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
|
/*-
* Copyright (c) 2006 John Baldwin <jhb@FreeBSD.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the author nor the names of any co-contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*
* Machine independent bits of reader/writer lock implementation.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
#include "opt_kdtrace.h"
#include "opt_no_adaptive_rwlocks.h"
#include <sys/param.h>
#include <sys/ktr.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/rwlock.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <sys/turnstile.h>
#include <machine/cpu.h>
#if defined(SMP) && !defined(NO_ADAPTIVE_RWLOCKS)
#define ADAPTIVE_RWLOCKS
#endif
#ifdef ADAPTIVE_RWLOCKS
static int rowner_retries = 10;
static int rowner_loops = 10000;
SYSCTL_NODE(_debug, OID_AUTO, rwlock, CTLFLAG_RD, NULL, "rwlock debugging");
SYSCTL_INT(_debug_rwlock, OID_AUTO, retry, CTLFLAG_RW, &rowner_retries, 0, "");
SYSCTL_INT(_debug_rwlock, OID_AUTO, loops, CTLFLAG_RW, &rowner_loops, 0, "");
#endif
#ifdef DDB
#include <ddb/ddb.h>
static void db_show_rwlock(struct lock_object *lock);
#endif
static void assert_rw(struct lock_object *lock, int what);
static void lock_rw(struct lock_object *lock, int how);
#ifdef KDTRACE_HOOKS
static int owner_rw(struct lock_object *lock, struct thread **owner);
#endif
static int unlock_rw(struct lock_object *lock);
struct lock_class lock_class_rw = {
.lc_name = "rw",
.lc_flags = LC_SLEEPLOCK | LC_RECURSABLE | LC_UPGRADABLE,
.lc_assert = assert_rw,
#ifdef DDB
.lc_ddb_show = db_show_rwlock,
#endif
.lc_lock = lock_rw,
.lc_unlock = unlock_rw,
#ifdef KDTRACE_HOOKS
.lc_owner = owner_rw,
#endif
};
/*
* Return a pointer to the owning thread if the lock is write-locked or
* NULL if the lock is unlocked or read-locked.
*/
#define rw_wowner(rw) \
((rw)->rw_lock & RW_LOCK_READ ? NULL : \
(struct thread *)RW_OWNER((rw)->rw_lock))
/*
* Returns if a write owner is recursed. Write ownership is not assured
* here and should be previously checked.
*/
#define rw_recursed(rw) ((rw)->rw_recurse != 0)
/*
* Return true if curthread helds the lock.
*/
#define rw_wlocked(rw) (rw_wowner((rw)) == curthread)
/*
* Return a pointer to the owning thread for this lock who should receive
* any priority lent by threads that block on this lock. Currently this
* is identical to rw_wowner().
*/
#define rw_owner(rw) rw_wowner(rw)
#ifndef INVARIANTS
#define _rw_assert(rw, what, file, line)
#endif
void
assert_rw(struct lock_object *lock, int what)
{
rw_assert((struct rwlock *)lock, what);
}
void
lock_rw(struct lock_object *lock, int how)
{
struct rwlock *rw;
rw = (struct rwlock *)lock;
if (how)
rw_wlock(rw);
else
rw_rlock(rw);
}
int
unlock_rw(struct lock_object *lock)
{
struct rwlock *rw;
rw = (struct rwlock *)lock;
rw_assert(rw, RA_LOCKED | LA_NOTRECURSED);
if (rw->rw_lock & RW_LOCK_READ) {
rw_runlock(rw);
return (0);
} else {
rw_wunlock(rw);
return (1);
}
}
#ifdef KDTRACE_HOOKS
int
owner_rw(struct lock_object *lock, struct thread **owner)
{
struct rwlock *rw = (struct rwlock *)lock;
uintptr_t x = rw->rw_lock;
