aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/core/lib/surface/completion_queue.cc
blob: c0ef8a8a1ca717710188b54c67c29f8781a0ee81 (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
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
/*
 *
 * Copyright 2015-2016 gRPC authors.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 */
#include <grpc/support/port_platform.h>

#include "src/core/lib/surface/completion_queue.h"

#include <inttypes.h>
#include <stdio.h>
#include <string.h>

#include <grpc/support/alloc.h>
#include <grpc/support/atm.h>
#include <grpc/support/log.h>
#include <grpc/support/string_util.h>
#include <grpc/support/time.h>

#include "src/core/lib/debug/stats.h"
#include "src/core/lib/iomgr/pollset.h"
#include "src/core/lib/iomgr/timer.h"
#include "src/core/lib/profiling/timers.h"
#include "src/core/lib/support/spinlock.h"
#include "src/core/lib/support/string.h"
#include "src/core/lib/surface/api_trace.h"
#include "src/core/lib/surface/call.h"
#include "src/core/lib/surface/event_string.h"

grpc_core::Tracer grpc_trace_operation_failures(false, "op_failure");
#ifndef NDEBUG
grpc_core::Tracer grpc_trace_pending_tags(false, "pending_tags");
grpc_core::Tracer grpc_trace_cq_refcount(false, "cq_refcount");
#endif

typedef struct {
  grpc_pollset_worker **worker;
  void *tag;
} plucker;

typedef struct {
  bool can_get_pollset;
  bool can_listen;
  size_t (*size)(void);
  void (*init)(grpc_pollset *pollset, gpr_mu **mu);
  grpc_error *(*kick)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                      grpc_pollset_worker *specific_worker);
  grpc_error *(*work)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                      grpc_pollset_worker **worker, grpc_millis deadline);
  void (*shutdown)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
                   grpc_closure *closure);
  void (*destroy)(grpc_exec_ctx *exec_ctx, grpc_pollset *pollset);
} cq_poller_vtable;

typedef struct non_polling_worker {
  gpr_cv cv;
  bool kicked;
  struct non_polling_worker *next;
  struct non_polling_worker *prev;
} non_polling_worker;

typedef struct {
  gpr_mu mu;
  non_polling_worker *root;
  grpc_closure *shutdown;
} non_polling_poller;

static size_t non_polling_poller_size(void) {
  return sizeof(non_polling_poller);
}

static void non_polling_poller_init(grpc_pollset *pollset, gpr_mu **mu) {
  non_polling_poller *npp = (non_polling_poller *)pollset;
  gpr_mu_init(&npp->mu);
  *mu = &npp->mu;
}

static void non_polling_poller_destroy(grpc_exec_ctx *exec_ctx,
                                       grpc_pollset *pollset) {
  non_polling_poller *npp = (non_polling_poller *)pollset;
  gpr_mu_destroy(&npp->mu);
}

static grpc_error *non_polling_poller_work(grpc_exec_ctx *exec_ctx,
                                           grpc_pollset *pollset,
                                           grpc_pollset_worker **worker,
                                           grpc_millis deadline) {
  non_polling_poller *npp = (non_polling_poller *)pollset;
  if (npp->shutdown) return GRPC_ERROR_NONE;
  non_polling_worker w;
  gpr_cv_init(&w.cv);
  if (worker != NULL) *worker = (grpc_pollset_worker *)&w;
  if (npp->root == NULL) {
    npp->root = w.next = w.prev = &w;
  } else {
    w.next = npp->root;
    w.prev = w.next->prev;
    w.next->prev = w.prev->next = &w;
  }
  w.kicked = false;
  gpr_timespec deadline_ts =
      grpc_millis_to_timespec(deadline, GPR_CLOCK_REALTIME);
  while (!npp->shutdown && !w.kicked &&
         !gpr_cv_wait(&w.cv, &npp->mu, deadline_ts))
    ;
  if (&w == npp->root) {
    npp->root = w.next;
    if (&w == npp->root) {
      if (npp->shutdown) {
        GRPC_CLOSURE_SCHED(exec_ctx, npp->shutdown, GRPC_ERROR_NONE);
      }
      npp->root = NULL;
    }
  }
  w.next->prev = w.prev;
  w.prev->next = w.next;
  gpr_cv_destroy(&w.cv);
  if (worker != NULL) *worker = NULL;
  return GRPC_ERROR_NONE;
}

static grpc_error *non_polling_poller_kick(
    grpc_exec_ctx *exec_ctx, grpc_pollset *pollset,
    grpc_pollset_worker *specific_worker) {
  non_polling_poller *p = (non_polling_poller *)pollset;
  if (specific_worker == NULL) specific_worker = (grpc_pollset_worker *)p->root;
  if (specific_worker != NULL) {
    non_polling_worker *w = (non_polling_worker *)specific_worker;
    if (!w->kicked) {
      w->kicked = true;
      gpr_cv_signal(&w->cv);
    }
  }
  return GRPC_ERROR_NONE;
}

static void non_polling_poller_shutdown(grpc_exec_ctx *exec_ctx,
                                        grpc_pollset *pollset,
                                        grpc_closure *closure) {
  non_polling_poller *p = (non_polling_poller *)pollset;
  GPR_ASSERT(closure != NULL);
  p->shutdown = closure;
  if (p->root == NULL) {
    GRPC_CLOSURE_SCHED(exec_ctx, closure, GRPC_ERROR_NONE);
  } else {
    non_polling_worker *w = p->root;
    do {
      gpr_cv_signal(&w->cv);
      w = w->next;
    } while (w != p->root);
  }
}

static const cq_poller_vtable g_poller_vtable_by_poller_type[] = {
    /* GRPC_CQ_DEFAULT_POLLING */
    {true, true, grpc_pollset_size, grpc_pollset_init, grpc_pollset_kick,
     grpc_pollset_work, grpc_pollset_shutdown, grpc_pollset_destroy},
    /* GRPC_CQ_NON_LISTENING */
    {true, false, grpc_pollset_size, grpc_pollset_init, grpc_pollset_kick,
     grpc_pollset_work, grpc_pollset_shutdown, grpc_pollset_destroy},
    /* GRPC_CQ_NON_POLLING */
    {false, false, non_polling_poller_size, non_polling_poller_init,
     non_polling_poller_kick, non_polling_poller_work,
     non_polling_poller_shutdown, non_polling_poller_destroy},
};

