1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2008 Attilio Rao <attilio@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice(s), this list of conditions and the following disclaimer as 12 * the first lines of this file unmodified other than the possible 13 * addition of one or more copyright notices. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice(s), this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY 19 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 20 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 21 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY 22 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 24 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 25 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH 28 * DAMAGE. 29 */ 30 31 #include "opt_ddb.h" 32 #include "opt_hwpmc_hooks.h" 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include <sys/param.h> 38 #include <sys/kdb.h> 39 #include <sys/ktr.h> 40 #include <sys/limits.h> 41 #include <sys/lock.h> 42 #include <sys/lock_profile.h> 43 #include <sys/lockmgr.h> 44 #include <sys/lockstat.h> 45 #include <sys/mutex.h> 46 #include <sys/proc.h> 47 #include <sys/sleepqueue.h> 48 #ifdef DEBUG_LOCKS 49 #include <sys/stack.h> 50 #endif 51 #include <sys/sysctl.h> 52 #include <sys/systm.h> 53 54 #include <machine/cpu.h> 55 56 #ifdef DDB 57 #include <ddb/ddb.h> 58 #endif 59 60 #ifdef HWPMC_HOOKS 61 #include <sys/pmckern.h> 62 PMC_SOFT_DECLARE( , , lock, failed); 63 #endif 64 65 /* 66 * Hack. There should be prio_t or similar so that this is not necessary. 67 */ 68 _Static_assert((PRILASTFLAG * 2) - 1 <= USHRT_MAX, 69 "prio flags wont fit in u_short pri in struct lock"); 70 71 CTASSERT(LK_UNLOCKED == (LK_UNLOCKED & 72 ~(LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS))); 73 74 #define SQ_EXCLUSIVE_QUEUE 0 75 #define SQ_SHARED_QUEUE 1 76 77 #ifndef INVARIANTS 78 #define _lockmgr_assert(lk, what, file, line) 79 #endif 80 81 #define TD_SLOCKS_INC(td) ((td)->td_lk_slocks++) 82 #define TD_SLOCKS_DEC(td) ((td)->td_lk_slocks--) 83 84 #ifndef DEBUG_LOCKS 85 #define STACK_PRINT(lk) 86 #define STACK_SAVE(lk) 87 #define STACK_ZERO(lk) 88 #else 89 #define STACK_PRINT(lk) stack_print_ddb(&(lk)->lk_stack) 90 #define STACK_SAVE(lk) stack_save(&(lk)->lk_stack) 91 #define STACK_ZERO(lk) stack_zero(&(lk)->lk_stack) 92 #endif 93 94 #define LOCK_LOG2(lk, string, arg1, arg2) \ 95 if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \ 96 CTR2(KTR_LOCK, (string), (arg1), (arg2)) 97 #define LOCK_LOG3(lk, string, arg1, arg2, arg3) \ 98 if (LOCK_LOG_TEST(&(lk)->lock_object, 0)) \ 99 CTR3(KTR_LOCK, (string), (arg1), (arg2), (arg3)) 100 101 #define GIANT_DECLARE \ 102 int _i = 0; \ 103 WITNESS_SAVE_DECL(Giant) 104 #define GIANT_RESTORE() do { \ 105 if (__predict_false(_i > 0)) { \ 106 while (_i--) \ 107 mtx_lock(&Giant); \ 108 WITNESS_RESTORE(&Giant.lock_object, Giant); \ 109 } \ 110 } while (0) 111 #define GIANT_SAVE() do { \ 112 if (__predict_false(mtx_owned(&Giant))) { \ 113 WITNESS_SAVE(&Giant.lock_object, Giant); \ 114 while (mtx_owned(&Giant)) { \ 115 _i++; \ 116 mtx_unlock(&Giant); \ 117 } \ 118 } \ 119 } while (0) 120 121 static bool __always_inline 122 LK_CAN_SHARE(uintptr_t x, int flags, bool fp) 123 { 124 125 if ((x & (LK_SHARE | LK_EXCLUSIVE_WAITERS | LK_EXCLUSIVE_SPINNERS)) == 126 LK_SHARE) 127 return (true); 128 if (fp || (!(x & LK_SHARE))) 129 return (false); 130 if ((curthread->td_lk_slocks != 0 && !(flags & LK_NODDLKTREAT)) || 131 (curthread->td_pflags & TDP_DEADLKTREAT)) 132 return (true); 133 return (false); 134 } 135 136 #define LK_TRYOP(x) \ 137 ((x) & LK_NOWAIT) 138 139 #define LK_CAN_WITNESS(x) \ 140 (((x) & LK_NOWITNESS) == 0 && !LK_TRYOP(x)) 141 #define LK_TRYWIT(x) \ 142 (LK_TRYOP(x) ? LOP_TRYLOCK : 0) 143 144 #define lockmgr_disowned(lk) \ 145 (((lk)->lk_lock & ~(LK_FLAGMASK & ~LK_SHARE)) == LK_KERNPROC) 146 147 #define lockmgr_xlocked_v(v) \ 148 (((v) & ~(LK_FLAGMASK & ~LK_SHARE)) == (uintptr_t)curthread) 149 150 #define lockmgr_xlocked(lk) lockmgr_xlocked_v(lockmgr_read_value(lk)) 151 152 static void assert_lockmgr(const struct lock_object *lock, int how); 153 #ifdef DDB 154 static void db_show_lockmgr(const struct lock_object *lock); 155 #endif 156 static void lock_lockmgr(struct lock_object *lock, uintptr_t how); 157 #ifdef KDTRACE_HOOKS 158 static int owner_lockmgr(const struct lock_object *lock, 159 struct thread **owner); 160 #endif 161 static uintptr_t unlock_lockmgr(struct lock_object *lock); 162 163 struct lock_class lock_class_lockmgr = { 164 .lc_name = "lockmgr", 165 .lc_flags = LC_RECURSABLE | LC_SLEEPABLE | LC_SLEEPLOCK | LC_UPGRADABLE, 166 .lc_assert = assert_lockmgr, 167 #ifdef DDB 168 .lc_ddb_show = db_show_lockmgr, 169 #endif 170 .lc_lock = lock_lockmgr, 171 .lc_unlock = unlock_lockmgr, 172 #ifdef KDTRACE_HOOKS 173 .lc_owner = owner_lockmgr, 174 #endif 175 }; 176 177 static __read_mostly bool lk_adaptive = true; 178 static SYSCTL_NODE(_debug, OID_AUTO, lockmgr, CTLFLAG_RD, NULL, "lockmgr debugging"); 179 SYSCTL_BOOL(_debug_lockmgr, OID_AUTO, adaptive_spinning, CTLFLAG_RW, &lk_adaptive, 180 0, ""); 181 #define lockmgr_delay locks_delay 182 183 struct lockmgr_wait { 184 const char *iwmesg; 185 int ipri; 186 int itimo; 187 }; 188 189 static bool __always_inline lockmgr_slock_try(struct lock *lk, uintptr_t *xp, 190 int flags, bool fp); 191 static bool __always_inline lockmgr_sunlock_try(struct lock *lk, uintptr_t *xp); 192 193 static void 194 lockmgr_exit(u_int flags, struct lock_object *ilk, int wakeup_swapper) 195 { 196 struct lock_class *class; 197 198 if (flags & LK_INTERLOCK) { 199 class = LOCK_CLASS(ilk); 200 class->lc_unlock(ilk); 201 } 202 203 if (__predict_false(wakeup_swapper)) 204 kick_proc0(); 205 } 206 207 static void 208 lockmgr_note_shared_acquire(struct lock *lk, int contested, 209 uint64_t waittime, const char *file, int line, int flags) 210 { 211 212 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(lockmgr__acquire, lk, contested, 213 waittime, file, line, LOCKSTAT_READER); 214 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, file, line); 215 WITNESS_LOCK(&lk->lock_object, LK_TRYWIT(flags), file, line); 216 TD_LOCKS_INC(curthread); 217 TD_SLOCKS_INC(curthread); 218 STACK_SAVE(lk); 219 } 220 221 static void 222 lockmgr_note_shared_release(struct lock *lk, const char *file, int line) 223 { 224 225 WITNESS_UNLOCK(&lk->lock_object, 0, file, line); 226 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, file, line); 227 TD_LOCKS_DEC(curthread); 228 TD_SLOCKS_DEC(curthread); 