FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_umtx.c
1 /*-
2 * Copyright (c) 2004, David Xu <davidxu@freebsd.org>
3 * Copyright (c) 2002, Jeffrey Roberson <jeff@freebsd.org>
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice unmodified, this list of conditions, and the following
11 * disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD: releng/10.3/sys/kern/kern_umtx.c 288494 2015-10-02 13:48:32Z vangyzen $");
30
31 #include "opt_compat.h"
32 #include "opt_umtx_profiling.h"
33
34 #include <sys/param.h>
35 #include <sys/kernel.h>
36 #include <sys/limits.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/mutex.h>
40 #include <sys/priv.h>
41 #include <sys/proc.h>
42 #include <sys/sbuf.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/sysctl.h>
46 #include <sys/sysent.h>
47 #include <sys/systm.h>
48 #include <sys/sysproto.h>
49 #include <sys/syscallsubr.h>
50 #include <sys/eventhandler.h>
51 #include <sys/umtx.h>
52
53 #include <vm/vm.h>
54 #include <vm/vm_param.h>
55 #include <vm/pmap.h>
56 #include <vm/vm_map.h>
57 #include <vm/vm_object.h>
58
59 #include <machine/cpu.h>
60
61 #ifdef COMPAT_FREEBSD32
62 #include <compat/freebsd32/freebsd32_proto.h>
63 #endif
64
65 #define _UMUTEX_TRY 1
66 #define _UMUTEX_WAIT 2
67
68 #ifdef UMTX_PROFILING
69 #define UPROF_PERC_BIGGER(w, f, sw, sf) \
70 (((w) > (sw)) || ((w) == (sw) && (f) > (sf)))
71 #endif
72
73 /* Priority inheritance mutex info. */
74 struct umtx_pi {
75 /* Owner thread */
76 struct thread *pi_owner;
77
78 /* Reference count */
79 int pi_refcount;
80
81 /* List entry to link umtx holding by thread */
82 TAILQ_ENTRY(umtx_pi) pi_link;
83
84 /* List entry in hash */
85 TAILQ_ENTRY(umtx_pi) pi_hashlink;
86
87 /* List for waiters */
88 TAILQ_HEAD(,umtx_q) pi_blocked;
89
90 /* Identify a userland lock object */
91 struct umtx_key pi_key;
92 };
93
94 /* A userland synchronous object user. */
95 struct umtx_q {
96 /* Linked list for the hash. */
97 TAILQ_ENTRY(umtx_q) uq_link;
98
99 /* Umtx key. */
100 struct umtx_key uq_key;
101
102 /* Umtx flags. */
103 int uq_flags;
104 #define UQF_UMTXQ 0x0001
105
106 /* The thread waits on. */
107 struct thread *uq_thread;
108
109 /*
110 * Blocked on PI mutex. read can use chain lock
111 * or umtx_lock, write must have both chain lock and
112 * umtx_lock being hold.
113 */
114 struct umtx_pi *uq_pi_blocked;
115
116 /* On blocked list */
117 TAILQ_ENTRY(umtx_q) uq_lockq;
118
119 /* Thread contending with us */
120 TAILQ_HEAD(,umtx_pi) uq_pi_contested;
121
122 /* Inherited priority from PP mutex */
123 u_char uq_inherited_pri;
124
125 /* Spare queue ready to be reused */
126 struct umtxq_queue *uq_spare_queue;
127
128 /* The queue we on */
129 struct umtxq_queue *uq_cur_queue;
130 };
131
132 TAILQ_HEAD(umtxq_head, umtx_q);
133
134 /* Per-key wait-queue */
135 struct umtxq_queue {
136 struct umtxq_head head;
137 struct umtx_key key;
138 LIST_ENTRY(umtxq_queue) link;
139 int length;
140 };
141
142 LIST_HEAD(umtxq_list, umtxq_queue);
143
144 /* Userland lock object's wait-queue chain */
145 struct umtxq_chain {
146 /* Lock for this chain. */
147 struct mtx uc_lock;
148
149 /* List of sleep queues. */
150 struct umtxq_list uc_queue[2];
151 #define UMTX_SHARED_QUEUE 0
152 #define UMTX_EXCLUSIVE_QUEUE 1
153
154 LIST_HEAD(, umtxq_queue) uc_spare_queue;
155
156 /* Busy flag */
157 char uc_busy;
158
159 /* Chain lock waiters */
160 int uc_waiters;
161
162 /* All PI in the list */
163 TAILQ_HEAD(,umtx_pi) uc_pi_list;
164
165 #ifdef UMTX_PROFILING
166 u_int length;
167 u_int max_length;
168 #endif
169 };
170
171 #define UMTXQ_LOCKED_ASSERT(uc) mtx_assert(&(uc)->uc_lock, MA_OWNED)
172
173 /*
174 * Don't propagate time-sharing priority, there is a security reason,
175 * a user can simply introduce PI-mutex, let thread A lock the mutex,
176 * and let another thread B block on the mutex, because B is
177 * sleeping, its priority will be boosted, this causes A's priority to
178 * be boosted via priority propagating too and will never be lowered even
179 * if it is using 100%CPU, this is unfair to other processes.
180 */
181
182 #define UPRI(td) (((td)->td_user_pri >= PRI_MIN_TIMESHARE &&\
183 (td)->td_user_pri <= PRI_MAX_TIMESHARE) ?\
184 PRI_MAX_TIMESHARE : (td)->td_user_pri)
185
186 #define GOLDEN_RATIO_PRIME 2654404609U
187 #define UMTX_CHAINS 512
188 #define UMTX_SHIFTS (__WORD_BIT - 9)
189
190 #define GET_SHARE(flags) \
191 (((flags) & USYNC_PROCESS_SHARED) == 0 ? THREAD_SHARE : PROCESS_SHARE)
192
193 #define BUSY_SPINS 200
194
195 struct abs_timeout {
196 int clockid;
197 struct timespec cur;
198 struct timespec end;
199 };
200
201 static uma_zone_t umtx_pi_zone;
202 static struct umtxq_chain umtxq_chains[2][UMTX_CHAINS];
203 static MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory");
204 static int umtx_pi_allocated;
205
206 static SYSCTL_NODE(_debug, OID_AUTO, umtx, CTLFLAG_RW, 0, "umtx debug");
207 SYSCTL_INT(_debug_umtx, OID_AUTO, umtx_pi_allocated, CTLFLAG_RD,
208 &umtx_pi_allocated, 0, "Allocated umtx_pi");
209
210 #ifdef UMTX_PROFILING
211 static long max_length;
212 SYSCTL_LONG(_debug_umtx, OID_AUTO, max_length, CTLFLAG_RD, &max_length, 0, "max_length");
213 static SYSCTL_NODE(_debug_umtx, OID_AUTO, chains, CTLFLAG_RD, 0, "umtx chain stats");
214 #endif
215
216 static void umtxq_sysinit(void *);
217 static void umtxq_hash(struct umtx_key *key);
218 static struct umtxq_chain *umtxq_getchain(struct umtx_key *key);
219 static void umtxq_lock(struct umtx_key *key);
220 static void umtxq_unlock(struct umtx_key *key);
221 static void umtxq_busy(struct umtx_key *key);
222 static void umtxq_unbusy(struct umtx_key *key);
223 static void umtxq_insert_queue(struct umtx_q *uq, int q);
224 static void umtxq_remove_queue(struct umtx_q *uq, int q);
225 static int umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *);
226 static int umtxq_count(struct umtx_key *key);
227 static struct umtx_pi *umtx_pi_alloc(int);
228 static void umtx_pi_free(struct umtx_pi *pi);
229 static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags);
230 static void umtx_thread_cleanup(struct thread *td);
231 static void umtx_exec_hook(void *arg __unused, struct proc *p __unused,
232 struct image_params *imgp __unused);
233 SYSINIT(umtx, SI_SUB_EVENTHANDLER+1, SI_ORDER_MIDDLE, umtxq_sysinit, NULL);
234
235 #define umtxq_signal(key, nwake) umtxq_signal_queue((key), (nwake), UMTX_SHARED_QUEUE)
236 #define umtxq_insert(uq) umtxq_insert_queue((uq), UMTX_SHARED_QUEUE)
237 #define umtxq_remove(uq) umtxq_remove_queue((uq), UMTX_SHARED_QUEUE)
238
239 static struct mtx umtx_lock;
240
241 #ifdef UMTX_PROFILING
242 static void
243 umtx_init_profiling(void)
244 {
245 struct sysctl_oid *chain_oid;
246 char chain_name[10];
247 int i;
248
249 for (i = 0; i < UMTX_CHAINS; ++i) {
250 snprintf(chain_name, sizeof(chain_name), "%d", i);
251 chain_oid = SYSCTL_ADD_NODE(NULL,
252 SYSCTL_STATIC_CHILDREN(_debug_umtx_chains), OID_AUTO,
253 chain_name, CTLFLAG_RD, NULL, "umtx hash stats");
254 SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
255 "max_length0", CTLFLAG_RD, &umtxq_chains[0][i].max_length, 0, NULL);
256 SYSCTL_ADD_INT(NULL, SYSCTL_CHILDREN(chain_oid), OID_AUTO,
257 "max_length1", CTLFLAG_RD, &umtxq_chains[1][i].max_length, 0, NULL);
258 }
259 }
260
261 static int
262 sysctl_debug_umtx_chains_peaks(SYSCTL_HANDLER_ARGS)
263 {
264 char buf[512];
265 struct sbuf sb;
266 struct umtxq_chain *uc;
267 u_int fract, i, j, tot, whole;
268 u_int sf0, sf1, sf2, sf3, sf4;
269 u_int si0, si1, si2, si3, si4;
270 u_int sw0, sw1, sw2, sw3, sw4;
271
272 sbuf_new(&sb, buf, sizeof(buf), SBUF_FIXEDLEN);
273 for (i = 0; i < 2; i++) {
274 tot = 0;
275 for (j = 0; j < UMTX_CHAINS; ++j) {
276 uc = &umtxq_chains[i][j];
277 mtx_lock(&uc->uc_lock);
278 tot += uc->max_length;
279 mtx_unlock(&uc->uc_lock);
280 }
281 if (tot == 0)
282 sbuf_printf(&sb, "%u) Empty ", i);
283 else {
284 sf0 = sf1 = sf2 = sf3 = sf4 = 0;
285 si0 = si1 = si2 = si3 = si4 = 0;
286 sw0 = sw1 = sw2 = sw3 = sw4 = 0;
287 for (j = 0; j < UMTX_CHAINS; j++) {
288 uc = &umtxq_chains[i][j];
289 mtx_lock(&uc->uc_lock);
290 whole = uc->max_length * 100;
291 mtx_unlock(&uc->uc_lock);
292 fract = (whole % tot) * 100;
293 if (UPROF_PERC_BIGGER(whole, fract, sw0, sf0)) {
294 sf0 = fract;
295 si0 = j;
296 sw0 = whole;
297 } else if (UPROF_PERC_BIGGER(whole, fract, sw1,
298 sf1)) {
299 sf1 = fract;
300 si1 = j;
301 sw1 = whole;
302 } else if (UPROF_PERC_BIGGER(whole, fract, sw2,
303 sf2)) {
304 sf2 = fract;
305 si2 = j;
306 sw2 = whole;
307 } else if (UPROF_PERC_BIGGER(whole, fract, sw3,
308 sf3)) {
309 sf3 = fract;
310 si3 = j;
311 sw3 = whole;
312 } else if (UPROF_PERC_BIGGER(whole, fract, sw4,
313 sf4)) {
314 sf4 = fract;
315 si4 = j;
316 sw4 = whole;
317 }
318 }
319 sbuf_printf(&sb, "queue %u:\n", i);
320 sbuf_printf(&sb, "1st: %u.%u%% idx: %u\n", sw0 / tot,
321 sf0 / tot, si0);
322 sbuf_printf(&sb, "2nd: %u.%u%% idx: %u\n", sw1 / tot,
323 sf1 / tot, si1);
324 sbuf_printf(&sb, "3rd: %u.%u%% idx: %u\n", sw2 / tot,
325 sf2 / tot, si2);
326 sbuf_printf(&sb, "4th: %u.%u%% idx: %u\n", sw3 / tot,
327 sf3 / tot, si3);
328 sbuf_printf(&sb, "5th: %u.%u%% idx: %u\n", sw4 / tot,
329 sf4 / tot, si4);
330 }
331 }
332 sbuf_trim(&sb);
333 sbuf_finish(&sb);
334 sysctl_handle_string(oidp, sbuf_data(&sb), sbuf_len(&sb), req);
335 sbuf_delete(&sb);
336 return (0);
337 }
338
339 static int
340 sysctl_debug_umtx_chains_clear(SYSCTL_HANDLER_ARGS)
341 {
342 struct umtxq_chain *uc;
343 u_int i, j;
344 int clear, error;
345
346 clear = 0;
347 error = sysctl_handle_int(oidp, &clear, 0, req);
348 if (error != 0 || req->newptr == NULL)
349 return (error);
350
351 if (clear != 0) {
352 for (i = 0; i < 2; ++i) {
353 for (j = 0; j < UMTX_CHAINS; ++j) {
354 uc = &umtxq_chains[i][j];
355 mtx_lock(&uc->uc_lock);
356 uc->length = 0;
357 uc->max_length = 0;
358 mtx_unlock(&uc->uc_lock);
359 }
360 }
361 }
362 return (0);
363 }
364
365 SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, clear,
366 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
367 sysctl_debug_umtx_chains_clear, "I", "Clear umtx chains statistics");
368 SYSCTL_PROC(_debug_umtx_chains, OID_AUTO, peaks,
369 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0,
370 sysctl_debug_umtx_chains_peaks, "A", "Highest peaks in chains max length");
371 #endif
372
373 static void
374 umtxq_sysinit(void *arg __unused)
375 {
376 int i, j;
377
378 umtx_pi_zone = uma_zcreate("umtx pi", sizeof(struct umtx_pi),
379 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
380 for (i = 0; i < 2; ++i) {
381 for (j = 0; j < UMTX_CHAINS; ++j) {
382 mtx_init(&umtxq_chains[i][j].uc_lock, "umtxql", NULL,
383 MTX_DEF | MTX_DUPOK);
384 LIST_INIT(&umtxq_chains[i][j].uc_queue[0]);
385 LIST_INIT(&umtxq_chains[i][j].uc_queue[1]);
386 LIST_INIT(&umtxq_chains[i][j].uc_spare_queue);
387 TAILQ_INIT(&umtxq_chains[i][j].uc_pi_list);
388 umtxq_chains[i][j].uc_busy = 0;
389 umtxq_chains[i][j].uc_waiters = 0;
390 #ifdef UMTX_PROFILING
391 umtxq_chains[i][j].length = 0;
392 umtxq_chains[i][j].max_length = 0;
393 #endif
394 }
395 }
396 #ifdef UMTX_PROFILING
397 umtx_init_profiling();
398 #endif
399 mtx_init(&umtx_lock, "umtx lock", NULL, MTX_DEF);
400 EVENTHANDLER_REGISTER(process_exec, umtx_exec_hook, NULL,
401 EVENTHANDLER_PRI_ANY);
402 }
403
404 struct umtx_q *
405 umtxq_alloc(void)
406 {
407 struct umtx_q *uq;
408
409 uq = malloc(sizeof(struct umtx_q), M_UMTX, M_WAITOK | M_ZERO);
410 uq->uq_spare_queue = malloc(sizeof(struct umtxq_queue), M_UMTX, M_WAITOK | M_ZERO);
411 TAILQ_INIT(&uq->uq_spare_queue->head);
412 TAILQ_INIT(&uq->uq_pi_contested);
413 uq->uq_inherited_pri = PRI_MAX;
414 return (uq);
415 }
416
417 void
418 umtxq_free(struct umtx_q *uq)
419 {
420 MPASS(uq->uq_spare_queue != NULL);
421 free(uq->uq_spare_queue, M_UMTX);
422 free(uq, M_UMTX);
423 }
424
425 static inline void
426 umtxq_hash(struct umtx_key *key)
427 {
428 unsigned n = (uintptr_t)key->info.both.a + key->info.both.b;
429 key->hash = ((n * GOLDEN_RATIO_PRIME) >> UMTX_SHIFTS) % UMTX_CHAINS;
430 }
431
432 static inline struct umtxq_chain *
433 umtxq_getchain(struct umtx_key *key)
434 {
435 if (key->type <= TYPE_SEM)
436 return (&umtxq_chains[1][key->hash]);
437 return (&umtxq_chains[0][key->hash]);
438 }
439
440 /*
441 * Lock a chain.
442 */
443 static inline void
444 umtxq_lock(struct umtx_key *key)
445 {
446 struct umtxq_chain *uc;
447
448 uc = umtxq_getchain(key);
449 mtx_lock(&uc->uc_lock);
450 }
451
452 /*
453 * Unlock a chain.
454 */
455 static inline void
456 umtxq_unlock(struct umtx_key *key)
457 {
458 struct umtxq_chain *uc;
459
460 uc = umtxq_getchain(key);
461 mtx_unlock(&uc->uc_lock);
462 }
463
464 /*
465 * Set chain to busy state when following operation
466 * may be blocked (kernel mutex can not be used).
