summaryrefslogtreecommitdiff
path: root/rts/Threads.c
blob: 07d0d0a180e5967eea3af20727ac6ab2bca3999b (plain)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
/* ---------------------------------------------------------------------------
 *
 * (c) The GHC Team, 2006
 *
 * Thread-related functionality
 *
 * --------------------------------------------------------------------------*/

#include "rts/PosixSource.h"
#include "Rts.h"

#include "Capability.h"
#include "Updates.h"
#include "Threads.h"
#include "STM.h"
#include "Schedule.h"
#include "Trace.h"
#include "ThreadLabels.h"
#include "Updates.h"
#include "Messages.h"
#include "RaiseAsync.h"
#include "Prelude.h"
#include "Printer.h"
#include "sm/Sanity.h"
#include "sm/Storage.h"

#include <string.h>

/* Next thread ID to allocate.
 * LOCK: sched_mutex
 */
static StgThreadID next_thread_id = 1;

/* The smallest stack size that makes any sense is:
 *    RESERVED_STACK_WORDS    (so we can get back from the stack overflow)
 *  + sizeofW(StgStopFrame)   (the stg_stop_thread_info frame)
 *  + 1                       (the closure to enter)
 *  + 1                       (stg_ap_v_ret)
 *  + 1                       (spare slot req'd by stg_ap_v_ret)
 *
 * A thread with this stack will bomb immediately with a stack
 * overflow, which will increase its stack size.
 */
#define MIN_STACK_WORDS (RESERVED_STACK_WORDS + sizeofW(StgStopFrame) + 3)

/* ---------------------------------------------------------------------------
   Create a new thread.

   The new thread starts with the given stack size.  Before the
   scheduler can run, however, this thread needs to have a closure
   (and possibly some arguments) pushed on its stack.  See
   pushClosure() in Schedule.h.

   createGenThread() and createIOThread() (in SchedAPI.h) are
   convenient packaged versions of this function.
   ------------------------------------------------------------------------ */
StgTSO *
createThread(Capability *cap, W_ size)
{
    StgTSO *tso;
    StgStack *stack;
    uint32_t stack_size;

    /* sched_mutex is *not* required */

    /* catch ridiculously small stack sizes */
    if (size < MIN_STACK_WORDS + sizeofW(StgStack) + sizeofW(StgTSO)) {
        size = MIN_STACK_WORDS + sizeofW(StgStack) + sizeofW(StgTSO);
    }

    /* The size argument we are given includes all the per-thread
     * overheads:
     *
     *    - The TSO structure
     *    - The STACK header
     *
     * This is so that we can use a nice round power of 2 for the
     * default stack size (e.g. 1k), and if we're allocating lots of
     * threads back-to-back they'll fit nicely in a block.  It's a bit
     * of a benchmark hack, but it doesn't do any harm.
     */
    stack_size = round_to_mblocks(size - sizeofW(StgTSO));
    stack = (StgStack *)allocate(cap, stack_size);
    TICK_ALLOC_STACK(stack_size);
    SET_HDR(stack, &stg_STACK_info, cap->r.rCCCS);
    stack->stack_size   = stack_size - sizeofW(StgStack);
    stack->sp           = stack->stack + stack->stack_size;
    stack->dirty        = STACK_DIRTY;
    stack->marking      = 0;

    tso = (StgTSO *)allocate(cap, sizeofW(StgTSO));
    TICK_ALLOC_TSO();
    SET_HDR(tso, &stg_TSO_info, CCS_SYSTEM);

    // Always start with the compiled code evaluator
    tso->what_next = ThreadRunGHC;
    tso->why_blocked  = NotBlocked;
    tso->block_info.closure = (StgClosure *)END_TSO_QUEUE;
    tso->blocked_exceptions = END_BLOCKED_EXCEPTIONS_QUEUE;
    tso->bq = (StgBlockingQueue *)END_TSO_QUEUE;
    tso->flags = 0;
    tso->dirty = 1;
    tso->_link = END_TSO_QUEUE;

    tso->saved_errno = 0;
    tso->bound = NULL;
    tso->cap = cap;

    tso->stackobj       = stack;
    tso->tot_stack_size = stack->stack_size;

    ASSIGN_Int64((W_*)&(tso->alloc_limit), 0);

    tso->trec = NO_TREC;
    tso->label = NULL;

