1 /*
2 * Copyright (c) 2021, Amazon.com, Inc. or its affiliates. All rights reserved.
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This code is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 only, as
8 * published by the Free Software Foundation.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 *
24 */
25
26 #ifndef SHARE_GC_SHENANDOAH_SHENANDOAHSCANREMEMBERED_HPP
27 #define SHARE_GC_SHENANDOAH_SHENANDOAHSCANREMEMBERED_HPP
28
29 // Terminology used within this source file:
30 //
31 // Card Entry: This is the information that identifies whether a
32 // particular card-table entry is Clean or Dirty. A clean
33 // card entry denotes that the associated memory does not
34 // hold references to young-gen memory.
35 //
36 // Card Region, aka
37 // Card Memory: This is the region of memory that is assocated with a
38 // particular card entry.
39 //
40 // Card Cluster: A card cluster represents 64 card entries. A card
41 // cluster is the minimal amount of work performed at a
42 // time by a parallel thread. Note that the work required
43 // to scan a card cluster is somewhat variable in that the
44 // required effort depends on how many cards are dirty, how
45 // many references are held within the objects that span a
46 // DIRTY card's memory, and on the size of the object
47 // that spans the end of a DIRTY card's memory (because
48 // that object will be scanned in its entirety). For these
49 // reasons, it is advisable for the multiple worker threads
50 // to be flexible in the number of clusters to be
51 // processed by each thread.
52 //
53 // A cluster represents a "natural" quantum of work to be performed by
54 // a parallel GC thread's background remembered set scanning efforts.
55 // The notion of cluster is similar to the notion of stripe in the
56 // implementation of parallel GC card scanning. However, a cluster is
57 // typically smaller than a stripe, enabling finer grain division of
58 // labor between multiple threads.
59 //
60 // For illustration, consider the following possible JVM configurations:
61 //
62 // Scenario 1:
63 // RegionSize is 128 MB
64 // Span of a card entry is 512 B
65 // Each card table entry consumes 1 B
66 // Assume one long word of card table entries represents a cluster.
67 // This long word holds 8 card table entries, spanning a
68 // total of 4KB
69 // The number of clusters per region is 128 MB / 4 KB = 32K
70 //
71 // Scenario 2:
72 // RegionSize is 128 MB
73 // Span of each card entry is 128 B
74 // Each card table entry consumes 1 bit
75 // Assume one int word of card tables represents a cluster.
76 // This int word holds 32 card table entries, spanning a
77 // total of 4KB
78 // The number of clusters per region is 128 MB / 4 KB = 32K
79 //
80 // Scenario 3:
81 // RegionSize is 128 MB
82 // Span of each card entry is 512 B
83 // Each card table entry consumes 1 bit
84 // Assume one long word of card tables represents a cluster.
85 // This long word holds 64 card table entries, spanning a
86 // total of 32 KB
87 // The number of clusters per region is 128 MB / 32 KB = 4K
88 //
89 // At the start of a new young-gen concurrent mark pass, the gang of
90 // Shenandoah worker threads collaborate in performing the following
91 // actions:
92 //
93 // Let old_regions = number of ShenandoahHeapRegion comprising
94 // old-gen memory
95 // Let region_size = ShenandoahHeapRegion::region_size_bytes()
96 // represent the number of bytes in each region
97 // Let clusters_per_region = region_size / 512
98 // Let rs represent the relevant RememberedSet implementation
99 // (an instance of ShenandoahDirectCardMarkRememberedSet or an instance
100 // of a to-be-implemented ShenandoahBufferWithSATBRememberedSet)
101 //
102 // for each ShenandoahHeapRegion old_region in the whole heap
103 // determine the cluster number of the first cluster belonging
104 // to that region
105 // for each cluster contained within that region
106 // Assure that exactly one worker thread initializes each
107 // cluster of overreach memory by invoking:
108 //
109 // rs->initialize_overreach(cluster_no, cluster_count)
110 //
111 // in separate threads. (Divide up the clusters so that
112 // different threads are responsible for initializing different
113 // clusters. Initialization cost is essentially identical for
114 // each cluster.)
115 //
116 // Next, we repeat the process for invocations of process_clusters.
117 // for each ShenandoahHeapRegion old_region in the whole heap
118 // determine the cluster number of the first cluster belonging
119 // to that region
120 // for each cluster contained within that region
121 // Assure that exactly one worker thread processes each
122 // cluster, each thread making a series of invocations of the
123 // following:
124 //
125 // rs->process_clusters(worker_id, ReferenceProcessor *,
126 // ShenandoahConcurrentMark *, cluster_no, cluster_count,
127 // HeapWord *end_of_range, OopClosure *oops);
128 //
129 // For efficiency, divide up the clusters so that different threads
130 // are responsible for processing different clusters. Processing costs
131 // may vary greatly between clusters for the following reasons:
132 //
133 // a) some clusters contain mostly dirty cards and other
134 // clusters contain mostly clean cards
135 // b) some clusters contain mostly primitive data and other
136 // clusters contain mostly reference data
137 // c) some clusters are spanned by very large objects that
138 // begin in some other cluster. When a large object
139 // beginning in a preceding cluster spans large portions of
140 // this cluster, the processing of this cluster gets a
141 // "free ride" because the thread responsible for processing
142 // the cluster that holds the object's header does the
143 // processing.
144 // d) in the case that the end of this cluster is spanned by a
145 // very large object, the processing of this cluster will
146 // be responsible for examining the entire object,
147 // potentially requiring this thread to process large amounts
148 // of memory pertaining to other clusters.
149 //
150 // Though an initial division of labor between marking threads may
151 // assign equal numbers of clusters to be scanned by each thread, it
152 // should be expected that some threads will finish their assigned
153 // work before others. Therefore, some amount of the full remembered
154 // set scanning effort should be held back and assigned incrementally
155 // to the threads that end up with excess capacity. Consider the
156 // following strategy for dividing labor:
157 //
158 // 1. Assume there are 8 marking threads and 1024 remembered
159 // set clusters to be scanned.
