1 /*- 2 * Copyright (c) 2008 Attilio Rao <attilio@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice(s), this list of conditions and the following disclaimer as 10 * the first lines of this file unmodified other than the possible 11 * addition of one or more copyright notices. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice(s), this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY 17 * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED 18 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE 19 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY 20 * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 23 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH 26 * DAMAGE. 27 */ 28 29 #include "opt_adaptive_lockmgrs.h" 30 #include "opt_ddb.h" 31 #include "opt_hwpmc_hooks.h" 32 33 #include <sys/cdefs.h> 34 __FBSDID("$FreeBSD$"); 35 36 #include <sys/param.h> 37 #include <sys/kdb.h> 38 #include <sys/ktr.h> 39 #include <sys/lock.h> 40 #include <sys/lock_profile.h> 41 #include <sys/lockmgr.h> 42 #include <sys/mutex.h> 43 #include <sys/proc.h> 44 #include <sys/sleepqueue.h> 45 #ifdef DEBUG_LOCKS 46 #include <sys/stack.h> 47 #endif 48 #include <sys/sysctl.h> 49 #include <sys/systm.h> 50 51 #include <machine/cpu.h> 52 53 #ifdef DDB 54 #include <ddb/ddb.h> 55 #endif 56 57 #ifdef HWPMC_HOOKS 58 #include <sys/pmckern.h> 59 PMC_SOFT_DECLARE( , , lock, failed); 60 #endif 61 62 CTASSERT(((LK_ADAPTIVE | LK_NOSHARE) & LO_CLASSFLAGS) == 63 (LK_ADAPTIVE | LK_NOSHARE)); 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 (_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 (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 #define LK_CAN_SHARE(x, flags) \ 115 (((x) & LK_SHARE) && \ 116 (((x) & (LK_EXCLUSIVE_WAITERS | LK_EXCLUSIVE_SPINNERS)) == 0 || \ 117 (curthread->td_lk_slocks != 0 && !(flags & LK_NODDLKTREAT)) || \ 118 (curthread->td_pflags & TDP_DEADLKTREAT))) 119 #define LK_TRYOP(x) \ 120 ((x) & LK_NOWAIT) 121 122 #define LK_CAN_WITNESS(x) \ 123 (((x) & LK_NOWITNESS) == 0 && !LK_TRYOP(x)) 124 #define LK_TRYWIT(x) \ 125 (LK_TRYOP(x) ? LOP_TRYLOCK : 0) 126 127 #define LK_CAN_ADAPT(lk, f) \ 128 (((lk)->lock_object.lo_flags & LK_ADAPTIVE) != 0 && \ 129 ((f) & LK_SLEEPFAIL) == 0) 130 131 #define lockmgr_disowned(lk) \ 132 (((lk)->lk_lock & ~(LK_FLAGMASK & ~LK_SHARE)) == LK_KERNPROC) 133 134 #define lockmgr_xlocked(lk) \ 135 (((lk)->lk_lock & ~(LK_FLAGMASK & ~LK_SHARE)) == (uintptr_t)curthread) 136 137 static void assert_lockmgr(const struct lock_object *lock, int how); 138 #ifdef DDB 139 static void db_show_lockmgr(const struct lock_object *lock); 140 #endif 141 static void lock_lockmgr(struct lock_object *lock, uintptr_t how); 142 #ifdef KDTRACE_HOOKS 143 static int owner_lockmgr(const struct lock_object *lock, 144 struct thread **owner); 145 #endif 146 static uintptr_t unlock_lockmgr(struct lock_object *lock); 147 148 struct lock_class lock_class_lockmgr = { 149 .lc_name = "lockmgr", 150 .lc_flags = LC_RECURSABLE | LC_SLEEPABLE | LC_SLEEPLOCK | LC_UPGRADABLE, 151 .lc_assert = assert_lockmgr, 152 #ifdef DDB 153 .lc_ddb_show = db_show_lockmgr, 154 #endif 155 .lc_lock = lock_lockmgr, 156 .lc_unlock = unlock_lockmgr, 157 #ifdef KDTRACE_HOOKS 158 .lc_owner = owner_lockmgr, 159 #endif 160 }; 161 162 #ifdef ADAPTIVE_LOCKMGRS 163 static u_int alk_retries = 10; 164 static u_int alk_loops = 10000; 165 static SYSCTL_NODE(_debug, OID_AUTO, lockmgr, CTLFLAG_RD, NULL, 166 "lockmgr debugging"); 167 SYSCTL_UINT(_debug_lockmgr, OID_AUTO, retries, CTLFLAG_RW, &alk_retries, 0, ""); 168 SYSCTL_UINT(_debug_lockmgr, OID_AUTO, loops, CTLFLAG_RW, &alk_loops, 0, ""); 169 #endif 170 171 static __inline struct thread * 172 lockmgr_xholder(const struct lock *lk) 173 { 174 uintptr_t x; 175 176 x = lk->lk_lock; 177 return ((x & LK_SHARE) ? NULL : (struct thread *)LK_HOLDER(x)); 178 } 179 180 /* 181 * It assumes sleepq_lock held and returns with this one unheld. 182 * It also assumes the generic interlock is sane and previously checked. 183 * If LK_INTERLOCK is specified the interlock is not reacquired after the 184 * sleep. 185 */ 186 static __inline int 187 sleeplk(struct lock *lk, u_int flags, struct lock_object *ilk, 188 const char *wmesg, int pri, int timo, int queue) 189 { 190 GIANT_DECLARE; 191 struct lock_class *class; 192 int catch, error; 193 194 class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; 195 catch = pri & PCATCH; 196 pri &= PRIMASK; 197 error = 0; 198 199 LOCK_LOG3(lk, "%s: %p blocking on the %s sleepqueue", __func__, lk, 200 (queue == SQ_EXCLUSIVE_QUEUE) ? "exclusive" : "shared"); 201 202 if (flags & LK_INTERLOCK) 203 class->lc_unlock(ilk); 204 if (queue == SQ_EXCLUSIVE_QUEUE && (flags & LK_SLEEPFAIL) != 0) 205 lk->lk_exslpfail++; 206 GIANT_SAVE(); 207 sleepq_add(&lk->lock_object, NULL, wmesg, SLEEPQ_LK | (catch ? 208 SLEEPQ_INTERRUPTIBLE : 0), queue); 209 if ((flags & LK_TIMELOCK) && timo) 210 sleepq_set_timeout(&lk->lock_object, timo); 211 212 /* 213 * Decisional switch for real sleeping. 