kern_umtx.c revision 201991
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: head/sys/kern/kern_umtx.c 201991 2010-01-10 09:31:57Z davidxu $"); 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_IA32 58#include <compat/freebsd32/freebsd32_proto.h> 59#endif 60 61enum { 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 */ 76struct 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. */ 97struct 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. */ 118struct 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 155TAILQ_HEAD(umtxq_head, umtx_q); 156 157/* Per-key wait-queue */ 158struct umtxq_queue { 159 struct umtxq_head head; 160 struct umtx_key key; 161 LIST_ENTRY(umtxq_queue) link; 162 int length; 163}; 164 165LIST_HEAD(umtxq_list, umtxq_queue); 166 167/* Userland lock object's wait-queue chain */ 168struct 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 219static uma_zone_t umtx_pi_zone; 220static struct umtxq_chain umtxq_chains[2][UMTX_CHAINS]; 221static MALLOC_DEFINE(M_UMTX, "umtx", "UMTX queue memory"); 222static int umtx_pi_allocated; 223 224SYSCTL_NODE(_debug, OID_AUTO, umtx, CTLFLAG_RW, 0, "umtx debug"); 225SYSCTL_INT(_debug_umtx, OID_AUTO, umtx_pi_allocated, CTLFLAG_RD, 226 &umtx_pi_allocated, 0, "Allocated umtx_pi"); 227 228static void umtxq_sysinit(void *); 229static void umtxq_hash(struct umtx_key *key); 230static struct umtxq_chain *umtxq_getchain(struct umtx_key *key); 231static void umtxq_lock(struct umtx_key *key); 232static void umtxq_unlock(struct umtx_key *key); 233static void umtxq_busy(struct umtx_key *key); 234static void umtxq_unbusy(struct umtx_key *key); 235static void umtxq_insert_queue(struct umtx_q *uq, int q); 236static void umtxq_remove_queue(struct umtx_q *uq, int q); 237static int umtxq_sleep(struct umtx_q *uq, const char *wmesg, int timo); 238static int umtxq_count(struct umtx_key *key); 239static int umtx_key_match(const struct umtx_key *k1, const struct umtx_key *k2); 240static int umtx_key_get(void *addr, int type, int share, 241 struct umtx_key *key); 242static void umtx_key_release(struct umtx_key *key); 243static struct umtx_pi *umtx_pi_alloc(int); 244static void umtx_pi_free(struct umtx_pi *pi); 245static void umtx_pi_adjust_locked(struct thread *td, u_char oldpri); 246static int do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags); 247static void umtx_thread_cleanup(struct thread *td); 248static void umtx_exec_hook(void *arg __unused, struct proc *p __unused, 249 struct image_params *imgp __unused); 250SYSINIT(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 256static struct mtx umtx_lock; 257 258static void 259umtxq_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 282struct umtx_q * 283umtxq_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 295void 296umtxq_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 303static inline void 304umtxq_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 310static inline int 311umtx_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 318static inline struct umtxq_chain * 319umtxq_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 */ 329static inline void 330umtxq_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 */ 341static inline void 342umtxq_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 */ 354static inline void 355umtxq_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 */ 385static inline void 386umtxq_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 398static struct umtxq_queue * 399umtxq_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 414static inline void 415umtxq_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, UMTX_SHARED_QUEUE); 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 440static inline void 441umtxq_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 */ 470static int 471umtxq_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 */ 488static int 489umtxq_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 509static int 510umtxq_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 */ 536static inline void 537umtxq_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 */ 551static inline int 552umtxq_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 */ 570static int 571umtx_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 */ 616static inline void 617umtx_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 */ 626static 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 */ 723static int 724do_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 */ 763static int 764do_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_IA32 822 823/* 824 * Lock a umtx object. 825 */ 826static 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 */ 922static int 923do_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 */ 962static int 963do_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 */ 1024static int 1025do_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 break; 1064 if (error != ETIMEDOUT) 1065 break; 1066 umtxq_unlock(&uq->uq_key); 1067 getnanouptime(&ts2); 1068 if (timespeccmp(&ts2, &ts, >=)) { 1069 error = ETIMEDOUT; 1070 umtxq_lock(&uq->uq_key); 1071 break; 1072 } 1073 ts3 = ts; 1074 timespecsub(&ts3, &ts2); 1075 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 1076 umtxq_lock(&uq->uq_key); 1077 } 1078 umtxq_remove(uq); 1079 umtxq_unlock(&uq->uq_key); 1080 } 1081 umtx_key_release(&uq->uq_key); 1082 if (error == ERESTART) 1083 error = EINTR; 1084 return (error); 1085} 1086 1087/* 1088 * Wake up threads sleeping on the specified address. 1089 */ 1090int 1091kern_umtx_wake(struct thread *td, void *uaddr, int n_wake, int is_private) 1092{ 1093 struct umtx_key key; 1094 int ret; 1095 1096 if ((ret = umtx_key_get(uaddr, TYPE_SIMPLE_WAIT, 1097 is_private ? THREAD_SHARE : AUTO_SHARE, &key)) != 0) 1098 return (ret); 1099 umtxq_lock(&key); 1100 ret = umtxq_signal(&key, n_wake); 1101 umtxq_unlock(&key); 1102 umtx_key_release(&key); 1103 return (0); 1104} 1105 1106/* 1107 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex. 1108 */ 1109static int 1110_do_lock_normal(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1111 int mode) 1112{ 1113 struct umtx_q *uq; 1114 uint32_t owner, old, id; 1115 int error = 0; 1116 1117 id = td->td_tid; 1118 uq = td->td_umtxq; 1119 1120 /* 1121 * Care must be exercised when dealing with umtx structure. It 1122 * can fault on any access. 1123 */ 1124 for (;;) { 1125 owner = fuword32(__DEVOLATILE(void *, &m->m_owner)); 1126 if (mode == _UMUTEX_WAIT) { 1127 if (owner == UMUTEX_UNOWNED || owner == UMUTEX_CONTESTED) 1128 return (0); 1129 } else { 1130 /* 1131 * Try the uncontested case. This should be done in userland. 1132 */ 1133 owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id); 1134 1135 /* The acquire succeeded. */ 1136 if (owner == UMUTEX_UNOWNED) 1137 return (0); 1138 1139 /* The address was invalid. */ 1140 if (owner == -1) 1141 return (EFAULT); 1142 1143 /* If no one owns it but it is contested try to acquire it. */ 1144 if (owner == UMUTEX_CONTESTED) { 1145 owner = casuword32(&m->m_owner, 1146 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 1147 1148 if (owner == UMUTEX_CONTESTED) 1149 return (0); 1150 1151 /* The address was invalid. */ 1152 if (owner == -1) 1153 return (EFAULT); 1154 1155 /* If this failed the lock has changed, restart. */ 1156 continue; 1157 } 1158 } 1159 1160 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 1161 (owner & ~UMUTEX_CONTESTED) == id) 1162 return (EDEADLK); 1163 1164 if (mode == _UMUTEX_TRY) 1165 return (EBUSY); 1166 1167 /* 1168 * If we caught a signal, we have retried and now 1169 * exit immediately. 1170 */ 1171 if (error != 0) 1172 return (error); 1173 1174 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, 1175 GET_SHARE(flags), &uq->uq_key)) != 0) 1176 return (error); 1177 1178 umtxq_lock(&uq->uq_key); 1179 umtxq_busy(&uq->uq_key); 1180 umtxq_insert(uq); 1181 umtxq_unlock(&uq->uq_key); 1182 1183 /* 1184 * Set the contested bit so that a release in user space 1185 * knows to use the system call for unlock. If this fails 1186 * either some one else has acquired the lock or it has been 1187 * released. 1188 */ 1189 old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED); 1190 1191 /* The address was invalid. */ 1192 if (old == -1) { 1193 umtxq_lock(&uq->uq_key); 1194 umtxq_remove(uq); 1195 umtxq_unbusy(&uq->uq_key); 1196 umtxq_unlock(&uq->uq_key); 1197 umtx_key_release(&uq->uq_key); 1198 return (EFAULT); 1199 } 1200 1201 /* 1202 * We set the contested bit, sleep. Otherwise the lock changed 1203 * and we need to retry or we lost a race to the thread 1204 * unlocking the umtx. 1205 */ 1206 umtxq_lock(&uq->uq_key); 1207 umtxq_unbusy(&uq->uq_key); 1208 if (old == owner) 1209 error = umtxq_sleep(uq, "umtxn", timo); 1210 umtxq_remove(uq); 1211 umtxq_unlock(&uq->uq_key); 1212 umtx_key_release(&uq->uq_key); 1213 } 1214 1215 return (0); 1216} 1217 1218/* 1219 * Lock PTHREAD_PRIO_NONE protocol POSIX mutex. 