160 // 2. Assign each thread to scan 64 clusters. This leaves
161 // 512 (1024 - (8*64)) clusters to still be scanned.
162 // 3. As the 8 server threads complete previous cluster
163 // scanning assignments, issue each of the next 8 scanning
164 // assignments as units of 32 additional cluster each.
165 // In the case that there is high variance in effort
166 // associated with previous cluster scanning assignments,
167 // multiples of these next assignments may be serviced by
168 // the server threads that were previously assigned lighter
169 // workloads.
170 // 4. Make subsequent scanning assignments as follows:
171 // a) 8 assignments of size 16 clusters
172 // b) 8 assignments of size 8 clusters
173 // c) 16 assignments of size 4 clusters
174 //
175 // When there is no more remembered set processing work to be
176 // assigned to a newly idled worker thread, that thread can move
177 // on to work on other tasks associated with root scanning until such
178 // time as all clusters have been examined.
179 //
180 // Once all clusters have been processed, the gang of GC worker
181 // threads collaborate to merge the overreach data.
182 //
183 // for each ShenandoahHeapRegion old_region in the whole heap
184 // determine the cluster number of the first cluster belonging
185 // to that region
186 // for each cluster contained within that region
187 // Assure that exactly one worker thread initializes each
188 // cluster of overreach memory by invoking:
189 //
190 // rs->merge_overreach(cluster_no, cluster_count)
191 //
192 // in separate threads. (Divide up the clusters so that
193 // different threads are responsible for merging different
194 // clusters. Merging cost is essentially identical for
195 // each cluster.)
196 //
197 // Though remembered set scanning is designed to run concurrently with
198 // mutator threads, the current implementation of remembered set
199 // scanning runs in parallel during a GC safepoint. Furthermore, the
200 // current implementation of remembered set scanning never clears a
201 // card once it has been marked. Since the current implementation
202 // never clears marked pages, the current implementation does not
203 // invoke initialize_overreach() or merge_overreach().
204 //
205 // These limitations will be addressed in future enhancements to the
206 // existing implementation.
207
208 #include <stdint.h>
209 #include "memory/iterator.hpp"
210 #include "gc/shared/workerThread.hpp"
211 #include "gc/shenandoah/shenandoahCardTable.hpp"
212 #include "gc/shenandoah/shenandoahHeapRegion.hpp"
213 #include "gc/shenandoah/shenandoahTaskqueue.hpp"
214
215 class ShenandoahReferenceProcessor;
216 class ShenandoahConcurrentMark;
217 class ShenandoahHeap;
218 class ShenandoahRegionIterator;
219 class ShenandoahMarkingContext;
220
221 class CardTable;
222
223 class ShenandoahDirectCardMarkRememberedSet: public CHeapObj<mtGC> {
224
225 private:
226
227 // Use symbolic constants defined in cardTable.hpp
228 // CardTable::card_shift = 9;
229 // CardTable::card_size = 512;
230 // CardTable::card_size_in_words = 64;
231
232 // CardTable::clean_card_val()
233 // CardTable::dirty_card_val()
234
235 ShenandoahHeap *_heap;
236 ShenandoahCardTable *_card_table;
237 size_t _card_shift;
238 size_t _total_card_count;
239 size_t _cluster_count;
240 HeapWord *_whole_heap_base; // Points to first HeapWord of data contained within heap memory
241 HeapWord *_whole_heap_end;
242 uint8_t *_byte_map; // Points to first entry within the card table
243 uint8_t *_byte_map_base; // Points to byte_map minus the bias computed from address of heap memory
244 uint8_t *_overreach_map; // Points to first entry within the overreach card table
245 uint8_t *_overreach_map_base; // Points to overreach_map minus the bias computed from address of heap memory
246
247 uint64_t _wide_clean_value;
248
249 public:
250 // count is the number of cards represented by the card table.
251 ShenandoahDirectCardMarkRememberedSet(ShenandoahCardTable *card_table, size_t total_card_count);
252 ~ShenandoahDirectCardMarkRememberedSet();
253
254 // Card index is zero-based relative to _byte_map.
255 size_t total_cards();
256 size_t card_index_for_addr(HeapWord *p);
257 HeapWord *addr_for_card_index(size_t card_index);
258 bool is_card_dirty(size_t card_index);
259 bool is_write_card_dirty(size_t card_index);
260 void mark_card_as_dirty(size_t card_index);
261 void mark_range_as_dirty(size_t card_index, size_t num_cards);
262 void mark_card_as_clean(size_t card_index);
263 void mark_read_card_as_clean(size_t card_index);
264 void mark_range_as_clean(size_t card_index, size_t num_cards);
265 void mark_overreach_card_as_dirty(size_t card_index);
266 bool is_card_dirty(HeapWord *p);
267 void mark_card_as_dirty(HeapWord *p);
268 void mark_range_as_dirty(HeapWord *p, size_t num_heap_words);
269 void mark_card_as_clean(HeapWord *p);
270 void mark_range_as_clean(HeapWord *p, size_t num_heap_words);
271 void mark_overreach_card_as_dirty(void *p);
272 size_t cluster_count();
273
274 // Called by multiple GC threads at start of concurrent mark and evacuation phases. Each parallel GC thread typically
275 // initializes a different subranges of all overreach entries.
276 void initialize_overreach(size_t first_cluster, size_t count);
277
278 // Called by GC thread at end of concurrent mark or evacuation phase. Each parallel GC thread typically merges different
279 // subranges of all overreach entries.
280 void merge_overreach(size_t first_cluster, size_t count);
281
282 // Called by GC thread at start of concurrent mark to exchange roles of read and write remembered sets.
