; } |
|
130 HeapWord* destination() const { return _destination; } |
|
131 |
|
132 // The destination count of the partial object referenced by this split |
|
133 // (either 1 or 2). This must be added to the destination count of the |
|
134 // remainder of the source region. |
|
135 unsigned int destination_count() const { return _destination_count; } |
|
136 |
|
137 // If a word within the partial object will be written to the first word of a |
|
138 // destination region, this is the address of the destination region; |
|
139 // otherwise this is NULL. |
|
140 HeapWord* dest_region_addr() const { return _dest_region_addr; } |
|
141 |
|
142 // If a word within the partial object will be written to the first word of a |
|
143 // destination region, this is the address of that word within the partial |
|
144 // object; otherwise this is NULL. |
|
145 HeapWord* first_src_addr() const { return _first_src_addr; } |
|
146 |
|
147 // Record the data necessary to split the region src_region_idx. |
|
148 void record(size_t src_region_idx, size_t partial_obj_size, |
|
149 HeapWord* destination); |
|
150 |
|
151 void clear(); |
|
152 |
|
153 DEBUG_ONLY(void verify_clear();) |
|
154 |
|
155 private: |
|
156 size_t _src_region_idx; |
|
157 size_t _partial_obj_size; |
|
158 HeapWord* _destination; |
|
159 unsigned int _destination_count; |
|
160 HeapWord* _dest_region_addr; |
|
161 HeapWord* _first_src_addr; |
|
162 }; |
|
163 |
|
164 inline bool SplitInfo::is_split(size_t region_idx) const |
|
165 { |
|
166 return _src_region_idx == region_idx && is_valid(); |
|
167 } |
|
168 |
|
169 class SpaceInfo |
|
170 { |
|
171 public: |
|
172 MutableSpace* space() const { return _space; } |
|
173 |
|
174 // Where the free space will start after the collection. Valid only after the |
|
175 // summary phase completes. |
|
176 HeapWord* new_top() const { return _new_top; } |
|
177 |
|
178 // Allows new_top to be set. |
|
179 HeapWord** new_top_addr() { return &_new_top; } |
|
180 |
|
181 // Where the smallest allowable dense prefix ends (used only for perm gen). |
|
182 HeapWord* min_dense_prefix() const { return _min_dense_prefix; } |
|
183 |
|
184 // Where the dense prefix ends, or the compacted region begins. |
|
185 HeapWord* dense_prefix() const { return _dense_prefix; } |
|
186 |
|
187 // The start array for the (generation containing the) space, or NULL if there |
|
188 // is no start array. |
|
189 ObjectStartArray* start_array() const { return _start_array; } |
|
190 |
|
191 SplitInfo& split_info() { return _split_info; } |
|
192 |
|
193 void set_space(MutableSpace* s) { _space = s; } |
|
194 void set_new_top(HeapWord* addr) { _new_top = addr; } |
|
195 void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; } |
|
196 void set_dense_prefix(HeapWord* addr) { _dense_prefix = addr; } |
|
197 void set_start_array(ObjectStartArray* s) { _start_array = s; } |
|
198 |
|
199 void publish_new_top() const { _space->set_top(_new_top); } |
|
200 |
|
201 private: |
|
202 MutableSpace* _space; |
|
203 HeapWord* _new_top; |
|
204 HeapWord* _min_dense_prefix; |
|
205 HeapWord* _dense_prefix; |
|
206 ObjectStartArray* _start_array; |
|
207 SplitInfo _split_info; |
|
208 }; |
|
209 |
|
210 class ParallelCompactData |
|
211 { |
|
212 public: |
|
213 // Sizes are in HeapWords, unless indicated otherwise. |
|
214 static const size_t Log2RegionSize; |
|
215 static const size_t RegionSize; |
|
216 static const size_t RegionSizeBytes; |
|
217 |
|
218 // Mask for the bits in a size_t to get an offset within a region. |
|
219 static const size_t RegionSizeOffsetMask; |
|
220 // Mask for the bits in a pointer to get an offset within a region. |
|
221 static const size_t RegionAddrOffsetMask; |
|
222 // Mask for the bits in a pointer to get the address of the start of a region. |
|
223 static const size_t RegionAddrMask; |
|
224 |
|
225 static const size_t Log2BlockSize; |
|
226 static const size_t BlockSize; |
|
227 static const size_t BlockSizeBytes; |
|
228 |
|
229 static const size_t BlockSizeOffsetMask; |
|
230 static const size_t BlockAddrOffsetMask; |
|
231 static const size_t BlockAddrMask; |
|
232 |
|
233 static const size_t BlocksPerRegion; |
|
234 static const size_t Log2BlocksPerRegion; |
|
235 |
|
236 class RegionData |
|
237 { |
|
238 public: |
|
239 // Destination address of the region. |
|
240 HeapWord* destination() const { return _destination; } |
|
241 |
|
242 // The first region containing data destined for this region. |
|
243 size_t source_region() const { return _source_region; } |
|
244 |
|
245 // The object (if any) starting in this region and ending in a different |
|
246 // region that could not be updated during the main (parallel) compaction |
|
247 // phase. This is different from _partial_obj_addr, which is an object that |
|
248 // extends onto a source region. However, the two uses do not overlap in |
|
249 // time, so the same field is used to save space. |
|
250 HeapWord* deferred_obj_addr() const { return _partial_obj_addr; } |
|
251 |
|
252 // The starting address of the partial object extending onto the region. |
|
253 HeapWord* partial_obj_addr() const { return _partial_obj_addr; } |
|
254 |
|
255 // Size of the partial object extending onto the region (words). |
|
256 size_t partial_obj_size() const { return _partial_obj_size; } |
|
257 |
|
258 // Size of live data that lies within this region due to objects that start |
|
259 // in this region (words). This does not include the partial object |
|
260 // extending onto the region (if any), or the part of an object that extends |
|
261 // onto the next region (if any). |
|
262 size_t live_obj_size() const { return _dc_and_los & los_mask; } |
|
263 |
|
264 // Total live data that lies within the region (words). |
|
265 size_t data_size() const { return partial_obj_size() + live_obj_size(); } |
|
266 |
|
267 // The destination_count is the number of other regions to which data from |
|
268 // this region will be copied. At the end of the summary phase, the valid |
|
269 // values of destination_count are |
|
270 // |
|
271 // 0 - data from the region will be compacted completely into itself, or the |
|
272 // region is empty. The region can be claimed and then filled. |
|
273 // 1 - data from the region will be compacted into 1 other region; some |
|
274 // data from the region may also be compacted into the region itself. |
|
275 // 2 - data from the region will be copied to 2 other regions. |
|
276 // |
|
277 // During compaction as regions are emptied, the destination_count is |
|
278 // decremented (atomically) and when it reaches 0, it can be claimed and |
|
279 // then filled. |
|
280 // |
|
281 // A region is claimed for processing by atomically changing the |
|
282 // destination_count to the claimed value (dc_claimed). After a region has |
|
283 // been filled, the destination_count should be set to the completed value |
|
284 // (dc_completed). |
|
285 inline uint destination_count() const; |
|
286 inline uint destination_count_raw() const; |
|
287 |
|
288 // Whether the block table for this region has been filled. |
|
289 inline bool blocks_filled() const; |
|
290 |
|
291 // Number of times the block table was filled. |
|
292 DEBUG_ONLY(inline size_t blocks_filled_count() const;) |
|
293 |
|
294 // The location of the java heap data that corresponds to this region. |