#if defined(PROFILING)
    tso->prof.cccs = CCS_MAIN;
#endif

    // put a stop frame on the stack
    stack->sp -= sizeofW(StgStopFrame);
    SET_HDR((StgClosure*)stack->sp,
            (StgInfoTable *)&stg_stop_thread_info,CCS_SYSTEM);

    /* Link the new thread on the global thread list.
     */
    ACQUIRE_LOCK(&sched_mutex);
    tso->id = next_thread_id++;  // while we have the mutex
    tso->global_link = g0->threads;
    /* Mutations above need no memory barrier since this lock will provide
     * a release barrier */
    g0->threads = tso;
    RELEASE_LOCK(&sched_mutex);

    // ToDo: report the stack size in the event?
    traceEventCreateThread(cap, tso);

    return tso;
}

/* ---------------------------------------------------------------------------
 * Equality on Thread ids.
 *
 * This is used from STG land in the implementation of the Eq instance
 * for ThreadIds.
 * ------------------------------------------------------------------------ */

bool
eq_thread(StgPtr tso1, StgPtr tso2)
{
  return tso1 == tso2;
}

/* ---------------------------------------------------------------------------
 * Comparing Thread ids.
 *
 * This is used from STG land in the implementation of the Ord instance
 * for ThreadIds.
 * ------------------------------------------------------------------------ */

int
cmp_thread(StgPtr tso1, StgPtr tso2)
{
  if (tso1 == tso2) return 0;

  StgThreadID id1 = ((StgTSO *)tso1)->id;
  StgThreadID id2 = ((StgTSO *)tso2)->id;

  ASSERT(id1 != id2);

  return id1 < id2 ? -1 : 1;
}

/* ---------------------------------------------------------------------------
 * Fetching the ThreadID from an StgTSO.
 *
 * This is used in the implementation of Show for ThreadIds.
 * ------------------------------------------------------------------------ */
StgThreadID
rts_getThreadId(StgPtr tso)
{
  return ((StgTSO *)tso)->id;
}

/* ---------------------------------------------------------------------------
 * Enabling and disabling the thread allocation limit
 * ------------------------------------------------------------------------ */

void rts_enableThreadAllocationLimit(StgPtr tso)
{
    ((StgTSO *)tso)->flags |= TSO_ALLOC_LIMIT;
}

void rts_disableThreadAllocationLimit(StgPtr tso)
{
    ((StgTSO *)tso)->flags &= ~TSO_ALLOC_LIMIT;
}

/* -----------------------------------------------------------------------------
   Remove a thread from a queue.
   Fails fatally if the TSO is not on the queue.
   -------------------------------------------------------------------------- */

bool // returns true if we modified queue
removeThreadFromQueue (Capability *cap, StgTSO **queue, StgTSO *tso)
{
    StgTSO *t, *prev;

    prev = NULL;
    for (t = *queue; t != END_TSO_QUEUE; prev = t, t = t->_link) {
        if (t == tso) {
            if (prev) {
                setTSOLink(cap,prev,t->_link);
                t->_link = END_TSO_QUEUE;
                return false;
            } else {
                *queue = t->_link;
                t->_link = END_TSO_QUEUE;
                return true;
            }
        }
    }
    barf("removeThreadFromQueue: not found");
}

bool // returns true if we modified head or tail
removeThreadFromDeQueue (Capability *cap,
                         StgTSO **head, StgTSO **tail, StgTSO *tso)
{
    StgTSO *t, *prev;
    bool flag = false;

    prev = NULL;
    for (t = *head; t != END_TSO_QUEUE; prev = t, t = t->_link) {
        if (t == tso) {
            if (prev) {
                setTSOLink(cap,prev,t->_link);
                flag = false;
            } else {
                *head = t->_link;
                flag = true;
            }
            t->_link = END_TSO_QUEUE;
            if (*tail == tso) {
                if (prev) {
                    *tail = prev;
                } else {
                    *tail = END_TSO_QUEUE;
                }
                return true;
            } else {
                return flag;
            }
        }
    }
    barf("removeThreadFromDeQueue: not found");
}

/* ----------------------------------------------------------------------------
   tryWakeupThread()

   Attempt to wake up a thread.  tryWakeupThread is idempotent: it is
   always safe to call it too many times, but it is not safe in
   general to omit a call.