467 */
468 static inline void
469 umtxq_busy(struct umtx_key *key)
470 {
471 struct umtxq_chain *uc;
472
473 uc = umtxq_getchain(key);
474 mtx_assert(&uc->uc_lock, MA_OWNED);
475 if (uc->uc_busy) {
476 #ifdef SMP
477 if (smp_cpus > 1) {
478 int count = BUSY_SPINS;
479 if (count > 0) {
480 umtxq_unlock(key);
481 while (uc->uc_busy && --count > 0)
482 cpu_spinwait();
483 umtxq_lock(key);
484 }
485 }
486 #endif
487 while (uc->uc_busy) {
488 uc->uc_waiters++;
489 msleep(uc, &uc->uc_lock, 0, "umtxqb", 0);
490 uc->uc_waiters--;
491 }
492 }
493 uc->uc_busy = 1;
494 }
495
496 /*
497 * Unbusy a chain.
498 */
499 static inline void
500 umtxq_unbusy(struct umtx_key *key)
501 {
502 struct umtxq_chain *uc;
503
504 uc = umtxq_getchain(key);
505 mtx_assert(&uc->uc_lock, MA_OWNED);
506 KASSERT(uc->uc_busy != 0, ("not busy"));
507 uc->uc_busy = 0;
508 if (uc->uc_waiters)
509 wakeup_one(uc);
510 }
511
512 static inline void
513 umtxq_unbusy_unlocked(struct umtx_key *key)
514 {
515
516 umtxq_lock(key);
517 umtxq_unbusy(key);
518 umtxq_unlock(key);
519 }
520
521 static struct umtxq_queue *
522 umtxq_queue_lookup(struct umtx_key *key, int q)
523 {
524 struct umtxq_queue *uh;
525 struct umtxq_chain *uc;
526
527 uc = umtxq_getchain(key);
528 UMTXQ_LOCKED_ASSERT(uc);
529 LIST_FOREACH(uh, &uc->uc_queue[q], link) {
530 if (umtx_key_match(&uh->key, key))
531 return (uh);
532 }
533
534 return (NULL);
535 }
536
537 static inline void
538 umtxq_insert_queue(struct umtx_q *uq, int q)
539 {
540 struct umtxq_queue *uh;
541 struct umtxq_chain *uc;
542
543 uc = umtxq_getchain(&uq->uq_key);
544 UMTXQ_LOCKED_ASSERT(uc);
545 KASSERT((uq->uq_flags & UQF_UMTXQ) == 0, ("umtx_q is already on queue"));
546 uh = umtxq_queue_lookup(&uq->uq_key, q);
547 if (uh != NULL) {
548 LIST_INSERT_HEAD(&uc->uc_spare_queue, uq->uq_spare_queue, link);
549 } else {
550 uh = uq->uq_spare_queue;
551 uh->key = uq->uq_key;
552 LIST_INSERT_HEAD(&uc->uc_queue[q], uh, link);
553 #ifdef UMTX_PROFILING
554 uc->length++;
555 if (uc->length > uc->max_length) {
556 uc->max_length = uc->length;
557 if (uc->max_length > max_length)
558 max_length = uc->max_length;
559 }
560 #endif
561 }
562 uq->uq_spare_queue = NULL;
563
564 TAILQ_INSERT_TAIL(&uh->head, uq, uq_link);
565 uh->length++;
566 uq->uq_flags |= UQF_UMTXQ;
567 uq->uq_cur_queue = uh;
568 return;
569 }
570
571 static inline void
572 umtxq_remove_queue(struct umtx_q *uq, int q)
573 {
574 struct umtxq_chain *uc;
575 struct umtxq_queue *uh;
576
577 uc = umtxq_getchain(&uq->uq_key);
578 UMTXQ_LOCKED_ASSERT(uc);
579 if (uq->uq_flags & UQF_UMTXQ) {
580 uh = uq->uq_cur_queue;
581 TAILQ_REMOVE(&uh->head, uq, uq_link);
582 uh->length--;
583 uq->uq_flags &= ~UQF_UMTXQ;
584 if (TAILQ_EMPTY(&uh->head)) {
585 KASSERT(uh->length == 0,
586 ("inconsistent umtxq_queue length"));
587 #ifdef UMTX_PROFILING
588 uc->length--;
589 #endif
590 LIST_REMOVE(uh, link);
591 } else {
592 uh = LIST_FIRST(&uc->uc_spare_queue);
593 KASSERT(uh != NULL, ("uc_spare_queue is empty"));
594 LIST_REMOVE(uh, link);
595 }
596 uq->uq_spare_queue = uh;
597 uq->uq_cur_queue = NULL;
598 }
599 }
600
601 /*
602 * Check if there are multiple waiters
603 */
604 static int
605 umtxq_count(struct umtx_key *key)
606 {
607 struct umtxq_chain *uc;
608 struct umtxq_queue *uh;
609
610 uc = umtxq_getchain(key);
611 UMTXQ_LOCKED_ASSERT(uc);
612 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
613 if (uh != NULL)
614 return (uh->length);
615 return (0);
616 }
617
618 /*
619 * Check if there are multiple PI waiters and returns first
620 * waiter.
621 */
622 static int
623 umtxq_count_pi(struct umtx_key *key, struct umtx_q **first)
624 {
625 struct umtxq_chain *uc;
626 struct umtxq_queue *uh;
627
628 *first = NULL;
629 uc = umtxq_getchain(key);
630 UMTXQ_LOCKED_ASSERT(uc);
631 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE);
632 if (uh != NULL) {
633 *first = TAILQ_FIRST(&uh->head);
634 return (uh->length);
635 }
636 return (0);
637 }
638
639 static int
640 umtxq_check_susp(struct thread *td)
641 {
642 struct proc *p;
643 int error;
644
645 /*
646 * The check for TDF_NEEDSUSPCHK is racy, but it is enough to
647 * eventually break the lockstep loop.
648 */
649 if ((td->td_flags & TDF_NEEDSUSPCHK) == 0)
650 return (0);
651 error = 0;
652 p = td->td_proc;
653 PROC_LOCK(p);
654 if (P_SHOULDSTOP(p) ||
655 ((p->p_flag & P_TRACED) && (td->td_dbgflags & TDB_SUSPEND))) {
656 if (p->p_flag & P_SINGLE_EXIT)
657 error = EINTR;
658 else
659 error = ERESTART;
660 }
661 PROC_UNLOCK(p);
662 return (error);
663 }
664
665 /*
666 * Wake up threads waiting on an userland object.
667 */
668
669 static int
670 umtxq_signal_queue(struct umtx_key *key, int n_wake, int q)
671 {
672 struct umtxq_chain *uc;
673 struct umtxq_queue *uh;
674 struct umtx_q *uq;
675 int ret;
676
677 ret = 0;
678 uc = umtxq_getchain(key);
679 UMTXQ_LOCKED_ASSERT(uc);
680 uh = umtxq_queue_lookup(key, q);
681 if (uh != NULL) {
682 while ((uq = TAILQ_FIRST(&uh->head)) != NULL) {
683 umtxq_remove_queue(uq, q);
684 wakeup(uq);
685 if (++ret >= n_wake)
686 return (ret);
687 }
688 }
689 return (ret);
690 }
691
692
693 /*
694 * Wake up specified thread.
695 */
696 static inline void
697 umtxq_signal_thread(struct umtx_q *uq)
698 {
699 struct umtxq_chain *uc;
700
701 uc = umtxq_getchain(&uq->uq_key);
702 UMTXQ_LOCKED_ASSERT(uc);
703 umtxq_remove(uq);
704 wakeup(uq);
705 }
706
707 static inline int
708 tstohz(const struct timespec *tsp)
709 {
710 struct timeval tv;
711
712 TIMESPEC_TO_TIMEVAL(&tv, tsp);
713 return tvtohz(&tv);
714 }
715
716 static void
717 abs_timeout_init(struct abs_timeout *timo, int clockid, int absolute,
718 const struct timespec *timeout)
719 {
720
721 timo->clockid = clockid;
722 if (!absolute) {
723 kern_clock_gettime(curthread, clockid, &timo->end);
724 timo->cur = timo->end;
725 timespecadd(&timo->end, timeout);
726 } else {
727 timo->end = *timeout;
728 kern_clock_gettime(curthread, clockid, &timo->cur);
729 }
730 }
731
732 static void
733 abs_timeout_init2(struct abs_timeout *timo, const struct _umtx_time *umtxtime)
734 {
735
736 abs_timeout_init(timo, umtxtime->_clockid,
737 (umtxtime->_flags & UMTX_ABSTIME) != 0,
738 &umtxtime->_timeout);
739 }
740
741 static inline void
742 abs_timeout_update(struct abs_timeout *timo)
743 {
744 kern_clock_gettime(curthread, timo->clockid, &timo->cur);
745 }
746
747 static int
748 abs_timeout_gethz(struct abs_timeout *timo)
749 {
750 struct timespec tts;
751
752 if (timespeccmp(&timo->end, &timo->cur, <=))
753 return (-1);
754 tts = timo->end;
755 timespecsub(&tts, &timo->cur);
756 return (tstohz(&tts));
757 }
758
759 /*
760 * Put thread into sleep state, before sleeping, check if
761 * thread was removed from umtx queue.
762 */
763 static inline int
764 umtxq_sleep(struct umtx_q *uq, const char *wmesg, struct abs_timeout *abstime)
765 {
766 struct umtxq_chain *uc;
767 int error, timo;
768
769 uc = umtxq_getchain(&uq->uq_key);
770 UMTXQ_LOCKED_ASSERT(uc);
771 for (;;) {
772 if (!(uq->uq_flags & UQF_UMTXQ))
773 return (0);
774 if (abstime != NULL) {
775 timo = abs_timeout_gethz(abstime);
776 if (timo < 0)
777 return (ETIMEDOUT);
778 } else
779 timo = 0;
780 error = msleep(uq, &uc->uc_lock, PCATCH | PDROP, wmesg, timo);
781 if (error != EWOULDBLOCK) {
782 umtxq_lock(&uq->uq_key);
783 break;
784 }
785 if (abstime != NULL)
786 abs_timeout_update(abstime);
787 umtxq_lock(&uq->uq_key);
788 }
789 return (error);
790 }
791
792 /*
793 * Convert userspace address into unique logical address.
794 */
795 int
796 umtx_key_get(void *addr, int type, int share, struct umtx_key *key)
797 {
798 struct thread *td = curthread;
799 vm_map_t map;
800 vm_map_entry_t entry;
801 vm_pindex_t pindex;
802 vm_prot_t prot;
803 boolean_t wired;
804
805 key->type = type;
806 if (share == THREAD_SHARE) {
807 key->shared = 0;
808 key->info.private.vs = td->td_proc->p_vmspace;
809 key->info.private.addr = (uintptr_t)addr;
810 } else {
811 MPASS(share == PROCESS_SHARE || share == AUTO_SHARE);
812 map = &td->td_proc->p_vmspace->vm_map;
813 if (vm_map_lookup(&map, (vm_offset_t)addr, VM_PROT_WRITE,
814 &entry, &key->info.shared.object, &pindex, &prot,
815 &wired) != KERN_SUCCESS) {
816 return EFAULT;
817 }
818
819 if ((share == PROCESS_SHARE) ||
820 (share == AUTO_SHARE &&
821 VM_INHERIT_SHARE == entry->inheritance)) {
822 key->shared = 1;
823 key->info.shared.offset = entry->offset + entry->start -
824 (vm_offset_t)addr;
825 vm_object_reference(key->info.shared.object);
826 } else {
827 key->shared = 0;
828 key->info.private.vs = td->td_proc->p_vmspace;
829 key->info.private.addr = (uintptr_t)addr;
830 }
831 vm_map_lookup_done(map, entry);
832 }
833
834 umtxq_hash(key);
835 return (0);
836 }
837
838 /*
839 * Release key.
840 */
841 void
842 umtx_key_release(struct umtx_key *key)
843 {
844 if (key->shared)
845 vm_object_deallocate(key->info.shared.object);
846 }
847
848 /*
849 * Lock a umtx object.
850 */
851 static int
852 do_lock_umtx(struct thread *td, struct umtx *umtx, u_long id,
853 const struct timespec *timeout)
854 {
855 struct abs_timeout timo;
856 struct umtx_q *uq;
857 u_long owner;
858 u_long old;
859 int error = 0;
860
861 uq = td->td_umtxq;
862 if (timeout != NULL)
863 abs_timeout_init(&timo, CLOCK_REALTIME, 0, timeout);
864
865 /*
866 * Care must be exercised when dealing with umtx structure. It
867 * can fault on any access.
868 */
869 for (;;) {
870 /*
871 * Try the uncontested case. This should be done in userland.
872 */
873 owner = casuword(&umtx->u_owner, UMTX_UNOWNED, id);
874
875 /* The acquire succeeded. */
876 if (owner == UMTX_UNOWNED)
877 return (0);
878
879 /* The address was invalid. */
880 if (owner == -1)
881 return (EFAULT);
882
883 /* If no one owns it but it is contested try to acquire it. */
884 if (owner == UMTX_CONTESTED) {
885 owner = casuword(&umtx->u_owner,
886 UMTX_CONTESTED, id | UMTX_CONTESTED);
887
888 if (owner == UMTX_CONTESTED)
889 return (0);
890
891 /* The address was invalid. */
892 if (owner == -1)
893 return (EFAULT);
894
895 error = umtxq_check_susp(td);
896 if (error != 0)
897 break;
898
899 /* If this failed the lock has changed, restart. */
900 continue;
901 }
902
903 /*
904 * If we caught a signal, we have retried and now
905 * exit immediately.
906 */
907 if (error != 0)
908 break;
909
910 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK,
911 AUTO_SHARE, &uq->uq_key)) != 0)
912 return (error);
913
914 umtxq_lock(&uq->uq_key);
915 umtxq_busy(&uq->uq_key);
916 umtxq_insert(uq);
917 umtxq_unbusy(&uq->uq_key);
918 umtxq_unlock(&uq->uq_key);
919
920 /*
921 * Set the contested bit so that a release in user space
922 * knows to use the system call for unlock. If this fails
923 * either some one else has acquired the lock or it has been
924 * released.
925 */
926 old = casuword(&umtx->u_owner, owner, owner | UMTX_CONTESTED);
927
928 /* The address was invalid. */
929 if (old == -1) {
930 umtxq_lock(&uq->uq_key);
931 umtxq_remove(uq);
932 umtxq_unlock(&uq->uq_key);
933 umtx_key_release(&uq->uq_key);
934 return (EFAULT);
935 }
936
937 /*
938 * We set the contested bit, sleep. Otherwise the lock changed
939 * and we need to retry or we lost a race to the thread
940 * unlocking the umtx.
941 */
942 umtxq_lock(&uq->uq_key);
943 if (old == owner)
944 error = umtxq_sleep(uq, "umtx", timeout == NULL ? NULL :
945 &timo);
946 umtxq_remove(uq);
947 umtxq_unlock(&uq->uq_key);
948 umtx_key_release(&uq->uq_key);
949
950 if (error == 0)
951 error = umtxq_check_susp(td);
952 }
953
954 if (timeout == NULL) {
955 /* Mutex locking is restarted if it is interrupted. */
956 if (error == EINTR)
957 error = ERESTART;
958 } else {
959 /* Timed-locking is not restarted. */
960 if (error == ERESTART)
961 error = EINTR;
962 }
963 return (error);
964 }
965
966 /*
967 * Unlock a umtx object.
968 */
969 static int
970 do_unlock_umtx(struct thread *td, struct umtx *umtx, u_long id)
971 {
972 struct umtx_key key;
973 u_long owner;
974 u_long old;
975 int error;
976 int count;
977
978 /*
979 * Make sure we own this mtx.
980 */
981 owner = fuword(__DEVOLATILE(u_long *, &umtx->u_owner));
982 if (owner == -1)
983 return (EFAULT);
984
985 if ((owner & ~UMTX_CONTESTED) != id)
986 return (EPERM);
987
988 /* This should be done in userland */
989 if ((owner & UMTX_CONTESTED) == 0) {
990 old = casuword(&umtx->u_owner, owner, UMTX_UNOWNED);
991 if (old == -1)
992 return (EFAULT);
993 if (old == owner)
994 return (0);
995 owner = old;
996 }
997
998 /* We should only ever be in here for contested locks */
999 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK, AUTO_SHARE,
1000 &key)) != 0)
1001 return (error);
1002
1003 umtxq_lock(&key);
1004 umtxq_busy(&key);
1005 count = umtxq_count(&key);
1006 umtxq_unlock(&key);
1007
1008 /*
1009 * When unlocking the umtx, it must be marked as unowned if
1010 * there is zero or one thread only waiting for it.
1011 * Otherwise, it must be marked as contested.
1012 */
1013 old = casuword(&umtx->u_owner, owner,
1014 count <= 1 ? UMTX_UNOWNED : UMTX_CONTESTED);
1015 umtxq_lock(&key);
1016 umtxq_signal(&key,1);
1017 umtxq_unbusy(&key);
1018 umtxq_unlock(&key);
1019 umtx_key_release(&key);
1020 if (old == -1)
1021 return (EFAULT);
1022 if (old != owner)
1023 return (EINVAL);
1024 return (0);
1025 }
1026
1027 #ifdef COMPAT_FREEBSD32
1028
1029 /*
1030 * Lock a umtx object.
