summaryrefslogtreecommitdiff
path: root/kern/sref.c
blob: 65fc396458beaf45e747b39da77f6ee65941b7ea (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
/*
 * Copyright (c) 2014-2019 Richard Braun.
 *
 * This program is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 *
 *
 * This implementation is based on the paper "RadixVM: Scalable address
 * spaces for multithreaded applications" by Austin T. Clements,
 * M. Frans Kaashoek, and Nickolai Zeldovich. Specifically, it implements
 * the Refcache component described in the paper, with a few differences
 * outlined below.
 *
 * Refcache flushes delta caches directly from an interrupt handler, and
 * disables interrupts and preemption on cache access. That behavior is
 * realtime-unfriendly because of the potentially large number of deltas
 * in a cache. This module uses dedicated manager threads to perform
 * cache flushes and queue reviews, and only disables preemption on
 * individual delta access.
 *
 * Locking protocol : cache -> counter -> global data
 */

#include <assert.h>
#include <errno.h>
#include <stdbool.h>
#include <stddef.h>
#include <stdint.h>
#include <stdio.h>

#include <kern/atomic.h>
#include <kern/clock.h>
#include <kern/cpumap.h>
#include <kern/init.h>
#include <kern/list.h>
#include <kern/log.h>
#include <kern/macros.h>
#include <kern/panic.h>
#include <kern/percpu.h>
#include <kern/slist.h>
#include <kern/spinlock.h>
#include <kern/sref.h>
#include <kern/sref_i.h>
#include <kern/syscnt.h>
#include <kern/thread.h>
#include <machine/cpu.h>

/*
 * Delay (in milliseconds) until a new global epoch starts.
 */
#define SREF_EPOCH_START_DELAY 10

/*
 * Per-cache delta table size.
 */
#define SREF_CACHE_DELTA_TABLE_SIZE 4096

#if !ISP2(SREF_CACHE_DELTA_TABLE_SIZE)
#error "delta table size must be a power-of-two"
#endif

#ifdef __LP64__
#define SREF_HASH_SHIFT 3
#else
#define SREF_HASH_SHIFT 2
#endif

/*
 * Negative close to 0 so that an overflow occurs early.
 */
#define SREF_EPOCH_ID_INIT_VALUE ((unsigned int)-500)

/*
 * Since review queues are processor-local, at least two local epochs
 * must have passed before a zero is considered a true zero. As a result,
 * three queues are required, one for the current epoch, and two more.
 * The queues are stored in an array used as a ring buffer that moves
 * forward with each new local epoch. Indexing in this array is done
 * with a binary mask instead of a modulo, for performance reasons, and
 * consequently, the array size must be at least the nearest power-of-two
 * above three.
 */
#define SREF_NR_QUEUES P2ROUND(3, 2)

/*
 * Number of counters in review queue beyond which to issue a warning.
 */
#define SREF_NR_COUNTERS_WARN 10000

/*
 * Global data.
 *
 * Processors regularly check the global epoch ID against their own,
 * locally cached epoch ID. If they're the same, a processor flushes
 * its cached deltas, acknowledges its flush by decrementing the number
 * of pending acknowledgment counter, and increments its local epoch ID,
 * preventing additional flushes during the same epoch.
 *
 * The last processor to acknowledge starts the next epoch.
 *
 * The epoch ID and the pending acknowledgments counter fill an entire
 * cache line each in order to avoid false sharing on SMP. Whenever
 * multiple processors may access them, they must use atomic operations
 * to avoid data races.
 *
 * Atomic operations on the pending acknowledgments counter are done
 * with acquire-release ordering to enforce the memory ordering
 * guarantees required by both the implementation and the interface.
 */
struct sref_data {
    struct {
        alignas(CPU_L1_SIZE) unsigned int epoch_id;
    };

    struct {
        alignas(CPU_L1_SIZE) unsigned int nr_pending_acks;
    };

    uint64_t start_ts;
    struct syscnt sc_epochs;
    struct syscnt sc_dirty_zeroes;
    struct syscnt sc_true_zeroes;
    struct syscnt sc_revives;
    struct syscnt sc_last_epoch_ms;
    struct syscnt sc_longest_epoch_ms;
};