|
295 inline HeapWord* data_location() const; |
|
296 |
|
297 // The highest address referenced by objects in this region. |
|
298 inline HeapWord* highest_ref() const; |
|
299 |
|
300 // Whether this region is available to be claimed, has been claimed, or has |
|
301 // been completed. |
|
302 // |
|
303 // Minor subtlety: claimed() returns true if the region is marked |
|
304 // completed(), which is desirable since a region must be claimed before it |
|
305 // can be completed. |
|
306 bool available() const { return _dc_and_los < dc_one; } |
|
307 bool claimed() const { return _dc_and_los >= dc_claimed; } |
|
308 bool completed() const { return _dc_and_los >= dc_completed; } |
|
309 |
|
310 // These are not atomic. |
|
311 void set_destination(HeapWord* addr) { _destination = addr; } |
|
312 void set_source_region(size_t region) { _source_region = region; } |
|
313 void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } |
|
314 void set_partial_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; } |
|
315 void set_partial_obj_size(size_t words) { |
|
316 _partial_obj_size = (region_sz_t) words; |
|
317 } |
|
318 inline void set_blocks_filled(); |
|
319 |
|
320 inline void set_destination_count(uint count); |
|
321 inline void set_live_obj_size(size_t words); |
|
322 inline void set_data_location(HeapWord* addr); |
|
323 inline void set_completed(); |
|
324 inline bool claim_unsafe(); |
|
325 |
|
326 // These are atomic. |
|
327 inline void add_live_obj(size_t words); |
|
328 inline void set_highest_ref(HeapWord* addr); |
|
329 inline void decrement_destination_count(); |
|
330 inline bool claim(); |
|
331 |
|
332 private: |
|
333 // The type used to represent object sizes within a region. |
|
334 typedef uint region_sz_t; |
|
335 |
|
336 // Constants for manipulating the _dc_and_los field, which holds both the |
|
337 // destination count and live obj size. The live obj size lives at the |
|
338 // least significant end so no masking is necessary when adding. |
|
339 static const region_sz_t dc_shift; // Shift amount. |
|
340 static const region_sz_t dc_mask; // Mask for destination count. |
|
341 static const region_sz_t dc_one; // 1, shifted appropriately. |
|
342 static const region_sz_t dc_claimed; // Region has been claimed. |
|
343 static const region_sz_t dc_completed; // Region has been completed. |
|
344 static const region_sz_t los_mask; // Mask for live obj size. |
|
345 |
|
346 HeapWord* _destination; |
|
347 size_t _source_region; |
|
348 HeapWord* _partial_obj_addr; |
|
349 region_sz_t _partial_obj_size; |
|
350 region_sz_t volatile _dc_and_los; |
|
351 bool _blocks_filled; |
|
352 |
|
353 #ifdef ASSERT |
|
354 size_t _blocks_filled_count; // Number of block table fills. |
|
355 |
|
356 // These enable optimizations that are only partially implemented. Use |
|
357 // debug builds to prevent the code fragments from breaking. |
|
358 HeapWord* _data_location; |
|
359 HeapWord* _highest_ref; |
|
360 #endif // #ifdef ASSERT |
|
361 |
|
362 #ifdef ASSERT |
|
363 public: |
|
364 uint _pushed; // 0 until region is pushed onto a stack |
|
365 private: |
|
366 #endif |
|
367 }; |
|
368 |
|
369 // "Blocks" allow shorter sections of the bitmap to be searched. Each Block |
|
370 // holds an offset, which is the amount of live data in the Region to the left |
|
371 // of the first live object that starts in the Block. |
|
372 class BlockData |
|
373 { |
|
374 public: |
|
375 typedef unsigned short int blk_ofs_t; |
|
376 |
|
377 blk_ofs_t offset() const { return _offset; } |
|
378 void set_offset(size_t val) { _offset = (blk_ofs_t)val; } |
|
379 |
|
380 private: |
|
381 blk_ofs_t _offset; |
|
382 }; |
|
383 |
|
384 public: |
|
385 ParallelCompactData(); |
|
386 bool initialize(MemRegion covered_region); |
|
387 |
|
388 size_t region_count() const { return _region_count; } |
|
389 size_t reserved_byte_size() const { return _reserved_byte_size; } |
|
390 |
|
391 // Convert region indices to/from RegionData pointers. |
|
392 inline RegionData* region(size_t region_idx) const; |
|
393 inline size_t region(const RegionData* const region_ptr) const; |
|
394 |
|
395 size_t block_count() const { return _block_count; } |
|
396 inline BlockData* block(size_t block_idx) const; |
|
397 inline size_t block(const BlockData* block_ptr) const; |
|
398 |
|
399 void add_obj(HeapWord* addr, size_t len); |
|
400 void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); } |
|
401 |
|
402 // Fill in the regions covering [beg, end) so that no data moves; i.e., the |
|
403 // destination of region n is simply the start of region n. The argument beg |
|
404 // must be region-aligned; end need not be. |
|
405 void summarize_dense_prefix(HeapWord* beg, HeapWord* end); |
|
406 |
|
407 HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info, |
|
408 HeapWord* destination, HeapWord* target_end, |
|
409 HeapWord** target_next); |
|
410 bool summarize(SplitInfo& split_info, |
|
411 HeapWord* source_beg, HeapWord* source_end, |
|
412 HeapWord** source_next, |
|
413 HeapWord* target_beg, HeapWord* target_end, |
|
414 HeapWord** target_next); |
|
415 |
|
416 void clear(); |
|
417 void clear_range(size_t beg_region, size_t end_region); |
|
418 void clear_range(HeapWord* beg, HeapWord* end) { |
|
419 clear_range(addr_to_region_idx(beg), addr_to_region_idx(end)); |
|
420 } |
|
421 |
|
422 // Return the number of words between addr and the start of the region |
|
423 // containing addr. |
|
424 inline size_t region_offset(const HeapWord* addr) const; |
|
425 |
|
426 // Convert addresses to/from a region index or region pointer. |
|
427 inline size_t addr_to_region_idx(const HeapWord* addr) const; |
|
428 inline RegionData* addr_to_region_ptr(const HeapWord* addr) const; |
|
429 inline HeapWord* region_to_addr(size_t region) const; |
|
430 inline HeapWord* region_to_addr(size_t region, size_t offset) const; |
|
431 inline HeapWord* region_to_addr(const RegionData* region) const; |
|
432 |
|
433 inline HeapWord* region_align_down(HeapWord* addr) const; |
|
434 inline HeapWord* region_align_up(HeapWord* addr) const; |
|
435 inline bool is_region_aligned(HeapWord* addr) const; |
|
436 |
|
437 // Analogous to region_offset() for blocks. |
|
438 size_t block_offset(const HeapWord* addr) const; |
|
439 size_t addr_to_block_idx(const HeapWord* addr) const; |
|
440 size_t addr_to_block_idx(const oop obj) const { |
|
441 return addr_to_block_idx((HeapWord*) obj); |
|
442 } |
|
443 inline BlockData* addr_to_block_ptr(const HeapWord* addr) const; |
|
444 inline HeapWord* block_to_addr(size_t block) const; |
|
445 inline size_t region_to_block_idx(size_t region) const; |
|
446 |
|
447 inline HeapWord* block_align_down(HeapWord* addr) const; |
|
448 inline HeapWord* block_align_up(HeapWord* addr) const; |
|
449 inline bool is_block_aligned(HeapWord* addr) const; |
|
450 |
|
451 // Return the address one past the end of the partial object. |
|
452 HeapWord* partial_obj_end(size_t region_idx) const; |
|
453 |
|
454 // Return the location of the object after compaction. |
|
455 HeapWord* calc_new_pointer(HeapWord* addr); |
|
456 |
|
457 HeapWord* calc_new_pointer(oop p) { |
|
458 return calc_new_pointer((HeapWord*) p); |
|
459 } |
|
460 |
|
461 #ifdef ASSERT |
|
462 void verify_clear(const PSVirtualSpace* vspace); |
|
463 void verify_clear(); |
|