typedef struct cq_vtable {
  grpc_cq_completion_type cq_completion_type;
  size_t data_size;
  void (*init)(void *data);
  void (*shutdown)(grpc_exec_ctx *exec_ctx, grpc_completion_queue *cq);
  void (*destroy)(void *data);
  bool (*begin_op)(grpc_completion_queue *cq, void *tag);
  void (*end_op)(grpc_exec_ctx *exec_ctx, grpc_completion_queue *cq, void *tag,
                 grpc_error *error,
                 void (*done)(grpc_exec_ctx *exec_ctx, void *done_arg,
                              grpc_cq_completion *storage),
                 void *done_arg, grpc_cq_completion *storage);
  grpc_event (*next)(grpc_completion_queue *cq, gpr_timespec deadline,
                     void *reserved);
  grpc_event (*pluck)(grpc_completion_queue *cq, void *tag,
                      gpr_timespec deadline, void *reserved);
} cq_vtable;

/* Queue that holds the cq_completion_events. Internally uses gpr_mpscq queue
 * (a lockfree multiproducer single consumer queue). It uses a queue_lock
 * to support multiple consumers.
 * Only used in completion queues whose completion_type is GRPC_CQ_NEXT */
typedef struct grpc_cq_event_queue {
  /* Spinlock to serialize consumers i.e pop() operations */
  gpr_spinlock queue_lock;

  gpr_mpscq queue;

  /* A lazy counter of number of items in the queue. This is NOT atomically
     incremented/decremented along with push/pop operations and hence is only
     eventually consistent */
  gpr_atm num_queue_items;
} grpc_cq_event_queue;

typedef struct cq_next_data {
  /** Completed events for completion-queues of type GRPC_CQ_NEXT */
  grpc_cq_event_queue queue;

  /** Counter of how many things have ever been queued on this completion queue
      useful for avoiding locks to check the queue */
  gpr_atm things_queued_ever;

  /* Number of outstanding events (+1 if not shut down) */
  gpr_atm pending_events;

  /** 0 initially. 1 once we initiated shutdown */
  bool shutdown_called;
} cq_next_data;

typedef struct cq_pluck_data {
  /** Completed events for completion-queues of type GRPC_CQ_PLUCK */
  grpc_cq_completion completed_head;
  grpc_cq_completion *completed_tail;

  /** Number of pending events (+1 if we're not shutdown) */
  gpr_atm pending_events;

  /** Counter of how many things have ever been queued on this completion queue
      useful for avoiding locks to check the queue */
  gpr_atm things_queued_ever;

  /** 0 initially. 1 once we completed shutting */
  /* TODO: (sreek) This is not needed since (shutdown == 1) if and only if
   * (pending_events == 0). So consider removing this in future and use
   * pending_events */
  gpr_atm shutdown;

  /** 0 initially. 1 once we initiated shutdown */
  bool shutdown_called;

  int num_pluckers;
  plucker pluckers[GRPC_MAX_COMPLETION_QUEUE_PLUCKERS];
} cq_pluck_data;

/* Completion queue structure */
struct grpc_completion_queue {
  /** Once owning_refs drops to zero, we will destroy the cq */
  gpr_refcount owning_refs;

  gpr_mu *mu;

  const cq_vtable *vtable;
  const cq_poller_vtable *poller_vtable;

#ifndef NDEBUG
  void **outstanding_tags;
  size_t outstanding_tag_count;
  size_t outstanding_tag_capacity;
#endif

  grpc_closure pollset_shutdown_done;
  int num_polls;
};

/* Forward declarations */
static void cq_finish_shutdown_next(grpc_exec_ctx *exec_ctx,
                                    grpc_completion_queue *cq);
static void cq_finish_shutdown_pluck(grpc_exec_ctx *exec_ctx,
                                     grpc_completion_queue *cq);
static void cq_shutdown_next(grpc_exec_ctx *exec_ctx,
                             grpc_completion_queue *cq);
static void cq_shutdown_pluck(grpc_exec_ctx *exec_ctx,
                              grpc_completion_queue *cq);

static bool cq_begin_op_for_next(grpc_completion_queue *cq, void *tag);
static bool cq_begin_op_for_pluck(grpc_completion_queue *cq, void *tag);

static void cq_end_op_for_next(grpc_exec_ctx *exec_ctx,
                               grpc_completion_queue *cq, void *tag,
                               grpc_error *error,
                               void (*done)(grpc_exec_ctx *exec_ctx,
                                            void *done_arg,
                                            grpc_cq_completion *storage),
                               void *done_arg, grpc_cq_completion *storage);

static void cq_end_op_for_pluck(grpc_exec_ctx *exec_ctx,
                                grpc_completion_queue *cq, void *tag,
                                grpc_error *error,
                                void (*done)(grpc_exec_ctx *exec_ctx,
                                             void *done_arg,
                                             grpc_cq_completion *storage),
                                void *done_arg, grpc_cq_completion *storage);

static grpc_event cq_next(grpc_completion_queue *cq, gpr_timespec deadline,
                          void *reserved);

static grpc_event cq_pluck(grpc_completion_queue *cq, void *tag,
                           gpr_timespec deadline, void *reserved);

static void cq_init_next(void *data);
static void cq_init_pluck(void *data);
static void cq_destroy_next(void *data);
static void cq_destroy_pluck(void *data);

/* Completion queue vtables based on the completion-type */
static const cq_vtable g_cq_vtable[] = {
    /* GRPC_CQ_NEXT */
    {GRPC_CQ_NEXT, sizeof(cq_next_data), cq_init_next, cq_shutdown_next,
     cq_destroy_next, cq_begin_op_for_next, cq_end_op_for_next, cq_next, NULL},
    /* GRPC_CQ_PLUCK */
    {GRPC_CQ_PLUCK, sizeof(cq_pluck_data), cq_init_pluck, cq_shutdown_pluck,
     cq_destroy_pluck, cq_begin_op_for_pluck, cq_end_op_for_pluck, NULL,
     cq_pluck},
};

