subr_turnstile.c revision 88088
1/*- 2 * Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 3. Berkeley Software Design Inc's name may not be used to endorse or 13 * promote products derived from this software without specific prior 14 * written permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER 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 26 * SUCH DAMAGE. 27 * 28 * from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $ 29 * and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $ 30 * $FreeBSD: head/sys/kern/subr_turnstile.c 88088 2001-12-18 00:27:18Z jhb $ 31 */ 32 33/* 34 * Machine independent bits of mutex implementation. 35 */ 36 37#include "opt_ddb.h" 38 39#include <sys/param.h> 40#include <sys/bus.h> 41#include <sys/kernel.h> 42#include <sys/lock.h> 43#include <sys/malloc.h> 44#include <sys/mutex.h> 45#include <sys/proc.h> 46#include <sys/resourcevar.h> 47#include <sys/sysctl.h> 48#include <sys/systm.h> 49#include <sys/vmmeter.h> 50#include <sys/ktr.h> 51 52#include <machine/atomic.h> 53#include <machine/bus.h> 54#include <machine/clock.h> 55#include <machine/cpu.h> 56 57#include <ddb/ddb.h> 58 59#include <vm/vm.h> 60#include <vm/vm_extern.h> 61 62/* 63 * Internal utility macros. 64 */ 65#define mtx_unowned(m) ((m)->mtx_lock == MTX_UNOWNED) 66 67#define mtx_owner(m) (mtx_unowned((m)) ? NULL \ 68 : (struct thread *)((m)->mtx_lock & MTX_FLAGMASK)) 69 70#define SET_PRIO(td, pri) (td)->td_ksegrp->kg_pri.pri_level = (pri) 71 72/* 73 * Lock classes for sleep and spin mutexes. 74 */ 75struct lock_class lock_class_mtx_sleep = { 76 "sleep mutex", 77 LC_SLEEPLOCK | LC_RECURSABLE 78}; 79struct lock_class lock_class_mtx_spin = { 80 "spin mutex", 81 LC_SPINLOCK | LC_RECURSABLE 82}; 83 84/* 85 * Prototypes for non-exported routines. 86 */ 87static void propagate_priority(struct thread *); 88 89static void 90propagate_priority(struct thread *td) 91{ 92 struct ksegrp *kg = td->td_ksegrp; 93 int pri = kg->kg_pri.pri_level; 94 struct mtx *m = td->td_blocked; 95 96 mtx_assert(&sched_lock, MA_OWNED); 97 for (;;) { 98 struct thread *td1; 99 100 td = mtx_owner(m); 101 102 if (td == NULL) { 103 /* 104 * This really isn't quite right. Really 105 * ought to bump priority of thread that 106 * next acquires the mutex. 107 */ 108 MPASS(m->mtx_lock == MTX_CONTESTED); 109 return; 110 } 111 kg = td->td_ksegrp; 112 113 MPASS(td->td_proc->p_magic == P_MAGIC); 114 KASSERT(td->td_proc->p_stat != SSLEEP, ("sleeping thread owns a mutex")); 115 if (kg->kg_pri.pri_level <= pri) /* lower is higher priority */ 116 return; 117 118 /* 119 * Bump this thread's priority. 120 */ 121 SET_PRIO(td, pri); 122 123 /* 124 * If lock holder is actually running, just bump priority. 125 */ 126 /* XXXKSE this test is not sufficient */ 127 if (td->td_kse && (td->td_kse->ke_oncpu != NOCPU)) { 128 MPASS(td->td_proc->p_stat == SRUN 129 || td->td_proc->p_stat == SZOMB 130 || td->td_proc->p_stat == SSTOP); 131 return; 132 } 133 134#ifndef SMP 135 /* 136 * For UP, we check to see if td is curthread (this shouldn't 137 * ever happen however as it would mean we are in a deadlock.) 