mirrored from https://gitlab.haskell.org/ghc/ghc.git
-
Notifications
You must be signed in to change notification settings - Fork 704
/
GC.c
1753 lines (1475 loc) · 52.7 KB
/
GC.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* -----------------------------------------------------------------------------
*
* (c) The GHC Team 1998-2008
*
* Generational garbage collector
*
* Documentation on the architecture of the Garbage Collector can be
* found in the online commentary:
*
* http://ghc.haskell.org/trac/ghc/wiki/Commentary/Rts/Storage/GC
*
* ---------------------------------------------------------------------------*/
#include "PosixSource.h"
#include "Rts.h"
#include "HsFFI.h"
#include "GC.h"
#include "GCThread.h"
#include "GCTDecl.h" // NB. before RtsSignals.h which
// clobbers REG_R1 on arm/Linux
#include "Compact.h"
#include "Evac.h"
#include "Scav.h"
#include "GCUtils.h"
#include "MarkStack.h"
#include "MarkWeak.h"
#include "Sparks.h"
#include "Sweep.h"
#include "Storage.h"
#include "RtsUtils.h"
#include "Apply.h"
#include "Updates.h"
#include "Stats.h"
#include "Schedule.h"
#include "Sanity.h"
#include "BlockAlloc.h"
#include "ProfHeap.h"
#include "Weak.h"
#include "Prelude.h"
#include "RtsSignals.h"
#include "STM.h"
#include "Trace.h"
#include "RetainerProfile.h"
#include "LdvProfile.h"
#include "RaiseAsync.h"
#include "Papi.h"
#include "Stable.h"
#include "CheckUnload.h"
#include <string.h> // for memset()
#include <unistd.h>
/* -----------------------------------------------------------------------------
Global variables
-------------------------------------------------------------------------- */
/* STATIC OBJECT LIST.
*
* During GC:
* We maintain a linked list of static objects that are still live.
* The requirements for this list are:
*
* - we need to scan the list while adding to it, in order to
* scavenge all the static objects (in the same way that
* breadth-first scavenging works for dynamic objects).
*
* - we need to be able to tell whether an object is already on
* the list, to break loops.
*
* Each static object has a "static link field", which we use for
* linking objects on to the list. We use a stack-type list, consing
* objects on the front as they are added (this means that the
* scavenge phase is depth-first, not breadth-first, but that
* shouldn't matter).
*
* A separate list is kept for objects that have been scavenged
* already - this is so that we can zero all the marks afterwards.
*
* An object is on the list if its static link field is non-zero; this
* means that we have to mark the end of the list with '1', not NULL.
*
* Extra notes for generational GC:
*
* Each generation has a static object list associated with it. When
* collecting generations up to N, we treat the static object lists
* from generations > N as roots.
*
* We build up a static object list while collecting generations 0..N,
* which is then appended to the static object list of generation N+1.
*/
/* N is the oldest generation being collected, where the generations
* are numbered starting at 0. A major GC (indicated by the major_gc
* flag) is when we're collecting all generations. We only attempt to
* deal with static objects and GC CAFs when doing a major GC.
*/
nat N;
rtsBool major_gc;
/* Data used for allocation area sizing.
*/
static W_ g0_pcnt_kept = 30; // percentage of g0 live at last minor GC
/* Mut-list stats */
#ifdef DEBUG
nat mutlist_MUTVARS,
mutlist_MUTARRS,
mutlist_MVARS,
mutlist_TVAR,
mutlist_TVAR_WATCH_QUEUE,
mutlist_TREC_CHUNK,
mutlist_TREC_HEADER,
mutlist_ATOMIC_INVARIANT,
mutlist_INVARIANT_CHECK_QUEUE,
mutlist_OTHERS;
#endif
/* Thread-local data for each GC thread
*/
gc_thread **gc_threads = NULL;
#if !defined(THREADED_RTS)
StgWord8 the_gc_thread[sizeof(gc_thread) + 64 * sizeof(gen_workspace)];
#endif
// Number of threads running in *this* GC. Affects how many
// step->todos[] lists we have to look in to find work.