#define DATA_FROM_CQ(cq) ((void *)(cq + 1))
#define POLLSET_FROM_CQ(cq) \
  ((grpc_pollset *)(cq->vtable->data_size + (char *)DATA_FROM_CQ(cq)))

grpc_core::Tracer grpc_cq_pluck_trace(true, "queue_pluck");
grpc_core::Tracer grpc_cq_event_timeout_trace(true, "queue_timeout");

#define GRPC_SURFACE_TRACE_RETURNED_EVENT(cq, event)                       \
  if (grpc_api_trace.enabled() && (grpc_cq_pluck_trace.enabled() ||        \
                                   (event)->type != GRPC_QUEUE_TIMEOUT)) { \
    char *_ev = grpc_event_string(event);                                  \
    gpr_log(GPR_INFO, "RETURN_EVENT[%p]: %s", cq, _ev);                    \
    gpr_free(_ev);                                                         \
  }

static void on_pollset_shutdown_done(grpc_exec_ctx *exec_ctx, void *cq,
                                     grpc_error *error);

static void cq_event_queue_init(grpc_cq_event_queue *q) {
  gpr_mpscq_init(&q->queue);
  q->queue_lock = GPR_SPINLOCK_INITIALIZER;
  gpr_atm_no_barrier_store(&q->num_queue_items, 0);
}

static void cq_event_queue_destroy(grpc_cq_event_queue *q) {
  gpr_mpscq_destroy(&q->queue);
}

static bool cq_event_queue_push(grpc_cq_event_queue *q, grpc_cq_completion *c) {
  gpr_mpscq_push(&q->queue, (gpr_mpscq_node *)c);
  return gpr_atm_no_barrier_fetch_add(&q->num_queue_items, 1) == 0;
}

static grpc_cq_completion *cq_event_queue_pop(grpc_cq_event_queue *q) {
  grpc_cq_completion *c = NULL;
  grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT;

  if (gpr_spinlock_trylock(&q->queue_lock)) {
    GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_SUCCESSES(&exec_ctx);

    bool is_empty = false;
    c = (grpc_cq_completion *)gpr_mpscq_pop_and_check_end(&q->queue, &is_empty);
    gpr_spinlock_unlock(&q->queue_lock);

    if (c == NULL && !is_empty) {
      GRPC_STATS_INC_CQ_EV_QUEUE_TRANSIENT_POP_FAILURES(&exec_ctx);
    }
  } else {
    GRPC_STATS_INC_CQ_EV_QUEUE_TRYLOCK_FAILURES(&exec_ctx);
  }

  grpc_exec_ctx_finish(&exec_ctx);

  if (c) {
    gpr_atm_no_barrier_fetch_add(&q->num_queue_items, -1);
  }

  return c;
}

/* Note: The counter is not incremented/decremented atomically with push/pop.
 * The count is only eventually consistent */
static long cq_event_queue_num_items(grpc_cq_event_queue *q) {
  return (long)gpr_atm_no_barrier_load(&q->num_queue_items);
}

grpc_completion_queue *grpc_completion_queue_create_internal(
    grpc_cq_completion_type completion_type,
    grpc_cq_polling_type polling_type) {
  grpc_completion_queue *cq;

  GPR_TIMER_BEGIN("grpc_completion_queue_create_internal", 0);

  GRPC_API_TRACE(
      "grpc_completion_queue_create_internal(completion_type=%d, "
      "polling_type=%d)",
      2, (completion_type, polling_type));

  const cq_vtable *vtable = &g_cq_vtable[completion_type];
  const cq_poller_vtable *poller_vtable =
      &g_poller_vtable_by_poller_type[polling_type];

  grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT;
  GRPC_STATS_INC_CQS_CREATED(&exec_ctx);
  grpc_exec_ctx_finish(&exec_ctx);

  cq = (grpc_completion_queue *)gpr_zalloc(sizeof(grpc_completion_queue) +
                                           vtable->data_size +
                                           poller_vtable->size());

  cq->vtable = vtable;
  cq->poller_vtable = poller_vtable;

  /* One for destroy(), one for pollset_shutdown */
  gpr_ref_init(&cq->owning_refs, 2);

  poller_vtable->init(POLLSET_FROM_CQ(cq), &cq->mu);
  vtable->init(DATA_FROM_CQ(cq));

  GRPC_CLOSURE_INIT(&cq->pollset_shutdown_done, on_pollset_shutdown_done, cq,
                    grpc_schedule_on_exec_ctx);

  GPR_TIMER_END("grpc_completion_queue_create_internal", 0);

  return cq;
}

static void cq_init_next(void *ptr) {
  cq_next_data *cqd = (cq_next_data *)ptr;
  /* Initial count is dropped by grpc_completion_queue_shutdown */
  gpr_atm_no_barrier_store(&cqd->pending_events, 1);
  cqd->shutdown_called = false;
  gpr_atm_no_barrier_store(&cqd->things_queued_ever, 0);
  cq_event_queue_init(&cqd->queue);
}

static void cq_destroy_next(void *ptr) {
  cq_next_data *cqd = (cq_next_data *)ptr;
  GPR_ASSERT(cq_event_queue_num_items(&cqd->queue) == 0);
  cq_event_queue_destroy(&cqd->queue);
}

static void cq_init_pluck(void *ptr) {
  cq_pluck_data *cqd = (cq_pluck_data *)ptr;
  /* Initial count is dropped by grpc_completion_queue_shutdown */
  gpr_atm_no_barrier_store(&cqd->pending_events, 1);
  cqd->completed_tail = &cqd->completed_head;
  cqd->completed_head.next = (uintptr_t)cqd->completed_tail;
  gpr_atm_no_barrier_store(&cqd->shutdown, 0);
  cqd->shutdown_called = false;
  cqd->num_pluckers = 0;
  gpr_atm_no_barrier_store(&cqd->things_queued_ever, 0);
}

static void cq_destroy_pluck(void *ptr) {
  cq_pluck_data *cqd = (cq_pluck_data *)ptr;
  GPR_ASSERT(cqd->completed_head.next == (uintptr_t)&cqd->completed_head);
}

grpc_cq_completion_type grpc_get_cq_completion_type(grpc_completion_queue *cq) {
  return cq->vtable->cq_completion_type;
}

int grpc_get_cq_poll_num(grpc_completion_queue *cq) {
  int cur_num_polls;
  gpr_mu_lock(cq->mu);
  cur_num_polls = cq->num_polls;
  gpr_mu_unlock(cq->mu);
  return cur_num_polls;
}