1220 */ 1221/* 1222 * Unlock PTHREAD_PRIO_NONE protocol POSIX mutex. 1223 */ 1224static int 1225do_unlock_normal(struct thread *td, struct umutex *m, uint32_t flags) 1226{ 1227 struct umtx_key key; 1228 uint32_t owner, old, id; 1229 int error; 1230 int count; 1231 1232 id = td->td_tid; 1233 /* 1234 * Make sure we own this mtx. 1235 */ 1236 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1237 if (owner == -1) 1238 return (EFAULT); 1239 1240 if ((owner & ~UMUTEX_CONTESTED) != id) 1241 return (EPERM); 1242 1243 if ((owner & UMUTEX_CONTESTED) == 0) { 1244 old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED); 1245 if (old == -1) 1246 return (EFAULT); 1247 if (old == owner) 1248 return (0); 1249 owner = old; 1250 } 1251 1252 /* We should only ever be in here for contested locks */ 1253 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags), 1254 &key)) != 0) 1255 return (error); 1256 1257 umtxq_lock(&key); 1258 umtxq_busy(&key); 1259 count = umtxq_count(&key); 1260 umtxq_unlock(&key); 1261 1262 /* 1263 * When unlocking the umtx, it must be marked as unowned if 1264 * there is zero or one thread only waiting for it. 1265 * Otherwise, it must be marked as contested. 1266 */ 1267 old = casuword32(&m->m_owner, owner, 1268 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED); 1269 umtxq_lock(&key); 1270 umtxq_signal(&key,1); 1271 umtxq_unbusy(&key); 1272 umtxq_unlock(&key); 1273 umtx_key_release(&key); 1274 if (old == -1) 1275 return (EFAULT); 1276 if (old != owner) 1277 return (EINVAL); 1278 return (0); 1279} 1280 1281/* 1282 * Check if the mutex is available and wake up a waiter, 1283 * only for simple mutex. 1284 */ 1285static int 1286do_wake_umutex(struct thread *td, struct umutex *m) 1287{ 1288 struct umtx_key key; 1289 uint32_t owner; 1290 uint32_t flags; 1291 int error; 1292 int count; 1293 1294 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1295 if (owner == -1) 1296 return (EFAULT); 1297 1298 if ((owner & ~UMUTEX_CONTESTED) != 0) 1299 return (0); 1300 1301 flags = fuword32(&m->m_flags); 1302 1303 /* We should only ever be in here for contested locks */ 1304 if ((error = umtx_key_get(m, TYPE_NORMAL_UMUTEX, GET_SHARE(flags), 1305 &key)) != 0) 1306 return (error); 1307 1308 umtxq_lock(&key); 1309 umtxq_busy(&key); 1310 count = umtxq_count(&key); 1311 umtxq_unlock(&key); 1312 1313 if (count <= 1) 1314 owner = casuword32(&m->m_owner, UMUTEX_CONTESTED, UMUTEX_UNOWNED); 1315 1316 umtxq_lock(&key); 1317 if (count != 0 && (owner & ~UMUTEX_CONTESTED) == 0) 1318 umtxq_signal(&key, 1); 1319 umtxq_unbusy(&key); 1320 umtxq_unlock(&key); 1321 umtx_key_release(&key); 1322 return (0); 1323} 1324 1325static inline struct umtx_pi * 1326umtx_pi_alloc(int flags) 1327{ 1328 struct umtx_pi *pi; 1329 1330 pi = uma_zalloc(umtx_pi_zone, M_ZERO | flags); 1331 TAILQ_INIT(&pi->pi_blocked); 1332 atomic_add_int(&umtx_pi_allocated, 1); 1333 return (pi); 1334} 1335 1336static inline void 1337umtx_pi_free(struct umtx_pi *pi) 1338{ 1339 uma_zfree(umtx_pi_zone, pi); 1340 atomic_add_int(&umtx_pi_allocated, -1); 1341} 1342 1343/* 1344 * Adjust the thread's position on a pi_state after its priority has been 1345 * changed. 1346 */ 1347static int 1348umtx_pi_adjust_thread(struct umtx_pi *pi, struct thread *td) 1349{ 1350 struct umtx_q *uq, *uq1, *uq2; 1351 struct thread *td1; 1352 1353 mtx_assert(&umtx_lock, MA_OWNED); 1354 if (pi == NULL) 1355 return (0); 1356 1357 uq = td->td_umtxq; 1358 1359 /* 1360 * Check if the thread needs to be moved on the blocked chain. 1361 * It needs to be moved if either its priority is lower than 1362 * the previous thread or higher than the next thread. 1363 */ 1364 uq1 = TAILQ_PREV(uq, umtxq_head, uq_lockq); 1365 uq2 = TAILQ_NEXT(uq, uq_lockq); 1366 if ((uq1 != NULL && UPRI(td) < UPRI(uq1->uq_thread)) || 1367 (uq2 != NULL && UPRI(td) > UPRI(uq2->uq_thread))) { 1368 /* 1369 * Remove thread from blocked chain and determine where 1370 * it should be moved to. 1371 */ 1372 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq); 1373 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) { 1374 td1 = uq1->uq_thread; 1375 MPASS(td1->td_proc->p_magic == P_MAGIC); 1376 if (UPRI(td1) > UPRI(td)) 1377 break; 1378 } 1379 1380 if (uq1 == NULL) 1381 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq); 1382 else 1383 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq); 1384 } 1385 return (1); 1386} 1387 1388/* 1389 * Propagate priority when a thread is blocked on POSIX 1390 * PI mutex. 1391 */ 1392static void 1393umtx_propagate_priority(struct thread *td) 1394{ 1395 struct umtx_q *uq; 1396 struct umtx_pi *pi; 1397 int pri; 1398 1399 mtx_assert(&umtx_lock, MA_OWNED); 1400 pri = UPRI(td); 1401 uq = td->td_umtxq; 1402 pi = uq->uq_pi_blocked; 1403 if (pi == NULL) 1404 return; 1405 1406 for (;;) { 1407 td = pi->pi_owner; 1408 if (td == NULL) 1409 return; 1410 1411 MPASS(td->td_proc != NULL); 1412 MPASS(td->td_proc->p_magic == P_MAGIC); 1413 1414 if (UPRI(td) <= pri) 1415 return; 1416 1417 thread_lock(td); 1418 sched_lend_user_prio(td, pri); 1419 thread_unlock(td); 1420 1421 /* 1422 * Pick up the lock that td is blocked on. 1423 */ 1424 uq = td->td_umtxq; 1425 pi = uq->uq_pi_blocked; 1426 /* Resort td on the list if needed. */ 1427 if (!umtx_pi_adjust_thread(pi, td)) 1428 break; 1429 } 1430} 1431 1432/* 1433 * Unpropagate priority for a PI mutex when a thread blocked on 1434 * it is interrupted by signal or resumed by others. 1435 */ 1436static void 1437umtx_unpropagate_priority(struct umtx_pi *pi) 1438{ 1439 struct umtx_q *uq, *uq_owner; 1440 struct umtx_pi *pi2; 1441 int pri, oldpri; 1442 1443 mtx_assert(&umtx_lock, MA_OWNED); 1444 1445 while (pi != NULL && pi->pi_owner != NULL) { 1446 pri = PRI_MAX; 1447 uq_owner = pi->pi_owner->td_umtxq; 1448 1449 TAILQ_FOREACH(pi2, &uq_owner->uq_pi_contested, pi_link) { 1450 uq = TAILQ_FIRST(&pi2->pi_blocked); 1451 if (uq != NULL) { 1452 if (pri > UPRI(uq->uq_thread)) 1453 pri = UPRI(uq->uq_thread); 1454 } 1455 } 1456 1457 if (pri > uq_owner->uq_inherited_pri) 1458 pri = uq_owner->uq_inherited_pri; 1459 thread_lock(pi->pi_owner); 1460 oldpri = pi->pi_owner->td_user_pri; 1461 sched_unlend_user_prio(pi->pi_owner, pri); 1462 thread_unlock(pi->pi_owner); 1463 if (uq_owner->uq_pi_blocked != NULL) 1464 umtx_pi_adjust_locked(pi->pi_owner, oldpri); 1465 pi = uq_owner->uq_pi_blocked; 1466 } 1467} 1468 1469/* 1470 * Insert a PI mutex into owned list. 1471 */ 1472static void 1473umtx_pi_setowner(struct umtx_pi *pi, struct thread *owner) 1474{ 1475 struct umtx_q *uq_owner; 1476 1477 uq_owner = owner->td_umtxq; 1478 mtx_assert(&umtx_lock, MA_OWNED); 1479 if (pi->pi_owner != NULL) 1480 panic("pi_ower != NULL"); 1481 pi->pi_owner = owner; 1482 TAILQ_INSERT_TAIL(&uq_owner->uq_pi_contested, pi, pi_link); 1483} 1484 1485/* 1486 * Claim ownership of a PI mutex. 1487 */ 1488static int 1489umtx_pi_claim(struct umtx_pi *pi, struct thread *owner) 1490{ 1491 struct umtx_q *uq, *uq_owner; 1492 1493 uq_owner = owner->td_umtxq; 1494 mtx_lock_spin(&umtx_lock); 1495 if (pi->pi_owner == owner) { 1496 mtx_unlock_spin(&umtx_lock); 1497 return (0); 1498 } 1499 1500 if (pi->pi_owner != NULL) { 1501 /* 1502 * userland may have already messed the mutex, sigh. 1503 */ 1504 mtx_unlock_spin(&umtx_lock); 1505 return (EPERM); 1506 } 1507 umtx_pi_setowner(pi, owner); 1508 uq = TAILQ_FIRST(&pi->pi_blocked); 1509 if (uq != NULL) { 1510 int pri; 1511 1512 pri = UPRI(uq->uq_thread); 1513 thread_lock(owner); 1514 if (pri < UPRI(owner)) 1515 sched_lend_user_prio(owner, pri); 1516 thread_unlock(owner); 1517 } 1518 mtx_unlock_spin(&umtx_lock); 1519 return (0); 1520} 1521 1522static void 1523umtx_pi_adjust_locked(struct thread *td, u_char oldpri) 1524{ 1525 struct umtx_q *uq; 1526 struct umtx_pi *pi; 1527 1528 uq = td->td_umtxq; 1529 /* 1530 * Pick up the lock that td is blocked on. 1531 */ 1532 pi = uq->uq_pi_blocked; 1533 MPASS(pi != NULL); 1534 1535 /* Resort the turnstile on the list. */ 1536 if (!umtx_pi_adjust_thread(pi, td)) 1537 return; 1538 1539 /* 1540 * If our priority was lowered and we are at the head of the 1541 * turnstile, then propagate our new priority up the chain. 1542 */ 1543 if (uq == TAILQ_FIRST(&pi->pi_blocked) && UPRI(td) < oldpri) 1544 umtx_propagate_priority(td); 1545} 1546 1547/* 1548 * Adjust a thread's order position in its blocked PI mutex, 1549 * this may result new priority propagating process. 1550 */ 1551void 1552umtx_pi_adjust(struct thread *td, u_char oldpri) 1553{ 1554 struct umtx_q *uq; 1555 struct umtx_pi *pi; 1556 1557 uq = td->td_umtxq; 1558 mtx_lock_spin(&umtx_lock); 1559 /* 1560 * Pick up the lock that td is blocked on. 1561 */ 1562 pi = uq->uq_pi_blocked; 1563 if (pi != NULL) 1564 umtx_pi_adjust_locked(td, oldpri); 1565 mtx_unlock_spin(&umtx_lock); 1566} 1567 1568/* 1569 * Sleep on a PI mutex. 