1031 */
1032 static int
1033 do_lock_umtx32(struct thread *td, uint32_t *m, uint32_t id,
1034 const struct timespec *timeout)
1035 {
1036 struct abs_timeout timo;
1037 struct umtx_q *uq;
1038 uint32_t owner;
1039 uint32_t old;
1040 int error = 0;
1041
1042 uq = td->td_umtxq;
1043
1044 if (timeout != NULL)
1045 abs_timeout_init(&timo, CLOCK_REALTIME, 0, timeout);
1046
1047 /*
1048 * Care must be exercised when dealing with umtx structure. It
1049 * can fault on any access.
1050 */
1051 for (;;) {
1052 /*
1053 * Try the uncontested case. This should be done in userland.
1054 */
1055 owner = casuword32(m, UMUTEX_UNOWNED, id);
1056
1057 /* The acquire succeeded. */
1058 if (owner == UMUTEX_UNOWNED)
1059 return (0);
1060
1061 /* The address was invalid. */
1062 if (owner == -1)
1063 return (EFAULT);
1064
1065 /* If no one owns it but it is contested try to acquire it. */
1066 if (owner == UMUTEX_CONTESTED) {
1067 owner = casuword32(m,
1068 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED);
1069 if (owner == UMUTEX_CONTESTED)
1070 return (0);
1071
1072 /* The address was invalid. */
1073 if (owner == -1)
1074 return (EFAULT);
1075
1076 error = umtxq_check_susp(td);
1077 if (error != 0)
1078 break;
1079
1080 /* If this failed the lock has changed, restart. */
1081 continue;
1082 }
1083
1084 /*
1085 * If we caught a signal, we have retried and now
1086 * exit immediately.
1087 */
1088 if (error != 0)
1089 return (error);
1090
1091 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK,
1092 AUTO_SHARE, &uq->uq_key)) != 0)
1093 return (error);
1094
1095 umtxq_lock(&uq->uq_key);
1096 umtxq_busy(&uq->uq_key);
1097 umtxq_insert(uq);
1098 umtxq_unbusy(&uq->uq_key);
1099 umtxq_unlock(&uq->uq_key);
1100
1101 /*
1102 * Set the contested bit so that a release in user space
1103 * knows to use the system call for unlock. If this fails
1104 * either some one else has acquired the lock or it has been
1105 * released.
1106 */
1107 old = casuword32(m, owner, owner | UMUTEX_CONTESTED);
1108
1109 /* The address was invalid. */
1110 if (old == -1) {
1111 umtxq_lock(&uq->uq_key);
1112 umtxq_remove(uq);
1113 umtxq_unlock(&uq->uq_key);
1114 umtx_key_release(&uq->uq_key);
1115 return (EFAULT);
1116 }
1117
1118 /*
1119 * We set the contested bit, sleep. Otherwise the lock changed
1120 * and we need to retry or we lost a race to the thread
1121 * unlocking the umtx.
1122 */
1123 umtxq_lock(&uq->uq_key);
1124 if (old == owner)
1125 error = umtxq_sleep(uq, "umtx", timeout == NULL ?
1126 NULL : &timo);
1127 umtxq_remove(uq);
1128 umtxq_unlock(&uq->uq_key);
1129 umtx_key_release(&uq->uq_key);
1130
1131 if (error == 0)
1132 error = umtxq_check_susp(td);
1133 }
1134
1135 if (timeout == NULL) {
1136 /* Mutex locking is restarted if it is interrupted. */
1137 if (error == EINTR)
1138 error = ERESTART;
1139 } else {
1140 /* Timed-locking is not restarted. */
1141 if (error == ERESTART)
1142 error = EINTR;
1143 }
1144 return (error);
1145 }
1146
1147 /*
1148 * Unlock a umtx object.
1149 */
1150 static int
1151 do_unlock_umtx32(struct thread *td, uint32_t *m, uint32_t id)
1152 {
1153 struct umtx_key key;
1154 uint32_t owner;
1155 uint32_t old;
1156 int error;
1157 int count;
1158
1159 /*
1160 * Make sure we own this mtx.
1161 */
1162 owner = fuword32(m);
1163 if (owner == -1)
1164 return (EFAULT);
1165
1166 if ((owner & ~UMUTEX_CONTESTED) != id)
1167 return (EPERM);
1168
1169 /* This should be done in userland */
1170 if ((owner & UMUTEX_CONTESTED) == 0) {
1171 old = casuword32(m, owner, UMUTEX_UNOWNED);
1172 if (old == -1)
1173 return (EFAULT);
1174 if (old == owner)
1175 return (0);
1176 owner = old;
1177 }
1178
1179 /* We should only ever be in here for contested locks */
1180 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK, AUTO_SHARE,
1181 &key)) != 0)
1182 return (error);
1183
1184 umtxq_lock(&key);
1185 umtxq_busy(&key);
1186 count = umtxq_count(&key);
1187 umtxq_unlock(&key);
1188
1189 /*
1190 * When unlocking the umtx, it must be marked as unowned if
1191 * there is zero or one thread only waiting for it.
1192 * Otherwise, it must be marked as contested.
1193 */
1194 old = casuword32(m, owner,
1195 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
1196 umtxq_lock(&key);
1197 umtxq_signal(&key,1);
1198 umtxq_unbusy(&key);
1199 umtxq_unlock(&key);
1200 umtx_key_release(&key);
1201 if (old == -1)
1202 return (EFAULT);
1203 if (old != owner)
1204 return (EINVAL);
1205 return (0);
1206 }
1207 #endif
1208
1209 /*
1210 * Fetch and compare value, sleep on the address if value is not changed.
1211 */
1212 static int
1213 do_wait(struct thread *td, void *addr, u_long id,
1214 struct _umtx_time *timeout, int compat32, int is_private)
1215 {
1216 struct abs_timeout timo;
1217 struct umtx_q *uq;
1218 u_long tmp;
1219 uint32_t tmp32;
1220 int error = 0;
1221
1222 uq = td->td_umtxq;
1223 if ((error = umtx_key_get(addr, TYPE_SIMPLE_WAIT,
1224 is_private ? THREAD_SHARE : AUTO_SHARE, &uq->uq_key)) != 0)
1225 return (error);
1226
1227 if (timeout != NULL)
1228 abs_timeout_init2(&timo, timeout);
1229
1230 umtxq_lock(&uq->uq_key);
1231 umtxq_insert(uq);
1232 umtxq_unlock(&uq->uq_key);
1233 if (compat32 == 0) {
1234 error = fueword(addr, &tmp);
1235 if (error != 0)
1236 error = EFAULT;
1237 } else {
1238 error = fueword32(addr, &tmp32);
1239 if (error == 0)
1240 tmp = tmp32;
1241 else
1242 error = EFAULT;
1243 }
1244 umtxq_lock(&uq->uq_key);
1245 if (error == 0) {
1246 if (tmp == id)
1247 error = umtxq_sleep(uq, "uwait", timeout == NULL ?
1248 NULL : &timo);
1249 if ((uq->uq_flags & UQF_UMTXQ) == 0)
1250 error = 0;
1251 else
1252 umtxq_remove(uq);
1253 } else if ((uq->uq_flags & UQF_UMTXQ) != 0) {
1254 umtxq_remove(uq);
1255 }
1256 umtxq_unlock(&uq->uq_key);
1257 umtx_key_release(&uq->uq_key);
1258 if (error == ERESTART)
1259 error = EINTR;
1260 return (error);
1261 }
1262
1263 /*
1264 * Wake up threads sleeping on the specified address.
1265 */
1266 int
1267 kern_umtx_wake(struct thread *td, void *uaddr, int n_wake, int is_private)
1268 {
1269 struct umtx_key key;
1270 int ret;
1271
1272 if ((ret = umtx_key_get(uaddr, TYPE_SIMPLE_WAIT,
1273 is_private ? THREAD_SHARE : AUTO_SHARE, &key)) != 0)
1274 return (ret);
1275 umtxq_lock(&key);
1276 umtxq_signal(&key, n_wake);
1277 umtxq_unlock(&key);
1278 umtx_key_release(&key);
1279 return (0);
1280 }
1281
1282 /*
1283 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex.
1284 */
1285 static int
1286 do_lock_normal(struct thread *td, struct umutex *m, uint32_t flags,
1287 struct _umtx_time *timeout, int mode)
1288 {
1289 struct abs_timeout timo;
1290 struct umtx_q *uq;
1291 uint32_t owner, old, id;
1292 int error, rv;
1293
1294 id = td->td_tid;
1295 uq = td->td_umtxq;
1296 error = 0;
1297 if (timeout != NULL)
1298 abs_timeout_init2(&timo, timeout);
1299
1300 /*
1301 * Care must be exercised when dealing with umtx structure. It
1302 * can fault on any access.
1303 */
1304 for (;;) {
1305 rv = fueword32(&m->m_owner, &owner);
1306 if (rv == -1)
1307 return (EFAULT);
1308 if (mode == _UMUTEX_WAIT) {
1309 if (owner == UMUTEX_UNOWNED || owner == UMUTEX_CONTESTED)
1310 return (0);
1311 } else {
1312 /*
1313 * Try the uncontested case. This should be done in userland.
1314 */
1315 rv = casueword32(&m->m_owner, UMUTEX_UNOWNED,
1316 &owner, id);
1317 /* The address was invalid. */
1318 if (rv == -1)
1319 return (EFAULT);
1320
1321 /* The acquire succeeded. */
1322 if (owner == UMUTEX_UNOWNED)
1323 return (0);
1324
1325 /* If no one owns it but it is contested try to acquire it. */
1326 if (owner == UMUTEX_CONTESTED) {
1327 rv = casueword32(&m->m_owner,
1328 UMUTEX_CONTESTED, &owner,
1329 id | UMUTEX_CONTESTED);
1330 /* The address was invalid. */
1331 if (rv == -1)
1332 return (EFAULT);
1333
1334 if (owner == UMUTEX_CONTESTED)
1335 return (0);
1336
1337 rv = umtxq_check_susp(td);
1338 if (rv != 0)
1339 return (rv);
1340
1341 /* If this failed the lock has changed, restart. */
1342 continue;
1343 }
1344 }
1345
1346 if ((flags & UMUTEX_ERROR_CHECK) != 0 &&
1347 (owner & ~UMUTEX_CONTESTED) == id)
1348 return (EDEADLK);
1349
1350 if (mode == _UMUTEX_TRY)
1351 return (EBUSY);
1352
1353 /*
1354 * If we caught a signal, we have retried and now
1355 * exit immediately.
1356 */
1357 if (error != 0)
1358 return (error);
1359
1360 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX,
1361 GET_SHARE(flags), &uq->uq_key)) != 0)
1362 return (error);
1363
1364 umtxq_lock(&uq->uq_key);
1365 umtxq_busy(&uq->uq_key);
1366 umtxq_insert(uq);
1367 umtxq_unlock(&uq->uq_key);
1368
1369 /*
1370 * Set the contested bit so that a release in user space
1371 * knows to use the system call for unlock. If this fails
1372 * either some one else has acquired the lock or it has been
1373 * released.
1374 */
1375 rv = casueword32(&m->m_owner, owner, &old,
1376 owner | UMUTEX_CONTESTED);
1377
1378 /* The address was invalid. */
1379 if (rv == -1) {
1380 umtxq_lock(&uq->uq_key);
1381 umtxq_remove(uq);
1382 umtxq_unbusy(&uq->uq_key);
1383 umtxq_unlock(&uq->uq_key);
1384 umtx_key_release(&uq->uq_key);
1385 return (EFAULT);
1386 }
1387
1388 /*
1389 * We set the contested bit, sleep. Otherwise the lock changed
1390 * and we need to retry or we lost a race to the thread
1391 * unlocking the umtx.
1392 */
1393 umtxq_lock(&uq->uq_key);
1394 umtxq_unbusy(&uq->uq_key);
1395 if (old == owner)
1396 error = umtxq_sleep(uq, "umtxn", timeout == NULL ?
1397 NULL : &timo);
1398 umtxq_remove(uq);
1399 umtxq_unlock(&uq->uq_key);
1400 umtx_key_release(&uq->uq_key);
1401
1402 if (error == 0)
1403 error = umtxq_check_susp(td);
1404 }
1405
1406 return (0);
1407 }
1408
1409 /*
1410 * Unlock PTHREAD_PRIO_NONE protocol POSIX mutex.
1411 */
1412 static int
1413 do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags)
1414 {
1415 struct umtx_key key;
1416 uint32_t owner, old, id;
1417 int error;
1418 int count;
1419
1420 id = td->td_tid;
1421 /*
1422 * Make sure we own this mtx.
1423 */
1424 error = fueword32(&m->m_owner, &owner);
1425 if (error == -1)
1426 return (EFAULT);
1427
1428 if ((owner & ~UMUTEX_CONTESTED) != id)
1429 return (EPERM);
1430
1431 if ((owner & UMUTEX_CONTESTED) == 0) {
1432 error = casueword32(&m->m_owner, owner, &old, UMUTEX_UNOWNED);
1433 if (error == -1)
1434 return (EFAULT);
1435 if (old == owner)
1436 return (0);
1437 owner = old;
1438 }
1439
1440 /* We should only ever be in here for contested locks */
1441 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
1442 &key)) != 0)
1443 return (error);
1444
1445 umtxq_lock(&key);
1446 umtxq_busy(&key);
1447 count = umtxq_count(&key);
1448 umtxq_unlock(&key);
1449
1450 /*
1451 * When unlocking the umtx, it must be marked as unowned if
1452 * there is zero or one thread only waiting for it.
1453 * Otherwise, it must be marked as contested.
1454 */
1455 error = casueword32(&m->m_owner, owner, &old,
1456 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
1457 umtxq_lock(&key);
1458 umtxq_signal(&key,1);
1459 umtxq_unbusy(&key);
1460 umtxq_unlock(&key);
1461 umtx_key_release(&key);
1462 if (error == -1)
1463 return (EFAULT);
1464 if (old != owner)
1465 return (EINVAL);
1466 return (0);
1467 }
1468
1469 /*
1470 * Check if the mutex is available and wake up a waiter,
1471 * only for simple mutex.
1472 */
1473 static int
1474 do_wake_umutex(struct thread *td, struct umutex *m)
1475 {
1476 struct umtx_key key;
1477 uint32_t owner;
1478 uint32_t flags;
1479 int error;
1480 int count;
1481
1482 error = fueword32(&m->m_owner, &owner);
1483 if (error == -1)
1484 return (EFAULT);
1485
1486 if ((owner & ~UMUTEX_CONTESTED) != 0)
1487 return (0);
1488
1489 error = fueword32(&m->m_flags, &flags);
1490 if (error == -1)
1491 return (EFAULT);
1492
1493 /* We should only ever be in here for contested locks */
1494 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags),
1495 &key)) != 0)
1496 return (error);
1497
1498 umtxq_lock(&key);
1499 umtxq_busy(&key);
1500 count = umtxq_count(&key);
1501 umtxq_unlock(&key);
1502
1503 if (count <= 1) {
1504 error = casueword32(&m->m_owner, UMUTEX_CONTESTED, &owner,
1505 UMUTEX_UNOWNED);
1506 if (error == -1)
1507 error = EFAULT;
1508 }
1509
1510 umtxq_lock(&key);
1511 if (error == 0 && count != 0 && (owner & ~UMUTEX_CONTESTED) == 0)
1512 umtxq_signal(&key, 1);
1513 umtxq_unbusy(&key);
1514 umtxq_unlock(&key);
1515 umtx_key_release(&key);
1516 return (error);
1517 }
1518
1519 /*
1520 * Check if the mutex has waiters and tries to fix contention bit.
1521 */
1522 static int
1523 do_wake2_umutex(struct thread *td, struct umutex *m, uint32_t flags)
1524 {
1525 struct umtx_key key;
1526 uint32_t owner, old;
1527 int type;
1528 int error;
1529 int count;
1530
1531 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
1532 case 0:
1533 type = TYPE_NORMAL_UMUTEX;
1534 break;
1535 case UMUTEX_PRIO_INHERIT:
1536 type = TYPE_PI_UMUTEX;
1537 break;
1538 case UMUTEX_PRIO_PROTECT:
1539 type = TYPE_PP_UMUTEX;
1540 break;
1541 default:
1542 return (EINVAL);
1543 }
1544 if ((error = umtx_key_get(m, type, GET_SHARE(flags),
1545 &key)) != 0)
1546 return (error);
1547
1548 owner = 0;
1549 umtxq_lock(&key);
1550 umtxq_busy(&key);
1551 count = umtxq_count(&key);
1552 umtxq_unlock(&key);
1553 /*
1554 * Only repair contention bit if there is a waiter, this means the mutex
1555 * is still being referenced by userland code, otherwise don't update
1556 * any memory.