/*
 * Temporary difference to apply on a reference counter.
 *
 * Deltas are stored in per-processor caches and added to their global
 * counter when flushed. A delta is valid if and only if the counter it
 * points to isn't NULL.
 *
 * On cache flush, if a delta is valid, it must be flushed whatever its
 * value because a delta can be a dirty zero too. By flushing all valid
 * deltas, and clearing them all after a flush, activity on a counter is
 * reliably reported.
 */
struct sref_delta {
    struct list node;
    struct sref_counter *counter;
    unsigned long value;
};

struct sref_queue {
    struct slist counters;
    unsigned long size;
};

/*
 * Per-processor cache of deltas.
 *
 * A cache is dirty if there is at least one delta that requires flushing.
 * It may only be flushed once per epoch.
 *
 * Delta caches are implemented with hash tables for quick ref count to
 * delta lookups. For now, a very simple replacement policy, similar to
 * that described in the RadixVM paper, is used. Improve with an LRU-like
 * algorithm if this turns out to be a problem.
 *
 * Periodic events (normally the system timer tick) trigger cache checks.
 * A cache check may wake up the manager thread if the cache needs management,
 * i.e. if it's dirty or if there are counters to review. Otherwise, the
 * flush acknowledgment is done directly to avoid the cost of a thread
 * wake-up.
 *
 * Interrupts and preemption must be disabled when accessing a delta cache.
 */
struct sref_cache {
    struct sref_data *data;
    bool dirty;
    bool flushed;
    unsigned int epoch_id;
    struct sref_delta deltas[SREF_CACHE_DELTA_TABLE_SIZE];
    struct list valid_deltas;
    struct sref_queue queues[SREF_NR_QUEUES];
    struct thread *manager;
    struct syscnt sc_collisions;
    struct syscnt sc_flushes;
};

static struct sref_data sref_data;
static struct sref_cache sref_cache __percpu;

static unsigned int
sref_data_get_epoch_id(const struct sref_data *data)
{
    return data->epoch_id;
}

static bool
sref_data_check_epoch_id(const struct sref_data *data, unsigned int epoch_id)
{
    unsigned int global_epoch_id;

    global_epoch_id = atomic_load(&data->epoch_id, ATOMIC_RELAXED);

    if (unlikely(global_epoch_id == epoch_id)) {
        atomic_fence(ATOMIC_ACQUIRE);
        return true;
    }

    return false;
}

static void
sref_data_start_epoch(struct sref_data *data)
{
    uint64_t now, duration;
    unsigned int epoch_id;

    now = clock_get_time();
    duration = clock_ticks_to_ms(now - data->start_ts);
    syscnt_set(&data->sc_last_epoch_ms, duration);

    if (duration > syscnt_read(&data->sc_longest_epoch_ms)) {
        syscnt_set(&data->sc_longest_epoch_ms, duration);
    }

    assert(data->nr_pending_acks == 0);
    data->nr_pending_acks = cpu_count();
    data->start_ts = now;

    epoch_id = atomic_load(&data->epoch_id, ATOMIC_RELAXED);
    atomic_store(&data->epoch_id, epoch_id + 1, ATOMIC_RELEASE);
}

static void
sref_data_ack_cpu(struct sref_data *data)
{
    unsigned int prev;

    prev = atomic_fetch_sub(&data->nr_pending_acks, 1, ATOMIC_ACQ_REL);

    if (prev != 1) {
        assert(prev != 0);
        return;
    }

    syscnt_inc(&data->sc_epochs);
    sref_data_start_epoch(data);
}

static void
sref_data_update_stats(struct sref_data *data, int64_t nr_dirty_zeroes,
                       int64_t nr_true_zeroes, int64_t nr_revives)
{
    syscnt_add(&data->sc_dirty_zeroes, nr_dirty_zeroes);
    syscnt_add(&data->sc_true_zeroes, nr_true_zeroes);
    syscnt_add(&data->sc_revives, nr_revives);
}

static bool
sref_counter_aligned(const struct sref_counter *counter)
{
    return ((uintptr_t)counter & (~SREF_WEAKREF_MASK)) == 0;
}

