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