1557 */
1558 if (count > 1) {
1559 error = fueword32(&m->m_owner, &owner);
1560 if (error == -1)
1561 error = EFAULT;
1562 while (error == 0 && (owner & UMUTEX_CONTESTED) == 0) {
1563 error = casueword32(&m->m_owner, owner, &old,
1564 owner | UMUTEX_CONTESTED);
1565 if (error == -1) {
1566 error = EFAULT;
1567 break;
1568 }
1569 if (old == owner)
1570 break;
1571 owner = old;
1572 error = umtxq_check_susp(td);
1573 if (error != 0)
1574 break;
1575 }
1576 } else if (count == 1) {
1577 error = fueword32(&m->m_owner, &owner);
1578 if (error == -1)
1579 error = EFAULT;
1580 while (error == 0 && (owner & ~UMUTEX_CONTESTED) != 0 &&
1581 (owner & UMUTEX_CONTESTED) == 0) {
1582 error = casueword32(&m->m_owner, owner, &old,
1583 owner | UMUTEX_CONTESTED);
1584 if (error == -1) {
1585 error = EFAULT;
1586 break;
1587 }
1588 if (old == owner)
1589 break;
1590 owner = old;
1591 error = umtxq_check_susp(td);
1592 if (error != 0)
1593 break;
1594 }
1595 }
1596 umtxq_lock(&key);
1597 if (error == EFAULT) {
1598 umtxq_signal(&key, INT_MAX);
1599 } else if (count != 0 && (owner & ~UMUTEX_CONTESTED) == 0)
1600 umtxq_signal(&key, 1);
1601 umtxq_unbusy(&key);
1602 umtxq_unlock(&key);
1603 umtx_key_release(&key);
1604 return (error);
1605 }
1606
1607 static inline struct umtx_pi *
1608 umtx_pi_alloc(int flags)
1609 {
1610 struct umtx_pi *pi;
1611
1612 pi = uma_zalloc(umtx_pi_zone, M_ZERO | flags);
1613 TAILQ_INIT(&pi->pi_blocked);
1614 atomic_add_int(&umtx_pi_allocated, 1);
1615 return (pi);
1616 }
1617
1618 static inline void
1619 umtx_pi_free(struct umtx_pi *pi)
1620 {
1621 uma_zfree(umtx_pi_zone, pi);
1622 atomic_add_int(&umtx_pi_allocated, -1);
1623 }
1624
1625 /*
1626 * Adjust the thread's position on a pi_state after its priority has been
1627 * changed.
1628 */
1629 static int
1630 umtx_pi_adjust_thread(struct umtx_pi *pi, struct thread *td)
1631 {
1632 struct umtx_q *uq, *uq1, *uq2;
1633 struct thread *td1;
1634
1635 mtx_assert(&umtx_lock, MA_OWNED);
1636 if (pi == NULL)
1637 return (0);
1638
1639 uq = td->td_umtxq;
1640
1641 /*
1642 * Check if the thread needs to be moved on the blocked chain.
1643 * It needs to be moved if either its priority is lower than
1644 * the previous thread or higher than the next thread.
1645 */
1646 uq1 = TAILQ_PREV(uq, umtxq_head, uq_lockq);
1647 uq2 = TAILQ_NEXT(uq, uq_lockq);
1648 if ((uq1 != NULL && UPRI(td) < UPRI(uq1->uq_thread)) ||
1649 (uq2 != NULL && UPRI(td) > UPRI(uq2->uq_thread))) {
1650 /*
1651 * Remove thread from blocked chain and determine where
1652 * it should be moved to.
1653 */
1654 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
1655 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
1656 td1 = uq1->uq_thread;
1657 MPASS(td1->td_proc->p_magic == P_MAGIC);
1658 if (UPRI(td1) > UPRI(td))
1659 break;
1660 }
1661
1662 if (uq1 == NULL)
1663 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
1664 else
1665 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
1666 }
1667 return (1);
1668 }
1669
1670 static struct umtx_pi *
1671 umtx_pi_next(struct umtx_pi *pi)
1672 {
1673 struct umtx_q *uq_owner;
1674
1675 if (pi->pi_owner == NULL)
1676 return (NULL);
1677 uq_owner = pi->pi_owner->td_umtxq;
1678 if (uq_owner == NULL)
1679 return (NULL);
1680 return (uq_owner->uq_pi_blocked);
1681 }
1682
1683 /*
1684 * Floyd's Cycle-Finding Algorithm.
1685 */
1686 static bool
1687 umtx_pi_check_loop(struct umtx_pi *pi)
1688 {
1689 struct umtx_pi *pi1; /* fast iterator */
1690
1691 mtx_assert(&umtx_lock, MA_OWNED);
1692 if (pi == NULL)
1693 return (false);
1694 pi1 = pi;
1695 for (;;) {
1696 pi = umtx_pi_next(pi);
1697 if (pi == NULL)
1698 break;
1699 pi1 = umtx_pi_next(pi1);
1700 if (pi1 == NULL)
1701 break;
1702 pi1 = umtx_pi_next(pi1);
1703 if (pi1 == NULL)
1704 break;
1705 if (pi == pi1)
1706 return (true);
1707 }
1708 return (false);
1709 }
1710
1711 /*
1712 * Propagate priority when a thread is blocked on POSIX
1713 * PI mutex.
1714 */
1715 static void
1716 umtx_propagate_priority(struct thread *td)
1717 {
1718 struct umtx_q *uq;
1719 struct umtx_pi *pi;
1720 int pri;
1721
1722 mtx_assert(&umtx_lock, MA_OWNED);
1723 pri = UPRI(td);
1724 uq = td->td_umtxq;
1725 pi = uq->uq_pi_blocked;
1726 if (pi == NULL)
1727 return;
1728 if (umtx_pi_check_loop(pi))
1729 return;
1730
1731 for (;;) {
1732 td = pi->pi_owner;
1733 if (td == NULL || td == curthread)
1734 return;
1735
1736 MPASS(td->td_proc != NULL);
1737 MPASS(td->td_proc->p_magic == P_MAGIC);
1738
1739 thread_lock(td);
1740 if (td->td_lend_user_pri > pri)
1741 sched_lend_user_prio(td, pri);
1742 else {
1743 thread_unlock(td);
1744 break;
1745 }
1746 thread_unlock(td);
1747
1748 /*
1749 * Pick up the lock that td is blocked on.
1750 */
1751 uq = td->td_umtxq;
1752 pi = uq->uq_pi_blocked;
1753 if (pi == NULL)
1754 break;
1755 /* Resort td on the list if needed. */
1756 umtx_pi_adjust_thread(pi, td);
1757 }
1758 }
1759
1760 /*
1761 * Unpropagate priority for a PI mutex when a thread blocked on
1762 * it is interrupted by signal or resumed by others.
1763 */
1764 static void
1765 umtx_repropagate_priority(struct umtx_pi *pi)
1766 {
1767 struct umtx_q *uq, *uq_owner;
1768 struct umtx_pi *pi2;
1769 int pri;
1770
1771 mtx_assert(&umtx_lock, MA_OWNED);
1772
1773 if (umtx_pi_check_loop(pi))
1774 return;
1775 while (pi != NULL && pi->pi_owner != NULL) {
1776 pri = PRI_MAX;
1777 uq_owner = pi->pi_owner->td_umtxq;
1778
1779 TAILQ_FOREACH(pi2, &uq_owner->uq_pi_contested, pi_link) {
1780 uq = TAILQ_FIRST(&pi2->pi_blocked);
1781 if (uq != NULL) {
1782 if (pri > UPRI(uq->uq_thread))
1783 pri = UPRI(uq->uq_thread);
1784 }
1785 }
1786
1787 if (pri > uq_owner->uq_inherited_pri)
1788 pri = uq_owner->uq_inherited_pri;
1789 thread_lock(pi->pi_owner);
1790 sched_lend_user_prio(pi->pi_owner, pri);
1791 thread_unlock(pi->pi_owner);
1792 if ((pi = uq_owner->uq_pi_blocked) != NULL)
1793 umtx_pi_adjust_thread(pi, uq_owner->uq_thread);
1794 }
1795 }
1796
1797 /*
1798 * Insert a PI mutex into owned list.
1799 */
1800 static void
1801 umtx_pi_setowner(struct umtx_pi *pi, struct thread *owner)
1802 {
1803 struct umtx_q *uq_owner;
1804
1805 uq_owner = owner->td_umtxq;
1806 mtx_assert(&umtx_lock, MA_OWNED);
1807 if (pi->pi_owner != NULL)
1808 panic("pi_owner != NULL");
1809 pi->pi_owner = owner;
1810 TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link);
1811 }
1812
1813
1814 /*
1815 * Disown a PI mutex, and remove it from the owned list.
1816 */
1817 static void
1818 umtx_pi_disown(struct umtx_pi *pi)
1819 {
1820
1821 mtx_assert(&umtx_lock, MA_OWNED);
1822 TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested, pi, pi_link);
1823 pi->pi_owner = NULL;
1824 }
1825
1826 /*
1827 * Claim ownership of a PI mutex.
1828 */
1829 static int
1830 umtx_pi_claim(struct umtx_pi *pi, struct thread *owner)
1831 {
1832 struct umtx_q *uq;
1833
1834 mtx_lock(&umtx_lock);
1835 if (pi->pi_owner == owner) {
1836 mtx_unlock(&umtx_lock);
1837 return (0);
1838 }
1839
1840 if (pi->pi_owner != NULL) {
1841 /*
1842 * userland may have already messed the mutex, sigh.
1843 */
1844 mtx_unlock(&umtx_lock);
1845 return (EPERM);
1846 }
1847 umtx_pi_setowner(pi, owner);
1848 uq = TAILQ_FIRST(&pi->pi_blocked);
1849 if (uq != NULL) {
1850 int pri;
1851
1852 pri = UPRI(uq->uq_thread);
1853 thread_lock(owner);
1854 if (pri < UPRI(owner))
1855 sched_lend_user_prio(owner, pri);
1856 thread_unlock(owner);
1857 }
1858 mtx_unlock(&umtx_lock);
1859 return (0);
1860 }
1861
1862 /*
1863 * Adjust a thread's order position in its blocked PI mutex,
1864 * this may result new priority propagating process.
1865 */
1866 void
1867 umtx_pi_adjust(struct thread *td, u_char oldpri)
1868 {
1869 struct umtx_q *uq;
1870 struct umtx_pi *pi;
1871
1872 uq = td->td_umtxq;
1873 mtx_lock(&umtx_lock);
1874 /*
1875 * Pick up the lock that td is blocked on.
1876 */
1877 pi = uq->uq_pi_blocked;
1878 if (pi != NULL) {
1879 umtx_pi_adjust_thread(pi, td);
1880 umtx_repropagate_priority(pi);
1881 }
1882 mtx_unlock(&umtx_lock);
1883 }
1884
1885 /*
1886 * Sleep on a PI mutex.
1887 */
1888 static int
1889 umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi,
1890 uint32_t owner, const char *wmesg, struct abs_timeout *timo)
1891 {
1892 struct umtxq_chain *uc;
1893 struct thread *td, *td1;
1894 struct umtx_q *uq1;
1895 int pri;
1896 int error = 0;
1897
1898 td = uq->uq_thread;
1899 KASSERT(td == curthread, ("inconsistent uq_thread"));
1900 uc = umtxq_getchain(&uq->uq_key);
1901 UMTXQ_LOCKED_ASSERT(uc);
1902 KASSERT(uc->uc_busy != 0, ("umtx chain is not busy"));
1903 umtxq_insert(uq);
1904 mtx_lock(&umtx_lock);
1905 if (pi->pi_owner == NULL) {
1906 mtx_unlock(&umtx_lock);
1907 /* XXX Only look up thread in current process. */
1908 td1 = tdfind(owner, curproc->p_pid);
1909 mtx_lock(&umtx_lock);
1910 if (td1 != NULL) {
1911 if (pi->pi_owner == NULL)
1912 umtx_pi_setowner(pi, td1);
1913 PROC_UNLOCK(td1->td_proc);
1914 }
1915 }
1916
1917 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) {
1918 pri = UPRI(uq1->uq_thread);
1919 if (pri > UPRI(td))
1920 break;
1921 }
1922
1923 if (uq1 != NULL)
1924 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq);
1925 else
1926 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq);
1927
1928 uq->uq_pi_blocked = pi;
1929 thread_lock(td);
1930 td->td_flags |= TDF_UPIBLOCKED;
1931 thread_unlock(td);
1932 umtx_propagate_priority(td);
1933 mtx_unlock(&umtx_lock);
1934 umtxq_unbusy(&uq->uq_key);
1935
1936 error = umtxq_sleep(uq, wmesg, timo);
1937 umtxq_remove(uq);
1938
1939 mtx_lock(&umtx_lock);
1940 uq->uq_pi_blocked = NULL;
1941 thread_lock(td);
1942 td->td_flags &= ~TDF_UPIBLOCKED;
1943 thread_unlock(td);
1944 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq);
1945 umtx_repropagate_priority(pi);
1946 mtx_unlock(&umtx_lock);
1947 umtxq_unlock(&uq->uq_key);
1948
1949 return (error);
1950 }
1951
1952 /*
1953 * Add reference count for a PI mutex.
1954 */
1955 static void
1956 umtx_pi_ref(struct umtx_pi *pi)
1957 {
1958 struct umtxq_chain *uc;
1959
1960 uc = umtxq_getchain(&pi->pi_key);
1961 UMTXQ_LOCKED_ASSERT(uc);
1962 pi->pi_refcount++;
1963 }
1964
1965 /*
1966 * Decrease reference count for a PI mutex, if the counter
1967 * is decreased to zero, its memory space is freed.
1968 */
1969 static void
1970 umtx_pi_unref(struct umtx_pi *pi)
1971 {
1972 struct umtxq_chain *uc;
1973
1974 uc = umtxq_getchain(&pi->pi_key);
1975 UMTXQ_LOCKED_ASSERT(uc);
1976 KASSERT(pi->pi_refcount > 0, ("invalid reference count"));
1977 if (--pi->pi_refcount == 0) {
1978 mtx_lock(&umtx_lock);
1979 if (pi->pi_owner != NULL)
1980 umtx_pi_disown(pi);
1981 KASSERT(TAILQ_EMPTY(&pi->pi_blocked),
1982 ("blocked queue not empty"));
1983 mtx_unlock(&umtx_lock);
1984 TAILQ_REMOVE(&uc->uc_pi_list, pi, pi_hashlink);
1985 umtx_pi_free(pi);
1986 }
1987 }
1988
1989 /*
1990 * Find a PI mutex in hash table.
1991 */
1992 static struct umtx_pi *
1993 umtx_pi_lookup(struct umtx_key *key)
1994 {
1995 struct umtxq_chain *uc;
1996 struct umtx_pi *pi;
1997
1998 uc = umtxq_getchain(key);
1999 UMTXQ_LOCKED_ASSERT(uc);
2000
2001 TAILQ_FOREACH(pi, &uc->uc_pi_list, pi_hashlink) {
2002 if (umtx_key_match(&pi->pi_key, key)) {
2003 return (pi);
2004 }
2005 }
2006 return (NULL);
2007 }
2008
2009 /*
2010 * Insert a PI mutex into hash table.
2011 */
2012 static inline void
2013 umtx_pi_insert(struct umtx_pi *pi)
2014 {
2015 struct umtxq_chain *uc;
2016
2017 uc = umtxq_getchain(&pi->pi_key);
2018 UMTXQ_LOCKED_ASSERT(uc);
2019 TAILQ_INSERT_TAIL(&uc->uc_pi_list, pi, pi_hashlink);
2020 }
2021
2022 /*
2023 * Lock a PI mutex.
2024 */
2025 static int
2026 do_lock_pi(struct thread *td, struct umutex *m, uint32_t flags,
2027 struct _umtx_time *timeout, int try)
2028 {
2029 struct abs_timeout timo;
2030 struct umtx_q *uq;
2031 struct umtx_pi *pi, *new_pi;
2032 uint32_t id, owner, old;
2033 int error, rv;
2034
2035 id = td->td_tid;
2036 uq = td->td_umtxq;
2037
2038 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags),
2039 &uq->uq_key)) != 0)
2040 return (error);
2041
2042 if (timeout != NULL)
2043 abs_timeout_init2(&timo, timeout);
2044
2045 umtxq_lock(&uq->uq_key);
2046 pi = umtx_pi_lookup(&uq->uq_key);
2047 if (pi == NULL) {
2048 new_pi = umtx_pi_alloc(M_NOWAIT);
2049 if (new_pi == NULL) {
2050 umtxq_unlock(&uq->uq_key);
2051 new_pi = umtx_pi_alloc(M_WAITOK);
2052 umtxq_lock(&uq->uq_key);
2053 pi = umtx_pi_lookup(&uq->uq_key);
2054 if (pi != NULL) {
2055 umtx_pi_free(new_pi);
2056 new_pi = NULL;
2057 }
2058 }
2059 if (new_pi != NULL) {
2060 new_pi->pi_key = uq->uq_key;
2061 umtx_pi_insert(new_pi);
2062 pi = new_pi;
2063 }
2064 }
2065 umtx_pi_ref(pi);
2066 umtxq_unlock(&uq->uq_key);
2067
2068 /*
2069 * Care must be exercised when dealing with umtx structure. It
2070 * can fault on any access.
2071 */
2072 for (;;) {
2073 /*
2074 * Try the uncontested case. This should be done in userland.
