1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2007 Attilio Rao <attilio@freebsd.org>
5 * Copyright (c) 2001 Jason Evans <jasone@freebsd.org>
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice(s), this list of conditions and the following disclaimer as
13 * the first lines of this file unmodified other than the possible
14 * addition of one or more copyright notices.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice(s), this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY
20 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
21 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY
23 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
24 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
25 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH
29 * DAMAGE.
30 */
31
32 /*
33 * Shared/exclusive locks. This implementation attempts to ensure
34 * deterministic lock granting behavior, so that slocks and xlocks are
35 * interleaved.
36 *
37 * Priority propagation will not generally raise the priority of lock holders,
38 * so should not be relied upon in combination with sx locks.
39 */
40
41 #include "opt_ddb.h"
42 #include "opt_hwpmc_hooks.h"
43 #include "opt_no_adaptive_sx.h"
44
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/kdb.h>
48 #include <sys/kernel.h>
49 #include <sys/ktr.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/sched.h>
54 #include <sys/sleepqueue.h>
55 #include <sys/sx.h>
56 #include <sys/smp.h>
57 #include <sys/sysctl.h>
58
59 #if defined(SMP) && !defined(NO_ADAPTIVE_SX)
60 #include <machine/cpu.h>
61 #endif
62
63 #ifdef DDB
64 #include <ddb/ddb.h>
65 #endif
66
67 #if defined(SMP) && !defined(NO_ADAPTIVE_SX)
68 #define ADAPTIVE_SX
69 #endif
70
71 #ifdef HWPMC_HOOKS
72 #include <sys/pmckern.h>
73 PMC_SOFT_DECLARE( , , lock, failed);
74 #endif
75
76 /* Handy macros for sleep queues. */
77 #define SQ_EXCLUSIVE_QUEUE 0
78 #define SQ_SHARED_QUEUE 1
79
80 /*
81 * Variations on DROP_GIANT()/PICKUP_GIANT() for use in this file. We
82 * drop Giant anytime we have to sleep or if we adaptively spin.
83 */
84 #define GIANT_DECLARE \
85 int _giantcnt = 0; \
86 WITNESS_SAVE_DECL(Giant) \
87
88 #define GIANT_SAVE(work) do { \
89 if (__predict_false(mtx_owned(&Giant))) { \
90 work++; \
91 WITNESS_SAVE(&Giant.lock_object, Giant); \
92 while (mtx_owned(&Giant)) { \
93 _giantcnt++; \
94 mtx_unlock(&Giant); \
95 } \
96 } \
97 } while (0)
98
99 #define GIANT_RESTORE() do { \
100 if (_giantcnt > 0) { \
101 mtx_assert(&Giant, MA_NOTOWNED); \
102 while (_giantcnt--) \
103 mtx_lock(&Giant); \
104 WITNESS_RESTORE(&Giant.lock_object, Giant); \
105 } \
106 } while (0)
107
108 /*
109 * Returns true if an exclusive lock is recursed. It assumes
110 * curthread currently has an exclusive lock.
111 */
112 #define sx_recursed(sx) ((sx)->sx_recurse != 0)
113
114 static void assert_sx(const struct lock_object *lock, int what);
115 #ifdef DDB
116 static void db_show_sx(const struct lock_object *lock);
117 #endif
118 static void lock_sx(struct lock_object *lock, uintptr_t how);
119 #ifdef KDTRACE_HOOKS
120 static int owner_sx(const struct lock_object *lock, struct thread **owner);
121 #endif
122 static uintptr_t unlock_sx(struct lock_object *lock);
123
124 struct lock_class lock_class_sx = {
125 .lc_name = "sx",
126 .lc_flags = LC_SLEEPLOCK | LC_SLEEPABLE | LC_RECURSABLE | LC_UPGRADABLE,
127 .lc_assert = assert_sx,
128 #ifdef DDB
129 .lc_ddb_show = db_show_sx,
130 #endif
131 .lc_lock = lock_sx,
132 .lc_unlock = unlock_sx,
133 #ifdef KDTRACE_HOOKS
134 .lc_owner = owner_sx,
135 #endif
136 };
137
138 #ifndef INVARIANTS
139 #define _sx_assert(sx, what, file, line)
140 #endif
141
142 #ifdef ADAPTIVE_SX
143 #ifdef SX_CUSTOM_BACKOFF
144 static u_short __read_frequently asx_retries;
145 static u_short __read_frequently asx_loops;
146 static SYSCTL_NODE(_debug, OID_AUTO, sx, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
147 "sxlock debugging");
148 SYSCTL_U16(_debug_sx, OID_AUTO, retries, CTLFLAG_RW, &asx_retries, 0, "");
149 SYSCTL_U16(_debug_sx, OID_AUTO, loops, CTLFLAG_RW, &asx_loops, 0, "");
150
151 static struct lock_delay_config __read_frequently sx_delay;
152
153 SYSCTL_U16(_debug_sx, OID_AUTO, delay_base, CTLFLAG_RW, &sx_delay.base,
154 0, "");
155 SYSCTL_U16(_debug_sx, OID_AUTO, delay_max, CTLFLAG_RW, &sx_delay.max,
156 0, "");
157
158 static void
sx_lock_delay_init(void * arg __unused)159 sx_lock_delay_init(void *arg __unused)
160 {
161
162 lock_delay_default_init(&sx_delay);
163 asx_retries = 10;
164 asx_loops = max(10000, sx_delay.max);
165 }
166 LOCK_DELAY_SYSINIT(sx_lock_delay_init);
167 #else
168 #define sx_delay locks_delay
169 #define asx_retries locks_delay_retries
170 #define asx_loops locks_delay_loops
171 #endif
172 #endif
173
174 void
assert_sx(const struct lock_object * lock,int what)175 assert_sx(const struct lock_object *lock, int what)
176 {
177
178 sx_assert((const struct sx *)lock, what);
179 }
180
181 void
lock_sx(struct lock_object * lock,uintptr_t how)182 lock_sx(struct lock_object *lock, uintptr_t how)
183 {
184 struct sx *sx;
185
186 sx = (struct sx *)lock;
187 if (how)
188 sx_slock(sx);
189 else
190 sx_xlock(sx);
191 }
192
193 uintptr_t
unlock_sx(struct lock_object * lock)194 unlock_sx(struct lock_object *lock)
195 {
196 struct sx *sx;
197
198 sx = (struct sx *)lock;
199 sx_assert(sx, SA_LOCKED | SA_NOTRECURSED);
200 if (sx_xlocked(sx)) {
201 sx_xunlock(sx);
202 return (0);
203 } else {
204 sx_sunlock(sx);
205 return (1);
206 }
207 }
208
209 #ifdef KDTRACE_HOOKS
210 int
owner_sx(const struct lock_object * lock,struct thread ** owner)211 owner_sx(const struct lock_object *lock, struct thread **owner)
212 {
213 const struct sx *sx;
214 uintptr_t x;
215
216 sx = (const struct sx *)lock;
217 x = sx->sx_lock;
218 *owner = NULL;
219 return ((x & SX_LOCK_SHARED) != 0 ? (SX_SHARERS(x) != 0) :
220 ((*owner = (struct thread *)SX_OWNER(x)) != NULL));
221 }
222 #endif
223
224 void
sx_sysinit(void * arg)225 sx_sysinit(void *arg)
226 {
227 struct sx_args *sargs = arg;
228
229 sx_init_flags(sargs->sa_sx, sargs->sa_desc, sargs->sa_flags);
230 }
231
232 void
sx_init_flags(struct sx * sx,const char * description,int opts)233 sx_init_flags(struct sx *sx, const char *description, int opts)
234 {
235 int flags;
236
237 MPASS((opts & ~(SX_QUIET | SX_RECURSE | SX_NOWITNESS | SX_DUPOK |
238 SX_NOPROFILE | SX_NEW)) == 0);