283 // Not currently used because mutator write barrier does not honor changes to the location of card table.
284 void swap_remset() { _card_table->swap_card_tables(); }
285
286 void merge_write_table(HeapWord* start, size_t word_count) {
287 size_t card_index = card_index_for_addr(start);
288 size_t num_cards = word_count / CardTable::card_size_in_words();
289 size_t iterations = num_cards / (sizeof (intptr_t) / sizeof (CardTable::CardValue));
290 intptr_t* read_table_ptr = (intptr_t*) &(_card_table->read_byte_map())[card_index];
291 intptr_t* write_table_ptr = (intptr_t*) &(_card_table->write_byte_map())[card_index];
292 for (size_t i = 0; i < iterations; i++) {
293 intptr_t card_value = *write_table_ptr;
294 *read_table_ptr++ &= card_value;
295 write_table_ptr++;
296 }
297 }
298
299 HeapWord* whole_heap_base() { return _whole_heap_base; }
300 HeapWord* whole_heap_end() { return _whole_heap_end; }
301
302 // Instead of swap_remset, the current implementation of concurrent remembered set scanning does reset_remset
303 // in parallel threads, each invocation processing one entire HeapRegion at a time. Processing of a region
304 // consists of copying the write table to the read table and cleaning the write table.
305 void reset_remset(HeapWord* start, size_t word_count) {
306 size_t card_index = card_index_for_addr(start);
307 size_t num_cards = word_count / CardTable::card_size_in_words();
308 size_t iterations = num_cards / (sizeof (intptr_t) / sizeof (CardTable::CardValue));
309 intptr_t* read_table_ptr = (intptr_t*) &(_card_table->read_byte_map())[card_index];
310 intptr_t* write_table_ptr = (intptr_t*) &(_card_table->write_byte_map())[card_index];
311 for (size_t i = 0; i < iterations; i++) {
312 *read_table_ptr++ = *write_table_ptr;
313 *write_table_ptr++ = CardTable::clean_card_row_val();
314 }
315 }
316
317 // Called by GC thread after scanning old remembered set in order to prepare for next GC pass
318 void clear_old_remset() { _card_table->clear_read_table(); }
319
320 };
321
322 // A ShenandoahCardCluster represents the minimal unit of work
323 // performed by independent parallel GC threads during scanning of
324 // remembered sets.
325 //
326 // The GC threads that perform card-table remembered set scanning may
327 // overwrite card-table entries to mark them as clean in the case that
328 // the associated memory no longer holds references to young-gen
329 // memory. Rather than access the card-table entries directly, all GC
330 // thread access to card-table information is made by way of the
331 // ShenandoahCardCluster data abstraction. This abstraction
332 // effectively manages access to multiple possible underlying
333 // remembered set implementations, including a traditional card-table
334 // approach and a SATB-based approach.
335 //
336 // The API services represent a compromise between efficiency and
337 // convenience.
338 //
339 // In the initial implementation, we assume that scanning of card
340 // table entries occurs only while the JVM is at a safe point. Thus,
341 // there is no synchronization required between GC threads that are
342 // scanning card-table entries and marking certain entries that were
343 // previously dirty as clean, and mutator threads which would possibly
344 // be marking certain card-table entries as dirty.
345 //
346 // There is however a need to implement concurrency control and memory
347 // coherency between multiple GC threads that scan the remembered set
348 // in parallel. The desire is to divide the complete scanning effort
349 // into multiple clusters of work that can be independently processed
350 // by individual threads without need for synchronizing efforts
351 // between the work performed by each task. The term "cluster" of
352 // work is similar to the term "stripe" as used in the implementation
353 // of Parallel GC.
354 //
355 // Complexity arises when an object to be scanned crosses the boundary
356 // between adjacent cluster regions. Here is the protocol that is
357 // followed:
358 //
359 // 1. We implement a supplemental data structure known as the overreach
360 // card table. The thread that is responsible for scanning each
361 // cluster of card-table entries is granted exclusive access to
362 // modify the associated card-table entries. In the case that a
363 // thread scans a very large object that reaches into one or more
364 // following clusters, that thread has exclusive access to the
365 // overreach card table for all of the entries belonging to the
366 // following clusters that are spanned by this large object.
367 // After all clusters have been scanned, the scanning threads
368 // briefly synchronize to merge the contents of the overreach
369 // entries with the traditional card table entries using logical-
370 // and operations.
371 // 2. Every object is scanned in its "entirety" by the thread that is
372 // responsible for the cluster that holds its starting address.
373 // Entirety is in quotes because there are various situations in
374 // which some portions of the object will not be scanned by this
375 // thread:
376 // a) If an object spans multiple card regions, all of which are
377 // contained within the same cluster, the scanning thread
378 // consults the existing card-table entries and does not scan
379 // portions of the object that are not currently dirty.
380 // b) For any cluster that is spanned in its entirety by a very
381 // large object, the GC thread that scans this object assumes
382 // full responsibility for maintenance of the associated
383 // card-table entries.
384 // c) If a cluster is partially spanned by an object originating
385 // in a preceding cluster, the portion of the object that
386 // partially spans the following cluster is scanned in its
387 // entirety (because the thread that is responsible for
388 // scanning the object cannot rely upon the card-table entries
389 // associated with the following cluster). Whenever references
390 // to young-gen memory are found within the scanned data, the
391 // associated overreach card table entries are marked as dirty
392 // by the scanning thread.
393 // 3. If a cluster is spanned in its entirety by an object that
394 // originates within a preceding cluster's memory, the thread
395 // assigned to examine this cluster does absolutely nothing. The
396 // thread assigned to scan the cluster that holds the object's
397 // starting address takes full responsibility for scanning the
398 // entire object and updating the associated card-table entries.