1570 */ 1571static int 1572umtxq_sleep_pi(struct umtx_q *uq, struct umtx_pi *pi, 1573 uint32_t owner, const char *wmesg, int timo) 1574{ 1575 struct umtxq_chain *uc; 1576 struct thread *td, *td1; 1577 struct umtx_q *uq1; 1578 int pri; 1579 int error = 0; 1580 1581 td = uq->uq_thread; 1582 KASSERT(td == curthread, ("inconsistent uq_thread")); 1583 uc = umtxq_getchain(&uq->uq_key); 1584 UMTXQ_LOCKED_ASSERT(uc); 1585 UMTXQ_BUSY_ASSERT(uc); 1586 umtxq_insert(uq); 1587 mtx_lock_spin(&umtx_lock); 1588 if (pi->pi_owner == NULL) { 1589 /* XXX 1590 * Current, We only support process private PI-mutex, 1591 * we need a faster way to find an owner thread for 1592 * process-shared mutex (not available yet). 1593 */ 1594 mtx_unlock_spin(&umtx_lock); 1595 PROC_LOCK(curproc); 1596 td1 = thread_find(curproc, owner); 1597 mtx_lock_spin(&umtx_lock); 1598 if (td1 != NULL && pi->pi_owner == NULL) { 1599 uq1 = td1->td_umtxq; 1600 umtx_pi_setowner(pi, td1); 1601 } 1602 PROC_UNLOCK(curproc); 1603 } 1604 1605 TAILQ_FOREACH(uq1, &pi->pi_blocked, uq_lockq) { 1606 pri = UPRI(uq1->uq_thread); 1607 if (pri > UPRI(td)) 1608 break; 1609 } 1610 1611 if (uq1 != NULL) 1612 TAILQ_INSERT_BEFORE(uq1, uq, uq_lockq); 1613 else 1614 TAILQ_INSERT_TAIL(&pi->pi_blocked, uq, uq_lockq); 1615 1616 uq->uq_pi_blocked = pi; 1617 thread_lock(td); 1618 td->td_flags |= TDF_UPIBLOCKED; 1619 thread_unlock(td); 1620 umtx_propagate_priority(td); 1621 mtx_unlock_spin(&umtx_lock); 1622 umtxq_unbusy(&uq->uq_key); 1623 1624 if (uq->uq_flags & UQF_UMTXQ) { 1625 error = msleep(uq, &uc->uc_lock, PCATCH, wmesg, timo); 1626 if (error == EWOULDBLOCK) 1627 error = ETIMEDOUT; 1628 if (uq->uq_flags & UQF_UMTXQ) { 1629 umtxq_remove(uq); 1630 } 1631 } 1632 mtx_lock_spin(&umtx_lock); 1633 uq->uq_pi_blocked = NULL; 1634 thread_lock(td); 1635 td->td_flags &= ~TDF_UPIBLOCKED; 1636 thread_unlock(td); 1637 TAILQ_REMOVE(&pi->pi_blocked, uq, uq_lockq); 1638 umtx_unpropagate_priority(pi); 1639 mtx_unlock_spin(&umtx_lock); 1640 umtxq_unlock(&uq->uq_key); 1641 1642 return (error); 1643} 1644 1645/* 1646 * Add reference count for a PI mutex. 1647 */ 1648static void 1649umtx_pi_ref(struct umtx_pi *pi) 1650{ 1651 struct umtxq_chain *uc; 1652 1653 uc = umtxq_getchain(&pi->pi_key); 1654 UMTXQ_LOCKED_ASSERT(uc); 1655 pi->pi_refcount++; 1656} 1657 1658/* 1659 * Decrease reference count for a PI mutex, if the counter 1660 * is decreased to zero, its memory space is freed. 1661 */ 1662static void 1663umtx_pi_unref(struct umtx_pi *pi) 1664{ 1665 struct umtxq_chain *uc; 1666 1667 uc = umtxq_getchain(&pi->pi_key); 1668 UMTXQ_LOCKED_ASSERT(uc); 1669 KASSERT(pi->pi_refcount > 0, ("invalid reference count")); 1670 if (--pi->pi_refcount == 0) { 1671 mtx_lock_spin(&umtx_lock); 1672 if (pi->pi_owner != NULL) { 1673 TAILQ_REMOVE(&pi->pi_owner->td_umtxq->uq_pi_contested, 1674 pi, pi_link); 1675 pi->pi_owner = NULL; 1676 } 1677 KASSERT(TAILQ_EMPTY(&pi->pi_blocked), 1678 ("blocked queue not empty")); 1679 mtx_unlock_spin(&umtx_lock); 1680 TAILQ_REMOVE(&uc->uc_pi_list, pi, pi_hashlink); 1681 umtx_pi_free(pi); 1682 } 1683} 1684 1685/* 1686 * Find a PI mutex in hash table. 1687 */ 1688static struct umtx_pi * 1689umtx_pi_lookup(struct umtx_key *key) 1690{ 1691 struct umtxq_chain *uc; 1692 struct umtx_pi *pi; 1693 1694 uc = umtxq_getchain(key); 1695 UMTXQ_LOCKED_ASSERT(uc); 1696 1697 TAILQ_FOREACH(pi, &uc->uc_pi_list, pi_hashlink) { 1698 if (umtx_key_match(&pi->pi_key, key)) { 1699 return (pi); 1700 } 1701 } 1702 return (NULL); 1703} 1704 1705/* 1706 * Insert a PI mutex into hash table. 1707 */ 1708static inline void 1709umtx_pi_insert(struct umtx_pi *pi) 1710{ 1711 struct umtxq_chain *uc; 1712 1713 uc = umtxq_getchain(&pi->pi_key); 1714 UMTXQ_LOCKED_ASSERT(uc); 1715 TAILQ_INSERT_TAIL(&uc->uc_pi_list, pi, pi_hashlink); 1716} 1717 1718/* 1719 * Lock a PI mutex. 1720 */ 1721static int 1722_do_lock_pi(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1723 int try) 1724{ 1725 struct umtx_q *uq; 1726 struct umtx_pi *pi, *new_pi; 1727 uint32_t id, owner, old; 1728 int error; 1729 1730 id = td->td_tid; 1731 uq = td->td_umtxq; 1732 1733 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags), 1734 &uq->uq_key)) != 0) 1735 return (error); 1736 umtxq_lock(&uq->uq_key); 1737 pi = umtx_pi_lookup(&uq->uq_key); 1738 if (pi == NULL) { 1739 new_pi = umtx_pi_alloc(M_NOWAIT); 1740 if (new_pi == NULL) { 1741 umtxq_unlock(&uq->uq_key); 1742 new_pi = umtx_pi_alloc(M_WAITOK); 1743 umtxq_lock(&uq->uq_key); 1744 pi = umtx_pi_lookup(&uq->uq_key); 1745 if (pi != NULL) { 1746 umtx_pi_free(new_pi); 1747 new_pi = NULL; 1748 } 1749 } 1750 if (new_pi != NULL) { 1751 new_pi->pi_key = uq->uq_key; 1752 umtx_pi_insert(new_pi); 1753 pi = new_pi; 1754 } 1755 } 1756 umtx_pi_ref(pi); 1757 umtxq_unlock(&uq->uq_key); 1758 1759 /* 1760 * Care must be exercised when dealing with umtx structure. It 1761 * can fault on any access. 1762 */ 1763 for (;;) { 1764 /* 1765 * Try the uncontested case. This should be done in userland. 1766 */ 1767 owner = casuword32(&m->m_owner, UMUTEX_UNOWNED, id); 1768 1769 /* The acquire succeeded. */ 1770 if (owner == UMUTEX_UNOWNED) { 1771 error = 0; 1772 break; 1773 } 1774 1775 /* The address was invalid. */ 1776 if (owner == -1) { 1777 error = EFAULT; 1778 break; 1779 } 1780 1781 /* If no one owns it but it is contested try to acquire it. */ 1782 if (owner == UMUTEX_CONTESTED) { 1783 owner = casuword32(&m->m_owner, 1784 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 1785 1786 if (owner == UMUTEX_CONTESTED) { 1787 umtxq_lock(&uq->uq_key); 1788 umtxq_busy(&uq->uq_key); 1789 error = umtx_pi_claim(pi, td); 1790 umtxq_unbusy(&uq->uq_key); 1791 umtxq_unlock(&uq->uq_key); 1792 break; 1793 } 1794 1795 /* The address was invalid. */ 1796 if (owner == -1) { 1797 error = EFAULT; 1798 break; 1799 } 1800 1801 /* If this failed the lock has changed, restart. */ 1802 continue; 1803 } 1804 1805 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 1806 (owner & ~UMUTEX_CONTESTED) == id) { 1807 error = EDEADLK; 1808 break; 1809 } 1810 1811 if (try != 0) { 1812 error = EBUSY; 1813 break; 1814 } 1815 1816 /* 1817 * If we caught a signal, we have retried and now 1818 * exit immediately. 1819 */ 1820 if (error != 0) 1821 break; 1822 1823 umtxq_lock(&uq->uq_key); 1824 umtxq_busy(&uq->uq_key); 1825 umtxq_unlock(&uq->uq_key); 1826 1827 /* 1828 * Set the contested bit so that a release in user space 1829 * knows to use the system call for unlock. If this fails 1830 * either some one else has acquired the lock or it has been 1831 * released. 1832 */ 1833 old = casuword32(&m->m_owner, owner, owner | UMUTEX_CONTESTED); 1834 1835 /* The address was invalid. */ 1836 if (old == -1) { 1837 umtxq_lock(&uq->uq_key); 1838 umtxq_unbusy(&uq->uq_key); 1839 umtxq_unlock(&uq->uq_key); 1840 error = EFAULT; 1841 break; 1842 } 1843 1844 umtxq_lock(&uq->uq_key); 1845 /* 1846 * We set the contested bit, sleep. Otherwise the lock changed 1847 * and we need to retry or we lost a race to the thread 1848 * unlocking the umtx. 1849 */ 1850 if (old == owner) 1851 error = umtxq_sleep_pi(uq, pi, owner & ~UMUTEX_CONTESTED, 1852 "umtxpi", timo); 1853 else { 1854 umtxq_unbusy(&uq->uq_key); 1855 umtxq_unlock(&uq->uq_key); 1856 } 1857 } 1858 1859 umtxq_lock(&uq->uq_key); 1860 umtx_pi_unref(pi); 1861 umtxq_unlock(&uq->uq_key); 1862 1863 umtx_key_release(&uq->uq_key); 1864 return (error); 1865} 1866 1867/* 1868 * Unlock a PI mutex. 1869 */ 1870static int 1871do_unlock_pi(struct thread *td, struct umutex *m, uint32_t flags) 1872{ 1873 struct umtx_key key; 1874 struct umtx_q *uq_first, *uq_first2, *uq_me; 1875 struct umtx_pi *pi, *pi2; 1876 uint32_t owner, old, id; 1877 int error; 1878 int count; 1879 int pri; 1880 1881 id = td->td_tid; 1882 /* 1883 * Make sure we own this mtx. 1884 */ 1885 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 1886 if (owner == -1) 1887 return (EFAULT); 1888 1889 if ((owner & ~UMUTEX_CONTESTED) != id) 1890 return (EPERM); 1891 1892 /* This should be done in userland */ 1893 if ((owner & UMUTEX_CONTESTED) == 0) { 1894 old = casuword32(&m->m_owner, owner, UMUTEX_UNOWNED); 1895 if (old == -1) 1896 return (EFAULT); 1897 if (old == owner) 1898 return (0); 1899 owner = old; 1900 } 1901 1902 /* We should only ever be in here for contested locks */ 1903 if ((error = umtx_key_get(m, TYPE_PI_UMUTEX, GET_SHARE(flags), 1904 &key)) != 0) 1905 return (error); 1906 1907 umtxq_lock(&key); 1908 umtxq_busy(&key); 1909 count = umtxq_count_pi(&key, &uq_first); 1910 if (uq_first != NULL) { 1911 mtx_lock_spin(&umtx_lock); 1912 pi = uq_first->uq_pi_blocked; 1913 KASSERT(pi != NULL, ("pi == NULL?")); 1914 if (pi->pi_owner != curthread) { 1915 mtx_unlock_spin(&umtx_lock); 1916 umtxq_unbusy(&key); 1917 umtxq_unlock(&key); 1918 umtx_key_release(&key); 1919 /* userland messed the mutex */ 1920 return (EPERM); 1921 } 1922 uq_me = curthread->td_umtxq; 1923 pi->pi_owner = NULL; 1924 TAILQ_REMOVE(&uq_me->uq_pi_contested, pi, pi_link); 1925 /* get highest priority thread which is still sleeping. */ 1926 uq_first = TAILQ_FIRST(&pi->pi_blocked); 1927 while (uq_first != NULL && 1928 (uq_first->uq_flags & UQF_UMTXQ) == 0) { 1929 uq_first = TAILQ_NEXT(uq_first, uq_lockq); 1930 } 1931 pri = PRI_MAX; 1932 TAILQ_FOREACH(pi2, &uq_me->uq_pi_contested, pi_link) { 1933 uq_first2 = TAILQ_FIRST(&pi2->pi_blocked); 1934 if (uq_first2 != NULL) { 1935 if (pri > UPRI(uq_first2->uq_thread)) 1936 pri = UPRI(uq_first2->uq_thread); 1937 } 1938 } 1939 thread_lock(curthread); 1940 sched_unlend_user_prio(curthread, pri); 1941 thread_unlock(curthread); 1942 mtx_unlock_spin(&umtx_lock); 1943 if (uq_first) 1944 umtxq_signal_thread(uq_first); 1945 } 1946 umtxq_unlock(&key); 1947 1948 /* 1949 * When unlocking the umtx, it must be marked as unowned if 1950 * there is zero or one thread only waiting for it. 1951 * Otherwise, it must be marked as contested. 