214 */ 215 if ((flags & LK_TIMELOCK) && timo && catch) 216 error = sleepq_timedwait_sig(&lk->lock_object, pri); 217 else if ((flags & LK_TIMELOCK) && timo) 218 error = sleepq_timedwait(&lk->lock_object, pri); 219 else if (catch) 220 error = sleepq_wait_sig(&lk->lock_object, pri); 221 else 222 sleepq_wait(&lk->lock_object, pri); 223 GIANT_RESTORE(); 224 if ((flags & LK_SLEEPFAIL) && error == 0) 225 error = ENOLCK; 226 227 return (error); 228 } 229 230 static __inline int 231 wakeupshlk(struct lock *lk, const char *file, int line) 232 { 233 uintptr_t v, x; 234 u_int realexslp; 235 int queue, wakeup_swapper; 236 237 WITNESS_UNLOCK(&lk->lock_object, 0, file, line); 238 LOCK_LOG_LOCK("SUNLOCK", &lk->lock_object, 0, 0, file, line); 239 240 wakeup_swapper = 0; 241 for (;;) { 242 x = lk->lk_lock; 243 244 /* 245 * If there is more than one shared lock held, just drop one 246 * and return. 247 */ 248 if (LK_SHARERS(x) > 1) { 249 if (atomic_cmpset_rel_ptr(&lk->lk_lock, x, 250 x - LK_ONE_SHARER)) 251 break; 252 continue; 253 } 254 255 /* 256 * If there are not waiters on the exclusive queue, drop the 257 * lock quickly. 258 */ 259 if ((x & LK_ALL_WAITERS) == 0) { 260 MPASS((x & ~LK_EXCLUSIVE_SPINNERS) == 261 LK_SHARERS_LOCK(1)); 262 if (atomic_cmpset_rel_ptr(&lk->lk_lock, x, LK_UNLOCKED)) 263 break; 264 continue; 265 } 266 267 /* 268 * We should have a sharer with waiters, so enter the hard 269 * path in order to handle wakeups correctly. 270 */ 271 sleepq_lock(&lk->lock_object); 272 x = lk->lk_lock & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 273 v = LK_UNLOCKED; 274 275 /* 276 * If the lock has exclusive waiters, give them preference in 277 * order to avoid deadlock with shared runners up. 278 * If interruptible sleeps left the exclusive queue empty 279 * avoid a starvation for the threads sleeping on the shared 280 * queue by giving them precedence and cleaning up the 281 * exclusive waiters bit anyway. 282 * Please note that lk_exslpfail count may be lying about 283 * the real number of waiters with the LK_SLEEPFAIL flag on 284 * because they may be used in conjunction with interruptible 285 * sleeps so lk_exslpfail might be considered an 'upper limit' 286 * bound, including the edge cases. 287 */ 288 realexslp = sleepq_sleepcnt(&lk->lock_object, 289 SQ_EXCLUSIVE_QUEUE); 290 if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { 291 if (lk->lk_exslpfail < realexslp) { 292 lk->lk_exslpfail = 0; 293 queue = SQ_EXCLUSIVE_QUEUE; 294 v |= (x & LK_SHARED_WAITERS); 295 } else { 296 lk->lk_exslpfail = 0; 297 LOCK_LOG2(lk, 298 "%s: %p has only LK_SLEEPFAIL sleepers", 299 __func__, lk); 300 LOCK_LOG2(lk, 301 "%s: %p waking up threads on the exclusive queue", 302 __func__, lk); 303 wakeup_swapper = 304 sleepq_broadcast(&lk->lock_object, 305 SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); 306 queue = SQ_SHARED_QUEUE; 307 } 308 309 } else { 310 311 /* 312 * Exclusive waiters sleeping with LK_SLEEPFAIL on 313 * and using interruptible sleeps/timeout may have 314 * left spourious lk_exslpfail counts on, so clean 315 * it up anyway. 316 */ 317 lk->lk_exslpfail = 0; 318 queue = SQ_SHARED_QUEUE; 319 } 320 321 if (!atomic_cmpset_rel_ptr(&lk->lk_lock, LK_SHARERS_LOCK(1) | x, 322 v)) { 323 sleepq_release(&lk->lock_object); 324 continue; 325 } 326 LOCK_LOG3(lk, "%s: %p waking up threads on the %s queue", 327 __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : 328 "exclusive"); 329 wakeup_swapper |= sleepq_broadcast(&lk->lock_object, SLEEPQ_LK, 330 0, queue); 331 sleepq_release(&lk->lock_object); 332 break; 333 } 334 335 lock_profile_release_lock(&lk->lock_object); 336 TD_LOCKS_DEC(curthread); 337 TD_SLOCKS_DEC(curthread); 338 return (wakeup_swapper); 339 } 340 341 static void 342 assert_lockmgr(const struct lock_object *lock, int what) 343 { 344 345 panic("lockmgr locks do not support assertions"); 346 } 347 348 static void 349 lock_lockmgr(struct lock_object *lock, uintptr_t how) 350 { 351 352 panic("lockmgr locks do not support sleep interlocking"); 353 } 354 355 static uintptr_t 356 unlock_lockmgr(struct lock_object *lock) 357 { 358 359 panic("lockmgr locks do not support sleep interlocking"); 360 } 361 362 #ifdef KDTRACE_HOOKS 363 static int 364 owner_lockmgr(const struct lock_object *lock, struct thread **owner) 365 { 366 367 panic("lockmgr locks do not support owner inquiring"); 368 } 369 #endif 370 371 void 372 lockinit(struct lock *lk, int pri, const char *wmesg, int timo, int flags) 373 { 374 int iflags; 375 376 MPASS((flags & ~LK_INIT_MASK) == 0); 377 ASSERT_ATOMIC_LOAD_PTR(lk->lk_lock, 378 ("%s: lockmgr not aligned for %s: %p", __func__, wmesg, 379 &lk->lk_lock)); 380 381 iflags = LO_SLEEPABLE | LO_UPGRADABLE; 382 if (flags & LK_CANRECURSE) 383 iflags |= LO_RECURSABLE; 384 if ((flags & LK_NODUP) == 0) 385 iflags |= LO_DUPOK; 386 if (flags & LK_NOPROFILE) 387 iflags |= LO_NOPROFILE; 388 if ((flags & LK_NOWITNESS) == 0) 389 iflags |= LO_WITNESS; 390 if (flags & LK_QUIET) 391 iflags |= LO_QUIET; 392 if (flags & LK_IS_VNODE) 393 iflags |= LO_IS_VNODE; 394 iflags |= flags & (LK_ADAPTIVE | LK_NOSHARE); 395 396 lock_init(&lk->lock_object, &lock_class_lockmgr, wmesg, NULL, iflags); 397 lk->lk_lock = LK_UNLOCKED; 398 lk->lk_recurse = 0; 399 lk->lk_exslpfail = 0; 400 lk->lk_timo = timo; 401 lk->lk_pri = pri; 402 STACK_ZERO(lk); 403 } 404 405 /* 406 * XXX: Gross hacks to manipulate external lock flags after 407 * initialization. Used for certain vnode and buf locks. 