399 // 4. If a cluster is spanned partially by an object that originates
400 // within a preceding cluster's memory, the thread assigned to
401 // examine this cluster marks the card-table entry as clean for
402 // each card table that is fully spanned by this overreaching
403 // object. If a card-table entry's memory is partially spanned
404 // by the overreaching object, the thread sets the card-table
405 // entry to clean if it was previously dirty and if the portion
406 // of the card-table entry's memory that is not spanned by the
407 // overreaching object does not hold pointers to young-gen
408 // memory.
409 // 5. While examining a particular card belonging to a particular
410 // cluster, if an object reaches beyond the end of its card
411 // memory, the thread "scans" all portions of the object that
412 // correspond to DIRTY card entries within the current cluster and
413 // all portions of the object that reach into following clustesr.
414 // After this object is scanned, continue scanning with the memory
415 // that follows this object if this memory pertains to the same
416 // cluster. Otherwise, consider this cluster's memory to have
417 // been fully examined.
418 //
419 // Discussion:
420 // Though this design results from careful consideration of multiple
421 // design objectives, it is subject to various criticisms. Some
422 // discussion of the design choices is provided here:
423 //
424 // 1. Note that remembered sets are a heuristic technique to avoid
425 // the need to scan all of old-gen memory with each young-gen
426 // collection. If we sometimes scan a bit more memory than is
427 // absolutely necessary, that should be considered a reasonable
428 // compromise. This compromise is already present in the sizing
429 // of card table memory areas. Note that a single dirty pointer
430 // within a 512-byte card region forces the "unnecessary" scanning
431 // of 63 = ((512 - 8 = 504) / 8) pointers.
432 // 2. One undesirable aspect of this design is that we sometimes have
433 // to scan large amounts of memory belonging to very large
434 // objects, even for parts of the very large object that do not
435 // correspond to dirty card table entries. Note that this design
436 // limits the amount of non-dirty scanning that might have to
437 // be performed for these very large objects. In particular, only
438 // the last part of the very large object that extends into but
439 // does not completely span a particular cluster is unnecessarily
440 // scanned. Thus, for each very large object, the maximum
441 // over-scan is the size of memory spanned by a single cluster.
442 // 3. The representation of pointer location descriptive information
443 // within Klass representations is not designed for efficient
444 // "random access". An alternative approach to this design would
445 // be to scan very large objects multiple times, once for each
446 // cluster that is spanned by the object's range. This reduces
447 // unnecessary overscan, but it introduces different sorts of
448 // overhead effort:
449 // i) For each spanned cluster, we have to look up the start of
450 // the crossing object.
451 // ii) Each time we scan the very large object, we have to
452 // sequentially walk through its pointer location
453 // descriptors, skipping over all of the pointers that
454 // precede the start of the range of addresses that we
455 // consider relevant.
456
457
458 // Because old-gen heap memory is not necessarily contiguous, and
459 // because cards are not necessarily maintained for young-gen memory,
460 // consecutive card numbers do not necessarily correspond to consecutive
461 // address ranges. For the traditional direct-card-marking
462 // implementation of this interface, consecutive card numbers are
463 // likely to correspond to contiguous regions of memory, but this
464 // should not be assumed. Instead, rely only upon the following:
465 //
466 // 1. All card numbers for cards pertaining to the same
467 // ShenandoahHeapRegion are consecutively numbered.
468 // 2. In the case that neighboring ShenandoahHeapRegions both
469 // represent old-gen memory, the card regions that span the
470 // boundary between these neighboring heap regions will be
471 // consecutively numbered.
472 // 3. (A corollary) In the case that an old-gen object spans the
473 // boundary between two heap regions, the card regions that
474 // correspond to the span of this object will be consecutively
475 // numbered.
476
477
478 // ShenandoahCardCluster abstracts access to the remembered set
479 // and also keeps track of crossing map information to allow efficient
480 // resolution of object start addresses.
481 //
482 // ShenandoahCardCluster supports all of the services of
483 // RememberedSet, plus it supports register_object() and lookup_object().
484 //
485 // There are two situations under which we need to know the location
486 // at which the object spanning the start of a particular card-table
487 // memory region begins:
488 //
489 // 1. When we begin to scan dirty card memory that is not the
490 // first card region within a cluster, and the object that
491 // crosses into this card memory was not previously scanned,
492 // we need to find where that object starts so we can scan it.
493 // (Asides: if the objects starts within a previous cluster, it
494 // has already been scanned. If the object starts within this
495 // cluster and it spans at least one card region that is dirty
496 // and precedes this card region within the cluster, then it has
497 // already been scanned.)
498 // 2. When we are otherwise done scanning a complete cluster, if the
499 // last object within the cluster reaches into the following
500 // cluster, we need to scan this object. Thus, we need to find
501 // its starting location.
502 //
503 // The RememberedSet template parameter is intended to represent either
504 // ShenandoahDirectCardMarkRememberedSet, or a to-be-implemented
505 // ShenandoahBufferWithSATBRememberedSet.
506 template<typename RememberedSet>
507 class ShenandoahCardCluster: public CHeapObj<mtGC> {
508
509 private:
510 RememberedSet *_rs;
511
512 public:
513 static const size_t CardsPerCluster = 64;
514
515 private:
516 typedef struct cross_map { uint8_t first; uint8_t last; } xmap;
517 typedef union crossing_info { uint16_t short_word; xmap offsets; } crossing_info;
518
519 // ObjectStartsInCardRegion bit is set within a crossing_info.offsets.start iff at least one object starts within
520 // a particular card region. We pack this bit into start byte under assumption that start byte is accessed less
521 // frequently that last byte. This is true when number of clean cards is greater than number of dirty cards.
522 static const uint16_t ObjectStartsInCardRegion = 0x80;
523 static const uint16_t FirstStartBits = 0x3f;
524
525 crossing_info *object_starts;
526
527 public:
528 // If we're setting first_start, assume the card has an object.
