sys/kern/kern_lock.c

   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                 for (;;) {
 791                         if (lk->lk_lock == LK_UNLOCKED &&
 792                             atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid))
 793                                 break;
 794 #ifdef HWPMC_HOOKS
 795                         PMC_SOFT_CALL( , , lock, failed);
 796 #endif
 797                         lock_profile_obtain_lock_failed(&lk->lock_object,
 798                             &contested, &waittime);
 799
 800                         /*
 801                          * If the lock is expected to not sleep just give up
 802                          * and return.
 803                          */
 804                         if (LK_TRYOP(flags)) {
 805                                 LOCK_LOG2(lk, "%s: %p fails the try operation",
 806                                     __func__, lk);
 807                                 error = EBUSY;
 808                                 break;
 809                         }
 810
 811 #ifdef ADAPTIVE_LOCKMGRS
 812                         /*
 813                          * If the owner is running on another CPU, spin until
 814                          * the owner stops running or the state of the lock
 815                          * changes.
 816                          */
 817                         x = lk->lk_lock;
 818                         if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
 819                             LK_HOLDER(x) != LK_KERNPROC) {
 820                                 owner = (struct thread *)LK_HOLDER(x);
 821                                 if (LOCK_LOG_TEST(&lk->lock_object, 0))
 822                                         CTR3(KTR_LOCK,
 823                                             "%s: spinning on %p held by %p",
 824                                             __func__, lk, owner);
 825                                 KTR_STATE1(KTR_SCHED, "thread",
 826                                     sched_tdname(td), "spinning",
 827                                     "lockname:\"%s\"", lk->lock_object.lo_name);
 828
 829                                 /*
 830                                  * If we are holding also an interlock drop it
 831                                  * in order to avoid a deadlock if the lockmgr
 832                                  * owner is adaptively spinning on the
 833                                  * interlock itself.
 834                                  */
 835                                 if (flags & LK_INTERLOCK) {
 836                                         class->lc_unlock(ilk);
 837                                         flags &= ~LK_INTERLOCK;
 838                                 }
 839                                 GIANT_SAVE();
 840                                 while (LK_HOLDER(lk->lk_lock) ==
 841                                     (uintptr_t)owner && TD_IS_RUNNING(owner))
 842                                         cpu_spinwait();
 843                                 KTR_STATE0(KTR_SCHED, "thread",
 844                                     sched_tdname(td), "running");
 845                                 GIANT_RESTORE();
 846                                 continue;
 847                         } else if (LK_CAN_ADAPT(lk, flags) &&
 848                             (x & LK_SHARE) != 0 && LK_SHARERS(x) &&
 849                             spintries < alk_retries) {
 850                                 if ((x & LK_EXCLUSIVE_SPINNERS) == 0 &&
 851                                     !atomic_cmpset_ptr(&lk->lk_lock, x,
 852                                     x | LK_EXCLUSIVE_SPINNERS))
 853                                         continue;
 854                                 KTR_STATE1(KTR_SCHED, "thread",
 855                                     sched_tdname(td), "spinning",
 856                                     "lockname:\"%s\"", lk->lock_object.lo_name);
 857                                 if (flags & LK_INTERLOCK) {
 858                                         class->lc_unlock(ilk);
 859                                         flags &= ~LK_INTERLOCK;
 860                                 }
 861                                 GIANT_SAVE();
 862                                 spintries++;
 863                                 for (i = 0; i < alk_loops; i++) {
 864                                         if (LOCK_LOG_TEST(&lk->lock_object, 0))
 865                                                 CTR4(KTR_LOCK,
 866                                     "%s: shared spinning on %p with %u and %u",
 867                                                     __func__, lk, spintries, i);
 868                                         if ((lk->lk_lock &
 869                                             LK_EXCLUSIVE_SPINNERS) == 0)
 870                                                 break;
 871                                         cpu_spinwait();
 872                                 }
 873                                 KTR_STATE0(KTR_SCHED, "thread",
 874                                     sched_tdname(td), "running");
 875                                 GIANT_RESTORE();
 876                                 if (i != alk_loops)
 877                                         continue;
 878                         }
 879 #endif
 880
 881                         /*
 882                          * Acquire the sleepqueue chain lock because we
 883                          * probabilly will need to manipulate waiters flags.