2075 */
2076 rv = casueword32(&m->m_owner, UMUTEX_UNOWNED, &owner, id);
2077 /* The address was invalid. */
2078 if (rv == -1) {
2079 error = EFAULT;
2080 break;
2081 }
2082
2083 /* The acquire succeeded. */
2084 if (owner == UMUTEX_UNOWNED) {
2085 error = 0;
2086 break;
2087 }
2088
2089 /* If no one owns it but it is contested try to acquire it. */
2090 if (owner == UMUTEX_CONTESTED) {
2091 rv = casueword32(&m->m_owner,
2092 UMUTEX_CONTESTED, &owner, id | UMUTEX_CONTESTED);
2093 /* The address was invalid. */
2094 if (rv == -1) {
2095 error = EFAULT;
2096 break;
2097 }
2098
2099 if (owner == UMUTEX_CONTESTED) {
2100 umtxq_lock(&uq->uq_key);
2101 umtxq_busy(&uq->uq_key);
2102 error = umtx_pi_claim(pi, td);
2103 umtxq_unbusy(&uq->uq_key);
2104 umtxq_unlock(&uq->uq_key);
2105 if (error != 0) {
2106 /*
2107 * Since we're going to return an
2108 * error, restore the m_owner to its
2109 * previous, unowned state to avoid
2110 * compounding the problem.
2111 */
2112 (void)casuword32(&m->m_owner,
2113 id | UMUTEX_CONTESTED,
2114 UMUTEX_CONTESTED);
2115 }
2116 break;
2117 }
2118
2119 error = umtxq_check_susp(td);
2120 if (error != 0)
2121 break;
2122
2123 /* If this failed the lock has changed, restart. */
2124 continue;
2125 }
2126
2127 if ((owner & ~UMUTEX_CONTESTED) == id) {
2128 error = EDEADLK;
2129 break;
2130 }
2131
2132 if (try != 0) {
2133 error = EBUSY;
2134 break;
2135 }
2136
2137 /*
2138 * If we caught a signal, we have retried and now
2139 * exit immediately.
2140 */
2141 if (error != 0)
2142 break;
2143
2144 umtxq_lock(&uq->uq_key);
2145 umtxq_busy(&uq->uq_key);
2146 umtxq_unlock(&uq->uq_key);
2147
2148 /*
2149 * Set the contested bit so that a release in user space
2150 * knows to use the system call for unlock. If this fails
2151 * either some one else has acquired the lock or it has been
2152 * released.
2153 */
2154 rv = casueword32(&m->m_owner, owner, &old,
2155 owner | UMUTEX_CONTESTED);
2156
2157 /* The address was invalid. */
2158 if (rv == -1) {
2159 umtxq_unbusy_unlocked(&uq->uq_key);
2160 error = EFAULT;
2161 break;
2162 }
2163
2164 umtxq_lock(&uq->uq_key);
2165 /*
2166 * We set the contested bit, sleep. Otherwise the lock changed
2167 * and we need to retry or we lost a race to the thread
2168 * unlocking the umtx.
2169 */
2170 if (old == owner) {
2171 error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED,
2172 "umtxpi", timeout == NULL ? NULL : &timo);
2173 if (error != 0)
2174 continue;
2175 } else {
2176 umtxq_unbusy(&uq->uq_key);
2177 umtxq_unlock(&uq->uq_key);
2178 }
2179
2180 error = umtxq_check_susp(td);
2181 if (error != 0)
2182 break;
2183 }
2184
2185 umtxq_lock(&uq->uq_key);
2186 umtx_pi_unref(pi);
2187 umtxq_unlock(&uq->uq_key);
2188
2189 umtx_key_release(&uq->uq_key);
2190 return (error);
2191 }
2192
2193 /*
2194 * Unlock a PI mutex.
2195 */
2196 static int
2197 do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags)
2198 {
2199 struct umtx_key key;
2200 struct umtx_q *uq_first, *uq_first2, *uq_me;
2201 struct umtx_pi *pi, *pi2;
2202 uint32_t owner, old, id;
2203 int error;
2204 int count;
2205 int pri;
2206
2207 id = td->td_tid;
2208 /*
2209 * Make sure we own this mtx.
2210 */
2211 error = fueword32(&m->m_owner, &owner);
2212 if (error == -1)
2213 return (EFAULT);
2214
2215 if ((owner & ~UMUTEX_CONTESTED) != id)
2216 return (EPERM);
2217
2218 /* This should be done in userland */
2219 if ((owner & UMUTEX_CONTESTED) == 0) {
2220 error = casueword32(&m->m_owner, owner, &old, UMUTEX_UNOWNED);
2221 if (error == -1)
2222 return (EFAULT);
2223 if (old == owner)
2224 return (0);
2225 owner = old;
2226 }
2227
2228 /* We should only ever be in here for contested locks */
2229 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags),
2230 &key)) != 0)
2231 return (error);
2232
2233 umtxq_lock(&key);
2234 umtxq_busy(&key);
2235 count = umtxq_count_pi(&key, &uq_first);
2236 if (uq_first != NULL) {
2237 mtx_lock(&umtx_lock);
2238 pi = uq_first->uq_pi_blocked;
2239 KASSERT(pi != NULL, ("pi == NULL?"));
2240 if (pi->pi_owner != td) {
2241 mtx_unlock(&umtx_lock);
2242 umtxq_unbusy(&key);
2243 umtxq_unlock(&key);
2244 umtx_key_release(&key);
2245 /* userland messed the mutex */
2246 return (EPERM);
2247 }
2248 uq_me = td->td_umtxq;
2249 umtx_pi_disown(pi);
2250 /* get highest priority thread which is still sleeping. */
2251 uq_first = TAILQ_FIRST(&pi->pi_blocked);
2252 while (uq_first != NULL &&
2253 (uq_first->uq_flags & UQF_UMTXQ) == 0) {
2254 uq_first = TAILQ_NEXT(uq_first, uq_lockq);
2255 }
2256 pri = PRI_MAX;
2257 TAILQ_FOREACH(pi2, &uq_me->uq_pi_contested, pi_link) {
2258 uq_first2 = TAILQ_FIRST(&pi2->pi_blocked);
2259 if (uq_first2 != NULL) {
2260 if (pri > UPRI(uq_first2->uq_thread))
2261 pri = UPRI(uq_first2->uq_thread);
2262 }
2263 }
2264 thread_lock(td);
2265 sched_lend_user_prio(td, pri);
2266 thread_unlock(td);
2267 mtx_unlock(&umtx_lock);
2268 if (uq_first)
2269 umtxq_signal_thread(uq_first);
2270 } else {
2271 pi = umtx_pi_lookup(&key);
2272 /*
2273 * A umtx_pi can exist if a signal or timeout removed the
2274 * last waiter from the umtxq, but there is still
2275 * a thread in do_lock_pi() holding the umtx_pi.
2276 */
2277 if (pi != NULL) {
2278 /*
2279 * The umtx_pi can be unowned, such as when a thread
2280 * has just entered do_lock_pi(), allocated the
2281 * umtx_pi, and unlocked the umtxq.
2282 * If the current thread owns it, it must disown it.
2283 */
2284 mtx_lock(&umtx_lock);
2285 if (pi->pi_owner == td)
2286 umtx_pi_disown(pi);
2287 mtx_unlock(&umtx_lock);
2288 }
2289 }
2290 umtxq_unlock(&key);
2291
2292 /*
2293 * When unlocking the umtx, it must be marked as unowned if
2294 * there is zero or one thread only waiting for it.
2295 * Otherwise, it must be marked as contested.
2296 */
2297 error = casueword32(&m->m_owner, owner, &old,
2298 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED);
2299
2300 umtxq_unbusy_unlocked(&key);
2301 umtx_key_release(&key);
2302 if (error == -1)
2303 return (EFAULT);
2304 if (old != owner)
2305 return (EINVAL);
2306 return (0);
2307 }
2308
2309 /*
2310 * Lock a PP mutex.
2311 */
2312 static int
2313 do_lock_pp(struct thread *td, struct umutex *m, uint32_t flags,
2314 struct _umtx_time *timeout, int try)
2315 {
2316 struct abs_timeout timo;
2317 struct umtx_q *uq, *uq2;
2318 struct umtx_pi *pi;
2319 uint32_t ceiling;
2320 uint32_t owner, id;
2321 int error, pri, old_inherited_pri, su, rv;
2322
2323 id = td->td_tid;
2324 uq = td->td_umtxq;
2325 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
2326 &uq->uq_key)) != 0)
2327 return (error);
2328
2329 if (timeout != NULL)
2330 abs_timeout_init2(&timo, timeout);
2331
2332 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
2333 for (;;) {
2334 old_inherited_pri = uq->uq_inherited_pri;
2335 umtxq_lock(&uq->uq_key);
2336 umtxq_busy(&uq->uq_key);
2337 umtxq_unlock(&uq->uq_key);
2338
2339 rv = fueword32(&m->m_ceilings[0], &ceiling);
2340 if (rv == -1) {
2341 error = EFAULT;
2342 goto out;
2343 }
2344 ceiling = RTP_PRIO_MAX - ceiling;
2345 if (ceiling > RTP_PRIO_MAX) {
2346 error = EINVAL;
2347 goto out;
2348 }
2349
2350 mtx_lock(&umtx_lock);
2351 if (UPRI(td) < PRI_MIN_REALTIME + ceiling) {
2352 mtx_unlock(&umtx_lock);
2353 error = EINVAL;
2354 goto out;
2355 }
2356 if (su && PRI_MIN_REALTIME + ceiling < uq->uq_inherited_pri) {
2357 uq->uq_inherited_pri = PRI_MIN_REALTIME + ceiling;
2358 thread_lock(td);
2359 if (uq->uq_inherited_pri < UPRI(td))
2360 sched_lend_user_prio(td, uq->uq_inherited_pri);
2361 thread_unlock(td);
2362 }
2363 mtx_unlock(&umtx_lock);
2364
2365 rv = casueword32(&m->m_owner,
2366 UMUTEX_CONTESTED, &owner, id | UMUTEX_CONTESTED);
2367 /* The address was invalid. */
2368 if (rv == -1) {
2369 error = EFAULT;
2370 break;
2371 }
2372
2373 if (owner == UMUTEX_CONTESTED) {
2374 error = 0;
2375 break;
2376 }
2377
2378 if ((flags & UMUTEX_ERROR_CHECK) != 0 &&
2379 (owner & ~UMUTEX_CONTESTED) == id) {
2380 error = EDEADLK;
2381 break;
2382 }
2383
2384 if (try != 0) {
2385 error = EBUSY;
2386 break;
2387 }
2388
2389 /*
2390 * If we caught a signal, we have retried and now
2391 * exit immediately.
2392 */
2393 if (error != 0)
2394 break;
2395
2396 umtxq_lock(&uq->uq_key);
2397 umtxq_insert(uq);
2398 umtxq_unbusy(&uq->uq_key);
2399 error = umtxq_sleep(uq, "umtxpp", timeout == NULL ?
2400 NULL : &timo);
2401 umtxq_remove(uq);
2402 umtxq_unlock(&uq->uq_key);
2403
2404 mtx_lock(&umtx_lock);
2405 uq->uq_inherited_pri = old_inherited_pri;
2406 pri = PRI_MAX;
2407 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2408 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2409 if (uq2 != NULL) {
2410 if (pri > UPRI(uq2->uq_thread))
2411 pri = UPRI(uq2->uq_thread);
2412 }
2413 }
2414 if (pri > uq->uq_inherited_pri)
2415 pri = uq->uq_inherited_pri;
2416 thread_lock(td);
2417 sched_lend_user_prio(td, pri);
2418 thread_unlock(td);
2419 mtx_unlock(&umtx_lock);
2420 }
2421
2422 if (error != 0) {
2423 mtx_lock(&umtx_lock);
2424 uq->uq_inherited_pri = old_inherited_pri;
2425 pri = PRI_MAX;
2426 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2427 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2428 if (uq2 != NULL) {
2429 if (pri > UPRI(uq2->uq_thread))
2430 pri = UPRI(uq2->uq_thread);
2431 }
2432 }
2433 if (pri > uq->uq_inherited_pri)
2434 pri = uq->uq_inherited_pri;
2435 thread_lock(td);
2436 sched_lend_user_prio(td, pri);
2437 thread_unlock(td);
2438 mtx_unlock(&umtx_lock);
2439 }
2440
2441 out:
2442 umtxq_unbusy_unlocked(&uq->uq_key);
2443 umtx_key_release(&uq->uq_key);
2444 return (error);
2445 }
2446
2447 /*
2448 * Unlock a PP mutex.
2449 */
2450 static int
2451 do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags)
2452 {
2453 struct umtx_key key;
2454 struct umtx_q *uq, *uq2;
2455 struct umtx_pi *pi;
2456 uint32_t owner, id;
2457 uint32_t rceiling;
2458 int error, pri, new_inherited_pri, su;
2459
2460 id = td->td_tid;
2461 uq = td->td_umtxq;
2462 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0);
2463
2464 /*
2465 * Make sure we own this mtx.
2466 */
2467 error = fueword32(&m->m_owner, &owner);
2468 if (error == -1)
2469 return (EFAULT);
2470
2471 if ((owner & ~UMUTEX_CONTESTED) != id)
2472 return (EPERM);
2473
2474 error = copyin(&m->m_ceilings[1], &rceiling, sizeof(uint32_t));
2475 if (error != 0)
2476 return (error);
2477
2478 if (rceiling == -1)
2479 new_inherited_pri = PRI_MAX;
2480 else {
2481 rceiling = RTP_PRIO_MAX - rceiling;
2482 if (rceiling > RTP_PRIO_MAX)
2483 return (EINVAL);
2484 new_inherited_pri = PRI_MIN_REALTIME + rceiling;
2485 }
2486
2487 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
2488 &key)) != 0)
2489 return (error);
2490 umtxq_lock(&key);
2491 umtxq_busy(&key);
2492 umtxq_unlock(&key);
2493 /*
2494 * For priority protected mutex, always set unlocked state
2495 * to UMUTEX_CONTESTED, so that userland always enters kernel
2496 * to lock the mutex, it is necessary because thread priority
2497 * has to be adjusted for such mutex.
2498 */
2499 error = suword32(&m->m_owner, UMUTEX_CONTESTED);
2500
2501 umtxq_lock(&key);
2502 if (error == 0)
2503 umtxq_signal(&key, 1);
2504 umtxq_unbusy(&key);
2505 umtxq_unlock(&key);
2506
2507 if (error == -1)
2508 error = EFAULT;
2509 else {
2510 mtx_lock(&umtx_lock);
2511 if (su != 0)
2512 uq->uq_inherited_pri = new_inherited_pri;
2513 pri = PRI_MAX;
2514 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) {
2515 uq2 = TAILQ_FIRST(&pi->pi_blocked);
2516 if (uq2 != NULL) {
2517 if (pri > UPRI(uq2->uq_thread))
2518 pri = UPRI(uq2->uq_thread);
2519 }
2520 }
2521 if (pri > uq->uq_inherited_pri)
2522 pri = uq->uq_inherited_pri;
2523 thread_lock(td);
2524 sched_lend_user_prio(td, pri);
2525 thread_unlock(td);
2526 mtx_unlock(&umtx_lock);
2527 }
2528 umtx_key_release(&key);
2529 return (error);
2530 }
2531
2532 static int
2533 do_set_ceiling(struct thread *td, struct umutex *m, uint32_t ceiling,
2534 uint32_t *old_ceiling)
2535 {
2536 struct umtx_q *uq;
2537 uint32_t save_ceiling;
2538 uint32_t owner, id;
2539 uint32_t flags;
2540 int error, rv;
2541
2542 error = fueword32(&m->m_flags, &flags);
2543 if (error == -1)
2544 return (EFAULT);
2545 if ((flags & UMUTEX_PRIO_PROTECT) == 0)
2546 return (EINVAL);
2547 if (ceiling > RTP_PRIO_MAX)
2548 return (EINVAL);
2549 id = td->td_tid;
2550 uq = td->td_umtxq;
2551 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags),
2552 &uq->uq_key)) != 0)
2553 return (error);
2554 for (;;) {
2555 umtxq_lock(&uq->uq_key);
2556 umtxq_busy(&uq->uq_key);
2557 umtxq_unlock(&uq->uq_key);
2558
2559 rv = fueword32(&m->m_ceilings[0], &save_ceiling);
2560 if (rv == -1) {
2561 error = EFAULT;
2562 break;
2563 }
2564
2565 rv = casueword32(&m->m_owner,
2566 UMUTEX_CONTESTED, &owner, id | UMUTEX_CONTESTED);
2567 if (rv == -1) {
2568 error = EFAULT;
2569 break;
2570 }
2571
2572 if (owner == UMUTEX_CONTESTED) {
2573 suword32(&m->m_ceilings[0], ceiling);
2574 suword32(&m->m_owner, UMUTEX_CONTESTED);
2575 error = 0;
2576 break;
2577 }
2578
2579 if ((owner & ~UMUTEX_CONTESTED) == id) {
2580 suword32(&m->m_ceilings[0], ceiling);
2581 error = 0;
2582 break;
2583 }
2584
2585 /*
2586 * If we caught a signal, we have retried and now
2587 * exit immediately.
2588 */
2589 if (error != 0)
2590 break;
2591
2592 /*
2593 * We set the contested bit, sleep. Otherwise the lock changed
2594 * and we need to retry or we lost a race to the thread
2595 * unlocking the umtx.