464 #endif // #ifdef ASSERT |
|
465 |
|
466 private: |
|
467 bool initialize_block_data(); |
|
468 bool initialize_region_data(size_t region_size); |
|
469 PSVirtualSpace* create_vspace(size_t count, size_t element_size); |
|
470 |
|
471 private: |
|
472 HeapWord* _region_start; |
|
473 #ifdef ASSERT |
|
474 HeapWord* _region_end; |
|
475 #endif // #ifdef ASSERT |
|
476 |
|
477 PSVirtualSpace* _region_vspace; |
|
478 size_t _reserved_byte_size; |
|
479 RegionData* _region_data; |
|
480 size_t _region_count; |
|
481 |
|
482 PSVirtualSpace* _block_vspace; |
|
483 BlockData* _block_data; |
|
484 size_t _block_count; |
|
485 }; |
|
486 |
|
487 inline uint |
|
488 ParallelCompactData::RegionData::destination_count_raw() const |
|
489 { |
|
490 return _dc_and_los & dc_mask; |
|
491 } |
|
492 |
|
493 inline uint |
|
494 ParallelCompactData::RegionData::destination_count() const |
|
495 { |
|
496 return destination_count_raw() >> dc_shift; |
|
497 } |
|
498 |
|
499 inline bool |
|
500 ParallelCompactData::RegionData::blocks_filled() const |
|
501 { |
|
502 return _blocks_filled; |
|
503 } |
|
504 |
|
505 #ifdef ASSERT |
|
506 inline size_t |
|
507 ParallelCompactData::RegionData::blocks_filled_count() const |
|
508 { |
|
509 return _blocks_filled_count; |
|
510 } |
|
511 #endif // #ifdef ASSERT |
|
512 |
|
513 inline void |
|
514 ParallelCompactData::RegionData::set_blocks_filled() |
|
515 { |
|
516 _blocks_filled = true; |
|
517 // Debug builds count the number of times the table was filled. |
|
518 DEBUG_ONLY(Atomic::inc_ptr(&_blocks_filled_count)); |
|
519 } |
|
520 |
|
521 inline void |
|
522 ParallelCompactData::RegionData::set_destination_count(uint count) |
|
523 { |
|
524 assert(count <= (dc_completed >> dc_shift), "count too large"); |
|
525 const region_sz_t live_sz = (region_sz_t) live_obj_size(); |
|
526 _dc_and_los = (count << dc_shift) | live_sz; |
|
527 } |
|
528 |
|
529 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words) |
|
530 { |
|
531 assert(words <= los_mask, "would overflow"); |
|
532 _dc_and_los = destination_count_raw() | (region_sz_t)words; |
|
533 } |
|
534 |
|
535 inline void ParallelCompactData::RegionData::decrement_destination_count() |
|
536 { |
|
537 assert(_dc_and_los < dc_claimed, "already claimed"); |
|
538 assert(_dc_and_los >= dc_one, "count would go negative"); |
|
539 Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los); |
|
540 } |
|
541 |
|
542 inline HeapWord* ParallelCompactData::RegionData::data_location() const |
|
543 { |
|
544 DEBUG_ONLY(return _data_location;) |
|
545 NOT_DEBUG(return NULL;) |
|
546 } |
|
547 |
|
548 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const |
|
549 { |
|
550 DEBUG_ONLY(return _highest_ref;) |
|
551 NOT_DEBUG(return NULL;) |
|
552 } |
|
553 |
|
554 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr) |
|
555 { |
|
556 DEBUG_ONLY(_data_location = addr;) |
|
557 } |
|
558 |
|
559 inline void ParallelCompactData::RegionData::set_completed() |
|
560 { |
|
561 assert(claimed(), "must be claimed first"); |
|
562 _dc_and_los = dc_completed | (region_sz_t) live_obj_size(); |
|
563 } |
|
564 |
|
565 // MT-unsafe claiming of a region. Should only be used during single threaded |
|
566 // execution. |
|
567 inline bool ParallelCompactData::RegionData::claim_unsafe() |
|
568 { |
|
569 if (available()) { |
|
570 _dc_and_los |= dc_claimed; |
|
571 return true; |
|
572 } |
|
573 return false; |
|
574 } |
|
575 |
|
576 inline void ParallelCompactData::RegionData::add_live_obj(size_t words) |
|
577 { |
|
578 assert(words <= (size_t)los_mask - live_obj_size(), "overflow"); |
|
579 Atomic::add((int) words, (volatile int*) &_dc_and_los); |
|
580 } |
|
581 |
|
582 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr) |
|
583 { |
|
584 #ifdef ASSERT |
|
585 HeapWord* tmp = _highest_ref; |
|
586 while (addr > tmp) { |
|
587 tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp); |
|
588 } |
|
589 #endif // #ifdef ASSERT |
|
590 } |
|
591 |
|
592 inline bool ParallelCompactData::RegionData::claim() |
|
593 { |
|
594 const int los = (int) live_obj_size(); |
|
595 const int old = Atomic::cmpxchg(dc_claimed | los, |
|
596 (volatile int*) &_dc_and_los, los); |
|
597 return old == los; |
|
598 } |
|
599 |
|
600 inline ParallelCompactData::RegionData* |
|
601 ParallelCompactData::region(size_t region_idx) const |
|
602 { |
|
603 assert(region_idx <= region_count(), "bad arg"); |
|
604 return _region_data + region_idx; |
|
605 } |
|
606 |
|
607 inline size_t |
|
608 ParallelCompactData::region(const RegionData* const region_ptr) const |
|
609 { |
|
610 assert(region_ptr >= _region_data, "bad arg"); |
|
611 assert(region_ptr <= _region_data + region_count(), "bad arg"); |
|
612 return pointer_delta(region_ptr, _region_data, sizeof(RegionData)); |
|
613 } |
|
614 |
|
615 inline ParallelCompactData::BlockData* |
|
616 ParallelCompactData::block(size_t n) const { |
|
617 assert(n < block_count(), "bad arg"); |
|
618 return _block_data + n; |
|
619 } |
|
620 |
|
621 inline size_t |
|
622 ParallelCompactData::region_offset(const HeapWord* addr) const |
|
623 { |
|
624 assert(addr >= _region_start, "bad addr"); |
|
625 assert(addr <= _region_end, "bad addr"); |
|
626 return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize; |
|
627 } |
|
628 |
|
629 inline size_t |
|
630 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const |
|
631 { |
|
632 assert(addr >= _region_start, "bad addr"); |
|
633 assert(addr <= _region_end, "bad addr"); |
|
634 return pointer_delta(addr, _region_start) >> Log2RegionSize; |
|
635 } |
|
636 |
|
637 inline ParallelCompactData::RegionData* |
|
638 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const |
|
639 { |
|
640 return region(addr_to_region_idx(addr)); |
|
641 } |
|
642 |
|
643 inline HeapWord* |
|
644 ParallelCompactData::region_to_addr(size_t region) const |
|
645 { |
|
646 assert(region <= _region_count, "region out of range"); |
|
647 return _region_start + (region << Log2RegionSize); |
|
648 } |
|
649 |
|
650 inline HeapWord* |
|
651 ParallelCompactData::region_to_addr(const RegionData* region) const |
|
652 { |
|
653 return region_to_addr(pointer_delta(region, _region_data, |
|
654 sizeof(RegionData))); |
|
655 } |
|
656 |
|
657 inline HeapWord* |
|
658 ParallelCompactData::region_to_addr(size_t region, size_t offset) const |
|
659 { |
|
660 assert(region <= _region_count, "region out of range"); |
|
661 assert(offset < RegionSize, "offset too big"); // This may be too strict. |
|
662 return region_to_addr(region) + offset; |
|
663 } |
|
664 |
|
665 inline HeapWord* |
|
666 ParallelCompactData::region_align_down(HeapWord* addr) const |
|
667 { |
|
668 assert(addr >= _region_start, "bad addr"); |
|
669 assert(addr < _region_end + RegionSize, "bad addr"); |
|
670 return (HeapWord*)(size_t(addr) & RegionAddrMask); |
|
671 } |
|
672 |
|
673 inline HeapWord* |
|
674 ParallelCompactData::region_align_up(HeapWord* addr) const |
|
675 { |
|
676 assert(addr >= _region_start, "bad addr"); |
|
677 assert(addr <= _region_end, "bad addr"); |
|
678 return region_align_down(addr + RegionSizeOffsetMask); |
|
679 } |
|
680 |
|
681 inline bool |
|
682 ParallelCompactData::is_region_aligned(HeapWord* addr) const |
|
683 { |
|
684 return region_offset(addr) == 0; |
|
685 } |
|
686 |
|
687 inline size_t |
|