 884                          */
 885                         sleepq_lock(&lk->lock_object);
 886                         x = lk->lk_lock;
 887
 888                         /*
 889                          * if the lock has been released while we spun on
 890                          * the sleepqueue chain lock just try again.
 891                          */
 892                         if (x == LK_UNLOCKED) {
 893                                 sleepq_release(&lk->lock_object);
 894                                 continue;
 895                         }
 896
 897 #ifdef ADAPTIVE_LOCKMGRS
 898                         /*
 899                          * The current lock owner might have started executing
 900                          * on another CPU (or the lock could have changed
 901                          * owner) while we were waiting on the turnstile
 902                          * chain lock.  If so, drop the turnstile lock and try
 903                          * again.
 904                          */
 905                         if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
 906                             LK_HOLDER(x) != LK_KERNPROC) {
 907                                 owner = (struct thread *)LK_HOLDER(x);
 908                                 if (TD_IS_RUNNING(owner)) {
 909                                         sleepq_release(&lk->lock_object);
 910                                         continue;
 911                                 }
 912                         }
 913 #endif
 914
 915                         /*
 916                          * The lock can be in the state where there is a
 917                          * pending queue of waiters, but still no owner.
 918                          * This happens when the lock is contested and an
 919                          * owner is going to claim the lock.
 920                          * If curthread is the one successfully acquiring it
 921                          * claim lock ownership and return, preserving waiters
 922                          * flags.
 923                          */
 924                         v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS);
 925                         if ((x & ~v) == LK_UNLOCKED) {
 926                                 v &= ~LK_EXCLUSIVE_SPINNERS;
 927                                 if (atomic_cmpset_acq_ptr(&lk->lk_lock, x,
 928                                     tid | v)) {
 929                                         sleepq_release(&lk->lock_object);
 930                                         LOCK_LOG2(lk,
 931                                             "%s: %p claimed by a new writer",
 932                                             __func__, lk);
 933                                         break;
 934                                 }
 935                                 sleepq_release(&lk->lock_object);
 936                                 continue;
 937                         }
 938
 939                         /*
 940                          * Try to set the LK_EXCLUSIVE_WAITERS flag.  If we
 941                          * fail, loop back and retry.
 942                          */
 943                         if ((x & LK_EXCLUSIVE_WAITERS) == 0) {
 944                                 if (!atomic_cmpset_ptr(&lk->lk_lock, x,
 945                                     x | LK_EXCLUSIVE_WAITERS)) {
 946                                         sleepq_release(&lk->lock_object);
 947                                         continue;
 948                                 }
 949                                 LOCK_LOG2(lk, "%s: %p set excl waiters flag",
 950                                     __func__, lk);
 951                         }
 952
 953                         /*
 954                          * As far as we have been unable to acquire the
 955                          * exclusive lock and the exclusive waiters flag
 956                          * is set, we will sleep.
 957                          */
 958                         error = sleeplk(lk, flags, ilk, iwmesg, ipri, itimo,
 959                             SQ_EXCLUSIVE_QUEUE);
 960                         flags &= ~LK_INTERLOCK;
 961                         if (error) {
 962                                 LOCK_LOG3(lk,
 963                                     "%s: interrupted sleep for %p with %d",
 964                                     __func__, lk, error);
 965                                 break;
 966                         }
 967                         LOCK_LOG2(lk, "%s: %p resuming from the sleep queue",
 968                             __func__, lk);
 969                 }
 970                 if (error == 0) {
 971                         lock_profile_obtain_lock_success(&lk->lock_object,
 972                             contested, waittime, file, line);
 973                         LOCK_LOG_LOCK("XLOCK", &lk->lock_object, 0,
 974                             lk->lk_recurse, file, line);
 975                         WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE |
 976                             LK_TRYWIT(flags), file, line);
 977                         TD_LOCKS_INC(curthread);
 978                         STACK_SAVE(lk);
 979                 }
 980                 break;
 981         case LK_DOWNGRADE:
 982                 _lockmgr_assert(lk, KA_XLOCKED, file, line);
 983                 LOCK_LOG_LOCK("XDOWNGRADE", &lk->lock_object, 0, 0, file, line);
 984                 WITNESS_DOWNGRADE(&lk->lock_object, 0, file, line);
 985
 986                 /*
 987                  * Panic if the lock is recursed.