408 */ 409 void 410 lockallowshare(struct lock *lk) 411 { 412 413 lockmgr_assert(lk, KA_XLOCKED); 414 lk->lock_object.lo_flags &= ~LK_NOSHARE; 415 } 416 417 void 418 lockdisableshare(struct lock *lk) 419 { 420 421 lockmgr_assert(lk, KA_XLOCKED); 422 lk->lock_object.lo_flags |= LK_NOSHARE; 423 } 424 425 void 426 lockallowrecurse(struct lock *lk) 427 { 428 429 lockmgr_assert(lk, KA_XLOCKED); 430 lk->lock_object.lo_flags |= LO_RECURSABLE; 431 } 432 433 void 434 lockdisablerecurse(struct lock *lk) 435 { 436 437 lockmgr_assert(lk, KA_XLOCKED); 438 lk->lock_object.lo_flags &= ~LO_RECURSABLE; 439 } 440 441 void 442 lockdestroy(struct lock *lk) 443 { 444 445 KASSERT(lk->lk_lock == LK_UNLOCKED, ("lockmgr still held")); 446 KASSERT(lk->lk_recurse == 0, ("lockmgr still recursed")); 447 KASSERT(lk->lk_exslpfail == 0, ("lockmgr still exclusive waiters")); 448 lock_destroy(&lk->lock_object); 449 } 450 451 int 452 __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk, 453 const char *wmesg, int pri, int timo, const char *file, int line) 454 { 455 GIANT_DECLARE; 456 struct lock_class *class; 457 const char *iwmesg; 458 uintptr_t tid, v, x; 459 u_int op, realexslp; 460 int error, ipri, itimo, queue, wakeup_swapper; 461 #ifdef LOCK_PROFILING 462 uint64_t waittime = 0; 463 int contested = 0; 464 #endif 465 #ifdef ADAPTIVE_LOCKMGRS 466 volatile struct thread *owner; 467 u_int i, spintries = 0; 468 #endif 469 470 error = 0; 471 tid = (uintptr_t)curthread; 472 op = (flags & LK_TYPE_MASK); 473 iwmesg = (wmesg == LK_WMESG_DEFAULT) ? lk->lock_object.lo_name : wmesg; 474 ipri = (pri == LK_PRIO_DEFAULT) ? lk->lk_pri : pri; 475 itimo = (timo == LK_TIMO_DEFAULT) ? lk->lk_timo : timo; 476 477 MPASS((flags & ~LK_TOTAL_MASK) == 0); 478 KASSERT((op & (op - 1)) == 0, 479 ("%s: Invalid requested operation @ %s:%d", __func__, file, line)); 480 KASSERT((flags & (LK_NOWAIT | LK_SLEEPFAIL)) == 0 || 481 (op != LK_DOWNGRADE && op != LK_RELEASE), 482 ("%s: Invalid flags in regard of the operation desired @ %s:%d", 483 __func__, file, line)); 484 KASSERT((flags & LK_INTERLOCK) == 0 || ilk != NULL, 485 ("%s: LK_INTERLOCK passed without valid interlock @ %s:%d", 486 __func__, file, line)); 487 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread), 488 ("%s: idle thread %p on lockmgr %s @ %s:%d", __func__, curthread, 489 lk->lock_object.lo_name, file, line)); 490 491 class = (flags & LK_INTERLOCK) ? LOCK_CLASS(ilk) : NULL; 492 if (panicstr != NULL) { 493 if (flags & LK_INTERLOCK) 494 class->lc_unlock(ilk); 495 return (0); 496 } 497 498 if (lk->lock_object.lo_flags & LK_NOSHARE) { 499 switch (op) { 500 case LK_SHARED: 501 op = LK_EXCLUSIVE; 502 break; 503 case LK_UPGRADE: 504 case LK_TRYUPGRADE: 505 case LK_DOWNGRADE: 506 _lockmgr_assert(lk, KA_XLOCKED | KA_NOTRECURSED, 507 file, line); 508 if (flags & LK_INTERLOCK) 509 class->lc_unlock(ilk); 510 return (0); 511 } 512 } 513 514 wakeup_swapper = 0; 515 switch (op) { 516 case LK_SHARED: 517 if (LK_CAN_WITNESS(flags)) 518 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER, 519 file, line, flags & LK_INTERLOCK ? ilk : NULL); 520 for (;;) { 521 x = lk->lk_lock; 522 523 /* 524 * If no other thread has an exclusive lock, or 525 * no exclusive waiter is present, bump the count of 526 * sharers. Since we have to preserve the state of 527 * waiters, if we fail to acquire the shared lock 528 * loop back and retry. 529 */ 530 if (LK_CAN_SHARE(x, flags)) { 531 if (atomic_cmpset_acq_ptr(&lk->lk_lock, x, 532 x + LK_ONE_SHARER)) 533 break; 534 continue; 535 } 536 #ifdef HWPMC_HOOKS 537 PMC_SOFT_CALL( , , lock, failed); 538 #endif 539 lock_profile_obtain_lock_failed(&lk->lock_object, 540 &contested, &waittime); 541 542 /* 543 * If the lock is already held by curthread in 544 * exclusive way avoid a deadlock. 545 */ 546 if (LK_HOLDER(x) == tid) { 547 LOCK_LOG2(lk, 548 "%s: %p already held in exclusive mode", 549 __func__, lk); 550 error = EDEADLK; 551 break; 552 } 553 554 /* 555 * If the lock is expected to not sleep just give up 556 * and return. 557 */ 558 if (LK_TRYOP(flags)) { 559 LOCK_LOG2(lk, "%s: %p fails the try operation", 560 __func__, lk); 561 error = EBUSY; 562 break; 563 } 564 565 #ifdef ADAPTIVE_LOCKMGRS 566 /* 567 * If the owner is running on another CPU, spin until 568 * the owner stops running or the state of the lock 569 * changes. We need a double-state handle here 570 * because for a failed acquisition the lock can be 571 * either held in exclusive mode or shared mode 572 * (for the writer starvation avoidance technique). 573 */ 574 if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 && 575 LK_HOLDER(x) != LK_KERNPROC) { 576 owner = (struct thread *)LK_HOLDER(x); 577 if (LOCK_LOG_TEST(&lk->lock_object, 0)) 578 CTR3(KTR_LOCK, 579 "%s: spinning on %p held by %p", 580 __func__, lk, owner); 581 KTR_STATE1(KTR_SCHED, "thread", 582 sched_tdname(td), "spinning", 583 "lockname:\"%s\"", lk->lock_object.lo_name); 584 585 /* 586 * If we are holding also an interlock drop it 587 * in order to avoid a deadlock if the lockmgr 588 * owner is adaptively spinning on the 589 * interlock itself. 