static void
sref_weakref_init(struct sref_weakref *weakref, struct sref_counter *counter)
{
    assert(sref_counter_aligned(counter));
    weakref->addr = (uintptr_t)counter;
}

static void
sref_weakref_mark_dying(struct sref_weakref *weakref)
{
    atomic_or(&weakref->addr, SREF_WEAKREF_DYING, ATOMIC_RELAXED);
}

static void
sref_weakref_clear_dying(struct sref_weakref *weakref)
{
    atomic_and(&weakref->addr, SREF_WEAKREF_MASK, ATOMIC_RELAXED);
}

static int
sref_weakref_kill(struct sref_weakref *weakref)
{
    uintptr_t addr, oldval;

    addr = atomic_load(&weakref->addr, ATOMIC_RELAXED) | SREF_WEAKREF_DYING;
    oldval = atomic_cas(&weakref->addr, addr, (uintptr_t)NULL, ATOMIC_RELAXED);

    if (oldval != addr) {
        assert((oldval & SREF_WEAKREF_MASK) == (addr & SREF_WEAKREF_MASK));
        return EBUSY;
    }

    return 0;
}

static struct sref_counter *
sref_weakref_tryget(struct sref_weakref *weakref)
{
    uintptr_t addr, oldval, newval;

    do {
        addr = atomic_load(&weakref->addr, ATOMIC_RELAXED);
        newval = addr & SREF_WEAKREF_MASK;
        oldval = atomic_cas(&weakref->addr, addr, newval, ATOMIC_RELAXED);
    } while (oldval != addr);

    return (struct sref_counter *)newval;
}

static uintptr_t
sref_counter_hash(const struct sref_counter *counter)
{
    uintptr_t va;

    va = (uintptr_t)counter;

    assert(P2ALIGNED(va, 1UL << SREF_HASH_SHIFT));
    return va >> SREF_HASH_SHIFT;
}

static bool
sref_counter_is_queued(const struct sref_counter *counter)
{
    return counter->flags & SREF_CNTF_QUEUED;
}

static void
sref_counter_mark_queued(struct sref_counter *counter)
{
    counter->flags |= SREF_CNTF_QUEUED;
}

static void
sref_counter_clear_queued(struct sref_counter *counter)
{
    counter->flags &= ~SREF_CNTF_QUEUED;
}

static bool
sref_counter_is_dirty(const struct sref_counter *counter)
{
    return counter->flags & SREF_CNTF_DIRTY;
}

static void
sref_counter_mark_dirty(struct sref_counter *counter)
{
    counter->flags |= SREF_CNTF_DIRTY;
}

static void
sref_counter_clear_dirty(struct sref_counter *counter)
{
    counter->flags &= ~SREF_CNTF_DIRTY;
}

#ifdef SREF_VERIFY

static bool
sref_counter_is_unreferenced(const struct sref_counter *counter)
{
    return counter->flags & SREF_CNTF_UNREF;
}

static void
sref_counter_mark_unreferenced(struct sref_counter *counter)
{
    counter->flags |= SREF_CNTF_UNREF;
}

#endif /* SREF_VERIFY */

static void
sref_counter_mark_dying(struct sref_counter *counter)
{
    if (counter->weakref == NULL) {
        return;
    }

    sref_weakref_mark_dying(counter->weakref);
}

static void
sref_counter_clear_dying(struct sref_counter *counter)
{
    if (counter->weakref == NULL) {
        return;
    }

    sref_weakref_clear_dying(counter->weakref);
}

static int
sref_counter_kill_weakref(struct sref_counter *counter)
{
    if (counter->weakref == NULL) {
        return 0;
    }

    return sref_weakref_kill(counter->weakref);
}

static void __init
sref_queue_init(struct sref_queue *queue)
{
    slist_init(&queue->counters);
    queue->size = 0;
}

static bool
sref_queue_empty(const struct sref_queue *queue)
{
    return queue->size == 0;
}

static void
sref_queue_push(struct sref_queue *queue, struct sref_counter *counter)
{
    slist_insert_tail(&queue->counters, &counter->node);
    queue->size++;
}

