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("$FreeBSD$"); 30 31 #include "opt_compat.h" 32 #include <sys/param.h> 33 #include <sys/kernel.h> 34 #include <sys/limits.h> 35 #include <sys/lock.h> 36 #include <sys/malloc.h> 37 #include <sys/mutex.h> 38 #include <sys/priv.h> 39 #include <sys/proc.h> 40 #include <sys/sched.h> 41 #include <sys/smp.h> 42 #include <sys/sysctl.h> 43 #include <sys/sysent.h> 44 #include <sys/systm.h> 45 #include <sys/sysproto.h> 46 #include <sys/eventhandler.h> 47 #include <sys/umtx.h> 48 49 #include <vm/vm.h> 50 #include <vm/vm_param.h> 51 #include <vm/pmap.h> 52 #include <vm/vm_map.h> 53 #include <vm/vm_object.h> 54 55 #include <machine/cpu.h> 56 57 #ifdef COMPAT_FREEBSD32 58 #include <compat/freebsd32/freebsd32_proto.h> 59 #endif 60 61 enum { 62 TYPE_SIMPLE_WAIT, 63 TYPE_CV, 64 TYPE_SEM, 65 TYPE_SIMPLE_LOCK, 66 TYPE_NORMAL_UMUTEX, 67 TYPE_PI_UMUTEX, 68 TYPE_PP_UMUTEX, 69 TYPE_RWLOCK 70 }; 71 72 #define _UMUTEX_TRY 1 73 #define _UMUTEX_WAIT 2 74 75 /* Key to represent a unique userland synchronous object */ 76 struct umtx_key { 77 int hash; 78 int type; 79 int shared; 80 union { 81 struct { 82 vm_object_t object; 83 uintptr_t offset; 84 } shared; 85 struct { 86 struct vmspace *vs; 87 uintptr_t addr; 88 } private; 89 struct { 90 void *a; 91 uintptr_t b; 92 } both; 93 } info; 94 }; 95 96 /* Priority inheritance mutex info. */ 97 struct umtx_pi { 98 /* Owner thread */ 99 struct thread *pi_owner; 100 101 /* Reference count */ 102 int pi_refcount; 103 104 /* List entry to link umtx holding by thread */ 105 TAILQ_ENTRY(umtx_pi) pi_link; 106 107 /* List entry in hash */ 108 TAILQ_ENTRY(umtx_pi) pi_hashlink; 109 110 /* List for waiters */ 111 TAILQ_HEAD(,umtx_q) pi_blocked; 112 113 /* Identify a userland lock object */ 114 struct umtx_key pi_key; 115 }; 116 117 /* A userland synchronous object user. */ 118 struct umtx_q { 119 /* Linked list for the hash. */ 120 TAILQ_ENTRY(umtx_q) uq_link; 121 122 /* Umtx key. */ 123 struct umtx_key uq_key; 124 125 /* Umtx flags. */ 126 int uq_flags; 127 #define UQF_UMTXQ 0x0001 128 129 /* The thread waits on. */ 130 struct thread *uq_thread; 131 132 /* 133 * Blocked on PI mutex. read can use chain lock 134 * or umtx_lock, write must have both chain lock and 135 * umtx_lock being hold. 136 */ 137 struct umtx_pi *uq_pi_blocked; 138 139 /* On blocked list */ 140 TAILQ_ENTRY(umtx_q) uq_lockq; 141 142 /* Thread contending with us */ 143 TAILQ_HEAD(,umtx_pi) uq_pi_contested; 144 145 /* Inherited priority from PP mutex */ 146 u_char uq_inherited_pri; 147 148 /* Spare queue ready to be reused */ 149 struct umtxq_queue *uq_spare_queue; 150 151 /* The queue we on */ 152 struct umtxq_queue *uq_cur_queue; 153 }; 154 155 TAILQ_HEAD(umtxq_head, umtx_q); 156 157 /* Per-key wait-queue */ 158 struct umtxq_queue { 159 struct umtxq_head head; 160 struct umtx_key key; 161 LIST_ENTRY(umtxq_queue) link; 162 int length; 163 }; 164 165 LIST_HEAD(umtxq_list, umtxq_queue); 166 167 /* Userland lock object's wait-queue chain */ 168 struct umtxq_chain { 169 /* Lock for this chain. */ 170 struct mtx uc_lock; 171 172 /* List of sleep queues. */ 173 struct umtxq_list uc_queue[2]; 174 #define UMTX_SHARED_QUEUE 0 175 #define UMTX_EXCLUSIVE_QUEUE 1 176 177 LIST_HEAD(, umtxq_queue) uc_spare_queue; 178 179 /* Busy flag */ 180 char uc_busy; 181 182 /* Chain lock waiters */ 183 int uc_waiters; 184 185 /* All PI in the list */ 186 TAILQ_HEAD(,umtx_pi) uc_pi_list; 187 188 }; 189 190 #define UMTXQ_LOCKED_ASSERT(uc) mtx_assert(&(uc)->uc_lock, MA_OWNED) 191 #define UMTXQ_BUSY_ASSERT(uc) KASSERT(&(uc)->uc_busy, ("umtx chain is not busy")) 192 193 /* 194 * Don't propagate time-sharing priority, there is a security reason, 195 * a user can simply introduce PI-mutex, let thread A lock the mutex, 196 * and let another thread B block on the mutex, because B is 197 * sleeping, its priority will be boosted, this causes A's priority to 198 * be boosted via priority propagating too and will never be lowered even 199 * if it is using 100%CPU, this is unfair to other processes. 200 */ 201 202 #define UPRI(td) (((td)->td_user_pri >= PRI_MIN_TIMESHARE &&\ 203 (td)->td_user_pri <= PRI_MAX_TIMESHARE) ?\ 204 PRI_MAX_TIMESHARE : (td)->td_user_pri) 205 206 #define GOLDEN_RATIO_PRIME 2654404609U 207 #define UMTX_CHAINS 128 208 #define UMTX_SHIFTS (__WORD_BIT - 7) 209 210 #define THREAD_SHARE 0 211 #define PROCESS_SHARE 1 212 #define AUTO_SHARE 2 213 214 #define GET_SHARE(flags) \ 215 (((flags) & USYNC_PROCESS_SHARED) == 0 ? THREAD_SHARE : PROCESS_SHARE) 216 217 #define BUSY_SPINS 200 218 219 static uma_zone_t umtx_pi_zone; 220 static struct umtxq_chain umtxq_chains[2][UMTX_CHAINS]; 221 static MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory"); 222 static int umtx_pi_allocated; 223 224 SYSCTL_NODE(_debug, OID_AUTO, umtx, CTLFLAG_RW, 0, "umtx debug"); 225 SYSCTL_INT(_debug_umtx, OID_AUTO, umtx_pi_allocated, CTLFLAG_RD, 226 &umtx_pi_allocated, 0, "Allocated umtx_pi"); 227 228 static void umtxq_sysinit(void *); 229 static void umtxq_hash(struct umtx_key *key); 230 static struct umtxq_chain *umtxq_getchain(struct umtx_key *key); 231 static void umtxq_lock(struct umtx_key *key); 232 static void umtxq_unlock(struct umtx_key *key); 233 static void umtxq_busy(struct umtx_key *key); 234 static void umtxq_unbusy(struct umtx_key *key); 235 static void umtxq_insert_queue(struct umtx_q *uq, int q); 236 static void umtxq_remove_queue(struct umtx_q *uq, int q); 237 static int umtxq_sleep(struct umtx_q *uq, const char *wmesg, int timo); 238 static int umtxq_count(struct umtx_key *key); 239 static int umtx_key_match(const struct umtx_key *k1, const struct umtx_key *k2); 240 static int umtx_key_get(void *addr, int type, int share, 241 struct umtx_key *key); 242 static void umtx_key_release(struct umtx_key *key); 243 static struct umtx_pi *umtx_pi_alloc(int); 244 static void umtx_pi_free(struct umtx_pi *pi); 245 static void umtx_pi_adjust_locked(struct thread *td, u_char oldpri); 246 static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags); 247 static void umtx_thread_cleanup(struct thread *td); 248 static void umtx_exec_hook(void *arg __unused, struct proc *p __unused, 249 struct image_params *imgp __unused); 250 SYSINIT(umtx, SI_SUB_EVENTHANDLER+1, SI_ORDER_MIDDLE, umtxq_sysinit, NULL); 251 252 #define umtxq_signal(key, nwake) umtxq_signal_queue((key), (nwake), UMTX_SHARED_QUEUE) 253 #define umtxq_insert(uq) umtxq_insert_queue((uq), UMTX_SHARED_QUEUE) 254 #define umtxq_remove(uq) umtxq_remove_queue((uq), UMTX_SHARED_QUEUE) 255 256 static struct mtx umtx_lock; 257 258 static void 259 umtxq_sysinit(void *arg __unused) 260 { 261 int i, j; 262 263 umtx_pi_zone = uma_zcreate("umtx pi", sizeof(struct umtx_pi), 264 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 265 for (i = 0; i < 2; ++i) { 266 for (j = 0; j < UMTX_CHAINS; ++j) { 267 mtx_init(&umtxq_chains[i][j].uc_lock, "umtxql", NULL, 268 MTX_DEF | MTX_DUPOK); 269 LIST_INIT(&umtxq_chains[i][j].uc_queue[0]); 270 LIST_INIT(&umtxq_chains[i][j].uc_queue[1]); 271 LIST_INIT(&umtxq_chains[i][j].uc_spare_queue); 272 TAILQ_INIT(&umtxq_chains[i][j].uc_pi_list); 273 umtxq_chains[i][j].uc_busy = 0; 274 umtxq_chains[i][j].uc_waiters = 0; 275 } 276 } 277 mtx_init(&umtx_lock, "umtx lock", NULL, MTX_SPIN); 278 EVENTHANDLER_REGISTER(process_exec, umtx_exec_hook, NULL, 279 EVENTHANDLER_PRI_ANY); 280 } 281 282 struct umtx_q * 283 umtxq_alloc(void) 284 { 285 struct umtx_q *uq; 286 287 uq = malloc(sizeof(struct umtx_q), M_UMTX, M_WAITOK | M_ZERO); 288 uq->uq_spare_queue = malloc(sizeof(struct umtxq_queue), M_UMTX, M_WAITOK | M_ZERO); 289 TAILQ_INIT(&uq->uq_spare_queue->head); 290 TAILQ_INIT(&uq->uq_pi_contested); 291 uq->uq_inherited_pri = PRI_MAX; 292 return (uq); 293 } 294 295 void 296 umtxq_free(struct umtx_q *uq) 297 { 298 MPASS(uq->uq_spare_queue != NULL); 299 free(uq->uq_spare_queue, M_UMTX); 300 free(uq, M_UMTX); 301 } 302 303 static inline void 304 umtxq_hash(struct umtx_key *key) 305 { 306 unsigned n = (uintptr_t)key->info.both.a + key->info.both.b; 307 key->hash = ((n * GOLDEN_RATIO_PRIME) >> UMTX_SHIFTS) % UMTX_CHAINS; 308 } 309 310 static inline int 311 umtx_key_match(const struct umtx_key *k1, const struct umtx_key *k2) 312 { 313 return (k1->type == k2->type && 314 k1->info.both.a == k2->info.both.a && 315 k1->info.both.b == k2->info.both.b); 316 } 317 318 static inline struct umtxq_chain * 319 umtxq_getchain(struct umtx_key *key) 320 { 321 if (key->type <= TYPE_SEM) 322 return (&umtxq_chains[1][key->hash]); 323 return (&umtxq_chains[0][key->hash]); 324 } 325 326 /* 327 * Lock a chain. 328 */ 329 static inline void 330 umtxq_lock(struct umtx_key *key) 331 { 332 struct umtxq_chain *uc; 333 334 uc = umtxq_getchain(key); 335 mtx_lock(&uc->uc_lock); 336 } 337 338 /* 339 * Unlock a chain. 340 */ 341 static inline void 342 umtxq_unlock(struct umtx_key *key) 343 { 344 struct umtxq_chain *uc; 345 346 uc = umtxq_getchain(key); 347 mtx_unlock(&uc->uc_lock); 348 } 349 350 /* 351 * Set chain to busy state when following operation 352 * may be blocked (kernel mutex can not be used). 353 */ 354 static inline void 355 umtxq_busy(struct umtx_key *key) 356 { 357 struct umtxq_chain *uc; 358 359 uc = umtxq_getchain(key); 360 mtx_assert(&uc->uc_lock, MA_OWNED); 361 if (uc->uc_busy) { 362 #ifdef SMP 363 if (smp_cpus > 1) { 364 int count = BUSY_SPINS; 365 if (count > 0) { 366 umtxq_unlock(key); 367 while (uc->uc_busy && --count > 0) 368 cpu_spinwait(); 369 umtxq_lock(key); 370 } 371 } 372 #endif 373 while (uc->uc_busy) { 374 uc->uc_waiters++; 375 msleep(uc, &uc->uc_lock, 0, "umtxqb", 0); 376 uc->uc_waiters--; 377 } 378 } 379 uc->uc_busy = 1; 380 } 381 382 /* 383 * Unbusy a chain. 384 */ 385 static inline void 386 umtxq_unbusy(struct umtx_key *key) 387 { 388 struct umtxq_chain *uc; 389 390 uc = umtxq_getchain(key); 391 mtx_assert(&uc->uc_lock, MA_OWNED); 392 KASSERT(uc->uc_busy != 0, ("not busy")); 393 uc->uc_busy = 0; 394 if (uc->uc_waiters) 395 wakeup_one(uc); 396 } 397 398 static struct umtxq_queue * 399 umtxq_queue_lookup(struct umtx_key *key, int q) 400 { 401 struct umtxq_queue *uh; 402 struct umtxq_chain *uc; 403 404 uc = umtxq_getchain(key); 405 UMTXQ_LOCKED_ASSERT(uc); 406 LIST_FOREACH(uh, &uc->uc_queue[q], link) { 407 if (umtx_key_match(&uh->key, key)) 408 return (uh); 409 } 410 411 return (NULL); 412 } 413 414 static inline void 415 umtxq_insert_queue(struct umtx_q *uq, int q) 416 { 417 struct umtxq_queue *uh; 418 struct umtxq_chain *uc; 419 420 uc = umtxq_getchain(&uq->uq_key); 421 UMTXQ_LOCKED_ASSERT(uc); 422 KASSERT((uq->uq_flags & UQF_UMTXQ) == 0, ("umtx_q is already on queue")); 423 uh = umtxq_queue_lookup(&uq->uq_key, q); 424 if (uh != NULL) { 425 LIST_INSERT_HEAD(&uc->uc_spare_queue, uq->uq_spare_queue, link); 426 } else { 427 uh = uq->uq_spare_queue; 428 uh->key = uq->uq_key; 429 LIST_INSERT_HEAD(&uc->uc_queue[q], uh, link); 430 } 431 uq->uq_spare_queue = NULL; 432 433 TAILQ_INSERT_TAIL(&uh->head, uq, uq_link); 434 uh->length++; 435 uq->uq_flags |= UQF_UMTXQ; 436 uq->uq_cur_queue = uh; 437 return; 438 } 439 440 static inline void 441 umtxq_remove_queue(struct umtx_q *uq, int q) 442 { 443 struct umtxq_chain *uc; 444 struct umtxq_queue *uh; 445 446 uc = umtxq_getchain(&uq->uq_key); 447 UMTXQ_LOCKED_ASSERT(uc); 448 if (uq->uq_flags & UQF_UMTXQ) { 449 uh = uq->uq_cur_queue; 450 TAILQ_REMOVE(&uh->head, uq, uq_link); 451 uh->length--; 452 uq->uq_flags &= ~UQF_UMTXQ; 453 if (TAILQ_EMPTY(&uh->head)) { 454 KASSERT(uh->length == 0, 455 ("inconsistent umtxq_queue length")); 456 LIST_REMOVE(uh, link); 457 } else { 458 uh = LIST_FIRST(&uc->uc_spare_queue); 459 KASSERT(uh != NULL, ("uc_spare_queue is empty")); 460 LIST_REMOVE(uh, link); 461 } 462 uq->uq_spare_queue = uh; 463 uq->uq_cur_queue = NULL; 464 } 465 } 466 467 /* 468 * Check if there are multiple waiters 469 */ 470 static int 471 umtxq_count(struct umtx_key *key) 472 { 473 struct umtxq_chain *uc; 474 struct umtxq_queue *uh; 475 476 uc = umtxq_getchain(key); 477 UMTXQ_LOCKED_ASSERT(uc); 478 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE); 479 if (uh != NULL) 480 return (uh->length); 481 return (0); 482 } 483 484 /* 485 * Check if there are multiple PI waiters and returns first 486 * waiter. 