2596 */
2597 umtxq_lock(&uq->uq_key);
2598 umtxq_insert(uq);
2599 umtxq_unbusy(&uq->uq_key);
2600 error = umtxq_sleep(uq, "umtxpp", NULL);
2601 umtxq_remove(uq);
2602 umtxq_unlock(&uq->uq_key);
2603 }
2604 umtxq_lock(&uq->uq_key);
2605 if (error == 0)
2606 umtxq_signal(&uq->uq_key, INT_MAX);
2607 umtxq_unbusy(&uq->uq_key);
2608 umtxq_unlock(&uq->uq_key);
2609 umtx_key_release(&uq->uq_key);
2610 if (error == 0 && old_ceiling != NULL)
2611 suword32(old_ceiling, save_ceiling);
2612 return (error);
2613 }
2614
2615 /*
2616 * Lock a userland POSIX mutex.
2617 */
2618 static int
2619 do_lock_umutex(struct thread *td, struct umutex *m,
2620 struct _umtx_time *timeout, int mode)
2621 {
2622 uint32_t flags;
2623 int error;
2624
2625 error = fueword32(&m->m_flags, &flags);
2626 if (error == -1)
2627 return (EFAULT);
2628
2629 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
2630 case 0:
2631 error = do_lock_normal(td, m, flags, timeout, mode);
2632 break;
2633 case UMUTEX_PRIO_INHERIT:
2634 error = do_lock_pi(td, m, flags, timeout, mode);
2635 break;
2636 case UMUTEX_PRIO_PROTECT:
2637 error = do_lock_pp(td, m, flags, timeout, mode);
2638 break;
2639 default:
2640 return (EINVAL);
2641 }
2642 if (timeout == NULL) {
2643 if (error == EINTR && mode != _UMUTEX_WAIT)
2644 error = ERESTART;
2645 } else {
2646 /* Timed-locking is not restarted. */
2647 if (error == ERESTART)
2648 error = EINTR;
2649 }
2650 return (error);
2651 }
2652
2653 /*
2654 * Unlock a userland POSIX mutex.
2655 */
2656 static int
2657 do_unlock_umutex(struct thread *td, struct umutex *m)
2658 {
2659 uint32_t flags;
2660 int error;
2661
2662 error = fueword32(&m->m_flags, &flags);
2663 if (error == -1)
2664 return (EFAULT);
2665
2666 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) {
2667 case 0:
2668 return (do_unlock_normal(td, m, flags));
2669 case UMUTEX_PRIO_INHERIT:
2670 return (do_unlock_pi(td, m, flags));
2671 case UMUTEX_PRIO_PROTECT:
2672 return (do_unlock_pp(td, m, flags));
2673 }
2674
2675 return (EINVAL);
2676 }
2677
2678 static int
2679 do_cv_wait(struct thread *td, struct ucond *cv, struct umutex *m,
2680 struct timespec *timeout, u_long wflags)
2681 {
2682 struct abs_timeout timo;
2683 struct umtx_q *uq;
2684 uint32_t flags, clockid, hasw;
2685 int error;
2686
2687 uq = td->td_umtxq;
2688 error = fueword32(&cv->c_flags, &flags);
2689 if (error == -1)
2690 return (EFAULT);
2691 error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &uq->uq_key);
2692 if (error != 0)
2693 return (error);
2694
2695 if ((wflags & CVWAIT_CLOCKID) != 0) {
2696 error = fueword32(&cv->c_clockid, &clockid);
2697 if (error == -1) {
2698 umtx_key_release(&uq->uq_key);
2699 return (EFAULT);
2700 }
2701 if (clockid < CLOCK_REALTIME ||
2702 clockid >= CLOCK_THREAD_CPUTIME_ID) {
2703 /* hmm, only HW clock id will work. */
2704 umtx_key_release(&uq->uq_key);
2705 return (EINVAL);
2706 }
2707 } else {
2708 clockid = CLOCK_REALTIME;
2709 }
2710
2711 umtxq_lock(&uq->uq_key);
2712 umtxq_busy(&uq->uq_key);
2713 umtxq_insert(uq);
2714 umtxq_unlock(&uq->uq_key);
2715
2716 /*
2717 * Set c_has_waiters to 1 before releasing user mutex, also
2718 * don't modify cache line when unnecessary.
2719 */
2720 error = fueword32(&cv->c_has_waiters, &hasw);
2721 if (error == 0 && hasw == 0)
2722 suword32(&cv->c_has_waiters, 1);
2723
2724 umtxq_unbusy_unlocked(&uq->uq_key);
2725
2726 error = do_unlock_umutex(td, m);
2727
2728 if (timeout != NULL)
2729 abs_timeout_init(&timo, clockid, ((wflags & CVWAIT_ABSTIME) != 0),
2730 timeout);
2731
2732 umtxq_lock(&uq->uq_key);
2733 if (error == 0) {
2734 error = umtxq_sleep(uq, "ucond", timeout == NULL ?
2735 NULL : &timo);
2736 }
2737
2738 if ((uq->uq_flags & UQF_UMTXQ) == 0)
2739 error = 0;
2740 else {
2741 /*
2742 * This must be timeout,interrupted by signal or
2743 * surprious wakeup, clear c_has_waiter flag when
2744 * necessary.
2745 */
2746 umtxq_busy(&uq->uq_key);
2747 if ((uq->uq_flags & UQF_UMTXQ) != 0) {
2748 int oldlen = uq->uq_cur_queue->length;
2749 umtxq_remove(uq);
2750 if (oldlen == 1) {
2751 umtxq_unlock(&uq->uq_key);
2752 suword32(&cv->c_has_waiters, 0);
2753 umtxq_lock(&uq->uq_key);
2754 }
2755 }
2756 umtxq_unbusy(&uq->uq_key);
2757 if (error == ERESTART)
2758 error = EINTR;
2759 }
2760
2761 umtxq_unlock(&uq->uq_key);
2762 umtx_key_release(&uq->uq_key);
2763 return (error);
2764 }
2765
2766 /*
2767 * Signal a userland condition variable.
2768 */
2769 static int
2770 do_cv_signal(struct thread *td, struct ucond *cv)
2771 {
2772 struct umtx_key key;
2773 int error, cnt, nwake;
2774 uint32_t flags;
2775
2776 error = fueword32(&cv->c_flags, &flags);
2777 if (error == -1)
2778 return (EFAULT);
2779 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
2780 return (error);
2781 umtxq_lock(&key);
2782 umtxq_busy(&key);
2783 cnt = umtxq_count(&key);
2784 nwake = umtxq_signal(&key, 1);
2785 if (cnt <= nwake) {
2786 umtxq_unlock(&key);
2787 error = suword32(&cv->c_has_waiters, 0);
2788 if (error == -1)
2789 error = EFAULT;
2790 umtxq_lock(&key);
2791 }
2792 umtxq_unbusy(&key);
2793 umtxq_unlock(&key);
2794 umtx_key_release(&key);
2795 return (error);
2796 }
2797
2798 static int
2799 do_cv_broadcast(struct thread *td, struct ucond *cv)
2800 {
2801 struct umtx_key key;
2802 int error;
2803 uint32_t flags;
2804
2805 error = fueword32(&cv->c_flags, &flags);
2806 if (error == -1)
2807 return (EFAULT);
2808 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0)
2809 return (error);
2810
2811 umtxq_lock(&key);
2812 umtxq_busy(&key);
2813 umtxq_signal(&key, INT_MAX);
2814 umtxq_unlock(&key);
2815
2816 error = suword32(&cv->c_has_waiters, 0);
2817 if (error == -1)
2818 error = EFAULT;
2819
2820 umtxq_unbusy_unlocked(&key);
2821
2822 umtx_key_release(&key);
2823 return (error);
2824 }
2825
2826 static int
2827 do_rw_rdlock(struct thread *td, struct urwlock *rwlock, long fflag, struct _umtx_time *timeout)
2828 {
2829 struct abs_timeout timo;
2830 struct umtx_q *uq;
2831 uint32_t flags, wrflags;
2832 int32_t state, oldstate;
2833 int32_t blocked_readers;
2834 int error, rv;
2835
2836 uq = td->td_umtxq;
2837 error = fueword32(&rwlock->rw_flags, &flags);
2838 if (error == -1)
2839 return (EFAULT);
2840 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
2841 if (error != 0)
2842 return (error);
2843
2844 if (timeout != NULL)
2845 abs_timeout_init2(&timo, timeout);
2846
2847 wrflags = URWLOCK_WRITE_OWNER;
2848 if (!(fflag & URWLOCK_PREFER_READER) && !(flags & URWLOCK_PREFER_READER))
2849 wrflags |= URWLOCK_WRITE_WAITERS;
2850
2851 for (;;) {
2852 rv = fueword32(&rwlock->rw_state, &state);
2853 if (rv == -1) {
2854 umtx_key_release(&uq->uq_key);
2855 return (EFAULT);
2856 }
2857
2858 /* try to lock it */
2859 while (!(state & wrflags)) {
2860 if (__predict_false(URWLOCK_READER_COUNT(state) == URWLOCK_MAX_READERS)) {
2861 umtx_key_release(&uq->uq_key);
2862 return (EAGAIN);
2863 }
2864 rv = casueword32(&rwlock->rw_state, state,
2865 &oldstate, state + 1);
2866 if (rv == -1) {
2867 umtx_key_release(&uq->uq_key);
2868 return (EFAULT);
2869 }
2870 if (oldstate == state) {
2871 umtx_key_release(&uq->uq_key);
2872 return (0);
2873 }
2874 error = umtxq_check_susp(td);
2875 if (error != 0)
2876 break;
2877 state = oldstate;
2878 }
2879
2880 if (error)
2881 break;
2882
2883 /* grab monitor lock */
2884 umtxq_lock(&uq->uq_key);
2885 umtxq_busy(&uq->uq_key);
2886 umtxq_unlock(&uq->uq_key);
2887
2888 /*
2889 * re-read the state, in case it changed between the try-lock above
2890 * and the check below
2891 */
2892 rv = fueword32(&rwlock->rw_state, &state);
2893 if (rv == -1)
2894 error = EFAULT;
2895
2896 /* set read contention bit */
2897 while (error == 0 && (state & wrflags) &&
2898 !(state & URWLOCK_READ_WAITERS)) {
2899 rv = casueword32(&rwlock->rw_state, state,
2900 &oldstate, state | URWLOCK_READ_WAITERS);
2901 if (rv == -1) {
2902 error = EFAULT;
2903 break;
2904 }
2905 if (oldstate == state)
2906 goto sleep;
2907 state = oldstate;
2908 error = umtxq_check_susp(td);
2909 if (error != 0)
2910 break;
2911 }
2912 if (error != 0) {
2913 umtxq_unbusy_unlocked(&uq->uq_key);
2914 break;
2915 }
2916
2917 /* state is changed while setting flags, restart */
2918 if (!(state & wrflags)) {
2919 umtxq_unbusy_unlocked(&uq->uq_key);
2920 error = umtxq_check_susp(td);
2921 if (error != 0)
2922 break;
2923 continue;
2924 }
2925
2926 sleep:
2927 /* contention bit is set, before sleeping, increase read waiter count */
2928 rv = fueword32(&rwlock->rw_blocked_readers,
2929 &blocked_readers);
2930 if (rv == -1) {
2931 umtxq_unbusy_unlocked(&uq->uq_key);
2932 error = EFAULT;
2933 break;
2934 }
2935 suword32(&rwlock->rw_blocked_readers, blocked_readers+1);
2936
2937 while (state & wrflags) {
2938 umtxq_lock(&uq->uq_key);
2939 umtxq_insert(uq);
2940 umtxq_unbusy(&uq->uq_key);
2941
2942 error = umtxq_sleep(uq, "urdlck", timeout == NULL ?
2943 NULL : &timo);
2944
2945 umtxq_busy(&uq->uq_key);
2946 umtxq_remove(uq);
2947 umtxq_unlock(&uq->uq_key);
2948 if (error)
2949 break;
2950 rv = fueword32(&rwlock->rw_state, &state);
2951 if (rv == -1) {
2952 error = EFAULT;
2953 break;
2954 }
2955 }
2956
2957 /* decrease read waiter count, and may clear read contention bit */
2958 rv = fueword32(&rwlock->rw_blocked_readers,
2959 &blocked_readers);
2960 if (rv == -1) {
2961 umtxq_unbusy_unlocked(&uq->uq_key);
2962 error = EFAULT;
2963 break;
2964 }
2965 suword32(&rwlock->rw_blocked_readers, blocked_readers-1);
2966 if (blocked_readers == 1) {
2967 rv = fueword32(&rwlock->rw_state, &state);
2968 if (rv == -1)
2969 error = EFAULT;
2970 while (error == 0) {
2971 rv = casueword32(&rwlock->rw_state, state,
2972 &oldstate, state & ~URWLOCK_READ_WAITERS);
2973 if (rv == -1) {
2974 error = EFAULT;
2975 break;
2976 }
2977 if (oldstate == state)
2978 break;
2979 state = oldstate;
2980 error = umtxq_check_susp(td);
2981 }
2982 }
2983
2984 umtxq_unbusy_unlocked(&uq->uq_key);
2985 if (error != 0)
2986 break;
2987 }
2988 umtx_key_release(&uq->uq_key);
2989 if (error == ERESTART)
2990 error = EINTR;
2991 return (error);
2992 }
2993
2994 static int
2995 do_rw_wrlock(struct thread *td, struct urwlock *rwlock, struct _umtx_time *timeout)
2996 {
2997 struct abs_timeout timo;
2998 struct umtx_q *uq;
2999 uint32_t flags;
3000 int32_t state, oldstate;
3001 int32_t blocked_writers;
3002 int32_t blocked_readers;
3003 int error, rv;
3004
3005 uq = td->td_umtxq;
3006 error = fueword32(&rwlock->rw_flags, &flags);
3007 if (error == -1)
3008 return (EFAULT);
3009 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
3010 if (error != 0)
3011 return (error);
3012
3013 if (timeout != NULL)
3014 abs_timeout_init2(&timo, timeout);
3015
3016 blocked_readers = 0;
3017 for (;;) {
3018 rv = fueword32(&rwlock->rw_state, &state);
3019 if (rv == -1) {
3020 umtx_key_release(&uq->uq_key);
3021 return (EFAULT);
3022 }
3023 while (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) {
3024 rv = casueword32(&rwlock->rw_state, state,
3025 &oldstate, state | URWLOCK_WRITE_OWNER);
3026 if (rv == -1) {
3027 umtx_key_release(&uq->uq_key);
3028 return (EFAULT);
3029 }
3030 if (oldstate == state) {
3031 umtx_key_release(&uq->uq_key);
3032 return (0);
3033 }
3034 state = oldstate;
3035 error = umtxq_check_susp(td);
3036 if (error != 0)
3037 break;
3038 }
3039
3040 if (error) {
3041 if (!(state & (URWLOCK_WRITE_OWNER|URWLOCK_WRITE_WAITERS)) &&
3042 blocked_readers != 0) {
3043 umtxq_lock(&uq->uq_key);
3044 umtxq_busy(&uq->uq_key);
3045 umtxq_signal_queue(&uq->uq_key, INT_MAX, UMTX_SHARED_QUEUE);
3046 umtxq_unbusy(&uq->uq_key);
3047 umtxq_unlock(&uq->uq_key);
3048 }
3049
3050 break;
3051 }
3052
3053 /* grab monitor lock */
3054 umtxq_lock(&uq->uq_key);
3055 umtxq_busy(&uq->uq_key);
3056 umtxq_unlock(&uq->uq_key);
3057
3058 /*
3059 * re-read the state, in case it changed between the try-lock above
3060 * and the check below
3061 */
3062 rv = fueword32(&rwlock->rw_state, &state);
3063 if (rv == -1)
3064 error = EFAULT;
3065
3066 while (error == 0 && ((state & URWLOCK_WRITE_OWNER) ||
3067 URWLOCK_READER_COUNT(state) != 0) &&
3068 (state & URWLOCK_WRITE_WAITERS) == 0) {
3069 rv = casueword32(&rwlock->rw_state, state,
3070 &oldstate, state | URWLOCK_WRITE_WAITERS);
3071 if (rv == -1) {
3072 error = EFAULT;
3073 break;
3074 }
3075 if (oldstate == state)
3076 goto sleep;
3077 state = oldstate;
3078 error = umtxq_check_susp(td);
3079 if (error != 0)
3080 break;
3081 }
3082 if (error != 0) {
3083 umtxq_unbusy_unlocked(&uq->uq_key);
3084 break;
3085 }
3086
3087 if (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) {
3088 umtxq_unbusy_unlocked(&uq->uq_key);
3089 error = umtxq_check_susp(td);
3090 if (error != 0)
3091 break;
3092 continue;
3093 }
3094 sleep:
3095 rv = fueword32(&rwlock->rw_blocked_writers,
3096 &blocked_writers);
3097 if (rv == -1) {
3098 umtxq_unbusy_unlocked(&uq->uq_key);
3099 error = EFAULT;
3100 break;
3101 }
3102 suword32(&rwlock->rw_blocked_writers, blocked_writers+1);
3103
3104 while ((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) {
3105 umtxq_lock(&uq->uq_key);
3106 umtxq_insert_queue(uq, UMTX_EXCLUSIVE_QUEUE);
3107 umtxq_unbusy(&uq->uq_key);
3108
3109 error = umtxq_sleep(uq, "uwrlck", timeout == NULL ?