1952 */ 1953 old = casuword32(&m->m_owner, owner, 1954 count <= 1 ? UMUTEX_UNOWNED : UMUTEX_CONTESTED); 1955 1956 umtxq_lock(&key); 1957 umtxq_unbusy(&key); 1958 umtxq_unlock(&key); 1959 umtx_key_release(&key); 1960 if (old == -1) 1961 return (EFAULT); 1962 if (old != owner) 1963 return (EINVAL); 1964 return (0); 1965} 1966 1967/* 1968 * Lock a PP mutex. 1969 */ 1970static int 1971_do_lock_pp(struct thread *td, struct umutex *m, uint32_t flags, int timo, 1972 int try) 1973{ 1974 struct umtx_q *uq, *uq2; 1975 struct umtx_pi *pi; 1976 uint32_t ceiling; 1977 uint32_t owner, id; 1978 int error, pri, old_inherited_pri, su; 1979 1980 id = td->td_tid; 1981 uq = td->td_umtxq; 1982 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 1983 &uq->uq_key)) != 0) 1984 return (error); 1985 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0); 1986 for (;;) { 1987 old_inherited_pri = uq->uq_inherited_pri; 1988 umtxq_lock(&uq->uq_key); 1989 umtxq_busy(&uq->uq_key); 1990 umtxq_unlock(&uq->uq_key); 1991 1992 ceiling = RTP_PRIO_MAX - fuword32(&m->m_ceilings[0]); 1993 if (ceiling > RTP_PRIO_MAX) { 1994 error = EINVAL; 1995 goto out; 1996 } 1997 1998 mtx_lock_spin(&umtx_lock); 1999 if (UPRI(td) < PRI_MIN_REALTIME + ceiling) { 2000 mtx_unlock_spin(&umtx_lock); 2001 error = EINVAL; 2002 goto out; 2003 } 2004 if (su && PRI_MIN_REALTIME + ceiling < uq->uq_inherited_pri) { 2005 uq->uq_inherited_pri = PRI_MIN_REALTIME + ceiling; 2006 thread_lock(td); 2007 if (uq->uq_inherited_pri < UPRI(td)) 2008 sched_lend_user_prio(td, uq->uq_inherited_pri); 2009 thread_unlock(td); 2010 } 2011 mtx_unlock_spin(&umtx_lock); 2012 2013 owner = casuword32(&m->m_owner, 2014 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 2015 2016 if (owner == UMUTEX_CONTESTED) { 2017 error = 0; 2018 break; 2019 } 2020 2021 /* The address was invalid. */ 2022 if (owner == -1) { 2023 error = EFAULT; 2024 break; 2025 } 2026 2027 if ((flags & UMUTEX_ERROR_CHECK) != 0 && 2028 (owner & ~UMUTEX_CONTESTED) == id) { 2029 error = EDEADLK; 2030 break; 2031 } 2032 2033 if (try != 0) { 2034 error = EBUSY; 2035 break; 2036 } 2037 2038 /* 2039 * If we caught a signal, we have retried and now 2040 * exit immediately. 2041 */ 2042 if (error != 0) 2043 break; 2044 2045 umtxq_lock(&uq->uq_key); 2046 umtxq_insert(uq); 2047 umtxq_unbusy(&uq->uq_key); 2048 error = umtxq_sleep(uq, "umtxpp", timo); 2049 umtxq_remove(uq); 2050 umtxq_unlock(&uq->uq_key); 2051 2052 mtx_lock_spin(&umtx_lock); 2053 uq->uq_inherited_pri = old_inherited_pri; 2054 pri = PRI_MAX; 2055 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2056 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2057 if (uq2 != NULL) { 2058 if (pri > UPRI(uq2->uq_thread)) 2059 pri = UPRI(uq2->uq_thread); 2060 } 2061 } 2062 if (pri > uq->uq_inherited_pri) 2063 pri = uq->uq_inherited_pri; 2064 thread_lock(td); 2065 sched_unlend_user_prio(td, pri); 2066 thread_unlock(td); 2067 mtx_unlock_spin(&umtx_lock); 2068 } 2069 2070 if (error != 0) { 2071 mtx_lock_spin(&umtx_lock); 2072 uq->uq_inherited_pri = old_inherited_pri; 2073 pri = PRI_MAX; 2074 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2075 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2076 if (uq2 != NULL) { 2077 if (pri > UPRI(uq2->uq_thread)) 2078 pri = UPRI(uq2->uq_thread); 2079 } 2080 } 2081 if (pri > uq->uq_inherited_pri) 2082 pri = uq->uq_inherited_pri; 2083 thread_lock(td); 2084 sched_unlend_user_prio(td, pri); 2085 thread_unlock(td); 2086 mtx_unlock_spin(&umtx_lock); 2087 } 2088 2089out: 2090 umtxq_lock(&uq->uq_key); 2091 umtxq_unbusy(&uq->uq_key); 2092 umtxq_unlock(&uq->uq_key); 2093 umtx_key_release(&uq->uq_key); 2094 return (error); 2095} 2096 2097/* 2098 * Unlock a PP mutex. 2099 */ 2100static int 2101do_unlock_pp(struct thread *td, struct umutex *m, uint32_t flags) 2102{ 2103 struct umtx_key key; 2104 struct umtx_q *uq, *uq2; 2105 struct umtx_pi *pi; 2106 uint32_t owner, id; 2107 uint32_t rceiling; 2108 int error, pri, new_inherited_pri, su; 2109 2110 id = td->td_tid; 2111 uq = td->td_umtxq; 2112 su = (priv_check(td, PRIV_SCHED_RTPRIO) == 0); 2113 2114 /* 2115 * Make sure we own this mtx. 2116 */ 2117 owner = fuword32(__DEVOLATILE(uint32_t *, &m->m_owner)); 2118 if (owner == -1) 2119 return (EFAULT); 2120 2121 if ((owner & ~UMUTEX_CONTESTED) != id) 2122 return (EPERM); 2123 2124 error = copyin(&m->m_ceilings[1], &rceiling, sizeof(uint32_t)); 2125 if (error != 0) 2126 return (error); 2127 2128 if (rceiling == -1) 2129 new_inherited_pri = PRI_MAX; 2130 else { 2131 rceiling = RTP_PRIO_MAX - rceiling; 2132 if (rceiling > RTP_PRIO_MAX) 2133 return (EINVAL); 2134 new_inherited_pri = PRI_MIN_REALTIME + rceiling; 2135 } 2136 2137 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 2138 &key)) != 0) 2139 return (error); 2140 umtxq_lock(&key); 2141 umtxq_busy(&key); 2142 umtxq_unlock(&key); 2143 /* 2144 * For priority protected mutex, always set unlocked state 2145 * to UMUTEX_CONTESTED, so that userland always enters kernel 2146 * to lock the mutex, it is necessary because thread priority 2147 * has to be adjusted for such mutex. 2148 */ 2149 error = suword32(__DEVOLATILE(uint32_t *, &m->m_owner), 2150 UMUTEX_CONTESTED); 2151 2152 umtxq_lock(&key); 2153 if (error == 0) 2154 umtxq_signal(&key, 1); 2155 umtxq_unbusy(&key); 2156 umtxq_unlock(&key); 2157 2158 if (error == -1) 2159 error = EFAULT; 2160 else { 2161 mtx_lock_spin(&umtx_lock); 2162 if (su != 0) 2163 uq->uq_inherited_pri = new_inherited_pri; 2164 pri = PRI_MAX; 2165 TAILQ_FOREACH(pi, &uq->uq_pi_contested, pi_link) { 2166 uq2 = TAILQ_FIRST(&pi->pi_blocked); 2167 if (uq2 != NULL) { 2168 if (pri > UPRI(uq2->uq_thread)) 2169 pri = UPRI(uq2->uq_thread); 2170 } 2171 } 2172 if (pri > uq->uq_inherited_pri) 2173 pri = uq->uq_inherited_pri; 2174 thread_lock(td); 2175 sched_unlend_user_prio(td, pri); 2176 thread_unlock(td); 2177 mtx_unlock_spin(&umtx_lock); 2178 } 2179 umtx_key_release(&key); 2180 return (error); 2181} 2182 2183static int 2184do_set_ceiling(struct thread *td, struct umutex *m, uint32_t ceiling, 2185 uint32_t *old_ceiling) 2186{ 2187 struct umtx_q *uq; 2188 uint32_t save_ceiling; 2189 uint32_t owner, id; 2190 uint32_t flags; 2191 int error; 2192 2193 flags = fuword32(&m->m_flags); 2194 if ((flags & UMUTEX_PRIO_PROTECT) == 0) 2195 return (EINVAL); 2196 if (ceiling > RTP_PRIO_MAX) 2197 return (EINVAL); 2198 id = td->td_tid; 2199 uq = td->td_umtxq; 2200 if ((error = umtx_key_get(m, TYPE_PP_UMUTEX, GET_SHARE(flags), 2201 &uq->uq_key)) != 0) 2202 return (error); 2203 for (;;) { 2204 umtxq_lock(&uq->uq_key); 2205 umtxq_busy(&uq->uq_key); 2206 umtxq_unlock(&uq->uq_key); 2207 2208 save_ceiling = fuword32(&m->m_ceilings[0]); 2209 2210 owner = casuword32(&m->m_owner, 2211 UMUTEX_CONTESTED, id | UMUTEX_CONTESTED); 2212 2213 if (owner == UMUTEX_CONTESTED) { 2214 suword32(&m->m_ceilings[0], ceiling); 2215 suword32(__DEVOLATILE(uint32_t *, &m->m_owner), 2216 UMUTEX_CONTESTED); 2217 error = 0; 2218 break; 2219 } 2220 2221 /* The address was invalid. */ 2222 if (owner == -1) { 2223 error = EFAULT; 2224 break; 2225 } 2226 2227 if ((owner & ~UMUTEX_CONTESTED) == id) { 2228 suword32(&m->m_ceilings[0], ceiling); 2229 error = 0; 2230 break; 2231 } 2232 2233 /* 2234 * If we caught a signal, we have retried and now 2235 * exit immediately. 2236 */ 2237 if (error != 0) 2238 break; 2239 2240 /* 2241 * We set the contested bit, sleep. Otherwise the lock changed 2242 * and we need to retry or we lost a race to the thread 2243 * unlocking the umtx. 2244 */ 2245 umtxq_lock(&uq->uq_key); 2246 umtxq_insert(uq); 2247 umtxq_unbusy(&uq->uq_key); 2248 error = umtxq_sleep(uq, "umtxpp", 0); 2249 umtxq_remove(uq); 2250 umtxq_unlock(&uq->uq_key); 2251 } 2252 umtxq_lock(&uq->uq_key); 2253 if (error == 0) 2254 umtxq_signal(&uq->uq_key, INT_MAX); 2255 umtxq_unbusy(&uq->uq_key); 2256 umtxq_unlock(&uq->uq_key); 2257 umtx_key_release(&uq->uq_key); 2258 if (error == 0 && old_ceiling != NULL) 2259 suword32(old_ceiling, save_ceiling); 2260 return (error); 2261} 2262 2263static int 2264_do_lock_umutex(struct thread *td, struct umutex *m, int flags, int timo, 2265 int mode) 2266{ 2267 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) { 2268 case 0: 2269 return (_do_lock_normal(td, m, flags, timo, mode)); 2270 case UMUTEX_PRIO_INHERIT: 2271 return (_do_lock_pi(td, m, flags, timo, mode)); 2272 case UMUTEX_PRIO_PROTECT: 2273 return (_do_lock_pp(td, m, flags, timo, mode)); 2274 } 2275 return (EINVAL); 2276} 2277 2278/* 2279 * Lock a userland POSIX mutex. 