static struct sref_counter *
sref_queue_pop(struct sref_queue *queue)
{
    struct sref_counter *counter;

    counter = slist_first_entry(&queue->counters, typeof(*counter), node);
    slist_remove(&queue->counters, NULL);
    queue->size--;
    return counter;
}

static void
sref_queue_move(struct sref_queue *dest, const struct sref_queue *src)
{
    slist_set_head(&dest->counters, &src->counters);
    dest->size = src->size;
}

static struct sref_queue *
sref_cache_get_queue(struct sref_cache *cache, size_t index)
{
    assert(index < ARRAY_SIZE(cache->queues));
    return &cache->queues[index];
}

static struct sref_queue *
sref_cache_get_queue_by_epoch_id(struct sref_cache *cache,
                                 unsigned int epoch_id)
{
    size_t mask;

    mask = ARRAY_SIZE(cache->queues) - 1;
    return sref_cache_get_queue(cache, epoch_id & mask);
}

static void
sref_cache_schedule_review(struct sref_cache *cache,
                           struct sref_counter *counter)
{
    struct sref_queue *queue;

    assert(!sref_counter_is_queued(counter));
    assert(!sref_counter_is_dirty(counter));

    sref_counter_mark_queued(counter);
    sref_counter_mark_dying(counter);

    queue = sref_cache_get_queue_by_epoch_id(cache, cache->epoch_id);
    sref_queue_push(queue, counter);
}

static void
sref_counter_add(struct sref_counter *counter, unsigned long delta,
                 struct sref_cache *cache)
{
    assert(!cpu_intr_enabled());

    spinlock_lock(&counter->lock);

    counter->value += delta;

    if (counter->value == 0) {
        if (sref_counter_is_queued(counter)) {
            sref_counter_mark_dirty(counter);
        } else {
            sref_cache_schedule_review(cache, counter);
        }
    }

    spinlock_unlock(&counter->lock);
}

static void
sref_counter_noref(struct work *work)
{
    struct sref_counter *counter;

    counter = structof(work, struct sref_counter, work);
    counter->noref_fn(counter);
}

static void __init
sref_delta_init(struct sref_delta *delta)
{
    delta->counter = NULL;
    delta->value = 0;
}

static struct sref_counter *
sref_delta_counter(struct sref_delta *delta)
{
    return delta->counter;
}

static void
sref_delta_set_counter(struct sref_delta *delta, struct sref_counter *counter)
{
    assert(delta->value == 0);
    delta->counter = counter;
}

static void
sref_delta_clear(struct sref_delta *delta)
{
    assert(delta->value == 0);
    delta->counter = NULL;
}

static void
sref_delta_inc(struct sref_delta *delta)
{
    delta->value++;
}

static void
sref_delta_dec(struct sref_delta *delta)
{
    delta->value--;
}

static bool
sref_delta_is_valid(const struct sref_delta *delta)
{
    return delta->counter;
}

static void
sref_delta_flush(struct sref_delta *delta, struct sref_cache *cache)
{
    sref_counter_add(delta->counter, delta->value, cache);
    delta->value = 0;
}

static void
sref_delta_evict(struct sref_delta *delta, struct sref_cache *cache)
{
    sref_delta_flush(delta, cache);
    sref_delta_clear(delta);
}

static struct sref_cache *
sref_get_local_cache(void)
{
    return cpu_local_ptr(sref_cache);
}

static uintptr_t
sref_cache_compute_counter_index(const struct sref_cache *cache,
                                 const struct sref_counter *counter)
{
    return sref_counter_hash(counter) & (ARRAY_SIZE(cache->deltas) - 1);
}

static struct sref_delta *
sref_cache_get_delta(struct sref_cache *cache, size_t index)
{
    assert(index < ARRAY_SIZE(cache->deltas));
    return &cache->deltas[index];
}

static struct sref_cache *
sref_cache_acquire(unsigned long *flags)
{
    thread_preempt_disable_intr_save(flags);
    return sref_get_local_cache();
}

static void
sref_cache_release(unsigned long flags)
{
    thread_preempt_enable_intr_restore(flags);
}

static bool
sref_cache_is_dirty(const struct sref_cache *cache)
{
    return cache->dirty;
}