239 ASSERT_ATOMIC_LOAD_PTR(sx->sx_lock,
240 ("%s: sx_lock not aligned for %s: %p", __func__, description,
241 &sx->sx_lock));
242
243 flags = LO_SLEEPABLE | LO_UPGRADABLE;
244 if (opts & SX_DUPOK)
245 flags |= LO_DUPOK;
246 if (opts & SX_NOPROFILE)
247 flags |= LO_NOPROFILE;
248 if (!(opts & SX_NOWITNESS))
249 flags |= LO_WITNESS;
250 if (opts & SX_RECURSE)
251 flags |= LO_RECURSABLE;
252 if (opts & SX_QUIET)
253 flags |= LO_QUIET;
254 if (opts & SX_NEW)
255 flags |= LO_NEW;
256
257 lock_init(&sx->lock_object, &lock_class_sx, description, NULL, flags);
258 sx->sx_lock = SX_LOCK_UNLOCKED;
259 sx->sx_recurse = 0;
260 }
261
262 void
sx_destroy(struct sx * sx)263 sx_destroy(struct sx *sx)
264 {
265
266 KASSERT(sx->sx_lock == SX_LOCK_UNLOCKED, ("sx lock still held"));
267 KASSERT(sx->sx_recurse == 0, ("sx lock still recursed"));
268 sx->sx_lock = SX_LOCK_DESTROYED;
269 lock_destroy(&sx->lock_object);
270 }
271
272 int
sx_try_slock_int(struct sx * sx LOCK_FILE_LINE_ARG_DEF)273 sx_try_slock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF)
274 {
275 uintptr_t x;
276
277 if (SCHEDULER_STOPPED())
278 return (1);
279
280 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
281 ("sx_try_slock() by idle thread %p on sx %s @ %s:%d",
282 curthread, sx->lock_object.lo_name, file, line));
283
284 x = sx->sx_lock;
285 for (;;) {
286 KASSERT(x != SX_LOCK_DESTROYED,
287 ("sx_try_slock() of destroyed sx @ %s:%d", file, line));
288 if (!(x & SX_LOCK_SHARED))
289 break;
290 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, x + SX_ONE_SHARER)) {
291 LOCK_LOG_TRY("SLOCK", &sx->lock_object, 0, 1, file, line);
292 WITNESS_LOCK(&sx->lock_object, LOP_TRYLOCK, file, line);
293 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire,
294 sx, 0, 0, file, line, LOCKSTAT_READER);
295 TD_LOCKS_INC(curthread);
296 curthread->td_sx_slocks++;
297 return (1);
298 }
299 }
300
301 LOCK_LOG_TRY("SLOCK", &sx->lock_object, 0, 0, file, line);
302 return (0);
303 }
304
305 int
sx_try_slock_(struct sx * sx,const char * file,int line)306 sx_try_slock_(struct sx *sx, const char *file, int line)
307 {
308
309 return (sx_try_slock_int(sx LOCK_FILE_LINE_ARG));
310 }
311
312 int
_sx_xlock(struct sx * sx,int opts,const char * file,int line)313 _sx_xlock(struct sx *sx, int opts, const char *file, int line)
314 {
315 uintptr_t tid, x;
316 int error = 0;
317
318 KASSERT(kdb_active != 0 || SCHEDULER_STOPPED() ||
319 !TD_IS_IDLETHREAD(curthread),
320 ("sx_xlock() by idle thread %p on sx %s @ %s:%d",
321 curthread, sx->lock_object.lo_name, file, line));
322 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
323 ("sx_xlock() of destroyed sx @ %s:%d", file, line));
324 WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file,
325 line, NULL);
326 tid = (uintptr_t)curthread;
327 x = SX_LOCK_UNLOCKED;
328 if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
329 error = _sx_xlock_hard(sx, x, opts LOCK_FILE_LINE_ARG);
330 else
331 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
332 0, 0, file, line, LOCKSTAT_WRITER);
333 if (!error) {
334 LOCK_LOG_LOCK("XLOCK", &sx->lock_object, 0, sx->sx_recurse,
335 file, line);
336 WITNESS_LOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line);
337 TD_LOCKS_INC(curthread);
338 }
339
340 return (error);
341 }
342
343 int
sx_try_xlock_int(struct sx * sx LOCK_FILE_LINE_ARG_DEF)344 sx_try_xlock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF)
345 {
346 struct thread *td;
347 uintptr_t tid, x;
348 int rval;
349 bool recursed;
350
351 td = curthread;
352 tid = (uintptr_t)td;
353 if (SCHEDULER_STOPPED())
354 return (1);
355
356 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(td),
357 ("sx_try_xlock() by idle thread %p on sx %s @ %s:%d",
358 curthread, sx->lock_object.lo_name, file, line));
359 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
360 ("sx_try_xlock() of destroyed sx @ %s:%d", file, line));
361
362 rval = 1;
363 recursed = false;
364 x = SX_LOCK_UNLOCKED;
365 for (;;) {
366 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
367 break;
368 if (x == SX_LOCK_UNLOCKED)
369 continue;
370 if (x == tid && (sx->lock_object.lo_flags & LO_RECURSABLE)) {
371 sx->sx_recurse++;
372 atomic_set_ptr(&sx->sx_lock, SX_LOCK_RECURSED);
373 break;
374 }
375 rval = 0;
376 break;
377 }
378
379 LOCK_LOG_TRY("XLOCK", &sx->lock_object, 0, rval, file, line);
380 if (rval) {
381 WITNESS_LOCK(&sx->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
382 file, line);
383 if (!recursed)
384 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire,
385 sx, 0, 0, file, line, LOCKSTAT_WRITER);
386 TD_LOCKS_INC(curthread);
387 }
388
389 return (rval);
390 }
391
392 int
sx_try_xlock_(struct sx * sx,const char * file,int line)393 sx_try_xlock_(struct sx *sx, const char *file, int line)
394 {
395
396 return (sx_try_xlock_int(sx LOCK_FILE_LINE_ARG));
397 }
398
399 void
_sx_xunlock(struct sx * sx,const char * file,int line)400 _sx_xunlock(struct sx *sx, const char *file, int line)
401 {
402
403 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
404 ("sx_xunlock() of destroyed sx @ %s:%d", file, line));
405 _sx_assert(sx, SA_XLOCKED, file, line);
406 WITNESS_UNLOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line);
407 LOCK_LOG_LOCK("XUNLOCK", &sx->lock_object, 0, sx->sx_recurse, file,
408 line);
409 #if LOCK_DEBUG > 0
410 _sx_xunlock_hard(sx, (uintptr_t)curthread, file, line);
411 #else
412 __sx_xunlock(sx, curthread, file, line);
413 #endif
414 TD_LOCKS_DEC(curthread);
415 }
416
417 /*
418 * Try to do a non-blocking upgrade from a shared lock to an exclusive lock.
419 * This will only succeed if this thread holds a single shared lock.
420 * Return 1 if if the upgrade succeed, 0 otherwise.
421 */
422 int
sx_try_upgrade_int(struct sx * sx LOCK_FILE_LINE_ARG_DEF)423 sx_try_upgrade_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF)
424 {
425 uintptr_t x;
426 uintptr_t waiters;
427 int success;
428
429 if (SCHEDULER_STOPPED())
430 return (1);
431
432 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
433 ("sx_try_upgrade() of destroyed sx @ %s:%d", file, line));
434 _sx_assert(sx, SA_SLOCKED, file, line);
435
436 /*
437 * Try to switch from one shared lock to an exclusive lock. We need
438 * to maintain the SX_LOCK_EXCLUSIVE_WAITERS flag if set so that
439 * we will wake up the exclusive waiters when we drop the lock.