487 */ 488 static int 489 umtxq_count_pi(struct umtx_key *key, struct umtx_q **first) 490 { 491 struct umtxq_chain *uc; 492 struct umtxq_queue *uh; 493 494 *first = NULL; 495 uc = umtxq_getchain(key); 496 UMTXQ_LOCKED_ASSERT(uc); 497 uh = umtxq_queue_lookup(key, UMTX_SHARED_QUEUE); 498 if (uh != NULL) { 499 *first = TAILQ_FIRST(&uh->head); 500 return (uh->length); 501 } 502 return (0); 503 } 504 505 /* 506 * Wake up threads waiting on an userland object. 507 */ 508 509 static int 510 umtxq_signal_queue(struct umtx_key *key, int n_wake, int q) 511 { 512 struct umtxq_chain *uc; 513 struct umtxq_queue *uh; 514 struct umtx_q *uq; 515 int ret; 516 517 ret = 0; 518 uc = umtxq_getchain(key); 519 UMTXQ_LOCKED_ASSERT(uc); 520 uh = umtxq_queue_lookup(key, q); 521 if (uh != NULL) { 522 while ((uq = TAILQ_FIRST(&uh->head)) != NULL) { 523 umtxq_remove_queue(uq, q); 524 wakeup(uq); 525 if (++ret >= n_wake) 526 return (ret); 527 } 528 } 529 return (ret); 530 } 531 532 533 /* 534 * Wake up specified thread. 535 */ 536 static inline void 537 umtxq_signal_thread(struct umtx_q *uq) 538 { 539 struct umtxq_chain *uc; 540 541 uc = umtxq_getchain(&uq->uq_key); 542 UMTXQ_LOCKED_ASSERT(uc); 543 umtxq_remove(uq); 544 wakeup(uq); 545 } 546 547 /* 548 * Put thread into sleep state, before sleeping, check if 549 * thread was removed from umtx queue. 550 */ 551 static inline int 552 umtxq_sleep(struct umtx_q *uq, const char *wmesg, int timo) 553 { 554 struct umtxq_chain *uc; 555 int error; 556 557 uc = umtxq_getchain(&uq->uq_key); 558 UMTXQ_LOCKED_ASSERT(uc); 559 if (!(uq->uq_flags & UQF_UMTXQ)) 560 return (0); 561 error = msleep(uq, &uc->uc_lock, PCATCH, wmesg, timo); 562 if (error == EWOULDBLOCK) 563 error = ETIMEDOUT; 564 return (error); 565 } 566 567 /* 568 * Convert userspace address into unique logical address. 569 */ 570 static int 571 umtx_key_get(void *addr, int type, int share, struct umtx_key *key) 572 { 573 struct thread *td = curthread; 574 vm_map_t map; 575 vm_map_entry_t entry; 576 vm_pindex_t pindex; 577 vm_prot_t prot; 578 boolean_t wired; 579 580 key->type = type; 581 if (share == THREAD_SHARE) { 582 key->shared = 0; 583 key->info.private.vs = td->td_proc->p_vmspace; 584 key->info.private.addr = (uintptr_t)addr; 585 } else { 586 MPASS(share == PROCESS_SHARE || share == AUTO_SHARE); 587 map = &td->td_proc->p_vmspace->vm_map; 588 if (vm_map_lookup(&map, (vm_offset_t)addr, VM_PROT_WRITE, 589 &entry, &key->info.shared.object, &pindex, &prot, 590 &wired) != KERN_SUCCESS) { 591 return EFAULT; 592 } 593 594 if ((share == PROCESS_SHARE) || 595 (share == AUTO_SHARE && 596 VM_INHERIT_SHARE == entry->inheritance)) { 597 key->shared = 1; 598 key->info.shared.offset = entry->offset + entry->start - 599 (vm_offset_t)addr; 600 vm_object_reference(key->info.shared.object); 601 } else { 602 key->shared = 0; 603 key->info.private.vs = td->td_proc->p_vmspace; 604 key->info.private.addr = (uintptr_t)addr; 605 } 606 vm_map_lookup_done(map, entry); 607 } 608 609 umtxq_hash(key); 610 return (0); 611 } 612 613 /* 614 * Release key. 615 */ 616 static inline void 617 umtx_key_release(struct umtx_key *key) 618 { 619 if (key->shared) 620 vm_object_deallocate(key->info.shared.object); 621 } 622 623 /* 624 * Lock a umtx object. 625 */ 626 static int 627 _do_lock_umtx(struct thread *td, struct umtx *umtx, u_long id, int timo) 628 { 629 struct umtx_q *uq; 630 u_long owner; 631 u_long old; 632 int error = 0; 633 634 uq = td->td_umtxq; 635 636 /* 637 * Care must be exercised when dealing with umtx structure. It 638 * can fault on any access. 639 */ 640 for (;;) { 641 /* 642 * Try the uncontested case. This should be done in userland. 643 */ 644 owner = casuword(&umtx->u_owner, UMTX_UNOWNED, id); 645 646 /* The acquire succeeded. */ 647 if (owner == UMTX_UNOWNED) 648 return (0); 649 650 /* The address was invalid. */ 651 if (owner == -1) 652 return (EFAULT); 653 654 /* If no one owns it but it is contested try to acquire it. */ 655 if (owner == UMTX_CONTESTED) { 656 owner = casuword(&umtx->u_owner, 657 UMTX_CONTESTED, id | UMTX_CONTESTED); 658 659 if (owner == UMTX_CONTESTED) 660 return (0); 661 662 /* The address was invalid. */ 663 if (owner == -1) 664 return (EFAULT); 665 666 /* If this failed the lock has changed, restart. */ 667 continue; 668 } 669 670 /* 671 * If we caught a signal, we have retried and now 672 * exit immediately. 673 */ 674 if (error != 0) 675 return (error); 676 677 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK, 678 AUTO_SHARE, &uq->uq_key)) != 0) 679 return (error); 680 681 umtxq_lock(&uq->uq_key); 682 umtxq_busy(&uq->uq_key); 683 umtxq_insert(uq); 684 umtxq_unbusy(&uq->uq_key); 685 umtxq_unlock(&uq->uq_key); 686 687 /* 688 * Set the contested bit so that a release in user space 689 * knows to use the system call for unlock. If this fails 690 * either some one else has acquired the lock or it has been 691 * released. 692 */ 693 old = casuword(&umtx->u_owner, owner, owner | UMTX_CONTESTED); 694 695 /* The address was invalid. */ 696 if (old == -1) { 697 umtxq_lock(&uq->uq_key); 698 umtxq_remove(uq); 699 umtxq_unlock(&uq->uq_key); 700 umtx_key_release(&uq->uq_key); 701 return (EFAULT); 702 } 703 704 /* 705 * We set the contested bit, sleep. Otherwise the lock changed 706 * and we need to retry or we lost a race to the thread 707 * unlocking the umtx. 708 */ 709 umtxq_lock(&uq->uq_key); 710 if (old == owner) 711 error = umtxq_sleep(uq, "umtx", timo); 712 umtxq_remove(uq); 713 umtxq_unlock(&uq->uq_key); 714 umtx_key_release(&uq->uq_key); 715 } 716 717 return (0); 718 } 719 720 /* 721 * Lock a umtx object. 722 */ 723 static int 724 do_lock_umtx(struct thread *td, struct umtx *umtx, u_long id, 725 struct timespec *timeout) 726 { 727 struct timespec ts, ts2, ts3; 728 struct timeval tv; 729 int error; 730 731 if (timeout == NULL) { 732 error = _do_lock_umtx(td, umtx, id, 0); 733 /* Mutex locking is restarted if it is interrupted. */ 734 if (error == EINTR) 735 error = ERESTART; 736 } else { 737 getnanouptime(&ts); 738 timespecadd(&ts, timeout); 739 TIMESPEC_TO_TIMEVAL(&tv, timeout); 740 for (;;) { 741 error = _do_lock_umtx(td, umtx, id, tvtohz(&tv)); 742 if (error != ETIMEDOUT) 743 break; 744 getnanouptime(&ts2); 745 if (timespeccmp(&ts2, &ts, >=)) { 746 error = ETIMEDOUT; 747 break; 748 } 749 ts3 = ts; 750 timespecsub(&ts3, &ts2); 751 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 752 } 753 /* Timed-locking is not restarted. */ 754 if (error == ERESTART) 755 error = EINTR; 756 } 757 return (error); 758 } 759 760 /* 761 * Unlock a umtx object. 762 */ 763 static int 764 do_unlock_umtx(struct thread *td, struct umtx *umtx, u_long id) 765 { 766 struct umtx_key key; 767 u_long owner; 768 u_long old; 769 int error; 770 int count; 771 772 /* 773 * Make sure we own this mtx. 774 */ 775 owner = fuword(__DEVOLATILE(u_long *, &umtx->u_owner)); 776 if (owner == -1) 777 return (EFAULT); 778 779 if ((owner & ~UMTX_CONTESTED) != id) 780 return (EPERM); 781 782 /* This should be done in userland */ 783 if ((owner & UMTX_CONTESTED) == 0) { 784 old = casuword(&umtx->u_owner, owner, UMTX_UNOWNED); 785 if (old == -1) 786 return (EFAULT); 787 if (old == owner) 788 return (0); 789 owner = old; 790 } 791 792 /* We should only ever be in here for contested locks */ 793 if ((error = umtx_key_get(umtx, TYPE_SIMPLE_LOCK, AUTO_SHARE, 794 &key)) != 0) 795 return (error); 796 797 umtxq_lock(&key); 798 umtxq_busy(&key); 799 count = umtxq_count(&key); 800 umtxq_unlock(&key); 801 802 /* 803 * When unlocking the umtx, it must be marked as unowned if 804 * there is zero or one thread only waiting for it. 805 * Otherwise, it must be marked as contested. 806 */ 807 old = casuword(&umtx->u_owner, owner, 808 count <= 1 ? UMTX_UNOWNED : UMTX_CONTESTED); 809 umtxq_lock(&key); 810 umtxq_signal(&key,1); 811 umtxq_unbusy(&key); 812 umtxq_unlock(&key); 813 umtx_key_release(&key); 814 if (old == -1) 815 return (EFAULT); 816 if (old != owner) 817 return (EINVAL); 818 return (0); 819 } 820 821 #ifdef COMPAT_FREEBSD32 822 823 /* 824 * Lock a umtx object. 825 */ 826 static int 827 _do_lock_umtx32(struct thread *td, uint32_t *m, uint32_t id, int timo) 828 { 829 struct umtx_q *uq; 830 uint32_t owner; 831 uint32_t old; 832 int error = 0; 833 834 uq = td->td_umtxq; 835 836 /* 837 * Care must be exercised when dealing with umtx structure. It 838 * can fault on any access. 839 */ 840 for (;;) { 841 /* 842 * Try the uncontested case. This should be done in userland. 843 */ 844 owner = casuword32(m, UMUTEX_UNOWNED, id); 845 846 /* The acquire succeeded. */ 847 if (owner == UMUTEX_UNOWNED) 848 return (0); 849 850 /* The address was invalid. */ 851 if (owner == -1) 852 return (EFAULT); 853 854 /* If no one owns it but it is contested try to acquire it. */ 855 if (owner == UMUTEX_CONTESTED) { 856 owner = casuword32(m, 857 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 858 if (owner == UMUTEX_CONTESTED) 859 return (0); 860 861 /* The address was invalid. */ 862 if (owner == -1) 863 return (EFAULT); 864 865 /* If this failed the lock has changed, restart. */ 866 continue; 867 } 868 869 /* 870 * If we caught a signal, we have retried and now 871 * exit immediately. 872 */ 873 if (error != 0) 874 return (error); 875 876 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK, 877 AUTO_SHARE, &uq->uq_key)) != 0) 878 return (error); 879 880 umtxq_lock(&uq->uq_key); 881 umtxq_busy(&uq->uq_key); 882 umtxq_insert(uq); 883 umtxq_unbusy(&uq->uq_key); 884 umtxq_unlock(&uq->uq_key); 885 886 /* 887 * Set the contested bit so that a release in user space 888 * knows to use the system call for unlock. If this fails 889 * either some one else has acquired the lock or it has been 890 * released. 891 */ 892 old = casuword32(m, owner, owner | UMUTEX_CONTESTED); 893 894 /* The address was invalid. */ 895 if (old == -1) { 896 umtxq_lock(&uq->uq_key); 897 umtxq_remove(uq); 898 umtxq_unlock(&uq->uq_key); 899 umtx_key_release(&uq->uq_key); 900 return (EFAULT); 901 } 902 903 /* 904 * We set the contested bit, sleep. Otherwise the lock changed 905 * and we need to retry or we lost a race to the thread 906 * unlocking the umtx. 907 */ 908 umtxq_lock(&uq->uq_key); 909 if (old == owner) 910 error = umtxq_sleep(uq, "umtx", timo); 911 umtxq_remove(uq); 912 umtxq_unlock(&uq->uq_key); 913 umtx_key_release(&uq->uq_key); 914 } 915 916 return (0); 917 } 918 919 /* 920 * Lock a umtx object. 921 */ 922 static int 923 do_lock_umtx32(struct thread *td, void *m, uint32_t id, 924 struct timespec *timeout) 925 { 926 struct timespec ts, ts2, ts3; 927 struct timeval tv; 928 int error; 929 930 if (timeout == NULL) { 931 error = _do_lock_umtx32(td, m, id, 0); 932 /* Mutex locking is restarted if it is interrupted. */ 933 if (error == EINTR) 934 error = ERESTART; 935 } else { 936 getnanouptime(&ts); 937 timespecadd(&ts, timeout); 938 TIMESPEC_TO_TIMEVAL(&tv, timeout); 939 for (;;) { 940 error = _do_lock_umtx32(td, m, id, tvtohz(&tv)); 941 if (error != ETIMEDOUT) 942 break; 943 getnanouptime(&ts2); 944 if (timespeccmp(&ts2, &ts, >=)) { 945 error = ETIMEDOUT; 946 break; 947 } 948 ts3 = ts; 949 timespecsub(&ts3, &ts2); 950 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 951 } 952 /* Timed-locking is not restarted. */ 953 if (error == ERESTART) 954 error = EINTR; 955 } 956 return (error); 957 } 958 959 /* 960 * Unlock a umtx object. 