229 } 230 231 static void 232 lockmgr_note_exclusive_acquire(struct lock *lk, int contested, 233 uint64_t waittime, const char *file, int line, int flags) 234 { 235 236 LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(lockmgr__acquire, lk, contested, 237 waittime, file, line, LOCKSTAT_WRITER); 238 LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, lk->lk_recurse, file, line); 239 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | LK_TRYWIT(flags), file, 240 line); 241 TD_LOCKS_INC(curthread); 242 STACK_SAVE(lk); 243 } 244 245 static void 246 lockmgr_note_exclusive_release(struct lock *lk, const char *file, int line) 247 { 248 249 if (LK_HOLDER(lockmgr_read_value(lk)) != LK_KERNPROC) { 250 WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line); 251 TD_LOCKS_DEC(curthread); 252 } 253 LOCK_LOG_LOCK("XUNLOCK", &lk->lock_object, 0, lk->lk_recurse, file, 254 line); 255 } 256 257 static __inline struct thread * 258 lockmgr_xholder(const struct lock *lk) 259 { 260 uintptr_t x; 261 262 x = lockmgr_read_value(lk); 263 return ((x & LK_SHARE) ? NULL : (struct thread *)LK_HOLDER(x)); 264 } 265 266 /* 267 * It assumes sleepq_lock held and returns with this one unheld. 268 * It also assumes the generic interlock is sane and previously checked. 269 * If LK_INTERLOCK is specified the interlock is not reacquired after the 270 * sleep. 271 */ 272 static __inline int 273 sleeplk(struct lock *lk, u_int flags, struct lock_object *ilk, 274 const char *wmesg, int pri, int timo, int queue) 275 { 276 GIANT_DECLARE; 277 struct lock_class *class; 278 int catch, error; 279 280 class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; 281 catch = pri & PCATCH; 282 pri &= PRIMASK; 283 error = 0; 284 285 LOCK_LOG3(lk, "%s: %p blocking on the %s sleepqueue", __func__, lk, 286 (queue == SQ_EXCLUSIVE_QUEUE) ? "exclusive" : "shared"); 287 288 if (flags & LK_INTERLOCK) 289 class->lc_unlock(ilk); 290 if (queue == SQ_EXCLUSIVE_QUEUE && (flags & LK_SLEEPFAIL) != 0) { 291 if (lk->lk_exslpfail < USHRT_MAX) 292 lk->lk_exslpfail++; 293 } 294 GIANT_SAVE(); 295 sleepq_add(&lk->lock_object, NULL, wmesg, SLEEPQ_LK | (catch ? 296 SLEEPQ_INTERRUPTIBLE : 0), queue); 297 if ((flags & LK_TIMELOCK) && timo) 298 sleepq_set_timeout(&lk->lock_object, timo); 299 300 /* 301 * Decisional switch for real sleeping. 302 */ 303 if ((flags & LK_TIMELOCK) && timo && catch) 304 error = sleepq_timedwait_sig(&lk->lock_object, pri); 305 else if ((flags & LK_TIMELOCK) && timo) 306 error = sleepq_timedwait(&lk->lock_object, pri); 307 else if (catch) 308 error = sleepq_wait_sig(&lk->lock_object, pri); 309 else 310 sleepq_wait(&lk->lock_object, pri); 311 GIANT_RESTORE(); 312 if ((flags & LK_SLEEPFAIL) && error == 0) 313 error = ENOLCK; 314 315 return (error); 316 } 317 318 static __inline int 319 wakeupshlk(struct lock *lk, const char *file, int line) 320 { 321 uintptr_t v, x, orig_x; 322 u_int realexslp; 323 int queue, wakeup_swapper; 324 325 wakeup_swapper = 0; 326 for (;;) { 327 x = lockmgr_read_value(lk); 328 if (lockmgr_sunlock_try(lk, &x)) 329 break; 330 331 /* 332 * We should have a sharer with waiters, so enter the hard 333 * path in order to handle wakeups correctly. 334 */ 335 sleepq_lock(&lk->lock_object); 336 orig_x = lockmgr_read_value(lk); 337 retry_sleepq: 338 x = orig_x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 339 v = LK_UNLOCKED; 340 341 /* 342 * If the lock has exclusive waiters, give them preference in 343 * order to avoid deadlock with shared runners up. 344 * If interruptible sleeps left the exclusive queue empty 345 * avoid a starvation for the threads sleeping on the shared 346 * queue by giving them precedence and cleaning up the 347 * exclusive waiters bit anyway. 348 * Please note that lk_exslpfail count may be lying about 349 * the real number of waiters with the LK_SLEEPFAIL flag on 350 * because they may be used in conjunction with interruptible 351 * sleeps so lk_exslpfail might be considered an 'upper limit' 352 * bound, including the edge cases. 353 */ 354 realexslp = sleepq_sleepcnt(&lk->lock_object, 355 SQ_EXCLUSIVE_QUEUE); 356 if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { 357 if (lk->lk_exslpfail != USHRT_MAX && lk->lk_exslpfail < realexslp) { 358 lk->lk_exslpfail = 0; 359 queue = SQ_EXCLUSIVE_QUEUE; 360 v |= (x & LK_SHARED_WAITERS); 361 } else { 362 lk->lk_exslpfail = 0; 363 LOCK_LOG2(lk, 364 "%s: %p has only LK_SLEEPFAIL sleepers", 365 __func__, lk); 366 LOCK_LOG2(lk, 367 "%s: %p waking up threads on the exclusive queue", 368 __func__, lk); 369 wakeup_swapper = 370 sleepq_broadcast(&lk->lock_object, 371 SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); 372 queue = SQ_SHARED_QUEUE; 373 } 374 } else { 375 /* 376 * Exclusive waiters sleeping with LK_SLEEPFAIL on 377 * and using interruptible sleeps/timeout may have 378 * left spourious lk_exslpfail counts on, so clean 379 * it up anyway. 380 */ 381 lk->lk_exslpfail = 0; 382 queue = SQ_SHARED_QUEUE; 383 } 384 385 if (lockmgr_sunlock_try(lk, &orig_x)) { 386 sleepq_release(&lk->lock_object); 387 break; 388 } 389 390 x |= LK_SHARERS_LOCK(1); 391 if (!atomic_fcmpset_rel_ptr(&lk->lk_lock, &x, v)) { 392 orig_x = x; 393 goto retry_sleepq; 394 } 395 LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue", 396 __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : 397 "exclusive"); 398 wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 399 0, queue); 400 sleepq_release(&lk->lock_object); 401 break; 402 } 403 404 LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, LOCKSTAT_READER); 405 return (wakeup_swapper); 406 } 407 408 static void 409 assert_lockmgr(const struct lock_object *lock, int what) 410 { 411 412 panic("lockmgr locks do not support assertions"); 413 } 414 415 static void 416 lock_lockmgr(struct lock_object *lock, uintptr_t how) 417 { 418 419 panic("lockmgr locks do not support sleep interlocking"); 420 } 421 422 static uintptr_t 423 unlock_lockmgr(struct lock_object *lock) 424 { 425 426 panic("lockmgr locks do not support sleep interlocking"); 427 } 428 429 #ifdef KDTRACE_HOOKS 430 static int 431 owner_lockmgr(const struct lock_object *lock, struct thread **owner) 432 { 433 434 panic("lockmgr locks do not support owner inquiring"); 435 } 436 #endif 437 438 void 439 lockinit(struct lock *lk, int pri, const char *wmesg, int timo, int flags) 440 { 441 int iflags; 442 443 MPASS((flags & ~LK_INIT_MASK) == 0); 444 ASSERT_ATOMIC_LOAD_PTR(lk->lk_lock, 445 ("%s: lockmgr not aligned for %s: %p", __func__, wmesg, 446 &lk->lk_lock)); 447 448 iflags = LO_SLEEPABLE | LO_UPGRADABLE; 449 if (flags & LK_CANRECURSE) 450 iflags |= LO_RECURSABLE; 451 if ((flags & LK_NODUP) == 0) 452 iflags |= LO_DUPOK; 453 if (flags & LK_NOPROFILE) 454 iflags |= LO_NOPROFILE; 455 if ((flags & LK_NOWITNESS) == 0) 456 iflags |= LO_WITNESS; 457 if (flags & LK_QUIET) 458 iflags |= LO_QUIET; 459 if (flags & LK_IS_VNODE) 460 iflags |= LO_IS_VNODE; 461 if (flags & LK_NEW) 462 iflags |= LO_NEW; 463 iflags |= flags & LK_NOSHARE; 464 465 lock_init(&lk->lock_object, &lock_class_lockmgr, wmesg, NULL, iflags); 466 lk->lk_lock = LK_UNLOCKED; 467 lk->lk_recurse = 0; 468 lk->lk_exslpfail = 0; 469 lk->lk_timo = timo; 470 lk->lk_pri = pri; 471 STACK_ZERO(lk); 472 } 473 474 /* 475 * XXX: Gross hacks to manipulate external lock flags after 476 * initialization. Used for certain vnode and buf locks. 