*owner = rw_wowner(rw);
return ((x & RW_LOCK_READ) != 0 ? (RW_READERS(x) != 0) :
(*owner != NULL));
}
#endif
void
rw_init_flags(struct rwlock *rw, const char *name, int opts)
{
int flags;
MPASS((opts & ~(RW_DUPOK | RW_NOPROFILE | RW_NOWITNESS | RW_QUIET |
RW_RECURSE)) == 0);
ASSERT_ATOMIC_LOAD_PTR(rw->rw_lock,
("%s: rw_lock not aligned for %s: %p", __func__, name,
&rw->rw_lock));
flags = LO_UPGRADABLE;
if (opts & RW_DUPOK)
flags |= LO_DUPOK;
if (opts & RW_NOPROFILE)
flags |= LO_NOPROFILE;
if (!(opts & RW_NOWITNESS))
flags |= LO_WITNESS;
if (opts & RW_RECURSE)
flags |= LO_RECURSABLE;
if (opts & RW_QUIET)
flags |= LO_QUIET;
rw->rw_lock = RW_UNLOCKED;
rw->rw_recurse = 0;
lock_init(&rw->lock_object, &lock_class_rw, name, NULL, flags);
}
void
rw_destroy(struct rwlock *rw)
{
KASSERT(rw->rw_lock == RW_UNLOCKED, ("rw lock %p not unlocked", rw));
KASSERT(rw->rw_recurse == 0, ("rw lock %p still recursed", rw));
rw->rw_lock = RW_DESTROYED;
lock_destroy(&rw->lock_object);
}
void
rw_sysinit(void *arg)
{
struct rw_args *args = arg;
rw_init(args->ra_rw, args->ra_desc);
}
void
rw_sysinit_flags(void *arg)
{
struct rw_args_flags *args = arg;
rw_init_flags(args->ra_rw, args->ra_desc, args->ra_flags);
}
int
rw_wowned(struct rwlock *rw)
{
return (rw_wowner(rw) == curthread);
}
void
_rw_wlock(struct rwlock *rw, const char *file, int line)
{
MPASS(curthread != NULL);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_wlock() of destroyed rwlock @ %s:%d", file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file,
line, NULL);
__rw_wlock(rw, curthread, file, line);
LOCK_LOG_LOCK("WLOCK", &rw->lock_object, 0, rw->rw_recurse, file, line);
WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line);
curthread->td_locks++;
}
int
_rw_try_wlock(struct rwlock *rw, const char *file, int line)
{
int rval;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_try_wlock() of destroyed rwlock @ %s:%d", file, line));
if (rw_wlocked(rw) &&
(rw->lock_object.lo_flags & LO_RECURSABLE) != 0) {
rw->rw_recurse++;
rval = 1;
} else
rval = atomic_cmpset_acq_ptr(&rw->rw_lock, RW_UNLOCKED,
(uintptr_t)curthread);
LOCK_LOG_TRY("WLOCK", &rw->lock_object, 0, rval, file, line);
if (rval) {
WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
file, line);
curthread->td_locks++;
}
return (rval);
}
void
_rw_wunlock(struct rwlock *rw, const char *file, int line)
{
MPASS(curthread != NULL);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_wunlock() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_WLOCKED, file, line);
curthread->td_locks--;
WITNESS_UNLOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("WUNLOCK", &rw->lock_object, 0, rw->rw_recurse, file,
line);
if (!rw_recursed(rw))
LOCKSTAT_PROFILE_RELEASE_LOCK(LS_RW_WUNLOCK_RELEASE, rw);
__rw_wunlock(rw, curthread, file, line);
}
/*
* Determines whether a new reader can acquire a lock. Succeeds if the
* reader already owns a read lock and the lock is locked for read to
* prevent deadlock from reader recursion. Also succeeds if the lock
* is unlocked and has no writer waiters or spinners. Failing otherwise
* prioritizes writers before readers.