961 */ 962 static int 963 do_unlock_umtx32(struct thread *td, uint32_t *m, uint32_t id) 964 { 965 struct umtx_key key; 966 uint32_t owner; 967 uint32_t old; 968 int error; 969 int count; 970 971 /* 972 * Make sure we own this mtx. 973 */ 974 owner = fuword32(m); 975 if (owner == -1) 976 return (EFAULT); 977 978 if ((owner & ~UMUTEX_CONTESTED) != id) 979 return (EPERM); 980 981 /* This should be done in userland */ 982 if ((owner & UMUTEX_CONTESTED) == 0) { 983 old = casuword32(m, owner, UMUTEX_UNOWNED); 984 if (old == -1) 985 return (EFAULT); 986 if (old == owner) 987 return (0); 988 owner = old; 989 } 990 991 /* We should only ever be in here for contested locks */ 992 if ((error = umtx_key_get(m, TYPE_SIMPLE_LOCK, AUTO_SHARE, 993 &key)) != 0) 994 return (error); 995 996 umtxq_lock(&key); 997 umtxq_busy(&key); 998 count = umtxq_count(&key); 999 umtxq_unlock(&key); 1000 1001 /* 1002 * When unlocking the umtx, it must be marked as unowned if 1003 * there is zero or one thread only waiting for it. 1004 * Otherwise, it must be marked as contested. 1005 */ 1006 old = casuword32(m, owner, 1007 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED); 1008 umtxq_lock(&key); 1009 umtxq_signal(&key,1); 1010 umtxq_unbusy(&key); 1011 umtxq_unlock(&key); 1012 umtx_key_release(&key); 1013 if (old == -1) 1014 return (EFAULT); 1015 if (old != owner) 1016 return (EINVAL); 1017 return (0); 1018 } 1019 #endif 1020 1021 /* 1022 * Fetch and compare value, sleep on the address if value is not changed. 1023 */ 1024 static int 1025 do_wait(struct thread *td, void *addr, u_long id, 1026 struct timespec *timeout, int compat32, int is_private) 1027 { 1028 struct umtx_q *uq; 1029 struct timespec ts, ts2, ts3; 1030 struct timeval tv; 1031 u_long tmp; 1032 int error = 0; 1033 1034 uq = td->td_umtxq; 1035 if ((error = umtx_key_get(addr, TYPE_SIMPLE_WAIT, 1036 is_private ? THREAD_SHARE : AUTO_SHARE, &uq->uq_key)) != 0) 1037 return (error); 1038 1039 umtxq_lock(&uq->uq_key); 1040 umtxq_insert(uq); 1041 umtxq_unlock(&uq->uq_key); 1042 if (compat32 == 0) 1043 tmp = fuword(addr); 1044 else 1045 tmp = (unsigned int)fuword32(addr); 1046 if (tmp != id) { 1047 umtxq_lock(&uq->uq_key); 1048 umtxq_remove(uq); 1049 umtxq_unlock(&uq->uq_key); 1050 } else if (timeout == NULL) { 1051 umtxq_lock(&uq->uq_key); 1052 error = umtxq_sleep(uq, "uwait", 0); 1053 umtxq_remove(uq); 1054 umtxq_unlock(&uq->uq_key); 1055 } else { 1056 getnanouptime(&ts); 1057 timespecadd(&ts, timeout); 1058 TIMESPEC_TO_TIMEVAL(&tv, timeout); 1059 umtxq_lock(&uq->uq_key); 1060 for (;;) { 1061 error = umtxq_sleep(uq, "uwait", tvtohz(&tv)); 1062 if (!(uq->uq_flags & UQF_UMTXQ)) { 1063 error = 0; 1064 break; 1065 } 1066 if (error != ETIMEDOUT) 1067 break; 1068 umtxq_unlock(&uq->uq_key); 1069 getnanouptime(&ts2); 1070 if (timespeccmp(&ts2, &ts, >=)) { 1071 error = ETIMEDOUT; 1072 umtxq_lock(&uq->uq_key); 1073 break; 1074 } 1075 ts3 = ts; 1076 timespecsub(&ts3, &ts2); 1077 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 1078 umtxq_lock(&uq->uq_key); 1079 } 1080 umtxq_remove(uq); 1081 umtxq_unlock(&uq->uq_key); 1082 } 1083 umtx_key_release(&uq->uq_key); 1084 if (error == ERESTART) 1085 error = EINTR; 1086 return (error); 1087 } 1088 1089 /* 1090 * Wake up threads sleeping on the specified address. 1091 */ 1092 int 1093 kern_umtx_wake(struct thread *td, void *uaddr, int n_wake, int is_private) 1094 { 1095 struct umtx_key key; 1096 int ret; 1097 1098 if ((ret = umtx_key_get(uaddr, TYPE_SIMPLE_WAIT, 1099 is_private ? THREAD_SHARE : AUTO_SHARE, &key)) != 0) 1100 return (ret); 1101 umtxq_lock(&key); 1102 ret = umtxq_signal(&key, n_wake); 1103 umtxq_unlock(&key); 1104 umtx_key_release(&key); 1105 return (0); 1106 } 1107 1108 /* 1109 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex. 1110 */ 1111 static int 1112 _do_lock_normal(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1113 int mode) 1114 { 1115 struct umtx_q *uq; 1116 uint32_t owner, old, id; 1117 int error = 0; 1118 1119 id = td->td_tid; 1120 uq = td->td_umtxq; 1121 1122 /* 1123 * Care must be exercised when dealing with umtx structure. It 1124 * can fault on any access. 1125 */ 1126 for (;;) { 1127 owner = fuword32(__DEVOLATILE(void *, &m->m_owner)); 1128 if (mode == _UMUTEX_WAIT) { 1129 if (owner == UMUTEX_UNOWNED || owner == UMUTEX_CONTESTED) 1130 return (0); 1131 } else { 1132 /* 1133 * Try the uncontested case. This should be done in userland. 1134 */ 1135 owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id); 1136 1137 /* The acquire succeeded. */ 1138 if (owner == UMUTEX_UNOWNED) 1139 return (0); 1140 1141 /* The address was invalid. */ 1142 if (owner == -1) 1143 return (EFAULT); 1144 1145 /* If no one owns it but it is contested try to acquire it. */ 1146 if (owner == UMUTEX_CONTESTED) { 1147 owner = casuword32(&m->m_owner, 1148 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 1149 1150 if (owner == UMUTEX_CONTESTED) 1151 return (0); 1152 1153 /* The address was invalid. */ 1154 if (owner == -1) 1155 return (EFAULT); 1156 1157 /* If this failed the lock has changed, restart. */ 1158 continue; 1159 } 1160 } 1161 1162 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 1163 (owner & ~UMUTEX_CONTESTED) == id) 1164 return (EDEADLK); 1165 1166 if (mode == _UMUTEX_TRY) 1167 return (EBUSY); 1168 1169 /* 1170 * If we caught a signal, we have retried and now 1171 * exit immediately. 1172 */ 1173 if (error != 0) 1174 return (error); 1175 1176 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, 1177 GET_SHARE(flags), &uq->uq_key)) != 0) 1178 return (error); 1179 1180 umtxq_lock(&uq->uq_key); 1181 umtxq_busy(&uq->uq_key); 1182 umtxq_insert(uq); 1183 umtxq_unlock(&uq->uq_key); 1184 1185 /* 1186 * Set the contested bit so that a release in user space 1187 * knows to use the system call for unlock. If this fails 1188 * either some one else has acquired the lock or it has been 1189 * released. 1190 */ 1191 old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED); 1192 1193 /* The address was invalid. */ 1194 if (old == -1) { 1195 umtxq_lock(&uq->uq_key); 1196 umtxq_remove(uq); 1197 umtxq_unbusy(&uq->uq_key); 1198 umtxq_unlock(&uq->uq_key); 1199 umtx_key_release(&uq->uq_key); 1200 return (EFAULT); 1201 } 1202 1203 /* 1204 * We set the contested bit, sleep. Otherwise the lock changed 1205 * and we need to retry or we lost a race to the thread 1206 * unlocking the umtx. 1207 */ 1208 umtxq_lock(&uq->uq_key); 1209 umtxq_unbusy(&uq->uq_key); 1210 if (old == owner) 1211 error = umtxq_sleep(uq, "umtxn", timo); 1212 umtxq_remove(uq); 1213 umtxq_unlock(&uq->uq_key); 1214 umtx_key_release(&uq->uq_key); 1215 } 1216 1217 return (0); 1218 } 1219 1220 /* 1221 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex. 1222 */ 1223 /* 1224 * Unlock PTHREAD_PRIO_NONE protocol POSIX mutex. 1225 */ 1226 static int 1227 do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags) 1228 { 1229 struct umtx_key key; 1230 uint32_t owner, old, id; 1231 int error; 1232 int count; 1233 1234 id = td->td_tid; 1235 /* 1236 * Make sure we own this mtx. 1237 */ 1238 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1239 if (owner == -1) 1240 return (EFAULT); 1241 1242 if ((owner & ~UMUTEX_CONTESTED) != id) 1243 return (EPERM); 1244 1245 if ((owner & UMUTEX_CONTESTED) == 0) { 1246 old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED); 1247 if (old == -1) 1248 return (EFAULT); 1249 if (old == owner) 1250 return (0); 1251 owner = old; 1252 } 1253 1254 /* We should only ever be in here for contested locks */ 1255 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags), 1256 &key)) != 0) 1257 return (error); 1258 1259 umtxq_lock(&key); 1260 umtxq_busy(&key); 1261 count = umtxq_count(&key); 1262 umtxq_unlock(&key); 1263 1264 /* 1265 * When unlocking the umtx, it must be marked as unowned if 1266 * there is zero or one thread only waiting for it. 1267 * Otherwise, it must be marked as contested. 1268 */ 1269 old = casuword32(&m->m_owner, owner, 1270 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED); 1271 umtxq_lock(&key); 1272 umtxq_signal(&key,1); 1273 umtxq_unbusy(&key); 1274 umtxq_unlock(&key); 1275 umtx_key_release(&key); 1276 if (old == -1) 1277 return (EFAULT); 1278 if (old != owner) 1279 return (EINVAL); 1280 return (0); 1281 } 1282 1283 /* 1284 * Check if the mutex is available and wake up a waiter, 1285 * only for simple mutex. 1286 */ 1287 static int 1288 do_wake_umutex(struct thread *td, struct umutex *m) 1289 { 1290 struct umtx_key key; 1291 uint32_t owner; 1292 uint32_t flags; 1293 int error; 1294 int count; 1295 1296 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1297 if (owner == -1) 1298 return (EFAULT); 1299 1300 if ((owner & ~UMUTEX_CONTESTED) != 0) 1301 return (0); 1302 1303 flags = fuword32(&m->m_flags); 1304 1305 /* We should only ever be in here for contested locks */ 1306 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags), 1307 &key)) != 0) 1308 return (error); 1309 1310 umtxq_lock(&key); 1311 umtxq_busy(&key); 1312 count = umtxq_count(&key); 1313 umtxq_unlock(&key); 1314 1315 if (count <= 1) 1316 owner = casuword32(&m->m_owner, UMUTEX_CONTESTED, UMUTEX_UNOWNED); 1317 1318 umtxq_lock(&key); 1319 if (count != 0 && (owner & ~UMUTEX_CONTESTED) == 0) 1320 umtxq_signal(&key, 1); 1321 umtxq_unbusy(&key); 1322 umtxq_unlock(&key); 1323 umtx_key_release(&key); 1324 return (0); 1325 } 1326 1327 static inline struct umtx_pi * 1328 umtx_pi_alloc(int flags) 1329 { 1330 struct umtx_pi *pi; 1331 1332 pi = uma_zalloc(umtx_pi_zone, M_ZERO | flags); 1333 TAILQ_INIT(&pi->pi_blocked); 1334 atomic_add_int(&umtx_pi_allocated, 1); 1335 return (pi); 1336 } 1337 1338 static inline void 1339 umtx_pi_free(struct umtx_pi *pi) 1340 { 1341 uma_zfree(umtx_pi_zone, pi); 1342 atomic_add_int(&umtx_pi_allocated, -1); 1343 } 1344 1345 /* 1346 * Adjust the thread's position on a pi_state after its priority has been 1347 * changed. 1348 */ 1349 static int 1350 umtx_pi_adjust_thread(struct umtx_pi *pi, struct thread *td) 1351 { 1352 struct umtx_q *uq, *uq1, *uq2; 1353 struct thread *td1; 1354 1355 mtx_assert(&umtx_lock, MA_OWNED); 1356 if (pi == NULL) 1357 return (0); 1358 1359 uq = td->td_umtxq; 1360 1361 /* 1362 * Check if the thread needs to be moved on the blocked chain. 1363 * It needs to be moved if either its priority is lower than 1364 * the previous thread or higher than the next thread. 1365 */ 1366 uq1 = TAILQ_PREV(uq, umtxq_head, uq_lockq); 1367 uq2 = TAILQ_NEXT(uq, uq_lockq); 1368 if ((uq1 != NULL && UPRI(td) < UPRI(uq1->uq_thread)) || 1369 (uq2 != NULL && UPRI(td) > UPRI(uq2->uq_thread))) { 1370 /* 1371 * Remove thread from blocked chain and determine where 1372 * it should be moved to. 1373 */ 1374 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq); 1375 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) { 1376 td1 = uq1->uq_thread; 1377 MPASS(td1->td_proc->p_magic == P_MAGIC); 1378 if (UPRI(td1) > UPRI(td)) 1379 break; 1380 } 1381 1382 if (uq1 == NULL) 1383 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq); 1384 else 1385 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq); 1386 } 1387 return (1); 1388 } 1389 1390 /* 1391 * Propagate priority when a thread is blocked on POSIX 1392 * PI mutex. 