477 */ 478 void 479 lockallowshare(struct lock *lk) 480 { 481 482 lockmgr_assert(lk, KA_XLOCKED); 483 lk->lock_object.lo_flags &= ~LK_NOSHARE; 484 } 485 486 void 487 lockdisableshare(struct lock *lk) 488 { 489 490 lockmgr_assert(lk, KA_XLOCKED); 491 lk->lock_object.lo_flags |= LK_NOSHARE; 492 } 493 494 void 495 lockallowrecurse(struct lock *lk) 496 { 497 498 lockmgr_assert(lk, KA_XLOCKED); 499 lk->lock_object.lo_flags |= LO_RECURSABLE; 500 } 501 502 void 503 lockdisablerecurse(struct lock *lk) 504 { 505 506 lockmgr_assert(lk, KA_XLOCKED); 507 lk->lock_object.lo_flags &= ~LO_RECURSABLE; 508 } 509 510 void 511 lockdestroy(struct lock *lk) 512 { 513 514 KASSERT(lk->lk_lock == LK_UNLOCKED, ("lockmgr still held")); 515 KASSERT(lk->lk_recurse == 0, ("lockmgr still recursed")); 516 KASSERT(lk->lk_exslpfail == 0, ("lockmgr still exclusive waiters")); 517 lock_destroy(&lk->lock_object); 518 } 519 520 static bool __always_inline 521 lockmgr_slock_try(struct lock *lk, uintptr_t *xp, int flags, bool fp) 522 { 523 524 /* 525 * If no other thread has an exclusive lock, or 526 * no exclusive waiter is present, bump the count of 527 * sharers. Since we have to preserve the state of 528 * waiters, if we fail to acquire the shared lock 529 * loop back and retry. 530 */ 531 while (LK_CAN_SHARE(*xp, flags, fp)) { 532 if (atomic_fcmpset_acq_ptr(&lk->lk_lock, xp, 533 *xp + LK_ONE_SHARER)) { 534 return (true); 535 } 536 } 537 return (false); 538 } 539 540 static bool __always_inline 541 lockmgr_sunlock_try(struct lock *lk, uintptr_t *xp) 542 { 543 544 for (;;) { 545 if (LK_SHARERS(*xp) > 1 || !(*xp & LK_ALL_WAITERS)) { 546 if (atomic_fcmpset_rel_ptr(&lk->lk_lock, xp, 547 *xp - LK_ONE_SHARER)) 548 return (true); 549 continue; 550 } 551 break; 552 } 553 return (false); 554 } 555 556 static bool 557 lockmgr_slock_adaptive(struct lock_delay_arg *lda, struct lock *lk, uintptr_t *xp, 558 int flags) 559 { 560 struct thread *owner; 561 uintptr_t x; 562 563 x = *xp; 564 MPASS(x != LK_UNLOCKED); 565 owner = (struct thread *)LK_HOLDER(x); 566 for (;;) { 567 MPASS(owner != curthread); 568 if (owner == (struct thread *)LK_KERNPROC) 569 return (false); 570 if ((x & LK_SHARE) && LK_SHARERS(x) > 0) 571 return (false); 572 if (owner == NULL) 573 return (false); 574 if (!TD_IS_RUNNING(owner)) 575 return (false); 576 if ((x & LK_ALL_WAITERS) != 0) 577 return (false); 578 lock_delay(lda); 579 x = lockmgr_read_value(lk); 580 if (LK_CAN_SHARE(x, flags, false)) { 581 *xp = x; 582 return (true); 583 } 584 owner = (struct thread *)LK_HOLDER(x); 585 } 586 } 587 588 static __noinline int 589 lockmgr_slock_hard(struct lock *lk, u_int flags, struct lock_object *ilk, 590 const char *file, int line, struct lockmgr_wait *lwa) 591 { 592 uintptr_t tid, x; 593 int error = 0; 594 const char *iwmesg; 595 int ipri, itimo; 596 597 #ifdef KDTRACE_HOOKS 598 uint64_t sleep_time = 0; 599 #endif 600 #ifdef LOCK_PROFILING 601 uint64_t waittime = 0; 602 int contested = 0; 603 #endif 604 struct lock_delay_arg lda; 605 606 if (SCHEDULER_STOPPED()) 607 goto out; 608 609 tid = (uintptr_t)curthread; 610 611 if (LK_CAN_WITNESS(flags)) 612 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, 613 file, line, flags & LK_INTERLOCK ? ilk : NULL); 614 x = lockmgr_read_value(lk); 615 lock_delay_arg_init(&lda, &lockmgr_delay); 616 if (!lk_adaptive) 617 flags &= ~LK_ADAPTIVE; 618 /* 619 * The lock may already be locked exclusive by curthread, 620 * avoid deadlock. 621 */ 622 if (LK_HOLDER(x) == tid) { 623 LOCK_LOG2(lk, 624 "%s: %p already held in exclusive mode", 625 __func__, lk); 626 error = EDEADLK; 627 goto out; 628 } 629 630 for (;;) { 631 if (lockmgr_slock_try(lk, &x, flags, false)) 632 break; 633 634 lock_profile_obtain_lock_failed(&lk->lock_object, false, 635 &contested, &waittime); 636 637 if ((flags & (LK_ADAPTIVE | LK_INTERLOCK)) == LK_ADAPTIVE) { 638 if (lockmgr_slock_adaptive(&lda, lk, &x, flags)) 639 continue; 640 } 641 642 #ifdef HWPMC_HOOKS 643 PMC_SOFT_CALL( , , lock, failed); 644 #endif 645 646 /* 647 * If the lock is expected to not sleep just give up 648 * and return. 649 */ 650 if (LK_TRYOP(flags)) { 651 LOCK_LOG2(lk, "%s: %p fails the try operation", 652 __func__, lk); 653 error = EBUSY; 654 break; 655 } 656 657 /* 658 * Acquire the sleepqueue chain lock because we 659 * probabilly will need to manipulate waiters flags. 660 */ 661 sleepq_lock(&lk->lock_object); 662 x = lockmgr_read_value(lk); 663 retry_sleepq: 664 665 /* 666 * if the lock can be acquired in shared mode, try 667 * again. 668 */ 669 if (LK_CAN_SHARE(x, flags, false)) { 670 sleepq_release(&lk->lock_object); 671 continue; 672 } 673 674 /* 675 * Try to set the LK_SHARED_WAITERS flag. If we fail, 676 * loop back and retry. 677 */ 678 if ((x & LK_SHARED_WAITERS) == 0) { 679 if (!atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, 680 x | LK_SHARED_WAITERS)) { 681 goto retry_sleepq; 682 } 683 LOCK_LOG2(lk, "%s: %p set shared waiters flag", 684 __func__, lk); 685 } 686 687 if (lwa == NULL) { 688 iwmesg = lk->lock_object.lo_name; 689 ipri = lk->lk_pri; 690 itimo = lk->lk_timo; 691 } else { 692 iwmesg = lwa->iwmesg; 693 ipri = lwa->ipri; 694 itimo = lwa->itimo; 695 } 696 697 /* 698 * As far as we have been unable to acquire the 699 * shared lock and the shared waiters flag is set, 700 * we will sleep. 701 */ 702 #ifdef KDTRACE_HOOKS 703 sleep_time -= lockstat_nsecs(&lk->lock_object); 704 #endif 705 error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, 706 SQ_SHARED_QUEUE); 707 #ifdef KDTRACE_HOOKS 708 sleep_time += lockstat_nsecs(&lk->lock_object); 709 #endif 710 flags &= ~LK_INTERLOCK; 711 if (error) { 712 LOCK_LOG3(lk, 713 "%s: interrupted sleep for %p with %d", 714 __func__, lk, error); 715 break; 716 } 717 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 718 __func__, lk); 719 x = lockmgr_read_value(lk); 720 } 721 if (error == 0) { 722 #ifdef KDTRACE_HOOKS 723 if (sleep_time != 0) 724 LOCKSTAT_RECORD4(lockmgr__block, lk, sleep_time, 725 LOCKSTAT_READER, (x & LK_SHARE) == 0, 726 (x & LK_SHARE) == 0 ? 