529 inline void set_first_start(size_t card_index, uint8_t value) {
530 object_starts[card_index].offsets.first = ObjectStartsInCardRegion | value;
531 }
532
533 inline void set_last_start(size_t card_index, uint8_t value) {
534 object_starts[card_index].offsets.last = value;
535 }
536
537 inline void set_has_object_bit(size_t card_index) {
538 object_starts[card_index].offsets.first |= ObjectStartsInCardRegion;
539 }
540
541 inline void clear_has_object_bit(size_t card_index) {
542 object_starts[card_index].offsets.first &= ~ObjectStartsInCardRegion;
543 }
544
545 // Returns true iff an object is known to start within the card memory associated with card card_index.
546 inline bool has_object(size_t card_index) {
547 return (object_starts[card_index].offsets.first & ObjectStartsInCardRegion) != 0;
548 }
549
550 inline void clear_objects_in_range(HeapWord *addr, size_t num_words) {
551 size_t card_index = _rs->card_index_for_addr(addr);
552 size_t last_card_index = _rs->card_index_for_addr(addr + num_words - 1);
553 while (card_index <= last_card_index)
554 object_starts[card_index++].short_word = 0;
555 }
556
557 ShenandoahCardCluster(RememberedSet *rs) {
558 _rs = rs;
559 // TODO: We don't really need object_starts entries for every card entry. We only need these for
560 // the card entries that correspond to old-gen memory. But for now, let's be quick and dirty.
561 object_starts = (crossing_info *) malloc(rs->total_cards() * sizeof(crossing_info));
562 if (object_starts == nullptr)
563 fatal("Insufficient memory for initializing heap");
564 for (size_t i = 0; i < rs->total_cards(); i++)
565 object_starts[i].short_word = 0;
566 }
567
568 ~ShenandoahCardCluster() {
569 if (object_starts != nullptr)
570 free(object_starts);
571 object_starts = nullptr;
572 }
573
574 // There is one entry within the object_starts array for each card entry.
575 //
576 // In the most recent implementation of ShenandoahScanRemembered::process_clusters(),
577 // there is no need for the get_crossing_object_start() method function, so there is no
578 // need to maintain the following information. The comment is left in place for now in
579 // case we find it necessary to add support for this service at a later time.
580 //
581 // Bits 0x7fff: If no object starts within this card region, the
582 // remaining bits of the object_starts array represent
583 // the absolute word offset within the enclosing
584 // cluster's memory of the starting address for the
585 // object that spans the start of this card region's
586 // memory. If the spanning object begins in memory
587 // that precedes this card region's cluster, the value
588 // stored in these bits is the special value 0x7fff.
589 // (Note that the maximum value required to represent a
590 // spanning object from within the current cluster is
591 // ((63 * 64) - 8), which equals 0x0fbf.
592 //
593 // In the absence of the need to support get_crossing_object_start(),
594 // here is discussion of performance:
595 //
596 // Suppose multiple garbage objects are coalesced during GC sweep
597 // into a single larger "free segment". As each two objects are
598 // coalesced together, the start information pertaining to the second
599 // object must be removed from the objects_starts array. If the
600 // second object had been been the first object within card memory,
601 // the new first object is the object that follows that object if
602 // that starts within the same card memory, or NoObject if the
603 // following object starts within the following cluster. If the
604 // second object had been the last object in the card memory,
605 // replace this entry with the newly coalesced object if it starts
606 // within the same card memory, or with NoObject if it starts in a
607 // preceding card's memory.
608 //
609 // Suppose a large free segment is divided into a smaller free
610 // segment and a new object. The second part of the newly divided
611 // memory must be registered as a new object, overwriting at most
612 // one first_start and one last_start entry. Note that one of the
613 // newly divided two objects might be a new GCLAB.
614 //
615 // Suppose postprocessing of a GCLAB finds that the original GCLAB
616 // has been divided into N objects. Each of the N newly allocated
617 // objects will be registered, overwriting at most one first_start
618 // and one last_start entries.
619 //
620 // No object registration operations are linear in the length of
621 // the registered objects.
622 //
623 // Consider further the following observations regarding object
624 // registration costs:
625 //
626 // 1. The cost is paid once for each old-gen object (Except when
627 // an object is demoted and repromoted, in which case we would
628 // pay the cost again).
629 // 2. The cost can be deferred so that there is no urgency during
630 // mutator copy-on-first-access promotion. Background GC
631 // threads will update the object_starts array by post-
632 // processing the contents of retired PLAB buffers.
633 // 3. The bet is that these costs are paid relatively rarely
634 // because:
635 // a) Most objects die young and objects that die in young-gen
636 // memory never need to be registered with the object_starts
637 // array.
638 // b) Most objects that are promoted into old-gen memory live
639 // there without further relocation for a relatively long
640 // time, so we get a lot of benefit from each investment
641 // in registering an object.
642
643 public:
644
645 // The starting locations of objects contained within old-gen memory
646 // are registered as part of the remembered set implementation. This
647 // information is required when scanning dirty card regions that are
648 // spanned by objects beginning within preceding card regions. It
649 // is necessary to find the first and last objects that begin within
650 // this card region. Starting addresses of objects are required to
651 // find the object headers, and object headers provide information
652 // about which fields within the object hold addresses.
653 //
654 // The old-gen memory allocator invokes register_object() for any
655 // object that is allocated within old-gen memory. This identifies
656 // the starting addresses of objects that span boundaries between
657 // card regions.
658 //
659 // It is not necessary to invoke register_object at the very instant
660 // an object is allocated. It is only necessary to invoke it
661 // prior to the next start of a garbage collection concurrent mark
662 // or concurrent update-references phase. An "ideal" time to register
663 // objects is during post-processing of a GCLAB after the GCLAB is
664 // retired due to depletion of its memory.