1393 */ 1394 static void 1395 umtx_propagate_priority(struct thread *td) 1396 { 1397 struct umtx_q *uq; 1398 struct umtx_pi *pi; 1399 int pri; 1400 1401 mtx_assert(&umtx_lock, MA_OWNED); 1402 pri = UPRI(td); 1403 uq = td->td_umtxq; 1404 pi = uq->uq_pi_blocked; 1405 if (pi == NULL) 1406 return; 1407 1408 for (;;) { 1409 td = pi->pi_owner; 1410 if (td == NULL) 1411 return; 1412 1413 MPASS(td->td_proc != NULL); 1414 MPASS(td->td_proc->p_magic == P_MAGIC); 1415 1416 if (UPRI(td) <= pri) 1417 return; 1418 1419 thread_lock(td); 1420 sched_lend_user_prio(td, pri); 1421 thread_unlock(td); 1422 1423 /* 1424 * Pick up the lock that td is blocked on. 1425 */ 1426 uq = td->td_umtxq; 1427 pi = uq->uq_pi_blocked; 1428 /* Resort td on the list if needed. */ 1429 if (!umtx_pi_adjust_thread(pi, td)) 1430 break; 1431 } 1432 } 1433 1434 /* 1435 * Unpropagate priority for a PI mutex when a thread blocked on 1436 * it is interrupted by signal or resumed by others. 1437 */ 1438 static void 1439 umtx_unpropagate_priority(struct umtx_pi *pi) 1440 { 1441 struct umtx_q *uq, *uq_owner; 1442 struct umtx_pi *pi2; 1443 int pri, oldpri; 1444 1445 mtx_assert(&umtx_lock, MA_OWNED); 1446 1447 while (pi != NULL && pi->pi_owner != NULL) { 1448 pri = PRI_MAX; 1449 uq_owner = pi->pi_owner->td_umtxq; 1450 1451 TAILQ_FOREACH(pi2, &uq_owner->uq_pi_contested, pi_link) { 1452 uq = TAILQ_FIRST(&pi2->pi_blocked); 1453 if (uq != NULL) { 1454 if (pri > UPRI(uq->uq_thread)) 1455 pri = UPRI(uq->uq_thread); 1456 } 1457 } 1458 1459 if (pri > uq_owner->uq_inherited_pri) 1460 pri = uq_owner->uq_inherited_pri; 1461 thread_lock(pi->pi_owner); 1462 oldpri = pi->pi_owner->td_user_pri; 1463 sched_unlend_user_prio(pi->pi_owner, pri); 1464 thread_unlock(pi->pi_owner); 1465 if (uq_owner->uq_pi_blocked != NULL) 1466 umtx_pi_adjust_locked(pi->pi_owner, oldpri); 1467 pi = uq_owner->uq_pi_blocked; 1468 } 1469 } 1470 1471 /* 1472 * Insert a PI mutex into owned list. 1473 */ 1474 static void 1475 umtx_pi_setowner(struct umtx_pi *pi, struct thread *owner) 1476 { 1477 struct umtx_q *uq_owner; 1478 1479 uq_owner = owner->td_umtxq; 1480 mtx_assert(&umtx_lock, MA_OWNED); 1481 if (pi->pi_owner != NULL) 1482 panic("pi_ower != NULL"); 1483 pi->pi_owner = owner; 1484 TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link); 1485 } 1486 1487 /* 1488 * Claim ownership of a PI mutex. 1489 */ 1490 static int 1491 umtx_pi_claim(struct umtx_pi *pi, struct thread *owner) 1492 { 1493 struct umtx_q *uq, *uq_owner; 1494 1495 uq_owner = owner->td_umtxq; 1496 mtx_lock_spin(&umtx_lock); 1497 if (pi->pi_owner == owner) { 1498 mtx_unlock_spin(&umtx_lock); 1499 return (0); 1500 } 1501 1502 if (pi->pi_owner != NULL) { 1503 /* 1504 * userland may have already messed the mutex, sigh. 1505 */ 1506 mtx_unlock_spin(&umtx_lock); 1507 return (EPERM); 1508 } 1509 umtx_pi_setowner(pi, owner); 1510 uq = TAILQ_FIRST(&pi->pi_blocked); 1511 if (uq != NULL) { 1512 int pri; 1513 1514 pri = UPRI(uq->uq_thread); 1515 thread_lock(owner); 1516 if (pri < UPRI(owner)) 1517 sched_lend_user_prio(owner, pri); 1518 thread_unlock(owner); 1519 } 1520 mtx_unlock_spin(&umtx_lock); 1521 return (0); 1522 } 1523 1524 static void 1525 umtx_pi_adjust_locked(struct thread *td, u_char oldpri) 1526 { 1527 struct umtx_q *uq; 1528 struct umtx_pi *pi; 1529 1530 uq = td->td_umtxq; 1531 /* 1532 * Pick up the lock that td is blocked on. 1533 */ 1534 pi = uq->uq_pi_blocked; 1535 MPASS(pi != NULL); 1536 1537 /* Resort the turnstile on the list. */ 1538 if (!umtx_pi_adjust_thread(pi, td)) 1539 return; 1540 1541 /* 1542 * If our priority was lowered and we are at the head of the 1543 * turnstile, then propagate our new priority up the chain. 1544 */ 1545 if (uq == TAILQ_FIRST(&pi->pi_blocked) && UPRI(td) < oldpri) 1546 umtx_propagate_priority(td); 1547 } 1548 1549 /* 1550 * Adjust a thread's order position in its blocked PI mutex, 1551 * this may result new priority propagating process. 1552 */ 1553 void 1554 umtx_pi_adjust(struct thread *td, u_char oldpri) 1555 { 1556 struct umtx_q *uq; 1557 struct umtx_pi *pi; 1558 1559 uq = td->td_umtxq; 1560 mtx_lock_spin(&umtx_lock); 1561 /* 1562 * Pick up the lock that td is blocked on. 1563 */ 1564 pi = uq->uq_pi_blocked; 1565 if (pi != NULL) 1566 umtx_pi_adjust_locked(td, oldpri); 1567 mtx_unlock_spin(&umtx_lock); 1568 } 1569 1570 /* 1571 * Sleep on a PI mutex. 1572 */ 1573 static int 1574 umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi, 1575 uint32_t owner, const char *wmesg, int timo) 1576 { 1577 struct umtxq_chain *uc; 1578 struct thread *td, *td1; 1579 struct umtx_q *uq1; 1580 int pri; 1581 int error = 0; 1582 1583 td = uq->uq_thread; 1584 KASSERT(td == curthread, ("inconsistent uq_thread")); 1585 uc = umtxq_getchain(&uq->uq_key); 1586 UMTXQ_LOCKED_ASSERT(uc); 1587 UMTXQ_BUSY_ASSERT(uc); 1588 umtxq_insert(uq); 1589 mtx_lock_spin(&umtx_lock); 1590 if (pi->pi_owner == NULL) { 1591 /* XXX 1592 * Current, We only support process private PI-mutex, 1593 * we need a faster way to find an owner thread for 1594 * process-shared mutex (not available yet). 1595 */ 1596 mtx_unlock_spin(&umtx_lock); 1597 PROC_LOCK(curproc); 1598 td1 = thread_find(curproc, owner); 1599 mtx_lock_spin(&umtx_lock); 1600 if (td1 != NULL && pi->pi_owner == NULL) { 1601 uq1 = td1->td_umtxq; 1602 umtx_pi_setowner(pi, td1); 1603 } 1604 PROC_UNLOCK(curproc); 1605 } 1606 1607 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) { 1608 pri = UPRI(uq1->uq_thread); 1609 if (pri > UPRI(td)) 1610 break; 1611 } 1612 1613 if (uq1 != NULL) 1614 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq); 1615 else 1616 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq); 1617 1618 uq->uq_pi_blocked = pi; 1619 thread_lock(td); 1620 td->td_flags |= TDF_UPIBLOCKED; 1621 thread_unlock(td); 1622 umtx_propagate_priority(td); 1623 mtx_unlock_spin(&umtx_lock); 1624 umtxq_unbusy(&uq->uq_key); 1625 1626 if (uq->uq_flags & UQF_UMTXQ) { 1627 error = msleep(uq, &uc->uc_lock, PCATCH, wmesg, timo); 1628 if (error == EWOULDBLOCK) 1629 error = ETIMEDOUT; 1630 if (uq->uq_flags & UQF_UMTXQ) { 1631 umtxq_remove(uq); 1632 } 1633 } 1634 mtx_lock_spin(&umtx_lock); 1635 uq->uq_pi_blocked = NULL; 1636 thread_lock(td); 1637 td->td_flags &= ~TDF_UPIBLOCKED; 1638 thread_unlock(td); 1639 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq); 1640 umtx_unpropagate_priority(pi); 1641 mtx_unlock_spin(&umtx_lock); 1642 umtxq_unlock(&uq->uq_key); 1643 1644 return (error); 1645 } 1646 1647 /* 1648 * Add reference count for a PI mutex. 1649 */ 1650 static void 1651 umtx_pi_ref(struct umtx_pi *pi) 1652 { 1653 struct umtxq_chain *uc; 1654 1655 uc = umtxq_getchain(&pi->pi_key); 1656 UMTXQ_LOCKED_ASSERT(uc); 1657 pi->pi_refcount++; 1658 } 1659 1660 /* 1661 * Decrease reference count for a PI mutex, if the counter 1662 * is decreased to zero, its memory space is freed. 1663 */ 1664 static void 1665 umtx_pi_unref(struct umtx_pi *pi) 1666 { 1667 struct umtxq_chain *uc; 1668 1669 uc = umtxq_getchain(&pi->pi_key); 1670 UMTXQ_LOCKED_ASSERT(uc); 1671 KASSERT(pi->pi_refcount > 0, ("invalid reference count")); 1672 if (--pi->pi_refcount == 0) { 1673 mtx_lock_spin(&umtx_lock); 1674 if (pi->pi_owner != NULL) { 1675 TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested, 1676 pi, pi_link); 1677 pi->pi_owner = NULL; 1678 } 1679 KASSERT(TAILQ_EMPTY(&pi->pi_blocked), 1680 ("blocked queue not empty")); 1681 mtx_unlock_spin(&umtx_lock); 1682 TAILQ_REMOVE(&uc->uc_pi_list, pi, pi_hashlink); 1683 umtx_pi_free(pi); 1684 } 1685 } 1686 1687 /* 1688 * Find a PI mutex in hash table. 1689 */ 1690 static struct umtx_pi * 1691 umtx_pi_lookup(struct umtx_key *key) 1692 { 1693 struct umtxq_chain *uc; 1694 struct umtx_pi *pi; 1695 1696 uc = umtxq_getchain(key); 1697 UMTXQ_LOCKED_ASSERT(uc); 1698 1699 TAILQ_FOREACH(pi, &uc->uc_pi_list, pi_hashlink) { 1700 if (umtx_key_match(&pi->pi_key, key)) { 1701 return (pi); 1702 } 1703 } 1704 return (NULL); 1705 } 1706 1707 /* 1708 * Insert a PI mutex into hash table. 1709 */ 1710 static inline void 1711 umtx_pi_insert(struct umtx_pi *pi) 1712 { 1713 struct umtxq_chain *uc; 1714 1715 uc = umtxq_getchain(&pi->pi_key); 1716 UMTXQ_LOCKED_ASSERT(uc); 1717 TAILQ_INSERT_TAIL(&uc->uc_pi_list, pi, pi_hashlink); 1718 } 1719 1720 /* 1721 * Lock a PI mutex. 1722 */ 1723 static int 1724 _do_lock_pi(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1725 int try) 1726 { 1727 struct umtx_q *uq; 1728 struct umtx_pi *pi, *new_pi; 1729 uint32_t id, owner, old; 1730 int error; 1731 1732 id = td->td_tid; 1733 uq = td->td_umtxq; 1734 1735 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags), 1736 &uq->uq_key)) != 0) 1737 return (error); 1738 umtxq_lock(&uq->uq_key); 1739 pi = umtx_pi_lookup(&uq->uq_key); 1740 if (pi == NULL) { 1741 new_pi = umtx_pi_alloc(M_NOWAIT); 1742 if (new_pi == NULL) { 1743 umtxq_unlock(&uq->uq_key); 1744 new_pi = umtx_pi_alloc(M_WAITOK); 1745 umtxq_lock(&uq->uq_key); 1746 pi = umtx_pi_lookup(&uq->uq_key); 1747 if (pi != NULL) { 1748 umtx_pi_free(new_pi); 1749 new_pi = NULL; 1750 } 1751 } 1752 if (new_pi != NULL) { 1753 new_pi->pi_key = uq->uq_key; 1754 umtx_pi_insert(new_pi); 1755 pi = new_pi; 1756 } 1757 } 1758 umtx_pi_ref(pi); 1759 umtxq_unlock(&uq->uq_key); 1760 1761 /* 1762 * Care must be exercised when dealing with umtx structure. It 1763 * can fault on any access. 1764 */ 1765 for (;;) { 1766 /* 1767 * Try the uncontested case. This should be done in userland. 1768 */ 1769 owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id); 1770 1771 /* The acquire succeeded. */ 1772 if (owner == UMUTEX_UNOWNED) { 1773 error = 0; 1774 break; 1775 } 1776 1777 /* The address was invalid. */ 1778 if (owner == -1) { 1779 error = EFAULT; 1780 break; 1781 } 1782 1783 /* If no one owns it but it is contested try to acquire it. */ 1784 if (owner == UMUTEX_CONTESTED) { 1785 owner = casuword32(&m->m_owner, 1786 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 1787 1788 if (owner == UMUTEX_CONTESTED) { 1789 umtxq_lock(&uq->uq_key); 1790 umtxq_busy(&uq->uq_key); 1791 error = umtx_pi_claim(pi, td); 1792 umtxq_unbusy(&uq->uq_key); 1793 umtxq_unlock(&uq->uq_key); 1794 break; 1795 } 1796 1797 /* The address was invalid. */ 1798 if (owner == -1) { 1799 error = EFAULT; 1800 break; 1801 } 1802 1803 /* If this failed the lock has changed, restart. */ 1804 continue; 1805 } 1806 1807 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 1808 (owner & ~UMUTEX_CONTESTED) == id) { 1809 error = EDEADLK; 1810 break; 1811 } 1812 1813 if (try != 0) { 1814 error = EBUSY; 1815 break; 1816 } 1817 1818 /* 1819 * If we caught a signal, we have retried and now 1820 * exit immediately. 1821 */ 1822 if (error != 0) 1823 break; 1824 1825 umtxq_lock(&uq->uq_key); 1826 umtxq_busy(&uq->uq_key); 1827 umtxq_unlock(&uq->uq_key); 1828 1829 /* 1830 * Set the contested bit so that a release in user space 1831 * knows to use the system call for unlock. If this fails 1832 * either some one else has acquired the lock or it has been 1833 * released. 1834 */ 1835 old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED); 1836 1837 /* The address was invalid. */ 1838 if (old == -1) { 1839 umtxq_lock(&uq->uq_key); 1840 umtxq_unbusy(&uq->uq_key); 1841 umtxq_unlock(&uq->uq_key); 1842 error = EFAULT; 1843 break; 1844 } 1845 1846 umtxq_lock(&uq->uq_key); 1847 /* 1848 * We set the contested bit, sleep. Otherwise the lock changed 1849 * and we need to retry or we lost a race to the thread 1850 * unlocking the umtx. 1851 */ 1852 if (old == owner) 1853 error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED, 1854 "umtxpi", timo); 1855 else { 1856 umtxq_unbusy(&uq->uq_key); 1857 umtxq_unlock(&uq->uq_key); 1858 } 1859 } 1860 1861 umtxq_lock(&uq->uq_key); 1862 umtx_pi_unref(pi); 1863 umtxq_unlock(&uq->uq_key); 1864 1865 umtx_key_release(&uq->uq_key); 1866 return (error); 1867 } 1868 1869 /* 1870 * Unlock a PI mutex. 