0 : LK_SHARERS(x)); 727 #endif 728 #ifdef LOCK_PROFILING 729 lockmgr_note_shared_acquire(lk, contested, waittime, 730 file, line, flags); 731 #else 732 lockmgr_note_shared_acquire(lk, 0, 0, file, line, 733 flags); 734 #endif 735 } 736 737 out: 738 lockmgr_exit(flags, ilk, 0); 739 return (error); 740 } 741 742 static bool 743 lockmgr_xlock_adaptive(struct lock_delay_arg *lda, struct lock *lk, uintptr_t *xp) 744 { 745 struct thread *owner; 746 uintptr_t x; 747 748 x = *xp; 749 MPASS(x != LK_UNLOCKED); 750 owner = (struct thread *)LK_HOLDER(x); 751 for (;;) { 752 MPASS(owner != curthread); 753 if (owner == NULL) 754 return (false); 755 if ((x & LK_SHARE) && LK_SHARERS(x) > 0) 756 return (false); 757 if (owner == (struct thread *)LK_KERNPROC) 758 return (false); 759 if (!TD_IS_RUNNING(owner)) 760 return (false); 761 if ((x & LK_ALL_WAITERS) != 0) 762 return (false); 763 lock_delay(lda); 764 x = lockmgr_read_value(lk); 765 if (x == LK_UNLOCKED) { 766 *xp = x; 767 return (true); 768 } 769 owner = (struct thread *)LK_HOLDER(x); 770 } 771 } 772 773 static __noinline int 774 lockmgr_xlock_hard(struct lock *lk, u_int flags, struct lock_object *ilk, 775 const char *file, int line, struct lockmgr_wait *lwa) 776 { 777 struct lock_class *class; 778 uintptr_t tid, x, v; 779 int error = 0; 780 const char *iwmesg; 781 int ipri, itimo; 782 783 #ifdef KDTRACE_HOOKS 784 uint64_t sleep_time = 0; 785 #endif 786 #ifdef LOCK_PROFILING 787 uint64_t waittime = 0; 788 int contested = 0; 789 #endif 790 struct lock_delay_arg lda; 791 792 if (SCHEDULER_STOPPED()) 793 goto out; 794 795 tid = (uintptr_t)curthread; 796 797 if (LK_CAN_WITNESS(flags)) 798 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 799 LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? 800 ilk : NULL); 801 802 /* 803 * If curthread already holds the lock and this one is 804 * allowed to recurse, simply recurse on it. 805 */ 806 if (lockmgr_xlocked(lk)) { 807 if ((flags & LK_CANRECURSE) == 0 && 808 (lk->lock_object.lo_flags & LO_RECURSABLE) == 0) { 809 /* 810 * If the lock is expected to not panic just 811 * give up and return. 812 */ 813 if (LK_TRYOP(flags)) { 814 LOCK_LOG2(lk, 815 "%s: %p fails the try operation", 816 __func__, lk); 817 error = EBUSY; 818 goto out; 819 } 820 if (flags & LK_INTERLOCK) { 821 class = LOCK_CLASS(ilk); 822 class->lc_unlock(ilk); 823 } 824 STACK_PRINT(lk); 825 panic("%s: recursing on non recursive lockmgr %p " 826 "@ %s:%d\n", __func__, lk, file, line); 827 } 828 atomic_set_ptr(&lk->lk_lock, LK_WRITER_RECURSED); 829 lk->lk_recurse++; 830 LOCK_LOG2(lk, "%s: %p recursing", __func__, lk); 831 LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, 832 lk->lk_recurse, file, line); 833 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | 834 LK_TRYWIT(flags), file, line); 835 TD_LOCKS_INC(curthread); 836 goto out; 837 } 838 839 x = LK_UNLOCKED; 840 lock_delay_arg_init(&lda, &lockmgr_delay); 841 if (!lk_adaptive) 842 flags &= ~LK_ADAPTIVE; 843 for (;;) { 844 if (x == LK_UNLOCKED) { 845 if (atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, tid)) 846 break; 847 continue; 848 } 849 850 lock_profile_obtain_lock_failed(&lk->lock_object, false, 851 &contested, &waittime); 852 853 if ((flags & (LK_ADAPTIVE | LK_INTERLOCK)) == LK_ADAPTIVE) { 854 if (lockmgr_xlock_adaptive(&lda, lk, &x)) 855 continue; 856 } 857 #ifdef HWPMC_HOOKS 858 PMC_SOFT_CALL( , , lock, failed); 859 #endif 860 861 /* 862 * If the lock is expected to not sleep just give up 863 * and return. 864 */ 865 if (LK_TRYOP(flags)) { 866 LOCK_LOG2(lk, "%s: %p fails the try operation", 867 __func__, lk); 868 error = EBUSY; 869 break; 870 } 871 872 /* 873 * Acquire the sleepqueue chain lock because we 874 * probabilly will need to manipulate waiters flags. 875 */ 876 sleepq_lock(&lk->lock_object); 877 x = lockmgr_read_value(lk); 878 retry_sleepq: 879 880 /* 881 * if the lock has been released while we spun on 882 * the sleepqueue chain lock just try again. 883 */ 884 if (x == LK_UNLOCKED) { 885 sleepq_release(&lk->lock_object); 886 continue; 887 } 888 889 /* 890 * The lock can be in the state where there is a 891 * pending queue of waiters, but still no owner. 892 * This happens when the lock is contested and an 893 * owner is going to claim the lock. 894 * If curthread is the one successfully acquiring it 895 * claim lock ownership and return, preserving waiters 896 * flags. 897 */ 898 v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 899 if ((x & ~v) == LK_UNLOCKED) { 900 v &= ~LK_EXCLUSIVE_SPINNERS; 901 if (atomic_fcmpset_acq_ptr(&lk->lk_lock, &x, 902 tid | v)) { 903 sleepq_release(&lk->lock_object); 904 LOCK_LOG2(lk, 905 "%s: %p claimed by a new writer", 906 __func__, lk); 907 break; 908 } 909 goto retry_sleepq; 910 } 911 912 /* 913 * Try to set the LK_EXCLUSIVE_WAITERS flag. If we 914 * fail, loop back and retry. 915 */ 916 if ((x & LK_EXCLUSIVE_WAITERS) == 0) { 917 if (!atomic_fcmpset_ptr(&lk->lk_lock, &x, 918 x | LK_EXCLUSIVE_WAITERS)) { 919 goto retry_sleepq; 920 } 921 LOCK_LOG2(lk, "%s: %p set excl waiters flag", 922 __func__, lk); 923 } 924 925 if (lwa == NULL) { 926 iwmesg = lk->lock_object.lo_name; 927 ipri = lk->lk_pri; 928 itimo = lk->lk_timo; 929 } else { 930 iwmesg = lwa->iwmesg; 931 ipri = lwa->ipri; 932 itimo = lwa->itimo; 933 } 934 935 /* 936 * As far as we have been unable to acquire the 937 * exclusive lock and the exclusive waiters flag 938 * is set, we will sleep. 939 */ 940 #ifdef KDTRACE_HOOKS 941 sleep_time -= lockstat_nsecs(&lk->lock_object); 942 #endif 943 error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, 944 SQ_EXCLUSIVE_QUEUE); 945 #ifdef KDTRACE_HOOKS 946 sleep_time += lockstat_nsecs(&lk->lock_object); 947 #endif 948 flags &= ~LK_INTERLOCK; 949 if (error) { 950 LOCK_LOG3(lk, 951 "%s: interrupted sleep for %p with %d", 952 __func__, lk, error); 953 break; 954 } 955 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 956 __func__, lk); 957 x = lockmgr_read_value(lk); 958 } 959 if (error == 0) { 960 #ifdef KDTRACE_HOOKS 961 if (sleep_time != 0) 962 LOCKSTAT_RECORD4(lockmgr__block, lk, sleep_time, 963 LOCKSTAT_WRITER, (x & LK_SHARE) == 0, 964 (x & LK_SHARE) == 0 ? 