665 //
666 // register_object() does not perform synchronization. In the case
667 // that multiple threads are registering objects whose starting
668 // addresses are within the same cluster, races between these
669 // threads may result in corruption of the object-start data
670 // structures. Parallel GC threads should avoid registering objects
671 // residing within the same cluster by adhering to the following
672 // coordination protocols:
673 //
674 // 1. Align thread-local GCLAB buffers with some TBD multiple of
675 // card clusters. The card cluster size is 32 KB. If the
676 // desired GCLAB size is 128 KB, align the buffer on a multiple
677 // of 4 card clusters.
678 // 2. Post-process the contents of GCLAB buffers to register the
679 // objects allocated therein. Allow one GC thread at a
680 // time to do the post-processing of each GCLAB.
681 // 3. Since only one GC thread at a time is registering objects
682 // belonging to a particular allocation buffer, no locking
683 // is performed when registering these objects.
684 // 4. Any remnant of unallocated memory within an expended GC
685 // allocation buffer is not returned to the old-gen allocation
686 // pool until after the GC allocation buffer has been post
687 // processed. Before any remnant memory is returned to the
688 // old-gen allocation pool, the GC thread that scanned this GC
689 // allocation buffer performs a write-commit memory barrier.
690 // 5. Background GC threads that perform tenuring of young-gen
691 // objects without a GCLAB use a CAS lock before registering
692 // each tenured object. The CAS lock assures both mutual
693 // exclusion and memory coherency/visibility. Note that an
694 // object tenured by a background GC thread will not overlap
695 // with any of the clusters that are receiving tenured objects
696 // by way of GCLAB buffers. Multiple independent GC threads may
697 // attempt to tenure objects into a shared cluster. This is why
698 // sychronization may be necessary. Consider the following
699 // scenarios:
700 //
701 // a) If two objects are tenured into the same card region, each
702 // registration may attempt to modify the first-start or
703 // last-start information associated with that card region.
704 // Furthermore, because the representations of first-start
705 // and last-start information within the object_starts array
706 // entry uses different bits of a shared uint_16 to represent
707 // each, it is necessary to lock the entire card entry
708 // before modifying either the first-start or last-start
709 // information within the entry.
710 // b) Suppose GC thread X promotes a tenured object into
711 // card region A and this tenured object spans into
712 // neighboring card region B. Suppose GC thread Y (not equal
713 // to X) promotes a tenured object into cluster B. GC thread X
714 // will update the object_starts information for card A. No
715 // synchronization is required.
716 // c) In summary, when background GC threads register objects
717 // newly tenured into old-gen memory, they must acquire a
718 // mutual exclusion lock on the card that holds the starting
719 // address of the newly tenured object. This can be achieved
720 // by using a CAS instruction to assure that the previous
721 // values of first-offset and last-offset have not been
722 // changed since the same thread inquired as to their most
723 // current values.
724 //
725 // One way to minimize the need for synchronization between
726 // background tenuring GC threads is for each tenuring GC thread
727 // to promote young-gen objects into distinct dedicated cluster
728 // ranges.
729 // 6. The object_starts information is only required during the
730 // starting of concurrent marking and concurrent evacuation
731 // phases of GC. Before we start either of these GC phases, the
732 // JVM enters a safe point and all GC threads perform
733 // commit-write barriers to assure that access to the
734 // object_starts information is coherent.
735
736
737 // Notes on synchronization of register_object():
738 //
739 // 1. For efficiency, there is no locking in the implementation of register_object()
740 // 2. Thus, it is required that users of this service assure that concurrent/parallel invocations of
741 // register_object() do pertain to the same card's memory range. See discussion below to undestand
742 // the risks.
743 // 3. When allocating from a TLAB or GCLAB, the mutual exclusion can be guaranteed by assuring that each
744 // LAB's start and end are aligned on card memory boundaries.
745 // 4. Use the same lock that guarantees exclusivity when performing free-list allocation within heap regions.
746 //
747 // Register the newly allocated object while we're holding the global lock since there's no synchronization
748 // built in to the implementation of register_object(). There are potential races when multiple independent
749 // threads are allocating objects, some of which might span the same card region. For example, consider
750 // a card table's memory region within which three objects are being allocated by three different threads:
751 //
752 // objects being "concurrently" allocated:
753 // [-----a------][-----b-----][--------------c------------------]
754 // [---- card table memory range --------------]
755 //
756 // Before any objects are allocated, this card's memory range holds no objects. Note that:
757 // allocation of object a wants to set the has-object, first-start, and last-start attributes of the preceding card region.
758 // allocation of object b wants to set the has-object, first-start, and last-start attributes of this card region.
759 // allocation of object c also wants to set the has-object, first-start, and last-start attributes of this card region.
760 //
761 // The thread allocating b and the thread allocating c can "race" in various ways, resulting in confusion, such as last-start
762 // representing object b while first-start represents object c. This is why we need to require all register_object()
763 // invocations associated with objects that are allocated from "free lists" to provide their own mutual exclusion locking
764 // mechanism.
765
766 // Reset the has_object() information to false for all cards in the range between from and to.
767 void reset_object_range(HeapWord *from, HeapWord *to);
768
769 // register_object() requires that the caller hold the heap lock
770 // before calling it.
771 void register_object(HeapWord* address);
772
773 // register_object_wo_lock() does not require that the caller hold
774 // the heap lock before calling it, under the assumption that the
775 // caller has assure no other thread will endeavor to concurrently
776 // register objects that start within the same card's memory region
777 // as address.
778 void register_object_wo_lock(HeapWord* address);
779
780 // During the reference updates phase of GC, we walk through each old-gen memory region that was
781 // not part of the collection set and we invalidate all unmarked objects. As part of this effort,
782 // we coalesce neighboring dead objects in order to make future remembered set scanning more
783 // efficient (since future remembered set scanning of any card region containing consecutive
784 // dead objects can skip over all of them at once by reading only a single dead object header
785 // instead of having to read the header of each of the coalesced dead objects.