1871 */ 1872 static int 1873 do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags) 1874 { 1875 struct umtx_key key; 1876 struct umtx_q *uq_first, *uq_first2, *uq_me; 1877 struct umtx_pi *pi, *pi2; 1878 uint32_t owner, old, id; 1879 int error; 1880 int count; 1881 int pri; 1882 1883 id = td->td_tid; 1884 /* 1885 * Make sure we own this mtx. 1886 */ 1887 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1888 if (owner == -1) 1889 return (EFAULT); 1890 1891 if ((owner & ~UMUTEX_CONTESTED) != id) 1892 return (EPERM); 1893 1894 /* This should be done in userland */ 1895 if ((owner & UMUTEX_CONTESTED) == 0) { 1896 old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED); 1897 if (old == -1) 1898 return (EFAULT); 1899 if (old == owner) 1900 return (0); 1901 owner = old; 1902 } 1903 1904 /* We should only ever be in here for contested locks */ 1905 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags), 1906 &key)) != 0) 1907 return (error); 1908 1909 umtxq_lock(&key); 1910 umtxq_busy(&key); 1911 count = umtxq_count_pi(&key, &uq_first); 1912 if (uq_first != NULL) { 1913 mtx_lock_spin(&umtx_lock); 1914 pi = uq_first->uq_pi_blocked; 1915 KASSERT(pi != NULL, ("pi == NULL?")); 1916 if (pi->pi_owner != curthread) { 1917 mtx_unlock_spin(&umtx_lock); 1918 umtxq_unbusy(&key); 1919 umtxq_unlock(&key); 1920 umtx_key_release(&key); 1921 /* userland messed the mutex */ 1922 return (EPERM); 1923 } 1924 uq_me = curthread->td_umtxq; 1925 pi->pi_owner = NULL; 1926 TAILQ_REMOVE(&uq_me->uq_pi_contested, pi, pi_link); 1927 /* get highest priority thread which is still sleeping. */ 1928 uq_first = TAILQ_FIRST(&pi->pi_blocked); 1929 while (uq_first != NULL && 1930 (uq_first->uq_flags & UQF_UMTXQ) == 0) { 1931 uq_first = TAILQ_NEXT(uq_first, uq_lockq); 1932 } 1933 pri = PRI_MAX; 1934 TAILQ_FOREACH(pi2, &uq_me->uq_pi_contested, pi_link) { 1935 uq_first2 = TAILQ_FIRST(&pi2->pi_blocked); 1936 if (uq_first2 != NULL) { 1937 if (pri > UPRI(uq_first2->uq_thread)) 1938 pri = UPRI(uq_first2->uq_thread); 1939 } 1940 } 1941 thread_lock(curthread); 1942 sched_unlend_user_prio(curthread, pri); 1943 thread_unlock(curthread); 1944 mtx_unlock_spin(&umtx_lock); 1945 if (uq_first) 1946 umtxq_signal_thread(uq_first); 1947 } 1948 umtxq_unlock(&key); 1949 1950 /* 1951 * When unlocking the umtx, it must be marked as unowned if 1952 * there is zero or one thread only waiting for it. 1953 * Otherwise, it must be marked as contested. 1954 */ 1955 old = casuword32(&m->m_owner, owner, 1956 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED); 1957 1958 umtxq_lock(&key); 1959 umtxq_unbusy(&key); 1960 umtxq_unlock(&key); 1961 umtx_key_release(&key); 1962 if (old == -1) 1963 return (EFAULT); 1964 if (old != owner) 1965 return (EINVAL); 1966 return (0); 1967 } 1968 1969 /* 1970 * Lock a PP mutex. 1971 */ 1972 static int 1973 _do_lock_pp(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1974 int try) 1975 { 1976 struct umtx_q *uq, *uq2; 1977 struct umtx_pi *pi; 1978 uint32_t ceiling; 1979 uint32_t owner, id; 1980 int error, pri, old_inherited_pri, su; 1981 1982 id = td->td_tid; 1983 uq = td->td_umtxq; 1984 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 1985 &uq->uq_key)) != 0) 1986 return (error); 1987 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0); 1988 for (;;) { 1989 old_inherited_pri = uq->uq_inherited_pri; 1990 umtxq_lock(&uq->uq_key); 1991 umtxq_busy(&uq->uq_key); 1992 umtxq_unlock(&uq->uq_key); 1993 1994 ceiling = RTP_PRIO_MAX - fuword32(&m->m_ceilings[0]); 1995 if (ceiling > RTP_PRIO_MAX) { 1996 error = EINVAL; 1997 goto out; 1998 } 1999 2000 mtx_lock_spin(&umtx_lock); 2001 if (UPRI(td) < PRI_MIN_REALTIME + ceiling) { 2002 mtx_unlock_spin(&umtx_lock); 2003 error = EINVAL; 2004 goto out; 2005 } 2006 if (su && PRI_MIN_REALTIME + ceiling < uq->uq_inherited_pri) { 2007 uq->uq_inherited_pri = PRI_MIN_REALTIME + ceiling; 2008 thread_lock(td); 2009 if (uq->uq_inherited_pri < UPRI(td)) 2010 sched_lend_user_prio(td, uq->uq_inherited_pri); 2011 thread_unlock(td); 2012 } 2013 mtx_unlock_spin(&umtx_lock); 2014 2015 owner = casuword32(&m->m_owner, 2016 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 2017 2018 if (owner == UMUTEX_CONTESTED) { 2019 error = 0; 2020 break; 2021 } 2022 2023 /* The address was invalid. */ 2024 if (owner == -1) { 2025 error = EFAULT; 2026 break; 2027 } 2028 2029 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 2030 (owner & ~UMUTEX_CONTESTED) == id) { 2031 error = EDEADLK; 2032 break; 2033 } 2034 2035 if (try != 0) { 2036 error = EBUSY; 2037 break; 2038 } 2039 2040 /* 2041 * If we caught a signal, we have retried and now 2042 * exit immediately. 2043 */ 2044 if (error != 0) 2045 break; 2046 2047 umtxq_lock(&uq->uq_key); 2048 umtxq_insert(uq); 2049 umtxq_unbusy(&uq->uq_key); 2050 error = umtxq_sleep(uq, "umtxpp", timo); 2051 umtxq_remove(uq); 2052 umtxq_unlock(&uq->uq_key); 2053 2054 mtx_lock_spin(&umtx_lock); 2055 uq->uq_inherited_pri = old_inherited_pri; 2056 pri = PRI_MAX; 2057 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2058 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2059 if (uq2 != NULL) { 2060 if (pri > UPRI(uq2->uq_thread)) 2061 pri = UPRI(uq2->uq_thread); 2062 } 2063 } 2064 if (pri > uq->uq_inherited_pri) 2065 pri = uq->uq_inherited_pri; 2066 thread_lock(td); 2067 sched_unlend_user_prio(td, pri); 2068 thread_unlock(td); 2069 mtx_unlock_spin(&umtx_lock); 2070 } 2071 2072 if (error != 0) { 2073 mtx_lock_spin(&umtx_lock); 2074 uq->uq_inherited_pri = old_inherited_pri; 2075 pri = PRI_MAX; 2076 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2077 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2078 if (uq2 != NULL) { 2079 if (pri > UPRI(uq2->uq_thread)) 2080 pri = UPRI(uq2->uq_thread); 2081 } 2082 } 2083 if (pri > uq->uq_inherited_pri) 2084 pri = uq->uq_inherited_pri; 2085 thread_lock(td); 2086 sched_unlend_user_prio(td, pri); 2087 thread_unlock(td); 2088 mtx_unlock_spin(&umtx_lock); 2089 } 2090 2091 out: 2092 umtxq_lock(&uq->uq_key); 2093 umtxq_unbusy(&uq->uq_key); 2094 umtxq_unlock(&uq->uq_key); 2095 umtx_key_release(&uq->uq_key); 2096 return (error); 2097 } 2098 2099 /* 2100 * Unlock a PP mutex. 2101 */ 2102 static int 2103 do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags) 2104 { 2105 struct umtx_key key; 2106 struct umtx_q *uq, *uq2; 2107 struct umtx_pi *pi; 2108 uint32_t owner, id; 2109 uint32_t rceiling; 2110 int error, pri, new_inherited_pri, su; 2111 2112 id = td->td_tid; 2113 uq = td->td_umtxq; 2114 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0); 2115 2116 /* 2117 * Make sure we own this mtx. 2118 */ 2119 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 2120 if (owner == -1) 2121 return (EFAULT); 2122 2123 if ((owner & ~UMUTEX_CONTESTED) != id) 2124 return (EPERM); 2125 2126 error = copyin(&m->m_ceilings[1], &rceiling, sizeof(uint32_t)); 2127 if (error != 0) 2128 return (error); 2129 2130 if (rceiling == -1) 2131 new_inherited_pri = PRI_MAX; 2132 else { 2133 rceiling = RTP_PRIO_MAX - rceiling; 2134 if (rceiling > RTP_PRIO_MAX) 2135 return (EINVAL); 2136 new_inherited_pri = PRI_MIN_REALTIME + rceiling; 2137 } 2138 2139 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 2140 &key)) != 0) 2141 return (error); 2142 umtxq_lock(&key); 2143 umtxq_busy(&key); 2144 umtxq_unlock(&key); 2145 /* 2146 * For priority protected mutex, always set unlocked state 2147 * to UMUTEX_CONTESTED, so that userland always enters kernel 2148 * to lock the mutex, it is necessary because thread priority 2149 * has to be adjusted for such mutex. 2150 */ 2151 error = suword32(__DEVOLATILE(uint32_t *, &m->m_owner), 2152 UMUTEX_CONTESTED); 2153 2154 umtxq_lock(&key); 2155 if (error == 0) 2156 umtxq_signal(&key, 1); 2157 umtxq_unbusy(&key); 2158 umtxq_unlock(&key); 2159 2160 if (error == -1) 2161 error = EFAULT; 2162 else { 2163 mtx_lock_spin(&umtx_lock); 2164 if (su != 0) 2165 uq->uq_inherited_pri = new_inherited_pri; 2166 pri = PRI_MAX; 2167 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2168 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2169 if (uq2 != NULL) { 2170 if (pri > UPRI(uq2->uq_thread)) 2171 pri = UPRI(uq2->uq_thread); 2172 } 2173 } 2174 if (pri > uq->uq_inherited_pri) 2175 pri = uq->uq_inherited_pri; 2176 thread_lock(td); 2177 sched_unlend_user_prio(td, pri); 2178 thread_unlock(td); 2179 mtx_unlock_spin(&umtx_lock); 2180 } 2181 umtx_key_release(&key); 2182 return (error); 2183 } 2184 2185 static int 2186 do_set_ceiling(struct thread *td, struct umutex *m, uint32_t ceiling, 2187 uint32_t *old_ceiling) 2188 { 2189 struct umtx_q *uq; 2190 uint32_t save_ceiling; 2191 uint32_t owner, id; 2192 uint32_t flags; 2193 int error; 2194 2195 flags = fuword32(&m->m_flags); 2196 if ((flags & UMUTEX_PRIO_PROTECT) == 0) 2197 return (EINVAL); 2198 if (ceiling > RTP_PRIO_MAX) 2199 return (EINVAL); 2200 id = td->td_tid; 2201 uq = td->td_umtxq; 2202 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 2203 &uq->uq_key)) != 0) 2204 return (error); 2205 for (;;) { 2206 umtxq_lock(&uq->uq_key); 2207 umtxq_busy(&uq->uq_key); 2208 umtxq_unlock(&uq->uq_key); 2209 2210 save_ceiling = fuword32(&m->m_ceilings[0]); 2211 2212 owner = casuword32(&m->m_owner, 2213 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 2214 2215 if (owner == UMUTEX_CONTESTED) { 2216 suword32(&m->m_ceilings[0], ceiling); 2217 suword32(__DEVOLATILE(uint32_t *, &m->m_owner), 2218 UMUTEX_CONTESTED); 2219 error = 0; 2220 break; 2221 } 2222 2223 /* The address was invalid. */ 2224 if (owner == -1) { 2225 error = EFAULT; 2226 break; 2227 } 2228 2229 if ((owner & ~UMUTEX_CONTESTED) == id) { 2230 suword32(&m->m_ceilings[0], ceiling); 2231 error = 0; 2232 break; 2233 } 2234 2235 /* 2236 * If we caught a signal, we have retried and now 2237 * exit immediately. 2238 */ 2239 if (error != 0) 2240 break; 2241 2242 /* 2243 * We set the contested bit, sleep. Otherwise the lock changed 2244 * and we need to retry or we lost a race to the thread 2245 * unlocking the umtx. 2246 */ 2247 umtxq_lock(&uq->uq_key); 2248 umtxq_insert(uq); 2249 umtxq_unbusy(&uq->uq_key); 2250 error = umtxq_sleep(uq, "umtxpp", 0); 2251 umtxq_remove(uq); 2252 umtxq_unlock(&uq->uq_key); 2253 } 2254 umtxq_lock(&uq->uq_key); 2255 if (error == 0) 2256 umtxq_signal(&uq->uq_key, INT_MAX); 2257 umtxq_unbusy(&uq->uq_key); 2258 umtxq_unlock(&uq->uq_key); 2259 umtx_key_release(&uq->uq_key); 2260 if (error == 0 && old_ceiling != NULL) 2261 suword32(old_ceiling, save_ceiling); 2262 return (error); 2263 } 2264 2265 static int 2266 _do_lock_umutex(struct thread *td, struct umutex *m, int flags, int timo, 2267 int mode) 2268 { 2269 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) { 2270 case 0: 2271 return (_do_lock_normal(td, m, flags, timo, mode)); 2272 case UMUTEX_PRIO_INHERIT: 2273 return (_do_lock_pi(td, m, flags, timo, mode)); 2274 case UMUTEX_PRIO_PROTECT: 2275 return (_do_lock_pp(td, m, flags, timo, mode)); 2276 } 2277 return (EINVAL); 2278 } 2279 2280 /* 2281 * Lock a userland POSIX mutex. 