*/
#define RW_CAN_READ(_rw) \
((curthread->td_rw_rlocks && (_rw) & RW_LOCK_READ) || ((_rw) & \
(RW_LOCK_READ | RW_LOCK_WRITE_WAITERS | RW_LOCK_WRITE_SPINNER)) == \
RW_LOCK_READ)
void
_rw_rlock(struct rwlock *rw, const char *file, int line)
{
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
uintptr_t v;
#ifdef KDTRACE_HOOKS
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
#endif
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_rlock() of destroyed rwlock @ %s:%d", file, line));
KASSERT(rw_wowner(rw) != curthread,
("%s (%s): wlock already held @ %s:%d", __func__,
rw->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL);
for (;;) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
/*
* Handle the easy case. If no other thread has a write
* lock, then try to bump up the count of read locks. Note
* that we have to preserve the current state of the
* RW_LOCK_WRITE_WAITERS flag. If we fail to acquire a
* read lock, then rw_lock must have changed, so restart
* the loop. Note that this handles the case of a
* completely unlocked rwlock since such a lock is encoded
* as a read lock with no waiters.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
/*
* The RW_LOCK_READ_WAITERS flag should only be set
* if the lock has been unlocked and write waiters
* were present.
*/
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v,
v + RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeed %p -> %p", __func__,
rw, (void *)v,
(void *)(v + RW_ONE_READER));
break;
}
continue;
}
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, rw, owner);
while ((struct thread*)RW_OWNER(rw->rw_lock) ==
owner && TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
continue;
}
} else if (spintries < rowner_retries) {
spintries++;
for (i = 0; i < rowner_loops; i++) {
v = rw->rw_lock;
if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(v))
break;
cpu_spinwait();
}
if (i != rowner_loops)
continue;
}
#endif
/*
* Okay, now it's the hard case. Some other thread already
* has a write lock or there are write waiters present,
* acquire the turnstile lock so we can begin the process
* of blocking.
*/
ts = turnstile_trywait(&rw->lock_object);
/*
* The lock might have been released while we spun, so
* recheck its state and restart the loop if needed.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* The lock is held in write mode or it already has waiters.
*/
MPASS(!RW_CAN_READ(v));
/*
* If the RW_LOCK_READ_WAITERS flag is already set, then
* we can go ahead and block. If it is not set then try
* to set it. If we fail to set it drop the turnstile
* lock and restart the loop.
*/
if (!(v & RW_LOCK_READ_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_READ_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set read waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the read waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs();
#endif
turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs();
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
}
/*
* TODO: acquire "owner of record" here. Here be turnstile dragons
* however. turnstiles don't like owners changing between calls to
* turnstile_wait() currently.
*/
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_RW_RLOCK_ACQUIRE, rw, contested,
waittime, file, line);
LOCK_LOG_LOCK("RLOCK", &rw->lock_object, 0, 0, file, line);
WITNESS_LOCK(&rw->lock_object, 0, file, line);
curthread->td_locks++;
curthread->td_rw_rlocks++;
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(LS_RW_RLOCK_BLOCK, rw, sleep_time);
/*
* Record only the loops spinning and not sleeping.
*/
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(LS_RW_RLOCK_SPIN, rw, (spin_cnt - sleep_cnt));
#endif
}
int
_rw_try_rlock(struct rwlock *rw, const char *file, int line)
{
uintptr_t x;
for (;;) {
x = rw->rw_lock;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_try_rlock() of destroyed rwlock @ %s:%d", file, line));
if (!(x & RW_LOCK_READ))
break;
if (atomic_cmpset_acq_ptr(&rw->rw_lock, x, x + RW_ONE_READER)) {
LOCK_LOG_TRY("RLOCK", &rw->lock_object, 0, 1, file,
line);
WITNESS_LOCK(&rw->lock_object, LOP_TRYLOCK, file, line);
curthread->td_locks++;
curthread->td_rw_rlocks++;
return (1);
}
}
LOCK_LOG_TRY("RLOCK", &rw->lock_object, 0, 0, file, line);
return (0);
}
void
_rw_runlock(struct rwlock *rw, const char *file, int line)
{
struct turnstile *ts;
uintptr_t x, v, queue;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_runlock() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_RLOCKED, file, line);
curthread->td_locks--;
curthread->td_rw_rlocks--;
WITNESS_UNLOCK(&rw->lock_object, 0, file, line);
LOCK_LOG_LOCK("RUNLOCK", &rw->lock_object, 0, 0, file, line);
/* TODO: drop "owner of record" here. */
for (;;) {
/*
* See if there is more than one read lock held. If so,
* just drop one and return.