688 ParallelCompactData::block_offset(const HeapWord* addr) const |
|
689 { |
|
690 assert(addr >= _region_start, "bad addr"); |
|
691 assert(addr <= _region_end, "bad addr"); |
|
692 return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize; |
|
693 } |
|
694 |
|
695 inline size_t |
|
696 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const |
|
697 { |
|
698 assert(addr >= _region_start, "bad addr"); |
|
699 assert(addr <= _region_end, "bad addr"); |
|
700 return pointer_delta(addr, _region_start) >> Log2BlockSize; |
|
701 } |
|
702 |
|
703 inline ParallelCompactData::BlockData* |
|
704 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const |
|
705 { |
|
706 return block(addr_to_block_idx(addr)); |
|
707 } |
|
708 |
|
709 inline HeapWord* |
|
710 ParallelCompactData::block_to_addr(size_t block) const |
|
711 { |
|
712 assert(block < _block_count, "block out of range"); |
|
713 return _region_start + (block << Log2BlockSize); |
|
714 } |
|
715 |
|
716 inline size_t |
|
717 ParallelCompactData::region_to_block_idx(size_t region) const |
|
718 { |
|
719 return region << Log2BlocksPerRegion; |
|
720 } |
|
721 |
|
722 inline HeapWord* |
|
723 ParallelCompactData::block_align_down(HeapWord* addr) const |
|
724 { |
|
725 assert(addr >= _region_start, "bad addr"); |
|
726 assert(addr < _region_end + RegionSize, "bad addr"); |
|
727 return (HeapWord*)(size_t(addr) & BlockAddrMask); |
|
728 } |
|
729 |
|
730 inline HeapWord* |
|
731 ParallelCompactData::block_align_up(HeapWord* addr) const |
|
732 { |
|
733 assert(addr >= _region_start, "bad addr"); |
|
734 assert(addr <= _region_end, "bad addr"); |
|
735 return block_align_down(addr + BlockSizeOffsetMask); |
|
736 } |
|
737 |
|
738 inline bool |
|
739 ParallelCompactData::is_block_aligned(HeapWord* addr) const |
|
740 { |
|
741 return block_offset(addr) == 0; |
|
742 } |
|
743 |
|
744 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the |
|
745 // do_addr() method. |
|
746 // |
|
747 // The closure is initialized with the number of heap words to process |
|
748 // (words_remaining()), and becomes 'full' when it reaches 0. The do_addr() |
|
749 // methods in subclasses should update the total as words are processed. Since |
|
750 // only one subclass actually uses this mechanism to terminate iteration, the |
|
751 // default initial value is > 0. The implementation is here and not in the |
|
752 // single subclass that uses it to avoid making is_full() virtual, and thus |
|
753 // adding a virtual call per live object. |
|
754 |
|
755 class ParMarkBitMapClosure: public StackObj { |
|
756 public: |
|
757 typedef ParMarkBitMap::idx_t idx_t; |
|
758 typedef ParMarkBitMap::IterationStatus IterationStatus; |
|
759 |
|
760 public: |
|
761 inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm, |
|
762 size_t words = max_uintx); |
|
763 |
|
764 inline ParCompactionManager* compaction_manager() const; |
|
765 inline ParMarkBitMap* bitmap() const; |
|
766 inline size_t words_remaining() const; |
|
767 inline bool is_full() const; |
|
768 inline HeapWord* source() const; |
|
769 |
|
770 inline void set_source(HeapWord* addr); |
|
771 |
|
772 virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0; |
|
773 |
|
774 protected: |
|
775 inline void decrement_words_remaining(size_t words); |
|
776 |
|
777 private: |
|
778 ParMarkBitMap* const _bitmap; |
|
779 ParCompactionManager* const _compaction_manager; |
|
780 DEBUG_ONLY(const size_t _initial_words_remaining;) // Useful in debugger. |
|
781 size_t _words_remaining; // Words left to copy. |
|
782 |
|
783 protected: |
|
784 HeapWord* _source; // Next addr that would be read. |
|
785 }; |
|
786 |
|
787 inline |
|
788 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap, |
|
789 ParCompactionManager* cm, |
|
790 size_t words): |
|
791 _bitmap(bitmap), _compaction_manager(cm) |
|
792 #ifdef ASSERT |
|
793 , _initial_words_remaining(words) |
|
794 #endif |
|
795 { |
|
796 _words_remaining = words; |
|
797 _source = NULL; |
|
798 } |
|
799 |
|
800 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const { |
|
801 return _compaction_manager; |
|
802 } |
|
803 |
|
804 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const { |
|
805 return _bitmap; |
|
806 } |
|
807 |
|
808 inline size_t ParMarkBitMapClosure::words_remaining() const { |
|
809 return _words_remaining; |
|
810 } |
|
811 |
|
812 inline bool ParMarkBitMapClosure::is_full() const { |
|
813 return words_remaining() == 0; |
|
814 } |
|
815 |
|
816 inline HeapWord* ParMarkBitMapClosure::source() const { |
|
817 return _source; |
|
818 } |
|
819 |
|
820 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) { |
|
821 _source = addr; |
|
822 } |
|
823 |
|
824 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) { |
|
825 assert(_words_remaining >= words, "processed too many words"); |
|
826 _words_remaining -= words; |
|
827 } |
|
828 |
|
829 // The UseParallelOldGC collector is a stop-the-world garbage collector that |
|
830 // does parts of the collection using parallel threads. The collection includes |
|
831 // the tenured generation and the young generation. The permanent generation is |
|
832 // collected at the same time as the other two generations but the permanent |
|
833 // generation is collect by a single GC thread. The permanent generation is |
|
834 // collected serially because of the requirement that during the processing of a |
|
835 // klass AAA, any objects reference by AAA must already have been processed. |
|
836 // This requirement is enforced by a left (lower address) to right (higher |
|
837 // address) sliding compaction. |
|
838 // |
|
839 // There are four phases of the collection. |
|
840 // |
|
841 // - marking phase |
|
842 // - summary phase |
|
843 // - compacting phase |
|
844 // - clean up phase |
|
845 // |
|
846 // Roughly speaking these phases correspond, respectively, to |
|
847 // - mark all the live objects |
|
848 // - calculate the destination of each object at the end of the collection |
|
849 // - move the objects to their destination |
|
850 // - update some references and reinitialize some variables |
|
851 // |
|
852 // These three phases are invoked in PSParallelCompact::invoke_no_policy(). The |
|
853 // marking phase is implemented in PSParallelCompact::marking_phase() and does a |
|
854 // complete marking of the heap. The summary phase is implemented in |
|
855 // PSParallelCompact::summary_phase(). The move and update phase is implemented |
|
856 // in PSParallelCompact::compact(). |
|
857 // |
|
858 // A space that is being collected is divided into regions and with each region |
|
859 // is associated an object of type ParallelCompactData. Each region is of a |
|
860 // fixed size and typically will contain more than 1 object and may have parts |
|
861 // of objects at the front and back of the region. |
|
862 // |
|
863 // region -----+---------------------+---------- |
|
864 // objects covered [ AAA )[ BBB )[ CCC )[ DDD ) |
|
865 // |
|
866 // The marking phase does a complete marking of all live objects in the heap. |
|
867 // The marking also compiles the size of the data for all live objects covered |
|
868 // by the region. This size includes the part of any live object spanning onto |
|
869 // the region (part of AAA if it is live) from the front, all live objects |
|
870 // contained in the region (BBB and/or CCC if they are live), and the part of |