138 */ 139 KASSERT(td != curthread, ("Deadlock detected")); 140#endif 141 142 /* 143 * If on run queue move to new run queue, and quit. 144 * XXXKSE this gets a lot more complicated under threads 145 * but try anyhow. 146 */ 147 if (td->td_proc->p_stat == SRUN) { 148 MPASS(td->td_blocked == NULL); 149 remrunqueue(td); 150 setrunqueue(td); 151 return; 152 } 153 154 /* 155 * If we aren't blocked on a mutex, we should be. 156 */ 157 KASSERT(td->td_proc->p_stat == SMTX, ( 158 "process %d(%s):%d holds %s but isn't blocked on a mutex\n", 159 td->td_proc->p_pid, td->td_proc->p_comm, td->td_proc->p_stat, 160 m->mtx_object.lo_name)); 161 162 /* 163 * Pick up the mutex that td is blocked on. 164 */ 165 m = td->td_blocked; 166 MPASS(m != NULL); 167 168 /* 169 * Check if the thread needs to be moved up on 170 * the blocked chain 171 */ 172 if (td == TAILQ_FIRST(&m->mtx_blocked)) { 173 continue; 174 } 175 176 td1 = TAILQ_PREV(td, threadqueue, td_blkq); 177 if (td1->td_ksegrp->kg_pri.pri_level <= pri) { 178 continue; 179 } 180 181 /* 182 * Remove thread from blocked chain and determine where 183 * it should be moved up to. Since we know that td1 has 184 * a lower priority than td, we know that at least one 185 * thread in the chain has a lower priority and that 186 * td1 will thus not be NULL after the loop. 187 */ 188 TAILQ_REMOVE(&m->mtx_blocked, td, td_blkq); 189 TAILQ_FOREACH(td1, &m->mtx_blocked, td_blkq) { 190 MPASS(td1->td_proc->p_magic == P_MAGIC); 191 if (td1->td_ksegrp->kg_pri.pri_level > pri) 192 break; 193 } 194 195 MPASS(td1 != NULL); 196 TAILQ_INSERT_BEFORE(td1, td, td_blkq); 197 CTR4(KTR_LOCK, 198 "propagate_priority: p %p moved before %p on [%p] %s", 199 td, td1, m, m->mtx_object.lo_name); 200 } 201} 202 203/* 204 * Function versions of the inlined __mtx_* macros. These are used by 205 * modules and can also be called from assembly language if needed. 206 */ 207void 208_mtx_lock_flags(struct mtx *m, int opts, const char *file, int line) 209{ 210 211 MPASS(curthread != NULL); 212 KASSERT((opts & MTX_NOSWITCH) == 0, 213 ("MTX_NOSWITCH used at %s:%d", file, line)); 214 _get_sleep_lock(m, curthread, opts, file, line); 215 LOCK_LOG_LOCK("LOCK", &m->mtx_object, opts, m->mtx_recurse, file, 216 line); 217 WITNESS_LOCK(&m->mtx_object, opts | LOP_EXCLUSIVE, file, line); 218} 219 220void 221_mtx_unlock_flags(struct mtx *m, int opts, const char *file, int line) 222{ 223 224 MPASS(curthread != NULL); 225 mtx_assert(m, MA_OWNED); 226 WITNESS_UNLOCK(&m->mtx_object, opts | LOP_EXCLUSIVE, file, line); 227 LOCK_LOG_LOCK("UNLOCK", &m->mtx_object, opts, m->mtx_recurse, file, 228 line); 229 _rel_sleep_lock(m, curthread, opts, file, line); 230} 231 232void 233_mtx_lock_spin_flags(struct mtx *m, int opts, const char *file, int line) 234{ 235 236 MPASS(curthread != NULL); 237 _get_spin_lock(m, curthread, opts, file, line); 238 LOCK_LOG_LOCK("LOCK", &m->mtx_object, opts, m->mtx_recurse, file, 239 line); 240 WITNESS_LOCK(&m->mtx_object, opts | LOP_EXCLUSIVE, file, line); 241} 242 243void 244_mtx_unlock_spin_flags(struct mtx *m, int opts, const char *file, int line) 245{ 246 247 MPASS(curthread != NULL); 248 mtx_assert(m, MA_OWNED); 249 WITNESS_UNLOCK(&m->mtx_object, opts | LOP_EXCLUSIVE, file, line); 250 LOCK_LOG_LOCK("UNLOCK", &m->mtx_object, opts, m->mtx_recurse, file, 251 line); 252 _rel_spin_lock(m); 253} 254 255/* 256 * The important part of mtx_trylock{,_flags}() 257 * Tries to acquire lock `m.' We do NOT handle recursion here; we assume that 258 * if we're called, it's because we know we don't already own this lock. 259 */ 260int 261_mtx_trylock(struct mtx *m, int opts, const char *file, int line) 262{ 263 int rval; 264 265 MPASS(curthread != NULL); 266 267 /* 268 * _mtx_trylock does not accept MTX_NOSWITCH option. 269 */ 270 KASSERT((opts & MTX_NOSWITCH) == 0, 271 ("mtx_trylock() called with invalid option flag(s) %d", opts)); 272 273 rval = _obtain_lock(m, curthread); 274 275 LOCK_LOG_TRY("LOCK", &m->mtx_object, opts, rval, file, line); 276 if (rval) { 277 /* 278 * We do not handle recursion in _mtx_trylock; see the 279 * note at the top of the routine. 280 */ 281 KASSERT(!mtx_recursed(m), 282 ("mtx_trylock() called on a recursed mutex")); 283 WITNESS_LOCK(&m->mtx_object, opts | LOP_EXCLUSIVE | LOP_TRYLOCK, 284 file, line); 285 } 286 287 return (rval); 288} 289 290/* 291 * _mtx_lock_sleep: the tougher part of acquiring an MTX_DEF lock. 292 * 293 * We call this if the lock is either contested (i.e. we need to go to 294 * sleep waiting for it), or if we need to recurse on it. 295 */ 296void 297_mtx_lock_sleep(struct mtx *m, int opts, const char *file, int line) 298{ 299 struct thread *td = curthread; 300 struct ksegrp *kg = td->td_ksegrp; 301 302 if ((m->mtx_lock & MTX_FLAGMASK) == (uintptr_t)td) { 303 m->mtx_recurse++; 304 atomic_set_ptr(&m->mtx_lock, MTX_RECURSED); 305 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 306 CTR1(KTR_LOCK, "_mtx_lock_sleep: %p recursing", m); 307 return; 308 } 309 310 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 311 CTR4(KTR_LOCK, 312 "_mtx_lock_sleep: %s contested (lock=%p) at %s:%d", 313 m->mtx_object.lo_name, (void *)m->mtx_lock, file, line); 314 315 while (!_obtain_lock(m, td)) { 316 uintptr_t v; 317 struct thread *td1; 318 319 mtx_lock_spin(&sched_lock); 320 /* 321 * Check if the lock has been released while spinning for 322 * the sched_lock. 323 */ 324 if ((v = m->mtx_lock) == MTX_UNOWNED) { 325 mtx_unlock_spin(&sched_lock); 326 continue; 327 } 328 329 /* 330 * The mutex was marked contested on release. This means that 331 * there are threads blocked on it. 332 */ 333 if (v == MTX_CONTESTED) { 334 td1 = TAILQ_FIRST(&m->mtx_blocked); 335 MPASS(td1 != NULL); 336 m->mtx_lock = (uintptr_t)td | MTX_CONTESTED; 337 338 if (td1->td_ksegrp->kg_pri.pri_level < kg->kg_pri.pri_level) 339 SET_PRIO(td, td1->td_ksegrp->kg_pri.pri_level); 340 mtx_unlock_spin(&sched_lock); 341 return; 342 } 343 344 /* 345 * If the mutex isn't already contested and a failure occurs 346 * setting the contested bit, the mutex was either released 347 * or the state of the MTX_RECURSED bit changed. 348 */ 349 if ((v & MTX_CONTESTED) == 0 && 350 !atomic_cmpset_ptr(&m->mtx_lock, (void *)v, 351 (void *)(v | MTX_CONTESTED))) { 352 mtx_unlock_spin(&sched_lock); 353 continue; 354 } 355 356 /* 357 * We deffinately must sleep for this lock. 358 */ 359 mtx_assert(m, MA_NOTOWNED); 360 361#ifdef notyet 362 /* 363 * If we're borrowing an interrupted thread's VM context, we 364 * must clean up before going to sleep. 