*/
x = rw->rw_lock;
if (RW_READERS(x) > 1) {
if (atomic_cmpset_rel_ptr(&rw->rw_lock, x,
x - RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeeded %p -> %p",
__func__, rw, (void *)x,
(void *)(x - RW_ONE_READER));
break;
}
continue;
}
/*
* If there aren't any waiters for a write lock, then try
* to drop it quickly.
*/
if (!(x & RW_LOCK_WAITERS)) {
MPASS((x & ~RW_LOCK_WRITE_SPINNER) ==
RW_READERS_LOCK(1));
if (atomic_cmpset_rel_ptr(&rw->rw_lock, x,
RW_UNLOCKED)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p last succeeded",
__func__, rw);
break;
}
continue;
}
/*
* Ok, we know we have waiters and we think we are the
* last reader, so grab the turnstile lock.
*/
turnstile_chain_lock(&rw->lock_object);
v = rw->rw_lock & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
MPASS(v & RW_LOCK_WAITERS);
/*
* Try to drop our lock leaving the lock in a unlocked
* state.
*
* If you wanted to do explicit lock handoff you'd have to
* do it here. You'd also want to use turnstile_signal()
* and you'd have to handle the race where a higher
* priority thread blocks on the write lock before the
* thread you wakeup actually runs and have the new thread
* "steal" the lock. For now it's a lot simpler to just
* wakeup all of the waiters.
*
* As above, if we fail, then another thread might have
* acquired a read lock, so drop the turnstile lock and
* restart.
*/
x = RW_UNLOCKED;
if (v & RW_LOCK_WRITE_WAITERS) {
queue = TS_EXCLUSIVE_QUEUE;
x |= (v & RW_LOCK_READ_WAITERS);
} else
queue = TS_SHARED_QUEUE;
if (!atomic_cmpset_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v,
x)) {
turnstile_chain_unlock(&rw->lock_object);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p last succeeded with waiters",
__func__, rw);
/*
* Ok. The lock is released and all that's left is to
* wake up the waiters. Note that the lock might not be
* free anymore, but in that case the writers will just
* block again if they run before the new lock holder(s)
* release the lock.
*/
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
turnstile_broadcast(ts, queue);
turnstile_unpend(ts, TS_SHARED_LOCK);
turnstile_chain_unlock(&rw->lock_object);
break;
}
LOCKSTAT_PROFILE_RELEASE_LOCK(LS_RW_RUNLOCK_RELEASE, rw);
}
/*
* This function is called when we are unable to obtain a write lock on the
* first try. This means that at least one other thread holds either a
* read or write lock.
*/
void
_rw_wlock_hard(struct rwlock *rw, uintptr_t tid, const char *file, int line)
{
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
uintptr_t v, x;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
#ifdef KDTRACE_HOOKS
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
#endif
if (rw_wlocked(rw)) {
KASSERT(rw->lock_object.lo_flags & LO_RECURSABLE,
("%s: recursing but non-recursive rw %s @ %s:%d\n",
__func__, rw->lock_object.lo_name, file, line));
rw->rw_recurse++;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p recursing", __func__, rw);
return;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
rw->lock_object.lo_name, (void *)rw->rw_lock, file, line);
while (!_rw_write_lock(rw, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the lock is write locked and the owner is
* running on another CPU, spin until the owner stops
* running or the state of the lock changes.