|
871 // any live objects covered by the region that extends off the region (part of |
|
872 // DDD if it is live). The marking phase uses multiple GC threads and marking |
|
873 // is done in a bit array of type ParMarkBitMap. The marking of the bit map is |
|
874 // done atomically as is the accumulation of the size of the live objects |
|
875 // covered by a region. |
|
876 // |
|
877 // The summary phase calculates the total live data to the left of each region |
|
878 // XXX. Based on that total and the bottom of the space, it can calculate the |
|
879 // starting location of the live data in XXX. The summary phase calculates for |
|
880 // each region XXX quantites such as |
|
881 // |
|
882 // - the amount of live data at the beginning of a region from an object |
|
883 // entering the region. |
|
884 // - the location of the first live data on the region |
|
885 // - a count of the number of regions receiving live data from XXX. |
|
886 // |
|
887 // See ParallelCompactData for precise details. The summary phase also |
|
888 // calculates the dense prefix for the compaction. The dense prefix is a |
|
889 // portion at the beginning of the space that is not moved. The objects in the |
|
890 // dense prefix do need to have their object references updated. See method |
|
891 // summarize_dense_prefix(). |
|
892 // |
|
893 // The summary phase is done using 1 GC thread. |
|
894 // |
|
895 // The compaction phase moves objects to their new location and updates all |
|
896 // references in the object. |
|
897 // |
|
898 // A current exception is that objects that cross a region boundary are moved |
|
899 // but do not have their references updated. References are not updated because |
|
900 // it cannot easily be determined if the klass pointer KKK for the object AAA |
|
901 // has been updated. KKK likely resides in a region to the left of the region |
|
902 // containing AAA. These AAA's have there references updated at the end in a |
|
903 // clean up phase. See the method PSParallelCompact::update_deferred_objects(). |
|
904 // An alternate strategy is being investigated for this deferral of updating. |
|
905 // |
|
906 // Compaction is done on a region basis. A region that is ready to be filled is |
|
907 // put on a ready list and GC threads take region off the list and fill them. A |
|
908 // region is ready to be filled if it empty of live objects. Such a region may |
|
909 // have been initially empty (only contained dead objects) or may have had all |
|
910 // its live objects copied out already. A region that compacts into itself is |
|
911 // also ready for filling. The ready list is initially filled with empty |
|
912 // regions and regions compacting into themselves. There is always at least 1 |
|
913 // region that can be put on the ready list. The regions are atomically added |
|
914 // and removed from the ready list. |
|
915 |
|
916 class PSParallelCompact : AllStatic { |
|
917 public: |
|
918 // Convenient access to type names. |
|
919 typedef ParMarkBitMap::idx_t idx_t; |
|
920 typedef ParallelCompactData::RegionData RegionData; |
|
921 typedef ParallelCompactData::BlockData BlockData; |
|
922 |
|
923 typedef enum { |
|
924 old_space_id, eden_space_id, |
|
925 from_space_id, to_space_id, last_space_id |
|
926 } SpaceId; |
|
927 |
|
928 public: |
|
929 // Inline closure decls |
|
930 // |
|
931 class IsAliveClosure: public BoolObjectClosure { |
|
932 public: |
|
933 virtual bool do_object_b(oop p); |
|
934 }; |
|
935 |
|
936 class KeepAliveClosure: public OopClosure { |
|
937 private: |
|
938 ParCompactionManager* _compaction_manager; |
|
939 protected: |
|
940 template <class T> inline void do_oop_work(T* p); |
|
941 public: |
|
942 KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } |
|
943 virtual void do_oop(oop* p); |
|
944 virtual void do_oop(narrowOop* p); |
|
945 }; |
|
946 |
|
947 class FollowStackClosure: public VoidClosure { |
|
948 private: |
|
949 ParCompactionManager* _compaction_manager; |
|
950 public: |
|
951 FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } |
|
952 virtual void do_void(); |
|
953 }; |
|
954 |
|
955 class AdjustPointerClosure: public OopClosure { |
|
956 public: |
|
957 virtual void do_oop(oop* p); |
|
958 virtual void do_oop(narrowOop* p); |
|
959 // do not walk from thread stacks to the code cache on this phase |
|
960 virtual void do_code_blob(CodeBlob* cb) const { } |
|
961 }; |
|
962 |
|
963 class AdjustKlassClosure : public KlassClosure { |
|
964 public: |
|
965 void do_klass(Klass* klass); |
|
966 }; |
|
967 |
|
968 friend class KeepAliveClosure; |
|
969 friend class FollowStackClosure; |
|
970 friend class AdjustPointerClosure; |
|
971 friend class AdjustKlassClosure; |
|
972 friend class FollowKlassClosure; |
|
973 friend class InstanceClassLoaderKlass; |
|
974 friend class RefProcTaskProxy; |
|
975 |
|
976 private: |
|
977 static STWGCTimer _gc_timer; |
|
978 static ParallelOldTracer _gc_tracer; |
|
979 static elapsedTimer _accumulated_time; |
|
980 static unsigned int _total_invocations; |
|
981 static unsigned int _maximum_compaction_gc_num; |
|
982 static jlong _time_of_last_gc; // ms |
|
983 static CollectorCounters* _counters; |
|
984 static ParMarkBitMap _mark_bitmap; |
|
985 static ParallelCompactData _summary_data; |
|
986 static IsAliveClosure _is_alive_closure; |
|
987 static SpaceInfo _space_info[last_space_id]; |
|
988 static bool _print_phases; |
|
989 static AdjustPointerClosure _adjust_pointer_closure; |
|
990 static AdjustKlassClosure _adjust_klass_closure; |
|
991 |
|
992 // Reference processing (used in ...follow_contents) |
|
993 static ReferenceProcessor* _ref_processor; |
|
994 |
|
995 // Updated location of intArrayKlassObj. |
|
996 static Klass* _updated_int_array_klass_obj; |
|
997 |
|
998 // Values computed at initialization and used by dead_wood_limiter(). |
|
999 static double _dwl_mean; |
|
1000 static double _dwl_std_dev; |
|
1001 static double _dwl_first_term; |
|
1002 static double _dwl_adjustment; |
|
1003 #ifdef ASSERT |
|
1004 static bool _dwl_initialized; |
|
1005 #endif // #ifdef ASSERT |
|
1006 |
|
1007 private: |
|
1008 |
|
1009 static void initialize_space_info(); |
|
1010 |
|
1011 // Return true if details about individual phases should be printed. |
|
1012 static inline bool print_phases(); |
|
1013 |
|
1014 // Clear the marking bitmap and summary data that cover the specified space. |
|
1015 static void clear_data_covering_space(SpaceId id); |
|
1016 |
|
1017 static void pre_compact(PreGCValues* pre_gc_values); |
|
1018 static void post_compact(); |
|
1019 |
|
1020 // Mark live objects |
|
1021 static void marking_phase(ParCompactionManager* cm, |
|
1022 bool maximum_heap_compaction, |
|
1023 ParallelOldTracer *gc_tracer); |
|
1024 |
|
1025 template <class T> |
|
1026 static inline void follow_root(ParCompactionManager* cm, T* p); |
|
1027 |
|
1028 // Compute the dense prefix for the designated space. This is an experimental |
|
1029 // implementation currently not used in production. |
|
1030 static HeapWord* compute_dense_prefix_via_density(const SpaceId id, |
|
1031 bool maximum_compaction); |
|
1032 |
|
1033 // Methods used to compute the dense prefix. |
|
1034 |
|
1035 // Compute the value of the normal distribution at x = density. The mean and |
|