0 : LK_SHARERS(x)); 965 #endif 966 #ifdef LOCK_PROFILING 967 lockmgr_note_exclusive_acquire(lk, contested, waittime, 968 file, line, flags); 969 #else 970 lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, 971 flags); 972 #endif 973 } 974 975 out: 976 lockmgr_exit(flags, ilk, 0); 977 return (error); 978 } 979 980 static __noinline int 981 lockmgr_upgrade(struct lock *lk, u_int flags, struct lock_object *ilk, 982 const char *file, int line, struct lockmgr_wait *lwa) 983 { 984 uintptr_t tid, v, setv; 985 int error = 0; 986 int op; 987 988 if (SCHEDULER_STOPPED()) 989 goto out; 990 991 tid = (uintptr_t)curthread; 992 993 _lockmgr_assert(lk, KA_SLOCKED, file, line); 994 995 op = flags & LK_TYPE_MASK; 996 v = lockmgr_read_value(lk); 997 for (;;) { 998 if (LK_SHARERS(v) > 1) { 999 if (op == LK_TRYUPGRADE) { 1000 LOCK_LOG2(lk, "%s: %p failed the nowait upgrade", 1001 __func__, lk); 1002 error = EBUSY; 1003 goto out; 1004 } 1005 if (atomic_fcmpset_rel_ptr(&lk->lk_lock, &v, 1006 v - LK_ONE_SHARER)) { 1007 lockmgr_note_shared_release(lk, file, line); 1008 goto out_xlock; 1009 } 1010 continue; 1011 } 1012 MPASS((v & ~LK_ALL_WAITERS) == LK_SHARERS_LOCK(1)); 1013 1014 setv = tid; 1015 setv |= (v & LK_ALL_WAITERS); 1016 1017 /* 1018 * Try to switch from one shared lock to an exclusive one. 1019 * We need to preserve waiters flags during the operation. 1020 */ 1021 if (atomic_fcmpset_ptr(&lk->lk_lock, &v, setv)) { 1022 LOCK_LOG_LOCK("XUPGRADE", &lk->lock_object, 0, 0, file, 1023 line); 1024 WITNESS_UPGRADE(&lk->lock_object, LOP_EXCLUSIVE | 1025 LK_TRYWIT(flags), file, line); 1026 LOCKSTAT_RECORD0(lockmgr__upgrade, lk); 1027 TD_SLOCKS_DEC(curthread); 1028 goto out; 1029 } 1030 } 1031 1032 out_xlock: 1033 error = lockmgr_xlock_hard(lk, flags, ilk, file, line, lwa); 1034 flags &= ~LK_INTERLOCK; 1035 out: 1036 lockmgr_exit(flags, ilk, 0); 1037 return (error); 1038 } 1039 1040 int 1041 lockmgr_lock_flags(struct lock *lk, u_int flags, struct lock_object *ilk, 1042 const char *file, int line) 1043 { 1044 struct lock_class *class; 1045 uintptr_t x, tid; 1046 u_int op; 1047 bool locked; 1048 1049 if (SCHEDULER_STOPPED()) 1050 return (0); 1051 1052 op = flags & LK_TYPE_MASK; 1053 locked = false; 1054 switch (op) { 1055 case LK_SHARED: 1056 if (LK_CAN_WITNESS(flags)) 1057 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, 1058 file, line, flags & LK_INTERLOCK ? ilk : NULL); 1059 if (__predict_false(lk->lock_object.lo_flags & LK_NOSHARE)) 1060 break; 1061 x = lockmgr_read_value(lk); 1062 if (lockmgr_slock_try(lk, &x, flags, true)) { 1063 lockmgr_note_shared_acquire(lk, 0, 0, 1064 file, line, flags); 1065 locked = true; 1066 } else { 1067 return (lockmgr_slock_hard(lk, flags, ilk, file, line, 1068 NULL)); 1069 } 1070 break; 1071 case LK_EXCLUSIVE: 1072 if (LK_CAN_WITNESS(flags)) 1073 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 1074 LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? 1075 ilk : NULL); 1076 tid = (uintptr_t)curthread; 1077 if (lockmgr_read_value(lk) == LK_UNLOCKED && 1078 atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) { 1079 lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, 1080 flags); 1081 locked = true; 1082 } else { 1083 return (lockmgr_xlock_hard(lk, flags, ilk, file, line, 1084 NULL)); 1085 } 1086 break; 1087 case LK_UPGRADE: 1088 case LK_TRYUPGRADE: 1089 return (lockmgr_upgrade(lk, flags, ilk, file, line, NULL)); 1090 default: 1091 break; 1092 } 1093 if (__predict_true(locked)) { 1094 if (__predict_false(flags & LK_INTERLOCK)) { 1095 class = LOCK_CLASS(ilk); 1096 class->lc_unlock(ilk); 1097 } 1098 return (0); 1099 } else { 1100 return (__lockmgr_args(lk, flags, ilk, LK_WMESG_DEFAULT, 1101 LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, file, line)); 1102 } 1103 } 1104 1105 static __noinline int 1106 lockmgr_sunlock_hard(struct lock *lk, uintptr_t x, u_int flags, struct lock_object *ilk, 1107 const char *file, int line) 1108 1109 { 1110 int wakeup_swapper = 0; 1111 1112 if (SCHEDULER_STOPPED()) 1113 goto out; 1114 1115 wakeup_swapper = wakeupshlk(lk, file, line); 1116 1117 out: 1118 lockmgr_exit(flags, ilk, wakeup_swapper); 1119 return (0); 1120 } 1121 1122 static __noinline int 1123 lockmgr_xunlock_hard(struct lock *lk, uintptr_t x, u_int flags, struct lock_object *ilk, 1124 const char *file, int line) 1125 { 1126 uintptr_t tid, v; 1127 int wakeup_swapper = 0; 1128 u_int realexslp; 1129 int queue; 1130 1131 if (SCHEDULER_STOPPED()) 1132 goto out; 1133 1134 tid = (uintptr_t)curthread; 1135 1136 /* 1137 * As first option, treact the lock as if it has not 1138 * any waiter. 1139 * Fix-up the tid var if the lock has been disowned. 1140 */ 1141 if (LK_HOLDER(x) == LK_KERNPROC) 1142 tid = LK_KERNPROC; 1143 1144 /* 1145 * The lock is held in exclusive mode. 1146 * If the lock is recursed also, then unrecurse it. 1147 */ 1148 if (lockmgr_recursed_v(x)) { 1149 LOCK_LOG2(lk, "%s: %p unrecursing", __func__, lk); 1150 lk->lk_recurse--; 1151 if (lk->lk_recurse == 0) 1152 atomic_clear_ptr(&lk->lk_lock, LK_WRITER_RECURSED); 1153 goto out; 1154 } 1155 if (tid != LK_KERNPROC) 1156 LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, 1157 LOCKSTAT_WRITER); 1158 1159 if (x == tid && atomic_cmpset_rel_ptr(&lk->lk_lock, tid, LK_UNLOCKED)) 1160 goto out; 1161 1162 sleepq_lock(&lk->lock_object); 1163 x = lockmgr_read_value(lk); 1164 v = LK_UNLOCKED; 1165 1166 /* 1167 * If the lock has exclusive waiters, give them 1168 * preference in order to avoid deadlock with 1169 * shared runners up. 1170 * If interruptible sleeps left the exclusive queue 1171 * empty avoid a starvation for the threads sleeping 1172 * on the shared queue by giving them precedence 1173 * and cleaning up the exclusive waiters bit anyway. 1174 * Please note that lk_exslpfail count may be lying 1175 * about the real number of waiters with the 1176 * LK_SLEEPFAIL flag on because they may be used in 1177 * conjunction with interruptible sleeps so 1178 * lk_exslpfail might be considered an 'upper limit' 1179 * bound, including the edge cases. 1180 */ 1181 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1182 realexslp = sleepq_sleepcnt(&lk->lock_object, SQ_EXCLUSIVE_QUEUE); 1183 if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { 1184 if (lk->lk_exslpfail != USHRT_MAX && lk->lk_exslpfail < realexslp) { 1185 lk->lk_exslpfail = 0; 1186 queue = SQ_EXCLUSIVE_QUEUE; 1187 v |= (x & LK_SHARED_WAITERS); 1188 } else { 1189 lk->lk_exslpfail = 0; 1190 LOCK_LOG2(lk, 1191 "%s: %p has only LK_SLEEPFAIL sleepers", 1192 __func__, lk); 1193 LOCK_LOG2(lk, 1194 "%s: %p waking up threads on the exclusive queue", 1195 __func__, lk); 1196 wakeup_swapper = sleepq_broadcast(&lk->lock_object, 1197 SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); 1198 queue = SQ_SHARED_QUEUE; 1199 } 1200 } else { 1201 /* 1202 * Exclusive waiters sleeping with LK_SLEEPFAIL 1203 * on and using interruptible sleeps/timeout 1204 * may have left spourious lk_exslpfail counts 1205 * on, so clean it up anyway. 