   ------------------------------------------------------------------------- */

void
tryWakeupThread (Capability *cap, StgTSO *tso)
{
    traceEventThreadWakeup (cap, tso, tso->cap->no);

#if defined(THREADED_RTS)
    if (tso->cap != cap)
    {
        MessageWakeup *msg;
        msg = (MessageWakeup *)allocate(cap,sizeofW(MessageWakeup));
        msg->tso = tso;
        SET_HDR(msg, &stg_MSG_TRY_WAKEUP_info, CCS_SYSTEM);
        sendMessage(cap, tso->cap, (Message*)msg);
        debugTraceCap(DEBUG_sched, cap, "message: try wakeup thread %"
                      FMT_StgThreadID " on cap %d", tso->id, tso->cap->no);
        return;
    }
#endif

    switch (tso->why_blocked)
    {
    case BlockedOnMVar:
    case BlockedOnMVarRead:
    {
        if (tso->_link == END_TSO_QUEUE) {
            tso->block_info.closure = (StgClosure*)END_TSO_QUEUE;
            goto unblock;
        } else {
            return;
        }
    }

    case BlockedOnMsgThrowTo:
    {
        const StgInfoTable *i;

        i = lockClosure(tso->block_info.closure);
        unlockClosure(tso->block_info.closure, i);
        if (i != &stg_MSG_NULL_info) {
            debugTraceCap(DEBUG_sched, cap, "thread %" FMT_StgThreadID " still "
                          "blocked on throwto (%p)", tso->id,
                          tso->block_info.throwto->header.info);
            return;
        }

        // remove the block frame from the stack
        ASSERT(tso->stackobj->sp[0] == (StgWord)&stg_block_throwto_info);
        tso->stackobj->sp += 3;
        goto unblock;
    }

    case BlockedOnSTM:
        tso->block_info.closure = &stg_STM_AWOKEN_closure;
        goto unblock;

    case BlockedOnBlackHole:
    case ThreadMigrating:
        goto unblock;

    default:
        // otherwise, do nothing
        return;
    }

unblock:
    // just run the thread now, if the BH is not really available,
    // we'll block again.
    tso->why_blocked = NotBlocked;
    appendToRunQueue(cap,tso);

    // We used to set the context switch flag here, which would
    // trigger a context switch a short time in the future (at the end
    // of the current nursery block).  The idea is that we have just
    // woken up a thread, so we may need to load-balance and migrate
    // threads to other CPUs.  On the other hand, setting the context
    // switch flag here unfairly penalises the current thread by
    // yielding its time slice too early.
    //
    // The synthetic benchmark nofib/smp/chan can be used to show the
    // difference quite clearly.

    // cap->context_switch = 1;
}

/* ----------------------------------------------------------------------------
   migrateThread
   ------------------------------------------------------------------------- */

// Precondition: The caller must own the `from` capability.
void
migrateThread (Capability *from, StgTSO *tso, Capability *to)
{
    // Sadly we can't assert this since migrateThread is called from
    // scheduleDoGC, where we implicitly own all capabilities.
    //ASSERT_FULL_CAPABILITY_INVARIANTS(from, getTask());

    traceEventMigrateThread (from, tso, to->no);
    // ThreadMigrating tells the target cap that it needs to be added to
    // the run queue when it receives the MSG_TRY_WAKEUP.
    tso->why_blocked = ThreadMigrating;
    tso->cap = to;
    tryWakeupThread(from, tso);
}

/* ----------------------------------------------------------------------------
   awakenBlockedQueue

   wakes up all the threads on the specified queue.
   ------------------------------------------------------------------------- */

static void
wakeBlockingQueue(Capability *cap, StgBlockingQueue *bq)
{
    MessageBlackHole *msg;
    const StgInfoTable *i;

    ASSERT(bq->header.info == &stg_BLOCKING_QUEUE_DIRTY_info  ||
           bq->header.info == &stg_BLOCKING_QUEUE_CLEAN_info  );

    for (msg = bq->queue; msg != (MessageBlackHole*)END_TSO_QUEUE;
         msg = msg->link) {
        i = ACQUIRE_LOAD(&msg->header.info);
        if (i != &stg_IND_info) {
            ASSERT(i == &stg_MSG_BLACKHOLE_info);
            tryWakeupThread(cap,msg->tso);
        }
    }

    // overwrite the BQ with an indirection so it will be
    // collected at the next GC.
    OVERWRITE_INFO(bq, &stg_IND_info);
}