590 */ 591 if (flags & LK_INTERLOCK) { 592 class->lc_unlock(ilk); 593 flags &= ~LK_INTERLOCK; 594 } 595 GIANT_SAVE(); 596 while (LK_HOLDER(lk->lk_lock) == 597 (uintptr_t)owner && TD_IS_RUNNING(owner)) 598 cpu_spinwait(); 599 KTR_STATE0(KTR_SCHED, "thread", 600 sched_tdname(td), "running"); 601 GIANT_RESTORE(); 602 continue; 603 } else if (LK_CAN_ADAPT(lk, flags) && 604 (x & LK_SHARE) != 0 && LK_SHARERS(x) && 605 spintries < alk_retries) { 606 KTR_STATE1(KTR_SCHED, "thread", 607 sched_tdname(td), "spinning", 608 "lockname:\"%s\"", lk->lock_object.lo_name); 609 if (flags & LK_INTERLOCK) { 610 class->lc_unlock(ilk); 611 flags &= ~LK_INTERLOCK; 612 } 613 GIANT_SAVE(); 614 spintries++; 615 for (i = 0; i < alk_loops; i++) { 616 if (LOCK_LOG_TEST(&lk->lock_object, 0)) 617 CTR4(KTR_LOCK, 618 "%s: shared spinning on %p with %u and %u", 619 __func__, lk, spintries, i); 620 x = lk->lk_lock; 621 if ((x & LK_SHARE) == 0 || 622 LK_CAN_SHARE(x, flags) != 0) 623 break; 624 cpu_spinwait(); 625 } 626 KTR_STATE0(KTR_SCHED, "thread", 627 sched_tdname(td), "running"); 628 GIANT_RESTORE(); 629 if (i != alk_loops) 630 continue; 631 } 632 #endif 633 634 /* 635 * Acquire the sleepqueue chain lock because we 636 * probabilly will need to manipulate waiters flags. 637 */ 638 sleepq_lock(&lk->lock_object); 639 x = lk->lk_lock; 640 641 /* 642 * if the lock can be acquired in shared mode, try 643 * again. 644 */ 645 if (LK_CAN_SHARE(x, flags)) { 646 sleepq_release(&lk->lock_object); 647 continue; 648 } 649 650 #ifdef ADAPTIVE_LOCKMGRS 651 /* 652 * The current lock owner might have started executing 653 * on another CPU (or the lock could have changed 654 * owner) while we were waiting on the turnstile 655 * chain lock. If so, drop the turnstile lock and try 656 * again. 657 */ 658 if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 && 659 LK_HOLDER(x) != LK_KERNPROC) { 660 owner = (struct thread *)LK_HOLDER(x); 661 if (TD_IS_RUNNING(owner)) { 662 sleepq_release(&lk->lock_object); 663 continue; 664 } 665 } 666 #endif 667 668 /* 669 * Try to set the LK_SHARED_WAITERS flag. If we fail, 670 * loop back and retry. 671 */ 672 if ((x & LK_SHARED_WAITERS) == 0) { 673 if (!atomic_cmpset_acq_ptr(&lk->lk_lock, x, 674 x | LK_SHARED_WAITERS)) { 675 sleepq_release(&lk->lock_object); 676 continue; 677 } 678 LOCK_LOG2(lk, "%s: %p set shared waiters flag", 679 __func__, lk); 680 } 681 682 /* 683 * As far as we have been unable to acquire the 684 * shared lock and the shared waiters flag is set, 685 * we will sleep. 686 */ 687 error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, 688 SQ_SHARED_QUEUE); 689 flags &= ~LK_INTERLOCK; 690 if (error) { 691 LOCK_LOG3(lk, 692 "%s: interrupted sleep for %p with %d", 693 __func__, lk, error); 694 break; 695 } 696 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 697 __func__, lk); 698 } 699 if (error == 0) { 700 lock_profile_obtain_lock_success(&lk->lock_object, 701 contested, waittime, file, line); 702 LOCK_LOG_LOCK("SLOCK", &lk->lock_object, 0, 0, file, 703 line); 704 WITNESS_LOCK(&lk->lock_object, LK_TRYWIT(flags), file, 705 line); 706 TD_LOCKS_INC(curthread); 707 TD_SLOCKS_INC(curthread); 708 STACK_SAVE(lk); 709 } 710 break; 711 case LK_UPGRADE: 712 case LK_TRYUPGRADE: 713 _lockmgr_assert(lk, KA_SLOCKED, file, line); 714 v = lk->lk_lock; 715 x = v & LK_ALL_WAITERS; 716 v &= LK_EXCLUSIVE_SPINNERS; 717 718 /* 719 * Try to switch from one shared lock to an exclusive one. 720 * We need to preserve waiters flags during the operation. 721 */ 722 if (atomic_cmpset_ptr(&lk->lk_lock, LK_SHARERS_LOCK(1) | x | v, 723 tid | x)) { 724 LOCK_LOG_LOCK("XUPGRADE", &lk->lock_object, 0, 0, file, 725 line); 726 WITNESS_UPGRADE(&lk->lock_object, LOP_EXCLUSIVE | 727 LK_TRYWIT(flags), file, line); 728 TD_SLOCKS_DEC(curthread); 729 break; 730 } 731 732 /* 733 * In LK_TRYUPGRADE mode, do not drop the lock, 734 * returning EBUSY instead. 735 */ 736 if (op == LK_TRYUPGRADE) { 737 LOCK_LOG2(lk, "%s: %p failed the nowait upgrade", 738 __func__, lk); 739 error = EBUSY; 740 break; 741 } 742 743 /* 744 * We have been unable to succeed in upgrading, so just 745 * give up the shared lock. 746 */ 747 wakeup_swapper |= wakeupshlk(lk, file, line); 748 749 /* FALLTHROUGH */ 750 case LK_EXCLUSIVE: 751 if (LK_CAN_WITNESS(flags)) 752 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 753 LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? 754 ilk : NULL); 755 756 /* 757 * If curthread already holds the lock and this one is 758 * allowed to recurse, simply recurse on it. 759 */ 760 if (lockmgr_xlocked(lk)) { 761 if ((flags & LK_CANRECURSE) == 0 && 762 (lk->lock_object.lo_flags & LO_RECURSABLE) == 0) { 763 764 /* 765 * If the lock is expected to not panic just 766 * give up and return. 767 */ 768 if (LK_TRYOP(flags)) { 769 LOCK_LOG2(lk, 770 "%s: %p fails the try operation", 771 __func__, lk); 772 error = EBUSY; 773 break; 774 } 775 if (flags & LK_INTERLOCK) 776 class->lc_unlock(ilk); 777 panic("%s: recursing on non recursive lockmgr %s @ %s:%d\n", 778 __func__, iwmesg, file, line); 779 } 780 lk->lk_recurse++; 781 LOCK_LOG2(lk, "%s: %p recursing", __func__, lk); 782 LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, 783 lk->lk_recurse, file, line); 784 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | 785 LK_TRYWIT(flags), file, line); 786 TD_LOCKS_INC(curthread); 787 break; 788 } 789 790 while (!atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, 791 tid)) { 792 #ifdef HWPMC_HOOKS 793 PMC_SOFT_CALL( , , lock, failed); 794 #endif 795 lock_profile_obtain_lock_failed(&lk->lock_object, 796 &contested, &waittime); 797 798 /* 799 * If the lock is expected to not sleep just give up 800 * and return. 