*/
v = rw->rw_lock;
owner = (struct thread *)RW_OWNER(v);
if (!(v & RW_LOCK_READ) && TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK, "%s: spinning on %p held by %p",
__func__, rw, owner);
while ((struct thread*)RW_OWNER(rw->rw_lock) == owner &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
continue;
}
if ((v & RW_LOCK_READ) && RW_READERS(v) &&
spintries < rowner_retries) {
if (!(v & RW_LOCK_WRITE_SPINNER)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_SPINNER)) {
continue;
}
}
spintries++;
for (i = 0; i < rowner_loops; i++) {
if ((rw->rw_lock & RW_LOCK_WRITE_SPINNER) == 0)
break;
cpu_spinwait();
}
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
if (i != rowner_loops)
continue;
}
#endif
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if (!(v & RW_LOCK_READ)) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* Check for the waiters flags about this rwlock.
* If the lock was released, without maintain any pending
* waiters queue, simply try to acquire it.
* If a pending waiters queue is present, claim the lock
* ownership and maintain the pending queue.
*/
x = v & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
if ((v & ~x) == RW_UNLOCKED) {
x &= ~RW_LOCK_WRITE_SPINNER;
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid | x)) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
continue;
}
/*
* If the RW_LOCK_WRITE_WAITERS flag isn't set, then try to
* set it. If we fail to set it, then loop back and try
* again.
*/
if (!(v & RW_LOCK_WRITE_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set write waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the write waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs();
#endif
turnstile_wait(ts, rw_owner(rw), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs();
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
#ifdef ADAPTIVE_RWLOCKS
spintries = 0;
#endif
}
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_RW_WLOCK_ACQUIRE, rw, contested,
waittime, file, line);
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(LS_RW_WLOCK_BLOCK, rw, sleep_time);
/*
* Record only the loops spinning and not sleeping.
*/
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(LS_RW_WLOCK_SPIN, rw, (spin_cnt - sleep_cnt));
#endif
}
/*
* This function is called if the first try at releasing a write lock failed.
* This means that one of the 2 waiter bits must be set indicating that at
* least one thread is waiting on this lock.
*/
void
_rw_wunlock_hard(struct rwlock *rw, uintptr_t tid, const char *file, int line)
{
struct turnstile *ts;
uintptr_t v;
int queue;
if (rw_wlocked(rw) && rw_recursed(rw)) {
rw->rw_recurse--;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p unrecursing", __func__, rw);
return;
}
KASSERT(rw->rw_lock & (RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS),
("%s: neither of the waiter flags are set", __func__));
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p contested", __func__, rw);
turnstile_chain_lock(&rw->lock_object);
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
/*
* Use the same algo as sx locks for now. Prefer waking up shared
* waiters if we have any over writers. This is probably not ideal.
*
* 'v' is the value we are going to write back to rw_lock. If we
* have waiters on both queues, we need to preserve the state of
* the waiter flag for the queue we don't wake up. For now this is
* hardcoded for the algorithm mentioned above.
*
* In the case of both readers and writers waiting we wakeup the
* readers but leave the RW_LOCK_WRITE_WAITERS flag set. If a
* new writer comes in before a reader it will claim the lock up
* above. There is probably a potential priority inversion in
* there that could be worked around either by waking both queues
* of waiters or doing some complicated lock handoff gymnastics.
*/
v = RW_UNLOCKED;
if (rw->rw_lock & RW_LOCK_WRITE_WAITERS) {
queue = TS_EXCLUSIVE_QUEUE;
v |= (rw->rw_lock & RW_LOCK_READ_WAITERS);
} else
queue = TS_SHARED_QUEUE;
/* Wake up all waiters for the specific queue. */
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK, "%s: %p waking up %s waiters", __func__, rw,
queue == TS_SHARED_QUEUE ? "read" : "write");
turnstile_broadcast(ts, queue);
atomic_store_rel_ptr(&rw->rw_lock, v);
turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
turnstile_chain_unlock(&rw->lock_object);
}
/*
* Attempt to do a non-blocking upgrade from a read lock to a write
* lock. This will only succeed if this thread holds a single read
* lock. Returns true if the upgrade succeeded and false otherwise.