2280 */ 2281static int 2282do_lock_umutex(struct thread *td, struct umutex *m, 2283 struct timespec *timeout, int mode) 2284{ 2285 struct timespec ts, ts2, ts3; 2286 struct timeval tv; 2287 uint32_t flags; 2288 int error; 2289 2290 flags = fuword32(&m->m_flags); 2291 if (flags == -1) 2292 return (EFAULT); 2293 2294 if (timeout == NULL) { 2295 error = _do_lock_umutex(td, m, flags, 0, mode); 2296 /* Mutex locking is restarted if it is interrupted. */ 2297 if (error == EINTR && mode != _UMUTEX_WAIT) 2298 error = ERESTART; 2299 } else { 2300 getnanouptime(&ts); 2301 timespecadd(&ts, timeout); 2302 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2303 for (;;) { 2304 error = _do_lock_umutex(td, m, flags, tvtohz(&tv), mode); 2305 if (error != ETIMEDOUT) 2306 break; 2307 getnanouptime(&ts2); 2308 if (timespeccmp(&ts2, &ts, >=)) { 2309 error = ETIMEDOUT; 2310 break; 2311 } 2312 ts3 = ts; 2313 timespecsub(&ts3, &ts2); 2314 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2315 } 2316 /* Timed-locking is not restarted. */ 2317 if (error == ERESTART) 2318 error = EINTR; 2319 } 2320 return (error); 2321} 2322 2323/* 2324 * Unlock a userland POSIX mutex. 2325 */ 2326static int 2327do_unlock_umutex(struct thread *td, struct umutex *m) 2328{ 2329 uint32_t flags; 2330 2331 flags = fuword32(&m->m_flags); 2332 if (flags == -1) 2333 return (EFAULT); 2334 2335 switch(flags & (UMUTEX_PRIO_INHERIT | UMUTEX_PRIO_PROTECT)) { 2336 case 0: 2337 return (do_unlock_normal(td, m, flags)); 2338 case UMUTEX_PRIO_INHERIT: 2339 return (do_unlock_pi(td, m, flags)); 2340 case UMUTEX_PRIO_PROTECT: 2341 return (do_unlock_pp(td, m, flags)); 2342 } 2343 2344 return (EINVAL); 2345} 2346 2347static int 2348do_cv_wait(struct thread *td, struct ucond *cv, struct umutex *m, 2349 struct timespec *timeout, u_long wflags) 2350{ 2351 struct umtx_q *uq; 2352 struct timeval tv; 2353 struct timespec cts, ets, tts; 2354 uint32_t flags; 2355 int error; 2356 2357 uq = td->td_umtxq; 2358 flags = fuword32(&cv->c_flags); 2359 error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &uq->uq_key); 2360 if (error != 0) 2361 return (error); 2362 umtxq_lock(&uq->uq_key); 2363 umtxq_busy(&uq->uq_key); 2364 umtxq_insert(uq); 2365 umtxq_unlock(&uq->uq_key); 2366 2367 /* 2368 * The magic thing is we should set c_has_waiters to 1 before 2369 * releasing user mutex. 2370 */ 2371 suword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 1); 2372 2373 umtxq_lock(&uq->uq_key); 2374 umtxq_unbusy(&uq->uq_key); 2375 umtxq_unlock(&uq->uq_key); 2376 2377 error = do_unlock_umutex(td, m); 2378 2379 umtxq_lock(&uq->uq_key); 2380 if (error == 0) { 2381 if ((wflags & UMTX_CHECK_UNPARKING) && 2382 (td->td_pflags & TDP_WAKEUP)) { 2383 td->td_pflags &= ~TDP_WAKEUP; 2384 error = EINTR; 2385 } else if (timeout == NULL) { 2386 error = umtxq_sleep(uq, "ucond", 0); 2387 } else { 2388 getnanouptime(&ets); 2389 timespecadd(&ets, timeout); 2390 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2391 for (;;) { 2392 error = umtxq_sleep(uq, "ucond", tvtohz(&tv)); 2393 if (error != ETIMEDOUT) 2394 break; 2395 getnanouptime(&cts); 2396 if (timespeccmp(&cts, &ets, >=)) { 2397 error = ETIMEDOUT; 2398 break; 2399 } 2400 tts = ets; 2401 timespecsub(&tts, &cts); 2402 TIMESPEC_TO_TIMEVAL(&tv, &tts); 2403 } 2404 } 2405 } 2406 2407 if (error != 0) { 2408 if ((uq->uq_flags & UQF_UMTXQ) == 0) { 2409 /* 2410 * If we concurrently got do_cv_signal()d 2411 * and we got an error or UNIX signals or a timeout, 2412 * then, perform another umtxq_signal to avoid 2413 * consuming the wakeup. This may cause supurious 2414 * wakeup for another thread which was just queued, 2415 * but SUSV3 explicitly allows supurious wakeup to 2416 * occur, and indeed a kernel based implementation 2417 * can not avoid it. 2418 */ 2419 if (!umtxq_signal(&uq->uq_key, 1)) 2420 error = 0; 2421 } 2422 if (error == ERESTART) 2423 error = EINTR; 2424 } 2425 umtxq_remove(uq); 2426 umtxq_unlock(&uq->uq_key); 2427 umtx_key_release(&uq->uq_key); 2428 return (error); 2429} 2430 2431/* 2432 * Signal a userland condition variable. 2433 */ 2434static int 2435do_cv_signal(struct thread *td, struct ucond *cv) 2436{ 2437 struct umtx_key key; 2438 int error, cnt, nwake; 2439 uint32_t flags; 2440 2441 flags = fuword32(&cv->c_flags); 2442 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0) 2443 return (error); 2444 umtxq_lock(&key); 2445 umtxq_busy(&key); 2446 cnt = umtxq_count(&key); 2447 nwake = umtxq_signal(&key, 1); 2448 if (cnt <= nwake) { 2449 umtxq_unlock(&key); 2450 error = suword32( 2451 __DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0); 2452 umtxq_lock(&key); 2453 } 2454 umtxq_unbusy(&key); 2455 umtxq_unlock(&key); 2456 umtx_key_release(&key); 2457 return (error); 2458} 2459 2460static int 2461do_cv_broadcast(struct thread *td, struct ucond *cv) 2462{ 2463 struct umtx_key key; 2464 int error; 2465 uint32_t flags; 2466 2467 flags = fuword32(&cv->c_flags); 2468 if ((error = umtx_key_get(cv, TYPE_CV, GET_SHARE(flags), &key)) != 0) 2469 return (error); 2470 2471 umtxq_lock(&key); 2472 umtxq_busy(&key); 2473 umtxq_signal(&key, INT_MAX); 2474 umtxq_unlock(&key); 2475 2476 error = suword32(__DEVOLATILE(uint32_t *, &cv->c_has_waiters), 0); 2477 2478 umtxq_lock(&key); 2479 umtxq_unbusy(&key); 2480 umtxq_unlock(&key); 2481 2482 umtx_key_release(&key); 2483 return (error); 2484} 2485 2486static int 2487do_rw_rdlock(struct thread *td, struct urwlock *rwlock, long fflag, int timo) 2488{ 2489 struct umtx_q *uq; 2490 uint32_t flags, wrflags; 2491 int32_t state, oldstate; 2492 int32_t blocked_readers; 2493 int error; 2494 2495 uq = td->td_umtxq; 2496 flags = fuword32(&rwlock->rw_flags); 2497 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2498 if (error != 0) 2499 return (error); 2500 2501 wrflags = URWLOCK_WRITE_OWNER; 2502 if (!(fflag & URWLOCK_PREFER_READER) && !(flags & URWLOCK_PREFER_READER)) 2503 wrflags |= URWLOCK_WRITE_WAITERS; 2504 2505 for (;;) { 2506 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2507 /* try to lock it */ 2508 while (!(state & wrflags)) { 2509 if (__predict_false(URWLOCK_READER_COUNT(state) == URWLOCK_MAX_READERS)) { 2510 umtx_key_release(&uq->uq_key); 2511 return (EAGAIN); 2512 } 2513 oldstate = casuword32(&rwlock->rw_state, state, state + 1); 2514 if (oldstate == state) { 2515 umtx_key_release(&uq->uq_key); 2516 return (0); 2517 } 2518 state = oldstate; 2519 } 2520 2521 if (error) 2522 break; 2523 2524 /* grab monitor lock */ 2525 umtxq_lock(&uq->uq_key); 2526 umtxq_busy(&uq->uq_key); 2527 umtxq_unlock(&uq->uq_key); 2528 2529 /* set read contention bit */ 2530 while ((state & wrflags) && !(state & URWLOCK_READ_WAITERS)) { 2531 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_READ_WAITERS); 2532 if (oldstate == state) 2533 goto sleep; 2534 state = oldstate; 2535 } 2536 2537 /* state is changed while setting flags, restart */ 2538 if (!(state & wrflags)) { 2539 umtxq_lock(&uq->uq_key); 2540 umtxq_unbusy(&uq->uq_key); 2541 umtxq_unlock(&uq->uq_key); 2542 continue; 2543 } 2544 2545sleep: 2546 /* contention bit is set, before sleeping, increase read waiter count */ 2547 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2548 suword32(&rwlock->rw_blocked_readers, blocked_readers+1); 2549 2550 while (state & wrflags) { 2551 umtxq_lock(&uq->uq_key); 2552 umtxq_insert(uq); 2553 umtxq_unbusy(&uq->uq_key); 2554 2555 error = umtxq_sleep(uq, "urdlck", timo); 2556 2557 umtxq_busy(&uq->uq_key); 2558 umtxq_remove(uq); 2559 umtxq_unlock(&uq->uq_key); 2560 if (error) 2561 break; 2562 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2563 } 2564 2565 /* decrease read waiter count, and may clear read contention bit */ 2566 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2567 suword32(&rwlock->rw_blocked_readers, blocked_readers-1); 2568 if (blocked_readers == 1) { 2569 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2570 for (;;) { 2571 oldstate = casuword32(&rwlock->rw_state, state, 2572 state & ~URWLOCK_READ_WAITERS); 2573 if (oldstate == state) 2574 break; 2575 state = oldstate; 2576 } 2577 } 2578 2579 umtxq_lock(&uq->uq_key); 2580 umtxq_unbusy(&uq->uq_key); 2581 umtxq_unlock(&uq->uq_key); 2582 } 2583 umtx_key_release(&uq->uq_key); 2584 return (error); 2585} 2586 2587static int 2588do_rw_rdlock2(struct thread *td, void *obj, long val, struct timespec *timeout) 2589{ 2590 struct timespec ts, ts2, ts3; 2591 struct timeval tv; 2592 int error; 2593 2594 getnanouptime(&ts); 2595 timespecadd(&ts, timeout); 2596 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2597 for (;;) { 2598 error = do_rw_rdlock(td, obj, val, tvtohz(&tv)); 2599 if (error != ETIMEDOUT) 2600 break; 2601 getnanouptime(&ts2); 2602 if (timespeccmp(&ts2, &ts, >=)) { 2603 error = ETIMEDOUT; 2604 break; 2605 } 2606 ts3 = ts; 2607 timespecsub(&ts3, &ts2); 2608 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2609 } 2610 if (error == ERESTART) 2611 error = EINTR; 2612 return (error); 2613} 2614 2615static int 2616do_rw_wrlock(struct thread *td, struct urwlock *rwlock, int timo) 2617{ 2618 struct umtx_q *uq; 2619 uint32_t flags; 2620 int32_t state, oldstate; 2621 int32_t blocked_writers; 2622 int32_t blocked_readers; 2623 int error; 2624 2625 uq = td->td_umtxq; 2626 flags = fuword32(&rwlock->rw_flags); 2627 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2628 if (error != 0) 2629 return (error); 2630 2631 blocked_readers = 0; 2632 for (;;) { 2633 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2634 while (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) { 2635 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_OWNER); 2636 if (oldstate == state) { 2637 umtx_key_release(&uq->uq_key); 2638 return (0); 2639 } 2640 state = oldstate; 2641 } 2642 2643 if (error) { 2644 if (!