786 //
787 // At some future time, we may implement a further optimization: satisfy future allocation requests
788 // by carving new objects out of the range of memory that represents the coalesced dead objects.
789 //
790 // Suppose we want to combine several dead objects into a single coalesced object. How does this
791 // impact our representation of crossing map information?
792 // 1. If the newly coalesced range is contained entirely within a card range, that card's last
793 // start entry either remains the same or it is changed to the start of the coalesced region.
794 // 2. For the card that holds the start of the coalesced object, it will not impact the first start
795 // but it may impact the last start.
796 // 3. For following cards spanned entirely by the newly coalesced object, it will change has_object
797 // to false (and make first-start and last-start "undefined").
798 // 4. For a following card that is spanned patially by the newly coalesced object, it may change
799 // first-start value, but it will not change the last-start value.
800 //
801 // The range of addresses represented by the arguments to coalesce_objects() must represent a range
802 // of memory that was previously occupied exactly by one or more previously registered objects. For
803 // convenience, it is legal to invoke coalesce_objects() with arguments that span a single previously
804 // registered object.
805 //
806 // The role of coalesce_objects is to change the crossing map information associated with all of the coalesced
807 // objects.
808 void coalesce_objects(HeapWord* address, size_t length_in_words);
809
810 // The typical use case is going to look something like this:
811 // for each heapregion that comprises old-gen memory
812 // for each card number that corresponds to this heap region
813 // scan the objects contained therein if the card is dirty
814 // To avoid excessive lookups in a sparse array, the API queries
815 // the card number pertaining to a particular address and then uses the
816 // card noumber for subsequent information lookups and stores.
817
818 // If has_object(card_index), this returns the word offset within this card
819 // memory at which the first object begins. If !has_object(card_index), the
820 // result is a don't care value.
821 size_t get_first_start(size_t card_index);
822
823 // If has_object(card_index), this returns the word offset within this card
824 // memory at which the last object begins. If !has_object(card_index), the
825 // result is a don't care value.
826 size_t get_last_start(size_t card_index);
827
828 };
829
830 // ShenandoahScanRemembered is a concrete class representing the
831 // ability to scan the old-gen remembered set for references to
832 // objects residing in young-gen memory.
833 //
834 // Scanning normally begins with an invocation of numRegions and ends
835 // after all clusters of all regions have been scanned.
836 //
837 // Throughout the scanning effort, the number of regions does not
838 // change.
839 //
840 // Even though the regions that comprise old-gen memory are not
841 // necessarily contiguous, the abstraction represented by this class
842 // identifies each of the old-gen regions with an integer value
843 // in the range from 0 to (numRegions() - 1) inclusive.
844 //
845
846 template<typename RememberedSet>
847 class ShenandoahScanRemembered: public CHeapObj<mtGC> {
848
849 private:
850
851 RememberedSet* _rs;
852 ShenandoahCardCluster<RememberedSet>* _scc;
853
854 public:
855 // How to instantiate this object?
856 // ShenandoahDirectCardMarkRememberedSet *rs =
857 // new ShenandoahDirectCardMarkRememberedSet();
858 // scr = new
859 // ShenandoahScanRememberd<ShenandoahDirectCardMarkRememberedSet>(rs);
860 //
861 // or, after the planned implementation of
862 // ShenandoahBufferWithSATBRememberedSet has been completed:
863 //
864 // ShenandoahBufferWithSATBRememberedSet *rs =
865 // new ShenandoahBufferWithSATBRememberedSet();
866 // scr = new
867 // ShenandoahScanRememberd<ShenandoahBufferWithSATBRememberedSet>(rs);
868
869
870 ShenandoahScanRemembered(RememberedSet *rs) {
871 _rs = rs;
872 _scc = new ShenandoahCardCluster<RememberedSet>(rs);
873 }
874
875 ~ShenandoahScanRemembered() {
876 delete _scc;
877 }
878
879 // TODO: We really don't want to share all of these APIs with arbitrary consumers of the ShenandoahScanRemembered abstraction.
880 // But in the spirit of quick and dirty for the time being, I'm going to go ahead and publish everything for right now. Some
881 // of existing code already depends on having access to these services (because existing code has not been written to honor
882 // full abstraction of remembered set scanning. In the not too distant future, we want to try to make most, if not all, of
883 // these services private. Two problems with publicizing:
884 // 1. Allowing arbitrary users to reach beneath the hood allows the users to make assumptions about underlying implementation.
885 // This will make it more difficult to change underlying implementation at a future time, such as when we eventually experiment
886 // with SATB-based implementation of remembered set representation.
887 // 2. If we carefully control sharing of certain of these services, we can reduce the overhead of synchronization by assuring
888 // that all users follow protocols that avoid contention that might require synchronization. When we publish these APIs, we
889 // lose control over who and how the data is accessed. As a result, we are required to insert more defensive measures into
890 // the implementation, including synchronization locks.
891
892
893 // Card index is zero-based relative to first spanned card region.
894 size_t total_cards();
895 size_t card_index_for_addr(HeapWord *p);
896 HeapWord *addr_for_card_index(size_t card_index);
897 bool is_card_dirty(size_t card_index);
898 bool is_write_card_dirty(size_t card_index) { return _rs->is_write_card_dirty(card_index); }
899 void mark_card_as_dirty(size_t card_index);
900 void mark_range_as_dirty(size_t card_index, size_t num_cards);
901 void mark_card_as_clean(size_t card_index);
902 void mark_read_card_as_clean(size_t card_index) { _rs->mark_read_card_clean(card_index); }
903 void mark_range_as_clean(size_t card_index, size_t num_cards);
904 void mark_overreach_card_as_dirty(size_t card_index);
905 bool is_card_dirty(HeapWord *p);
906 void mark_card_as_dirty(HeapWord *p);
907 void mark_range_as_dirty(HeapWord *p, size_t num_heap_words);
908 void mark_card_as_clean(HeapWord *p);
909 void mark_range_as_clean(HeapWord *p, size_t num_heap_words);
910 void mark_overreach_card_as_dirty(void *p);
911 size_t cluster_count();
912 void initialize_overreach(size_t first_cluster, size_t count);
913 void merge_overreach(size_t first_cluster, size_t count);
914
915 // Called by GC thread at start of concurrent mark to exchange roles of read and write remembered sets.