801 */ 802 if (LK_TRYOP(flags)) { 803 LOCK_LOG2(lk, "%s: %p fails the try operation", 804 __func__, lk); 805 error = EBUSY; 806 break; 807 } 808 809 #ifdef ADAPTIVE_LOCKMGRS 810 /* 811 * If the owner is running on another CPU, spin until 812 * the owner stops running or the state of the lock 813 * changes. 814 */ 815 x = lk->lk_lock; 816 if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 && 817 LK_HOLDER(x) != LK_KERNPROC) { 818 owner = (struct thread *)LK_HOLDER(x); 819 if (LOCK_LOG_TEST(&lk->lock_object, 0)) 820 CTR3(KTR_LOCK, 821 "%s: spinning on %p held by %p", 822 __func__, lk, owner); 823 KTR_STATE1(KTR_SCHED, "thread", 824 sched_tdname(td), "spinning", 825 "lockname:\"%s\"", lk->lock_object.lo_name); 826 827 /* 828 * If we are holding also an interlock drop it 829 * in order to avoid a deadlock if the lockmgr 830 * owner is adaptively spinning on the 831 * interlock itself. 832 */ 833 if (flags & LK_INTERLOCK) { 834 class->lc_unlock(ilk); 835 flags &= ~LK_INTERLOCK; 836 } 837 GIANT_SAVE(); 838 while (LK_HOLDER(lk->lk_lock) == 839 (uintptr_t)owner && TD_IS_RUNNING(owner)) 840 cpu_spinwait(); 841 KTR_STATE0(KTR_SCHED, "thread", 842 sched_tdname(td), "running"); 843 GIANT_RESTORE(); 844 continue; 845 } else if (LK_CAN_ADAPT(lk, flags) && 846 (x & LK_SHARE) != 0 && LK_SHARERS(x) && 847 spintries < alk_retries) { 848 if ((x & LK_EXCLUSIVE_SPINNERS) == 0 && 849 !atomic_cmpset_ptr(&lk->lk_lock, x, 850 x | LK_EXCLUSIVE_SPINNERS)) 851 continue; 852 KTR_STATE1(KTR_SCHED, "thread", 853 sched_tdname(td), "spinning", 854 "lockname:\"%s\"", lk->lock_object.lo_name); 855 if (flags & LK_INTERLOCK) { 856 class->lc_unlock(ilk); 857 flags &= ~LK_INTERLOCK; 858 } 859 GIANT_SAVE(); 860 spintries++; 861 for (i = 0; i < alk_loops; i++) { 862 if (LOCK_LOG_TEST(&lk->lock_object, 0)) 863 CTR4(KTR_LOCK, 864 "%s: shared spinning on %p with %u and %u", 865 __func__, lk, spintries, i); 866 if ((lk->lk_lock & 867 LK_EXCLUSIVE_SPINNERS) == 0) 868 break; 869 cpu_spinwait(); 870 } 871 KTR_STATE0(KTR_SCHED, "thread", 872 sched_tdname(td), "running"); 873 GIANT_RESTORE(); 874 if (i != alk_loops) 875 continue; 876 } 877 #endif 878 879 /* 880 * Acquire the sleepqueue chain lock because we 881 * probabilly will need to manipulate waiters flags. 882 */ 883 sleepq_lock(&lk->lock_object); 884 x = lk->lk_lock; 885 886 /* 887 * if the lock has been released while we spun on 888 * the sleepqueue chain lock just try again. 889 */ 890 if (x == LK_UNLOCKED) { 891 sleepq_release(&lk->lock_object); 892 continue; 893 } 894 895 #ifdef ADAPTIVE_LOCKMGRS 896 /* 897 * The current lock owner might have started executing 898 * on another CPU (or the lock could have changed 899 * owner) while we were waiting on the turnstile 900 * chain lock. If so, drop the turnstile lock and try 901 * again. 902 */ 903 if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 && 904 LK_HOLDER(x) != LK_KERNPROC) { 905 owner = (struct thread *)LK_HOLDER(x); 906 if (TD_IS_RUNNING(owner)) { 907 sleepq_release(&lk->lock_object); 908 continue; 909 } 910 } 911 #endif 912 913 /* 914 * The lock can be in the state where there is a 915 * pending queue of waiters, but still no owner. 916 * This happens when the lock is contested and an 917 * owner is going to claim the lock. 918 * If curthread is the one successfully acquiring it 919 * claim lock ownership and return, preserving waiters 920 * flags. 921 */ 922 v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 923 if ((x & ~v) == LK_UNLOCKED) { 924 v &= ~LK_EXCLUSIVE_SPINNERS; 925 if (atomic_cmpset_acq_ptr(&lk->lk_lock, x, 926 tid | v)) { 927 sleepq_release(&lk->lock_object); 928 LOCK_LOG2(lk, 929 "%s: %p claimed by a new writer", 930 __func__, lk); 931 break; 932 } 933 sleepq_release(&lk->lock_object); 934 continue; 935 } 936 937 /* 938 * Try to set the LK_EXCLUSIVE_WAITERS flag. If we 939 * fail, loop back and retry. 940 */ 941 if ((x & LK_EXCLUSIVE_WAITERS) == 0) { 942 if (!atomic_cmpset_ptr(&lk->lk_lock, x, 943 x | LK_EXCLUSIVE_WAITERS)) { 944 sleepq_release(&lk->lock_object); 945 continue; 946 } 947 LOCK_LOG2(lk, "%s: %p set excl waiters flag", 948 __func__, lk); 949 } 950 951 /* 952 * As far as we have been unable to acquire the 953 * exclusive lock and the exclusive waiters flag 954 * is set, we will sleep. 955 */ 956 error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo, 957 SQ_EXCLUSIVE_QUEUE); 958 flags &= ~LK_INTERLOCK; 959 if (error) { 960 LOCK_LOG3(lk, 961 "%s: interrupted sleep for %p with %d", 962 __func__, lk, error); 963 break; 964 } 965 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 966 __func__, lk); 967 } 968 if (error == 0) { 969 lock_profile_obtain_lock_success(&lk->lock_object, 970 contested, waittime, file, line); 971 LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0, 972 lk->lk_recurse, file, line); 973 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | 974 LK_TRYWIT(flags), file, line); 975 TD_LOCKS_INC(curthread); 976 STACK_SAVE(lk); 977 } 978 break; 979 case LK_DOWNGRADE: 980 _lockmgr_assert(lk, KA_XLOCKED, file, line); 981 LOCK_LOG_LOCK("XDOWNGRADE", &lk->lock_object, 0, 0, file, line); 982 WITNESS_DOWNGRADE(&lk->lock_object, 0, file, line); 983 984 /* 985 * Panic if the lock is recursed. 