*/
int
_rw_try_upgrade(struct rwlock *rw, const char *file, int line)
{
uintptr_t v, x, tid;
struct turnstile *ts;
int success;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_try_upgrade() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_RLOCKED, file, line);
/*
* Attempt to switch from one reader to a writer. If there
* are any write waiters, then we will have to lock the
* turnstile first to prevent races with another writer
* calling turnstile_wait() before we have claimed this
* turnstile. So, do the simple case of no waiters first.
*/
tid = (uintptr_t)curthread;
success = 0;
for (;;) {
v = rw->rw_lock;
if (RW_READERS(v) > 1)
break;
if (!(v & RW_LOCK_WAITERS)) {
success = atomic_cmpset_ptr(&rw->rw_lock, v, tid);
if (!success)
continue;
break;
}
/*
* Ok, we think we have waiters, so lock the turnstile.
*/
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
if (RW_READERS(v) > 1) {
turnstile_cancel(ts);
break;
}
/*
* Try to switch from one reader to a writer again. This time
* we honor the current state of the waiters flags.
* If we obtain the lock with the flags set, then claim
* ownership of the turnstile.
*/
x = rw->rw_lock & RW_LOCK_WAITERS;
success = atomic_cmpset_ptr(&rw->rw_lock, v, tid | x);
if (success) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
}
LOCK_LOG_TRY("WUPGRADE", &rw->lock_object, 0, success, file, line);
if (success) {
curthread->td_rw_rlocks--;
WITNESS_UPGRADE(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
file, line);
LOCKSTAT_RECORD0(LS_RW_TRYUPGRADE_UPGRADE, rw);
}
return (success);
}
/*
* Downgrade a write lock into a single read lock.
*/
void
_rw_downgrade(struct rwlock *rw, const char *file, int line)
{
struct turnstile *ts;
uintptr_t tid, v;
int rwait, wwait;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_downgrade() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_WLOCKED | RA_NOTRECURSED, file, line);
#ifndef INVARIANTS
if (rw_recursed(rw))
panic("downgrade of a recursed lock");
#endif
WITNESS_DOWNGRADE(&rw->lock_object, 0, file, line);
/*
* Convert from a writer to a single reader. First we handle
* the easy case with no waiters. If there are any waiters, we
* lock the turnstile and "disown" the lock.
*/
tid = (uintptr_t)curthread;
if (atomic_cmpset_rel_ptr(&rw->rw_lock, tid, RW_READERS_LOCK(1)))
goto out;
/*
* Ok, we think we have waiters, so lock the turnstile so we can
* read the waiter flags without any races.
*/
turnstile_chain_lock(&rw->lock_object);
v = rw->rw_lock & RW_LOCK_WAITERS;
rwait = v & RW_LOCK_READ_WAITERS;
wwait = v & RW_LOCK_WRITE_WAITERS;
MPASS(rwait | wwait);
/*
* Downgrade from a write lock while preserving waiters flag
* and give up ownership of the turnstile.
*/
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
if (!wwait)
v &= ~RW_LOCK_READ_WAITERS;
atomic_store_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v);
/*
* Wake other readers if there are no writers pending. Otherwise they
* won't be able to acquire the lock anyway.
*/
if (rwait && !wwait) {
turnstile_broadcast(ts, TS_SHARED_QUEUE);
turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
} else
turnstile_disown(ts);
turnstile_chain_unlock(&rw->lock_object);
out:
curthread->td_rw_rlocks++;
LOCK_LOG_LOCK("WDOWNGRADE", &rw->lock_object, 0, 0, file, line);
LOCKSTAT_RECORD0(LS_RW_DOWNGRADE_DOWNGRADE, rw);
}
#ifdef INVARIANT_SUPPORT
#ifndef INVARIANTS
#undef _rw_assert
#endif
/*
* In the non-WITNESS case, rw_assert() can only detect that at least
* *some* thread owns an rlock, but it cannot guarantee that *this*
* thread owns an rlock.