#ifndef NDEBUG
void grpc_cq_internal_ref(grpc_completion_queue *cq, const char *reason,
                          const char *file, int line) {
  if (grpc_trace_cq_refcount.enabled()) {
    gpr_atm val = gpr_atm_no_barrier_load(&cq->owning_refs.count);
    gpr_log(file, line, GPR_LOG_SEVERITY_DEBUG,
            "CQ:%p   ref %" PRIdPTR " -> %" PRIdPTR " %s", cq, val, val + 1,
            reason);
  }
#else
void grpc_cq_internal_ref(grpc_completion_queue *cq) {
#endif
  gpr_ref(&cq->owning_refs);
}

static void on_pollset_shutdown_done(grpc_exec_ctx *exec_ctx, void *arg,
                                     grpc_error *error) {
  grpc_completion_queue *cq = (grpc_completion_queue *)arg;
  GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "pollset_destroy");
}

#ifndef NDEBUG
void grpc_cq_internal_unref(grpc_exec_ctx *exec_ctx, grpc_completion_queue *cq,
                            const char *reason, const char *file, int line) {
  if (grpc_trace_cq_refcount.enabled()) {
    gpr_atm val = gpr_atm_no_barrier_load(&cq->owning_refs.count);
    gpr_log(file, line, GPR_LOG_SEVERITY_DEBUG,
            "CQ:%p unref %" PRIdPTR " -> %" PRIdPTR " %s", cq, val, val - 1,
            reason);
  }
#else
void grpc_cq_internal_unref(grpc_exec_ctx *exec_ctx,
                            grpc_completion_queue *cq) {
#endif
  if (gpr_unref(&cq->owning_refs)) {
    cq->vtable->destroy(DATA_FROM_CQ(cq));
    cq->poller_vtable->destroy(exec_ctx, POLLSET_FROM_CQ(cq));
#ifndef NDEBUG
    gpr_free(cq->outstanding_tags);
#endif
    gpr_free(cq);
  }
}

#ifndef NDEBUG
static void cq_check_tag(grpc_completion_queue *cq, void *tag, bool lock_cq) {
  int found = 0;
  if (lock_cq) {
    gpr_mu_lock(cq->mu);
  }

  for (int i = 0; i < (int)cq->outstanding_tag_count; i++) {
    if (cq->outstanding_tags[i] == tag) {
      cq->outstanding_tag_count--;
      GPR_SWAP(void *, cq->outstanding_tags[i],
               cq->outstanding_tags[cq->outstanding_tag_count]);
      found = 1;
      break;
    }
  }

  if (lock_cq) {
    gpr_mu_unlock(cq->mu);
  }

  GPR_ASSERT(found);
}
#else
static void cq_check_tag(grpc_completion_queue *cq, void *tag, bool lock_cq) {}
#endif

/* Atomically increments a counter only if the counter is not zero. Returns
 * true if the increment was successful; false if the counter is zero */
static bool atm_inc_if_nonzero(gpr_atm *counter) {
  while (true) {
    gpr_atm count = gpr_atm_acq_load(counter);
    /* If zero, we are done. If not, we must to a CAS (instead of an atomic
     * increment) to maintain the contract: do not increment the counter if it
     * is zero. */
    if (count == 0) {
      return false;
    } else if (gpr_atm_full_cas(counter, count, count + 1)) {
      break;
    }
  }

  return true;
}

static bool cq_begin_op_for_next(grpc_completion_queue *cq, void *tag) {
  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);
  return atm_inc_if_nonzero(&cqd->pending_events);
}

static bool cq_begin_op_for_pluck(grpc_completion_queue *cq, void *tag) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);
  return atm_inc_if_nonzero(&cqd->pending_events);
}

bool grpc_cq_begin_op(grpc_completion_queue *cq, void *tag) {
#ifndef NDEBUG
  gpr_mu_lock(cq->mu);
  if (cq->outstanding_tag_count == cq->outstanding_tag_capacity) {
    cq->outstanding_tag_capacity = GPR_MAX(4, 2 * cq->outstanding_tag_capacity);
    cq->outstanding_tags = (void **)gpr_realloc(
        cq->outstanding_tags,
        sizeof(*cq->outstanding_tags) * cq->outstanding_tag_capacity);
  }
  cq->outstanding_tags[cq->outstanding_tag_count++] = tag;
  gpr_mu_unlock(cq->mu);
#endif
  return cq->vtable->begin_op(cq, tag);
}

/* Queue a GRPC_OP_COMPLETED operation to a completion queue (with a
 * completion
 * type of GRPC_CQ_NEXT) */
static void cq_end_op_for_next(grpc_exec_ctx *exec_ctx,
                               grpc_completion_queue *cq, void *tag,
                               grpc_error *error,
                               void (*done)(grpc_exec_ctx *exec_ctx,
                                            void *done_arg,
                                            grpc_cq_completion *storage),
                               void *done_arg, grpc_cq_completion *storage) {
  GPR_TIMER_BEGIN("cq_end_op_for_next", 0);

  if (grpc_api_trace.enabled() ||
      (grpc_trace_operation_failures.enabled() && error != GRPC_ERROR_NONE)) {
    const char *errmsg = grpc_error_string(error);
    GRPC_API_TRACE(
        "cq_end_op_for_next(exec_ctx=%p, cq=%p, tag=%p, error=%s, "
        "done=%p, done_arg=%p, storage=%p)",
        7, (exec_ctx, cq, tag, errmsg, done, done_arg, storage));
    if (grpc_trace_operation_failures.enabled() && error != GRPC_ERROR_NONE) {
      gpr_log(GPR_ERROR, "Operation failed: tag=%p, error=%s", tag, errmsg);
    }
  }

  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);
  int is_success = (error == GRPC_ERROR_NONE);

  storage->tag = tag;
  storage->done = done;
  storage->done_arg = done_arg;
  storage->next = (uintptr_t)(is_success);

  cq_check_tag(cq, tag, true); /* Used in debug builds only */

  /* Add the completion to the queue */
  bool is_first = cq_event_queue_push(&cqd->queue, storage);
  gpr_atm_no_barrier_fetch_add(&cqd->things_queued_ever, 1);