440 */
441 success = 0;
442 x = SX_READ_VALUE(sx);
443 for (;;) {
444 if (SX_SHARERS(x) > 1)
445 break;
446 waiters = (x & SX_LOCK_WAITERS);
447 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x,
448 (uintptr_t)curthread | waiters)) {
449 success = 1;
450 break;
451 }
452 }
453 LOCK_LOG_TRY("XUPGRADE", &sx->lock_object, 0, success, file, line);
454 if (success) {
455 curthread->td_sx_slocks--;
456 WITNESS_UPGRADE(&sx->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
457 file, line);
458 LOCKSTAT_RECORD0(sx__upgrade, sx);
459 }
460 return (success);
461 }
462
463 int
sx_try_upgrade_(struct sx * sx,const char * file,int line)464 sx_try_upgrade_(struct sx *sx, const char *file, int line)
465 {
466
467 return (sx_try_upgrade_int(sx LOCK_FILE_LINE_ARG));
468 }
469
470 /*
471 * Downgrade an unrecursed exclusive lock into a single shared lock.
472 */
473 void
sx_downgrade_int(struct sx * sx LOCK_FILE_LINE_ARG_DEF)474 sx_downgrade_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF)
475 {
476 uintptr_t x;
477 int wakeup_swapper;
478
479 if (SCHEDULER_STOPPED())
480 return;
481
482 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
483 ("sx_downgrade() of destroyed sx @ %s:%d", file, line));
484 _sx_assert(sx, SA_XLOCKED | SA_NOTRECURSED, file, line);
485 #ifndef INVARIANTS
486 if (sx_recursed(sx))
487 panic("downgrade of a recursed lock");
488 #endif
489
490 WITNESS_DOWNGRADE(&sx->lock_object, 0, file, line);
491
492 /*
493 * Try to switch from an exclusive lock with no shared waiters
494 * to one sharer with no shared waiters. If there are
495 * exclusive waiters, we don't need to lock the sleep queue so
496 * long as we preserve the flag. We do one quick try and if
497 * that fails we grab the sleepq lock to keep the flags from
498 * changing and do it the slow way.
499 *
500 * We have to lock the sleep queue if there are shared waiters
501 * so we can wake them up.
502 */
503 x = sx->sx_lock;
504 if (!(x & SX_LOCK_SHARED_WAITERS) &&
505 atomic_cmpset_rel_ptr(&sx->sx_lock, x, SX_SHARERS_LOCK(1) |
506 (x & SX_LOCK_EXCLUSIVE_WAITERS)))
507 goto out;
508
509 /*
510 * Lock the sleep queue so we can read the waiters bits
511 * without any races and wakeup any shared waiters.
512 */
513 sleepq_lock(&sx->lock_object);
514
515 /*
516 * Preserve SX_LOCK_EXCLUSIVE_WAITERS while downgraded to a single
517 * shared lock. If there are any shared waiters, wake them up.
518 */
519 wakeup_swapper = 0;
520 x = sx->sx_lock;
521 atomic_store_rel_ptr(&sx->sx_lock, SX_SHARERS_LOCK(1) |
522 (x & SX_LOCK_EXCLUSIVE_WAITERS));
523 if (x & SX_LOCK_SHARED_WAITERS)
524 wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX,
525 0, SQ_SHARED_QUEUE);
526 sleepq_release(&sx->lock_object);
527
528 if (wakeup_swapper)
529 kick_proc0();
530
531 out:
532 curthread->td_sx_slocks++;
533 LOCK_LOG_LOCK("XDOWNGRADE", &sx->lock_object, 0, 0, file, line);
534 LOCKSTAT_RECORD0(sx__downgrade, sx);
535 }
536
537 void
sx_downgrade_(struct sx * sx,const char * file,int line)538 sx_downgrade_(struct sx *sx, const char *file, int line)
539 {
540
541 sx_downgrade_int(sx LOCK_FILE_LINE_ARG);
542 }
543
544 #ifdef ADAPTIVE_SX
545 static inline void
sx_drop_critical(uintptr_t x,bool * in_critical,int * extra_work)546 sx_drop_critical(uintptr_t x, bool *in_critical, int *extra_work)
547 {
548
549 if (x & SX_LOCK_WRITE_SPINNER)
550 return;
551 if (*in_critical) {
552 critical_exit();
553 *in_critical = false;
554 (*extra_work)--;
555 }
556 }
557 #else
558 #define sx_drop_critical(x, in_critical, extra_work) do { } while (0)
559 #endif
560
561 /*
562 * This function represents the so-called 'hard case' for sx_xlock
563 * operation. All 'easy case' failures are redirected to this. Note
564 * that ideally this would be a static function, but it needs to be
565 * accessible from at least sx.h.
566 */
567 int
_sx_xlock_hard(struct sx * sx,uintptr_t x,int opts LOCK_FILE_LINE_ARG_DEF)568 _sx_xlock_hard(struct sx *sx, uintptr_t x, int opts LOCK_FILE_LINE_ARG_DEF)
569 {
570 GIANT_DECLARE;
571 uintptr_t tid, setx;
572 #ifdef ADAPTIVE_SX
573 struct thread *owner;
574 u_int i, n, spintries = 0;
575 enum { READERS, WRITER } sleep_reason = READERS;
576 bool in_critical = false;
577 #endif
578 #ifdef LOCK_PROFILING
579 uint64_t waittime = 0;
580 int contested = 0;
581 #endif
582 int error = 0;
583 #if defined(ADAPTIVE_SX) || defined(KDTRACE_HOOKS)
584 struct lock_delay_arg lda;
585 #endif
586 #ifdef KDTRACE_HOOKS
587 u_int sleep_cnt = 0;
588 int64_t sleep_time = 0;
589 int64_t all_time = 0;
590 #endif
591 #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
592 uintptr_t state = 0;
593 int doing_lockprof = 0;
594 #endif
595 int extra_work = 0;
596
597 tid = (uintptr_t)curthread;
598
599 #ifdef KDTRACE_HOOKS
600 if (LOCKSTAT_PROFILE_ENABLED(sx__acquire)) {
601 while (x == SX_LOCK_UNLOCKED) {
602 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
603 goto out_lockstat;
604 }
605 extra_work = 1;
606 doing_lockprof = 1;
607 all_time -= lockstat_nsecs(&sx->lock_object);
608 state = x;
609 }
610 #endif
611 #ifdef LOCK_PROFILING
612 extra_work = 1;
613 doing_lockprof = 1;
614 state = x;
615 #endif
616
617 if (SCHEDULER_STOPPED())
618 return (0);
619
620 if (__predict_false(x == SX_LOCK_UNLOCKED))
621 x = SX_READ_VALUE(sx);
622
623 /* If we already hold an exclusive lock, then recurse. */
624 if (__predict_false(lv_sx_owner(x) == (struct thread *)tid)) {
625 KASSERT((sx->lock_object.lo_flags & LO_RECURSABLE) != 0,
626 ("_sx_xlock_hard: recursed on non-recursive sx %s @ %s:%d\n",
627 sx->lock_object.lo_name, file, line));
628 sx->sx_recurse++;
629 atomic_set_ptr(&sx->sx_lock, SX_LOCK_RECURSED);
630 if (LOCK_LOG_TEST(&sx->lock_object, 0))
631 CTR2(KTR_LOCK, "%s: %p recursing", __func__, sx);
632 return (0);
633 }
634
635 if (LOCK_LOG_TEST(&sx->lock_object, 0))
636 CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
637 sx->lock_object.lo_name, (void *)sx->sx_lock, file, line);
638
639 #if defined(ADAPTIVE_SX)
640 lock_delay_arg_init(&lda, &sx_delay);
641 #elif defined(KDTRACE_HOOKS)
642 lock_delay_arg_init_noadapt(&lda);
643 #endif
644
645 #ifdef HWPMC_HOOKS
646 PMC_SOFT_CALL( , , lock, failed);
647 #endif
648 lock_profile_obtain_lock_failed(&sx->lock_object, false, &contested,
649 &waittime);
650
651 #ifndef INVARIANTS
652 GIANT_SAVE(extra_work);
653 #endif
654
655 THREAD_CONTENDS_ON_LOCK(&sx->lock_object);
656
657 for (;;) {
658 if (x == SX_LOCK_UNLOCKED) {
659 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
660 break;
661 continue;
662 }
663 #ifdef INVARIANTS
664 GIANT_SAVE(extra_work);
665 #endif
666 #ifdef KDTRACE_HOOKS
667 lda.spin_cnt++;
668 #endif
669 #ifdef ADAPTIVE_SX
670 if (x == (SX_LOCK_SHARED | SX_LOCK_WRITE_SPINNER)) {
671 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid))
672 break;
673 continue;
674 }
675
676 /*
677 * If the lock is write locked and the owner is
678 * running on another CPU, spin until the owner stops
679 * running or the state of the lock changes.
