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