3110 NULL : &timo);
3111
3112 umtxq_busy(&uq->uq_key);
3113 umtxq_remove_queue(uq, UMTX_EXCLUSIVE_QUEUE);
3114 umtxq_unlock(&uq->uq_key);
3115 if (error)
3116 break;
3117 rv = fueword32(&rwlock->rw_state, &state);
3118 if (rv == -1) {
3119 error = EFAULT;
3120 break;
3121 }
3122 }
3123
3124 rv = fueword32(&rwlock->rw_blocked_writers,
3125 &blocked_writers);
3126 if (rv == -1) {
3127 umtxq_unbusy_unlocked(&uq->uq_key);
3128 error = EFAULT;
3129 break;
3130 }
3131 suword32(&rwlock->rw_blocked_writers, blocked_writers-1);
3132 if (blocked_writers == 1) {
3133 rv = fueword32(&rwlock->rw_state, &state);
3134 if (rv == -1) {
3135 umtxq_unbusy_unlocked(&uq->uq_key);
3136 error = EFAULT;
3137 break;
3138 }
3139 for (;;) {
3140 rv = casueword32(&rwlock->rw_state, state,
3141 &oldstate, state & ~URWLOCK_WRITE_WAITERS);
3142 if (rv == -1) {
3143 error = EFAULT;
3144 break;
3145 }
3146 if (oldstate == state)
3147 break;
3148 state = oldstate;
3149 error = umtxq_check_susp(td);
3150 /*
3151 * We are leaving the URWLOCK_WRITE_WAITERS
3152 * behind, but this should not harm the
3153 * correctness.
3154 */
3155 if (error != 0)
3156 break;
3157 }
3158 rv = fueword32(&rwlock->rw_blocked_readers,
3159 &blocked_readers);
3160 if (rv == -1) {
3161 umtxq_unbusy_unlocked(&uq->uq_key);
3162 error = EFAULT;
3163 break;
3164 }
3165 } else
3166 blocked_readers = 0;
3167
3168 umtxq_unbusy_unlocked(&uq->uq_key);
3169 }
3170
3171 umtx_key_release(&uq->uq_key);
3172 if (error == ERESTART)
3173 error = EINTR;
3174 return (error);
3175 }
3176
3177 static int
3178 do_rw_unlock(struct thread *td, struct urwlock *rwlock)
3179 {
3180 struct umtx_q *uq;
3181 uint32_t flags;
3182 int32_t state, oldstate;
3183 int error, rv, q, count;
3184
3185 uq = td->td_umtxq;
3186 error = fueword32(&rwlock->rw_flags, &flags);
3187 if (error == -1)
3188 return (EFAULT);
3189 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key);
3190 if (error != 0)
3191 return (error);
3192
3193 error = fueword32(&rwlock->rw_state, &state);
3194 if (error == -1) {
3195 error = EFAULT;
3196 goto out;
3197 }
3198 if (state & URWLOCK_WRITE_OWNER) {
3199 for (;;) {
3200 rv = casueword32(&rwlock->rw_state, state,
3201 &oldstate, state & ~URWLOCK_WRITE_OWNER);
3202 if (rv == -1) {
3203 error = EFAULT;
3204 goto out;
3205 }
3206 if (oldstate != state) {
3207 state = oldstate;
3208 if (!(oldstate & URWLOCK_WRITE_OWNER)) {
3209 error = EPERM;
3210 goto out;
3211 }
3212 error = umtxq_check_susp(td);
3213 if (error != 0)
3214 goto out;
3215 } else
3216 break;
3217 }
3218 } else if (URWLOCK_READER_COUNT(state) != 0) {
3219 for (;;) {
3220 rv = casueword32(&rwlock->rw_state, state,
3221 &oldstate, state - 1);
3222 if (rv == -1) {
3223 error = EFAULT;
3224 goto out;
3225 }
3226 if (oldstate != state) {
3227 state = oldstate;
3228 if (URWLOCK_READER_COUNT(oldstate) == 0) {
3229 error = EPERM;
3230 goto out;
3231 }
3232 error = umtxq_check_susp(td);
3233 if (error != 0)
3234 goto out;
3235 } else
3236 break;
3237 }
3238 } else {
3239 error = EPERM;
3240 goto out;
3241 }
3242
3243 count = 0;
3244
3245 if (!(flags & URWLOCK_PREFER_READER)) {
3246 if (state & URWLOCK_WRITE_WAITERS) {
3247 count = 1;
3248 q = UMTX_EXCLUSIVE_QUEUE;
3249 } else if (state & URWLOCK_READ_WAITERS) {
3250 count = INT_MAX;
3251 q = UMTX_SHARED_QUEUE;
3252 }
3253 } else {
3254 if (state & URWLOCK_READ_WAITERS) {
3255 count = INT_MAX;
3256 q = UMTX_SHARED_QUEUE;
3257 } else if (state & URWLOCK_WRITE_WAITERS) {
3258 count = 1;
3259 q = UMTX_EXCLUSIVE_QUEUE;
3260 }
3261 }
3262
3263 if (count) {
3264 umtxq_lock(&uq->uq_key);
3265 umtxq_busy(&uq->uq_key);
3266 umtxq_signal_queue(&uq->uq_key, count, q);
3267 umtxq_unbusy(&uq->uq_key);
3268 umtxq_unlock(&uq->uq_key);
3269 }
3270 out:
3271 umtx_key_release(&uq->uq_key);
3272 return (error);
3273 }
3274
3275 static int
3276 do_sem_wait(struct thread *td, struct _usem *sem, struct _umtx_time *timeout)
3277 {
3278 struct abs_timeout timo;
3279 struct umtx_q *uq;
3280 uint32_t flags, count, count1;
3281 int error, rv;
3282
3283 uq = td->td_umtxq;
3284 error = fueword32(&sem->_flags, &flags);
3285 if (error == -1)
3286 return (EFAULT);
3287 error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key);
3288 if (error != 0)
3289 return (error);
3290
3291 if (timeout != NULL)
3292 abs_timeout_init2(&timo, timeout);
3293
3294 umtxq_lock(&uq->uq_key);
3295 umtxq_busy(&uq->uq_key);
3296 umtxq_insert(uq);
3297 umtxq_unlock(&uq->uq_key);
3298 rv = casueword32(&sem->_has_waiters, 0, &count1, 1);
3299 if (rv == 0)
3300 rv = fueword32(&sem->_count, &count);
3301 if (rv == -1 || count != 0) {
3302 umtxq_lock(&uq->uq_key);
3303 umtxq_unbusy(&uq->uq_key);
3304 umtxq_remove(uq);
3305 umtxq_unlock(&uq->uq_key);
3306 umtx_key_release(&uq->uq_key);
3307 return (rv == -1 ? EFAULT : 0);
3308 }
3309 umtxq_lock(&uq->uq_key);
3310 umtxq_unbusy(&uq->uq_key);
3311
3312 error = umtxq_sleep(uq, "usem", timeout == NULL ? NULL : &timo);
3313
3314 if ((uq->uq_flags & UQF_UMTXQ) == 0)
3315 error = 0;
3316 else {
3317 umtxq_remove(uq);
3318 /* A relative timeout cannot be restarted. */
3319 if (error == ERESTART && timeout != NULL &&
3320 (timeout->_flags & UMTX_ABSTIME) == 0)
3321 error = EINTR;
3322 }
3323 umtxq_unlock(&uq->uq_key);
3324 umtx_key_release(&uq->uq_key);
3325 return (error);
3326 }
3327
3328 /*
3329 * Signal a userland condition variable.
3330 */
3331 static int
3332 do_sem_wake(struct thread *td, struct _usem *sem)
3333 {
3334 struct umtx_key key;
3335 int error, cnt;
3336 uint32_t flags;
3337
3338 error = fueword32(&sem->_flags, &flags);
3339 if (error == -1)
3340 return (EFAULT);
3341 if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0)
3342 return (error);
3343 umtxq_lock(&key);
3344 umtxq_busy(&key);
3345 cnt = umtxq_count(&key);
3346 if (cnt > 0) {
3347 umtxq_signal(&key, 1);
3348 /*
3349 * Check if count is greater than 0, this means the memory is
3350 * still being referenced by user code, so we can safely
3351 * update _has_waiters flag.
3352 */
3353 if (cnt == 1) {
3354 umtxq_unlock(&key);
3355 error = suword32(&sem->_has_waiters, 0);
3356 umtxq_lock(&key);
3357 if (error == -1)
3358 error = EFAULT;
3359 }
3360 }
3361 umtxq_unbusy(&key);
3362 umtxq_unlock(&key);
3363 umtx_key_release(&key);
3364 return (error);
3365 }
3366
3367 int
3368 sys__umtx_lock(struct thread *td, struct _umtx_lock_args *uap)
3369 /* struct umtx *umtx */
3370 {
3371 return do_lock_umtx(td, uap->umtx, td->td_tid, 0);
3372 }
3373
3374 int
3375 sys__umtx_unlock(struct thread *td, struct _umtx_unlock_args *uap)
3376 /* struct umtx *umtx */
3377 {
3378 return do_unlock_umtx(td, uap->umtx, td->td_tid);
3379 }
3380
3381 inline int
3382 umtx_copyin_timeout(const void *addr, struct timespec *tsp)
3383 {
3384 int error;
3385
3386 error = copyin(addr, tsp, sizeof(struct timespec));
3387 if (error == 0) {
3388 if (tsp->tv_sec < 0 ||
3389 tsp->tv_nsec >= 1000000000 ||
3390 tsp->tv_nsec < 0)
3391 error = EINVAL;
3392 }
3393 return (error);
3394 }
3395
3396 static inline int
3397 umtx_copyin_umtx_time(const void *addr, size_t size, struct _umtx_time *tp)
3398 {
3399 int error;
3400
3401 if (size <= sizeof(struct timespec)) {
3402 tp->_clockid = CLOCK_REALTIME;
3403 tp->_flags = 0;
3404 error = copyin(addr, &tp->_timeout, sizeof(struct timespec));
3405 } else
3406 error = copyin(addr, tp, sizeof(struct _umtx_time));
3407 if (error != 0)
3408 return (error);
3409 if (tp->_timeout.tv_sec < 0 ||
3410 tp->_timeout.tv_nsec >= 1000000000 || tp->_timeout.tv_nsec < 0)
3411 return (EINVAL);
3412 return (0);
3413 }
3414
3415 static int
3416 __umtx_op_lock_umtx(struct thread *td, struct _umtx_op_args *uap)
3417 {
3418 struct timespec *ts, timeout;
3419 int error;
3420
3421 /* Allow a null timespec (wait forever). */
3422 if (uap->uaddr2 == NULL)
3423 ts = NULL;
3424 else {
3425 error = umtx_copyin_timeout(uap->uaddr2, &timeout);
3426 if (error != 0)
3427 return (error);
3428 ts = &timeout;
3429 }
3430 return (do_lock_umtx(td, uap->obj, uap->val, ts));
3431 }
3432
3433 static int
3434 __umtx_op_unlock_umtx(struct thread *td, struct _umtx_op_args *uap)
3435 {
3436 return (do_unlock_umtx(td, uap->obj, uap->val));
3437 }
3438
3439 static int
3440 __umtx_op_wait(struct thread *td, struct _umtx_op_args *uap)
3441 {
3442 struct _umtx_time timeout, *tm_p;
3443 int error;
3444
3445 if (uap->uaddr2 == NULL)
3446 tm_p = NULL;
3447 else {
3448 error = umtx_copyin_umtx_time(
3449 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3450 if (error != 0)
3451 return (error);
3452 tm_p = &timeout;
3453 }
3454 return do_wait(td, uap->obj, uap->val, tm_p, 0, 0);
3455 }
3456
3457 static int
3458 __umtx_op_wait_uint(struct thread *td, struct _umtx_op_args *uap)
3459 {
3460 struct _umtx_time timeout, *tm_p;
3461 int error;
3462
3463 if (uap->uaddr2 == NULL)
3464 tm_p = NULL;
3465 else {
3466 error = umtx_copyin_umtx_time(
3467 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3468 if (error != 0)
3469 return (error);
3470 tm_p = &timeout;
3471 }
3472 return do_wait(td, uap->obj, uap->val, tm_p, 1, 0);
3473 }
3474
3475 static int
3476 __umtx_op_wait_uint_private(struct thread *td, struct _umtx_op_args *uap)
3477 {
3478 struct _umtx_time *tm_p, timeout;
3479 int error;
3480
3481 if (uap->uaddr2 == NULL)
3482 tm_p = NULL;
3483 else {
3484 error = umtx_copyin_umtx_time(
3485 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3486 if (error != 0)
3487 return (error);
3488 tm_p = &timeout;
3489 }
3490 return do_wait(td, uap->obj, uap->val, tm_p, 1, 1);
3491 }
3492
3493 static int
3494 __umtx_op_wake(struct thread *td, struct _umtx_op_args *uap)
3495 {
3496 return (kern_umtx_wake(td, uap->obj, uap->val, 0));
3497 }
3498
3499 #define BATCH_SIZE 128
3500 static int
3501 __umtx_op_nwake_private(struct thread *td, struct _umtx_op_args *uap)
3502 {
3503 int count = uap->val;
3504 void *uaddrs[BATCH_SIZE];
3505 char **upp = (char **)uap->obj;
3506 int tocopy;
3507 int error = 0;
3508 int i, pos = 0;
3509
3510 while (count > 0) {
3511 tocopy = count;
3512 if (tocopy > BATCH_SIZE)
3513 tocopy = BATCH_SIZE;
3514 error = copyin(upp+pos, uaddrs, tocopy * sizeof(char *));
3515 if (error != 0)
3516 break;
3517 for (i = 0; i < tocopy; ++i)
3518 kern_umtx_wake(td, uaddrs[i], INT_MAX, 1);
3519 count -= tocopy;
3520 pos += tocopy;
3521 }
3522 return (error);
3523 }
3524
3525 static int
3526 __umtx_op_wake_private(struct thread *td, struct _umtx_op_args *uap)
3527 {
3528 return (kern_umtx_wake(td, uap->obj, uap->val, 1));
3529 }
3530
3531 static int
3532 __umtx_op_lock_umutex(struct thread *td, struct _umtx_op_args *uap)
3533 {
3534 struct _umtx_time *tm_p, timeout;
3535 int error;
3536
3537 /* Allow a null timespec (wait forever). */
3538 if (uap->uaddr2 == NULL)
3539 tm_p = NULL;
3540 else {
3541 error = umtx_copyin_umtx_time(
3542 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3543 if (error != 0)
3544 return (error);
3545 tm_p = &timeout;
3546 }
3547 return do_lock_umutex(td, uap->obj, tm_p, 0);
3548 }
3549
3550 static int
3551 __umtx_op_trylock_umutex(struct thread *td, struct _umtx_op_args *uap)
3552 {
3553 return do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY);
3554 }
3555
3556 static int
3557 __umtx_op_wait_umutex(struct thread *td, struct _umtx_op_args *uap)
3558 {
3559 struct _umtx_time *tm_p, timeout;
3560 int error;
3561
3562 /* Allow a null timespec (wait forever). */
3563 if (uap->uaddr2 == NULL)
3564 tm_p = NULL;
3565 else {
3566 error = umtx_copyin_umtx_time(
3567 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3568 if (error != 0)
3569 return (error);
3570 tm_p = &timeout;
3571 }
3572 return do_lock_umutex(td, uap->obj, tm_p, _UMUTEX_WAIT);
3573 }
3574
3575 static int
3576 __umtx_op_wake_umutex(struct thread *td, struct _umtx_op_args *uap)
3577 {
3578 return do_wake_umutex(td, uap->obj);
3579 }
3580
3581 static int
3582 __umtx_op_unlock_umutex(struct thread *td, struct _umtx_op_args *uap)
3583 {
3584 return do_unlock_umutex(td, uap->obj);
3585 }
3586
3587 static int
3588 __umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap)
3589 {
3590 return do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1);
3591 }
3592
3593 static int
3594 __umtx_op_cv_wait(struct thread *td, struct _umtx_op_args *uap)
3595 {
3596 struct timespec *ts, timeout;
3597 int error;
3598
3599 /* Allow a null timespec (wait forever). */
3600 if (uap->uaddr2 == NULL)
3601 ts = NULL;
3602 else {
3603 error = umtx_copyin_timeout(uap->uaddr2, &timeout);
3604 if (error != 0)
3605 return (error);
3606 ts = &timeout;
3607 }
3608 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val));
3609 }
3610
3611 static int
3612 __umtx_op_cv_signal(struct thread *td, struct _umtx_op_args *uap)
3613 {
3614 return do_cv_signal(td, uap->obj);
3615 }
3616
3617 static int
3618 __umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap)
3619 {
3620 return do_cv_broadcast(td, uap->obj);
3621 }
3622
3623 static int
3624 __umtx_op_rw_rdlock(struct thread *td, struct _umtx_op_args *uap)
3625 {
3626 struct _umtx_time timeout;
3627 int error;
3628
3629 /* Allow a null timespec (wait forever). */
3630 if (uap->uaddr2 == NULL) {
3631 error = do_rw_rdlock(td, uap->obj, uap->val, 0);
3632 } else {
3633 error = umtx_copyin_umtx_time(uap->uaddr2,
3634 (size_t)uap->uaddr1, &timeout);
3635 if (error != 0)
3636 return (error);
3637 error = do_rw_rdlock(td, uap->obj, uap->val, &timeout);
3638 }
3639 return (error);
3640 }
3641
3642 static int