(state & (URWLOCK_WRITE_OWNER|URWLOCK_WRITE_WAITERS)) && 2645 blocked_readers != 0) { 2646 umtxq_lock(&uq->uq_key); 2647 umtxq_busy(&uq->uq_key); 2648 umtxq_signal_queue(&uq->uq_key, INT_MAX, UMTX_SHARED_QUEUE); 2649 umtxq_unbusy(&uq->uq_key); 2650 umtxq_unlock(&uq->uq_key); 2651 } 2652 2653 break; 2654 } 2655 2656 /* grab monitor lock */ 2657 umtxq_lock(&uq->uq_key); 2658 umtxq_busy(&uq->uq_key); 2659 umtxq_unlock(&uq->uq_key); 2660 2661 while (((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) && 2662 (state & URWLOCK_WRITE_WAITERS) == 0) { 2663 oldstate = casuword32(&rwlock->rw_state, state, state | URWLOCK_WRITE_WAITERS); 2664 if (oldstate == state) 2665 goto sleep; 2666 state = oldstate; 2667 } 2668 2669 if (!(state & URWLOCK_WRITE_OWNER) && URWLOCK_READER_COUNT(state) == 0) { 2670 umtxq_lock(&uq->uq_key); 2671 umtxq_unbusy(&uq->uq_key); 2672 umtxq_unlock(&uq->uq_key); 2673 continue; 2674 } 2675sleep: 2676 blocked_writers = fuword32(&rwlock->rw_blocked_writers); 2677 suword32(&rwlock->rw_blocked_writers, blocked_writers+1); 2678 2679 while ((state & URWLOCK_WRITE_OWNER) || URWLOCK_READER_COUNT(state) != 0) { 2680 umtxq_lock(&uq->uq_key); 2681 umtxq_insert_queue(uq, UMTX_EXCLUSIVE_QUEUE); 2682 umtxq_unbusy(&uq->uq_key); 2683 2684 error = umtxq_sleep(uq, "uwrlck", timo); 2685 2686 umtxq_busy(&uq->uq_key); 2687 umtxq_remove_queue(uq, UMTX_EXCLUSIVE_QUEUE); 2688 umtxq_unlock(&uq->uq_key); 2689 if (error) 2690 break; 2691 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2692 } 2693 2694 blocked_writers = fuword32(&rwlock->rw_blocked_writers); 2695 suword32(&rwlock->rw_blocked_writers, blocked_writers-1); 2696 if (blocked_writers == 1) { 2697 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2698 for (;;) { 2699 oldstate = casuword32(&rwlock->rw_state, state, 2700 state & ~URWLOCK_WRITE_WAITERS); 2701 if (oldstate == state) 2702 break; 2703 state = oldstate; 2704 } 2705 blocked_readers = fuword32(&rwlock->rw_blocked_readers); 2706 } else 2707 blocked_readers = 0; 2708 2709 umtxq_lock(&uq->uq_key); 2710 umtxq_unbusy(&uq->uq_key); 2711 umtxq_unlock(&uq->uq_key); 2712 } 2713 2714 umtx_key_release(&uq->uq_key); 2715 return (error); 2716} 2717 2718static int 2719do_rw_wrlock2(struct thread *td, void *obj, struct timespec *timeout) 2720{ 2721 struct timespec ts, ts2, ts3; 2722 struct timeval tv; 2723 int error; 2724 2725 getnanouptime(&ts); 2726 timespecadd(&ts, timeout); 2727 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2728 for (;;) { 2729 error = do_rw_wrlock(td, obj, tvtohz(&tv)); 2730 if (error != ETIMEDOUT) 2731 break; 2732 getnanouptime(&ts2); 2733 if (timespeccmp(&ts2, &ts, >=)) { 2734 error = ETIMEDOUT; 2735 break; 2736 } 2737 ts3 = ts; 2738 timespecsub(&ts3, &ts2); 2739 TIMESPEC_TO_TIMEVAL(&tv, &ts3); 2740 } 2741 if (error == ERESTART) 2742 error = EINTR; 2743 return (error); 2744} 2745 2746static int 2747do_rw_unlock(struct thread *td, struct urwlock *rwlock) 2748{ 2749 struct umtx_q *uq; 2750 uint32_t flags; 2751 int32_t state, oldstate; 2752 int error, q, count; 2753 2754 uq = td->td_umtxq; 2755 flags = fuword32(&rwlock->rw_flags); 2756 error = umtx_key_get(rwlock, TYPE_RWLOCK, GET_SHARE(flags), &uq->uq_key); 2757 if (error != 0) 2758 return (error); 2759 2760 state = fuword32(__DEVOLATILE(int32_t *, &rwlock->rw_state)); 2761 if (state & URWLOCK_WRITE_OWNER) { 2762 for (;;) { 2763 oldstate = casuword32(&rwlock->rw_state, state, 2764 state & ~URWLOCK_WRITE_OWNER); 2765 if (oldstate != state) { 2766 state = oldstate; 2767 if (!(oldstate & URWLOCK_WRITE_OWNER)) { 2768 error = EPERM; 2769 goto out; 2770 } 2771 } else 2772 break; 2773 } 2774 } else if (URWLOCK_READER_COUNT(state) != 0) { 2775 for (;;) { 2776 oldstate = casuword32(&rwlock->rw_state, state, 2777 state - 1); 2778 if (oldstate != state) { 2779 state = oldstate; 2780 if (URWLOCK_READER_COUNT(oldstate) == 0) { 2781 error = EPERM; 2782 goto out; 2783 } 2784 } 2785 else 2786 break; 2787 } 2788 } else { 2789 error = EPERM; 2790 goto out; 2791 } 2792 2793 count = 0; 2794 2795 if (!(flags & URWLOCK_PREFER_READER)) { 2796 if (state & URWLOCK_WRITE_WAITERS) { 2797 count = 1; 2798 q = UMTX_EXCLUSIVE_QUEUE; 2799 } else if (state & URWLOCK_READ_WAITERS) { 2800 count = INT_MAX; 2801 q = UMTX_SHARED_QUEUE; 2802 } 2803 } else { 2804 if (state & URWLOCK_READ_WAITERS) { 2805 count = INT_MAX; 2806 q = UMTX_SHARED_QUEUE; 2807 } else if (state & URWLOCK_WRITE_WAITERS) { 2808 count = 1; 2809 q = UMTX_EXCLUSIVE_QUEUE; 2810 } 2811 } 2812 2813 if (count) { 2814 umtxq_lock(&uq->uq_key); 2815 umtxq_busy(&uq->uq_key); 2816 umtxq_signal_queue(&uq->uq_key, count, q); 2817 umtxq_unbusy(&uq->uq_key); 2818 umtxq_unlock(&uq->uq_key); 2819 } 2820out: 2821 umtx_key_release(&uq->uq_key); 2822 return (error); 2823} 2824 2825static int 2826do_sem_wait(struct thread *td, struct _usem *sem, struct timespec *timeout) 2827{ 2828 struct umtx_q *uq; 2829 struct timeval tv; 2830 struct timespec cts, ets, tts; 2831 uint32_t flags, count; 2832 int error; 2833 2834 uq = td->td_umtxq; 2835 flags = fuword32(&sem->_flags); 2836 error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &uq->uq_key); 2837 if (error != 0) 2838 return (error); 2839 umtxq_lock(&uq->uq_key); 2840 umtxq_busy(&uq->uq_key); 2841 umtxq_insert(uq); 2842 umtxq_unlock(&uq->uq_key); 2843 2844 count = fuword32(__DEVOLATILE(uint32_t *, &sem->_count)); 2845 if (count != 0) { 2846 umtxq_lock(&uq->uq_key); 2847 umtxq_unbusy(&uq->uq_key); 2848 umtxq_remove(uq); 2849 umtxq_unlock(&uq->uq_key); 2850 umtx_key_release(&uq->uq_key); 2851 return (0); 2852 } 2853 2854 /* 2855 * The magic thing is we should set c_has_waiters to 1 before 2856 * releasing user mutex. 2857 */ 2858 suword32(__DEVOLATILE(uint32_t *, &sem->_has_waiters), 1); 2859 2860 umtxq_lock(&uq->uq_key); 2861 umtxq_unbusy(&uq->uq_key); 2862 umtxq_unlock(&uq->uq_key); 2863 2864 umtxq_lock(&uq->uq_key); 2865 if (timeout == NULL) { 2866 error = umtxq_sleep(uq, "usem", 0); 2867 } else { 2868 getnanouptime(&ets); 2869 timespecadd(&ets, timeout); 2870 TIMESPEC_TO_TIMEVAL(&tv, timeout); 2871 for (;;) { 2872 error = umtxq_sleep(uq, "usem", tvtohz(&tv)); 2873 if (error != ETIMEDOUT) 2874 break; 2875 getnanouptime(&cts); 2876 if (timespeccmp(&cts, &ets, >=)) { 2877 error = ETIMEDOUT; 2878 break; 2879 } 2880 tts = ets; 2881 timespecsub(&tts, &cts); 2882 TIMESPEC_TO_TIMEVAL(&tv, &tts); 2883 } 2884 } 2885 2886 if (error != 0) { 2887 if ((uq->uq_flags & UQF_UMTXQ) == 0) { 2888 if (!umtxq_signal(&uq->uq_key, 1)) 2889 error = 0; 2890 } 2891 if (error == ERESTART) 2892 error = EINTR; 2893 } 2894 umtxq_remove(uq); 2895 umtxq_unlock(&uq->uq_key); 2896 umtx_key_release(&uq->uq_key); 2897 return (error); 2898} 2899 2900/* 2901 * Signal a userland condition variable. 