986 */ 987 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) { 988 if (flags & LK_INTERLOCK) 989 class->lc_unlock(ilk); 990 panic("%s: downgrade a recursed lockmgr %s @ %s:%d\n", 991 __func__, iwmesg, file, line); 992 } 993 TD_SLOCKS_INC(curthread); 994 995 /* 996 * In order to preserve waiters flags, just spin. 997 */ 998 for (;;) { 999 x = lk->lk_lock; 1000 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1001 x &= LK_ALL_WAITERS; 1002 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, 1003 LK_SHARERS_LOCK(1) | x)) 1004 break; 1005 cpu_spinwait(); 1006 } 1007 break; 1008 case LK_RELEASE: 1009 _lockmgr_assert(lk, KA_LOCKED, file, line); 1010 x = lk->lk_lock; 1011 1012 if ((x & LK_SHARE) == 0) { 1013 1014 /* 1015 * As first option, treact the lock as if it has not 1016 * any waiter. 1017 * Fix-up the tid var if the lock has been disowned. 1018 */ 1019 if (LK_HOLDER(x) == LK_KERNPROC) 1020 tid = LK_KERNPROC; 1021 else { 1022 WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, 1023 file, line); 1024 TD_LOCKS_DEC(curthread); 1025 } 1026 LOCK_LOG_LOCK("XUNLOCK", &lk->lock_object, 0, 1027 lk->lk_recurse, file, line); 1028 1029 /* 1030 * The lock is held in exclusive mode. 1031 * If the lock is recursed also, then unrecurse it. 1032 */ 1033 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) { 1034 LOCK_LOG2(lk, "%s: %p unrecursing", __func__, 1035 lk); 1036 lk->lk_recurse--; 1037 break; 1038 } 1039 if (tid != LK_KERNPROC) 1040 lock_profile_release_lock(&lk->lock_object); 1041 1042 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid, 1043 LK_UNLOCKED)) 1044 break; 1045 1046 sleepq_lock(&lk->lock_object); 1047 x = lk->lk_lock; 1048 v = LK_UNLOCKED; 1049 1050 /* 1051 * If the lock has exclusive waiters, give them 1052 * preference in order to avoid deadlock with 1053 * shared runners up. 1054 * If interruptible sleeps left the exclusive queue 1055 * empty avoid a starvation for the threads sleeping 1056 * on the shared queue by giving them precedence 1057 * and cleaning up the exclusive waiters bit anyway. 1058 * Please note that lk_exslpfail count may be lying 1059 * about the real number of waiters with the 1060 * LK_SLEEPFAIL flag on because they may be used in 1061 * conjunction with interruptible sleeps so 1062 * lk_exslpfail might be considered an 'upper limit' 1063 * bound, including the edge cases. 1064 */ 1065 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1066 realexslp = sleepq_sleepcnt(&lk->lock_object, 1067 SQ_EXCLUSIVE_QUEUE); 1068 if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) { 1069 if (lk->lk_exslpfail < realexslp) { 1070 lk->lk_exslpfail = 0; 1071 queue = SQ_EXCLUSIVE_QUEUE; 1072 v |= (x & LK_SHARED_WAITERS); 1073 } else { 1074 lk->lk_exslpfail = 0; 1075 LOCK_LOG2(lk, 1076 "%s: %p has only LK_SLEEPFAIL sleepers", 1077 __func__, lk); 1078 LOCK_LOG2(lk, 1079 "%s: %p waking up threads on the exclusive queue", 1080 __func__, lk); 1081 wakeup_swapper = 1082 sleepq_broadcast(&lk->lock_object, 1083 SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE); 1084 queue = SQ_SHARED_QUEUE; 1085 } 1086 } else { 1087 1088 /* 1089 * Exclusive waiters sleeping with LK_SLEEPFAIL 1090 * on and using interruptible sleeps/timeout 1091 * may have left spourious lk_exslpfail counts 1092 * on, so clean it up anyway. 1093 */ 1094 lk->lk_exslpfail = 0; 1095 queue = SQ_SHARED_QUEUE; 1096 } 1097 1098 LOCK_LOG3(lk, 1099 "%s: %p waking up threads on the %s queue", 1100 __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" : 1101 "exclusive"); 1102 atomic_store_rel_ptr(&lk->lk_lock, v); 1103 wakeup_swapper |= sleepq_broadcast(&lk->lock_object, 1104 SLEEPQ_LK, 0, queue); 1105 sleepq_release(&lk->lock_object); 1106 break; 1107 } else 1108 wakeup_swapper = wakeupshlk(lk, file, line); 1109 break; 1110 case LK_DRAIN: 1111 if (LK_CAN_WITNESS(flags)) 1112 WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER | 1113 LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ? 1114 ilk : NULL); 1115 1116 /* 1117 * Trying to drain a lock we already own will result in a 1118 * deadlock. 1119 */ 1120 if (lockmgr_xlocked(lk)) { 1121 if (flags & LK_INTERLOCK) 1122 class->lc_unlock(ilk); 1123 panic("%s: draining %s with the lock held @ %s:%d\n", 1124 __func__, iwmesg, file, line); 1125 } 1126 1127 while (!atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid)) { 1128 #ifdef HWPMC_HOOKS 1129 PMC_SOFT_CALL( , , lock, failed); 1130 #endif 1131 lock_profile_obtain_lock_failed(&lk->lock_object, 1132 &contested, &waittime); 1133 1134 /* 1135 * If the lock is expected to not sleep just give up 1136 * and return. 1137 */ 1138 if (LK_TRYOP(flags)) { 1139 LOCK_LOG2(lk, "%s: %p fails the try operation", 1140 __func__, lk); 1141 error = EBUSY; 1142 break; 1143 } 1144 1145 /* 1146 * Acquire the sleepqueue chain lock because we 1147 * probabilly will need to manipulate waiters flags. 1148 */ 1149 sleepq_lock(&lk->lock_object); 1150 x = lk->lk_lock; 1151 1152 /* 1153 * if the lock has been released while we spun on 1154 * the sleepqueue chain lock just try again. 1155 */ 1156 if (x == LK_UNLOCKED) { 1157 sleepq_release(&lk->lock_object); 1158 continue; 1159 } 1160 1161 v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS); 1162 if ((x & ~v) == LK_UNLOCKED) { 1163 v = (x & ~LK_EXCLUSIVE_SPINNERS); 1164 1165 /* 1166 * If interruptible sleeps left the exclusive 1167 * queue empty avoid a starvation for the 1168 * threads sleeping on the shared queue by 1169 * giving them precedence and cleaning up the 1170 * exclusive waiters bit anyway. 