1206 */ 1207 lk->lk_exslpfail = 0; 1208 queue = SQ_SHARED_QUEUE; 1209 } 1210 1211 LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue", 1212 __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : 1213 "exclusive"); 1214 atomic_store_rel_ptr(&lk->lk_lock, v); 1215 wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 0, queue); 1216 sleepq_release(&lk->lock_object); 1217 1218 out: 1219 lockmgr_exit(flags, ilk, wakeup_swapper); 1220 return (0); 1221 } 1222 1223 /* 1224 * Lightweight entry points for common operations. 1225 * 1226 * Functionality is similar to sx locks, in that none of the additional lockmgr 1227 * features are supported. To be clear, these are NOT supported: 1228 * 1. shared locking disablement 1229 * 2. returning with an error after sleep 1230 * 3. unlocking the interlock 1231 * 1232 * If in doubt, use lockmgr_lock_flags. 1233 */ 1234 int 1235 lockmgr_slock(struct lock *lk, u_int flags, const char *file, int line) 1236 { 1237 uintptr_t x; 1238 1239 MPASS((flags & LK_TYPE_MASK) == LK_SHARED); 1240 MPASS((flags & LK_INTERLOCK) == 0); 1241 MPASS((lk->lock_object.lo_flags & LK_NOSHARE) == 0); 1242 1243 if (LK_CAN_WITNESS(flags)) 1244 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, 1245 file, line, NULL); 1246 x = lockmgr_read_value(lk); 1247 if (__predict_true(lockmgr_slock_try(lk, &x, flags, true))) { 1248 lockmgr_note_shared_acquire(lk, 0, 0, file, line, flags); 1249 return (0); 1250 } 1251 1252 return (lockmgr_slock_hard(lk, flags | LK_ADAPTIVE, NULL, file, line, NULL)); 1253 } 1254 1255 int 1256 lockmgr_xlock(struct lock *lk, u_int flags, const char *file, int line) 1257 { 1258 uintptr_t tid; 1259 1260 MPASS((flags & LK_TYPE_MASK) == LK_EXCLUSIVE); 1261 MPASS((flags & LK_INTERLOCK) == 0); 1262 1263 if (LK_CAN_WITNESS(flags)) 1264 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 1265 LOP_EXCLUSIVE, file, line, NULL); 1266 tid = (uintptr_t)curthread; 1267 if (atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) { 1268 lockmgr_note_exclusive_acquire(lk, 0, 0, file, line, 1269 flags); 1270 return (0); 1271 } 1272 1273 return (lockmgr_xlock_hard(lk, flags | LK_ADAPTIVE, NULL, file, line, NULL)); 1274 } 1275 1276 int 1277 lockmgr_unlock(struct lock *lk) 1278 { 1279 uintptr_t x, tid; 1280 const char *file; 1281 int line; 1282 1283 file = __FILE__; 1284 line = __LINE__; 1285 1286 _lockmgr_assert(lk, KA_LOCKED, file, line); 1287 x = lockmgr_read_value(lk); 1288 if (__predict_true(x & LK_SHARE) != 0) { 1289 lockmgr_note_shared_release(lk, file, line); 1290 if (lockmgr_sunlock_try(lk, &x)) { 1291 LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk, LOCKSTAT_READER); 1292 } else { 1293 return (lockmgr_sunlock_hard(lk, x, LK_RELEASE, NULL, file, line)); 1294 } 1295 } else { 1296 tid = (uintptr_t)curthread; 1297 lockmgr_note_exclusive_release(lk, file, line); 1298 if (x == tid && atomic_cmpset_rel_ptr(&lk->lk_lock, tid, LK_UNLOCKED)) { 1299 LOCKSTAT_PROFILE_RELEASE_RWLOCK(lockmgr__release, lk,LOCKSTAT_WRITER); 1300 } else { 1301 return (lockmgr_xunlock_hard(lk, x, LK_RELEASE, NULL, file, line)); 1302 } 1303 } 1304 return (0); 1305 } 1306 1307 int 1308 __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk, 1309 const char *wmesg, int pri, int timo, const char *file, int line) 1310 { 1311 GIANT_DECLARE; 1312 struct lockmgr_wait lwa; 1313 struct lock_class *class; 1314 const char *iwmesg; 1315 uintptr_t tid, v, x; 1316 u_int op, realexslp; 1317 int error, ipri, itimo, queue, wakeup_swapper; 1318 #ifdef LOCK_PROFILING 1319 uint64_t waittime = 0; 1320 int contested = 0; 1321 #endif 1322 1323 if (SCHEDULER_STOPPED()) 1324 return (0); 1325 1326 error = 0; 1327 tid = (uintptr_t)curthread; 1328 op = (flags & LK_TYPE_MASK); 1329 iwmesg = (wmesg == LK_WMESG_DEFAULT) ? lk->lock_object.lo_name : wmesg; 1330 ipri = (pri == LK_PRIO_DEFAULT) ? lk->lk_pri : pri; 1331 itimo = (timo == LK_TIMO_DEFAULT) ? lk->lk_timo : timo; 1332 1333 lwa.iwmesg = iwmesg; 1334 lwa.ipri = ipri; 1335 lwa.itimo = itimo; 1336 1337 MPASS((flags & ~LK_TOTAL_MASK) == 0); 1338 KASSERT((op & (op - 1)) == 0, 1339 ("%s: Invalid requested operation @ %s:%d", __func__, file, line)); 1340 KASSERT((flags & (LK_NOWAIT | LK_SLEEPFAIL)) == 0 || 1341 (op != LK_DOWNGRADE && op != LK_RELEASE), 1342 ("%s: Invalid flags in regard of the operation desired @ %s:%d", 1343 __func__, file, line)); 1344 KASSERT((flags & LK_INTERLOCK) == 0 || ilk != NULL, 1345 ("%s: LK_INTERLOCK passed without valid interlock @ %s:%d", 1346 __func__, file, line)); 1347 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), 1348 ("%s: idle thread %p on lockmgr %s @ %s:%d", __func__, curthread, 1349 lk->lock_object.lo_name, file, line)); 1350 1351 class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; 1352 1353 if (lk->lock_object.lo_flags & LK_NOSHARE) { 1354 switch (op) { 1355 case LK_SHARED: 1356 op = LK_EXCLUSIVE; 1357 break; 1358 case LK_UPGRADE: 1359 case LK_TRYUPGRADE: 1360 case LK_DOWNGRADE: 1361 _lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED, 1362 file, line); 1363 if (flags & LK_INTERLOCK) 1364 class->lc_unlock(ilk); 1365 return (0); 1366 } 1367 } 1368 1369 wakeup_swapper = 0; 1370 switch (op) { 1371 case LK_SHARED: 1372 return (lockmgr_slock_hard(lk, flags, ilk, file, line, &lwa)); 1373 break; 1374 case LK_UPGRADE: 1375 case LK_TRYUPGRADE: 1376 return (lockmgr_upgrade(lk, flags, ilk, file, line, &lwa)); 1377 break; 1378 case LK_EXCLUSIVE: 1379 return (lockmgr_xlock_hard(lk, flags, ilk, file, line, &lwa)); 1380 break; 1381 case LK_DOWNGRADE: 1382 _lockmgr_assert(lk, KA_XLOCKED, file, line); 1383 WITNESS_DOWNGRADE(&lk->lock_object, 0, file, line); 1384 1385 /* 1386 * Panic if the lock is recursed. 