// If we update a closure that we know we BLACKHOLE'd, and the closure
// no longer points to the current TSO as its owner, then there may be
// an orphaned BLOCKING_QUEUE closure with blocked threads attached to
// it.  We therefore traverse the BLOCKING_QUEUEs attached to the
// current TSO to see if any can now be woken up.
void
checkBlockingQueues (Capability *cap, StgTSO *tso)
{
    StgBlockingQueue *bq, *next;
    StgClosure *p;

    debugTraceCap(DEBUG_sched, cap, "collision occurred; checking blocking "
                  "queues for thread %" FMT_StgThreadID, tso->id);

    for (bq = tso->bq; bq != (StgBlockingQueue*)END_TSO_QUEUE; bq = next) {
        next = bq->link;

        const StgInfoTable *bqinfo = ACQUIRE_LOAD(&bq->header.info);
        if (bqinfo == &stg_IND_info) {
            // ToDo: could short it out right here, to avoid
            // traversing this IND multiple times.
            continue;
        }

        // We need to always ensure we untag bh.  While it might seem a
        // sensible assumption that bh will never be tagged, the GC could
        // shortcut the indirection and put a tagged pointer into the
        // indirection.
        //
        // This blew up on aarch64-darwin with misaligned access.  bh pointing
        // to an address that always ended in 0xa.  Thus on architectures that
        // are a little less strict about alignment, this would have read a
        // garbage pinfo, which very, very unlikely would have been equal to
        // stg_BLACKHOLE_info.  Thus while the code would have done the wrong
        // thing the result would be the same in almost all cases. See #20093.
        p = UNTAG_CLOSURE(bq->bh);
        const StgInfoTable *pinfo = ACQUIRE_LOAD(&p->header.info);
        if (pinfo != &stg_BLACKHOLE_info ||
            ((StgInd *)p)->indirectee != (StgClosure*)bq)
        {
            wakeBlockingQueue(cap,bq);
        }
    }
}

/* ----------------------------------------------------------------------------
   updateThunk

   Update a thunk with a value.  In order to do this, we need to know
   which TSO owns (or is evaluating) the thunk, in case we need to
   awaken any threads that are blocked on it.
   ------------------------------------------------------------------------- */

void
updateThunk (Capability *cap, StgTSO *tso, StgClosure *thunk, StgClosure *val)
{
    StgClosure *v;
    StgTSO *owner;
    const StgInfoTable *i;

    i = ACQUIRE_LOAD(&thunk->header.info);
    if (i != &stg_BLACKHOLE_info &&
        i != &stg_CAF_BLACKHOLE_info &&
        i != &__stg_EAGER_BLACKHOLE_info &&
        i != &stg_WHITEHOLE_info) {
        updateWithIndirection(cap, thunk, val);
        return;
    }

    v = UNTAG_CLOSURE(((StgInd*)thunk)->indirectee);

    updateWithIndirection(cap, thunk, val);

    // sometimes the TSO is locked when we reach here, so its header
    // might be WHITEHOLE.  Hence check for the correct owner using
    // pointer equality first.
    if ((StgTSO*)v == tso) {
        return;
    }

    i = ACQUIRE_LOAD(&v->header.info);
    if (i == &stg_TSO_info) {
        checkBlockingQueues(cap, tso);
        return;
    }

    if (i != &stg_BLOCKING_QUEUE_CLEAN_info &&
        i != &stg_BLOCKING_QUEUE_DIRTY_info) {
        checkBlockingQueues(cap, tso);
        return;
    }

    owner = ((StgBlockingQueue*)v)->owner;

    if (owner != tso) {
        checkBlockingQueues(cap, tso);
    } else {
        wakeBlockingQueue(cap, (StgBlockingQueue*)v);
    }
}

/* ---------------------------------------------------------------------------
 * rtsSupportsBoundThreads(): is the RTS built to support bound threads?
 * used by Control.Concurrent for error checking.
 * ------------------------------------------------------------------------- */

HsBool
rtsSupportsBoundThreads(void)
{
#if defined(THREADED_RTS)
  return HS_BOOL_TRUE;
#else
  return HS_BOOL_FALSE;
#endif
}

/* ---------------------------------------------------------------------------
 * isThreadBound(tso): check whether tso is bound to an OS thread.
 * ------------------------------------------------------------------------- */

StgBool
isThreadBound(StgTSO* tso USED_IF_THREADS)
{
#if defined(THREADED_RTS)
  return (tso->bound != NULL);
#endif
  return false;
}