static void
sref_cache_set_dirty(struct sref_cache *cache)
{
    cache->dirty = true;
}

static void
sref_cache_clear_dirty(struct sref_cache *cache)
{
    cache->dirty = false;
}

static bool
sref_cache_is_flushed(const struct sref_cache *cache)
{
    return cache->flushed;
}

static void
sref_cache_set_flushed(struct sref_cache *cache)
{
    cache->flushed = true;
}

static void
sref_cache_clear_flushed(struct sref_cache *cache)
{
    cache->flushed = false;
}

static void
sref_cache_add_delta(struct sref_cache *cache, struct sref_delta *delta,
                     struct sref_counter *counter)
{
    assert(!sref_delta_is_valid(delta));
    assert(counter);

    sref_delta_set_counter(delta, counter);
    list_insert_tail(&cache->valid_deltas, &delta->node);
}

static void
sref_cache_remove_delta(struct sref_cache *cache, struct sref_delta *delta)
{
    assert(sref_delta_is_valid(delta));

    sref_delta_evict(delta, cache);
    list_remove(&delta->node);
}

static struct sref_delta *
sref_cache_take_delta(struct sref_cache *cache, struct sref_counter *counter)
{
    struct sref_delta *delta;
    size_t index;

    index = sref_cache_compute_counter_index(cache, counter);
    delta = sref_cache_get_delta(cache, index);

    if (!sref_delta_is_valid(delta)) {
        sref_cache_add_delta(cache, delta, counter);
    } else if (sref_delta_counter(delta) != counter) {
        sref_cache_remove_delta(cache, delta);
        sref_cache_add_delta(cache, delta, counter);
        syscnt_inc(&cache->sc_collisions);
    }

    return delta;
}

static bool
sref_cache_needs_management(struct sref_cache *cache)
{
    const struct sref_queue *queue;

    assert(!cpu_intr_enabled());
    assert(!thread_preempt_enabled());

    queue = sref_cache_get_queue_by_epoch_id(cache, cache->epoch_id - 2);
    return sref_cache_is_dirty(cache) || !sref_queue_empty(queue);
}

static void
sref_cache_end_epoch(struct sref_cache *cache)
{
    assert(!sref_cache_needs_management(cache));

    sref_data_ack_cpu(cache->data);
    cache->epoch_id++;
}

static void
sref_cache_flush(struct sref_cache *cache, struct sref_queue *queue)
{
    struct sref_queue *prev_queue;
    unsigned long flags;

    for (;;) {
        struct sref_delta *delta;

        thread_preempt_disable_intr_save(&flags);

        if (list_empty(&cache->valid_deltas)) {
            break;
        }

        delta = list_first_entry(&cache->valid_deltas, typeof(*delta), node);
        sref_cache_remove_delta(cache, delta);

        thread_preempt_enable_intr_restore(flags);
    }

    sref_cache_clear_dirty(cache);
    sref_cache_set_flushed(cache);

    prev_queue = sref_cache_get_queue_by_epoch_id(cache, cache->epoch_id - 2);
    sref_queue_move(queue, prev_queue);
    sref_queue_init(prev_queue);

    sref_cache_end_epoch(cache);

    thread_preempt_enable_intr_restore(flags);

    syscnt_inc(&cache->sc_flushes);
}

static void
sref_queue_review(struct sref_queue *queue, struct sref_cache *cache)
{
    int64_t nr_dirty_zeroes, nr_true_zeroes, nr_revives;
    struct sref_counter *counter;
    struct work_queue works;
    unsigned long flags;
    bool requeue;
    int error;

    nr_dirty_zeroes = 0;
    nr_true_zeroes = 0;
    nr_revives = 0;
    work_queue_init(&works);

    while (!sref_queue_empty(queue)) {
        counter = sref_queue_pop(queue);

        spinlock_lock_intr_save(&counter->lock, &flags);