1387 */ 1388 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) { 1389 if (flags & LK_INTERLOCK) 1390 class->lc_unlock(ilk); 1391 panic("%s: downgrade a recursed lockmgr %s @ %s:%d\n", 1392 __func__, iwmesg, file, line); 1393 } 1394 TD_SLOCKS_INC(curthread); 1395 1396 /* 1397 * In order to preserve waiters flags, just spin. 1398 */ 1399 for (;;) { 1400 x = lockmgr_read_value(lk); 1401 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1402 x &= LK_ALL_WAITERS; 1403 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, 1404 LK_SHARERS_LOCK(1) | x)) 1405 break; 1406 cpu_spinwait(); 1407 } 1408 LOCK_LOG_LOCK("XDOWNGRADE", &lk->lock_object, 0, 0, file, line); 1409 LOCKSTAT_RECORD0(lockmgr__downgrade, lk); 1410 break; 1411 case LK_RELEASE: 1412 _lockmgr_assert(lk, KA_LOCKED, file, line); 1413 x = lockmgr_read_value(lk); 1414 1415 if (__predict_true(x & LK_SHARE) != 0) { 1416 lockmgr_note_shared_release(lk, file, line); 1417 return (lockmgr_sunlock_hard(lk, x, flags, ilk, file, line)); 1418 } else { 1419 lockmgr_note_exclusive_release(lk, file, line); 1420 return (lockmgr_xunlock_hard(lk, x, flags, ilk, file, line)); 1421 } 1422 break; 1423 case LK_DRAIN: 1424 if (LK_CAN_WITNESS(flags)) 1425 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 1426 LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? 1427 ilk : NULL); 1428 1429 /* 1430 * Trying to drain a lock we already own will result in a 1431 * deadlock. 1432 */ 1433 if (lockmgr_xlocked(lk)) { 1434 if (flags & LK_INTERLOCK) 1435 class->lc_unlock(ilk); 1436 panic("%s: draining %s with the lock held @ %s:%d\n", 1437 __func__, iwmesg, file, line); 1438 } 1439 1440 for (;;) { 1441 if (lk->lk_lock == LK_UNLOCKED && 1442 atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) 1443 break; 1444 1445 #ifdef HWPMC_HOOKS 1446 PMC_SOFT_CALL( , , lock, failed); 1447 #endif 1448 lock_profile_obtain_lock_failed(&lk->lock_object, false, 1449 &contested, &waittime); 1450 1451 /* 1452 * If the lock is expected to not sleep just give up 1453 * and return. 1454 */ 1455 if (LK_TRYOP(flags)) { 1456 LOCK_LOG2(lk, "%s: %p fails the try operation", 1457 __func__, lk); 1458 error = EBUSY; 1459 break; 1460 } 1461 1462 /* 1463 * Acquire the sleepqueue chain lock because we 1464 * probabilly will need to manipulate waiters flags. 1465 */ 1466 sleepq_lock(&lk->lock_object); 1467 x = lockmgr_read_value(lk); 1468 1469 /* 1470 * if the lock has been released while we spun on 1471 * the sleepqueue chain lock just try again. 1472 */ 1473 if (x == LK_UNLOCKED) { 1474 sleepq_release(&lk->lock_object); 1475 continue; 1476 } 1477 1478 v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 1479 if ((x & ~v) == LK_UNLOCKED) { 1480 v = (x & ~LK_EXCLUSIVE_SPINNERS); 1481 1482 /* 1483 * If interruptible sleeps left the exclusive 1484 * queue empty avoid a starvation for the 1485 * threads sleeping on the shared queue by 1486 * giving them precedence and cleaning up the 1487 * exclusive waiters bit anyway. 1488 * Please note that lk_exslpfail count may be 1489 * lying about the real number of waiters with 1490 * the LK_SLEEPFAIL flag on because they may 1491 * be used in conjunction with interruptible 1492 * sleeps so lk_exslpfail might be considered 1493 * an 'upper limit' bound, including the edge 1494 * cases. 1495 */ 1496 if (v & LK_EXCLUSIVE_WAITERS) { 1497 queue = SQ_EXCLUSIVE_QUEUE; 1498 v &= ~LK_EXCLUSIVE_WAITERS; 1499 } else { 1500 /* 1501 * Exclusive waiters sleeping with 1502 * LK_SLEEPFAIL on and using 1503 * interruptible sleeps/timeout may 1504 * have left spourious lk_exslpfail 1505 * counts on, so clean it up anyway. 1506 */ 1507 MPASS(v & LK_SHARED_WAITERS); 1508 lk->lk_exslpfail = 0; 1509 queue = SQ_SHARED_QUEUE; 1510 v &= ~LK_SHARED_WAITERS; 1511 } 1512 if (queue == SQ_EXCLUSIVE_QUEUE) { 1513 realexslp = 1514 sleepq_sleepcnt(&lk->lock_object, 1515 SQ_EXCLUSIVE_QUEUE); 1516 if (lk->lk_exslpfail >= realexslp) { 1517 lk->lk_exslpfail = 0; 1518 queue = SQ_SHARED_QUEUE; 1519 v &= ~LK_SHARED_WAITERS; 1520 if (realexslp != 0) { 1521 LOCK_LOG2(lk, 1522 "%s: %p has only LK_SLEEPFAIL sleepers", 1523 __func__, lk); 1524 LOCK_LOG2(lk, 1525 "%s: %p waking up threads on the exclusive queue", 1526 __func__, lk); 1527 wakeup_swapper = 1528 sleepq_broadcast( 1529 &lk->lock_object, 1530 SLEEPQ_LK, 0, 1531 SQ_EXCLUSIVE_QUEUE); 1532 } 1533 } else 1534 lk->lk_exslpfail = 0; 1535 } 1536 if (!atomic_cmpset_ptr(&lk->lk_lock, x, v)) { 1537 sleepq_release(&lk->lock_object); 1538 continue; 1539 } 1540 LOCK_LOG3(lk, 1541 "%s: %p waking up all threads on the %s queue", 1542 __func__, lk, queue == SQ_SHARED_QUEUE ? 1543 "shared" : "exclusive"); 1544 wakeup_swapper |= sleepq_broadcast( 1545 &lk->lock_object, SLEEPQ_LK, 0, queue); 1546 1547 /* 1548 * If shared waiters have been woken up we need 1549 * to wait for one of them to acquire the lock 1550 * before to set the exclusive waiters in 1551 * order to avoid a deadlock. 1552 */ 1553 if (queue == SQ_SHARED_QUEUE) { 1554 for (v = lk->lk_lock; 1555 (v & LK_SHARE) && !LK_SHARERS(v); 1556 v = lk->lk_lock) 1557 cpu_spinwait(); 1558 } 1559 } 1560 1561 /* 1562 * Try to set the LK_EXCLUSIVE_WAITERS flag. If we 1563 * fail, loop back and retry. 1564 */ 1565 if ((x & LK_EXCLUSIVE_WAITERS) == 0) { 1566 if (!atomic_cmpset_ptr(&lk->lk_lock, x, 1567 x | LK_EXCLUSIVE_WAITERS)) { 1568 sleepq_release(&lk->lock_object); 1569 continue; 1570 } 1571 LOCK_LOG2(lk, "%s: %p set drain waiters flag", 1572 __func__, lk); 1573 } 1574 1575 /* 1576 * As far as we have been unable to acquire the 1577 * exclusive lock and the exclusive waiters flag 1578 * is set, we will sleep. 1579 */ 1580 if (flags & LK_INTERLOCK) { 1581 class->lc_unlock(ilk); 1582 flags &= ~LK_INTERLOCK; 1583 } 1584 GIANT_SAVE(); 1585 sleepq_add(&lk->lock_object, NULL, iwmesg, SLEEPQ_LK, 1586 SQ_EXCLUSIVE_QUEUE); 1587 sleepq_wait(&lk->lock_object, ipri & PRIMASK); 1588 GIANT_RESTORE(); 1589 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 1590 __func__, lk); 1591 } 1592 1593 if (error == 0) { 1594 lock_profile_obtain_lock_success(&lk->lock_object, 1595 false, contested, waittime, file, line); 1596 LOCK_LOG_LOCK("DRAIN", &lk->lock_object, 0, 1597 lk->lk_recurse, file, line); 1598 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | 1599 LK_TRYWIT(flags), file, line); 1600 TD_LOCKS_INC(curthread); 1601 STACK_SAVE(lk); 1602 } 1603 break; 1604 default: 1605 if (flags & LK_INTERLOCK) 1606 class->lc_unlock(ilk); 1607 panic("%s: unknown lockmgr request 0x%x\n", __func__, op); 1608 } 1609 1610 if (flags & LK_INTERLOCK) 1611 class->lc_unlock(ilk); 1612 if (wakeup_swapper) 1613 kick_proc0(); 1614 1615 return (error); 1616 } 1617 1618 void 1619 _lockmgr_disown(struct lock *lk, const char *file, int line) 1620 { 1621 uintptr_t tid, x; 1622 1623 if (SCHEDULER_STOPPED()) 1624 return; 1625 1626 tid = (uintptr_t)curthread; 1627 _lockmgr_assert(lk, KA_XLOCKED, file, line); 1628 1629 /* 1630 * Panic if the lock is recursed. 