 988                  */
 989                 if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) {
 990                         if (flags & LK_INTERLOCK)
 991                                 class->lc_unlock(ilk);
 992                         panic("%s: downgrade a recursed lockmgr %s @ %s:%d\n",
 993                             __func__, iwmesg, file, line);
 994                 }
 995                 TD_SLOCKS_INC(curthread);
 996
 997                 /*
 998                  * In order to preserve waiters flags, just spin.
 999                  */
1000                 for (;;) {
1001                         x = lk->lk_lock;
1002                         MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
1003                         x &= LK_ALL_WAITERS;
1004                         if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x,
1005                             LK_SHARERS_LOCK(1) | x))
1006                                 break;
1007                         cpu_spinwait();
1008                 }
1009                 break;
1010         case LK_RELEASE:
1011                 _lockmgr_assert(lk, KA_LOCKED, file, line);
1012                 x = lk->lk_lock;
1013
1014                 if ((x & LK_SHARE) == 0) {
1015
1016                         /*
1017                          * As first option, treact the lock as if it has not
1018                          * any waiter.
1019                          * Fix-up the tid var if the lock has been disowned.
1020                          */
1021                         if (LK_HOLDER(x) == LK_KERNPROC)
1022                                 tid = LK_KERNPROC;
1023                         else {
1024                                 WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE,
1025                                     file, line);
1026                                 TD_LOCKS_DEC(curthread);
1027                         }
1028                         LOCK_LOG_LOCK("XUNLOCK", &lk->lock_object, 0,
1029                             lk->lk_recurse, file, line);
1030
1031                         /*
1032                          * The lock is held in exclusive mode.
1033                          * If the lock is recursed also, then unrecurse it.
1034                          */
1035                         if (lockmgr_xlocked(lk) && lockmgr_recursed(lk)) {
1036                                 LOCK_LOG2(lk, "%s: %p unrecursing", __func__,
1037                                     lk);
1038                                 lk->lk_recurse--;
1039                                 break;
1040                         }
1041                         if (tid != LK_KERNPROC)
1042                                 lock_profile_release_lock(&lk->lock_object);
1043
1044                         if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid,
1045                             LK_UNLOCKED))
1046                                 break;
1047
1048                         sleepq_lock(&lk->lock_object);
1049                         x = lk->lk_lock;
1050                         v = LK_UNLOCKED;
1051
1052                         /*
1053                          * If the lock has exclusive waiters, give them
1054                          * preference in order to avoid deadlock with
1055                          * shared runners up.
1056                          * If interruptible sleeps left the exclusive queue
1057                          * empty avoid a starvation for the threads sleeping
1058                          * on the shared queue by giving them precedence
1059                          * and cleaning up the exclusive waiters bit anyway.
1060                          * Please note that lk_exslpfail count may be lying
1061                          * about the real number of waiters with the
1062                          * LK_SLEEPFAIL flag on because they may be used in
1063                          * conjunction with interruptible sleeps so
1064                          * lk_exslpfail might be considered an 'upper limit'
1065                          * bound, including the edge cases.