#ifdef SREF_VERIFY
        assert(!sref_counter_is_unreferenced(counter));
#endif /* SREF_VERIFY */

        assert(sref_counter_is_queued(counter));
        sref_counter_clear_queued(counter);

        if (counter->value != 0) {
            sref_counter_clear_dirty(counter);
            sref_counter_clear_dying(counter);
            spinlock_unlock_intr_restore(&counter->lock, flags);
            continue;
        }

        if (sref_counter_is_dirty(counter)) {
            requeue = true;
            nr_dirty_zeroes++;
            sref_counter_clear_dirty(counter);
        } else {
            error = sref_counter_kill_weakref(counter);

            if (!error) {
                requeue = false;
            } else {
                requeue = true;
                nr_revives++;
            }
        }

        if (requeue) {
            sref_cache_schedule_review(cache, counter);
            spinlock_unlock_intr_restore(&counter->lock, flags);
        } else {

            /*
             * Keep in mind that the work structure shares memory with
             * the counter data.
             */

#ifdef SREF_VERIFY
            sref_counter_mark_unreferenced(counter);
#endif /* SREF_VERIFY */

            /*
             * Unlocking isn't needed here, since this counter is now
             * really at 0, but do it for consistency.
             */
            spinlock_unlock_intr_restore(&counter->lock, flags);

            nr_true_zeroes++;
            work_init(&counter->work, sref_counter_noref);
            work_queue_push(&works, &counter->work);
        }
    }

    if (work_queue_nr_works(&works) != 0) {
        work_queue_schedule(&works, WORK_HIGHPRIO);
    }

    sref_data_update_stats(cache->data, nr_dirty_zeroes,
                           nr_true_zeroes, nr_revives);
}

static void
sref_cache_manage(void *arg)
{
    struct sref_cache *cache;
    struct sref_queue queue;
    unsigned long flags;

    cache = arg;

    thread_preempt_disable_intr_save(&flags);

    for (;;) {

        while (sref_cache_is_flushed(cache)) {
            thread_sleep(NULL, cache, "sref");
        }

        thread_preempt_enable_intr_restore(flags);

        sref_cache_flush(cache, &queue);
        sref_queue_review(&queue, cache);

        thread_preempt_disable_intr_save(&flags);
    }

    /* Never reached */
}

static void
sref_cache_check(struct sref_cache *cache)
{
    bool same_epoch;

    same_epoch = sref_data_check_epoch_id(&sref_data, cache->epoch_id);

    if (!same_epoch) {
        return;
    }

    if (!sref_cache_needs_management(cache)) {
        sref_cache_end_epoch(cache);
        return;
    }

    sref_cache_clear_flushed(cache);
    thread_wakeup(cache->manager);
}

static void __init
sref_cache_init(struct sref_cache *cache, unsigned int cpu,
                struct sref_data *data)
{
    char name[SYSCNT_NAME_SIZE];

    cache->data = data;
    cache->dirty = false;
    cache->flushed = true;
    cache->epoch_id = sref_data_get_epoch_id(&sref_data) + 1;

    for (size_t i = 0; i < ARRAY_SIZE(cache->deltas); i++) {
        sref_delta_init(sref_cache_get_delta(cache, i));
    }

    list_init(&cache->valid_deltas);

    for (size_t i = 0; i < ARRAY_SIZE(cache->queues); i++) {
        sref_queue_init(sref_cache_get_queue(cache, i));
    }

    snprintf(name, sizeof(name), "sref_collisions/%u", cpu);
    syscnt_register(&cache->sc_collisions, name);
    snprintf(name, sizeof(name), "sref_flushes/%u", cpu);
    syscnt_register(&cache->sc_flushes, name);
    cache->manager = NULL;
}

static void __init
sref_cache_init_manager(struct sref_cache *cache, unsigned int cpu)
{
    char name[THREAD_NAME_SIZE];
    struct thread_attr attr;
    struct thread *manager;
    struct cpumap *cpumap;
    int error;

    error = cpumap_create(&cpumap);

    if (error) {
        panic("sref: unable to create manager thread CPU map");
    }

    cpumap_zero(cpumap);
    cpumap_set(cpumap, cpu);
    snprintf(name, sizeof(name), THREAD_KERNEL_PREFIX "sref_cache_manage/%u",
             cpu);
    thread_attr_init(&attr, name);
    thread_attr_set_cpumap(&attr, cpumap);
    thread_attr_set_priority(&attr, THREAD_SCHED_FS_PRIO_MAX);
    error = thread_create(&manager, &attr, sref_cache_manage, cache);
    cpumap_destroy(cpumap);