680 */
681 if ((x & SX_LOCK_SHARED) == 0) {
682 sx_drop_critical(x, &in_critical, &extra_work);
683 sleep_reason = WRITER;
684 owner = lv_sx_owner(x);
685 if (!TD_IS_RUNNING(owner))
686 goto sleepq;
687 if (LOCK_LOG_TEST(&sx->lock_object, 0))
688 CTR3(KTR_LOCK, "%s: spinning on %p held by %p",
689 __func__, sx, owner);
690 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
691 "spinning", "lockname:\"%s\"",
692 sx->lock_object.lo_name);
693 do {
694 lock_delay(&lda);
695 x = SX_READ_VALUE(sx);
696 owner = lv_sx_owner(x);
697 } while (owner != NULL && TD_IS_RUNNING(owner));
698 KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
699 "running");
700 continue;
701 } else if (SX_SHARERS(x) > 0) {
702 sleep_reason = READERS;
703 if (spintries == asx_retries)
704 goto sleepq;
705 if (!(x & SX_LOCK_WRITE_SPINNER)) {
706 if (!in_critical) {
707 critical_enter();
708 in_critical = true;
709 extra_work++;
710 }
711 if (!atomic_fcmpset_ptr(&sx->sx_lock, &x,
712 x | SX_LOCK_WRITE_SPINNER)) {
713 critical_exit();
714 in_critical = false;
715 extra_work--;
716 continue;
717 }
718 }
719 spintries++;
720 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
721 "spinning", "lockname:\"%s\"",
722 sx->lock_object.lo_name);
723 n = SX_SHARERS(x);
724 for (i = 0; i < asx_loops; i += n) {
725 lock_delay_spin(n);
726 x = SX_READ_VALUE(sx);
727 if (!(x & SX_LOCK_WRITE_SPINNER))
728 break;
729 if (!(x & SX_LOCK_SHARED))
730 break;
731 n = SX_SHARERS(x);
732 if (n == 0)
733 break;
734 }
735 #ifdef KDTRACE_HOOKS
736 lda.spin_cnt += i;
737 #endif
738 KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
739 "running");
740 if (i < asx_loops)
741 continue;
742 }
743 sleepq:
744 #endif
745 sleepq_lock(&sx->lock_object);
746 x = SX_READ_VALUE(sx);
747 retry_sleepq:
748
749 /*
750 * If the lock was released while spinning on the
751 * sleep queue chain lock, try again.
752 */
753 if (x == SX_LOCK_UNLOCKED) {
754 sleepq_release(&sx->lock_object);
755 sx_drop_critical(x, &in_critical, &extra_work);
756 continue;
757 }
758
759 #ifdef ADAPTIVE_SX
760 /*
761 * The current lock owner might have started executing
762 * on another CPU (or the lock could have changed
763 * owners) while we were waiting on the sleep queue
764 * chain lock. If so, drop the sleep queue lock and try
765 * again.
766 */
767 if (!(x & SX_LOCK_SHARED)) {
768 owner = (struct thread *)SX_OWNER(x);
769 if (TD_IS_RUNNING(owner)) {
770 sleepq_release(&sx->lock_object);
771 sx_drop_critical(x, &in_critical,
772 &extra_work);
773 continue;
774 }
775 } else if (SX_SHARERS(x) > 0 && sleep_reason == WRITER) {
776 sleepq_release(&sx->lock_object);
777 sx_drop_critical(x, &in_critical, &extra_work);
778 continue;
779 }
780 #endif
781
782 /*
783 * If an exclusive lock was released with both shared
784 * and exclusive waiters and a shared waiter hasn't
785 * woken up and acquired the lock yet, sx_lock will be
786 * set to SX_LOCK_UNLOCKED | SX_LOCK_EXCLUSIVE_WAITERS.
787 * If we see that value, try to acquire it once. Note
788 * that we have to preserve SX_LOCK_EXCLUSIVE_WAITERS
789 * as there are other exclusive waiters still. If we
790 * fail, restart the loop.
791 */
792 setx = x & (SX_LOCK_WAITERS | SX_LOCK_WRITE_SPINNER);
793 if ((x & ~setx) == SX_LOCK_SHARED) {
794 setx &= ~SX_LOCK_WRITE_SPINNER;
795 if (!atomic_fcmpset_acq_ptr(&sx->sx_lock, &x, tid | setx))
796 goto retry_sleepq;
797 sleepq_release(&sx->lock_object);
798 CTR2(KTR_LOCK, "%s: %p claimed by new writer",
799 __func__, sx);
800 break;
801 }
802
803 #ifdef ADAPTIVE_SX
804 /*
805 * It is possible we set the SX_LOCK_WRITE_SPINNER bit.
806 * It is an invariant that when the bit is set, there is
807 * a writer ready to grab the lock. Thus clear the bit since
808 * we are going to sleep.
809 */
810 if (in_critical) {
811 if ((x & SX_LOCK_WRITE_SPINNER) ||
812 !((x & SX_LOCK_EXCLUSIVE_WAITERS))) {
813 setx = x & ~SX_LOCK_WRITE_SPINNER;
814 setx |= SX_LOCK_EXCLUSIVE_WAITERS;
815 if (!atomic_fcmpset_ptr(&sx->sx_lock, &x,
816 setx)) {
817 goto retry_sleepq;
818 }
819 }
820 critical_exit();
821 in_critical = false;
822 } else {
823 #endif
824 /*
825 * Try to set the SX_LOCK_EXCLUSIVE_WAITERS. If we fail,
826 * than loop back and retry.
827 */
828 if (!(x & SX_LOCK_EXCLUSIVE_WAITERS)) {
829 if (!atomic_fcmpset_ptr(&sx->sx_lock, &x,
830 x | SX_LOCK_EXCLUSIVE_WAITERS)) {
831 goto retry_sleepq;
832 }
833 if (LOCK_LOG_TEST(&sx->lock_object, 0))
834 CTR2(KTR_LOCK, "%s: %p set excl waiters flag",
835 __func__, sx);
836 }
837 #ifdef ADAPTIVE_SX
838 }
839 #endif
840
841 /*
842 * Since we have been unable to acquire the exclusive
843 * lock and the exclusive waiters flag is set, we have
844 * to sleep.