2282 */ 2283 static int 2284 do_lock_umutex(struct thread *td, struct umutex *m, 2285 struct timespec *timeout, int mode) 2286 { 2287 struct timespec ts, ts2, ts3; 2288 struct timeval tv; 2289 uint32_t flags; 2290 int error; 2291 2292 flags = fuword32(&m->m_flags); 2293 if (flags == -1) 2294 return (EFAULT); 2295 2296 if (timeout == NULL) { 2297 error = _do_lock_umutex(td, m, flags, 0, mode); 2298 /* Mutex locking is restarted if it is interrupted. */ 2299 if (error == EINTR && mode != _UMUTEX_WAIT) 2300 error = ERESTART; 2301 } else { 2302 getnanouptime(&ts); 2303 timespecadd(&ts, timeout); 2304 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2305 for (;;) { 2306 error = _do_lock_umutex(td, m, flags, tvtohz(&tv), mode); 2307 if (error != ETIMEDOUT) 2308 break; 2309 getnanouptime(&ts2); 2310 if (timespeccmp(&ts2, &ts, >=)) { 2311 error = ETIMEDOUT; 2312 break; 2313 } 2314 ts3 = ts; 2315 timespecsub(&ts3, &ts2); 2316 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2317 } 2318 /* Timed-locking is not restarted. */ 2319 if (error == ERESTART) 2320 error = EINTR; 2321 } 2322 return (error); 2323 } 2324 2325 /* 2326 * Unlock a userland POSIX mutex. 2327 */ 2328 static int 2329 do_unlock_umutex(struct thread *td, struct umutex *m) 2330 { 2331 uint32_t flags; 2332 2333 flags = fuword32(&m->m_flags); 2334 if (flags == -1) 2335 return (EFAULT); 2336 2337 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) { 2338 case 0: 2339 return (do_unlock_normal(td, m, flags)); 2340 case UMUTEX_PRIO_INHERIT: 2341 return (do_unlock_pi(td, m, flags)); 2342 case UMUTEX_PRIO_PROTECT: 2343 return (do_unlock_pp(td, m, flags)); 2344 } 2345 2346 return (EINVAL); 2347 } 2348 2349 static int 2350 do_cv_wait(struct thread *td, struct ucond *cv, struct umutex *m, 2351 struct timespec *timeout, u_long wflags) 2352 { 2353 struct umtx_q *uq; 2354 struct timeval tv; 2355 struct timespec cts, ets, tts; 2356 uint32_t flags; 2357 int error; 2358 2359 uq = td->td_umtxq; 2360 flags = fuword32(&cv->c_flags); 2361 error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &uq->uq_key); 2362 if (error != 0) 2363 return (error); 2364 umtxq_lock(&uq->uq_key); 2365 umtxq_busy(&uq->uq_key); 2366 umtxq_insert(uq); 2367 umtxq_unlock(&uq->uq_key); 2368 2369 /* 2370 * The magic thing is we should set c_has_waiters to 1 before 2371 * releasing user mutex. 2372 */ 2373 suword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 1); 2374 2375 umtxq_lock(&uq->uq_key); 2376 umtxq_unbusy(&uq->uq_key); 2377 umtxq_unlock(&uq->uq_key); 2378 2379 error = do_unlock_umutex(td, m); 2380 2381 umtxq_lock(&uq->uq_key); 2382 if (error == 0) { 2383 if ((wflags & UMTX_CHECK_UNPARKING) && 2384 (td->td_pflags & TDP_WAKEUP)) { 2385 td->td_pflags &= ~TDP_WAKEUP; 2386 error = EINTR; 2387 } else if (timeout == NULL) { 2388 error = umtxq_sleep(uq, "ucond", 0); 2389 } else { 2390 getnanouptime(&ets); 2391 timespecadd(&ets, timeout); 2392 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2393 for (;;) { 2394 error = umtxq_sleep(uq, "ucond", tvtohz(&tv)); 2395 if (error != ETIMEDOUT) 2396 break; 2397 getnanouptime(&cts); 2398 if (timespeccmp(&cts, &ets, >=)) { 2399 error = ETIMEDOUT; 2400 break; 2401 } 2402 tts = ets; 2403 timespecsub(&tts, &cts); 2404 TIMESPEC_TO_TIMEVAL(&tv, &tts); 2405 } 2406 } 2407 } 2408 2409 if ((uq->uq_flags & UQF_UMTXQ) == 0) 2410 error = 0; 2411 else { 2412 umtxq_remove(uq); 2413 if (error == ERESTART) 2414 error = EINTR; 2415 } 2416 2417 umtxq_unlock(&uq->uq_key); 2418 umtx_key_release(&uq->uq_key); 2419 return (error); 2420 } 2421 2422 /* 2423 * Signal a userland condition variable. 2424 */ 2425 static int 2426 do_cv_signal(struct thread *td, struct ucond *cv) 2427 { 2428 struct umtx_key key; 2429 int error, cnt, nwake; 2430 uint32_t flags; 2431 2432 flags = fuword32(&cv->c_flags); 2433 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0) 2434 return (error); 2435 umtxq_lock(&key); 2436 umtxq_busy(&key); 2437 cnt = umtxq_count(&key); 2438 nwake = umtxq_signal(&key, 1); 2439 if (cnt <= nwake) { 2440 umtxq_unlock(&key); 2441 error = suword32( 2442 __DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0); 2443 umtxq_lock(&key); 2444 } 2445 umtxq_unbusy(&key); 2446 umtxq_unlock(&key); 2447 umtx_key_release(&key); 2448 return (error); 2449 } 2450 2451 static int 2452 do_cv_broadcast(struct thread *td, struct ucond *cv) 2453 { 2454 struct umtx_key key; 2455 int error; 2456 uint32_t flags; 2457 2458 flags = fuword32(&cv->c_flags); 2459 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0) 2460 return (error); 2461 2462 umtxq_lock(&key); 2463 umtxq_busy(&key); 2464 umtxq_signal(&key, INT_MAX); 2465 umtxq_unlock(&key); 2466 2467 error = suword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0); 2468 2469 umtxq_lock(&key); 2470 umtxq_unbusy(&key); 2471 umtxq_unlock(&key); 2472 2473 umtx_key_release(&key); 2474 return (error); 2475 } 2476 2477 static int 2478 do_rw_rdlock(struct thread *td, struct urwlock *rwlock, long fflag, int timo) 2479 { 2480 struct umtx_q *uq; 2481 uint32_t flags, wrflags; 2482 int32_t state, oldstate; 2483 int32_t blocked_readers; 2484 int error; 2485 2486 uq = td->td_umtxq; 2487 flags = fuword32(&rwlock->rw_flags); 2488 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2489 if (error != 0) 2490 return (error); 2491 2492 wrflags = URWLOCK_WRITE_OWNER; 2493 if (!(fflag & URWLOCK_PREFER_READER) && !(flags & URWLOCK_PREFER_READER)) 2494 wrflags |= URWLOCK_WRITE_WAITERS; 2495 2496 for (;;) { 2497 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2498 /* try to lock it */ 2499 while (!(state & wrflags)) { 2500 if (__predict_false(URWLOCK_READER_COUNT(state) == URWLOCK_MAX_READERS)) { 2501 umtx_key_release(&uq->uq_key); 2502 return (EAGAIN); 2503 } 2504 oldstate = casuword32(&rwlock->rw_state, state, state + 1); 2505 if (oldstate == state) { 2506 umtx_key_release(&uq->uq_key); 2507 return (0); 2508 } 2509 state = oldstate; 2510 } 2511 2512 if (error) 2513 break; 2514 2515 /* grab monitor lock */ 2516 umtxq_lock(&uq->uq_key); 2517 umtxq_busy(&uq->uq_key); 2518 umtxq_unlock(&uq->uq_key); 2519 2520 /* 2521 * re-read the state, in case it changed between the try-lock above 2522 * and the check below 2523 */ 2524 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2525 2526 /* set read contention bit */ 2527 while ((state & wrflags) && !(state & URWLOCK_READ_WAITERS)) { 2528 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_READ_WAITERS); 2529 if (oldstate == state) 2530 goto sleep; 2531 state = oldstate; 2532 } 2533 2534 /* state is changed while setting flags, restart */ 2535 if (!(state & wrflags)) { 2536 umtxq_lock(&uq->uq_key); 2537 umtxq_unbusy(&uq->uq_key); 2538 umtxq_unlock(&uq->uq_key); 2539 continue; 2540 } 2541 2542 sleep: 2543 /* contention bit is set, before sleeping, increase read waiter count */ 2544 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2545 suword32(&rwlock->rw_blocked_readers, blocked_readers+1); 2546 2547 while (state & wrflags) { 2548 umtxq_lock(&uq->uq_key); 2549 umtxq_insert(uq); 2550 umtxq_unbusy(&uq->uq_key); 2551 2552 error = umtxq_sleep(uq, "urdlck", timo); 2553 2554 umtxq_busy(&uq->uq_key); 2555 umtxq_remove(uq); 2556 umtxq_unlock(&uq->uq_key); 2557 if (error) 2558 break; 2559 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2560 } 2561 2562 /* decrease read waiter count, and may clear read contention bit */ 2563 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2564 suword32(&rwlock->rw_blocked_readers, blocked_readers-1); 2565 if (blocked_readers == 1) { 2566 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2567 for (;;) { 2568 oldstate = casuword32(&rwlock->rw_state, state, 2569 state & ~URWLOCK_READ_WAITERS); 2570 if (oldstate == state) 2571 break; 2572 state = oldstate; 2573 } 2574 } 2575 2576 umtxq_lock(&uq->uq_key); 2577 umtxq_unbusy(&uq->uq_key); 2578 umtxq_unlock(&uq->uq_key); 2579 } 2580 umtx_key_release(&uq->uq_key); 2581 return (error); 2582 } 2583 2584 static int 2585 do_rw_rdlock2(struct thread *td, void *obj, long val, struct timespec *timeout) 2586 { 2587 struct timespec ts, ts2, ts3; 2588 struct timeval tv; 2589 int error; 2590 2591 getnanouptime(&ts); 2592 timespecadd(&ts, timeout); 2593 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2594 for (;;) { 2595 error = do_rw_rdlock(td, obj, val, tvtohz(&tv)); 2596 if (error != ETIMEDOUT) 2597 break; 2598 getnanouptime(&ts2); 2599 if (timespeccmp(&ts2, &ts, >=)) { 2600 error = ETIMEDOUT; 2601 break; 2602 } 2603 ts3 = ts; 2604 timespecsub(&ts3, &ts2); 2605 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2606 } 2607 if (error == ERESTART) 2608 error = EINTR; 2609 return (error); 2610 } 2611 2612 static int 2613 do_rw_wrlock(struct thread *td, struct urwlock *rwlock, int timo) 2614 { 2615 struct umtx_q *uq; 2616 uint32_t flags; 2617 int32_t state, oldstate; 2618 int32_t blocked_writers; 2619 int32_t blocked_readers; 2620 int error; 2621 2622 uq = td->td_umtxq; 2623 flags = fuword32(&rwlock->rw_flags); 2624 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2625 if (error != 0) 2626 return (error); 2627 2628 blocked_readers = 0; 2629 for (;;) { 2630 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2631 while (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) { 2632 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_OWNER); 2633 if (oldstate == state) { 2634 umtx_key_release(&uq->uq_key); 2635 return (0); 2636 } 2637 state = oldstate; 2638 } 2639 2640 if (error) { 2641 if (!(state & (URWLOCK_WRITE_OWNER|URWLOCK_WRITE_WAITERS)) && 2642 blocked_readers != 0) { 2643 umtxq_lock(&uq->uq_key); 2644 umtxq_busy(&uq->uq_key); 2645 umtxq_signal_queue(&uq->uq_key, INT_MAX, UMTX_SHARED_QUEUE); 2646 umtxq_unbusy(&uq->uq_key); 2647 umtxq_unlock(&uq->uq_key); 2648 } 2649 2650 break; 2651 } 2652 2653 /* grab monitor lock */ 2654 umtxq_lock(&uq->uq_key); 2655 umtxq_busy(&uq->uq_key); 2656 umtxq_unlock(&uq->uq_key); 2657 2658 /* 2659 * re-read the state, in case it changed between the try-lock above 2660 * and the check below 2661 */ 2662 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2663 2664 while (((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) && 2665 (state & URWLOCK_WRITE_WAITERS) == 0) { 2666 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_WAITERS); 2667 if (oldstate == state) 2668 goto sleep; 2669 state = oldstate; 2670 } 2671 2672 if (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) { 2673 umtxq_lock(&uq->uq_key); 2674 umtxq_unbusy(&uq->uq_key); 2675 umtxq_unlock(&uq->uq_key); 2676 continue; 2677 } 2678 sleep: 2679 blocked_writers = fuword32(&rwlock->rw_blocked_writers); 2680 suword32(&rwlock->rw_blocked_writers, blocked_writers+1); 2681 2682 while ((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) { 2683 umtxq_lock(&uq->uq_key); 2684 umtxq_insert_queue(uq, UMTX_EXCLUSIVE_QUEUE); 2685 umtxq_unbusy(&uq->uq_key); 2686 2687 error = umtxq_sleep(uq, "uwrlck", timo); 2688 2689 umtxq_busy(&uq->uq_key); 2690 umtxq_remove_queue(uq, UMTX_EXCLUSIVE_QUEUE); 2691 umtxq_unlock(&uq->uq_key); 2692 if (error) 2693 break; 2694 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2695 } 2696 2697 blocked_writers = fuword32(&rwlock->rw_blocked_writers); 2698 suword32(&rwlock->rw_blocked_writers, blocked_writers-1); 2699 if (blocked_writers == 1) { 2700 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2701 for (;;) { 2702 oldstate = casuword32(&rwlock->rw_state, state, 2703 state & ~URWLOCK_WRITE_WAITERS); 2704 if (oldstate == state) 2705 break; 2706 state = oldstate; 2707 } 2708 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2709 } else 2710 blocked_readers = 0; 2711 2712 umtxq_lock(&uq->uq_key); 2713 umtxq_unbusy(&uq->uq_key); 2714 umtxq_unlock(&uq->uq_key); 2715 } 2716 2717 umtx_key_release(&uq->uq_key); 2718 return (error); 2719 } 2720 2721 static int 2722 do_rw_wrlock2(struct thread *td, void *obj, struct timespec *timeout) 2723 { 2724 struct timespec ts, ts2, ts3; 2725 struct timeval tv; 2726 int error; 2727 2728 getnanouptime(&ts); 2729 timespecadd(&ts, timeout); 2730 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2731 for (;;) { 2732 error = do_rw_wrlock(td, obj, tvtohz(&tv)); 2733 if (error != ETIMEDOUT) 2734 break; 2735 getnanouptime(&ts2); 2736 if (timespeccmp(&ts2, &ts, >=)) { 2737 error = ETIMEDOUT; 2738 break; 2739 } 2740 ts3 = ts; 2741 timespecsub(&ts3, &ts2); 2742 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2743 } 2744 if (error == ERESTART) 2745 error = EINTR; 2746 return (error); 2747 } 2748 2749 static int 2750 do_rw_unlock(struct thread *td, struct urwlock *rwlock) 2751 { 2752 struct umtx_q *uq; 2753 uint32_t flags; 2754 int32_t state, oldstate; 2755 int error, q, count; 2756 2757 uq = td->td_umtxq; 2758 flags = fuword32(&rwlock->rw_flags); 2759 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2760 if (error != 0) 2761 return (error); 2762 2763 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2764 if (state & URWLOCK_WRITE_OWNER) { 2765 for (;;) { 2766 oldstate = casuword32(&rwlock->rw_state, state, 2767 state & ~URWLOCK_WRITE_OWNER); 2768 if (oldstate != state) { 2769 state = oldstate; 2770 if (!