  /* Since we do not hold the cq lock here, it is important to do an 'acquire'
     load here (instead of a 'no_barrier' load) to match with the release store
     (done via gpr_atm_full_fetch_add(pending_events, -1)) in cq_shutdown_next
     */
  bool will_definitely_shutdown = gpr_atm_acq_load(&cqd->pending_events) == 1;

  if (!will_definitely_shutdown) {
    /* Only kick if this is the first item queued */
    if (is_first) {
      gpr_mu_lock(cq->mu);
      grpc_error *kick_error =
          cq->poller_vtable->kick(exec_ctx, POLLSET_FROM_CQ(cq), NULL);
      gpr_mu_unlock(cq->mu);

      if (kick_error != GRPC_ERROR_NONE) {
        const char *msg = grpc_error_string(kick_error);
        gpr_log(GPR_ERROR, "Kick failed: %s", msg);
        GRPC_ERROR_UNREF(kick_error);
      }
    }
    if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) {
      GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
      gpr_mu_lock(cq->mu);
      cq_finish_shutdown_next(exec_ctx, cq);
      gpr_mu_unlock(cq->mu);
      GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down");
    }
  } else {
    GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
    gpr_atm_rel_store(&cqd->pending_events, 0);
    gpr_mu_lock(cq->mu);
    cq_finish_shutdown_next(exec_ctx, cq);
    gpr_mu_unlock(cq->mu);
    GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down");
  }

  GPR_TIMER_END("cq_end_op_for_next", 0);

  GRPC_ERROR_UNREF(error);
}

/* Queue a GRPC_OP_COMPLETED operation to a completion queue (with a
 * completion
 * type of GRPC_CQ_PLUCK) */
static void cq_end_op_for_pluck(grpc_exec_ctx *exec_ctx,
                                grpc_completion_queue *cq, void *tag,
                                grpc_error *error,
                                void (*done)(grpc_exec_ctx *exec_ctx,
                                             void *done_arg,
                                             grpc_cq_completion *storage),
                                void *done_arg, grpc_cq_completion *storage) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);
  int is_success = (error == GRPC_ERROR_NONE);

  GPR_TIMER_BEGIN("cq_end_op_for_pluck", 0);

  if (grpc_api_trace.enabled() ||
      (grpc_trace_operation_failures.enabled() && error != GRPC_ERROR_NONE)) {
    const char *errmsg = grpc_error_string(error);
    GRPC_API_TRACE(
        "cq_end_op_for_pluck(exec_ctx=%p, cq=%p, tag=%p, error=%s, "
        "done=%p, done_arg=%p, storage=%p)",
        7, (exec_ctx, cq, tag, errmsg, done, done_arg, storage));
    if (grpc_trace_operation_failures.enabled() && error != GRPC_ERROR_NONE) {
      gpr_log(GPR_ERROR, "Operation failed: tag=%p, error=%s", tag, errmsg);
    }
  }

  storage->tag = tag;
  storage->done = done;
  storage->done_arg = done_arg;
  storage->next = ((uintptr_t)&cqd->completed_head) | ((uintptr_t)(is_success));

  gpr_mu_lock(cq->mu);
  cq_check_tag(cq, tag, false); /* Used in debug builds only */

  /* Add to the list of completions */
  gpr_atm_no_barrier_fetch_add(&cqd->things_queued_ever, 1);
  cqd->completed_tail->next =
      ((uintptr_t)storage) | (1u & (uintptr_t)cqd->completed_tail->next);
  cqd->completed_tail = storage;

  if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) {
    cq_finish_shutdown_pluck(exec_ctx, cq);
    gpr_mu_unlock(cq->mu);
  } else {
    grpc_pollset_worker *pluck_worker = NULL;
    for (int i = 0; i < cqd->num_pluckers; i++) {
      if (cqd->pluckers[i].tag == tag) {
        pluck_worker = *cqd->pluckers[i].worker;
        break;
      }
    }

    grpc_error *kick_error =
        cq->poller_vtable->kick(exec_ctx, POLLSET_FROM_CQ(cq), pluck_worker);

    gpr_mu_unlock(cq->mu);

    if (kick_error != GRPC_ERROR_NONE) {
      const char *msg = grpc_error_string(kick_error);
      gpr_log(GPR_ERROR, "Kick failed: %s", msg);

      GRPC_ERROR_UNREF(kick_error);
    }
  }

  GPR_TIMER_END("cq_end_op_for_pluck", 0);

  GRPC_ERROR_UNREF(error);
}

void grpc_cq_end_op(grpc_exec_ctx *exec_ctx, grpc_completion_queue *cq,
                    void *tag, grpc_error *error,
                    void (*done)(grpc_exec_ctx *exec_ctx, void *done_arg,
                                 grpc_cq_completion *storage),
                    void *done_arg, grpc_cq_completion *storage) {
  cq->vtable->end_op(exec_ctx, cq, tag, error, done, done_arg, storage);
}

typedef struct {
  gpr_atm last_seen_things_queued_ever;
  grpc_completion_queue *cq;
  grpc_millis deadline;
  grpc_cq_completion *stolen_completion;
  void *tag; /* for pluck */
  bool first_loop;
} cq_is_finished_arg;

static bool cq_is_next_finished(grpc_exec_ctx *exec_ctx, void *arg) {
  cq_is_finished_arg *a = (cq_is_finished_arg *)arg;
  grpc_completion_queue *cq = a->cq;
  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);
  GPR_ASSERT(a->stolen_completion == NULL);

  gpr_atm current_last_seen_things_queued_ever =
      gpr_atm_no_barrier_load(&cqd->things_queued_ever);

  if (current_last_seen_things_queued_ever != a->last_seen_things_queued_ever) {
    a->last_seen_things_queued_ever =
        gpr_atm_no_barrier_load(&cqd->things_queued_ever);

    /* Pop a cq_completion from the queue. Returns NULL if the queue is empty
     * might return NULL in some cases even if the queue is not empty; but
     * that
     * is ok and doesn't affect correctness. Might effect the tail latencies a
     * bit) */
    a->stolen_completion = cq_event_queue_pop(&cqd->queue);
    if (a->stolen_completion != NULL) {
      return true;
    }
  }
  return !a->first_loop && a->deadline < grpc_exec_ctx_now(exec_ctx);
}