1066                          */
1067                         MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
1068                         realexslp = sleepq_sleepcnt(&lk->lock_object,
1069                             SQ_EXCLUSIVE_QUEUE);
1070                         if ((x & LK_EXCLUSIVE_WAITERS) != 0 && realexslp != 0) {
1071                                 if (lk->lk_exslpfail < realexslp) {
1072                                         lk->lk_exslpfail = 0;
1073                                         queue = SQ_EXCLUSIVE_QUEUE;
1074                                         v |= (x & LK_SHARED_WAITERS);
1075                                 } else {
1076                                         lk->lk_exslpfail = 0;
1077                                         LOCK_LOG2(lk,
1078                                         "%s: %p has only LK_SLEEPFAIL sleepers",
1079                                             __func__, lk);
1080                                         LOCK_LOG2(lk,
1081                         "%s: %p waking up threads on the exclusive queue",
1082                                             __func__, lk);
1083                                         wakeup_swapper =
1084                                             sleepq_broadcast(&lk->lock_object,
1085                                             SLEEPQ_LK, 0, SQ_EXCLUSIVE_QUEUE);
1086                                         queue = SQ_SHARED_QUEUE;
1087                                 }
1088                         } else {
1089
1090                                 /*
1091                                  * Exclusive waiters sleeping with LK_SLEEPFAIL
1092                                  * on and using interruptible sleeps/timeout
1093                                  * may have left spourious lk_exslpfail counts
1094                                  * on, so clean it up anyway.
1095                                  */
1096                                 lk->lk_exslpfail = 0;
1097                                 queue = SQ_SHARED_QUEUE;
1098                         }
1099
1100                         LOCK_LOG3(lk,
1101                             "%s: %p waking up threads on the %s queue",
1102                             __func__, lk, queue == SQ_SHARED_QUEUE ? "shared" :
1103                             "exclusive");
1104                         atomic_store_rel_ptr(&lk->lk_lock, v);
1105                         wakeup_swapper |= sleepq_broadcast(&lk->lock_object,
1106                             SLEEPQ_LK, 0, queue);
1107                         sleepq_release(&lk->lock_object);
1108                         break;
1109                 } else
1110                         wakeup_swapper = wakeupshlk(lk, file, line);
1111                 break;
1112         case LK_DRAIN:
1113                 if (LK_CAN_WITNESS(flags))
1114                         WITNESS_CHECKORDER(&lk->lock_object, LOP_NEWORDER |
1115                             LOP_EXCLUSIVE, file, line, flags & LK_INTERLOCK ?
1116                             ilk : NULL);
1117
1118                 /*
1119                  * Trying to drain a lock we already own will result in a
1120                  * deadlock.
1121                  */
1122                 if (lockmgr_xlocked(lk)) {
1123                         if (flags & LK_INTERLOCK)
1124                                 class->lc_unlock(ilk);
1125                         panic("%s: draining %s with the lock held @ %s:%d\n",
1126                             __func__, iwmesg, file, line);
1127                 }
1128
1129                 for (;;) {
1130                         if (lk->lk_lock == LK_UNLOCKED &&
1131                             atomic_cmpset_acq_ptr(&lk->lk_lock, LK_UNLOCKED, tid))
1132                                 break;
1133
1134 #ifdef HWPMC_HOOKS
1135                         PMC_SOFT_CALL( , , lock, failed);
1136 #endif
1137                         lock_profile_obtain_lock_failed(&lk->lock_object,
1138                             &contested, &waittime);
1139
1140                         /*
1141                          * If the lock is expected to not sleep just give up
1142                          * and return.
1143                          */
1144                         if (LK_TRYOP(flags)) {
1145                                 LOCK_LOG2(lk, "%s: %p fails the try operation",
1146                                     __func__, lk);
1147                                 error = EBUSY;
1148                                 break;
1149                         }
1150
1151                         /*
1152                          * Acquire the sleepqueue chain lock because we
1153                          * probabilly will need to manipulate waiters flags.
1154                          */
1155                         sleepq_lock(&lk->lock_object);
1156                         x = lk->lk_lock;
1157
1158                         /*
1159                          * if the lock has been released while we spun on
1160                          * the sleepqueue chain lock just try again.
1161                          */
1162                         if (x == LK_UNLOCKED) {
1163                                 sleepq_release(&lk->lock_object);
1164                                 continue;
1165                         }
1166
1167                         v = x & (LK_ALL_WAITERS | LK_EXCLUSIVE_SPINNERS);
1168                         if ((x & ~v) == LK_UNLOCKED) {
1169                                 v = (x & ~LK_EXCLUSIVE_SPINNERS);
1170
1171                                 /*
1172                                  * If interruptible sleeps left the exclusive
1173                                  * queue empty avoid a starvation for the
1174                                  * threads sleeping on the shared queue by
1175                                  * giving them precedence and cleaning up the
1176                                  * exclusive waiters bit anyway.