845 */
846 if (LOCK_LOG_TEST(&sx->lock_object, 0))
847 CTR2(KTR_LOCK, "%s: %p blocking on sleep queue",
848 __func__, sx);
849
850 #ifdef KDTRACE_HOOKS
851 sleep_time -= lockstat_nsecs(&sx->lock_object);
852 #endif
853 sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name,
854 SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ?
855 SLEEPQ_INTERRUPTIBLE : 0), SQ_EXCLUSIVE_QUEUE);
856 /*
857 * Hack: this can land in thread_suspend_check which will
858 * conditionally take a mutex, tripping over an assert if a
859 * lock we are waiting for is set.
860 */
861 THREAD_CONTENTION_DONE(&sx->lock_object);
862 if (!(opts & SX_INTERRUPTIBLE))
863 sleepq_wait(&sx->lock_object, 0);
864 else
865 error = sleepq_wait_sig(&sx->lock_object, 0);
866 THREAD_CONTENDS_ON_LOCK(&sx->lock_object);
867 #ifdef KDTRACE_HOOKS
868 sleep_time += lockstat_nsecs(&sx->lock_object);
869 sleep_cnt++;
870 #endif
871 if (error) {
872 if (LOCK_LOG_TEST(&sx->lock_object, 0))
873 CTR2(KTR_LOCK,
874 "%s: interruptible sleep by %p suspended by signal",
875 __func__, sx);
876 break;
877 }
878 if (LOCK_LOG_TEST(&sx->lock_object, 0))
879 CTR2(KTR_LOCK, "%s: %p resuming from sleep queue",
880 __func__, sx);
881 x = SX_READ_VALUE(sx);
882 }
883 THREAD_CONTENTION_DONE(&sx->lock_object);
884 if (__predict_true(!extra_work))
885 return (error);
886 #ifdef ADAPTIVE_SX
887 if (in_critical)
888 critical_exit();
889 #endif
890 GIANT_RESTORE();
891 #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
892 if (__predict_true(!doing_lockprof))
893 return (error);
894 #endif
895 #ifdef KDTRACE_HOOKS
896 all_time += lockstat_nsecs(&sx->lock_object);
897 if (sleep_time)
898 LOCKSTAT_RECORD4(sx__block, sx, sleep_time,
899 LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0,
900 (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
901 if (lda.spin_cnt > sleep_cnt)
902 LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time,
903 LOCKSTAT_WRITER, (state & SX_LOCK_SHARED) == 0,
904 (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
905 out_lockstat:
906 #endif
907 if (!error)
908 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
909 contested, waittime, file, line, LOCKSTAT_WRITER);
910 return (error);
911 }
912
913 /*
914 * This function represents the so-called 'hard case' for sx_xunlock
915 * operation. All 'easy case' failures are redirected to this. Note
916 * that ideally this would be a static function, but it needs to be
917 * accessible from at least sx.h.
918 */
919 void
_sx_xunlock_hard(struct sx * sx,uintptr_t x LOCK_FILE_LINE_ARG_DEF)920 _sx_xunlock_hard(struct sx *sx, uintptr_t x LOCK_FILE_LINE_ARG_DEF)
921 {
922 uintptr_t tid, setx;
923 int queue, wakeup_swapper;
924
925 if (SCHEDULER_STOPPED())
926 return;
927
928 tid = (uintptr_t)curthread;
929
930 if (__predict_false(x == tid))
931 x = SX_READ_VALUE(sx);
932
933 MPASS(!(x & SX_LOCK_SHARED));
934
935 if (__predict_false(x & SX_LOCK_RECURSED)) {
936 /* The lock is recursed, unrecurse one level. */
937 if ((--sx->sx_recurse) == 0)
938 atomic_clear_ptr(&sx->sx_lock, SX_LOCK_RECURSED);
939 if (LOCK_LOG_TEST(&sx->lock_object, 0))
940 CTR2(KTR_LOCK, "%s: %p unrecursing", __func__, sx);
941 return;
942 }
943
944 LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx, LOCKSTAT_WRITER);
945 if (x == tid &&
946 atomic_cmpset_rel_ptr(&sx->sx_lock, tid, SX_LOCK_UNLOCKED))
947 return;
948
949 if (LOCK_LOG_TEST(&sx->lock_object, 0))
950 CTR2(KTR_LOCK, "%s: %p contested", __func__, sx);
951
952 sleepq_lock(&sx->lock_object);
953 x = SX_READ_VALUE(sx);
954 MPASS(x & (SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS));
955
956 /*
957 * The wake up algorithm here is quite simple and probably not
958 * ideal. It gives precedence to shared waiters if they are
959 * present. For this condition, we have to preserve the
960 * state of the exclusive waiters flag.
961 * If interruptible sleeps left the shared queue empty avoid a
962 * starvation for the threads sleeping on the exclusive queue by giving
963 * them precedence and cleaning up the shared waiters bit anyway.
964 */
965 setx = SX_LOCK_UNLOCKED;
966 queue = SQ_SHARED_QUEUE;
967 if ((x & SX_LOCK_EXCLUSIVE_WAITERS) != 0 &&
968 sleepq_sleepcnt(&sx->lock_object, SQ_EXCLUSIVE_QUEUE) != 0) {
969 queue = SQ_EXCLUSIVE_QUEUE;
970 setx |= (x & SX_LOCK_SHARED_WAITERS);
971 }
972 atomic_store_rel_ptr(&sx->sx_lock, setx);
973
974 /* Wake up all the waiters for the specific queue. */
975 if (LOCK_LOG_TEST(&sx->lock_object, 0))
976 CTR3(KTR_LOCK, "%s: %p waking up all threads on %s queue",
977 __func__, sx, queue == SQ_SHARED_QUEUE ? "shared" :
978 "exclusive");
979
980 wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX, 0,
981 queue);
982 sleepq_release(&sx->lock_object);
983 if (wakeup_swapper)
984 kick_proc0();
985 }
986
987 static bool __always_inline
__sx_can_read(struct thread * td,uintptr_t x,bool fp)988 __sx_can_read(struct thread *td, uintptr_t x, bool fp)
989 {
990
991 if ((x & (SX_LOCK_SHARED | SX_LOCK_EXCLUSIVE_WAITERS | SX_LOCK_WRITE_SPINNER))
992 == SX_LOCK_SHARED)
993 return (true);
994 if (!fp && td->td_sx_slocks && (x & SX_LOCK_SHARED))
995 return (true);
996 return (false);
997 }
998
999 static bool __always_inline
__sx_slock_try(struct sx * sx,struct thread * td,uintptr_t * xp,bool fp LOCK_FILE_LINE_ARG_DEF)1000 __sx_slock_try(struct sx *sx, struct thread *td, uintptr_t *xp, bool fp
1001 LOCK_FILE_LINE_ARG_DEF)
1002 {
1003
1004 /*
1005 * If no other thread has an exclusive lock then try to bump up
1006 * the count of sharers. Since we have to preserve the state
1007 * of SX_LOCK_EXCLUSIVE_WAITERS, if we fail to acquire the
1008 * shared lock loop back and retry.