#ifndef NDEBUG
static void dump_pending_tags(grpc_completion_queue *cq) {
  if (!grpc_trace_pending_tags.enabled()) return;

  gpr_strvec v;
  gpr_strvec_init(&v);
  gpr_strvec_add(&v, gpr_strdup("PENDING TAGS:"));
  gpr_mu_lock(cq->mu);
  for (size_t i = 0; i < cq->outstanding_tag_count; i++) {
    char *s;
    gpr_asprintf(&s, " %p", cq->outstanding_tags[i]);
    gpr_strvec_add(&v, s);
  }
  gpr_mu_unlock(cq->mu);
  char *out = gpr_strvec_flatten(&v, NULL);
  gpr_strvec_destroy(&v);
  gpr_log(GPR_DEBUG, "%s", out);
  gpr_free(out);
}
#else
static void dump_pending_tags(grpc_completion_queue *cq) {}
#endif

static grpc_event cq_next(grpc_completion_queue *cq, gpr_timespec deadline,
                          void *reserved) {
  grpc_event ret;
  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);

  GPR_TIMER_BEGIN("grpc_completion_queue_next", 0);

  GRPC_API_TRACE(
      "grpc_completion_queue_next("
      "cq=%p, "
      "deadline=gpr_timespec { tv_sec: %" PRId64
      ", tv_nsec: %d, clock_type: %d }, "
      "reserved=%p)",
      5, (cq, deadline.tv_sec, deadline.tv_nsec, (int)deadline.clock_type,
          reserved));
  GPR_ASSERT(!reserved);

  dump_pending_tags(cq);

  GRPC_CQ_INTERNAL_REF(cq, "next");

  grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline);
  cq_is_finished_arg is_finished_arg = {
      gpr_atm_no_barrier_load(&cqd->things_queued_ever),
      cq,
      deadline_millis,
      NULL,
      NULL,
      true};
  grpc_exec_ctx exec_ctx =
      GRPC_EXEC_CTX_INITIALIZER(0, cq_is_next_finished, &is_finished_arg);
  for (;;) {
    grpc_millis iteration_deadline = deadline_millis;

    if (is_finished_arg.stolen_completion != NULL) {
      grpc_cq_completion *c = is_finished_arg.stolen_completion;
      is_finished_arg.stolen_completion = NULL;
      ret.type = GRPC_OP_COMPLETE;
      ret.success = c->next & 1u;
      ret.tag = c->tag;
      c->done(&exec_ctx, c->done_arg, c);
      break;
    }

    grpc_cq_completion *c = cq_event_queue_pop(&cqd->queue);

    if (c != NULL) {
      ret.type = GRPC_OP_COMPLETE;
      ret.success = c->next & 1u;
      ret.tag = c->tag;
      c->done(&exec_ctx, c->done_arg, c);
      break;
    } else {
      /* If c == NULL it means either the queue is empty OR in an transient
         inconsistent state. If it is the latter, we shold do a 0-timeout poll
         so that the thread comes back quickly from poll to make a second
         attempt at popping. Not doing this can potentially deadlock this
         thread forever (if the deadline is infinity) */
      if (cq_event_queue_num_items(&cqd->queue) > 0) {
        iteration_deadline = 0;
      }
    }

    if (gpr_atm_acq_load(&cqd->pending_events) == 0) {
      /* Before returning, check if the queue has any items left over (since
         gpr_mpscq_pop() can sometimes return NULL even if the queue is not
         empty. If so, keep retrying but do not return GRPC_QUEUE_SHUTDOWN */
      if (cq_event_queue_num_items(&cqd->queue) > 0) {
        /* Go to the beginning of the loop. No point doing a poll because
           (cq->shutdown == true) is only possible when there is no pending
           work (i.e cq->pending_events == 0) and any outstanding completion
           events should have already been queued on this cq */
        continue;
      }

      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_SHUTDOWN;
      break;
    }

    if (!is_finished_arg.first_loop &&
        grpc_exec_ctx_now(&exec_ctx) >= deadline_millis) {
      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_TIMEOUT;
      dump_pending_tags(cq);
      break;
    }

    /* The main polling work happens in grpc_pollset_work */
    gpr_mu_lock(cq->mu);
    cq->num_polls++;
    grpc_error *err = cq->poller_vtable->work(&exec_ctx, POLLSET_FROM_CQ(cq),
                                              NULL, iteration_deadline);
    gpr_mu_unlock(cq->mu);

    if (err != GRPC_ERROR_NONE) {
      const char *msg = grpc_error_string(err);
      gpr_log(GPR_ERROR, "Completion queue next failed: %s", msg);

      GRPC_ERROR_UNREF(err);
      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_TIMEOUT;
      dump_pending_tags(cq);
      break;
    }
    is_finished_arg.first_loop = false;
  }

  if (cq_event_queue_num_items(&cqd->queue) > 0 &&
      gpr_atm_acq_load(&cqd->pending_events) > 0) {
    gpr_mu_lock(cq->mu);
    cq->poller_vtable->kick(&exec_ctx, POLLSET_FROM_CQ(cq), NULL);
    gpr_mu_unlock(cq->mu);
  }

  GRPC_SURFACE_TRACE_RETURNED_EVENT(cq, &ret);
  GRPC_CQ_INTERNAL_UNREF(&exec_ctx, cq, "next");
  grpc_exec_ctx_finish(&exec_ctx);
  GPR_ASSERT(is_finished_arg.stolen_completion == NULL);

  GPR_TIMER_END("grpc_completion_queue_next", 0);

  return ret;
}

/* Finishes the completion queue shutdown. This means that there are no more
   completion events / tags expected from the completion queue
   - Must be called under completion queue lock
   - Must be called only once in completion queue's lifetime
   - grpc_completion_queue_shutdown() MUST have been called before calling
   this function */
static void cq_finish_shutdown_next(grpc_exec_ctx *exec_ctx,
                                    grpc_completion_queue *cq) {
  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);