2902 */ 2903static int 2904do_sem_wake(struct thread *td, struct _usem *sem) 2905{ 2906 struct umtx_key key; 2907 int error, cnt, nwake; 2908 uint32_t flags; 2909 2910 flags = fuword32(&sem->_flags); 2911 if ((error = umtx_key_get(sem, TYPE_SEM, GET_SHARE(flags), &key)) != 0) 2912 return (error); 2913 umtxq_lock(&key); 2914 umtxq_busy(&key); 2915 cnt = umtxq_count(&key); 2916 nwake = umtxq_signal(&key, 1); 2917 if (cnt <= nwake) { 2918 umtxq_unlock(&key); 2919 error = suword32( 2920 __DEVOLATILE(uint32_t *, &sem->_has_waiters), 0); 2921 umtxq_lock(&key); 2922 } 2923 umtxq_unbusy(&key); 2924 umtxq_unlock(&key); 2925 umtx_key_release(&key); 2926 return (error); 2927} 2928 2929int 2930_umtx_lock(struct thread *td, struct _umtx_lock_args *uap) 2931 /* struct umtx *umtx */ 2932{ 2933 return _do_lock_umtx(td, uap->umtx, td->td_tid, 0); 2934} 2935 2936int 2937_umtx_unlock(struct thread *td, struct _umtx_unlock_args *uap) 2938 /* struct umtx *umtx */ 2939{ 2940 return do_unlock_umtx(td, uap->umtx, td->td_tid); 2941} 2942 2943static int 2944__umtx_op_lock_umtx(struct thread *td, struct _umtx_op_args *uap) 2945{ 2946 struct timespec *ts, timeout; 2947 int error; 2948 2949 /* Allow a null timespec (wait forever). */ 2950 if (uap->uaddr2 == NULL) 2951 ts = NULL; 2952 else { 2953 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 2954 if (error != 0) 2955 return (error); 2956 if (timeout.tv_nsec >= 1000000000 || 2957 timeout.tv_nsec < 0) { 2958 return (EINVAL); 2959 } 2960 ts = &timeout; 2961 } 2962 return (do_lock_umtx(td, uap->obj, uap->val, ts)); 2963} 2964 2965static int 2966__umtx_op_unlock_umtx(struct thread *td, struct _umtx_op_args *uap) 2967{ 2968 return (do_unlock_umtx(td, uap->obj, uap->val)); 2969} 2970 2971static int 2972__umtx_op_wait(struct thread *td, struct _umtx_op_args *uap) 2973{ 2974 struct timespec *ts, timeout; 2975 int error; 2976 2977 if (uap->uaddr2 == NULL) 2978 ts = NULL; 2979 else { 2980 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 2981 if (error != 0) 2982 return (error); 2983 if (timeout.tv_nsec >= 1000000000 || 2984 timeout.tv_nsec < 0) 2985 return (EINVAL); 2986 ts = &timeout; 2987 } 2988 return do_wait(td, uap->obj, uap->val, ts, 0, 0); 2989} 2990 2991static int 2992__umtx_op_wait_uint(struct thread *td, struct _umtx_op_args *uap) 2993{ 2994 struct timespec *ts, timeout; 2995 int error; 2996 2997 if (uap->uaddr2 == NULL) 2998 ts = NULL; 2999 else { 3000 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 3001 if (error != 0) 3002 return (error); 3003 if (timeout.tv_nsec >= 1000000000 || 3004 timeout.tv_nsec < 0) 3005 return (EINVAL); 3006 ts = &timeout; 3007 } 3008 return do_wait(td, uap->obj, uap->val, ts, 1, 0); 3009} 3010 3011static int 3012__umtx_op_wait_uint_private(struct thread *td, struct _umtx_op_args *uap) 3013{ 3014 struct timespec *ts, timeout; 3015 int error; 3016 3017 if (uap->uaddr2 == NULL) 3018 ts = NULL; 3019 else { 3020 error = copyin(uap->uaddr2, &timeout, sizeof(timeout)); 3021 if (error != 0) 3022 return (error); 3023 if (timeout.tv_nsec >= 1000000000 || 3024 timeout.tv_nsec < 0) 3025 return (EINVAL); 3026 ts = &timeout; 3027 } 3028 return do_wait(td, uap->obj, uap->val, ts, 1, 1); 3029} 3030 3031static int 3032__umtx_op_wake(struct thread *td, struct _umtx_op_args *uap) 3033{ 3034 return (kern_umtx_wake(td, uap->obj, uap->val, 0)); 3035} 3036 3037static int 3038__umtx_op_wake_private(struct thread *td, struct _umtx_op_args *uap) 3039{ 3040 return (kern_umtx_wake(td, uap->obj, uap->val, 1)); 3041} 3042 3043static int 3044__umtx_op_lock_umutex(struct thread *td, struct _umtx_op_args *uap) 3045{ 3046 struct timespec *ts, timeout; 3047 int error; 3048 3049 /* Allow a null timespec (wait forever). */ 3050 if (uap->uaddr2 == NULL) 3051 ts = NULL; 3052 else { 3053 error = copyin(uap->uaddr2, &timeout, 3054 sizeof(timeout)); 3055 if (error != 0) 3056 return (error); 3057 if (timeout.tv_nsec >= 1000000000 || 3058 timeout.tv_nsec < 0) { 3059 return (EINVAL); 3060 } 3061 ts = &timeout; 3062 } 3063 return do_lock_umutex(td, uap->obj, ts, 0); 3064} 3065 3066static int 3067__umtx_op_trylock_umutex(struct thread *td, struct _umtx_op_args *uap) 3068{ 3069 return do_lock_umutex(td, uap->obj, NULL, _UMUTEX_TRY); 3070} 3071 3072static int 3073__umtx_op_wait_umutex(struct thread *td, struct _umtx_op_args *uap) 3074{ 3075 struct timespec *ts, timeout; 3076 int error; 3077 3078 /* Allow a null timespec (wait forever). */ 3079 if (uap->uaddr2 == NULL) 3080 ts = NULL; 3081 else { 3082 error = copyin(uap->uaddr2, &timeout, 3083 sizeof(timeout)); 3084 if (error != 0) 3085 return (error); 3086 if (timeout.tv_nsec >= 1000000000 || 3087 timeout.tv_nsec < 0) { 3088 return (EINVAL); 3089 } 3090 ts = &timeout; 3091 } 3092 return do_lock_umutex(td, uap->obj, ts, _UMUTEX_WAIT); 3093} 3094 3095static int 3096__umtx_op_wake_umutex(struct thread *td, struct _umtx_op_args *uap) 3097{ 3098 return do_wake_umutex(td, uap->obj); 3099} 3100 3101static int 3102__umtx_op_unlock_umutex(struct thread *td, struct _umtx_op_args *uap) 3103{ 3104 return do_unlock_umutex(td, uap->obj); 3105} 3106 3107static int 3108__umtx_op_set_ceiling(struct thread *td, struct _umtx_op_args *uap) 3109{ 3110 return do_set_ceiling(td, uap->obj, uap->val, uap->uaddr1); 3111} 3112 3113static int 3114__umtx_op_cv_wait(struct thread *td, struct _umtx_op_args *uap) 3115{ 3116 struct timespec *ts, timeout; 3117 int error; 3118 3119 /* Allow a null timespec (wait forever). */ 3120 if (uap->uaddr2 == NULL) 3121 ts = NULL; 3122 else { 3123 error = copyin(uap->uaddr2, &timeout, 3124 sizeof(timeout)); 3125 if (error != 0) 3126 return (error); 3127 if (timeout.tv_nsec >= 1000000000 || 3128 timeout.tv_nsec < 0) { 3129 return (EINVAL); 3130 } 3131 ts = &timeout; 3132 } 3133 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val)); 3134} 3135 3136static int 3137__umtx_op_cv_signal(struct thread *td, struct _umtx_op_args *uap) 3138{ 3139 return do_cv_signal(td, uap->obj); 3140} 3141 3142static int 3143__umtx_op_cv_broadcast(struct thread *td, struct _umtx_op_args *uap) 3144{ 3145 return do_cv_broadcast(td, uap->obj); 3146} 3147 3148static int 3149__umtx_op_rw_rdlock(struct thread *td, struct _umtx_op_args *uap) 3150{ 3151 struct timespec timeout; 3152 int error; 3153 3154 /* Allow a null timespec (wait forever). */ 3155 if (uap->uaddr2 == NULL) { 3156 error = do_rw_rdlock(td, uap->obj, uap->val, 0); 3157 } else { 3158 error = copyin(uap->uaddr2, &timeout, 3159 sizeof(timeout)); 3160 if (error != 0) 3161 return (error); 3162 if (timeout.tv_nsec >= 1000000000 || 3163 timeout.tv_nsec < 0) { 3164 return (EINVAL); 3165 } 3166 error = do_rw_rdlock2(td, uap->obj, uap->val, &timeout); 3167 } 3168 return (error); 3169} 3170 3171static int 3172__umtx_op_rw_wrlock(struct thread *td, struct _umtx_op_args *uap) 3173{ 3174 struct timespec timeout; 3175 int error; 3176 3177 /* Allow a null timespec (wait forever). */ 3178 if (uap->uaddr2 == NULL) { 3179 error = do_rw_wrlock(td, uap->obj, 0); 3180 } else { 3181 error = copyin(uap->uaddr2, &timeout, 3182 sizeof(timeout)); 3183 if (error != 0) 3184 return (error); 3185 if (timeout.tv_nsec >= 1000000000 || 3186 timeout.tv_nsec < 0) { 3187 return (EINVAL); 3188 } 3189 3190 error = do_rw_wrlock2(td, uap->obj, &timeout); 3191 } 3192 return (error); 3193} 3194 3195static int 3196__umtx_op_rw_unlock(struct thread *td, struct _umtx_op_args *uap) 3197{ 3198 return do_rw_unlock(td, uap->obj); 3199} 3200 3201static int 3202__umtx_op_sem_wait(struct thread *td, struct _umtx_op_args *uap) 3203{ 3204 struct timespec *ts, timeout; 3205 int error; 3206 3207 /* Allow a null timespec (wait forever). */ 3208 if (uap->uaddr2 == NULL) 3209 ts = NULL; 3210 else { 3211 error = copyin(uap->uaddr2, &timeout, 3212 sizeof(timeout)); 3213 if (error != 0) 3214 return (error); 3215 if (timeout.tv_nsec >= 1000000000 || 3216 timeout.tv_nsec < 0) { 3217 return (EINVAL); 3218 } 3219 ts = &timeout; 3220 } 3221 return (do_sem_wait(td, uap->obj, ts)); 3222} 3223 3224static int 3225__umtx_op_sem_wake(struct thread *td, struct _umtx_op_args *uap) 3226{ 3227 return do_sem_wake(td, uap->obj); 3228} 3229 3230typedef int (*_umtx_op_func)(struct thread *td, struct _umtx_op_args *uap); 3231 3232static _umtx_op_func op_table[] = { 3233 __umtx_op_lock_umtx, /* UMTX_OP_LOCK */ 3234 __umtx_op_unlock_umtx, /* UMTX_OP_UNLOCK */ 3235 __umtx_op_wait, /* UMTX_OP_WAIT */ 3236 __umtx_op_wake, /* UMTX_OP_WAKE */ 3237 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_TRYLOCK */ 3238 __umtx_op_lock_umutex, /* UMTX_OP_MUTEX_LOCK */ 3239 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */ 3240 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */ 3241 __umtx_op_cv_wait, /* UMTX_OP_CV_WAIT*/ 3242 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */ 3243 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */ 3244 __umtx_op_wait_uint, /* UMTX_OP_WAIT_UINT */ 3245 __umtx_op_rw_rdlock, /* UMTX_OP_RW_RDLOCK */ 3246 __umtx_op_rw_wrlock, /* UMTX_OP_RW_WRLOCK */ 3247 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */ 3248 __umtx_op_wait_uint_private, /* UMTX_OP_WAIT_UINT_PRIVATE */ 3249 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */ 3250 __umtx_op_wait_umutex, /* UMTX_OP_UMUTEX_WAIT */ 3251 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */ 3252 __umtx_op_sem_wait, /* UMTX_OP_SEM_WAIT */ 3253 __umtx_op_sem_wake /* UMTX_OP_SEM_WAKE */ 3254}; 3255 3256int 3257_umtx_op(struct thread *td, struct _umtx_op_args *uap) 3258{ 3259 if ((unsigned)uap->op < UMTX_OP_MAX) 3260 return (*op_table[uap->op])(td, uap); 3261 return (EINVAL); 3262} 3263 3264#ifdef COMPAT_IA32 3265int 3266freebsd32_umtx_lock(struct thread *td, struct freebsd32_umtx_lock_args *uap) 3267 /* struct umtx *umtx */ 3268{ 3269 