1177                                  * Please note that lk_exslpfail count may be
1178                                  * lying about the real number of waiters with
1179                                  * the LK_SLEEPFAIL flag on because they may
1180                                  * be used in conjunction with interruptible
1181                                  * sleeps so lk_exslpfail might be considered
1182                                  * an 'upper limit' bound, including the edge
1183                                  * cases.
1184                                  */
1185                                 if (v & LK_EXCLUSIVE_WAITERS) {
1186                                         queue = SQ_EXCLUSIVE_QUEUE;
1187                                         v &= ~LK_EXCLUSIVE_WAITERS;
1188                                 } else {
1189
1190                                         /*
1191                                          * Exclusive waiters sleeping with
1192                                          * LK_SLEEPFAIL on and using
1193                                          * interruptible sleeps/timeout may
1194                                          * have left spourious lk_exslpfail
1195                                          * counts on, so clean it up anyway.
1196                                          */
1197                                         MPASS(v & LK_SHARED_WAITERS);
1198                                         lk->lk_exslpfail = 0;
1199                                         queue = SQ_SHARED_QUEUE;
1200                                         v &= ~LK_SHARED_WAITERS;
1201                                 }
1202                                 if (queue == SQ_EXCLUSIVE_QUEUE) {
1203                                         realexslp =
1204                                             sleepq_sleepcnt(&lk->lock_object,
1205                                             SQ_EXCLUSIVE_QUEUE);
1206                                         if (lk->lk_exslpfail >= realexslp) {
1207                                                 lk->lk_exslpfail = 0;
1208                                                 queue = SQ_SHARED_QUEUE;
1209                                                 v &= ~LK_SHARED_WAITERS;
1210                                                 if (realexslp != 0) {
1211                                                         LOCK_LOG2(lk,
1212                                         "%s: %p has only LK_SLEEPFAIL sleepers",
1213                                                             __func__, lk);
1214                                                         LOCK_LOG2(lk,
1215                         "%s: %p waking up threads on the exclusive queue",
1216                                                             __func__, lk);
1217                                                         wakeup_swapper =
1218                                                             sleepq_broadcast(
1219                                                             &lk->lock_object,
1220                                                             SLEEPQ_LK, 0,
1221                                                             SQ_EXCLUSIVE_QUEUE);
1222                                                 }
1223                                         } else
1224                                                 lk->lk_exslpfail = 0;
1225                                 }
1226                                 if (!atomic_cmpset_ptr(&lk->lk_lock, x, v)) {
1227                                         sleepq_release(&lk->lock_object);
1228                                         continue;
1229                                 }
1230                                 LOCK_LOG3(lk,
1231                                 "%s: %p waking up all threads on the %s queue",
1232                                     __func__, lk, queue == SQ_SHARED_QUEUE ?
1233                                     "shared" : "exclusive");
1234                                 wakeup_swapper |= sleepq_broadcast(
1235                                     &lk->lock_object, SLEEPQ_LK, 0, queue);
1236
1237                                 /*
1238                                  * If shared waiters have been woken up we need
1239                                  * to wait for one of them to acquire the lock
1240                                  * before to set the exclusive waiters in
1241                                  * order to avoid a deadlock.
1242                                  */
1243                                 if (queue == SQ_SHARED_QUEUE) {
1244                                         for (v = lk->lk_lock;
1245                                             (v & LK_SHARE) && !LK_SHARERS(v);
1246                                             v = lk->lk_lock)
1247                                                 cpu_spinwait();
1248                                 }
1249                         }
1250
1251                         /*
1252                          * Try to set the LK_EXCLUSIVE_WAITERS flag.  If we
1253                          * fail, loop back and retry.