    if (error) {
        panic("sref: unable to create manager thread");
    }

    cache->manager = manager;
}

static void __init
sref_data_init(struct sref_data *data)
{
    data->epoch_id = SREF_EPOCH_ID_INIT_VALUE;
    data->nr_pending_acks = 0;
    data->start_ts = clock_get_time();

    syscnt_register(&data->sc_epochs, "sref_epochs");
    syscnt_register(&data->sc_dirty_zeroes, "sref_dirty_zeroes");
    syscnt_register(&data->sc_true_zeroes, "sref_true_zeroes");
    syscnt_register(&data->sc_revives, "sref_revives");
    syscnt_register(&data->sc_last_epoch_ms, "sref_last_epoch_ms");
    syscnt_register(&data->sc_longest_epoch_ms, "sref_longest_epoch_ms");
}

static int __init
sref_bootstrap(void)
{
    sref_data_init(&sref_data);
    sref_cache_init(sref_get_local_cache(), 0, &sref_data);
    return 0;
}

INIT_OP_DEFINE(sref_bootstrap,
               INIT_OP_DEP(cpu_setup, true),
               INIT_OP_DEP(spinlock_setup, true),
               INIT_OP_DEP(syscnt_setup, true),
               INIT_OP_DEP(thread_bootstrap, true));

static int __init
sref_setup(void)
{
    for (unsigned int i = 1; i < cpu_count(); i++) {
        sref_cache_init(percpu_ptr(sref_cache, i), i, &sref_data);
    }

    for (unsigned int i = 0; i < cpu_count(); i++) {
        sref_cache_init_manager(percpu_ptr(sref_cache, i), i);
    }

    sref_data_start_epoch(&sref_data);

    return 0;
}

INIT_OP_DEFINE(sref_setup,
               INIT_OP_DEP(cpu_mp_probe, true),
               INIT_OP_DEP(cpumap_setup, true),
               INIT_OP_DEP(log_setup, true),
               INIT_OP_DEP(sref_bootstrap, true),
               INIT_OP_DEP(thread_setup, true));

void
sref_report_periodic_event(void)
{
    sref_cache_check(sref_get_local_cache());
}

void
sref_counter_init(struct sref_counter *counter,
                  unsigned long init_value,
                  struct sref_weakref *weakref,
                  sref_noref_fn_t noref_fn)
{
    assert(init_value != 0);

    counter->noref_fn = noref_fn;
    spinlock_init(&counter->lock);
    counter->flags = 0;
    counter->value = init_value;
    counter->weakref = weakref;

    if (weakref) {
        sref_weakref_init(weakref, counter);
    }
}

static void
sref_counter_inc_common(struct sref_counter *counter, struct sref_cache *cache)
{
    struct sref_delta *delta;

    sref_cache_set_dirty(cache);
    delta = sref_cache_take_delta(cache, counter);
    sref_delta_inc(delta);
}

void
sref_counter_inc(struct sref_counter *counter)
{
    struct sref_cache *cache;
    unsigned long flags;

    cache = sref_cache_acquire(&flags);
    sref_counter_inc_common(counter, cache);
    sref_cache_release(flags);
}

void
sref_counter_dec(struct sref_counter *counter)
{
    struct sref_cache *cache;
    struct sref_delta *delta;
    unsigned long flags;

    cache = sref_cache_acquire(&flags);
    sref_cache_set_dirty(cache);
    delta = sref_cache_take_delta(cache, counter);
    sref_delta_dec(delta);
    sref_cache_release(flags);
}

struct sref_counter *
sref_weakref_get(struct sref_weakref *weakref)
{
    struct sref_counter *counter;
    struct sref_cache *cache;
    unsigned long flags;

    cache = sref_cache_acquire(&flags);

    counter = sref_weakref_tryget(weakref);

    if (counter) {
        sref_counter_inc_common(counter, cache);
    }

    sref_cache_release(flags);

    return counter;
}