1171 * Please note that lk_exslpfail count may be 1172 * lying about the real number of waiters with 1173 * the LK_SLEEPFAIL flag on because they may 1174 * be used in conjunction with interruptible 1175 * sleeps so lk_exslpfail might be considered 1176 * an 'upper limit' bound, including the edge 1177 * cases. 1178 */ 1179 if (v & LK_EXCLUSIVE_WAITERS) { 1180 queue = SQ_EXCLUSIVE_QUEUE; 1181 v &= ~LK_EXCLUSIVE_WAITERS; 1182 } else { 1183 1184 /* 1185 * Exclusive waiters sleeping with 1186 * LK_SLEEPFAIL on and using 1187 * interruptible sleeps/timeout may 1188 * have left spourious lk_exslpfail 1189 * counts on, so clean it up anyway. 1190 */ 1191 MPASS(v & LK_SHARED_WAITERS); 1192 lk->lk_exslpfail = 0; 1193 queue = SQ_SHARED_QUEUE; 1194 v &= ~LK_SHARED_WAITERS; 1195 } 1196 if (queue == SQ_EXCLUSIVE_QUEUE) { 1197 realexslp = 1198 sleepq_sleepcnt(&lk->lock_object, 1199 SQ_EXCLUSIVE_QUEUE); 1200 if (lk->lk_exslpfail >= realexslp) { 1201 lk->lk_exslpfail = 0; 1202 queue = SQ_SHARED_QUEUE; 1203 v &= ~LK_SHARED_WAITERS; 1204 if (realexslp != 0) { 1205 LOCK_LOG2(lk, 1206 "%s: %p has only LK_SLEEPFAIL sleepers", 1207 __func__, lk); 1208 LOCK_LOG2(lk, 1209 "%s: %p waking up threads on the exclusive queue", 1210 __func__, lk); 1211 wakeup_swapper = 1212 sleepq_broadcast( 1213 &lk->lock_object, 1214 SLEEPQ_LK, 0, 1215 SQ_EXCLUSIVE_QUEUE); 1216 } 1217 } else 1218 lk->lk_exslpfail = 0; 1219 } 1220 if (!atomic_cmpset_ptr(&lk->lk_lock, x, v)) { 1221 sleepq_release(&lk->lock_object); 1222 continue; 1223 } 1224 LOCK_LOG3(lk, 1225 "%s: %p waking up all threads on the %s queue", 1226 __func__, lk, queue == SQ_SHARED_QUEUE ? 1227 "shared" : "exclusive"); 1228 wakeup_swapper |= sleepq_broadcast( 1229 &lk->lock_object, SLEEPQ_LK, 0, queue); 1230 1231 /* 1232 * If shared waiters have been woken up we need 1233 * to wait for one of them to acquire the lock 1234 * before to set the exclusive waiters in 1235 * order to avoid a deadlock. 1236 */ 1237 if (queue == SQ_SHARED_QUEUE) { 1238 for (v = lk->lk_lock; 1239 (v & LK_SHARE) && !LK_SHARERS(v); 1240 v = lk->lk_lock) 1241 cpu_spinwait(); 1242 } 1243 } 1244 1245 /* 1246 * Try to set the LK_EXCLUSIVE_WAITERS flag. If we 1247 * fail, loop back and retry. 1248 */ 1249 if ((x & LK_EXCLUSIVE_WAITERS) == 0) { 1250 if (!atomic_cmpset_ptr(&lk->lk_lock, x, 1251 x | LK_EXCLUSIVE_WAITERS)) { 1252 sleepq_release(&lk->lock_object); 1253 continue; 1254 } 1255 LOCK_LOG2(lk, "%s: %p set drain waiters flag", 1256 __func__, lk); 1257 } 1258 1259 /* 1260 * As far as we have been unable to acquire the 1261 * exclusive lock and the exclusive waiters flag 1262 * is set, we will sleep. 1263 */ 1264 if (flags & LK_INTERLOCK) { 1265 class->lc_unlock(ilk); 1266 flags &= ~LK_INTERLOCK; 1267 } 1268 GIANT_SAVE(); 1269 sleepq_add(&lk->lock_object, NULL, iwmesg, SLEEPQ_LK, 1270 SQ_EXCLUSIVE_QUEUE); 1271 sleepq_wait(&lk->lock_object, ipri & PRIMASK); 1272 GIANT_RESTORE(); 1273 LOCK_LOG2(lk, "%s: %p resuming from the sleep queue", 1274 __func__, lk); 1275 } 1276 1277 if (error == 0) { 1278 lock_profile_obtain_lock_success(&lk->lock_object, 1279 contested, waittime, file, line); 1280 LOCK_LOG_LOCK("DRAIN", &lk->lock_object, 0, 1281 lk->lk_recurse, file, line); 1282 WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE | 1283 LK_TRYWIT(flags), file, line); 1284 TD_LOCKS_INC(curthread); 1285 STACK_SAVE(lk); 1286 } 1287 break; 1288 default: 1289 if (flags & LK_INTERLOCK) 1290 class->lc_unlock(ilk); 1291 panic("%s: unknown lockmgr request 0x%x\n", __func__, op); 1292 } 1293 1294 if (flags & LK_INTERLOCK) 1295 class->lc_unlock(ilk); 1296 if (wakeup_swapper) 1297 kick_proc0(); 1298 1299 return (error); 1300 } 1301 1302 void 1303 _lockmgr_disown(struct lock *lk, const char *file, int line) 1304 { 1305 uintptr_t tid, x; 1306 1307 if (SCHEDULER_STOPPED()) 1308 return; 1309 1310 tid = (uintptr_t)curthread; 1311 _lockmgr_assert(lk, KA_XLOCKED, file, line); 1312 1313 /* 1314 * Panic if the lock is recursed. 1315 */ 1316 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) 1317 panic("%s: disown a recursed lockmgr @ %s:%d\n", 1318 __func__, file, line); 1319 1320 /* 1321 * If the owner is already LK_KERNPROC just skip the whole operation. 1322 */ 1323 if (LK_HOLDER(lk->lk_lock) != tid) 1324 return; 1325 lock_profile_release_lock(&lk->lock_object); 1326 LOCK_LOG_LOCK("XDISOWN", &lk->lock_object, 0, 0, file, line); 1327 WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line); 1328 TD_LOCKS_DEC(curthread); 1329 STACK_SAVE(lk); 1330 1331 /* 1332 * In order to preserve waiters flags, just spin. 1333 */ 1334 for (;;) { 1335 x = lk->lk_lock; 1336 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0); 1337 x &= LK_ALL_WAITERS; 1338 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x, 1339 LK_KERNPROC | x)) 1340 return; 1341 cpu_spinwait(); 1342 } 1343 } 1344 1345 void 1346 lockmgr_printinfo(const struct lock *lk) 1347 { 1348 struct thread *td; 1349 uintptr_t x; 1350 1351 if (lk->lk_lock == LK_UNLOCKED) 1352 printf("lock type %s: UNLOCKED\n", lk->lock_object.lo_name); 1353 else if (lk->lk_lock & LK_SHARE) 1354 printf("lock type %s: SHARED (count %ju)\n", 1355 lk->lock_object.lo_name, 1356 (uintmax_t)LK_SHARERS(lk->lk_lock)); 