*/
void
_rw_assert(struct rwlock *rw, int what, const char *file, int line)
{
if (panicstr != NULL)
return;
switch (what) {
case RA_LOCKED:
case RA_LOCKED | RA_RECURSED:
case RA_LOCKED | RA_NOTRECURSED:
case RA_RLOCKED:
#ifdef WITNESS
witness_assert(&rw->lock_object, what, file, line);
#else
/*
* If some other thread has a write lock or we have one
* and are asserting a read lock, fail. Also, if no one
* has a lock at all, fail.
*/
if (rw->rw_lock == RW_UNLOCKED ||
(!(rw->rw_lock & RW_LOCK_READ) && (what == RA_RLOCKED ||
rw_wowner(rw) != curthread)))
panic("Lock %s not %slocked @ %s:%d\n",
rw->lock_object.lo_name, (what == RA_RLOCKED) ?
"read " : "", file, line);
if (!(rw->rw_lock & RW_LOCK_READ)) {
if (rw_recursed(rw)) {
if (what & RA_NOTRECURSED)
panic("Lock %s recursed @ %s:%d\n",
rw->lock_object.lo_name, file,
line);
} else if (what & RA_RECURSED)
panic("Lock %s not recursed @ %s:%d\n",
rw->lock_object.lo_name, file, line);
}
#endif
break;
case RA_WLOCKED:
case RA_WLOCKED | RA_RECURSED:
case RA_WLOCKED | RA_NOTRECURSED:
if (rw_wowner(rw) != curthread)
panic("Lock %s not exclusively locked @ %s:%d\n",
rw->lock_object.lo_name, file, line);
if (rw_recursed(rw)) {
if (what & RA_NOTRECURSED)
panic("Lock %s recursed @ %s:%d\n",
rw->lock_object.lo_name, file, line);
} else if (what & RA_RECURSED)
panic("Lock %s not recursed @ %s:%d\n",
rw->lock_object.lo_name, file, line);
break;
case RA_UNLOCKED:
#ifdef WITNESS
witness_assert(&rw->lock_object, what, file, line);
#else
/*
* If we hold a write lock fail. We can't reliably check
* to see if we hold a read lock or not.
*/
if (rw_wowner(rw) == curthread)
panic("Lock %s exclusively locked @ %s:%d\n",
rw->lock_object.lo_name, file, line);
#endif
break;
default:
panic("Unknown rw lock assertion: %d @ %s:%d", what, file,
line);
}
}
#endif /* INVARIANT_SUPPORT */
#ifdef DDB
void
db_show_rwlock(struct lock_object *lock)
{
struct rwlock *rw;
struct thread *td;
rw = (struct rwlock *)lock;
db_printf(" state: ");
if (rw->rw_lock == RW_UNLOCKED)
db_printf("UNLOCKED\n");
else if (rw->rw_lock == RW_DESTROYED) {
db_printf("DESTROYED\n");
return;
} else if (rw->rw_lock & RW_LOCK_READ)
db_printf("RLOCK: %ju locks\n",
(uintmax_t)(RW_READERS(rw->rw_lock)));
else {
td = rw_wowner(rw);
db_printf("WLOCK: %p (tid %d, pid %d, \"%s\")\n", td,
td->td_tid, td->td_proc->p_pid, td->td_name);
if (rw_recursed(rw))
db_printf(" recursed: %u\n", rw->rw_recurse);
}
db_printf(" waiters: ");
switch (rw->rw_lock & (RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS)) {
case RW_LOCK_READ_WAITERS:
db_printf("readers\n");
break;
case RW_LOCK_WRITE_WAITERS:
db_printf("writers\n");
break;
case RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS:
db_printf("readers and writers\n");
break;
default:
db_printf("none\n");
break;
}
}
#endif
|