return (do_lock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid, NULL)); 3270} 3271 3272int 3273freebsd32_umtx_unlock(struct thread *td, struct freebsd32_umtx_unlock_args *uap) 3274 /* struct umtx *umtx */ 3275{ 3276 return (do_unlock_umtx32(td, (uint32_t *)uap->umtx, td->td_tid)); 3277} 3278 3279struct timespec32 { 3280 u_int32_t tv_sec; 3281 u_int32_t tv_nsec; 3282}; 3283 3284static inline int 3285copyin_timeout32(void *addr, struct timespec *tsp) 3286{ 3287 struct timespec32 ts32; 3288 int error; 3289 3290 error = copyin(addr, &ts32, sizeof(struct timespec32)); 3291 if (error == 0) { 3292 tsp->tv_sec = ts32.tv_sec; 3293 tsp->tv_nsec = ts32.tv_nsec; 3294 } 3295 return (error); 3296} 3297 3298static int 3299__umtx_op_lock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap) 3300{ 3301 struct timespec *ts, timeout; 3302 int error; 3303 3304 /* Allow a null timespec (wait forever). */ 3305 if (uap->uaddr2 == NULL) 3306 ts = NULL; 3307 else { 3308 error = copyin_timeout32(uap->uaddr2, &timeout); 3309 if (error != 0) 3310 return (error); 3311 if (timeout.tv_nsec >= 1000000000 || 3312 timeout.tv_nsec < 0) { 3313 return (EINVAL); 3314 } 3315 ts = &timeout; 3316 } 3317 return (do_lock_umtx32(td, uap->obj, uap->val, ts)); 3318} 3319 3320static int 3321__umtx_op_unlock_umtx_compat32(struct thread *td, struct _umtx_op_args *uap) 3322{ 3323 return (do_unlock_umtx32(td, uap->obj, (uint32_t)uap->val)); 3324} 3325 3326static int 3327__umtx_op_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3328{ 3329 struct timespec *ts, timeout; 3330 int error; 3331 3332 if (uap->uaddr2 == NULL) 3333 ts = NULL; 3334 else { 3335 error = copyin_timeout32(uap->uaddr2, &timeout); 3336 if (error != 0) 3337 return (error); 3338 if (timeout.tv_nsec >= 1000000000 || 3339 timeout.tv_nsec < 0) 3340 return (EINVAL); 3341 ts = &timeout; 3342 } 3343 return do_wait(td, uap->obj, uap->val, ts, 1, 0); 3344} 3345 3346static int 3347__umtx_op_lock_umutex_compat32(struct thread *td, struct _umtx_op_args *uap) 3348{ 3349 struct timespec *ts, timeout; 3350 int error; 3351 3352 /* Allow a null timespec (wait forever). */ 3353 if (uap->uaddr2 == NULL) 3354 ts = NULL; 3355 else { 3356 error = copyin_timeout32(uap->uaddr2, &timeout); 3357 if (error != 0) 3358 return (error); 3359 if (timeout.tv_nsec >= 1000000000 || 3360 timeout.tv_nsec < 0) 3361 return (EINVAL); 3362 ts = &timeout; 3363 } 3364 return do_lock_umutex(td, uap->obj, ts, 0); 3365} 3366 3367static int 3368__umtx_op_wait_umutex_compat32(struct thread *td, struct _umtx_op_args *uap) 3369{ 3370 struct timespec *ts, timeout; 3371 int error; 3372 3373 /* Allow a null timespec (wait forever). */ 3374 if (uap->uaddr2 == NULL) 3375 ts = NULL; 3376 else { 3377 error = copyin_timeout32(uap->uaddr2, &timeout); 3378 if (error != 0) 3379 return (error); 3380 if (timeout.tv_nsec >= 1000000000 || 3381 timeout.tv_nsec < 0) 3382 return (EINVAL); 3383 ts = &timeout; 3384 } 3385 return do_lock_umutex(td, uap->obj, ts, _UMUTEX_WAIT); 3386} 3387 3388static int 3389__umtx_op_cv_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3390{ 3391 struct timespec *ts, timeout; 3392 int error; 3393 3394 /* Allow a null timespec (wait forever). */ 3395 if (uap->uaddr2 == NULL) 3396 ts = NULL; 3397 else { 3398 error = copyin_timeout32(uap->uaddr2, &timeout); 3399 if (error != 0) 3400 return (error); 3401 if (timeout.tv_nsec >= 1000000000 || 3402 timeout.tv_nsec < 0) 3403 return (EINVAL); 3404 ts = &timeout; 3405 } 3406 return (do_cv_wait(td, uap->obj, uap->uaddr1, ts, uap->val)); 3407} 3408 3409static int 3410__umtx_op_rw_rdlock_compat32(struct thread *td, struct _umtx_op_args *uap) 3411{ 3412 struct timespec timeout; 3413 int error; 3414 3415 /* Allow a null timespec (wait forever). */ 3416 if (uap->uaddr2 == NULL) { 3417 error = do_rw_rdlock(td, uap->obj, uap->val, 0); 3418 } else { 3419 error = copyin(uap->uaddr2, &timeout, 3420 sizeof(timeout)); 3421 if (error != 0) 3422 return (error); 3423 if (timeout.tv_nsec >= 1000000000 || 3424 timeout.tv_nsec < 0) { 3425 return (EINVAL); 3426 } 3427 error = do_rw_rdlock2(td, uap->obj, uap->val, &timeout); 3428 } 3429 return (error); 3430} 3431 3432static int 3433__umtx_op_rw_wrlock_compat32(struct thread *td, struct _umtx_op_args *uap) 3434{ 3435 struct timespec timeout; 3436 int error; 3437 3438 /* Allow a null timespec (wait forever). */ 3439 if (uap->uaddr2 == NULL) { 3440 error = do_rw_wrlock(td, uap->obj, 0); 3441 } else { 3442 error = copyin_timeout32(uap->uaddr2, &timeout); 3443 if (error != 0) 3444 return (error); 3445 if (timeout.tv_nsec >= 1000000000 || 3446 timeout.tv_nsec < 0) { 3447 return (EINVAL); 3448 } 3449 3450 error = do_rw_wrlock2(td, uap->obj, &timeout); 3451 } 3452 return (error); 3453} 3454 3455static int 3456__umtx_op_wait_uint_private_compat32(struct thread *td, struct _umtx_op_args *uap) 3457{ 3458 struct timespec *ts, timeout; 3459 int error; 3460 3461 if (uap->uaddr2 == NULL) 3462 ts = NULL; 3463 else { 3464 error = copyin_timeout32(uap->uaddr2, &timeout); 3465 if (error != 0) 3466 return (error); 3467 if (timeout.tv_nsec >= 1000000000 || 3468 timeout.tv_nsec < 0) 3469 return (EINVAL); 3470 ts = &timeout; 3471 } 3472 return do_wait(td, uap->obj, uap->val, ts, 1, 1); 3473} 3474 3475static int 3476__umtx_op_sem_wait_compat32(struct thread *td, struct _umtx_op_args *uap) 3477{ 3478 struct timespec *ts, timeout; 3479 int error; 3480 3481 /* Allow a null timespec (wait forever). */ 3482 if (uap->uaddr2 == NULL) 3483 ts = NULL; 3484 else { 3485 error = copyin_timeout32(uap->uaddr2, &timeout); 3486 if (error != 0) 3487 return (error); 3488 if (timeout.tv_nsec >= 1000000000 || 3489 timeout.tv_nsec < 0) 3490 return (EINVAL); 3491 ts = &timeout; 3492 } 3493 return (do_sem_wait(td, uap->obj, ts)); 3494} 3495 3496static _umtx_op_func op_table_compat32[] = { 3497 __umtx_op_lock_umtx_compat32, /* UMTX_OP_LOCK */ 3498 __umtx_op_unlock_umtx_compat32, /* UMTX_OP_UNLOCK */ 3499 __umtx_op_wait_compat32, /* UMTX_OP_WAIT */ 3500 __umtx_op_wake, /* UMTX_OP_WAKE */ 3501 __umtx_op_trylock_umutex, /* UMTX_OP_MUTEX_LOCK */ 3502 __umtx_op_lock_umutex_compat32, /* UMTX_OP_MUTEX_TRYLOCK */ 3503 __umtx_op_unlock_umutex, /* UMTX_OP_MUTEX_UNLOCK */ 3504 __umtx_op_set_ceiling, /* UMTX_OP_SET_CEILING */ 3505 __umtx_op_cv_wait_compat32, /* UMTX_OP_CV_WAIT*/ 3506 __umtx_op_cv_signal, /* UMTX_OP_CV_SIGNAL */ 3507 __umtx_op_cv_broadcast, /* UMTX_OP_CV_BROADCAST */ 3508 __umtx_op_wait_compat32, /* UMTX_OP_WAIT_UINT */ 3509 __umtx_op_rw_rdlock_compat32, /* UMTX_OP_RW_RDLOCK */ 3510 __umtx_op_rw_wrlock_compat32, /* UMTX_OP_RW_WRLOCK */ 3511 __umtx_op_rw_unlock, /* UMTX_OP_RW_UNLOCK */ 3512 __umtx_op_wait_uint_private_compat32, /* UMTX_OP_WAIT_UINT_PRIVATE */ 3513 __umtx_op_wake_private, /* UMTX_OP_WAKE_PRIVATE */ 3514 __umtx_op_wait_umutex_compat32, /* UMTX_OP_UMUTEX_WAIT */ 3515 __umtx_op_wake_umutex, /* UMTX_OP_UMUTEX_WAKE */ 3516 __umtx_op_sem_wait_compat32, /* UMTX_OP_SEM_WAIT */ 3517 __umtx_op_sem_wake /* UMTX_OP_SEM_WAKE */ 3518}; 3519 3520int 3521freebsd32_umtx_op(struct thread *td, struct freebsd32_umtx_op_args *uap) 3522{ 3523 if ((unsigned)uap->op < UMTX_OP_MAX) 3524 return (*op_table_compat32[uap->op])(td, 3525 (struct _umtx_op_args *)uap); 3526 return (EINVAL); 3527} 3528#endif 3529 3530void 3531umtx_thread_init(struct thread *td) 3532{ 3533 td->td_umtxq = umtxq_alloc(); 3534 td->td_umtxq->uq_thread = td; 3535} 3536 3537void 3538umtx_thread_fini(struct thread *td) 3539{ 3540 umtxq_free(td->td_umtxq); 3541} 3542 3543/* 3544 * It will be called when new thread is created, e.g fork(). 3545 */ 3546void 3547umtx_thread_alloc(struct thread *td) 3548{ 3549 struct umtx_q *uq; 3550 3551 uq = td->td_umtxq; 3552 uq->uq_inherited_pri = PRI_MAX; 3553 3554 KASSERT(uq->uq_flags == 0, ("uq_flags != 0")); 3555 KASSERT(uq->uq_thread == td, ("uq_thread != td")); 3556 KASSERT(uq->uq_pi_blocked == NULL, ("uq_pi_blocked != NULL")); 3557 KASSERT(TAILQ_EMPTY(&uq->uq_pi_contested), ("uq_pi_contested is not empty")); 3558} 3559 3560/* 3561 * exec() hook. 3562 */ 3563static void 3564umtx_exec_hook(void *arg __unused, struct proc *p __unused, 3565 struct image_params *imgp __unused) 3566{ 3567 umtx_thread_cleanup(curthread); 3568} 3569 3570/* 3571 * thread_exit() hook. 3572 */ 3573void 3574umtx_thread_exit(struct thread *td) 3575{ 3576 umtx_thread_cleanup(td); 3577} 3578 3579/* 3580 * clean up umtx data. 3581 */ 3582static void 3583umtx_thread_cleanup(struct thread *td) 3584{ 3585 struct umtx_q *uq; 3586 struct umtx_pi *pi; 3587 3588 if ((uq = td->td_umtxq) == NULL) 3589 return; 3590 3591 mtx_lock_spin(&umtx_lock); 3592 uq->uq_inherited_pri = PRI_MAX; 3593 while ((pi = TAILQ_FIRST(&uq->uq_pi_contested)) != NULL) { 3594 pi->pi_owner = NULL; 3595 TAILQ_REMOVE(&uq->uq_pi_contested, pi, pi_link); 3596 } 3597 thread_lock(td); 3598 td->td_flags &= ~TDF_UBORROWING; 3599 thread_unlock(td); 3600 mtx_unlock_spin(&umtx_lock); 3601} 3602