  GPR_ASSERT(cqd->shutdown_called);
  GPR_ASSERT(gpr_atm_no_barrier_load(&cqd->pending_events) == 0);

  cq->poller_vtable->shutdown(exec_ctx, POLLSET_FROM_CQ(cq),
                              &cq->pollset_shutdown_done);
}

static void cq_shutdown_next(grpc_exec_ctx *exec_ctx,
                             grpc_completion_queue *cq) {
  cq_next_data *cqd = (cq_next_data *)DATA_FROM_CQ(cq);

  /* Need an extra ref for cq here because:
   * We call cq_finish_shutdown_next() below, that would call pollset shutdown.
   * Pollset shutdown decrements the cq ref count which can potentially destroy
   * the cq (if that happens to be the last ref).
   * Creating an extra ref here prevents the cq from getting destroyed while
   * this function is still active */
  GRPC_CQ_INTERNAL_REF(cq, "shutting_down");
  gpr_mu_lock(cq->mu);
  if (cqd->shutdown_called) {
    gpr_mu_unlock(cq->mu);
    GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down");
    return;
  }
  cqd->shutdown_called = true;
  /* Doing a full_fetch_add (i.e acq/release) here to match with
   * cq_begin_op_for_next and and cq_end_op_for_next functions which read/write
   * on this counter without necessarily holding a lock on cq */
  if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) {
    cq_finish_shutdown_next(exec_ctx, cq);
  }
  gpr_mu_unlock(cq->mu);
  GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down");
}

grpc_event grpc_completion_queue_next(grpc_completion_queue *cq,
                                      gpr_timespec deadline, void *reserved) {
  return cq->vtable->next(cq, deadline, reserved);
}

static int add_plucker(grpc_completion_queue *cq, void *tag,
                       grpc_pollset_worker **worker) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);
  if (cqd->num_pluckers == GRPC_MAX_COMPLETION_QUEUE_PLUCKERS) {
    return 0;
  }
  cqd->pluckers[cqd->num_pluckers].tag = tag;
  cqd->pluckers[cqd->num_pluckers].worker = worker;
  cqd->num_pluckers++;
  return 1;
}

static void del_plucker(grpc_completion_queue *cq, void *tag,
                        grpc_pollset_worker **worker) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);
  for (int i = 0; i < cqd->num_pluckers; i++) {
    if (cqd->pluckers[i].tag == tag && cqd->pluckers[i].worker == worker) {
      cqd->num_pluckers--;
      GPR_SWAP(plucker, cqd->pluckers[i], cqd->pluckers[cqd->num_pluckers]);
      return;
    }
  }
  GPR_UNREACHABLE_CODE(return );
}

static bool cq_is_pluck_finished(grpc_exec_ctx *exec_ctx, void *arg) {
  cq_is_finished_arg *a = (cq_is_finished_arg *)arg;
  grpc_completion_queue *cq = a->cq;
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);

  GPR_ASSERT(a->stolen_completion == NULL);
  gpr_atm current_last_seen_things_queued_ever =
      gpr_atm_no_barrier_load(&cqd->things_queued_ever);
  if (current_last_seen_things_queued_ever != a->last_seen_things_queued_ever) {
    gpr_mu_lock(cq->mu);
    a->last_seen_things_queued_ever =
        gpr_atm_no_barrier_load(&cqd->things_queued_ever);
    grpc_cq_completion *c;
    grpc_cq_completion *prev = &cqd->completed_head;
    while ((c = (grpc_cq_completion *)(prev->next & ~(uintptr_t)1)) !=
           &cqd->completed_head) {
      if (c->tag == a->tag) {
        prev->next = (prev->next & (uintptr_t)1) | (c->next & ~(uintptr_t)1);
        if (c == cqd->completed_tail) {
          cqd->completed_tail = prev;
        }
        gpr_mu_unlock(cq->mu);
        a->stolen_completion = c;
        return true;
      }
      prev = c;
    }
    gpr_mu_unlock(cq->mu);
  }
  return !a->first_loop && a->deadline < grpc_exec_ctx_now(exec_ctx);
}

static grpc_event cq_pluck(grpc_completion_queue *cq, void *tag,
                           gpr_timespec deadline, void *reserved) {
  grpc_event ret;
  grpc_cq_completion *c;
  grpc_cq_completion *prev;
  grpc_pollset_worker *worker = NULL;
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);

  GPR_TIMER_BEGIN("grpc_completion_queue_pluck", 0);

  if (grpc_cq_pluck_trace.enabled()) {
    GRPC_API_TRACE(
        "grpc_completion_queue_pluck("
        "cq=%p, tag=%p, "
        "deadline=gpr_timespec { tv_sec: %" PRId64
        ", tv_nsec: %d, clock_type: %d }, "
        "reserved=%p)",
        6, (cq, tag, deadline.tv_sec, deadline.tv_nsec,
            (int)deadline.clock_type, reserved));
  }
  GPR_ASSERT(!reserved);

  dump_pending_tags(cq);