/* -----------------------------------------------------------------------------
   Stack overflow

   If the thread has reached its maximum stack size, then raise the
   StackOverflow exception in the offending thread.  Otherwise
   relocate the TSO into a larger chunk of memory and adjust its stack
   size appropriately.
   -------------------------------------------------------------------------- */

void
threadStackOverflow (Capability *cap, StgTSO *tso)
{
    StgStack *new_stack, *old_stack;
    StgUnderflowFrame *frame;
    W_ chunk_size;

    IF_DEBUG(sanity,checkTSO(tso));

    if (RtsFlags.GcFlags.maxStkSize > 0
        && tso->tot_stack_size >= RtsFlags.GcFlags.maxStkSize) {
        // #3677: In a stack overflow situation, stack squeezing may
        // reduce the stack size, but we don't know whether it has been
        // reduced enough for the stack check to succeed if we try
        // again.  Fortunately stack squeezing is idempotent, so all we
        // need to do is record whether *any* squeezing happened.  If we
        // are at the stack's absolute -K limit, and stack squeezing
        // happened, then we try running the thread again.  The
        // TSO_SQUEEZED flag is set by threadPaused() to tell us whether
        // squeezing happened or not.
        if (tso->flags & TSO_SQUEEZED) {
            return;
        }

        debugTrace(DEBUG_gc,
                   "threadStackOverflow of TSO %" FMT_StgThreadID " (%p): stack"
                   " too large (now %ld; max is %ld)", tso->id, tso,
                   (long)tso->stackobj->stack_size, RtsFlags.GcFlags.maxStkSize);
        IF_DEBUG(gc,
                 /* If we're debugging, just print out the top of the stack */
                 printStackChunk(tso->stackobj->sp,
                                 stg_min(tso->stackobj->stack + tso->stackobj->stack_size,
                                         tso->stackobj->sp+64)));

        // See Note [Throw to self when masked], also #767 and #8303.
        throwToSelf(cap, tso, (StgClosure *)stackOverflow_closure);
        return;
    }


    // We also want to avoid enlarging the stack if squeezing has
    // already released some of it.  However, we don't want to get into
    // a pathological situation where a thread has a nearly full stack
    // (near its current limit, but not near the absolute -K limit),
    // keeps allocating a little bit, squeezing removes a little bit,
    // and then it runs again.  So to avoid this, if we squeezed *and*
    // there is still less than BLOCK_SIZE_W words free, then we enlarge
    // the stack anyway.
    //
    // NB: This reasoning only applies if the stack has been squeezed;
    // if no squeezing has occurred, then BLOCK_SIZE_W free space does
    // not mean there is enough stack to run; the thread may have
    // requested a large amount of stack (see below).  If the amount
    // we squeezed is not enough to run the thread, we'll come back
    // here (no squeezing will have occurred and thus we'll enlarge the
    // stack.)
    if ((tso->flags & TSO_SQUEEZED) &&
        ((W_)(tso->stackobj->sp - tso->stackobj->stack) >= BLOCK_SIZE_W)) {
        return;
    }

    old_stack = tso->stackobj;

    // If we used less than half of the previous stack chunk, then we
    // must have failed a stack check for a large amount of stack.  In
    // this case we allocate a double-sized chunk to try to
    // accommodate the large stack request.  If that also fails, the
    // next chunk will be 4x normal size, and so on.
    //
    // It would be better to have the mutator tell us how much stack
    // was needed, as we do with heap allocations, but this works for
    // now.
    //
    if (old_stack->sp > old_stack->stack + old_stack->stack_size / 2)
    {
        chunk_size = stg_max(2 * (old_stack->stack_size + sizeofW(StgStack)),
                             RtsFlags.GcFlags.stkChunkSize);
    }
    else
    {
        chunk_size = RtsFlags.GcFlags.stkChunkSize;
    }

    debugTraceCap(DEBUG_sched, cap,
                  "allocating new stack chunk of size %d bytes",
                  chunk_size * sizeof(W_));

    // Charge the current thread for allocating stack.  Stack usage is
    // non-deterministic, because the chunk boundaries might vary from
    // run to run, but accounting for this is better than not
    // accounting for it, since a deep recursion will otherwise not be
    // subject to allocation limits.
    cap->r.rCurrentTSO = tso;
    new_stack = (StgStack*) allocate(cap, chunk_size);
    cap->r.rCurrentTSO = NULL;