365 */ 366 if (td->td_ithd != NULL) { 367 struct ithd *it = td->td_ithd; 368 369 if (it->it_interrupted) { 370 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 371 CTR2(KTR_LOCK, 372 "_mtx_lock_sleep: %p interrupted %p", 373 it, it->it_interrupted); 374 intr_thd_fixup(it); 375 } 376 } 377#endif 378 379 /* 380 * Put us on the list of threads blocked on this mutex. 381 */ 382 if (TAILQ_EMPTY(&m->mtx_blocked)) { 383 td1 = (struct thread *)(m->mtx_lock & MTX_FLAGMASK); 384 LIST_INSERT_HEAD(&td1->td_contested, m, mtx_contested); 385 TAILQ_INSERT_TAIL(&m->mtx_blocked, td, td_blkq); 386 } else { 387 TAILQ_FOREACH(td1, &m->mtx_blocked, td_blkq) 388 if (td1->td_ksegrp->kg_pri.pri_level > kg->kg_pri.pri_level) 389 break; 390 if (td1) 391 TAILQ_INSERT_BEFORE(td1, td, td_blkq); 392 else 393 TAILQ_INSERT_TAIL(&m->mtx_blocked, td, td_blkq); 394 } 395 396 /* 397 * Save who we're blocked on. 398 */ 399 td->td_blocked = m; 400 td->td_mtxname = m->mtx_object.lo_name; 401 td->td_proc->p_stat = SMTX; 402 propagate_priority(td); 403 404 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 405 CTR3(KTR_LOCK, 406 "_mtx_lock_sleep: p %p blocked on [%p] %s", td, m, 407 m->mtx_object.lo_name); 408 409 td->td_proc->p_stats->p_ru.ru_nvcsw++; 410 mi_switch(); 411 412 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 413 CTR3(KTR_LOCK, 414 "_mtx_lock_sleep: p %p free from blocked on [%p] %s", 415 td, m, m->mtx_object.lo_name); 416 417 mtx_unlock_spin(&sched_lock); 418 } 419 420 return; 421} 422 423/* 424 * _mtx_lock_spin: the tougher part of acquiring an MTX_SPIN lock. 425 * 426 * This is only called if we need to actually spin for the lock. Recursion 427 * is handled inline. 428 */ 429void 430_mtx_lock_spin(struct mtx *m, int opts, const char *file, int line) 431{ 432 int i = 0; 433 434 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 435 CTR1(KTR_LOCK, "_mtx_lock_spin: %p spinning", m); 436 437 for (;;) { 438 if (_obtain_lock(m, curthread)) 439 break; 440 441 /* Give interrupts a chance while we spin. */ 442 critical_exit(); 443 while (m->mtx_lock != MTX_UNOWNED) { 444 if (i++ < 1000000) 445 continue; 446 if (i++ < 6000000) 447 DELAY(1); 448#ifdef DDB 449 else if (!db_active) 450#else 451 else 452#endif 453 panic("spin lock %s held by %p for > 5 seconds", 454 m->mtx_object.lo_name, (void *)m->mtx_lock); 455 } 456 critical_enter(); 457 } 458 459 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 460 CTR1(KTR_LOCK, "_mtx_lock_spin: %p spin done", m); 461 462 return; 463} 464 465/* 466 * _mtx_unlock_sleep: the tougher part of releasing an MTX_DEF lock. 467 * 468 * We are only called here if the lock is recursed or contested (i.e. we 469 * need to wake up a blocked thread). 