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