1009 */
1010 while (__sx_can_read(td, *xp, fp)) {
1011 if (atomic_fcmpset_acq_ptr(&sx->sx_lock, xp,
1012 *xp + SX_ONE_SHARER)) {
1013 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1014 CTR4(KTR_LOCK, "%s: %p succeed %p -> %p",
1015 __func__, sx, (void *)*xp,
1016 (void *)(*xp + SX_ONE_SHARER));
1017 td->td_sx_slocks++;
1018 return (true);
1019 }
1020 }
1021 return (false);
1022 }
1023
1024 static int __noinline
_sx_slock_hard(struct sx * sx,int opts,uintptr_t x LOCK_FILE_LINE_ARG_DEF)1025 _sx_slock_hard(struct sx *sx, int opts, uintptr_t x LOCK_FILE_LINE_ARG_DEF)
1026 {
1027 GIANT_DECLARE;
1028 struct thread *td;
1029 #ifdef ADAPTIVE_SX
1030 struct thread *owner;
1031 u_int i, n, spintries = 0;
1032 #endif
1033 #ifdef LOCK_PROFILING
1034 uint64_t waittime = 0;
1035 int contested = 0;
1036 #endif
1037 int error = 0;
1038 #if defined(ADAPTIVE_SX) || defined(KDTRACE_HOOKS)
1039 struct lock_delay_arg lda;
1040 #endif
1041 #ifdef KDTRACE_HOOKS
1042 u_int sleep_cnt = 0;
1043 int64_t sleep_time = 0;
1044 int64_t all_time = 0;
1045 #endif
1046 #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
1047 uintptr_t state = 0;
1048 #endif
1049 int extra_work __sdt_used = 0;
1050
1051 td = curthread;
1052
1053 #ifdef KDTRACE_HOOKS
1054 if (LOCKSTAT_PROFILE_ENABLED(sx__acquire)) {
1055 if (__sx_slock_try(sx, td, &x, false LOCK_FILE_LINE_ARG))
1056 goto out_lockstat;
1057 extra_work = 1;
1058 all_time -= lockstat_nsecs(&sx->lock_object);
1059 state = x;
1060 }
1061 #endif
1062 #ifdef LOCK_PROFILING
1063 extra_work = 1;
1064 state = x;
1065 #endif
1066
1067 if (SCHEDULER_STOPPED())
1068 return (0);
1069
1070 #if defined(ADAPTIVE_SX)
1071 lock_delay_arg_init(&lda, &sx_delay);
1072 #elif defined(KDTRACE_HOOKS)
1073 lock_delay_arg_init_noadapt(&lda);
1074 #endif
1075
1076 #ifdef HWPMC_HOOKS
1077 PMC_SOFT_CALL( , , lock, failed);
1078 #endif
1079 lock_profile_obtain_lock_failed(&sx->lock_object, false, &contested,
1080 &waittime);
1081
1082 #ifndef INVARIANTS
1083 GIANT_SAVE(extra_work);
1084 #endif
1085
1086 THREAD_CONTENDS_ON_LOCK(&sx->lock_object);
1087
1088 /*
1089 * As with rwlocks, we don't make any attempt to try to block
1090 * shared locks once there is an exclusive waiter.
1091 */
1092 for (;;) {
1093 if (__sx_slock_try(sx, td, &x, false LOCK_FILE_LINE_ARG))
1094 break;
1095 #ifdef INVARIANTS
1096 GIANT_SAVE(extra_work);
1097 #endif
1098 #ifdef KDTRACE_HOOKS
1099 lda.spin_cnt++;
1100 #endif
1101
1102 #ifdef ADAPTIVE_SX
1103 /*
1104 * If the owner is running on another CPU, spin until
1105 * the owner stops running or the state of the lock
1106 * changes.
1107 */
1108 if ((x & SX_LOCK_SHARED) == 0) {
1109 owner = lv_sx_owner(x);
1110 if (TD_IS_RUNNING(owner)) {
1111 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1112 CTR3(KTR_LOCK,
1113 "%s: spinning on %p held by %p",
1114 __func__, sx, owner);
1115 KTR_STATE1(KTR_SCHED, "thread",
1116 sched_tdname(curthread), "spinning",
1117 "lockname:\"%s\"", sx->lock_object.lo_name);
1118 do {
1119 lock_delay(&lda);
1120 x = SX_READ_VALUE(sx);
1121 owner = lv_sx_owner(x);
1122 } while (owner != NULL && TD_IS_RUNNING(owner));
1123 KTR_STATE0(KTR_SCHED, "thread",
1124 sched_tdname(curthread), "running");
1125 continue;
1126 }
1127 } else {
1128 if ((x & SX_LOCK_WRITE_SPINNER) && SX_SHARERS(x) == 0) {
1129 MPASS(!__sx_can_read(td, x, false));
1130 lock_delay_spin(2);
1131 x = SX_READ_VALUE(sx);
1132 continue;
1133 }
1134 if (spintries < asx_retries) {
1135 KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
1136 "spinning", "lockname:\"%s\"",
1137 sx->lock_object.lo_name);
1138 n = SX_SHARERS(x);
1139 for (i = 0; i < asx_loops; i += n) {
1140 lock_delay_spin(n);
1141 x = SX_READ_VALUE(sx);
1142 if (!(x & SX_LOCK_SHARED))
1143 break;
1144 n = SX_SHARERS(x);
1145 if (n == 0)
1146 break;
1147 if (__sx_can_read(td, x, false))
1148 break;
1149 }
1150 #ifdef KDTRACE_HOOKS
1151 lda.spin_cnt += i;
1152 #endif
1153 KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
1154 "running");
1155 if (i < asx_loops)
1156 continue;
1157 }
1158 }
1159 #endif
1160
1161 /*
1162 * Some other thread already has an exclusive lock, so
1163 * start the process of blocking.
1164 */
1165 sleepq_lock(&sx->lock_object);
1166 x = SX_READ_VALUE(sx);
1167 retry_sleepq:
1168 if (((x & SX_LOCK_WRITE_SPINNER) && SX_SHARERS(x) == 0) ||
1169 __sx_can_read(td, x, false)) {
1170 sleepq_release(&sx->lock_object);
1171 continue;
1172 }
1173
1174 #ifdef ADAPTIVE_SX
1175 /*
1176 * If the owner is running on another CPU, spin until
1177 * the owner stops running or the state of the lock
1178 * changes.
1179 */
1180 if (!(x & SX_LOCK_SHARED)) {
1181 owner = (struct thread *)SX_OWNER(x);
1182 if (TD_IS_RUNNING(owner)) {
1183 sleepq_release(&sx->lock_object);
1184 x = SX_READ_VALUE(sx);
1185 continue;
1186 }
1187 }
1188 #endif
1189
1190 /*
1191 * Try to set the SX_LOCK_SHARED_WAITERS flag. If we
1192 * fail to set it drop the sleep queue lock and loop
1193 * back.
1194 */
1195 if (!(x & SX_LOCK_SHARED_WAITERS)) {
1196 if (!atomic_fcmpset_ptr(&sx->sx_lock, &x,
1197 x | SX_LOCK_SHARED_WAITERS))
1198 goto retry_sleepq;
1199 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1200 CTR2(KTR_LOCK, "%s: %p set shared waiters flag",
1201 __func__, sx);
1202 }
1203
1204 /*
1205 * Since we have been unable to acquire the shared lock,
1206 * we have to sleep.
1207 */
1208 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1209 CTR2(KTR_LOCK, "%s: %p blocking on sleep queue",
1210 __func__, sx);
1211
1212 #ifdef KDTRACE_HOOKS
1213 sleep_time -= lockstat_nsecs(&sx->lock_object);
1214 #endif
1215 sleepq_add(&sx->lock_object, NULL, sx->lock_object.lo_name,
1216 SLEEPQ_SX | ((opts & SX_INTERRUPTIBLE) ?
1217 SLEEPQ_INTERRUPTIBLE : 0), SQ_SHARED_QUEUE);
1218 /*
1219 * Hack: this can land in thread_suspend_check which will
1220 * conditionally take a mutex, tripping over an assert if a
1221 * lock we are waiting for is set.