1036 // standard deviation are values saved by initialize_dead_wood_limiter(). |
|
1037 static inline double normal_distribution(double density); |
|
1038 |
|
1039 // Initialize the static vars used by dead_wood_limiter(). |
|
1040 static void initialize_dead_wood_limiter(); |
|
1041 |
|
1042 // Return the percentage of space that can be treated as "dead wood" (i.e., |
|
1043 // not reclaimed). |
|
1044 static double dead_wood_limiter(double density, size_t min_percent); |
|
1045 |
|
1046 // Find the first (left-most) region in the range [beg, end) that has at least |
|
1047 // dead_words of dead space to the left. The argument beg must be the first |
|
1048 // region in the space that is not completely live. |
|
1049 static RegionData* dead_wood_limit_region(const RegionData* beg, |
|
1050 const RegionData* end, |
|
1051 size_t dead_words); |
|
1052 |
|
1053 // Return a pointer to the first region in the range [beg, end) that is not |
|
1054 // completely full. |
|
1055 static RegionData* first_dead_space_region(const RegionData* beg, |
|
1056 const RegionData* end); |
|
1057 |
|
1058 // Return a value indicating the benefit or 'yield' if the compacted region |
|
1059 // were to start (or equivalently if the dense prefix were to end) at the |
|
1060 // candidate region. Higher values are better. |
|
1061 // |
|
1062 // The value is based on the amount of space reclaimed vs. the costs of (a) |
|
1063 // updating references in the dense prefix plus (b) copying objects and |
|
1064 // updating references in the compacted region. |
|
1065 static inline double reclaimed_ratio(const RegionData* const candidate, |
|
1066 HeapWord* const bottom, |
|
1067 HeapWord* const top, |
|
1068 HeapWord* const new_top); |
|
1069 |
|
1070 // Compute the dense prefix for the designated space. |
|
1071 static HeapWord* compute_dense_prefix(const SpaceId id, |
|
1072 bool maximum_compaction); |
|
1073 |
|
1074 // Return true if dead space crosses onto the specified Region; bit must be |
|
1075 // the bit index corresponding to the first word of the Region. |
|
1076 static inline bool dead_space_crosses_boundary(const RegionData* region, |
|
1077 idx_t bit); |
|
1078 |
|
1079 // Summary phase utility routine to fill dead space (if any) at the dense |
|
1080 // prefix boundary. Should only be called if the the dense prefix is |
|
1081 // non-empty. |
|
1082 static void fill_dense_prefix_end(SpaceId id); |
|
1083 |
|
1084 // Clear the summary data source_region field for the specified addresses. |
|
1085 static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr); |
|
1086 |
|
1087 #ifndef PRODUCT |
|
1088 // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot). |
|
1089 |
|
1090 // Fill the region [start, start + words) with live object(s). Only usable |
|
1091 // for the old and permanent generations. |
|
1092 static void fill_with_live_objects(SpaceId id, HeapWord* const start, |
|
1093 size_t words); |
|
1094 // Include the new objects in the summary data. |
|
1095 static void summarize_new_objects(SpaceId id, HeapWord* start); |
|
1096 |
|
1097 // Add live objects to a survivor space since it's rare that both survivors |
|
1098 // are non-empty. |
|
1099 static void provoke_split_fill_survivor(SpaceId id); |
|
1100 |
|
1101 // Add live objects and/or choose the dense prefix to provoke splitting. |
|
1102 static void provoke_split(bool & maximum_compaction); |
|
1103 #endif |
|
1104 |
|
1105 static void summarize_spaces_quick(); |
|
1106 static void summarize_space(SpaceId id, bool maximum_compaction); |
|
1107 static void summary_phase(ParCompactionManager* cm, bool maximum_compaction); |
|
1108 |
|
1109 // Adjust addresses in roots. Does not adjust addresses in heap. |
|
1110 static void adjust_roots(); |
|
1111 |
|
1112 DEBUG_ONLY(static void write_block_fill_histogram(outputStream* const out);) |
|
1113 |
|
1114 // Move objects to new locations. |
|
1115 static void compact_perm(ParCompactionManager* cm); |
|
1116 static void compact(); |
|
1117 |
|
1118 // Add available regions to the stack and draining tasks to the task queue. |
|
1119 static void enqueue_region_draining_tasks(GCTaskQueue* q, |
|
1120 uint parallel_gc_threads); |
|
1121 |
|
1122 // Add dense prefix update tasks to the task queue. |
|
1123 static void enqueue_dense_prefix_tasks(GCTaskQueue* q, |
|
1124 uint parallel_gc_threads); |
|
1125 |
|
1126 // Add region stealing tasks to the task queue. |
|
1127 static void enqueue_region_stealing_tasks( |
|
1128 GCTaskQueue* q, |
|
1129 ParallelTaskTerminator* terminator_ptr, |
|
1130 uint parallel_gc_threads); |
|
1131 |
|
1132 // If objects are left in eden after a collection, try to move the boundary |
|
1133 // and absorb them into the old gen. Returns true if eden was emptied. |
|
1134 static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy, |
|
1135 PSYoungGen* young_gen, |
|
1136 PSOldGen* old_gen); |
|
1137 |
|
1138 // Reset time since last full gc |
|
1139 static void reset_millis_since_last_gc(); |
|
1140 |
|
1141 public: |
|
1142 class MarkAndPushClosure: public OopClosure { |
|
1143 private: |
|
1144 ParCompactionManager* _compaction_manager; |
|
1145 public: |
|
1146 MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { } |
|
1147 virtual void do_oop(oop* p); |
|
1148 virtual void do_oop(narrowOop* p); |
|
1149 }; |
|
1150 |
|
1151 // The one and only place to start following the classes. |
|
1152 // Should only be applied to the ClassLoaderData klasses list. |
|
1153 class FollowKlassClosure : public KlassClosure { |
|
1154 private: |
|
1155 MarkAndPushClosure* _mark_and_push_closure; |
|
1156 public: |
|
1157 FollowKlassClosure(MarkAndPushClosure* mark_and_push_closure) : |
|
1158 _mark_and_push_closure(mark_and_push_closure) { } |
|
1159 void do_klass(Klass* klass); |
|
1160 }; |
|
1161 |
|
1162 PSParallelCompact(); |
|
1163 |
|
1164 // Convenient accessor for Universe::heap(). |
|
1165 static ParallelScavengeHeap* gc_heap() { |
|
1166 return (ParallelScavengeHeap*)Universe::heap(); |
|
1167 } |
|
1168 |
|
1169 static void invoke(bool maximum_heap_compaction); |
|
1170 static bool invoke_no_policy(bool maximum_heap_compaction); |
|
1171 |
|
1172 static void post_initialize(); |
|
1173 // Perform initialization for PSParallelCompact that requires |
|
1174 // allocations. This should be called during the VM initialization |
|
1175 // at a pointer where it would be appropriate to return a JNI_ENOMEM |
|
1176 // in the event of a failure. |
|
1177 static bool initialize(); |
|
1178 |
|
1179 // Closure accessors |
|
1180 static OopClosure* adjust_pointer_closure() { return (OopClosure*)&_adjust_pointer_closure; } |
|
1181 static KlassClosure* adjust_klass_closure() { return (KlassClosure*)&_adjust_klass_closure; } |
|
1182 static BoolObjectClosure* is_alive_closure() { return (BoolObjectClosure*)&_is_alive_closure; } |
|
1183 |
|
1184 // Public accessors |
|
1185 static elapsedTimer* accumulated_time() { return &_accumulated_time; } |
|
1186 static unsigned int total_invocations() { return _total_invocations; } |
|
1187 static CollectorCounters* counters() { return _counters; } |
|
1188 |
|
1189 // Used to add tasks |
|
1190 static GCTaskManager* const gc_task_manager(); |
|
1191 static Klass* updated_int_array_klass_obj() { |
|
1192 return _updated_int_array_klass_obj; |
|
1193 } |
|
1194 |
|
1195 // Marking support |