  GRPC_CQ_INTERNAL_REF(cq, "pluck");
  gpr_mu_lock(cq->mu);
  grpc_millis deadline_millis = grpc_timespec_to_millis_round_up(deadline);
  cq_is_finished_arg is_finished_arg = {
      gpr_atm_no_barrier_load(&cqd->things_queued_ever),
      cq,
      deadline_millis,
      NULL,
      tag,
      true};
  grpc_exec_ctx exec_ctx =
      GRPC_EXEC_CTX_INITIALIZER(0, cq_is_pluck_finished, &is_finished_arg);
  for (;;) {
    if (is_finished_arg.stolen_completion != NULL) {
      gpr_mu_unlock(cq->mu);
      c = is_finished_arg.stolen_completion;
      is_finished_arg.stolen_completion = NULL;
      ret.type = GRPC_OP_COMPLETE;
      ret.success = c->next & 1u;
      ret.tag = c->tag;
      c->done(&exec_ctx, c->done_arg, c);
      break;
    }
    prev = &cqd->completed_head;
    while ((c = (grpc_cq_completion *)(prev->next & ~(uintptr_t)1)) !=
           &cqd->completed_head) {
      if (c->tag == tag) {
        prev->next = (prev->next & (uintptr_t)1) | (c->next & ~(uintptr_t)1);
        if (c == cqd->completed_tail) {
          cqd->completed_tail = prev;
        }
        gpr_mu_unlock(cq->mu);
        ret.type = GRPC_OP_COMPLETE;
        ret.success = c->next & 1u;
        ret.tag = c->tag;
        c->done(&exec_ctx, c->done_arg, c);
        goto done;
      }
      prev = c;
    }
    if (gpr_atm_no_barrier_load(&cqd->shutdown)) {
      gpr_mu_unlock(cq->mu);
      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_SHUTDOWN;
      break;
    }
    if (!add_plucker(cq, tag, &worker)) {
      gpr_log(GPR_DEBUG,
              "Too many outstanding grpc_completion_queue_pluck calls: maximum "
              "is %d",
              GRPC_MAX_COMPLETION_QUEUE_PLUCKERS);
      gpr_mu_unlock(cq->mu);
      memset(&ret, 0, sizeof(ret));
      /* TODO(ctiller): should we use a different result here */
      ret.type = GRPC_QUEUE_TIMEOUT;
      dump_pending_tags(cq);
      break;
    }
    if (!is_finished_arg.first_loop &&
        grpc_exec_ctx_now(&exec_ctx) >= deadline_millis) {
      del_plucker(cq, tag, &worker);
      gpr_mu_unlock(cq->mu);
      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_TIMEOUT;
      dump_pending_tags(cq);
      break;
    }
    cq->num_polls++;
    grpc_error *err = cq->poller_vtable->work(&exec_ctx, POLLSET_FROM_CQ(cq),
                                              &worker, deadline_millis);
    if (err != GRPC_ERROR_NONE) {
      del_plucker(cq, tag, &worker);
      gpr_mu_unlock(cq->mu);
      const char *msg = grpc_error_string(err);
      gpr_log(GPR_ERROR, "Completion queue pluck failed: %s", msg);

      GRPC_ERROR_UNREF(err);
      memset(&ret, 0, sizeof(ret));
      ret.type = GRPC_QUEUE_TIMEOUT;
      dump_pending_tags(cq);
      break;
    }
    is_finished_arg.first_loop = false;
    del_plucker(cq, tag, &worker);
  }
done:
  GRPC_SURFACE_TRACE_RETURNED_EVENT(cq, &ret);
  GRPC_CQ_INTERNAL_UNREF(&exec_ctx, cq, "pluck");
  grpc_exec_ctx_finish(&exec_ctx);
  GPR_ASSERT(is_finished_arg.stolen_completion == NULL);

  GPR_TIMER_END("grpc_completion_queue_pluck", 0);

  return ret;
}

grpc_event grpc_completion_queue_pluck(grpc_completion_queue *cq, void *tag,
                                       gpr_timespec deadline, void *reserved) {
  return cq->vtable->pluck(cq, tag, deadline, reserved);
}

static void cq_finish_shutdown_pluck(grpc_exec_ctx *exec_ctx,
                                     grpc_completion_queue *cq) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);

  GPR_ASSERT(cqd->shutdown_called);
  GPR_ASSERT(!gpr_atm_no_barrier_load(&cqd->shutdown));
  gpr_atm_no_barrier_store(&cqd->shutdown, 1);

  cq->poller_vtable->shutdown(exec_ctx, POLLSET_FROM_CQ(cq),
                              &cq->pollset_shutdown_done);
}

/* NOTE: This function is almost exactly identical to cq_shutdown_next() but
 * merging them is a bit tricky and probably not worth it */
static void cq_shutdown_pluck(grpc_exec_ctx *exec_ctx,
                              grpc_completion_queue *cq) {
  cq_pluck_data *cqd = (cq_pluck_data *)DATA_FROM_CQ(cq);

  /* Need an extra ref for cq here because:
   * We call cq_finish_shutdown_pluck() below, that would call pollset shutdown.
   * Pollset shutdown decrements the cq ref count which can potentially destroy
   * the cq (if that happens to be the last ref).
   * Creating an extra ref here prevents the cq from getting destroyed while
   * this function is still active */
  GRPC_CQ_INTERNAL_REF(cq, "shutting_down (pluck cq)");
  gpr_mu_lock(cq->mu);
  if (cqd->shutdown_called) {
    gpr_mu_unlock(cq->mu);
    GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down (pluck cq)");
    return;
  }
  cqd->shutdown_called = true;
  if (gpr_atm_full_fetch_add(&cqd->pending_events, -1) == 1) {
    cq_finish_shutdown_pluck(exec_ctx, cq);
  }
  gpr_mu_unlock(cq->mu);
  GRPC_CQ_INTERNAL_UNREF(exec_ctx, cq, "shutting_down (pluck cq)");
}

/* Shutdown simply drops a ref that we reserved at creation time; if we drop
   to zero here, then enter shutdown mode and wake up any waiters */
void grpc_completion_queue_shutdown(grpc_completion_queue *cq) {
  grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT;
  GPR_TIMER_BEGIN("grpc_completion_queue_shutdown", 0);
  GRPC_API_TRACE("grpc_completion_queue_shutdown(cq=%p)", 1, (cq));
  cq->vtable->shutdown(&exec_ctx, cq);
  grpc_exec_ctx_finish(&exec_ctx);
  GPR_TIMER_END("grpc_completion_queue_shutdown", 0);
}

void grpc_completion_queue_destroy(grpc_completion_queue *cq) {
  GRPC_API_TRACE("grpc_completion_queue_destroy(cq=%p)", 1, (cq));
  GPR_TIMER_BEGIN("grpc_completion_queue_destroy", 0);
  grpc_completion_queue_shutdown(cq);

  grpc_exec_ctx exec_ctx = GRPC_EXEC_CTX_INIT;
  GRPC_CQ_INTERNAL_UNREF(&exec_ctx, cq, "destroy");
  grpc_exec_ctx_finish(&exec_ctx);
  GPR_TIMER_END("grpc_completion_queue_destroy", 0);
}

grpc_pollset *grpc_cq_pollset(grpc_completion_queue *cq) {
  return cq->poller_vtable->can_get_pollset ? POLLSET_FROM_CQ(cq) : NULL;
}

bool grpc_cq_can_listen(grpc_completion_queue *cq) {
  return cq->poller_vtable->can_listen;
}