3643 __umtx_op_rw_wrlock(struct thread *td, struct _umtx_op_args *uap)
3644 {
3645 struct _umtx_time timeout;
3646 int error;
3647
3648 /* Allow a null timespec (wait forever). */
3649 if (uap->uaddr2 == NULL) {
3650 error = do_rw_wrlock(td, uap->obj, 0);
3651 } else {
3652 error = umtx_copyin_umtx_time(uap->uaddr2,
3653 (size_t)uap->uaddr1, &timeout);
3654 if (error != 0)
3655 return (error);
3656
3657 error = do_rw_wrlock(td, uap->obj, &timeout);
3658 }
3659 return (error);
3660 }
3661
3662 static int
3663 __umtx_op_rw_unlock(struct thread *td, struct _umtx_op_args *uap)
3664 {
3665 return do_rw_unlock(td, uap->obj);
3666 }
3667
3668 static int
3669 __umtx_op_sem_wait(struct thread *td, struct _umtx_op_args *uap)
3670 {
3671 struct _umtx_time *tm_p, timeout;
3672 int error;
3673
3674 /* Allow a null timespec (wait forever). */
3675 if (uap->uaddr2 == NULL)
3676 tm_p = NULL;
3677 else {
3678 error = umtx_copyin_umtx_time(
3679 uap->uaddr2, (size_t)uap->uaddr1, &timeout);
3680 if (error != 0)
3681 return (error);
3682 tm_p = &timeout;
3683 }
3684 return (do_sem_wait(td, uap->obj, tm_p));
3685 }
3686
3687 static int
3688 __umtx_op_sem_wake(struct thread *td, struct _umtx_op_args *uap)
3689 {
3690 return do_sem_wake(td, uap->obj);
3691 }
3692
3693 static int
3694 __umtx_op_wake2_umutex(struct thread *td, struct _umtx_op_args *uap)
3695 {
3696 return do_wake2_umutex(td, uap->obj, uap->val);
3697 }
3698
3699 typedef int (*_umtx_op_func)(struct thread *td, struct _umtx_op_args *uap);
3700
3701 static _umtx_op_func op_table[] = {
3702 __umtx_op_lock_umtx, /* UMTX_OP_LOCK */
3703 __umtx_op_unlock_umtx, /* UMTX_OP_UNLOCK */
3704 __umtx_op_wait, /* UMTX_OP_WAIT */
3705 __umtx_op_wake, /* UMTX_OP_WAKE */
3706 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_TRYLOCK */
3707 __umtx_op_lock_umutex, /* UMTX_OP_MUTEX_LOCK */
3708 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */
3709 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */
3710 __umtx_op_cv_wait, /* UMTX_OP_CV_WAIT*/
3711 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */
3712 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */
3713 __umtx_op_wait_uint, /* UMTX_OP_WAIT_UINT */
3714 __umtx_op_rw_rdlock, /* UMTX_OP_RW_RDLOCK */
3715 __umtx_op_rw_wrlock, /* UMTX_OP_RW_WRLOCK */
3716 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */
3717 __umtx_op_wait_uint_private, /* UMTX_OP_WAIT_UINT_PRIVATE */
3718 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */
3719 __umtx_op_wait_umutex, /* UMTX_OP_UMUTEX_WAIT */
3720 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */
3721 __umtx_op_sem_wait, /* UMTX_OP_SEM_WAIT */
3722 __umtx_op_sem_wake, /* UMTX_OP_SEM_WAKE */
3723 __umtx_op_nwake_private, /* UMTX_OP_NWAKE_PRIVATE */
3724 __umtx_op_wake2_umutex /* UMTX_OP_UMUTEX_WAKE2 */
3725 };
3726
3727 int
3728 sys__umtx_op(struct thread *td, struct _umtx_op_args *uap)
3729 {
3730 if ((unsigned)uap->op < UMTX_OP_MAX)
3731 return (*op_table[uap->op])(td, uap);
3732 return (EINVAL);
3733 }
3734
3735 #ifdef COMPAT_FREEBSD32
3736 int
3737 freebsd32_umtx_lock(struct thread *td, struct freebsd32_umtx_lock_args *uap)
3738 /* struct umtx *umtx */
3739 {
3740 return (do_lock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid, NULL));
3741 }
3742
3743 int
3744 freebsd32_umtx_unlock(struct thread *td, struct freebsd32_umtx_unlock_args *uap)
3745 /* struct umtx *umtx */
3746 {
3747 return (do_unlock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid));
3748 }
3749
3750 struct timespec32 {
3751 int32_t tv_sec;
3752 int32_t tv_nsec;
3753 };
3754
3755 struct umtx_time32 {
3756 struct timespec32 timeout;
3757 uint32_t flags;
3758 uint32_t clockid;
3759 };
3760
3761 static inline int
3762 umtx_copyin_timeout32(void *addr, struct timespec *tsp)
3763 {
3764 struct timespec32 ts32;
3765 int error;
3766
3767 error = copyin(addr, &ts32, sizeof(struct timespec32));
3768 if (error == 0) {
3769 if (ts32.tv_sec < 0 ||
3770 ts32.tv_nsec >= 1000000000 ||
3771 ts32.tv_nsec < 0)
3772 error = EINVAL;
3773 else {
3774 tsp->tv_sec = ts32.tv_sec;
3775 tsp->tv_nsec = ts32.tv_nsec;
3776 }
3777 }
3778 return (error);
3779 }
3780
3781 static inline int
3782 umtx_copyin_umtx_time32(const void *addr, size_t size, struct _umtx_time *tp)
3783 {
3784 struct umtx_time32 t32;
3785 int error;
3786
3787 t32.clockid = CLOCK_REALTIME;
3788 t32.flags = 0;
3789 if (size <= sizeof(struct timespec32))
3790 error = copyin(addr, &t32.timeout, sizeof(struct timespec32));
3791 else
3792 error = copyin(addr, &t32, sizeof(struct umtx_time32));
3793 if (error != 0)
3794 return (error);
3795 if (t32.timeout.tv_sec < 0 ||
3796 t32.timeout.tv_nsec >= 1000000000 || t32.timeout.tv_nsec < 0)
3797 return (EINVAL);
3798 tp->_timeout.tv_sec = t32.timeout.tv_sec;
3799 tp->_timeout.tv_nsec = t32.timeout.tv_nsec;
3800 tp->_flags = t32.flags;
3801 tp->_clockid = t32.clockid;
3802 return (0);
3803 }
3804
3805 static int
3806 __umtx_op_lock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap)
3807 {
3808 struct timespec *ts, timeout;
3809 int error;
3810
3811 /* Allow a null timespec (wait forever). */
3812 if (uap->uaddr2 == NULL)
3813 ts = NULL;
3814 else {
3815 error = umtx_copyin_timeout32(uap->uaddr2, &timeout);
3816 if (error != 0)
3817 return (error);
3818 ts = &timeout;
3819 }
3820 return (do_lock_umtx32(td, uap->obj, uap->val, ts));
3821 }
3822
3823 static int
3824 __umtx_op_unlock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap)
3825 {
3826 return (do_unlock_umtx32(td, uap->obj, (uint32_t)uap->val));
3827 }
3828
3829 static int
3830 __umtx_op_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
3831 {
3832 struct _umtx_time *tm_p, timeout;
3833 int error;
3834
3835 if (uap->uaddr2 == NULL)
3836 tm_p = NULL;
3837 else {
3838 error = umtx_copyin_umtx_time32(uap->uaddr2,
3839 (size_t)uap->uaddr1, &timeout);
3840 if (error != 0)
3841 return (error);
3842 tm_p = &timeout;
3843 }
3844 return do_wait(td, uap->obj, uap->val, tm_p, 1, 0);
3845 }
3846
3847 static int
3848 __umtx_op_lock_umutex_compat32(struct thread *td, struct _umtx_op_args *uap)
3849 {
3850 struct _umtx_time *tm_p, timeout;
3851 int error;
3852
3853 /* Allow a null timespec (wait forever). */
3854 if (uap->uaddr2 == NULL)
3855 tm_p = NULL;
3856 else {
3857 error = umtx_copyin_umtx_time(uap->uaddr2,
3858 (size_t)uap->uaddr1, &timeout);
3859 if (error != 0)
3860 return (error);
3861 tm_p = &timeout;
3862 }
3863 return do_lock_umutex(td, uap->obj, tm_p, 0);
3864 }
3865
3866 static int
3867 __umtx_op_wait_umutex_compat32(struct thread *td, struct _umtx_op_args *uap)
3868 {
3869 struct _umtx_time *tm_p, timeout;
3870 int error;
3871
3872 /* Allow a null timespec (wait forever). */
3873 if (uap->uaddr2 == NULL)
3874 tm_p = NULL;
3875 else {
3876 error = umtx_copyin_umtx_time32(uap->uaddr2,
3877 (size_t)uap->uaddr1, &timeout);
3878 if (error != 0)
3879 return (error);
3880 tm_p = &timeout;
3881 }
3882 return do_lock_umutex(td, uap->obj, tm_p, _UMUTEX_WAIT);
3883 }
3884
3885 static int
3886 __umtx_op_cv_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
3887 {
3888 struct timespec *ts, timeout;
3889 int error;
3890
3891 /* Allow a null timespec (wait forever). */
3892 if (uap->uaddr2 == NULL)
3893 ts = NULL;
3894 else {
3895 error = umtx_copyin_timeout32(uap->uaddr2, &timeout);
3896 if (error != 0)
3897 return (error);
3898 ts = &timeout;
3899 }
3900 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val));
3901 }
3902
3903 static int
3904 __umtx_op_rw_rdlock_compat32(struct thread *td, struct _umtx_op_args *uap)
3905 {
3906 struct _umtx_time timeout;
3907 int error;
3908
3909 /* Allow a null timespec (wait forever). */
3910 if (uap->uaddr2 == NULL) {
3911 error = do_rw_rdlock(td, uap->obj, uap->val, 0);
3912 } else {
3913 error = umtx_copyin_umtx_time32(uap->uaddr2,
3914 (size_t)uap->uaddr1, &timeout);
3915 if (error != 0)
3916 return (error);
3917 error = do_rw_rdlock(td, uap->obj, uap->val, &timeout);
3918 }
3919 return (error);
3920 }
3921
3922 static int
3923 __umtx_op_rw_wrlock_compat32(struct thread *td, struct _umtx_op_args *uap)
3924 {
3925 struct _umtx_time timeout;
3926 int error;
3927
3928 /* Allow a null timespec (wait forever). */
3929 if (uap->uaddr2 == NULL) {
3930 error = do_rw_wrlock(td, uap->obj, 0);
3931 } else {
3932 error = umtx_copyin_umtx_time32(uap->uaddr2,
3933 (size_t)uap->uaddr1, &timeout);
3934 if (error != 0)
3935 return (error);
3936 error = do_rw_wrlock(td, uap->obj, &timeout);
3937 }
3938 return (error);
3939 }
3940
3941 static int
3942 __umtx_op_wait_uint_private_compat32(struct thread *td, struct _umtx_op_args *uap)
3943 {
3944 struct _umtx_time *tm_p, timeout;
3945 int error;
3946
3947 if (uap->uaddr2 == NULL)
3948 tm_p = NULL;
3949 else {
3950 error = umtx_copyin_umtx_time32(
3951 uap->uaddr2, (size_t)uap->uaddr1,&timeout);
3952 if (error != 0)
3953 return (error);
3954 tm_p = &timeout;
3955 }
3956 return do_wait(td, uap->obj, uap->val, tm_p, 1, 1);
3957 }
3958
3959 static int
3960 __umtx_op_sem_wait_compat32(struct thread *td, struct _umtx_op_args *uap)
3961 {
3962 struct _umtx_time *tm_p, timeout;
3963 int error;
3964
3965 /* Allow a null timespec (wait forever). */
3966 if (uap->uaddr2 == NULL)
3967 tm_p = NULL;
3968 else {
3969 error = umtx_copyin_umtx_time32(uap->uaddr2,
3970 (size_t)uap->uaddr1, &timeout);
3971 if (error != 0)
3972 return (error);
3973 tm_p = &timeout;
3974 }
3975 return (do_sem_wait(td, uap->obj, tm_p));
3976 }
3977
3978 static int
3979 __umtx_op_nwake_private32(struct thread *td, struct _umtx_op_args *uap)
3980 {
3981 int count = uap->val;
3982 uint32_t uaddrs[BATCH_SIZE];
3983 uint32_t **upp = (uint32_t **)uap->obj;
3984 int tocopy;
3985 int error = 0;
3986 int i, pos = 0;
3987
3988 while (count > 0) {
3989 tocopy = count;
3990 if (tocopy > BATCH_SIZE)
3991 tocopy = BATCH_SIZE;
3992 error = copyin(upp+pos, uaddrs, tocopy * sizeof(uint32_t));
3993 if (error != 0)
3994 break;
3995 for (i = 0; i < tocopy; ++i)
3996 kern_umtx_wake(td, (void *)(intptr_t)uaddrs[i],
3997 INT_MAX, 1);
3998 count -= tocopy;
3999 pos += tocopy;
4000 }
4001 return (error);
4002 }
4003
4004 static _umtx_op_func op_table_compat32[] = {
4005 __umtx_op_lock_umtx_compat32, /* UMTX_OP_LOCK */
4006 __umtx_op_unlock_umtx_compat32, /* UMTX_OP_UNLOCK */
4007 __umtx_op_wait_compat32, /* UMTX_OP_WAIT */
4008 __umtx_op_wake, /* UMTX_OP_WAKE */
4009 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_LOCK */
4010 __umtx_op_lock_umutex_compat32, /* UMTX_OP_MUTEX_TRYLOCK */
4011 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */
4012 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */
4013 __umtx_op_cv_wait_compat32, /* UMTX_OP_CV_WAIT*/
4014 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */
4015 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */
4016 __umtx_op_wait_compat32, /* UMTX_OP_WAIT_UINT */
4017 __umtx_op_rw_rdlock_compat32, /* UMTX_OP_RW_RDLOCK */
4018 __umtx_op_rw_wrlock_compat32, /* UMTX_OP_RW_WRLOCK */
4019 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */
4020 __umtx_op_wait_uint_private_compat32, /* UMTX_OP_WAIT_UINT_PRIVATE */
4021 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */
4022 __umtx_op_wait_umutex_compat32, /* UMTX_OP_UMUTEX_WAIT */
4023 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */
4024 __umtx_op_sem_wait_compat32, /* UMTX_OP_SEM_WAIT */
4025 __umtx_op_sem_wake, /* UMTX_OP_SEM_WAKE */
4026 __umtx_op_nwake_private32, /* UMTX_OP_NWAKE_PRIVATE */
4027 __umtx_op_wake2_umutex /* UMTX_OP_UMUTEX_WAKE2 */
4028 };
4029
4030 int
4031 freebsd32_umtx_op(struct thread *td, struct freebsd32_umtx_op_args *uap)
4032 {
4033 if ((unsigned)uap->op < UMTX_OP_MAX)
4034 return (*op_table_compat32[uap->op])(td,
4035 (struct _umtx_op_args *)uap);
4036 return (EINVAL);
4037 }
4038 #endif
4039
4040 void
4041 umtx_thread_init(struct thread *td)
4042 {
4043 td->td_umtxq = umtxq_alloc();
4044 td->td_umtxq->uq_thread = td;
4045 }
4046
4047 void
4048 umtx_thread_fini(struct thread *td)
4049 {
4050 umtxq_free(td->td_umtxq);
4051 }
4052
4053 /*
4054 * It will be called when new thread is created, e.g fork().
4055 */
4056 void
4057 umtx_thread_alloc(struct thread *td)
4058 {
4059 struct umtx_q *uq;
4060
4061 uq = td->td_umtxq;
4062 uq->uq_inherited_pri = PRI_MAX;
4063
4064 KASSERT(uq->uq_flags == 0, ("uq_flags != 0"));
4065 KASSERT(uq->uq_thread == td, ("uq_thread != td"));
4066 KASSERT(uq->uq_pi_blocked == NULL, ("uq_pi_blocked != NULL"));
4067 KASSERT(TAILQ_EMPTY(&uq->uq_pi_contested), ("uq_pi_contested is not empty"));
4068 }
4069
4070 /*
4071 * exec() hook.
4072 */
4073 static void
4074 umtx_exec_hook(void *arg __unused, struct proc *p __unused,
4075 struct image_params *imgp __unused)
4076 {
4077 umtx_thread_cleanup(curthread);
4078 }
4079
4080 /*
4081 * thread_exit() hook.
4082 */
4083 void
4084 umtx_thread_exit(struct thread *td)
4085 {
4086 umtx_thread_cleanup(td);
4087 }
4088
4089 /*
4090 * clean up umtx data.
4091 */
4092 static void
4093 umtx_thread_cleanup(struct thread *td)
4094 {
4095 struct umtx_q *uq;
4096 struct umtx_pi *pi;
4097
4098 if ((uq = td->td_umtxq) == NULL)
4099 return;
4100
4101 mtx_lock(&umtx_lock);
4102 uq->uq_inherited_pri = PRI_MAX;
4103 while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) {
4104 pi->pi_owner = NULL;
4105 TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link);
4106 }
4107 mtx_unlock(&umtx_lock);
4108 thread_lock(td);
4109 sched_lend_user_prio(td, PRI_MAX);
4110 thread_unlock(td);
4111 }
Cache object: 2e2a80e7c5728c01111a1fd56e9c9e29
|