nat n_gc_threads;
// For stats:
long copied; // *words* copied & scavenged during this GC
rtsBool work_stealing;
DECLARE_GCT
/* -----------------------------------------------------------------------------
Static function declarations
-------------------------------------------------------------------------- */
static void mark_root (void *user, StgClosure **root);
static void zero_static_object_list (StgClosure* first_static);
static void prepare_collected_gen (generation *gen);
static void prepare_uncollected_gen (generation *gen);
static void init_gc_thread (gc_thread *t);
static void resize_generations (void);
static void resize_nursery (void);
static void start_gc_threads (void);
static void scavenge_until_all_done (void);
static StgWord inc_running (void);
static StgWord dec_running (void);
static void wakeup_gc_threads (nat me);
static void shutdown_gc_threads (nat me);
static void collect_gct_blocks (void);
static void collect_pinned_object_blocks (void);
#if defined(DEBUG)
static void gcCAFs (void);
#endif
/* -----------------------------------------------------------------------------
The mark stack.
-------------------------------------------------------------------------- */
bdescr *mark_stack_top_bd; // topmost block in the mark stack
bdescr *mark_stack_bd; // current block in the mark stack
StgPtr mark_sp; // pointer to the next unallocated mark stack entry
/* -----------------------------------------------------------------------------
GarbageCollect: the main entry point to the garbage collector.
The collect_gen parameter is gotten by calling calcNeeded().
Locks held: all capabilities are held throughout GarbageCollect().
-------------------------------------------------------------------------- */
void
GarbageCollect (nat collect_gen,
rtsBool do_heap_census,
nat gc_type USED_IF_THREADS,
Capability *cap)
{
bdescr *bd;
generation *gen;
StgWord live_blocks, live_words, par_max_copied, par_tot_copied;
#if defined(THREADED_RTS)
gc_thread *saved_gct;
#endif
nat g, n;
// necessary if we stole a callee-saves register for gct:
#if defined(THREADED_RTS)
saved_gct = gct;
#endif
#ifdef PROFILING
CostCentreStack *save_CCS[n_capabilities];
#endif
ACQUIRE_SM_LOCK;
#if defined(RTS_USER_SIGNALS)
if (RtsFlags.MiscFlags.install_signal_handlers) {
// block signals
blockUserSignals();
}
#endif
ASSERT(sizeof(gen_workspace) == 16 * sizeof(StgWord));
// otherwise adjust the padding in gen_workspace.
// this is the main thread
SET_GCT(gc_threads[cap->no]);
// tell the stats department that we've started a GC
stat_startGC(cap, gct);
// lock the StablePtr table
stableLock();
#ifdef DEBUG
mutlist_MUTVARS = 0;
mutlist_MUTARRS = 0;
mutlist_MVARS = 0;
mutlist_TVAR = 0;
mutlist_TVAR_WATCH_QUEUE = 0;
mutlist_TREC_CHUNK = 0;
mutlist_TREC_HEADER = 0;
mutlist_ATOMIC_INVARIANT = 0;
mutlist_INVARIANT_CHECK_QUEUE = 0;
mutlist_OTHERS = 0;
#endif
// attribute any costs to CCS_GC
#ifdef PROFILING
for (n = 0; n < n_capabilities; n++) {
save_CCS[n] = capabilities[n]->r.rCCCS;
capabilities[n]->r.rCCCS = CCS_GC;
}
#endif
/* Figure out which generation to collect
*/
N = collect_gen;
major_gc = (N == RtsFlags.GcFlags.generations-1);
#if defined(THREADED_RTS)
work_stealing = RtsFlags.ParFlags.parGcLoadBalancingEnabled &&
N >= RtsFlags.ParFlags.parGcLoadBalancingGen;
// It's not always a good idea to do load balancing in parallel
// GC. In particular, for a parallel program we don't want to
// lose locality by moving cached data into another CPU's cache
// (this effect can be quite significant).
//
// We could have a more complex way to deterimine whether to do
// work stealing or not, e.g. it might be a good idea to do it
// if the heap is big. For now, we just turn it on or off with
// a flag.
#endif
/* Start threads, so they can be spinning up while we finish initialisation.
*/
start_gc_threads();
#if defined(THREADED_RTS)
/* How many threads will be participating in this GC?
* We don't try to parallelise minor GCs (unless the user asks for
* it with +RTS -gn0), or mark/compact/sweep GC.
*/
if (gc_type == SYNC_GC_PAR) {
n_gc_threads = n_capabilities;
} else {
n_gc_threads = 1;
}
#else
n_gc_threads = 1;
#endif
debugTrace(DEBUG_gc, "GC (gen %d, using %d thread(s))",
N, n_gc_threads);
#ifdef DEBUG
// check for memory leaks if DEBUG is on
memInventory(DEBUG_gc);
#endif
// do this *before* we start scavenging
collectFreshWeakPtrs();
// check sanity *before* GC
IF_DEBUG(sanity, checkSanity(rtsFalse /* before GC */, major_gc));
// gather blocks allocated using allocatePinned() from each capability
// and put them on the g0->large_object list.
collect_pinned_object_blocks();
// Initialise all the generations/steps that we're collecting.
for (g = 0; g <= N; g++) {
prepare_collected_gen(&generations[g]);
}
// Initialise all the generations/steps that we're *not* collecting.
for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
prepare_uncollected_gen(&generations[g]);
}
// Prepare this gc_thread
init_gc_thread(gct);
/* Allocate a mark stack if we're doing a major collection.
*/
if (major_gc && oldest_gen->mark) {
mark_stack_bd = allocBlock();
mark_stack_top_bd = mark_stack_bd;
mark_stack_bd->link = NULL;
mark_stack_bd->u.back = NULL;
mark_sp = mark_stack_bd->start;
} else {
mark_stack_bd = NULL;
mark_stack_top_bd = NULL;
mark_sp = NULL;
}
/* -----------------------------------------------------------------------
* follow all the roots that we know about:
*/
// the main thread is running: this prevents any other threads from
// exiting prematurely, so we can start them now.
// NB. do this after the mutable lists have been saved above, otherwise
// the other GC threads will be writing into the old mutable lists.
inc_running();
wakeup_gc_threads(gct->thread_index);
traceEventGcWork(gct->cap);
// scavenge the capability-private mutable lists. This isn't part
// of markSomeCapabilities() because markSomeCapabilities() can only
// call back into the GC via mark_root() (due to the gct register
// variable).
if (n_gc_threads == 1) {
for (n = 0; n < n_capabilities; n++) {
#if defined(THREADED_RTS)
scavenge_capability_mut_Lists1(capabilities[n]);
#else
scavenge_capability_mut_lists(capabilities[n]);
#endif
}
} else {
scavenge_capability_mut_lists(gct->cap);
for (n = 0; n < n_capabilities; n++) {
if (gc_threads[n]->idle) {
markCapability(mark_root, gct, capabilities[n],
rtsTrue/*don't mark sparks*/);
scavenge_capability_mut_lists(capabilities[n]);
}
}
}
// follow roots from the CAF list (used by GHCi)
gct->evac_gen_no = 0;
markCAFs(mark_root, gct);
// follow all the roots that the application knows about.
gct->evac_gen_no = 0;
if (n_gc_threads == 1) {
for (n = 0; n < n_capabilities; n++) {
markCapability(mark_root, gct, capabilities[n],
rtsTrue/*don't mark sparks*/);
}
} else {
markCapability(mark_root, gct, cap, rtsTrue/*don't mark sparks*/);
}
markScheduler(mark_root, gct);
#if defined(RTS_USER_SIGNALS)
// mark the signal handlers (signals should be already blocked)
markSignalHandlers(mark_root, gct);
#endif
// Mark the weak pointer list, and prepare to detect dead weak pointers.
markWeakPtrList();
initWeakForGC();
// Mark the stable pointer table.
markStableTables(mark_root, gct);
/* -------------------------------------------------------------------------
* Repeatedly scavenge all the areas we know about until there's no
* more scavenging to be done.
*/
for (;;)
{
scavenge_until_all_done();
// The other threads are now stopped. We might recurse back to
// here, but from now on this is the only thread.
// must be last... invariant is that everything is fully
// scavenged at this point.
if (traverseWeakPtrList()) { // returns rtsTrue if evaced something
inc_running();
continue;
}
// If we get to here, there's really nothing left to do.
break;
}
shutdown_gc_threads(gct->thread_index);
// Now see which stable names are still alive.
gcStableTables();
#ifdef THREADED_RTS
if (n_gc_threads == 1) {
for (n = 0; n < n_capabilities; n++) {
pruneSparkQueue(capabilities[n]);
}
} else {
for (n = 0; n < n_capabilities; n++) {
if (n == cap->no || gc_threads[n]->idle) {
pruneSparkQueue(capabilities[n]);
}
}
}
#endif
#ifdef PROFILING
// We call processHeapClosureForDead() on every closure destroyed during
// the current garbage collection, so we invoke LdvCensusForDead().
if (RtsFlags.ProfFlags.doHeapProfile == HEAP_BY_LDV
|| RtsFlags.ProfFlags.bioSelector != NULL) {
RELEASE_SM_LOCK; // LdvCensusForDead may need to take the lock
LdvCensusForDead(N);
ACQUIRE_SM_LOCK;
}
#endif
// NO MORE EVACUATION AFTER THIS POINT!
// Finally: compact or sweep the oldest generation.
if (major_gc && oldest_gen->mark) {
if (oldest_gen->compact)
compact(gct->scavenged_static_objects);
else
sweep(oldest_gen);
}
copied = 0;
par_max_copied = 0;
par_tot_copied = 0;
{
nat i;
for (i=0; i < n_gc_threads; i++) {
if (n_gc_threads > 1) {
debugTrace(DEBUG_gc,"thread %d:", i);
debugTrace(DEBUG_gc," copied %ld", gc_threads[i]->copied * sizeof(W_));
debugTrace(DEBUG_gc," scanned %ld", gc_threads[i]->scanned * sizeof(W_));
debugTrace(DEBUG_gc," any_work %ld", gc_threads[i]->any_work);
debugTrace(DEBUG_gc," no_work %ld", gc_threads[i]->no_work);
debugTrace(DEBUG_gc," scav_find_work %ld", gc_threads[i]->scav_find_work);
}
copied += gc_threads[i]->copied;
par_max_copied = stg_max(gc_threads[i]->copied, par_max_copied);
}
par_tot_copied = copied;
if (n_gc_threads == 1) {
par_max_copied = 0;
par_tot_copied = 0;
}
}
// Run through all the generations/steps and tidy up.
// We're going to:
// - count the amount of "live" data (live_words, live_blocks)
// - count the amount of "copied" data in this GC (copied)
// - free from-space
// - make to-space the new from-space (set BF_EVACUATED on all blocks)
//
live_words = 0;
live_blocks = 0;
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
if (g == N) {
generations[g].collections++; // for stats
if (n_gc_threads > 1) generations[g].par_collections++;
}
// Count the mutable list as bytes "copied" for the purposes of
// stats. Every mutable list is copied during every GC.
if (g > 0) {
W_ mut_list_size = 0;
for (n = 0; n < n_capabilities; n++) {
mut_list_size += countOccupied(capabilities[n]->mut_lists[g]);
}
copied += mut_list_size;
debugTrace(DEBUG_gc,
"mut_list_size: %lu (%d vars, %d arrays, %d MVARs, %d TVARs, %d TVAR_WATCH_QUEUEs, %d TREC_CHUNKs, %d TREC_HEADERs, %d ATOMIC_INVARIANTs, %d INVARIANT_CHECK_QUEUEs, %d others)",
(unsigned long)(mut_list_size * sizeof(W_)),
mutlist_MUTVARS, mutlist_MUTARRS, mutlist_MVARS,
mutlist_TVAR, mutlist_TVAR_WATCH_QUEUE,
mutlist_TREC_CHUNK, mutlist_TREC_HEADER,
mutlist_ATOMIC_INVARIANT,
mutlist_INVARIANT_CHECK_QUEUE,
mutlist_OTHERS);
}
bdescr *next, *prev;
gen = &generations[g];
// for generations we collected...
if (g <= N) {
/* free old memory and shift to-space into from-space for all
* the collected steps (except the allocation area). These
* freed blocks will probaby be quickly recycled.
*/
if (gen->mark)
{
// tack the new blocks on the end of the existing blocks
if (gen->old_blocks != NULL) {
prev = NULL;
for (bd = gen->old_blocks; bd != NULL; bd = next) {
next = bd->link;
if (!(bd->flags & BF_MARKED))
{
if (prev == NULL) {
gen->old_blocks = next;
} else {
prev->link = next;
}
freeGroup(bd);
gen->n_old_blocks--;
}
else
{
gen->n_words += bd->free - bd->start;
// NB. this step might not be compacted next
// time, so reset the BF_MARKED flags.
// They are set before GC if we're going to
// compact. (search for BF_MARKED above).
bd->flags &= ~BF_MARKED;
// between GCs, all blocks in the heap except
// for the nursery have the BF_EVACUATED flag set.
bd->flags |= BF_EVACUATED;
prev = bd;
}
}
if (prev != NULL) {
prev->link = gen->blocks;
gen->blocks = gen->old_blocks;
}
}
// add the new blocks to the block tally
gen->n_blocks += gen->n_old_blocks;
ASSERT(countBlocks(gen->blocks) == gen->n_blocks);
ASSERT(countOccupied(gen->blocks) == gen->n_words);
}
else // not copacted
{
freeChain(gen->old_blocks);
}
gen->old_blocks = NULL;
gen->n_old_blocks = 0;
/* LARGE OBJECTS. The current live large objects are chained on
* scavenged_large, having been moved during garbage
* collection from large_objects. Any objects left on the
* large_objects list are therefore dead, so we free them here.
*/
freeChain(gen->large_objects);
gen->large_objects = gen->scavenged_large_objects;
gen->n_large_blocks = gen->n_scavenged_large_blocks;
gen->n_large_words = countOccupied(gen->large_objects);
gen->n_new_large_words = 0;
}
else // for generations > N
{
/* For older generations, we need to append the
* scavenged_large_object list (i.e. large objects that have been
* promoted during this GC) to the large_object list for that step.
*/
for (bd = gen->scavenged_large_objects; bd; bd = next) {
next = bd->link;
dbl_link_onto(bd, &gen->large_objects);
gen->n_large_words += bd->free - bd->start;
}
// add the new blocks we promoted during this GC
gen->n_large_blocks += gen->n_scavenged_large_blocks;
}
ASSERT(countBlocks(gen->large_objects) == gen->n_large_blocks);
ASSERT(countOccupied(gen->large_objects) == gen->n_large_words);
gen->scavenged_large_objects = NULL;
gen->n_scavenged_large_blocks = 0;
// Count "live" data
live_words += genLiveWords(gen);
live_blocks += genLiveBlocks(gen);
// add in the partial blocks in the gen_workspaces, but ignore gen 0
// if this is a local GC (we can't count another capability's part_list)
{
nat i;
for (i = 0; i < n_capabilities; i++) {
live_words += gcThreadLiveWords(i, gen->no);
live_blocks += gcThreadLiveBlocks(i, gen->no);
}
}
} // for all generations
// update the max size of older generations after a major GC
resize_generations();
// Free the mark stack.
if (mark_stack_top_bd != NULL) {
debugTrace(DEBUG_gc, "mark stack: %d blocks",
countBlocks(mark_stack_top_bd));
freeChain(mark_stack_top_bd);
}
// Free any bitmaps.
for (g = 0; g <= N; g++) {
gen = &generations[g];
if (gen->bitmap != NULL) {
freeGroup(gen->bitmap);
gen->bitmap = NULL;
}
}
resize_nursery();
resetNurseries();
// mark the garbage collected CAFs as dead
#if defined(DEBUG)
if (major_gc) { gcCAFs(); }
#endif
// Update the stable pointer hash table.
updateStableTables(major_gc);
// unlock the StablePtr table. Must be before scheduleFinalizers(),
// because a finalizer may call hs_free_fun_ptr() or
// hs_free_stable_ptr(), both of which access the StablePtr table.
stableUnlock();
// Must be after stableUnlock(), because it might free stable ptrs.
if (major_gc) {
checkUnload (gct->scavenged_static_objects);
}
#ifdef PROFILING
// resetStaticObjectForRetainerProfiling() must be called before
// zeroing below.
// ToDo: fix the gct->scavenged_static_objects below
resetStaticObjectForRetainerProfiling(gct->scavenged_static_objects);
#endif
// zero the scavenged static object list
if (major_gc) {
nat i;
if (n_gc_threads == 1) {
zero_static_object_list(gct->scavenged_static_objects);
} else {
for (i = 0; i < n_gc_threads; i++) {
if (!gc_threads[i]->idle) {
zero_static_object_list(gc_threads[i]->scavenged_static_objects);
}
}
}
}
// Start any pending finalizers. Must be after
// updateStableTables() and stableUnlock() (see #4221).
RELEASE_SM_LOCK;
scheduleFinalizers(cap, dead_weak_ptr_list);
ACQUIRE_SM_LOCK;
// check sanity after GC
// before resurrectThreads(), because that might overwrite some
// closures, which will cause problems with THREADED where we don't
// fill slop.
IF_DEBUG(sanity, checkSanity(rtsTrue /* after GC */, major_gc));
// If a heap census is due, we need to do it before
// resurrectThreads(), for the same reason as checkSanity above:
// resurrectThreads() will overwrite some closures and leave slop
// behind.
if (do_heap_census) {
debugTrace(DEBUG_sched, "performing heap census");
RELEASE_SM_LOCK;
heapCensus(gct->gc_start_cpu);
ACQUIRE_SM_LOCK;
}
// send exceptions to any threads which were about to die
RELEASE_SM_LOCK;
resurrectThreads(resurrected_threads);
ACQUIRE_SM_LOCK;
if (major_gc) {
W_ need, got;
need = BLOCKS_TO_MBLOCKS(n_alloc_blocks);
got = mblocks_allocated;
/* If the amount of data remains constant, next major GC we'll
require (F+1)*need. We leave (F+2)*need in order to reduce
repeated deallocation and reallocation. */
need = (RtsFlags.GcFlags.oldGenFactor + 2) * need;
if (got > need) {
returnMemoryToOS(got - need);
}
}
// extra GC trace info
IF_DEBUG(gc, statDescribeGens());
#ifdef DEBUG
// symbol-table based profiling
/* heapCensus(to_blocks); */ /* ToDo */
#endif
// restore enclosing cost centre
#ifdef PROFILING
for (n = 0; n < n_capabilities; n++) {
capabilities[n]->r.rCCCS = save_CCS[n];
}
#endif
#ifdef DEBUG
// check for memory leaks if DEBUG is on
memInventory(DEBUG_gc);
#endif
// ok, GC over: tell the stats department what happened.
stat_endGC(cap, gct, live_words, copied,
live_blocks * BLOCK_SIZE_W - live_words /* slop */,
N, n_gc_threads, par_max_copied, par_tot_copied);
#if defined(RTS_USER_SIGNALS)
if (RtsFlags.MiscFlags.install_signal_handlers) {
// unblock signals again
unblockUserSignals();
}
#endif
RELEASE_SM_LOCK;
SET_GCT(saved_gct);
}
/* -----------------------------------------------------------------------------
Initialise the gc_thread structures.
-------------------------------------------------------------------------- */
#define GC_THREAD_INACTIVE 0
#define GC_THREAD_STANDING_BY 1
#define GC_THREAD_RUNNING 2
#define GC_THREAD_WAITING_TO_CONTINUE 3
static void
new_gc_thread (nat n, gc_thread *t)
{
nat g;
gen_workspace *ws;
t->cap = capabilities[n];
#ifdef THREADED_RTS
t->id = 0;
initSpinLock(&t->gc_spin);
initSpinLock(&t->mut_spin);
ACQUIRE_SPIN_LOCK(&t->gc_spin);
ACQUIRE_SPIN_LOCK(&t->mut_spin);
t->wakeup = GC_THREAD_INACTIVE; // starts true, so we can wait for the
// thread to start up, see wakeup_gc_threads
#endif
t->thread_index = n;
t->idle = rtsFalse;
t->free_blocks = NULL;
t->gc_count = 0;
init_gc_thread(t);
#ifdef USE_PAPI
t->papi_events = -1;
#endif
for (g = 0; g < RtsFlags.GcFlags.generations; g++)
{
ws = &t->gens[g];
ws->gen = &generations[g];
ASSERT(g == ws->gen->no);
ws->my_gct = t;
// We want to call
// alloc_todo_block(ws,0);
// but can't, because it uses gct which isn't set up at this point.
// Hence, allocate a block for todo_bd manually:
{
bdescr *bd = allocBlock(); // no lock, locks aren't initialised yet
initBdescr(bd, ws->gen, ws->gen->to);
bd->flags = BF_EVACUATED;
bd->u.scan = bd->free = bd->start;
ws->todo_bd = bd;
ws->todo_free = bd->free;
ws->todo_lim = bd->start + BLOCK_SIZE_W;
}
ws->todo_q = newWSDeque(128);
ws->todo_overflow = NULL;
ws->n_todo_overflow = 0;
ws->todo_large_objects = NULL;
ws->part_list = NULL;
ws->n_part_blocks = 0;
ws->scavd_list = NULL;
ws->n_scavd_blocks = 0;
}
}
void
initGcThreads (nat from USED_IF_THREADS, nat to USED_IF_THREADS)
{
#if defined(THREADED_RTS)
nat i;
if (from > 0) {
gc_threads = stgReallocBytes (gc_threads, to * sizeof(gc_thread*),
"initGcThreads");
} else {
gc_threads = stgMallocBytes (to * sizeof(gc_thread*),
"initGcThreads");
}
for (i = from; i < to; i++) {
gc_threads[i] =
stgMallocBytes(sizeof(gc_thread) +
RtsFlags.GcFlags.generations * sizeof(gen_workspace),
"alloc_gc_threads");
new_gc_thread(i, gc_threads[i]);
}
#else
ASSERT(from == 0 && to == 1);
gc_threads = stgMallocBytes (sizeof(gc_thread*),"alloc_gc_threads");
gc_threads[0] = gct;
new_gc_thread(0,gc_threads[0]);
#endif
}
void
freeGcThreads (void)
{
nat g;
if (gc_threads != NULL) {
#if defined(THREADED_RTS)
nat i;
for (i = 0; i < n_capabilities; i++) {
for (g = 0; g < RtsFlags.GcFlags.generations; g++)
{
freeWSDeque(gc_threads[i]->gens[g].todo_q);
}
stgFree (gc_threads[i]);
}
stgFree (gc_threads);
#else
for (g = 0; g < RtsFlags.GcFlags.generations; g++)
{
freeWSDeque(gc_threads[0]->gens[g].todo_q);
}
stgFree (gc_threads);
#endif
gc_threads = NULL;
}
}
/* ----------------------------------------------------------------------------
Start GC threads
------------------------------------------------------------------------- */
static volatile StgWord gc_running_threads;
static StgWord
inc_running (void)
{
StgWord new;
new = atomic_inc(&gc_running_threads, 1);
ASSERT(new <= n_gc_threads);
return new;
}
static StgWord
dec_running (void)
{
ASSERT(gc_running_threads != 0);
return atomic_dec(&gc_running_threads);
}
static rtsBool
any_work (void)
{
int g;
gen_workspace *ws;
gct->any_work++;
write_barrier();
// scavenge objects in compacted generation
if (mark_stack_bd != NULL && !mark_stack_empty()) {
return rtsTrue;
}
// Check for global work in any step. We don't need to check for
// local work, because we have already exited scavenge_loop(),
// which means there is no local work for this thread.
for (g = 0; g < (int)RtsFlags.GcFlags.generations; g++) {
ws = &gct->gens[g];
if (ws->todo_large_objects) return rtsTrue;
if (!looksEmptyWSDeque(ws->todo_q)) return rtsTrue;
if (ws->todo_overflow) return rtsTrue;
}
#if defined(THREADED_RTS)
if (work_stealing) {
nat n;
// look for work to steal
for (n = 0; n < n_gc_threads; n++) {
if (n == gct->thread_index) continue;
for (g = RtsFlags.GcFlags.generations-1; g >= 0; g--) {
ws = &gc_threads[n]->gens[g];
if (!looksEmptyWSDeque(ws->todo_q)) return rtsTrue;
}
}
}
#endif
gct->no_work++;
#if defined(THREADED_RTS)
yieldThread();
#endif
return rtsFalse;
}
static void
scavenge_until_all_done (void)
{
DEBUG_ONLY( nat r );
loop:
#if defined(THREADED_RTS)
if (n_gc_threads > 1) {
scavenge_loop();
} else {
scavenge_loop1();
}
#else
scavenge_loop();
#endif
collect_gct_blocks();
// scavenge_loop() only exits when there's no work to do
#ifdef DEBUG
r = dec_running();
#else
dec_running();
#endif
traceEventGcIdle(gct->cap);
debugTrace(DEBUG_gc, "%d GC threads still running", r);
while (gc_running_threads != 0) {
// usleep(1);
if (any_work()) {
inc_running();