1254                          */
1255                         if ((x & LK_EXCLUSIVE_WAITERS) == 0) {
1256                                 if (!atomic_cmpset_ptr(&lk->lk_lock, x,
1257                                     x | LK_EXCLUSIVE_WAITERS)) {
1258                                         sleepq_release(&lk->lock_object);
1259                                         continue;
1260                                 }
1261                                 LOCK_LOG2(lk, "%s: %p set drain waiters flag",
1262                                     __func__, lk);
1263                         }
1264
1265                         /*
1266                          * As far as we have been unable to acquire the
1267                          * exclusive lock and the exclusive waiters flag
1268                          * is set, we will sleep.
1269                          */
1270                         if (flags & LK_INTERLOCK) {
1271                                 class->lc_unlock(ilk);
1272                                 flags &= ~LK_INTERLOCK;
1273                         }
1274                         GIANT_SAVE();
1275                         sleepq_add(&lk->lock_object, NULL, iwmesg, SLEEPQ_LK,
1276                             SQ_EXCLUSIVE_QUEUE);
1277                         sleepq_wait(&lk->lock_object, ipri & PRIMASK);
1278                         GIANT_RESTORE();
1279                         LOCK_LOG2(lk, "%s: %p resuming from the sleep queue",
1280                             __func__, lk);
1281                 }
1282
1283                 if (error == 0) {
1284                         lock_profile_obtain_lock_success(&lk->lock_object,
1285                             contested, waittime, file, line);
1286                         LOCK_LOG_LOCK("DRAIN", &lk->lock_object, 0,
1287                             lk->lk_recurse, file, line);
1288                         WITNESS_LOCK(&lk->lock_object, LOP_EXCLUSIVE |
1289                             LK_TRYWIT(flags), file, line);
1290                         TD_LOCKS_INC(curthread);
1291                         STACK_SAVE(lk);
1292                 }
1293                 break;
1294         default:
1295                 if (flags & LK_INTERLOCK)
1296                         class->lc_unlock(ilk);
1297                 panic("%s: unknown lockmgr request 0x%x\n", __func__, op);
1298         }
1299
1300         if (flags & LK_INTERLOCK)
1301                 class->lc_unlock(ilk);
1302         if (wakeup_swapper)
1303                 kick_proc0();
1304
1305         return (error);
1306 }
1307
1308 void
1309 _lockmgr_disown(struct lock *lk, const char *file, int line)
1310 {
1311         uintptr_t tid, x;
1312
1313         if (SCHEDULER_STOPPED())
1314                 return;
1315
1316         tid = (uintptr_t)curthread;
1317         _lockmgr_assert(lk, KA_XLOCKED, file, line);
1318
1319         /*
1320          * Panic if the lock is recursed.
1321          */
1322         if (lockmgr_xlocked(lk) && lockmgr_recursed(lk))
1323                 panic("%s: disown a recursed lockmgr @ %s:%d\n",
1324                     __func__,  file, line);
1325
1326         /*
1327          * If the owner is already LK_KERNPROC just skip the whole operation.
1328          */
1329         if (LK_HOLDER(lk->lk_lock) != tid)
1330                 return;
1331         lock_profile_release_lock(&lk->lock_object);
1332         LOCK_LOG_LOCK("XDISOWN", &lk->lock_object, 0, 0, file, line);
1333         WITNESS_UNLOCK(&lk->lock_object, LOP_EXCLUSIVE, file, line);
1334         TD_LOCKS_DEC(curthread);
1335         STACK_SAVE(lk);
1336
1337         /*
1338          * In order to preserve waiters flags, just spin.