|
1196 static inline bool mark_obj(oop obj); |
|
1197 static inline bool is_marked(oop obj); |
|
1198 // Check mark and maybe push on marking stack |
|
1199 template <class T> static inline void mark_and_push(ParCompactionManager* cm, |
|
1200 T* p); |
|
1201 template <class T> static inline void adjust_pointer(T* p); |
|
1202 |
|
1203 static inline void follow_klass(ParCompactionManager* cm, Klass* klass); |
|
1204 |
|
1205 static void follow_class_loader(ParCompactionManager* cm, |
|
1206 ClassLoaderData* klass); |
|
1207 |
|
1208 // Compaction support. |
|
1209 // Return true if p is in the range [beg_addr, end_addr). |
|
1210 static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr); |
|
1211 static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr); |
|
1212 |
|
1213 // Convenience wrappers for per-space data kept in _space_info. |
|
1214 static inline MutableSpace* space(SpaceId space_id); |
|
1215 static inline HeapWord* new_top(SpaceId space_id); |
|
1216 static inline HeapWord* dense_prefix(SpaceId space_id); |
|
1217 static inline ObjectStartArray* start_array(SpaceId space_id); |
|
1218 |
|
1219 // Move and update the live objects in the specified space. |
|
1220 static void move_and_update(ParCompactionManager* cm, SpaceId space_id); |
|
1221 |
|
1222 // Process the end of the given region range in the dense prefix. |
|
1223 // This includes saving any object not updated. |
|
1224 static void dense_prefix_regions_epilogue(ParCompactionManager* cm, |
|
1225 size_t region_start_index, |
|
1226 size_t region_end_index, |
|
1227 idx_t exiting_object_offset, |
|
1228 idx_t region_offset_start, |
|
1229 idx_t region_offset_end); |
|
1230 |
|
1231 // Update a region in the dense prefix. For each live object |
|
1232 // in the region, update it's interior references. For each |
|
1233 // dead object, fill it with deadwood. Dead space at the end |
|
1234 // of a region range will be filled to the start of the next |
|
1235 // live object regardless of the region_index_end. None of the |
|
1236 // objects in the dense prefix move and dead space is dead |
|
1237 // (holds only dead objects that don't need any processing), so |
|
1238 // dead space can be filled in any order. |
|
1239 static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm, |
|
1240 SpaceId space_id, |
|
1241 size_t region_index_start, |
|
1242 size_t region_index_end); |
|
1243 |
|
1244 // Return the address of the count + 1st live word in the range [beg, end). |
|
1245 static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count); |
|
1246 |
|
1247 // Return the address of the word to be copied to dest_addr, which must be |
|
1248 // aligned to a region boundary. |
|
1249 static HeapWord* first_src_addr(HeapWord* const dest_addr, |
|
1250 SpaceId src_space_id, |
|
1251 size_t src_region_idx); |
|
1252 |
|
1253 // Determine the next source region, set closure.source() to the start of the |
|
1254 // new region return the region index. Parameter end_addr is the address one |
|
1255 // beyond the end of source range just processed. If necessary, switch to a |
|
1256 // new source space and set src_space_id (in-out parameter) and src_space_top |
|
1257 // (out parameter) accordingly. |
|
1258 static size_t next_src_region(MoveAndUpdateClosure& closure, |
|
1259 SpaceId& src_space_id, |
|
1260 HeapWord*& src_space_top, |
|
1261 HeapWord* end_addr); |
|
1262 |
|
1263 // Decrement the destination count for each non-empty source region in the |
|
1264 // range [beg_region, region(region_align_up(end_addr))). If the destination |
|
1265 // count for a region goes to 0 and it needs to be filled, enqueue it. |
|
1266 static void decrement_destination_counts(ParCompactionManager* cm, |
|
1267 SpaceId src_space_id, |
|
1268 size_t beg_region, |
|
1269 HeapWord* end_addr); |
|
1270 |
|
1271 // Fill a region, copying objects from one or more source regions. |
|
1272 static void fill_region(ParCompactionManager* cm, size_t region_idx); |
|
1273 static void fill_and_update_region(ParCompactionManager* cm, size_t region) { |
|
1274 fill_region(cm, region); |
|
1275 } |
|
1276 |
|
1277 // Fill in the block table for the specified region. |
|
1278 static void fill_blocks(size_t region_idx); |
|
1279 |
|
1280 // Update the deferred objects in the space. |
|
1281 static void update_deferred_objects(ParCompactionManager* cm, SpaceId id); |
|
1282 |
|
1283 static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; } |
|
1284 static ParallelCompactData& summary_data() { return _summary_data; } |
|
1285 |
|
1286 // Reference Processing |
|
1287 static ReferenceProcessor* const ref_processor() { return _ref_processor; } |
|
1288 |
|
1289 static STWGCTimer* gc_timer() { return &_gc_timer; } |
|
1290 |
|
1291 // Return the SpaceId for the given address. |
|
1292 static SpaceId space_id(HeapWord* addr); |
|
1293 |
|
1294 // Time since last full gc (in milliseconds). |
|
1295 static jlong millis_since_last_gc(); |
|
1296 |
|
1297 static void print_on_error(outputStream* st); |
|
1298 |
|
1299 #ifndef PRODUCT |
|
1300 // Debugging support. |
|
1301 static const char* space_names[last_space_id]; |
|
1302 static void print_region_ranges(); |
|
1303 static void print_dense_prefix_stats(const char* const algorithm, |
|
1304 const SpaceId id, |
|
1305 const bool maximum_compaction, |
|
1306 HeapWord* const addr); |
|
1307 static void summary_phase_msg(SpaceId dst_space_id, |
|
1308 HeapWord* dst_beg, HeapWord* dst_end, |
|
1309 SpaceId src_space_id, |
|
1310 HeapWord* src_beg, HeapWord* src_end); |
|
1311 #endif // #ifndef PRODUCT |
|
1312 |
|
1313 #ifdef ASSERT |
|
1314 // Sanity check the new location of a word in the heap. |
|
1315 static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr); |
|
1316 // Verify that all the regions have been emptied. |
|
1317 static void verify_complete(SpaceId space_id); |
|
1318 #endif // #ifdef ASSERT |
|
1319 }; |
|
1320 |
|
1321 inline bool PSParallelCompact::mark_obj(oop obj) { |
|
1322 const int obj_size = obj->size(); |
|
1323 if (mark_bitmap()->mark_obj(obj, obj_size)) { |
|
1324 _summary_data.add_obj(obj, obj_size); |
|
1325 return true; |
|
1326 } else { |
|
1327 return false; |
|
1328 } |
|
1329 } |
|
1330 |
|
1331 inline bool PSParallelCompact::is_marked(oop obj) { |
|
1332 return mark_bitmap()->is_marked(obj); |
|
1333 } |
|
1334 |
|
1335 template <class T> |
|
1336 inline void PSParallelCompact::follow_root(ParCompactionManager* cm, T* p) { |
|
1337 assert(!Universe::heap()->is_in_reserved(p), |
|
1338 "roots shouldn't be things within the heap"); |
|
1339 |
|
1340 T heap_oop = oopDesc::load_heap_oop(p); |
|
1341 if (!oopDesc::is_null(heap_oop)) { |
|
1342 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
|
1343 if (mark_bitmap()->is_unmarked(obj)) { |
|
1344 if (mark_obj(obj)) { |
|
1345 obj->follow_contents(cm); |
|
1346 } |
|
1347 } |
|
1348 } |
|
1349 cm->follow_marking_stacks(); |
|
1350 } |
|
1351 |
|
1352 template <class T> |
|
1353 inline void PSParallelCompact::mark_and_push(ParCompactionManager* cm, T* p) { |
|
1354 T heap_oop = oopDesc::load_heap_oop(p); |
|
1355 if (!oopDesc::is_null(heap_oop)) { |
|
1356 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
|
1357 if (mark_bitmap()->is_unmarked(obj) && mark_obj(obj)) { |
|
1358 cm->push(obj); |
|
1359 } |
|
1360 } |
|
1361 } |
|
1362 |
|