916 void swap_remset() { _rs->swap_remset(); }
917
918 void reset_remset(HeapWord* start, size_t word_count) { _rs->reset_remset(start, word_count); }
919
920 void merge_write_table(HeapWord* start, size_t word_count) { _rs->merge_write_table(start, word_count); }
921
922 // Called by GC thread after scanning old remembered set in order to prepare for next GC pass
923 void clear_old_remset() { _rs->clear_old_remset(); }
924
925 size_t cluster_for_addr(HeapWord *addr);
926
927 void reset_object_range(HeapWord *from, HeapWord *to);
928 void register_object(HeapWord *addr);
929 void register_object_wo_lock(HeapWord *addr);
930 void coalesce_objects(HeapWord *addr, size_t length_in_words);
931
932 // Return true iff this object is "properly" registered.
933 bool verify_registration(HeapWord* address, ShenandoahMarkingContext* ctx);
934
935 // clear the cards to clean, and clear the object_starts info to no objects
936 void mark_range_as_empty(HeapWord *addr, size_t length_in_words);
937
938 // process_clusters() scans a portion of the remembered set during a JVM
939 // safepoint as part of the root scanning activities that serve to
940 // initiate concurrent scanning and concurrent evacuation. Multiple
941 // threads may scan different portions of the remembered set by
942 // making parallel invocations of process_clusters() with each
943 // invocation scanning different clusters of the remembered set.
944 //
945 // An invocation of process_clusters() examines all of the
946 // intergenerational references spanned by count clusters starting
947 // with first_cluster. The oops argument is assumed to represent a
948 // thread-local OopClosure into which addresses of intergenerational
949 // pointer values will be accumulated for the purposes of root scanning.
950 //
951 // A side effect of executing process_clusters() is to update the card
952 // table entries, marking dirty cards as clean if they no longer
953 // hold references to young-gen memory. (THIS IS NOT YET IMPLEMENTED.)
954 //
955 // The implementation of process_clusters() is designed to efficiently
956 // minimize work in the large majority of cases for which the
957 // associated cluster has very few dirty card-table entries.
958 //
959 // At initialization of concurrent marking, invoke process_clusters with
960 // ClosureType equal to ShenandoahInitMarkRootsClosure.
961 //
962 // At initialization of concurrent evacuation, invoke process_clusters with
963 // ClosureType equal to ShenandoahEvacuateUpdateRootsClosure.
964
965 // This is big enough it probably shouldn't be in-lined. On the other hand, there are only a few places this
966 // code is called from, so it might as well be in-lined. The "real" reason I'm inlining at the moment is because
967 // the template expansions were making it difficult for the link/loader to resolve references to the template-
968 // parameterized implementations of this service.
969 template <typename ClosureType>
970 inline void process_clusters(size_t first_cluster, size_t count, HeapWord *end_of_range, ClosureType *oops);
971
972 template <typename ClosureType>
973 inline void process_clusters(size_t first_cluster, size_t count, HeapWord *end_of_range, ClosureType *oops, bool use_write_table);
974
975 template <typename ClosureType>
976 inline void process_region(ShenandoahHeapRegion* region, ClosureType *cl);
977
978 template <typename ClosureType>
979 inline void process_region(ShenandoahHeapRegion* region, ClosureType *cl, bool use_write_table);
980
981 // To Do:
982 // Create subclasses of ShenandoahInitMarkRootsClosure and
983 // ShenandoahEvacuateUpdateRootsClosure and any other closures
984 // that need to participate in remembered set scanning. Within the
985 // subclasses, add a (probably templated) instance variable that
986 // refers to the associated ShenandoahCardCluster object. Use this
987 // ShenandoahCardCluster instance to "enhance" the do_oops
988 // processing so that we can:
989 //
990 // 1. Avoid processing references that correspond to clean card
991 // regions, and
992 // 2. Set card status to CLEAN when the associated card region no
993 // longer holds inter-generatioanal references.
994 //
995 // To enable efficient implementation of these behaviors, we
996 // probably also want to add a few fields into the
997 // ShenandoahCardCluster object that allow us to precompute and
998 // remember the addresses at which card status is going to change
999 // from dirty to clean and clean to dirty. The do_oops
1000 // implementations will want to update this value each time they
1001 // cross one of these boundaries.
1002 void roots_do(OopIterateClosure* cl);
1003 };
1004
1005 typedef ShenandoahScanRemembered<ShenandoahDirectCardMarkRememberedSet> RememberedScanner;
1006
1007 class ShenandoahScanRememberedTask : public WorkerTask {
1008 private:
1009 ShenandoahObjToScanQueueSet* _queue_set;
1010 ShenandoahObjToScanQueueSet* _old_queue_set;
1011 ShenandoahReferenceProcessor* _rp;
1012 ShenandoahRegionIterator* _regions;
1013 public:
1014 ShenandoahScanRememberedTask(ShenandoahObjToScanQueueSet* queue_set,
1015 ShenandoahObjToScanQueueSet* old_queue_set,
1016 ShenandoahReferenceProcessor* rp,
1017 ShenandoahRegionIterator* regions);
1018
1019 void work(uint worker_id);
1020 };
1021 #endif // SHARE_GC_SHENANDOAH_SHENANDOAHSCANREMEMBERED_HPP