1631 */ 1632 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) 1633 panic("%s: disown a recursed lockmgr @ %s:%d\n", 1634 __func__, file, line); 1635 1636 /* 1637 * If the owner is already LK_KERNPROC just skip the whole operation. 1638 */ 1639 if (LK_HOLDER(lk->lk_lock) != tid) 1640 return; 1641 lock_profile_release_lock(&lk->lock_object, false); 1642 LOCKSTAT_RECORD1(lockmgr__disown, lk, LOCKSTAT_WRITER); 1643 LOCK_LOG_LOCK("XDISOWN", &lk->lock_object, 0, 0, file, line); 1644 WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line); 1645 TD_LOCKS_DEC(curthread); 1646 STACK_SAVE(lk); 1647 1648 /* 1649 * In order to preserve waiters flags, just spin. 1650 */ 1651 for (;;) { 1652 x = lockmgr_read_value(lk); 1653 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1654 x &= LK_ALL_WAITERS; 1655 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, 1656 LK_KERNPROC | x)) 1657 return; 1658 cpu_spinwait(); 1659 } 1660 } 1661 1662 void 1663 lockmgr_printinfo(const struct lock *lk) 1664 { 1665 struct thread *td; 1666 uintptr_t x; 1667 1668 if (lk->lk_lock == LK_UNLOCKED) 1669 printf("lock type %s: UNLOCKED\n", lk->lock_object.lo_name); 1670 else if (lk->lk_lock & LK_SHARE) 1671 printf("lock type %s: SHARED (count %ju)\n", 1672 lk->lock_object.lo_name, 1673 (uintmax_t)LK_SHARERS(lk->lk_lock)); 1674 else { 1675 td = lockmgr_xholder(lk); 1676 if (td == (struct thread *)LK_KERNPROC) 1677 printf("lock type %s: EXCL by KERNPROC\n", 1678 lk->lock_object.lo_name); 1679 else 1680 printf("lock type %s: EXCL by thread %p " 1681 "(pid %d, %s, tid %d)\n", lk->lock_object.lo_name, 1682 td, td->td_proc->p_pid, td->td_proc->p_comm, 1683 td->td_tid); 1684 } 1685 1686 x = lk->lk_lock; 1687 if (x & LK_EXCLUSIVE_WAITERS) 1688 printf(" with exclusive waiters pending\n"); 1689 if (x & LK_SHARED_WAITERS) 1690 printf(" with shared waiters pending\n"); 1691 if (x & LK_EXCLUSIVE_SPINNERS) 1692 printf(" with exclusive spinners pending\n"); 1693 1694 STACK_PRINT(lk); 1695 } 1696 1697 int 1698 lockstatus(const struct lock *lk) 1699 { 1700 uintptr_t v, x; 1701 int ret; 1702 1703 ret = LK_SHARED; 1704 x = lockmgr_read_value(lk); 1705 v = LK_HOLDER(x); 1706 1707 if ((x & LK_SHARE) == 0) { 1708 if (v == (uintptr_t)curthread || v == LK_KERNPROC) 1709 ret = LK_EXCLUSIVE; 1710 else 1711 ret = LK_EXCLOTHER; 1712 } else if (x == LK_UNLOCKED) 1713 ret = 0; 1714 1715 return (ret); 1716 } 1717 1718 #ifdef INVARIANT_SUPPORT 1719 1720 FEATURE(invariant_support, 1721 "Support for modules compiled with INVARIANTS option"); 1722 1723 #ifndef INVARIANTS 1724 #undef _lockmgr_assert 1725 #endif 1726 1727 void 1728 _lockmgr_assert(const struct lock *lk, int what, const char *file, int line) 1729 { 1730 int slocked = 0; 1731 1732 if (SCHEDULER_STOPPED()) 1733 return; 1734 switch (what) { 1735 case KA_SLOCKED: 1736 case KA_SLOCKED | KA_NOTRECURSED: 1737 case KA_SLOCKED | KA_RECURSED: 1738 slocked = 1; 1739 case KA_LOCKED: 1740 case KA_LOCKED | KA_NOTRECURSED: 1741 case KA_LOCKED | KA_RECURSED: 1742 #ifdef WITNESS 1743 1744 /* 1745 * We cannot trust WITNESS if the lock is held in exclusive 1746 * mode and a call to lockmgr_disown() happened. 1747 * Workaround this skipping the check if the lock is held in 1748 * exclusive mode even for the KA_LOCKED case. 1749 */ 1750 if (slocked || (lk->lk_lock & LK_SHARE)) { 1751 witness_assert(&lk->lock_object, what, file, line); 1752 break; 1753 } 1754 #endif 1755 if (lk->lk_lock == LK_UNLOCKED || 1756 ((lk->lk_lock & LK_SHARE) == 0 && (slocked || 1757 (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk))))) 1758 panic("Lock %s not %slocked @ %s:%d\n", 1759 lk->lock_object.lo_name, slocked ? "share" : "", 1760 file, line); 1761 1762 if ((lk->lk_lock & LK_SHARE) == 0) { 1763 if (lockmgr_recursed(lk)) { 1764 if (what & KA_NOTRECURSED) 1765 panic("Lock %s recursed @ %s:%d\n", 1766 lk->lock_object.lo_name, file, 1767 line); 1768 } else if (what & KA_RECURSED) 1769 panic("Lock %s not recursed @ %s:%d\n", 1770 lk->lock_object.lo_name, file, line); 1771 } 1772 break; 1773 case KA_XLOCKED: 1774 case KA_XLOCKED | KA_NOTRECURSED: 1775 case KA_XLOCKED | KA_RECURSED: 1776 if (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk)) 1777 panic("Lock %s not exclusively locked @ %s:%d\n", 1778 lk->lock_object.lo_name, file, line); 1779 if (lockmgr_recursed(lk)) { 1780 if (what & KA_NOTRECURSED) 1781 panic("Lock %s recursed @ %s:%d\n", 1782 lk->lock_object.lo_name, file, line); 1783 } else if (what & KA_RECURSED) 1784 panic("Lock %s not recursed @ %s:%d\n", 1785 lk->lock_object.lo_name, file, line); 1786 break; 1787 case KA_UNLOCKED: 1788 if (lockmgr_xlocked(lk) || lockmgr_disowned(lk)) 1789 panic("Lock %s exclusively locked @ %s:%d\n", 1790 lk->lock_object.lo_name, file, line); 1791 break; 1792 default: 1793 panic("Unknown lockmgr assertion: %d @ %s:%d\n", what, file, 1794 line); 1795 } 1796 } 1797 #endif 1798 1799 #ifdef DDB 1800 int 1801 lockmgr_chain(struct thread *td, struct thread **ownerp) 1802 { 1803 const struct lock *lk; 1804 1805 lk = td->td_wchan; 1806 1807 if (LOCK_CLASS(&lk->lock_object) != &lock_class_lockmgr) 1808 return (0); 1809 db_printf("blocked on lockmgr %s", lk->lock_object.lo_name); 1810 if (lk->lk_lock & LK_SHARE) 1811 db_printf("SHARED (count %ju)\n", 1812 (uintmax_t)LK_SHARERS(lk->lk_lock)); 1813 else 1814 db_printf("EXCL\n"); 1815 *ownerp = lockmgr_xholder(lk); 1816 1817 return (1); 1818 } 1819 1820 static void 1821 db_show_lockmgr(const struct lock_object *lock) 1822 { 1823 struct thread *td; 1824 const struct lock *lk; 1825 1826 lk = (const struct lock *)lock; 1827 1828 db_printf(" state: "); 1829 if (lk->lk_lock == LK_UNLOCKED) 1830 db_printf("UNLOCKED\n"); 1831 else if (lk->lk_lock & LK_SHARE) 1832 db_printf("SLOCK: %ju\n", (uintmax_t)LK_SHARERS(lk->lk_lock)); 1833 else { 1834 td = lockmgr_xholder(lk); 1835 if (td == (struct thread *)LK_KERNPROC) 1836 db_printf("XLOCK: LK_KERNPROC\n"); 1837 else 1838 db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td, 1839 td->td_tid, td->td_proc->p_pid, 1840 td->td_proc->p_comm); 1841 if (lockmgr_recursed(lk)) 1842 db_printf(" recursed: %d\n", lk->lk_recurse); 1843 } 1844 db_printf(" waiters: "); 1845 switch (lk->lk_lock & LK_ALL_WAITERS) { 1846 case LK_SHARED_WAITERS: 1847 db_printf("shared\n"); 1848 break; 1849 case LK_EXCLUSIVE_WAITERS: 1850 db_printf("exclusive\n"); 1851 break; 1852 case LK_ALL_WAITERS: 1853 db_printf("shared and exclusive\n"); 1854 break; 1855 default: 1856 db_printf("none\n"); 1857 } 1858 db_printf(" spinners: "); 1859 if (lk->lk_lock & LK_EXCLUSIVE_SPINNERS) 1860 db_printf("exclusive\n"); 1861 else 1862 db_printf("none\n"); 1863 } 1864 #endif 1865