1357 else { 1358 td = lockmgr_xholder(lk); 1359 if (td == (struct thread *)LK_KERNPROC) 1360 printf("lock type %s: EXCL by KERNPROC\n", 1361 lk->lock_object.lo_name); 1362 else 1363 printf("lock type %s: EXCL by thread %p " 1364 "(pid %d, %s, tid %d)\n", lk->lock_object.lo_name, 1365 td, td->td_proc->p_pid, td->td_proc->p_comm, 1366 td->td_tid); 1367 } 1368 1369 x = lk->lk_lock; 1370 if (x & LK_EXCLUSIVE_WAITERS) 1371 printf(" with exclusive waiters pending\n"); 1372 if (x & LK_SHARED_WAITERS) 1373 printf(" with shared waiters pending\n"); 1374 if (x & LK_EXCLUSIVE_SPINNERS) 1375 printf(" with exclusive spinners pending\n"); 1376 1377 STACK_PRINT(lk); 1378 } 1379 1380 int 1381 lockstatus(const struct lock *lk) 1382 { 1383 uintptr_t v, x; 1384 int ret; 1385 1386 ret = LK_SHARED; 1387 x = lk->lk_lock; 1388 v = LK_HOLDER(x); 1389 1390 if ((x & LK_SHARE) == 0) { 1391 if (v == (uintptr_t)curthread || v == LK_KERNPROC) 1392 ret = LK_EXCLUSIVE; 1393 else 1394 ret = LK_EXCLOTHER; 1395 } else if (x == LK_UNLOCKED) 1396 ret = 0; 1397 1398 return (ret); 1399 } 1400 1401 #ifdef INVARIANT_SUPPORT 1402 1403 FEATURE(invariant_support, 1404 "Support for modules compiled with INVARIANTS option"); 1405 1406 #ifndef INVARIANTS 1407 #undef _lockmgr_assert 1408 #endif 1409 1410 void 1411 _lockmgr_assert(const struct lock *lk, int what, const char *file, int line) 1412 { 1413 int slocked = 0; 1414 1415 if (panicstr != NULL) 1416 return; 1417 switch (what) { 1418 case KA_SLOCKED: 1419 case KA_SLOCKED | KA_NOTRECURSED: 1420 case KA_SLOCKED | KA_RECURSED: 1421 slocked = 1; 1422 case KA_LOCKED: 1423 case KA_LOCKED | KA_NOTRECURSED: 1424 case KA_LOCKED | KA_RECURSED: 1425 #ifdef WITNESS 1426 1427 /* 1428 * We cannot trust WITNESS if the lock is held in exclusive 1429 * mode and a call to lockmgr_disown() happened. 1430 * Workaround this skipping the check if the lock is held in 1431 * exclusive mode even for the KA_LOCKED case. 1432 */ 1433 if (slocked || (lk->lk_lock & LK_SHARE)) { 1434 witness_assert(&lk->lock_object, what, file, line); 1435 break; 1436 } 1437 #endif 1438 if (lk->lk_lock == LK_UNLOCKED || 1439 ((lk->lk_lock & LK_SHARE) == 0 && (slocked || 1440 (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk))))) 1441 panic("Lock %s not %slocked @ %s:%d\n", 1442 lk->lock_object.lo_name, slocked ? "share" : "", 1443 file, line); 1444 1445 if ((lk->lk_lock & LK_SHARE) == 0) { 1446 if (lockmgr_recursed(lk)) { 1447 if (what & KA_NOTRECURSED) 1448 panic("Lock %s recursed @ %s:%d\n", 1449 lk->lock_object.lo_name, file, 1450 line); 1451 } else if (what & KA_RECURSED) 1452 panic("Lock %s not recursed @ %s:%d\n", 1453 lk->lock_object.lo_name, file, line); 1454 } 1455 break; 1456 case KA_XLOCKED: 1457 case KA_XLOCKED | KA_NOTRECURSED: 1458 case KA_XLOCKED | KA_RECURSED: 1459 if (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk)) 1460 panic("Lock %s not exclusively locked @ %s:%d\n", 1461 lk->lock_object.lo_name, file, line); 1462 if (lockmgr_recursed(lk)) { 1463 if (what & KA_NOTRECURSED) 1464 panic("Lock %s recursed @ %s:%d\n", 1465 lk->lock_object.lo_name, file, line); 1466 } else if (what & KA_RECURSED) 1467 panic("Lock %s not recursed @ %s:%d\n", 1468 lk->lock_object.lo_name, file, line); 1469 break; 1470 case KA_UNLOCKED: 1471 if (lockmgr_xlocked(lk) || lockmgr_disowned(lk)) 1472 panic("Lock %s exclusively locked @ %s:%d\n", 1473 lk->lock_object.lo_name, file, line); 1474 break; 1475 default: 1476 panic("Unknown lockmgr assertion: %d @ %s:%d\n", what, file, 1477 line); 1478 } 1479 } 1480 #endif 1481 1482 #ifdef DDB 1483 int 1484 lockmgr_chain(struct thread *td, struct thread **ownerp) 1485 { 1486 struct lock *lk; 1487 1488 lk = td->td_wchan; 1489 1490 if (LOCK_CLASS(&lk->lock_object) != &lock_class_lockmgr) 1491 return (0); 1492 db_printf("blocked on lockmgr %s", lk->lock_object.lo_name); 1493 if (lk->lk_lock & LK_SHARE) 1494 db_printf("SHARED (count %ju)\n", 1495 (uintmax_t)LK_SHARERS(lk->lk_lock)); 1496 else 1497 db_printf("EXCL\n"); 1498 *ownerp = lockmgr_xholder(lk); 1499 1500 return (1); 1501 } 1502 1503 static void 1504 db_show_lockmgr(const struct lock_object *lock) 1505 { 1506 struct thread *td; 1507 const struct lock *lk; 1508 1509 lk = (const struct lock *)lock; 1510 1511 db_printf(" state: "); 1512 if (lk->lk_lock == LK_UNLOCKED) 1513 db_printf("UNLOCKED\n"); 1514 else if (lk->lk_lock & LK_SHARE) 1515 db_printf("SLOCK: %ju\n", (uintmax_t)LK_SHARERS(lk->lk_lock)); 1516 else { 1517 td = lockmgr_xholder(lk); 1518 if (td == (struct thread *)LK_KERNPROC) 1519 db_printf("XLOCK: LK_KERNPROC\n"); 1520 else 1521 db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td, 1522 td->td_tid, td->td_proc->p_pid, 1523 td->td_proc->p_comm); 1524 if (lockmgr_recursed(lk)) 1525 db_printf(" recursed: %d\n", lk->lk_recurse); 1526 } 1527 db_printf(" waiters: "); 1528 switch (lk->lk_lock & LK_ALL_WAITERS) { 1529 case LK_SHARED_WAITERS: 1530 db_printf("shared\n"); 1531 break; 1532 case LK_EXCLUSIVE_WAITERS: 1533 db_printf("exclusive\n"); 1534 break; 1535 case LK_ALL_WAITERS: 1536 db_printf("shared and exclusive\n"); 1537 break; 1538 default: 1539 db_printf("none\n"); 1540 } 1541 db_printf(" spinners: "); 1542 if (lk->lk_lock & LK_EXCLUSIVE_SPINNERS) 1543 db_printf("exclusive\n"); 1544 else 1545 db_printf("none\n"); 1546 } 1547 #endif 1548