1222 */
1223 THREAD_CONTENTION_DONE(&sx->lock_object);
1224 if (!(opts & SX_INTERRUPTIBLE))
1225 sleepq_wait(&sx->lock_object, 0);
1226 else
1227 error = sleepq_wait_sig(&sx->lock_object, 0);
1228 THREAD_CONTENDS_ON_LOCK(&sx->lock_object);
1229 #ifdef KDTRACE_HOOKS
1230 sleep_time += lockstat_nsecs(&sx->lock_object);
1231 sleep_cnt++;
1232 #endif
1233 if (error) {
1234 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1235 CTR2(KTR_LOCK,
1236 "%s: interruptible sleep by %p suspended by signal",
1237 __func__, sx);
1238 break;
1239 }
1240 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1241 CTR2(KTR_LOCK, "%s: %p resuming from sleep queue",
1242 __func__, sx);
1243 x = SX_READ_VALUE(sx);
1244 }
1245 THREAD_CONTENTION_DONE(&sx->lock_object);
1246 #if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
1247 if (__predict_true(!extra_work))
1248 return (error);
1249 #endif
1250 #ifdef KDTRACE_HOOKS
1251 all_time += lockstat_nsecs(&sx->lock_object);
1252 if (sleep_time)
1253 LOCKSTAT_RECORD4(sx__block, sx, sleep_time,
1254 LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0,
1255 (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
1256 if (lda.spin_cnt > sleep_cnt)
1257 LOCKSTAT_RECORD4(sx__spin, sx, all_time - sleep_time,
1258 LOCKSTAT_READER, (state & SX_LOCK_SHARED) == 0,
1259 (state & SX_LOCK_SHARED) == 0 ? 0 : SX_SHARERS(state));
1260 out_lockstat:
1261 #endif
1262 if (error == 0) {
1263 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(sx__acquire, sx,
1264 contested, waittime, file, line, LOCKSTAT_READER);
1265 }
1266 GIANT_RESTORE();
1267 return (error);
1268 }
1269
1270 int
_sx_slock_int(struct sx * sx,int opts LOCK_FILE_LINE_ARG_DEF)1271 _sx_slock_int(struct sx *sx, int opts LOCK_FILE_LINE_ARG_DEF)
1272 {
1273 struct thread *td;
1274 uintptr_t x;
1275 int error;
1276
1277 KASSERT(kdb_active != 0 || SCHEDULER_STOPPED() ||
1278 !TD_IS_IDLETHREAD(curthread),
1279 ("sx_slock() by idle thread %p on sx %s @ %s:%d",
1280 curthread, sx->lock_object.lo_name, file, line));
1281 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
1282 ("sx_slock() of destroyed sx @ %s:%d", file, line));
1283 WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER, file, line, NULL);
1284
1285 error = 0;
1286 td = curthread;
1287 x = SX_READ_VALUE(sx);
1288 if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__acquire) ||
1289 !__sx_slock_try(sx, td, &x, true LOCK_FILE_LINE_ARG)))
1290 error = _sx_slock_hard(sx, opts, x LOCK_FILE_LINE_ARG);
1291 else
1292 lock_profile_obtain_lock_success(&sx->lock_object, false, 0, 0,
1293 file, line);
1294 if (error == 0) {
1295 LOCK_LOG_LOCK("SLOCK", &sx->lock_object, 0, 0, file, line);
1296 WITNESS_LOCK(&sx->lock_object, 0, file, line);
1297 TD_LOCKS_INC(curthread);
1298 }
1299 return (error);
1300 }
1301
1302 int
_sx_slock(struct sx * sx,int opts,const char * file,int line)1303 _sx_slock(struct sx *sx, int opts, const char *file, int line)
1304 {
1305
1306 return (_sx_slock_int(sx, opts LOCK_FILE_LINE_ARG));
1307 }
1308
1309 static bool __always_inline
_sx_sunlock_try(struct sx * sx,struct thread * td,uintptr_t * xp)1310 _sx_sunlock_try(struct sx *sx, struct thread *td, uintptr_t *xp)
1311 {
1312
1313 for (;;) {
1314 if (SX_SHARERS(*xp) > 1 || !(*xp & SX_LOCK_WAITERS)) {
1315 if (atomic_fcmpset_rel_ptr(&sx->sx_lock, xp,
1316 *xp - SX_ONE_SHARER)) {
1317 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1318 CTR4(KTR_LOCK,
1319 "%s: %p succeeded %p -> %p",
1320 __func__, sx, (void *)*xp,
1321 (void *)(*xp - SX_ONE_SHARER));
1322 td->td_sx_slocks--;
1323 return (true);
1324 }
1325 continue;
1326 }
1327 break;
1328 }
1329 return (false);
1330 }
1331
1332 static void __noinline
_sx_sunlock_hard(struct sx * sx,struct thread * td,uintptr_t x LOCK_FILE_LINE_ARG_DEF)1333 _sx_sunlock_hard(struct sx *sx, struct thread *td, uintptr_t x
1334 LOCK_FILE_LINE_ARG_DEF)
1335 {
1336 int wakeup_swapper = 0;
1337 uintptr_t setx, queue;
1338
1339 if (SCHEDULER_STOPPED())
1340 return;
1341
1342 if (_sx_sunlock_try(sx, td, &x))
1343 goto out_lockstat;
1344
1345 sleepq_lock(&sx->lock_object);
1346 x = SX_READ_VALUE(sx);
1347 for (;;) {
1348 if (_sx_sunlock_try(sx, td, &x))
1349 break;
1350
1351 /*
1352 * Wake up semantic here is quite simple:
1353 * Just wake up all the exclusive waiters.
1354 * Note that the state of the lock could have changed,
1355 * so if it fails loop back and retry.
1356 */
1357 setx = SX_LOCK_UNLOCKED;
1358 queue = SQ_SHARED_QUEUE;
1359 if (x & SX_LOCK_EXCLUSIVE_WAITERS) {
1360 setx |= (x & SX_LOCK_SHARED_WAITERS);
1361 queue = SQ_EXCLUSIVE_QUEUE;
1362 }
1363 setx |= (x & SX_LOCK_WRITE_SPINNER);
1364 if (!atomic_fcmpset_rel_ptr(&sx->sx_lock, &x, setx))
1365 continue;
1366 if (LOCK_LOG_TEST(&sx->lock_object, 0))
1367 CTR2(KTR_LOCK, "%s: %p waking up all thread on"
1368 "exclusive queue", __func__, sx);
1369 wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX,
1370 0, queue);
1371 td->td_sx_slocks--;
1372 break;
1373 }
1374 sleepq_release(&sx->lock_object);
1375 if (wakeup_swapper)
1376 kick_proc0();
1377 out_lockstat:
1378 LOCKSTAT_PROFILE_RELEASE_RWLOCK(sx__release, sx, LOCKSTAT_READER);
1379 }
1380
1381 void
_sx_sunlock_int(struct sx * sx LOCK_FILE_LINE_ARG_DEF)1382 _sx_sunlock_int(struct sx *sx LOCK_FILE_LINE_ARG_DEF)
1383 {
1384 struct thread *td;
1385 uintptr_t x;
1386
1387 KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
1388 ("sx_sunlock() of destroyed sx @ %s:%d", file, line));
1389 _sx_assert(sx, SA_SLOCKED, file, line);
1390 WITNESS_UNLOCK(&sx->lock_object, 0, file, line);
1391 LOCK_LOG_LOCK("SUNLOCK", &sx->lock_object, 0, 0, file, line);
1392
1393 td = curthread;
1394 x = SX_READ_VALUE(sx);
1395 if (__predict_false(LOCKSTAT_PROFILE_ENABLED(sx__release) ||
1396 !_sx_sunlock_try(sx, td, &x)))
1397 _sx_sunlock_hard(sx, td, x LOCK_FILE_LINE_ARG);
1398 else
1399 lock_profile_release_lock(&sx->lock_object, false);
1400
1401 TD_LOCKS_DEC(curthread);
1402 }
1403
1404 void
_sx_sunlock(struct sx * sx,const char * file,int line)1405 _sx_sunlock(struct sx *sx, const char *file, int line)
1406 {
1407
1408 _sx_sunlock_int(sx LOCK_FILE_LINE_ARG);
1409 }
1410
1411 #ifdef INVARIANT_SUPPORT
1412 #ifndef INVARIANTS
1413 #undef _sx_assert
1414 #endif
1415
1416 /*
1417 * In the non-WITNESS case, sx_assert() can only detect that at least
1418 * *some* thread owns an slock, but it cannot guarantee that *this*
1419 * thread owns an slock.