    SET_HDR(new_stack, &stg_STACK_info, old_stack->header.prof.ccs);
    TICK_ALLOC_STACK(chunk_size);

    new_stack->dirty = 0; // begin clean, we'll mark it dirty below
    new_stack->marking = 0;
    new_stack->stack_size = chunk_size - sizeofW(StgStack);
    new_stack->sp = new_stack->stack + new_stack->stack_size;

    tso->tot_stack_size += new_stack->stack_size;

    {
        StgWord *sp;
        W_ chunk_words, size;

        // find the boundary of the chunk of old stack we're going to
        // copy to the new stack.  We skip over stack frames until we
        // reach the smaller of
        //
        //   * the chunk buffer size (+RTS -kb)
        //   * the end of the old stack
        //
        for (sp = old_stack->sp;
             sp < stg_min(old_stack->sp + RtsFlags.GcFlags.stkChunkBufferSize,
                          old_stack->stack + old_stack->stack_size); )
        {
            size = stack_frame_sizeW((StgClosure*)sp);

            // if including this frame would exceed the size of the
            // new stack (taking into account the underflow frame),
            // then stop at the previous frame.
            if (sp + size > old_stack->sp + (new_stack->stack_size -
                                             sizeofW(StgUnderflowFrame))) {
                break;
            }
            sp += size;
        }

        if (sp == old_stack->stack + old_stack->stack_size) {
            //
            // the old stack chunk is now empty, so we do *not* insert
            // an underflow frame pointing back to it.  There are two
            // cases: either the old stack chunk was the last one, in
            // which case it ends with a STOP_FRAME, or it is not the
            // last one, and it already ends with an UNDERFLOW_FRAME
            // pointing to the previous chunk.  In the latter case, we
            // will copy the UNDERFLOW_FRAME into the new stack chunk.
            // In both cases, the old chunk will be subsequently GC'd.
            //
            // With the default settings, -ki1k -kb1k, this means the
            // first stack chunk will be discarded after the first
            // overflow, being replaced by a non-moving 32k chunk.
            //
        } else {
            new_stack->sp -= sizeofW(StgUnderflowFrame);
            frame = (StgUnderflowFrame*)new_stack->sp;
            frame->info = &stg_stack_underflow_frame_info;
            frame->next_chunk  = old_stack;
        }

        // copy the stack chunk between tso->sp and sp to
        //   new_tso->sp + (tso->sp - sp)
        chunk_words = sp - old_stack->sp;

        memcpy(/* dest   */ new_stack->sp - chunk_words,
               /* source */ old_stack->sp,
               /* size   */ chunk_words * sizeof(W_));

        old_stack->sp += chunk_words;
        new_stack->sp -= chunk_words;
    }

    // No write barriers needed; all of the writes above are to structured
    // owned by our capability.
    tso->stackobj = new_stack;

    // we're about to run it, better mark it dirty
    dirty_STACK(cap, new_stack);

    IF_DEBUG(sanity,checkTSO(tso));
    // IF_DEBUG(scheduler,printTSO(new_tso));
}



/* ---------------------------------------------------------------------------
   Stack underflow - called from the stg_stack_underflow_info frame
   ------------------------------------------------------------------------ */

W_ // returns offset to the return address
threadStackUnderflow (Capability *cap, StgTSO *tso)
{
    StgStack *new_stack, *old_stack;
    StgUnderflowFrame *frame;
    uint32_t retvals;

    debugTraceCap(DEBUG_sched, cap, "stack underflow");

    old_stack = tso->stackobj;

    frame = (StgUnderflowFrame*)(old_stack->stack + old_stack->stack_size
                                 - sizeofW(StgUnderflowFrame));
    ASSERT(frame->info == &stg_stack_underflow_frame_info);

    new_stack = (StgStack*)frame->next_chunk;
    tso->stackobj = new_stack;

    retvals = (P_)frame - old_stack->sp;
    if (retvals != 0)
    {
        // we have some return values to copy to the old stack
        if ((W_)(new_stack->sp - new_stack->stack) < retvals)
        {
            barf("threadStackUnderflow: not enough space for return values");
        }

        memcpy(/* dest */ new_stack->sp - retvals,
               /* src  */ old_stack->sp,
               /* size */ retvals * sizeof(W_));
    }

    // empty the old stack.  The GC may still visit this object
    // because it is on the mutable list.
    old_stack->sp = old_stack->stack + old_stack->stack_size;