1339          */
1340         for (;;) {
1341                 x = lk->lk_lock;
1342                 MPASS((x & LK_EXCLUSIVE_SPINNERS) == 0);
1343                 x &= LK_ALL_WAITERS;
1344                 if (atomic_cmpset_rel_ptr(&lk->lk_lock, tid | x,
1345                     LK_KERNPROC | x))
1346                         return;
1347                 cpu_spinwait();
1348         }
1349 }
1350
1351 void
1352 lockmgr_printinfo(const struct lock *lk)
1353 {
1354         struct thread *td;
1355         uintptr_t x;
1356
1357         if (lk->lk_lock == LK_UNLOCKED)
1358                 printf("lock type %s: UNLOCKED\n", lk->lock_object.lo_name);
1359         else if (lk->lk_lock & LK_SHARE)
1360                 printf("lock type %s: SHARED (count %ju)\n",
1361                     lk->lock_object.lo_name,
1362                     (uintmax_t)LK_SHARERS(lk->lk_lock));
1363         else {
1364                 td = lockmgr_xholder(lk);
1365                 if (td == (struct thread *)LK_KERNPROC)
1366                         printf("lock type %s: EXCL by KERNPROC\n",
1367                             lk->lock_object.lo_name);
1368                 else
1369                         printf("lock type %s: EXCL by thread %p "
1370                             "(pid %d, %s, tid %d)\n", lk->lock_object.lo_name,
1371                             td, td->td_proc->p_pid, td->td_proc->p_comm,
1372                             td->td_tid);
1373         }
1374
1375         x = lk->lk_lock;
1376         if (x & LK_EXCLUSIVE_WAITERS)
1377                 printf(" with exclusive waiters pending\n");
1378         if (x & LK_SHARED_WAITERS)
1379                 printf(" with shared waiters pending\n");
1380         if (x & LK_EXCLUSIVE_SPINNERS)
1381                 printf(" with exclusive spinners pending\n");
1382
1383         STACK_PRINT(lk);
1384 }
1385
1386 int
1387 lockstatus(const struct lock *lk)
1388 {
1389         uintptr_t v, x;
1390         int ret;
1391
1392         ret = LK_SHARED;
1393         x = lk->lk_lock;
1394         v = LK_HOLDER(x);
1395
1396         if ((x & LK_SHARE) == 0) {
1397                 if (v == (uintptr_t)curthread || v == LK_KERNPROC)
1398                         ret = LK_EXCLUSIVE;
1399                 else
1400                         ret = LK_EXCLOTHER;
1401         } else if (x == LK_UNLOCKED)
1402                 ret = 0;
1403
1404         return (ret);
1405 }
1406
1407 #ifdef INVARIANT_SUPPORT
1408
1409 FEATURE(invariant_support,
1410     "Support for modules compiled with INVARIANTS option");
1411
1412 #ifndef INVARIANTS
1413 #undef  _lockmgr_assert
1414 #endif
1415
1416 void
1417 _lockmgr_assert(const struct lock *lk, int what, const char *file, int line)
1418 {
1419         int slocked = 0;
1420
1421         if (panicstr != NULL)
1422                 return;
1423         switch (what) {
1424         case KA_SLOCKED:
1425         case KA_SLOCKED | KA_NOTRECURSED:
1426         case KA_SLOCKED | KA_RECURSED:
1427                 slocked = 1;
1428         case KA_LOCKED:
1429         case KA_LOCKED | KA_NOTRECURSED:
1430         case KA_LOCKED | KA_RECURSED:
1431 #ifdef WITNESS
1432
1433                 /*
1434                  * We cannot trust WITNESS if the lock is held in exclusive
1435                  * mode and a call to lockmgr_disown() happened.
1436                  * Workaround this skipping the check if the lock is held in
1437                  * exclusive mode even for the KA_LOCKED case.