(oldstate & URWLOCK_WRITE_OWNER)) { 2771 error = EPERM; 2772 goto out; 2773 } 2774 } else 2775 break; 2776 } 2777 } else if (URWLOCK_READER_COUNT(state) != 0) { 2778 for (;;) { 2779 oldstate = casuword32(&rwlock->rw_state, state, 2780 state - 1); 2781 if (oldstate != state) { 2782 state = oldstate; 2783 if (URWLOCK_READER_COUNT(oldstate) == 0) { 2784 error = EPERM; 2785 goto out; 2786 } 2787 } 2788 else 2789 break; 2790 } 2791 } else { 2792 error = EPERM; 2793 goto out; 2794 } 2795 2796 count = 0; 2797 2798 if (!(flags & URWLOCK_PREFER_READER)) { 2799 if (state & URWLOCK_WRITE_WAITERS) { 2800 count = 1; 2801 q = UMTX_EXCLUSIVE_QUEUE; 2802 } else if (state & URWLOCK_READ_WAITERS) { 2803 count = INT_MAX; 2804 q = UMTX_SHARED_QUEUE; 2805 } 2806 } else { 2807 if (state & URWLOCK_READ_WAITERS) { 2808 count = INT_MAX; 2809 q = UMTX_SHARED_QUEUE; 2810 } else if (state & URWLOCK_WRITE_WAITERS) { 2811 count = 1; 2812 q = UMTX_EXCLUSIVE_QUEUE; 2813 } 2814 } 2815 2816 if (count) { 2817 umtxq_lock(&uq->uq_key); 2818 umtxq_busy(&uq->uq_key); 2819 umtxq_signal_queue(&uq->uq_key, count, q); 2820 umtxq_unbusy(&uq->uq_key); 2821 umtxq_unlock(&uq->uq_key); 2822 } 2823 out: 2824 umtx_key_release(&uq->uq_key); 2825 return (error); 2826 } 2827 2828 static int 2829 do_sem_wait(struct thread *td, struct _usem *sem, struct timespec *timeout) 2830 { 2831 struct umtx_q *uq; 2832 struct timeval tv; 2833 struct timespec cts, ets, tts; 2834 uint32_t flags, count; 2835 int error; 2836 2837 uq = td->td_umtxq; 2838 flags = fuword32(&sem->_flags); 2839 error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key); 2840 if (error != 0) 2841 return (error); 2842 umtxq_lock(&uq->uq_key); 2843 umtxq_busy(&uq->uq_key); 2844 umtxq_insert(uq); 2845 umtxq_unlock(&uq->uq_key); 2846 2847 suword32(__DEVOLATILE(uint32_t *, &sem->_has_waiters), 1); 2848 2849 count = fuword32(__DEVOLATILE(uint32_t *, &sem->_count)); 2850 if (count != 0) { 2851 umtxq_lock(&uq->uq_key); 2852 umtxq_unbusy(&uq->uq_key); 2853 umtxq_remove(uq); 2854 umtxq_unlock(&uq->uq_key); 2855 umtx_key_release(&uq->uq_key); 2856 return (0); 2857 } 2858 2859 umtxq_lock(&uq->uq_key); 2860 umtxq_unbusy(&uq->uq_key); 2861 umtxq_unlock(&uq->uq_key); 2862 2863 umtxq_lock(&uq->uq_key); 2864 if (timeout == NULL) { 2865 error = umtxq_sleep(uq, "usem", 0); 2866 } else { 2867 getnanouptime(&ets); 2868 timespecadd(&ets, timeout); 2869 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2870 for (;;) { 2871 error = umtxq_sleep(uq, "usem", tvtohz(&tv)); 2872 if (error != ETIMEDOUT) 2873 break; 2874 getnanouptime(&cts); 2875 if (timespeccmp(&cts, &ets, >=)) { 2876 error = ETIMEDOUT; 2877 break; 2878 } 2879 tts = ets; 2880 timespecsub(&tts, &cts); 2881 TIMESPEC_TO_TIMEVAL(&tv, &tts); 2882 } 2883 } 2884 2885 if ((uq->uq_flags & UQF_UMTXQ) == 0) 2886 error = 0; 2887 else { 2888 umtxq_remove(uq); 2889 if (error == ERESTART) 2890 error = EINTR; 2891 } 2892 umtxq_unlock(&uq->uq_key); 2893 umtx_key_release(&uq->uq_key); 2894 return (error); 2895 } 2896 2897 /* 2898 * Signal a userland condition variable. 2899 */ 2900 static int 2901 do_sem_wake(struct thread *td, struct _usem *sem) 2902 { 2903 struct umtx_key key; 2904 int error, cnt, nwake; 2905 uint32_t flags; 2906 2907 flags = fuword32(&sem->_flags); 2908 if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0) 2909 return (error); 2910 umtxq_lock(&key); 2911 umtxq_busy(&key); 2912 cnt = umtxq_count(&key); 2913 nwake = umtxq_signal(&key, 1); 2914 if (cnt <= nwake) { 2915 umtxq_unlock(&key); 2916 error = suword32( 2917 __DEVOLATILE(uint32_t *, &sem->_has_waiters), 0); 2918 umtxq_lock(&key); 2919 } 2920 umtxq_unbusy(&key); 2921 umtxq_unlock(&key); 2922 umtx_key_release(&key); 2923 return (error); 2924 } 2925 2926 int 2927 _umtx_lock(struct thread *td, struct _umtx_lock_args *uap) 2928 /* struct umtx *umtx */ 2929 { 2930 return _do_lock_umtx(td, uap->umtx, td->td_tid, 0); 2931 } 2932 2933 int 2934 _umtx_unlock(struct thread *td, struct _umtx_unlock_args *uap) 2935 /* struct umtx *umtx */ 2936 { 2937 return do_unlock_umtx(td, uap->umtx, td->td_tid); 2938 } 2939 2940 static int 2941 __umtx_op_lock_umtx(struct thread *td, struct _umtx_op_args *uap) 2942 { 2943 struct timespec *ts, timeout; 2944 int error; 2945 2946 /* Allow a null timespec (wait forever). */ 2947 if (uap->uaddr2 == NULL) 2948 ts = NULL; 2949 else { 2950 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 2951 if (error != 0) 2952 return (error); 2953 if (timeout.tv_nsec >= 1000000000 || 2954 timeout.tv_nsec < 0) { 2955 return (EINVAL); 2956 } 2957 ts = &timeout; 2958 } 2959 return (do_lock_umtx(td, uap->obj, uap->val, ts)); 2960 } 2961 2962 static int 2963 __umtx_op_unlock_umtx(struct thread *td, struct _umtx_op_args *uap) 2964 { 2965 return (do_unlock_umtx(td, uap->obj, uap->val)); 2966 } 2967 2968 static int 2969 __umtx_op_wait(struct thread *td, struct _umtx_op_args *uap) 2970 { 2971 struct timespec *ts, timeout; 2972 int error; 2973 2974 if (uap->uaddr2 == NULL) 2975 ts = NULL; 2976 else { 2977 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 2978 if (error != 0) 2979 return (error); 2980 if (timeout.tv_nsec >= 1000000000 || 2981 timeout.tv_nsec < 0) 2982 return (EINVAL); 2983 ts = &timeout; 2984 } 2985 return do_wait(td, uap->obj, uap->val, ts, 0, 0); 2986 } 2987 2988 static int 2989 __umtx_op_wait_uint(struct thread *td, struct _umtx_op_args *uap) 2990 { 2991 struct timespec *ts, timeout; 2992 int error; 2993 2994 if (uap->uaddr2 == NULL) 2995 ts = NULL; 2996 else { 2997 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 2998 if (error != 0) 2999 return (error); 3000 if (timeout.tv_nsec >= 1000000000 || 3001 timeout.tv_nsec < 0) 3002 return (EINVAL); 3003 ts = &timeout; 3004 } 3005 return do_wait(td, uap->obj, uap->val, ts, 1, 0); 3006 } 3007 3008 static int 3009 __umtx_op_wait_uint_private(struct thread *td, struct _umtx_op_args *uap) 3010 { 3011 struct timespec *ts, timeout; 3012 int error; 3013 3014 if (uap->uaddr2 == NULL) 3015 ts = NULL; 3016 else { 3017 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 3018 if (error != 0) 3019 return (error); 3020 if (timeout.tv_nsec >= 1000000000 || 3021 timeout.tv_nsec < 0) 3022 return (EINVAL); 3023 ts = &timeout; 3024 } 3025 return do_wait(td, uap->obj, uap->val, ts, 1, 1); 3026 } 3027 3028 static int 3029 __umtx_op_wake(struct thread *td, struct _umtx_op_args *uap) 3030 { 3031 return (kern_umtx_wake(td, uap->obj, uap->val, 0)); 3032 } 3033 3034 static int 3035 __umtx_op_wake_private(struct thread *td, struct _umtx_op_args *uap) 3036 { 3037 return (kern_umtx_wake(td, uap->obj, uap->val, 1)); 3038 } 3039 3040 static int 3041 __umtx_op_lock_umutex(struct thread *td, struct _umtx_op_args *uap) 3042 { 3043 struct timespec *ts, timeout; 3044 int error; 3045 3046 /* Allow a null timespec (wait forever). */ 3047 if (uap->uaddr2 == NULL) 3048 ts = NULL; 3049 else { 3050 error = copyin(uap->uaddr2, &timeout, 3051 sizeof(timeout)); 3052 if (error != 0) 3053 return (error); 3054 if (timeout.tv_nsec >= 1000000000 || 3055 timeout.tv_nsec < 0) { 3056 return (EINVAL); 3057 } 3058 ts = &timeout; 3059 } 3060 return do_lock_umutex(td, uap->obj, ts, 0); 3061 } 3062 3063 static int 3064 __umtx_op_trylock_umutex(struct thread *td, struct _umtx_op_args *uap) 3065 { 3066 return do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY); 3067 } 3068 3069 static int 3070 __umtx_op_wait_umutex(struct thread *td, struct _umtx_op_args *uap) 3071 { 3072 struct timespec *ts, timeout; 3073 int error; 3074 3075 /* Allow a null timespec (wait forever). */ 3076 if (uap->uaddr2 == NULL) 3077 ts = NULL; 3078 else { 3079 error = copyin(uap->uaddr2, &timeout, 3080 sizeof(timeout)); 3081 if (error != 0) 3082 return (error); 3083 if (timeout.tv_nsec >= 1000000000 || 3084 timeout.tv_nsec < 0) { 3085 return (EINVAL); 3086 } 3087 ts = &timeout; 3088 } 3089 return do_lock_umutex(td, uap->obj, ts, _UMUTEX_WAIT); 3090 } 3091 3092 static int 3093 __umtx_op_wake_umutex(struct thread *td, struct _umtx_op_args *uap) 3094 { 3095 return do_wake_umutex(td, uap->obj); 3096 } 3097 3098 static int 3099 __umtx_op_unlock_umutex(struct thread *td, struct _umtx_op_args *uap) 3100 { 3101 return do_unlock_umutex(td, uap->obj); 3102 } 3103 3104 static int 3105 __umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap) 3106 { 3107 return do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1); 3108 } 3109 3110 static int 3111 __umtx_op_cv_wait(struct thread *td, struct _umtx_op_args *uap) 3112 { 3113 struct timespec *ts, timeout; 3114 int error; 3115 3116 /* Allow a null timespec (wait forever). */ 3117 if (uap->uaddr2 == NULL) 3118 ts = NULL; 3119 else { 3120 error = copyin(uap->uaddr2, &timeout, 3121 sizeof(timeout)); 3122 if (error != 0) 3123 return (error); 3124 if (timeout.tv_nsec >= 1000000000 || 3125 timeout.tv_nsec < 0) { 3126 return (EINVAL); 3127 } 3128 ts = &timeout; 3129 } 3130 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val)); 3131 } 3132 3133 static int 3134 __umtx_op_cv_signal(struct thread *td, struct _umtx_op_args *uap) 3135 { 3136 return do_cv_signal(td, uap->obj); 3137 } 3138 3139 static int 3140 __umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap) 3141 { 3142 return do_cv_broadcast(td, uap->obj); 3143 } 3144 3145 static int 3146 __umtx_op_rw_rdlock(struct thread *td, struct _umtx_op_args *uap) 3147 { 3148 struct timespec timeout; 3149 int error; 3150 3151 /* Allow a null timespec (wait forever). */ 3152 if (uap->uaddr2 == NULL) { 3153 error = do_rw_rdlock(td, uap->obj, uap->val, 0); 3154 } else { 3155 error = copyin(uap->uaddr2, &timeout, 3156 sizeof(timeout)); 3157 if (error != 0) 3158 return (error); 3159 if (timeout.tv_nsec >= 1000000000 || 3160 timeout.tv_nsec < 0) { 3161 return (EINVAL); 3162 } 3163 error = do_rw_rdlock2(td, uap->obj, uap->val, &timeout); 3164 } 3165 return (error); 3166 } 3167 3168 static int 3169 __umtx_op_rw_wrlock(struct thread *td, struct _umtx_op_args *uap) 3170 { 3171 struct timespec timeout; 3172 int error; 3173 3174 /* Allow a null timespec (wait forever). */ 3175 if (uap->uaddr2 == NULL) { 3176 error = do_rw_wrlock(td, uap->obj, 0); 3177 } else { 3178 error = copyin(uap->uaddr2, &timeout, 3179 sizeof(timeout)); 3180 if (error != 0) 3181 return (error); 3182 if (timeout.tv_nsec >= 1000000000 || 3183 timeout.tv_nsec < 0) { 3184 return (EINVAL); 3185 } 3186 3187 error = do_rw_wrlock2(td, uap->obj, &timeout); 