1363 template <class T> |
|
1364 inline void PSParallelCompact::adjust_pointer(T* p) { |
|
1365 T heap_oop = oopDesc::load_heap_oop(p); |
|
1366 if (!oopDesc::is_null(heap_oop)) { |
|
1367 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
|
1368 oop new_obj = (oop)summary_data().calc_new_pointer(obj); |
|
1369 assert(new_obj != NULL, // is forwarding ptr? |
|
1370 "should be forwarded"); |
|
1371 // Just always do the update unconditionally? |
|
1372 if (new_obj != NULL) { |
|
1373 assert(Universe::heap()->is_in_reserved(new_obj), |
|
1374 "should be in object space"); |
|
1375 oopDesc::encode_store_heap_oop_not_null(p, new_obj); |
|
1376 } |
|
1377 } |
|
1378 } |
|
1379 |
|
1380 inline void PSParallelCompact::follow_klass(ParCompactionManager* cm, Klass* klass) { |
|
1381 oop holder = klass->klass_holder(); |
|
1382 PSParallelCompact::mark_and_push(cm, &holder); |
|
1383 } |
|
1384 |
|
1385 template <class T> |
|
1386 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) { |
|
1387 mark_and_push(_compaction_manager, p); |
|
1388 } |
|
1389 |
|
1390 inline bool PSParallelCompact::print_phases() { |
|
1391 return _print_phases; |
|
1392 } |
|
1393 |
|
1394 inline double PSParallelCompact::normal_distribution(double density) { |
|
1395 assert(_dwl_initialized, "uninitialized"); |
|
1396 const double squared_term = (density - _dwl_mean) / _dwl_std_dev; |
|
1397 return _dwl_first_term * exp(-0.5 * squared_term * squared_term); |
|
1398 } |
|
1399 |
|
1400 inline bool |
|
1401 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region, |
|
1402 idx_t bit) |
|
1403 { |
|
1404 assert(bit > 0, "cannot call this for the first bit/region"); |
|
1405 assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit), |
|
1406 "sanity check"); |
|
1407 |
|
1408 // Dead space crosses the boundary if (1) a partial object does not extend |
|
1409 // onto the region, (2) an object does not start at the beginning of the |
|
1410 // region, and (3) an object does not end at the end of the prior region. |
|
1411 return region->partial_obj_size() == 0 && |
|
1412 !_mark_bitmap.is_obj_beg(bit) && |
|
1413 !_mark_bitmap.is_obj_end(bit - 1); |
|
1414 } |
|
1415 |
|
1416 inline bool |
|
1417 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) { |
|
1418 return p >= beg_addr && p < end_addr; |
|
1419 } |
|
1420 |
|
1421 inline bool |
|
1422 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) { |
|
1423 return is_in((HeapWord*)p, beg_addr, end_addr); |
|
1424 } |
|
1425 |
|
1426 inline MutableSpace* PSParallelCompact::space(SpaceId id) { |
|
1427 assert(id < last_space_id, "id out of range"); |
|
1428 return _space_info[id].space(); |
|
1429 } |
|
1430 |
|
1431 inline HeapWord* PSParallelCompact::new_top(SpaceId id) { |
|
1432 assert(id < last_space_id, "id out of range"); |
|
1433 return _space_info[id].new_top(); |
|
1434 } |
|
1435 |
|
1436 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) { |
|
1437 assert(id < last_space_id, "id out of range"); |
|
1438 return _space_info[id].dense_prefix(); |
|
1439 } |
|
1440 |
|
1441 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) { |
|
1442 assert(id < last_space_id, "id out of range"); |
|
1443 return _space_info[id].start_array(); |
|
1444 } |
|
1445 |
|
1446 #ifdef ASSERT |
|
1447 inline void |
|
1448 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr) |
|
1449 { |
|
1450 assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr), |
|
1451 "must move left or to a different space"); |
|
1452 assert(is_object_aligned((intptr_t)old_addr) && is_object_aligned((intptr_t)new_addr), |
|
1453 "checking alignment"); |
|
1454 } |
|
1455 #endif // ASSERT |
|
1456 |
|
1457 class MoveAndUpdateClosure: public ParMarkBitMapClosure { |
|
1458 public: |
|
1459 inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm, |
|
1460 ObjectStartArray* start_array, |
|
1461 HeapWord* destination, size_t words); |
|
1462 |
|
1463 // Accessors. |
|
1464 HeapWord* destination() const { return _destination; } |
|
1465 |
|
1466 // If the object will fit (size <= words_remaining()), copy it to the current |
|
1467 // destination, update the interior oops and the start array and return either |
|
1468 // full (if the closure is full) or incomplete. If the object will not fit, |
|
1469 // return would_overflow. |
|
1470 virtual IterationStatus do_addr(HeapWord* addr, size_t size); |
|
1471 |
|
1472 // Copy enough words to fill this closure, starting at source(). Interior |
|
1473 // oops and the start array are not updated. Return full. |
|
1474 IterationStatus copy_until_full(); |
|
1475 |
|
1476 // Copy enough words to fill this closure or to the end of an object, |
|
1477 // whichever is smaller, starting at source(). Interior oops and the start |
|
1478 // array are not updated. |
|
1479 void copy_partial_obj(); |
|
1480 |
|
1481 protected: |
|
1482 // Update variables to indicate that word_count words were processed. |
|
1483 inline void update_state(size_t word_count); |
|
1484 |
|
1485 protected: |
|
1486 ObjectStartArray* const _start_array; |
|
1487 HeapWord* _destination; // Next addr to be written. |
|
1488 }; |
|
1489 |
|
1490 inline |
|
1491 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap, |
|
1492 ParCompactionManager* cm, |
|
1493 ObjectStartArray* start_array, |
|
1494 HeapWord* destination, |
|
1495 size_t words) : |
|
1496 ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array) |
|
1497 { |
|
1498 _destination = destination; |
|
1499 } |
|
1500 |
|
1501 inline void MoveAndUpdateClosure::update_state(size_t words) |
|
1502 { |
|
1503 decrement_words_remaining(words); |
|
1504 _source += words; |
|
1505 _destination += words; |
|
1506 } |
|
1507 |
|
1508 class UpdateOnlyClosure: public ParMarkBitMapClosure { |
|
1509 private: |
|
1510 const PSParallelCompact::SpaceId _space_id; |
|
1511 ObjectStartArray* const _start_array; |
|
1512 |
|
1513 public: |
|
1514 UpdateOnlyClosure(ParMarkBitMap* mbm, |
|
1515 ParCompactionManager* cm, |
|
1516 PSParallelCompact::SpaceId space_id); |
|
1517 |
|
1518 // Update the object. |
|
1519 virtual IterationStatus do_addr(HeapWord* addr, size_t words); |
|
1520 |
|
1521 inline void do_addr(HeapWord* addr); |
|
1522 }; |
|
1523 |
|
1524 inline void UpdateOnlyClosure::do_addr(HeapWord* addr) |
|
1525 { |
|
1526 _start_array->allocate_block(addr); |
|
1527 oop(addr)->update_contents(compaction_manager()); |
|
1528 } |
|
1529 |
|
1530 class FillClosure: public ParMarkBitMapClosure |
|
1531 { |
|
1532 public: |
|
1533 FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) : |
|
1534 ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm), |
|
1535 _start_array(PSParallelCompact::start_array(space_id)) |
|
1536 { |
|
1537 assert(space_id == PSParallelCompact::old_space_id, |
|
1538 "cannot use FillClosure in the young gen"); |
|
1539 } |
|
1540 |
|
1541 virtual IterationStatus do_addr(HeapWord* addr, size_t size) { |
|
1542 CollectedHeap::fill_with_objects(addr, size); |
|
1543 HeapWord* const end = addr + size; |
|
1544 do { |
|
1545 _start_array->allocate_block(addr); |
|
1546 addr += oop(addr)->size(); |
|
1547 } while (addr < end); |
|
1548 return ParMarkBitMap::incomplete; |
|
1549 } |
|
1550 |
|
1551 private: |
|
1552 ObjectStartArray* const _start_array; |
|
1553 }; |
|
1554 |
|
1555 #endif // SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP |