mirrored from https://gitlab.haskell.org/ghc/ghc.git
/
GC.c
4339 lines (3680 loc) · 118 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 1998-2003
*
* Generational garbage collector
*
* ---------------------------------------------------------------------------*/
#include "PosixSource.h"
#include "Rts.h"
#include "RtsFlags.h"
#include "RtsUtils.h"
#include "Apply.h"
#include "OSThreads.h"
#include "Storage.h"
#include "LdvProfile.h"
#include "Updates.h"
#include "Stats.h"
#include "Schedule.h"
#include "Sanity.h"
#include "BlockAlloc.h"
#include "MBlock.h"
#include "ProfHeap.h"
#include "SchedAPI.h"
#include "Weak.h"
#include "Prelude.h"
#include "ParTicky.h" // ToDo: move into Rts.h
#include "GCCompact.h"
#include "Signals.h"
#include "STM.h"
#if defined(GRAN) || defined(PAR)
# include "GranSimRts.h"
# include "ParallelRts.h"
# include "FetchMe.h"
# if defined(DEBUG)
# include "Printer.h"
# include "ParallelDebug.h"
# endif
#endif
#include "HsFFI.h"
#include "Linker.h"
#if defined(RTS_GTK_FRONTPANEL)
#include "FrontPanel.h"
#endif
#include "RetainerProfile.h"
#include <string.h>
// Turn off inlining when debugging - it obfuscates things
#ifdef DEBUG
# undef STATIC_INLINE
# define STATIC_INLINE static
#endif
/* 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.
*/
static StgClosure* static_objects; // live static objects
StgClosure* scavenged_static_objects; // static objects scavenged so far
/* 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.
*/
static nat N;
static rtsBool major_gc;
/* Youngest generation that objects should be evacuated to in
* evacuate(). (Logically an argument to evacuate, but it's static
* a lot of the time so we optimise it into a global variable).
*/
static nat evac_gen;
/* Weak pointers
*/
StgWeak *old_weak_ptr_list; // also pending finaliser list
/* Which stage of processing various kinds of weak pointer are we at?
* (see traverse_weak_ptr_list() below for discussion).
*/
typedef enum { WeakPtrs, WeakThreads, WeakDone } WeakStage;
static WeakStage weak_stage;
/* List of all threads during GC
*/
static StgTSO *old_all_threads;
StgTSO *resurrected_threads;
/* Flag indicating failure to evacuate an object to the desired
* generation.
*/
static rtsBool failed_to_evac;
/* Old to-space (used for two-space collector only)
*/
static bdescr *old_to_blocks;
/* Data used for allocation area sizing.
*/
static lnat new_blocks; // blocks allocated during this GC
static lnat g0s0_pcnt_kept = 30; // percentage of g0s0 live at last minor GC
/* Used to avoid long recursion due to selector thunks
*/
static lnat thunk_selector_depth = 0;
#define MAX_THUNK_SELECTOR_DEPTH 8
/* -----------------------------------------------------------------------------
Static function declarations
-------------------------------------------------------------------------- */
static bdescr * gc_alloc_block ( step *stp );
static void mark_root ( StgClosure **root );
// Use a register argument for evacuate, if available.
#if __GNUC__ >= 2
#define REGPARM1 __attribute__((regparm(1)))
#else
#define REGPARM1
#endif
REGPARM1 static StgClosure * evacuate (StgClosure *q);
static void zero_static_object_list ( StgClosure* first_static );
static rtsBool traverse_weak_ptr_list ( void );
static void mark_weak_ptr_list ( StgWeak **list );
static StgClosure * eval_thunk_selector ( nat field, StgSelector * p );
static void scavenge ( step * );
static void scavenge_mark_stack ( void );
static void scavenge_stack ( StgPtr p, StgPtr stack_end );
static rtsBool scavenge_one ( StgPtr p );
static void scavenge_large ( step * );
static void scavenge_static ( void );
static void scavenge_mutable_list ( generation *g );
static void scavenge_large_bitmap ( StgPtr p,
StgLargeBitmap *large_bitmap,
nat size );
#if 0 && defined(DEBUG)
static void gcCAFs ( void );
#endif
/* -----------------------------------------------------------------------------
inline functions etc. for dealing with the mark bitmap & stack.
-------------------------------------------------------------------------- */
#define MARK_STACK_BLOCKS 4
static bdescr *mark_stack_bdescr;
static StgPtr *mark_stack;
static StgPtr *mark_sp;
static StgPtr *mark_splim;
// Flag and pointers used for falling back to a linear scan when the
// mark stack overflows.
static rtsBool mark_stack_overflowed;
static bdescr *oldgen_scan_bd;
static StgPtr oldgen_scan;
STATIC_INLINE rtsBool
mark_stack_empty(void)
{
return mark_sp == mark_stack;
}
STATIC_INLINE rtsBool
mark_stack_full(void)
{
return mark_sp >= mark_splim;
}
STATIC_INLINE void
reset_mark_stack(void)
{
mark_sp = mark_stack;
}
STATIC_INLINE void
push_mark_stack(StgPtr p)
{
*mark_sp++ = p;
}
STATIC_INLINE StgPtr
pop_mark_stack(void)
{
return *--mark_sp;
}
/* -----------------------------------------------------------------------------
Allocate a new to-space block in the given step.
-------------------------------------------------------------------------- */
static bdescr *
gc_alloc_block(step *stp)
{
bdescr *bd = allocBlock();
bd->gen_no = stp->gen_no;
bd->step = stp;
bd->link = NULL;
// blocks in to-space in generations up to and including N
// get the BF_EVACUATED flag.
if (stp->gen_no <= N) {
bd->flags = BF_EVACUATED;
} else {
bd->flags = 0;
}
// Start a new to-space block, chain it on after the previous one.
if (stp->hp_bd == NULL) {
stp->hp_bd = bd;
} else {
stp->hp_bd->free = stp->hp;
stp->hp_bd->link = bd;
stp->hp_bd = bd;
}
stp->hp = bd->start;
stp->hpLim = stp->hp + BLOCK_SIZE_W;
stp->n_to_blocks++;
new_blocks++;
return bd;
}
/* -----------------------------------------------------------------------------
GarbageCollect
Rough outline of the algorithm: for garbage collecting generation N
(and all younger generations):
- follow all pointers in the root set. the root set includes all
mutable objects in all generations (mutable_list).
- for each pointer, evacuate the object it points to into either
+ to-space of the step given by step->to, which is the next
highest step in this generation or the first step in the next
generation if this is the last step.
+ to-space of generations[evac_gen]->steps[0], if evac_gen != 0.
When we evacuate an object we attempt to evacuate
everything it points to into the same generation - this is
achieved by setting evac_gen to the desired generation. If
we can't do this, then an entry in the mut list has to
be made for the cross-generation pointer.
+ if the object is already in a generation > N, then leave
it alone.
- repeatedly scavenge to-space from each step in each generation
being collected until no more objects can be evacuated.
- free from-space in each step, and set from-space = to-space.
Locks held: sched_mutex
-------------------------------------------------------------------------- */
void
GarbageCollect ( void (*get_roots)(evac_fn), rtsBool force_major_gc )
{
bdescr *bd;
step *stp;
lnat live, allocated, collected = 0, copied = 0;
lnat oldgen_saved_blocks = 0;
nat g, s;
#ifdef PROFILING
CostCentreStack *prev_CCS;
#endif
#if defined(DEBUG) && defined(GRAN)
IF_DEBUG(gc, debugBelch("@@ Starting garbage collection at %ld (%lx)\n",
Now, Now));
#endif
#if defined(RTS_USER_SIGNALS)
// block signals
blockUserSignals();
#endif
// tell the STM to discard any cached closures its hoping to re-use
stmPreGCHook();
// tell the stats department that we've started a GC
stat_startGC();
// Init stats and print par specific (timing) info
PAR_TICKY_PAR_START();
// attribute any costs to CCS_GC
#ifdef PROFILING
prev_CCS = CCCS;
CCCS = CCS_GC;
#endif
/* Approximate how much we allocated.
* Todo: only when generating stats?
*/
allocated = calcAllocated();
/* Figure out which generation to collect
*/
if (force_major_gc) {
N = RtsFlags.GcFlags.generations - 1;
major_gc = rtsTrue;
} else {
N = 0;
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
if (generations[g].steps[0].n_blocks +
generations[g].steps[0].n_large_blocks
>= generations[g].max_blocks) {
N = g;
}
}
major_gc = (N == RtsFlags.GcFlags.generations-1);
}
#ifdef RTS_GTK_FRONTPANEL
if (RtsFlags.GcFlags.frontpanel) {
updateFrontPanelBeforeGC(N);
}
#endif
// check stack sanity *before* GC (ToDo: check all threads)
#if defined(GRAN)
// ToDo!: check sanity IF_DEBUG(sanity, checkTSOsSanity());
#endif
IF_DEBUG(sanity, checkFreeListSanity());
/* Initialise the static object lists
*/
static_objects = END_OF_STATIC_LIST;
scavenged_static_objects = END_OF_STATIC_LIST;
/* Save the old to-space if we're doing a two-space collection
*/
if (RtsFlags.GcFlags.generations == 1) {
old_to_blocks = g0s0->to_blocks;
g0s0->to_blocks = NULL;
g0s0->n_to_blocks = 0;
}
/* Keep a count of how many new blocks we allocated during this GC
* (used for resizing the allocation area, later).
*/
new_blocks = 0;
// Initialise to-space in all the generations/steps that we're
// collecting.
//
for (g = 0; g <= N; g++) {
// throw away the mutable list. Invariant: the mutable list
// always has at least one block; this means we can avoid a check for
// NULL in recordMutable().
if (g != 0) {
freeChain(generations[g].mut_list);
generations[g].mut_list = allocBlock();
}
for (s = 0; s < generations[g].n_steps; s++) {
// generation 0, step 0 doesn't need to-space
if (g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1) {
continue;
}
stp = &generations[g].steps[s];
ASSERT(stp->gen_no == g);
// start a new to-space for this step.
stp->hp = NULL;
stp->hp_bd = NULL;
stp->to_blocks = NULL;
// allocate the first to-space block; extra blocks will be
// chained on as necessary.
bd = gc_alloc_block(stp);
stp->to_blocks = bd;
stp->scan = bd->start;
stp->scan_bd = bd;
// initialise the large object queues.
stp->new_large_objects = NULL;
stp->scavenged_large_objects = NULL;
stp->n_scavenged_large_blocks = 0;
// mark the large objects as not evacuated yet
for (bd = stp->large_objects; bd; bd = bd->link) {
bd->flags &= ~BF_EVACUATED;
}
// for a compacted step, we need to allocate the bitmap
if (stp->is_compacted) {
nat bitmap_size; // in bytes
bdescr *bitmap_bdescr;
StgWord *bitmap;
bitmap_size = stp->n_blocks * BLOCK_SIZE / (sizeof(W_)*BITS_PER_BYTE);
if (bitmap_size > 0) {
bitmap_bdescr = allocGroup((lnat)BLOCK_ROUND_UP(bitmap_size)
/ BLOCK_SIZE);
stp->bitmap = bitmap_bdescr;
bitmap = bitmap_bdescr->start;
IF_DEBUG(gc, debugBelch("bitmap_size: %d, bitmap: %p",
bitmap_size, bitmap););
// don't forget to fill it with zeros!
memset(bitmap, 0, bitmap_size);
// For each block in this step, point to its bitmap from the
// block descriptor.
for (bd=stp->blocks; bd != NULL; bd = bd->link) {
bd->u.bitmap = bitmap;
bitmap += BLOCK_SIZE_W / (sizeof(W_)*BITS_PER_BYTE);
// Also at this point we set the BF_COMPACTED flag
// for this block. The invariant is that
// BF_COMPACTED is always unset, except during GC
// when it is set on those blocks which will be
// compacted.
bd->flags |= BF_COMPACTED;
}
}
}
}
}
/* make sure the older generations have at least one block to
* allocate into (this makes things easier for copy(), see below).
*/
for (g = N+1; g < RtsFlags.GcFlags.generations; g++) {
for (s = 0; s < generations[g].n_steps; s++) {
stp = &generations[g].steps[s];
if (stp->hp_bd == NULL) {
ASSERT(stp->blocks == NULL);
bd = gc_alloc_block(stp);
stp->blocks = bd;
stp->n_blocks = 1;
}
/* Set the scan pointer for older generations: remember we
* still have to scavenge objects that have been promoted. */
stp->scan = stp->hp;
stp->scan_bd = stp->hp_bd;
stp->to_blocks = NULL;
stp->n_to_blocks = 0;
stp->new_large_objects = NULL;
stp->scavenged_large_objects = NULL;
stp->n_scavenged_large_blocks = 0;
}
}
/* Allocate a mark stack if we're doing a major collection.
*/
if (major_gc) {
mark_stack_bdescr = allocGroup(MARK_STACK_BLOCKS);
mark_stack = (StgPtr *)mark_stack_bdescr->start;
mark_sp = mark_stack;
mark_splim = mark_stack + (MARK_STACK_BLOCKS * BLOCK_SIZE_W);
} else {
mark_stack_bdescr = NULL;
}
/* -----------------------------------------------------------------------
* follow all the roots that we know about:
* - mutable lists from each generation > N
* we want to *scavenge* these roots, not evacuate them: they're not
* going to move in this GC.
* Also: do them in reverse generation order. This is because we
* often want to promote objects that are pointed to by older
* generations early, so we don't have to repeatedly copy them.
* Doing the generations in reverse order ensures that we don't end
* up in the situation where we want to evac an object to gen 3 and
* it has already been evaced to gen 2.
*/
{
int st;
for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
generations[g].saved_mut_list = generations[g].mut_list;
generations[g].mut_list = allocBlock();
// mut_list always has at least one block.
}
for (g = RtsFlags.GcFlags.generations-1; g > N; g--) {
IF_PAR_DEBUG(verbose, printMutableList(&generations[g]));
scavenge_mutable_list(&generations[g]);
evac_gen = g;
for (st = generations[g].n_steps-1; st >= 0; st--) {
scavenge(&generations[g].steps[st]);
}
}
}
/* follow roots from the CAF list (used by GHCi)
*/
evac_gen = 0;
markCAFs(mark_root);
/* follow all the roots that the application knows about.
*/
evac_gen = 0;
get_roots(mark_root);
#if defined(PAR)
/* And don't forget to mark the TSO if we got here direct from
* Haskell! */
/* Not needed in a seq version?
if (CurrentTSO) {
CurrentTSO = (StgTSO *)MarkRoot((StgClosure *)CurrentTSO);
}
*/
// Mark the entries in the GALA table of the parallel system
markLocalGAs(major_gc);
// Mark all entries on the list of pending fetches
markPendingFetches(major_gc);
#endif
/* Mark the weak pointer list, and prepare to detect dead weak
* pointers.
*/
mark_weak_ptr_list(&weak_ptr_list);
old_weak_ptr_list = weak_ptr_list;
weak_ptr_list = NULL;
weak_stage = WeakPtrs;
/* The all_threads list is like the weak_ptr_list.
* See traverse_weak_ptr_list() for the details.
*/
old_all_threads = all_threads;
all_threads = END_TSO_QUEUE;
resurrected_threads = END_TSO_QUEUE;
/* Mark the stable pointer table.
*/
markStablePtrTable(mark_root);
/* -------------------------------------------------------------------------
* Repeatedly scavenge all the areas we know about until there's no
* more scavenging to be done.
*/
{
rtsBool flag;
loop:
flag = rtsFalse;
// scavenge static objects
if (major_gc && static_objects != END_OF_STATIC_LIST) {
IF_DEBUG(sanity, checkStaticObjects(static_objects));
scavenge_static();
}
/* When scavenging the older generations: Objects may have been
* evacuated from generations <= N into older generations, and we
* need to scavenge these objects. We're going to try to ensure that
* any evacuations that occur move the objects into at least the
* same generation as the object being scavenged, otherwise we
* have to create new entries on the mutable list for the older
* generation.
*/
// scavenge each step in generations 0..maxgen
{
long gen;
int st;
loop2:
// scavenge objects in compacted generation
if (mark_stack_overflowed || oldgen_scan_bd != NULL ||
(mark_stack_bdescr != NULL && !mark_stack_empty())) {
scavenge_mark_stack();
flag = rtsTrue;
}
for (gen = RtsFlags.GcFlags.generations; --gen >= 0; ) {
for (st = generations[gen].n_steps; --st >= 0; ) {
if (gen == 0 && st == 0 && RtsFlags.GcFlags.generations > 1) {
continue;
}
stp = &generations[gen].steps[st];
evac_gen = gen;
if (stp->hp_bd != stp->scan_bd || stp->scan < stp->hp) {
scavenge(stp);
flag = rtsTrue;
goto loop2;
}
if (stp->new_large_objects != NULL) {
scavenge_large(stp);
flag = rtsTrue;
goto loop2;
}
}
}
}
if (flag) { goto loop; }
// must be last... invariant is that everything is fully
// scavenged at this point.
if (traverse_weak_ptr_list()) { // returns rtsTrue if evaced something
goto loop;
}
}
/* Update the pointers from the "main thread" list - these are
* treated as weak pointers because we want to allow a main thread
* to get a BlockedOnDeadMVar exception in the same way as any other
* thread. Note that the threads should all have been retained by
* GC by virtue of being on the all_threads list, we're just
* updating pointers here.
*/
{
StgMainThread *m;
StgTSO *tso;
for (m = main_threads; m != NULL; m = m->link) {
tso = (StgTSO *) isAlive((StgClosure *)m->tso);
if (tso == NULL) {
barf("main thread has been GC'd");
}
m->tso = tso;
}
}
#if defined(PAR)
// Reconstruct the Global Address tables used in GUM
rebuildGAtables(major_gc);
IF_DEBUG(sanity, checkLAGAtable(rtsTrue/*check closures, too*/));
#endif
// Now see which stable names are still alive.
gcStablePtrTable();
// Tidy the end of the to-space chains
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
for (s = 0; s < generations[g].n_steps; s++) {
stp = &generations[g].steps[s];
if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
ASSERT(Bdescr(stp->hp) == stp->hp_bd);
stp->hp_bd->free = stp->hp;
}
}
}
#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)
LdvCensusForDead(N);
#endif
// NO MORE EVACUATION AFTER THIS POINT!
// Finally: compaction of the oldest generation.
if (major_gc && oldest_gen->steps[0].is_compacted) {
// save number of blocks for stats
oldgen_saved_blocks = oldest_gen->steps[0].n_blocks;
compact(get_roots);
}
IF_DEBUG(sanity, checkGlobalTSOList(rtsFalse));
/* run through all the generations/steps and tidy up
*/
copied = new_blocks * BLOCK_SIZE_W;
for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
if (g <= N) {
generations[g].collections++; // for stats
}
// Count the mutable list as bytes "copied" for the purposes of
// stats. Every mutable list is copied during every GC.
if (g > 0) {
for (bd = generations[g].mut_list; bd != NULL; bd = bd->link) {
copied += (bd->free - bd->start) * sizeof(StgWord);
}
}
for (s = 0; s < generations[g].n_steps; s++) {
bdescr *next;
stp = &generations[g].steps[s];
if (!(g == 0 && s == 0 && RtsFlags.GcFlags.generations > 1)) {
// stats information: how much we copied
if (g <= N) {
copied -= stp->hp_bd->start + BLOCK_SIZE_W -
stp->hp_bd->free;
}
}
// for generations we collected...
if (g <= N) {
// rough calculation of garbage collected, for stats output
if (stp->is_compacted) {
collected += (oldgen_saved_blocks - stp->n_blocks) * BLOCK_SIZE_W;
} else {
if (g == 0 && s == 0) {
collected += countNurseryBlocks() * BLOCK_SIZE_W;
collected += alloc_blocks;
} else {
collected += stp->n_blocks * BLOCK_SIZE_W;
}
}
/* 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 (!(g == 0 && s == 0)) {
if (stp->is_compacted) {
// for a compacted step, just shift the new to-space
// onto the front of the now-compacted existing blocks.
for (bd = stp->to_blocks; bd != NULL; bd = bd->link) {
bd->flags &= ~BF_EVACUATED; // now from-space
}
// tack the new blocks on the end of the existing blocks
if (stp->blocks == NULL) {
stp->blocks = stp->to_blocks;
} else {
for (bd = stp->blocks; bd != NULL; bd = next) {
next = bd->link;
if (next == NULL) {
bd->link = stp->to_blocks;
}
// NB. this step might not be compacted next
// time, so reset the BF_COMPACTED flags.
// They are set before GC if we're going to
// compact. (search for BF_COMPACTED above).
bd->flags &= ~BF_COMPACTED;
}
}
// add the new blocks to the block tally
stp->n_blocks += stp->n_to_blocks;
} else {
freeChain(stp->blocks);
stp->blocks = stp->to_blocks;
stp->n_blocks = stp->n_to_blocks;
for (bd = stp->blocks; bd != NULL; bd = bd->link) {
bd->flags &= ~BF_EVACUATED; // now from-space
}
}
stp->to_blocks = NULL;
stp->n_to_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
* large_objects list are therefore dead, so we free them here.
*/
for (bd = stp->large_objects; bd != NULL; bd = next) {
next = bd->link;
freeGroup(bd);
bd = next;
}
// update the count of blocks used by large objects
for (bd = stp->scavenged_large_objects; bd != NULL; bd = bd->link) {
bd->flags &= ~BF_EVACUATED;
}
stp->large_objects = stp->scavenged_large_objects;
stp->n_large_blocks = stp->n_scavenged_large_blocks;
} else {
// for older generations...
/* 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 = stp->scavenged_large_objects; bd; bd = next) {
next = bd->link;
bd->flags &= ~BF_EVACUATED;
dbl_link_onto(bd, &stp->large_objects);
}
// add the new blocks we promoted during this GC
stp->n_blocks += stp->n_to_blocks;
stp->n_to_blocks = 0;
stp->n_large_blocks += stp->n_scavenged_large_blocks;
}
}
}
/* Reset the sizes of the older generations when we do a major
* collection.
*
* CURRENT STRATEGY: make all generations except zero the same size.
* We have to stay within the maximum heap size, and leave a certain
* percentage of the maximum heap size available to allocate into.
*/
if (major_gc && RtsFlags.GcFlags.generations > 1) {
nat live, size, min_alloc;
nat max = RtsFlags.GcFlags.maxHeapSize;
nat gens = RtsFlags.GcFlags.generations;
// live in the oldest generations
live = oldest_gen->steps[0].n_blocks +
oldest_gen->steps[0].n_large_blocks;
// default max size for all generations except zero
size = stg_max(live * RtsFlags.GcFlags.oldGenFactor,
RtsFlags.GcFlags.minOldGenSize);
// minimum size for generation zero
min_alloc = stg_max((RtsFlags.GcFlags.pcFreeHeap * max) / 200,
RtsFlags.GcFlags.minAllocAreaSize);
// Auto-enable compaction when the residency reaches a
// certain percentage of the maximum heap size (default: 30%).
if (RtsFlags.GcFlags.generations > 1 &&
(RtsFlags.GcFlags.compact ||
(max > 0 &&
oldest_gen->steps[0].n_blocks >
(RtsFlags.GcFlags.compactThreshold * max) / 100))) {
oldest_gen->steps[0].is_compacted = 1;
// debugBelch("compaction: on\n", live);
} else {
oldest_gen->steps[0].is_compacted = 0;
// debugBelch("compaction: off\n", live);
}
// if we're going to go over the maximum heap size, reduce the
// size of the generations accordingly. The calculation is
// different if compaction is turned on, because we don't need
// to double the space required to collect the old generation.
if (max != 0) {
// this test is necessary to ensure that the calculations
// below don't have any negative results - we're working
// with unsigned values here.
if (max < min_alloc) {
heapOverflow();
}
if (oldest_gen->steps[0].is_compacted) {
if ( (size + (size - 1) * (gens - 2) * 2) + min_alloc > max ) {
size = (max - min_alloc) / ((gens - 1) * 2 - 1);
}
} else {
if ( (size * (gens - 1) * 2) + min_alloc > max ) {
size = (max - min_alloc) / ((gens - 1) * 2);
}
}
if (size < live) {
heapOverflow();
}
}
#if 0
debugBelch("live: %d, min_alloc: %d, size : %d, max = %d\n", live,
min_alloc, size, max);
#endif
for (g = 0; g < gens; g++) {
generations[g].max_blocks = size;
}
}
// Guess the amount of live data for stats.
live = calcLive();
/* Free the small objects allocated via allocate(), since this will
* all have been copied into G0S1 now.
*/
if (small_alloc_list != NULL) {
freeChain(small_alloc_list);
}
small_alloc_list = NULL;
alloc_blocks = 0;
alloc_Hp = NULL;
alloc_HpLim = NULL;
alloc_blocks_lim = RtsFlags.GcFlags.minAllocAreaSize;
// Start a new pinned_object_block
pinned_object_block = NULL;
/* Free the mark stack.
*/
if (mark_stack_bdescr != NULL) {
freeGroup(mark_stack_bdescr);
}
/* Free any bitmaps.
*/
for (g = 0; g <= N; g++) {
for (s = 0; s < generations[g].n_steps; s++) {
stp = &generations[g].steps[s];
if (stp->is_compacted && stp->bitmap != NULL) {
freeGroup(stp->bitmap);
}
}
}
/* Two-space collector:
* Free the old to-space, and estimate the amount of live data.
*/
if (RtsFlags.GcFlags.generations == 1) {
nat blocks;
if (old_to_blocks != NULL) {
freeChain(old_to_blocks);
}
for (bd = g0s0->to_blocks; bd != NULL; bd = bd->link) {
bd->flags = 0; // now from-space
}
/* For a two-space collector, we need to resize the nursery. */
/* set up a new nursery. Allocate a nursery size based on a
* function of the amount of live data (by default a factor of 2)
* Use the blocks from the old nursery if possible, freeing up any
* left over blocks.
*
* If we get near the maximum heap size, then adjust our nursery
* size accordingly. If the nursery is the same size as the live
* data (L), then we need 3L bytes. We can reduce the size of the
* nursery to bring the required memory down near 2L bytes.
*
* A normal 2-space collector would need 4L bytes to give the same
* performance we get from 3L bytes, reducing to the same
* performance at 2L bytes.
*/
blocks = g0s0->n_to_blocks;
if ( RtsFlags.GcFlags.maxHeapSize != 0 &&
blocks * RtsFlags.GcFlags.oldGenFactor * 2 >
RtsFlags.GcFlags.maxHeapSize ) {
long adjusted_blocks; // signed on purpose
int pc_free;
adjusted_blocks = (RtsFlags.GcFlags.maxHeapSize - 2 * blocks);
IF_DEBUG(gc, debugBelch("@@ Near maximum heap size of 0x%x blocks, blocks = %d, adjusted to %ld", RtsFlags.GcFlags.maxHeapSize, blocks, adjusted_blocks));
pc_free = adjusted_blocks * 100 / RtsFlags.GcFlags.maxHeapSize;
if (pc_free < RtsFlags.GcFlags.pcFreeHeap) /* might even be < 0 */ {
heapOverflow();
}
blocks = adjusted_blocks;
} else {
blocks *= RtsFlags.GcFlags.oldGenFactor;
if (blocks < RtsFlags.GcFlags.minAllocAreaSize) {
blocks = RtsFlags.GcFlags.minAllocAreaSize;
}
}
resizeNurseries(blocks);
} else {
/* Generational collector:
* If the user has given us a suggested heap size, adjust our
* allocation area to make best use of the memory available.
*/
if (RtsFlags.GcFlags.heapSizeSuggestion) {