1420 */
1421 void
_sx_assert(const struct sx * sx,int what,const char * file,int line)1422 _sx_assert(const struct sx *sx, int what, const char *file, int line)
1423 {
1424 #ifndef WITNESS
1425 int slocked = 0;
1426 #endif
1427
1428 if (SCHEDULER_STOPPED())
1429 return;
1430 switch (what) {
1431 case SA_SLOCKED:
1432 case SA_SLOCKED | SA_NOTRECURSED:
1433 case SA_SLOCKED | SA_RECURSED:
1434 #ifndef WITNESS
1435 slocked = 1;
1436 /* FALLTHROUGH */
1437 #endif
1438 case SA_LOCKED:
1439 case SA_LOCKED | SA_NOTRECURSED:
1440 case SA_LOCKED | SA_RECURSED:
1441 #ifdef WITNESS
1442 witness_assert(&sx->lock_object, what, file, line);
1443 #else
1444 /*
1445 * If some other thread has an exclusive lock or we
1446 * have one and are asserting a shared lock, fail.
1447 * Also, if no one has a lock at all, fail.
1448 */
1449 if (sx->sx_lock == SX_LOCK_UNLOCKED ||
1450 (!(sx->sx_lock & SX_LOCK_SHARED) && (slocked ||
1451 sx_xholder(sx) != curthread)))
1452 panic("Lock %s not %slocked @ %s:%d\n",
1453 sx->lock_object.lo_name, slocked ? "share " : "",
1454 file, line);
1455
1456 if (!(sx->sx_lock & SX_LOCK_SHARED)) {
1457 if (sx_recursed(sx)) {
1458 if (what & SA_NOTRECURSED)
1459 panic("Lock %s recursed @ %s:%d\n",
1460 sx->lock_object.lo_name, file,
1461 line);
1462 } else if (what & SA_RECURSED)
1463 panic("Lock %s not recursed @ %s:%d\n",
1464 sx->lock_object.lo_name, file, line);
1465 }
1466 #endif
1467 break;
1468 case SA_XLOCKED:
1469 case SA_XLOCKED | SA_NOTRECURSED:
1470 case SA_XLOCKED | SA_RECURSED:
1471 if (sx_xholder(sx) != curthread)
1472 panic("Lock %s not exclusively locked @ %s:%d\n",
1473 sx->lock_object.lo_name, file, line);
1474 if (sx_recursed(sx)) {
1475 if (what & SA_NOTRECURSED)
1476 panic("Lock %s recursed @ %s:%d\n",
1477 sx->lock_object.lo_name, file, line);
1478 } else if (what & SA_RECURSED)
1479 panic("Lock %s not recursed @ %s:%d\n",
1480 sx->lock_object.lo_name, file, line);
1481 break;
1482 case SA_UNLOCKED:
1483 #ifdef WITNESS
1484 witness_assert(&sx->lock_object, what, file, line);
1485 #else
1486 /*
1487 * If we hold an exclusve lock fail. We can't
1488 * reliably check to see if we hold a shared lock or
1489 * not.
1490 */
1491 if (sx_xholder(sx) == curthread)
1492 panic("Lock %s exclusively locked @ %s:%d\n",
1493 sx->lock_object.lo_name, file, line);
1494 #endif
1495 break;
1496 default:
1497 panic("Unknown sx lock assertion: %d @ %s:%d", what, file,
1498 line);
1499 }
1500 }
1501 #endif /* INVARIANT_SUPPORT */
1502
1503 #ifdef DDB
1504 static void
db_show_sx(const struct lock_object * lock)1505 db_show_sx(const struct lock_object *lock)
1506 {
1507 struct thread *td;
1508 const struct sx *sx;
1509
1510 sx = (const struct sx *)lock;
1511
1512 db_printf(" state: ");
1513 if (sx->sx_lock == SX_LOCK_UNLOCKED)
1514 db_printf("UNLOCKED\n");
1515 else if (sx->sx_lock == SX_LOCK_DESTROYED) {
1516 db_printf("DESTROYED\n");
1517 return;
1518 } else if (sx->sx_lock & SX_LOCK_SHARED)
1519 db_printf("SLOCK: %ju\n", (uintmax_t)SX_SHARERS(sx->sx_lock));
1520 else {
1521 td = sx_xholder(sx);
1522 db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td,
1523 td->td_tid, td->td_proc->p_pid, td->td_name);
1524 if (sx_recursed(sx))
1525 db_printf(" recursed: %d\n", sx->sx_recurse);
1526 }
1527
1528 db_printf(" waiters: ");
1529 switch(sx->sx_lock &
1530 (SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS)) {
1531 case SX_LOCK_SHARED_WAITERS:
1532 db_printf("shared\n");
1533 break;
1534 case SX_LOCK_EXCLUSIVE_WAITERS:
1535 db_printf("exclusive\n");
1536 break;
1537 case SX_LOCK_SHARED_WAITERS | SX_LOCK_EXCLUSIVE_WAITERS:
1538 db_printf("exclusive and shared\n");
1539 break;
1540 default:
1541 db_printf("none\n");
1542 }
1543 }
1544
1545 /*
1546 * Check to see if a thread that is blocked on a sleep queue is actually
1547 * blocked on an sx lock. If so, output some details and return true.
1548 * If the lock has an exclusive owner, return that in *ownerp.
1549 */
1550 int
sx_chain(struct thread * td,struct thread ** ownerp)1551 sx_chain(struct thread *td, struct thread **ownerp)
1552 {
1553 const struct sx *sx;
1554
1555 /*
1556 * Check to see if this thread is blocked on an sx lock.
1557 * First, we check the lock class. If that is ok, then we
1558 * compare the lock name against the wait message.
1559 */
1560 sx = td->td_wchan;
1561 if (LOCK_CLASS(&sx->lock_object) != &lock_class_sx ||
1562 sx->lock_object.lo_name != td->td_wmesg)
1563 return (0);
1564
1565 /* We think we have an sx lock, so output some details. */
1566 db_printf("blocked on sx \"%s\" ", td->td_wmesg);
1567 *ownerp = sx_xholder(sx);
1568 if (sx->sx_lock & SX_LOCK_SHARED)
1569 db_printf("SLOCK (count %ju)\n",
1570 (uintmax_t)SX_SHARERS(sx->sx_lock));
1571 else
1572 db_printf("XLOCK\n");
1573 return (1);
1574 }
1575 #endif
1576