    // restore the stack parameters, and update tot_stack_size
    tso->tot_stack_size -= old_stack->stack_size;

    // we're about to run it, better mark it dirty.
    //
    // N.B. the nonmoving collector may mark the stack, meaning that sp must
    // point at a valid stack frame.
    dirty_STACK(cap, new_stack);
    new_stack->sp -= retvals;

    return retvals;
}

/* ----------------------------------------------------------------------------
   Implementation of tryPutMVar#

   NOTE: this should be kept in sync with stg_tryPutMVarzh in PrimOps.cmm
   ------------------------------------------------------------------------- */

bool performTryPutMVar(Capability *cap, StgMVar *mvar, StgClosure *value)
{
    const StgInfoTable *info;
    const StgInfoTable *qinfo;
    StgMVarTSOQueue *q;
    StgTSO *tso;

    info = lockClosure((StgClosure*)mvar);

    if (mvar->value != &stg_END_TSO_QUEUE_closure) {
#if defined(THREADED_RTS)
        unlockClosure((StgClosure*)mvar, info);
#endif
        return false;
    }

    q = mvar->head;
loop:
    if (q == (StgMVarTSOQueue*)&stg_END_TSO_QUEUE_closure) {
        /* No further takes, the MVar is now full. */
        if (info == &stg_MVAR_CLEAN_info) {
            dirty_MVAR(&cap->r, (StgClosure*)mvar, mvar->value);
        }

        mvar->value = value;
        unlockClosure((StgClosure*)mvar, &stg_MVAR_DIRTY_info);
        return true;
    }

    qinfo = ACQUIRE_LOAD(&q->header.info);
    if (qinfo == &stg_IND_info ||
        qinfo == &stg_MSG_NULL_info) {
        q = (StgMVarTSOQueue*)((StgInd*)q)->indirectee;
        goto loop;
    }

    // There are takeMVar(s) waiting: wake up the first one
    tso = q->tso;
    mvar->head = q = q->link;
    if (q == (StgMVarTSOQueue*)&stg_END_TSO_QUEUE_closure) {
        mvar->tail = (StgMVarTSOQueue*)&stg_END_TSO_QUEUE_closure;
    } else {
        if (info == &stg_MVAR_CLEAN_info) {
            // Resolve #18919.
            dirty_MVAR(&cap->r, (StgClosure*)mvar, mvar->value);
            info = &stg_MVAR_DIRTY_info;
        }
    }

    ASSERT(tso->block_info.closure == (StgClosure*)mvar);
    // save why_blocked here, because waking up the thread destroys
    // this information
    StgWord why_blocked = RELAXED_LOAD(&tso->why_blocked);

    // actually perform the takeMVar
    StgStack* stack = tso->stackobj;
    RELAXED_STORE(&stack->sp[1], (W_)value);
    RELAXED_STORE(&stack->sp[0], (W_)&stg_ret_p_info);

    // indicate that the MVar operation has now completed.
    RELEASE_STORE(&tso->_link, (StgTSO*)&stg_END_TSO_QUEUE_closure);

    if ((stack->dirty & STACK_DIRTY) == 0) {
        dirty_STACK(cap, stack);
    }

    tryWakeupThread(cap, tso);

    // If it was a readMVar, then we can still do work,
    // so loop back. (XXX: This could take a while)
    if (why_blocked == BlockedOnMVarRead)
        goto loop;

    ASSERT(why_blocked == BlockedOnMVar);

    unlockClosure((StgClosure*)mvar, info);

    return true;
}

StgMutArrPtrs *listThreads(Capability *cap)
{
    ACQUIRE_LOCK(&sched_mutex);

    // First count how many threads we have...
    StgWord n_threads = 0;
    for (unsigned g = 0; g < RtsFlags.GcFlags.generations; g++) {
        for (StgTSO *t = generations[g].threads; t != END_TSO_QUEUE; t = t->global_link) {
            n_threads++;
        }
    }

    // Allocate a suitably-sized array...
    const StgWord size = n_threads + mutArrPtrsCardTableSize(n_threads);
    StgMutArrPtrs *arr =
        (StgMutArrPtrs *)allocate(cap, sizeofW(StgMutArrPtrs) + size);
    TICK_ALLOC_PRIM(sizeofW(StgMutArrPtrs), n, 0);
    arr->ptrs = n_threads;
    arr->size = size;