1438                  */
1439                 if (slocked || (lk->lk_lock & LK_SHARE)) {
1440                         witness_assert(&lk->lock_object, what, file, line);
1441                         break;
1442                 }
1443 #endif
1444                 if (lk->lk_lock == LK_UNLOCKED ||
1445                     ((lk->lk_lock & LK_SHARE) == 0 && (slocked ||
1446                     (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk)))))
1447                         panic("Lock %s not %slocked @ %s:%d\n",
1448                             lk->lock_object.lo_name, slocked ? "share" : "",
1449                             file, line);
1450
1451                 if ((lk->lk_lock & LK_SHARE) == 0) {
1452                         if (lockmgr_recursed(lk)) {
1453                                 if (what & KA_NOTRECURSED)
1454                                         panic("Lock %s recursed @ %s:%d\n",
1455                                             lk->lock_object.lo_name, file,
1456                                             line);
1457                         } else if (what & KA_RECURSED)
1458                                 panic("Lock %s not recursed @ %s:%d\n",
1459                                     lk->lock_object.lo_name, file, line);
1460                 }
1461                 break;
1462         case KA_XLOCKED:
1463         case KA_XLOCKED | KA_NOTRECURSED:
1464         case KA_XLOCKED | KA_RECURSED:
1465                 if (!lockmgr_xlocked(lk) && !lockmgr_disowned(lk))
1466                         panic("Lock %s not exclusively locked @ %s:%d\n",
1467                             lk->lock_object.lo_name, file, line);
1468                 if (lockmgr_recursed(lk)) {
1469                         if (what & KA_NOTRECURSED)
1470                                 panic("Lock %s recursed @ %s:%d\n",
1471                                     lk->lock_object.lo_name, file, line);
1472                 } else if (what & KA_RECURSED)
1473                         panic("Lock %s not recursed @ %s:%d\n",
1474                             lk->lock_object.lo_name, file, line);
1475                 break;
1476         case KA_UNLOCKED:
1477                 if (lockmgr_xlocked(lk) || lockmgr_disowned(lk))
1478                         panic("Lock %s exclusively locked @ %s:%d\n",
1479                             lk->lock_object.lo_name, file, line);
1480                 break;
1481         default:
1482                 panic("Unknown lockmgr assertion: %d @ %s:%d\n", what, file,
1483                     line);
1484         }
1485 }
1486 #endif
1487
1488 #ifdef DDB
1489 int
1490 lockmgr_chain(struct thread *td, struct thread **ownerp)
1491 {
1492         struct lock *lk;
1493
1494         lk = td->td_wchan;
1495
1496         if (LOCK_CLASS(&lk->lock_object) != &lock_class_lockmgr)
1497                 return (0);
1498         db_printf("blocked on lockmgr %s", lk->lock_object.lo_name);
1499         if (lk->lk_lock & LK_SHARE)
1500                 db_printf("SHARED (count %ju)\n",
1501                     (uintmax_t)LK_SHARERS(lk->lk_lock));
1502         else
1503                 db_printf("EXCL\n");
1504         *ownerp = lockmgr_xholder(lk);
1505
1506         return (1);
1507 }
1508
1509 static void
1510 db_show_lockmgr(const struct lock_object *lock)
1511 {
1512         struct thread *td;
1513         const struct lock *lk;
1514
1515         lk = (const struct lock *)lock;
1516
1517         db_printf(" state: ");
1518         if (lk->lk_lock == LK_UNLOCKED)
1519                 db_printf("UNLOCKED\n");
1520         else if (lk->lk_lock & LK_SHARE)
1521                 db_printf("SLOCK: %ju\n", (uintmax_t)LK_SHARERS(lk->lk_lock));
1522         else {
1523                 td = lockmgr_xholder(lk);
1524                 if (td == (struct thread *)LK_KERNPROC)
1525                         db_printf("XLOCK: LK_KERNPROC\n");
1526                 else
1527                         db_printf("XLOCK: %p (tid %d, pid %d, \"%s\")\n", td,
1528                             td->td_tid, td->td_proc->p_pid,
1529                             td->td_proc->p_comm);
1530                 if (lockmgr_recursed(lk))
1531                         db_printf(" recursed: %d\n", lk->lk_recurse);
1532         }
1533         db_printf(" waiters: ");
1534         switch (lk->lk_lock & LK_ALL_WAITERS) {
1535         case LK_SHARED_WAITERS:
1536                 db_printf("shared\n");
1537                 break;
1538         case LK_EXCLUSIVE_WAITERS:
1539                 db_printf("exclusive\n");
1540                 break;
1541         case LK_ALL_WAITERS:
1542                 db_printf("shared and exclusive\n");
1543                 break;
1544         default:
1545                 db_printf("none\n");
1546         }
1547         db_printf(" spinners: ");
1548         if (lk->lk_lock & LK_EXCLUSIVE_SPINNERS)
1549                 db_printf("exclusive\n");
1550         else
1551                 db_printf("none\n");
1552 }
1553 #endif