3188 } 3189 return (error); 3190 } 3191 3192 static int 3193 __umtx_op_rw_unlock(struct thread *td, struct _umtx_op_args *uap) 3194 { 3195 return do_rw_unlock(td, uap->obj); 3196 } 3197 3198 static int 3199 __umtx_op_sem_wait(struct thread *td, struct _umtx_op_args *uap) 3200 { 3201 struct timespec *ts, timeout; 3202 int error; 3203 3204 /* Allow a null timespec (wait forever). */ 3205 if (uap->uaddr2 == NULL) 3206 ts = NULL; 3207 else { 3208 error = copyin(uap->uaddr2, &timeout, 3209 sizeof(timeout)); 3210 if (error != 0) 3211 return (error); 3212 if (timeout.tv_nsec >= 1000000000 || 3213 timeout.tv_nsec < 0) { 3214 return (EINVAL); 3215 } 3216 ts = &timeout; 3217 } 3218 return (do_sem_wait(td, uap->obj, ts)); 3219 } 3220 3221 static int 3222 __umtx_op_sem_wake(struct thread *td, struct _umtx_op_args *uap) 3223 { 3224 return do_sem_wake(td, uap->obj); 3225 } 3226 3227 typedef int (*_umtx_op_func)(struct thread *td, struct _umtx_op_args *uap); 3228 3229 static _umtx_op_func op_table[] = { 3230 __umtx_op_lock_umtx, /* UMTX_OP_LOCK */ 3231 __umtx_op_unlock_umtx, /* UMTX_OP_UNLOCK */ 3232 __umtx_op_wait, /* UMTX_OP_WAIT */ 3233 __umtx_op_wake, /* UMTX_OP_WAKE */ 3234 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_TRYLOCK */ 3235 __umtx_op_lock_umutex, /* UMTX_OP_MUTEX_LOCK */ 3236 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */ 3237 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */ 3238 __umtx_op_cv_wait, /* UMTX_OP_CV_WAIT*/ 3239 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */ 3240 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */ 3241 __umtx_op_wait_uint, /* UMTX_OP_WAIT_UINT */ 3242 __umtx_op_rw_rdlock, /* UMTX_OP_RW_RDLOCK */ 3243 __umtx_op_rw_wrlock, /* UMTX_OP_RW_WRLOCK */ 3244 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */ 3245 __umtx_op_wait_uint_private, /* UMTX_OP_WAIT_UINT_PRIVATE */ 3246 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */ 3247 __umtx_op_wait_umutex, /* UMTX_OP_UMUTEX_WAIT */ 3248 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */ 3249 __umtx_op_sem_wait, /* UMTX_OP_SEM_WAIT */ 3250 __umtx_op_sem_wake /* UMTX_OP_SEM_WAKE */ 3251 }; 3252 3253 int 3254 _umtx_op(struct thread *td, struct _umtx_op_args *uap) 3255 { 3256 if ((unsigned)uap->op < UMTX_OP_MAX) 3257 return (*op_table[uap->op])(td, uap); 3258 return (EINVAL); 3259 } 3260 3261 #ifdef COMPAT_FREEBSD32 3262 int 3263 freebsd32_umtx_lock(struct thread *td, struct freebsd32_umtx_lock_args *uap) 3264 /* struct umtx *umtx */ 3265 { 3266 return (do_lock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid, NULL)); 3267 } 3268 3269 int 3270 freebsd32_umtx_unlock(struct thread *td, struct freebsd32_umtx_unlock_args *uap) 3271 /* struct umtx *umtx */ 3272 { 3273 return (do_unlock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid)); 3274 } 3275 3276 struct timespec32 { 3277 uint32_t tv_sec; 3278 uint32_t tv_nsec; 3279 }; 3280 3281 static inline int 3282 copyin_timeout32(void *addr, struct timespec *tsp) 3283 { 3284 struct timespec32 ts32; 3285 int error; 3286 3287 error = copyin(addr, &ts32, sizeof(struct timespec32)); 3288 if (error == 0) { 3289 tsp->tv_sec = ts32.tv_sec; 3290 tsp->tv_nsec = ts32.tv_nsec; 3291 } 3292 return (error); 3293 } 3294 3295 static int 3296 __umtx_op_lock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap) 3297 { 3298 struct timespec *ts, timeout; 3299 int error; 3300 3301 /* Allow a null timespec (wait forever). */ 3302 if (uap->uaddr2 == NULL) 3303 ts = NULL; 3304 else { 3305 error = copyin_timeout32(uap->uaddr2, &timeout); 3306 if (error != 0) 3307 return (error); 3308 if (timeout.tv_nsec >= 1000000000 || 3309 timeout.tv_nsec < 0) { 3310 return (EINVAL); 3311 } 3312 ts = &timeout; 3313 } 3314 return (do_lock_umtx32(td, uap->obj, uap->val, ts)); 3315 } 3316 3317 static int 3318 __umtx_op_unlock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap) 3319 { 3320 return (do_unlock_umtx32(td, uap->obj, (uint32_t)uap->val)); 3321 } 3322 3323 static int 3324 __umtx_op_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3325 { 3326 struct timespec *ts, timeout; 3327 int error; 3328 3329 if (uap->uaddr2 == NULL) 3330 ts = NULL; 3331 else { 3332 error = copyin_timeout32(uap->uaddr2, &timeout); 3333 if (error != 0) 3334 return (error); 3335 if (timeout.tv_nsec >= 1000000000 || 3336 timeout.tv_nsec < 0) 3337 return (EINVAL); 3338 ts = &timeout; 3339 } 3340 return do_wait(td, uap->obj, uap->val, ts, 1, 0); 3341 } 3342 3343 static int 3344 __umtx_op_lock_umutex_compat32(struct thread *td, struct _umtx_op_args *uap) 3345 { 3346 struct timespec *ts, timeout; 3347 int error; 3348 3349 /* Allow a null timespec (wait forever). */ 3350 if (uap->uaddr2 == NULL) 3351 ts = NULL; 3352 else { 3353 error = copyin_timeout32(uap->uaddr2, &timeout); 3354 if (error != 0) 3355 return (error); 3356 if (timeout.tv_nsec >= 1000000000 || 3357 timeout.tv_nsec < 0) 3358 return (EINVAL); 3359 ts = &timeout; 3360 } 3361 return do_lock_umutex(td, uap->obj, ts, 0); 3362 } 3363 3364 static int 3365 __umtx_op_wait_umutex_compat32(struct thread *td, struct _umtx_op_args *uap) 3366 { 3367 struct timespec *ts, timeout; 3368 int error; 3369 3370 /* Allow a null timespec (wait forever). */ 3371 if (uap->uaddr2 == NULL) 3372 ts = NULL; 3373 else { 3374 error = copyin_timeout32(uap->uaddr2, &timeout); 3375 if (error != 0) 3376 return (error); 3377 if (timeout.tv_nsec >= 1000000000 || 3378 timeout.tv_nsec < 0) 3379 return (EINVAL); 3380 ts = &timeout; 3381 } 3382 return do_lock_umutex(td, uap->obj, ts, _UMUTEX_WAIT); 3383 } 3384 3385 static int 3386 __umtx_op_cv_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3387 { 3388 struct timespec *ts, timeout; 3389 int error; 3390 3391 /* Allow a null timespec (wait forever). */ 3392 if (uap->uaddr2 == NULL) 3393 ts = NULL; 3394 else { 3395 error = copyin_timeout32(uap->uaddr2, &timeout); 3396 if (error != 0) 3397 return (error); 3398 if (timeout.tv_nsec >= 1000000000 || 3399 timeout.tv_nsec < 0) 3400 return (EINVAL); 3401 ts = &timeout; 3402 } 3403 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val)); 3404 } 3405 3406 static int 3407 __umtx_op_rw_rdlock_compat32(struct thread *td, struct _umtx_op_args *uap) 3408 { 3409 struct timespec timeout; 3410 int error; 3411 3412 /* Allow a null timespec (wait forever). */ 3413 if (uap->uaddr2 == NULL) { 3414 error = do_rw_rdlock(td, uap->obj, uap->val, 0); 3415 } else { 3416 error = copyin(uap->uaddr2, &timeout, 3417 sizeof(timeout)); 3418 if (error != 0) 3419 return (error); 3420 if (timeout.tv_nsec >= 1000000000 || 3421 timeout.tv_nsec < 0) { 3422 return (EINVAL); 3423 } 3424 error = do_rw_rdlock2(td, uap->obj, uap->val, &timeout); 3425 } 3426 return (error); 3427 } 3428 3429 static int 3430 __umtx_op_rw_wrlock_compat32(struct thread *td, struct _umtx_op_args *uap) 3431 { 3432 struct timespec timeout; 3433 int error; 3434 3435 /* Allow a null timespec (wait forever). */ 3436 if (uap->uaddr2 == NULL) { 3437 error = do_rw_wrlock(td, uap->obj, 0); 3438 } else { 3439 error = copyin_timeout32(uap->uaddr2, &timeout); 3440 if (error != 0) 3441 return (error); 3442 if (timeout.tv_nsec >= 1000000000 || 3443 timeout.tv_nsec < 0) { 3444 return (EINVAL); 3445 } 3446 3447 error = do_rw_wrlock2(td, uap->obj, &timeout); 3448 } 3449 return (error); 3450 } 3451 3452 static int 3453 __umtx_op_wait_uint_private_compat32(struct thread *td, struct _umtx_op_args *uap) 3454 { 3455 struct timespec *ts, timeout; 3456 int error; 3457 3458 if (uap->uaddr2 == NULL) 3459 ts = NULL; 3460 else { 3461 error = copyin_timeout32(uap->uaddr2, &timeout); 3462 if (error != 0) 3463 return (error); 3464 if (timeout.tv_nsec >= 1000000000 || 3465 timeout.tv_nsec < 0) 3466 return (EINVAL); 3467 ts = &timeout; 3468 } 3469 return do_wait(td, uap->obj, uap->val, ts, 1, 1); 3470 } 3471 3472 static int 3473 __umtx_op_sem_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3474 { 3475 struct timespec *ts, timeout; 3476 int error; 3477 3478 /* Allow a null timespec (wait forever). */ 3479 if (uap->uaddr2 == NULL) 3480 ts = NULL; 3481 else { 3482 error = copyin_timeout32(uap->uaddr2, &timeout); 3483 if (error != 0) 3484 return (error); 3485 if (timeout.tv_nsec >= 1000000000 || 3486 timeout.tv_nsec < 0) 3487 return (EINVAL); 3488 ts = &timeout; 3489 } 3490 return (do_sem_wait(td, uap->obj, ts)); 3491 } 3492 3493 static _umtx_op_func op_table_compat32[] = { 3494 __umtx_op_lock_umtx_compat32, /* UMTX_OP_LOCK */ 3495 __umtx_op_unlock_umtx_compat32, /* UMTX_OP_UNLOCK */ 3496 __umtx_op_wait_compat32, /* UMTX_OP_WAIT */ 3497 __umtx_op_wake, /* UMTX_OP_WAKE */ 3498 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_LOCK */ 3499 __umtx_op_lock_umutex_compat32, /* UMTX_OP_MUTEX_TRYLOCK */ 3500 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */ 3501 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */ 3502 __umtx_op_cv_wait_compat32, /* UMTX_OP_CV_WAIT*/ 3503 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */ 3504 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */ 3505 __umtx_op_wait_compat32, /* UMTX_OP_WAIT_UINT */ 3506 __umtx_op_rw_rdlock_compat32, /* UMTX_OP_RW_RDLOCK */ 3507 __umtx_op_rw_wrlock_compat32, /* UMTX_OP_RW_WRLOCK */ 3508 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */ 3509 __umtx_op_wait_uint_private_compat32, /* UMTX_OP_WAIT_UINT_PRIVATE */ 3510 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */ 3511 __umtx_op_wait_umutex_compat32, /* UMTX_OP_UMUTEX_WAIT */ 3512 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */ 3513 __umtx_op_sem_wait_compat32, /* UMTX_OP_SEM_WAIT */ 3514 __umtx_op_sem_wake /* UMTX_OP_SEM_WAKE */ 3515 }; 3516 3517 int 3518 freebsd32_umtx_op(struct thread *td, struct freebsd32_umtx_op_args *uap) 3519 { 3520 if ((unsigned)uap->op < UMTX_OP_MAX) 3521 return (*op_table_compat32[uap->op])(td, 3522 (struct _umtx_op_args *)uap); 3523 return (EINVAL); 3524 } 3525 #endif 3526 3527 void 3528 umtx_thread_init(struct thread *td) 3529 { 3530 td->td_umtxq = umtxq_alloc(); 3531 td->td_umtxq->uq_thread = td; 3532 } 3533 3534 void 3535 umtx_thread_fini(struct thread *td) 3536 { 3537 umtxq_free(td->td_umtxq); 3538 } 3539 3540 /* 3541 * It will be called when new thread is created, e.g fork(). 3542 */ 3543 void 3544 umtx_thread_alloc(struct thread *td) 3545 { 3546 struct umtx_q *uq; 3547 3548 uq = td->td_umtxq; 3549 uq->uq_inherited_pri = PRI_MAX; 3550 3551 KASSERT(uq->uq_flags == 0, ("uq_flags != 0")); 3552 KASSERT(uq->uq_thread == td, ("uq_thread != td")); 3553 KASSERT(uq->uq_pi_blocked == NULL, ("uq_pi_blocked != NULL")); 3554 KASSERT(TAILQ_EMPTY(&uq->uq_pi_contested), ("uq_pi_contested is not empty")); 3555 } 3556 3557 /* 3558 * exec() hook. 3559 */ 3560 static void 3561 umtx_exec_hook(void *arg __unused, struct proc *p __unused, 3562 struct image_params *imgp __unused) 3563 { 3564 umtx_thread_cleanup(curthread); 3565 } 3566 3567 /* 3568 * thread_exit() hook. 3569 */ 3570 void 3571 umtx_thread_exit(struct thread *td) 3572 { 3573 umtx_thread_cleanup(td); 3574 } 3575 3576 /* 3577 * clean up umtx data. 3578 */ 3579 static void 3580 umtx_thread_cleanup(struct thread *td) 3581 { 3582 struct umtx_q *uq; 3583 struct umtx_pi *pi; 3584 3585 if ((uq = td->td_umtxq) == NULL) 3586 return; 3587 3588 mtx_lock_spin(&umtx_lock); 3589 uq->uq_inherited_pri = PRI_MAX; 3590 while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) { 3591 pi->pi_owner = NULL; 3592 TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link); 3593 } 3594 thread_lock(td); 3595 td->td_flags &= ~TDF_UBORROWING; 3596 thread_unlock(td); 3597 mtx_unlock_spin(&umtx_lock); 3598 } 3599