    // Populate it...
    StgWord i = 0;
    for (unsigned g = 0; g < RtsFlags.GcFlags.generations; g++) {
        for (StgTSO *t = generations[g].threads; t != END_TSO_QUEUE; t = t->global_link) {
            // It's possible that new threads have been created since we counted.
            // Ignore them.
            if (i == n_threads)
                break;
            arr->payload[i] = (StgClosure *) t;
            i++;
        }
    }
    CHECKM(i == n_threads, "listThreads: Found too few threads");
    RELEASE_LOCK(&sched_mutex);
    return arr;
}

/* ----------------------------------------------------------------------------
 * Debugging: why is a thread blocked
 * ------------------------------------------------------------------------- */

#if defined(DEBUG)
void
printThreadBlockage(StgTSO *tso)
{
  switch (tso->why_blocked) {
#if defined(mingw32_HOST_OS)
    case BlockedOnDoProc:
    debugBelch("is blocked on proc (request: %u)", tso->block_info.async_result->reqID);
    break;
#endif
#if !defined(THREADED_RTS)
  case BlockedOnRead:
    debugBelch("is blocked on read from fd %d", (int)(tso->block_info.fd));
    break;
  case BlockedOnWrite:
    debugBelch("is blocked on write to fd %d", (int)(tso->block_info.fd));
    break;
  case BlockedOnDelay:
    debugBelch("is blocked until %ld", (long)(tso->block_info.target));
    break;
#endif
    break;
  case BlockedOnMVar:
    debugBelch("is blocked on an MVar @ %p", tso->block_info.closure);
    break;
  case BlockedOnMVarRead:
    debugBelch("is blocked on atomic MVar read @ %p", tso->block_info.closure);
    break;
    break;
  case BlockedOnBlackHole:
      debugBelch("is blocked on a black hole %p",
                 ((StgBlockingQueue*)tso->block_info.bh->bh));
    break;
  case BlockedOnMsgThrowTo:
    debugBelch("is blocked on a throwto message");
    break;
  case NotBlocked:
    debugBelch("is not blocked");
    break;
  case ThreadMigrating:
    debugBelch("is runnable, but not on the run queue");
    break;
  case BlockedOnCCall:
    debugBelch("is blocked on an external call");
    break;
  case BlockedOnCCall_Interruptible:
    debugBelch("is blocked on an external call (but may be interrupted)");
    break;
  case BlockedOnSTM:
    debugBelch("is blocked on an STM operation");
    break;
  default:
    barf("printThreadBlockage: strange tso->why_blocked: %d for TSO %"
         FMT_StgThreadID " (%p)", tso->why_blocked, tso->id, tso);
  }
}


void
printThreadStatus(StgTSO *t)
{
    debugBelch("\tthread %4lu @ %p ", (unsigned long)t->id, (void *)t);
    if (t->label) {
        debugBelch("[\"%.*s\"] ", (int)t->label->bytes, (char *)t->label->payload);
    }
    switch (t->what_next) {
    case ThreadKilled:
        debugBelch("has been killed");
        break;
    case ThreadComplete:
        debugBelch("has completed");
        break;
    default:
        printThreadBlockage(t);
    }
    if (t->dirty) {
        debugBelch(" (TSO_DIRTY)");
    }
    debugBelch("\n");
}

void
printAllThreads(void)
{
  StgTSO *t, *next;
  uint32_t i, g;
  Capability *cap;

  debugBelch("all threads:\n");

  for (i = 0; i < n_capabilities; i++) {
      cap = capabilities[i];
      debugBelch("threads on capability %d:\n", cap->no);
      for (t = cap->run_queue_hd; t != END_TSO_QUEUE; t = t->_link) {
          printThreadStatus(t);
      }
  }

  debugBelch("other threads:\n");
  for (g = 0; g < RtsFlags.GcFlags.generations; g++) {
    for (t = generations[g].threads; t != END_TSO_QUEUE; t = next) {
      if (t->why_blocked != NotBlocked) {
          printThreadStatus(t);
      }
      next = t->global_link;
    }
  }
}

// useful from gdb
void
printThreadQueue(StgTSO *t)
{
    uint32_t i = 0;
    for (; t != END_TSO_QUEUE; t = t->_link) {
        printThreadStatus(t);
        i++;
    }
    debugBelch("%d threads on queue\n", i);
}

#endif /* DEBUG */