470 */ 471void 472_mtx_unlock_sleep(struct mtx *m, int opts, const char *file, int line) 473{ 474 struct thread *td, *td1; 475 struct mtx *m1; 476 int pri; 477 struct ksegrp *kg; 478 479 td = curthread; 480 kg = td->td_ksegrp; 481 482 if (mtx_recursed(m)) { 483 if (--(m->mtx_recurse) == 0) 484 atomic_clear_ptr(&m->mtx_lock, MTX_RECURSED); 485 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 486 CTR1(KTR_LOCK, "_mtx_unlock_sleep: %p unrecurse", m); 487 return; 488 } 489 490 mtx_lock_spin(&sched_lock); 491 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 492 CTR1(KTR_LOCK, "_mtx_unlock_sleep: %p contested", m); 493 494 td1 = TAILQ_FIRST(&m->mtx_blocked); 495 MPASS(td->td_proc->p_magic == P_MAGIC); 496 MPASS(td1->td_proc->p_magic == P_MAGIC); 497 498 TAILQ_REMOVE(&m->mtx_blocked, td1, td_blkq); 499 500 if (TAILQ_EMPTY(&m->mtx_blocked)) { 501 LIST_REMOVE(m, mtx_contested); 502 _release_lock_quick(m); 503 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 504 CTR1(KTR_LOCK, "_mtx_unlock_sleep: %p not held", m); 505 } else 506 atomic_store_rel_ptr(&m->mtx_lock, (void *)MTX_CONTESTED); 507 508 pri = PRI_MAX; 509 LIST_FOREACH(m1, &td->td_contested, mtx_contested) { 510 int cp = TAILQ_FIRST(&m1->mtx_blocked)->td_ksegrp->kg_pri.pri_level; 511 if (cp < pri) 512 pri = cp; 513 } 514 515 if (pri > kg->kg_pri.pri_native) 516 pri = kg->kg_pri.pri_native; 517 SET_PRIO(td, pri); 518 519 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 520 CTR2(KTR_LOCK, "_mtx_unlock_sleep: %p contested setrunqueue %p", 521 m, td1); 522 523 td1->td_blocked = NULL; 524 td1->td_proc->p_stat = SRUN; 525 setrunqueue(td1); 526 527 if ((opts & MTX_NOSWITCH) == 0 && td1->td_ksegrp->kg_pri.pri_level < pri) { 528#ifdef notyet 529 if (td->td_ithd != NULL) { 530 struct ithd *it = td->td_ithd; 531 532 if (it->it_interrupted) { 533 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 534 CTR2(KTR_LOCK, 535 "_mtx_unlock_sleep: %p interrupted %p", 536 it, it->it_interrupted); 537 intr_thd_fixup(it); 538 } 539 } 540#endif 541 setrunqueue(td); 542 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 543 CTR2(KTR_LOCK, 544 "_mtx_unlock_sleep: %p switching out lock=%p", m, 545 (void *)m->mtx_lock); 546 547 td->td_proc->p_stats->p_ru.ru_nivcsw++; 548 mi_switch(); 549 if (LOCK_LOG_TEST(&m->mtx_object, opts)) 550 CTR2(KTR_LOCK, "_mtx_unlock_sleep: %p resuming lock=%p", 551 m, (void *)m->mtx_lock); 552 } 553 554 mtx_unlock_spin(&sched_lock); 555 556 return; 557} 558 559/* 560 * All the unlocking of MTX_SPIN locks is done inline. 561 * See the _rel_spin_lock() macro for the details. 562 */ 563 564/* 565 * The backing function for the INVARIANTS-enabled mtx_assert() 566 */ 567#ifdef INVARIANT_SUPPORT 568void 569_mtx_assert(struct mtx *m, int what, const char *file, int line) 570{ 571 572 if (panicstr != NULL) 573 return; 574 switch (what) { 575 case MA_OWNED: 576 case MA_OWNED | MA_RECURSED: 577 case MA_OWNED | MA_NOTRECURSED: 578 if (!mtx_owned(m)) 579 panic("mutex %s not owned at %s:%d", 580 m->mtx_object.lo_name, file, line); 581 if (mtx_recursed(m)) { 582 if ((what & MA_NOTRECURSED) != 0) 583 panic("mutex %s recursed at %s:%d", 584 m->mtx_object.lo_name, file, line); 585 } else if ((what & MA_RECURSED) != 0) { 586 panic("mutex %s unrecursed at %s:%d", 587 m->mtx_object.lo_name, file, line); 588 } 589 break; 590 case MA_NOTOWNED: 591 if (mtx_owned(m)) 592 panic("mutex %s owned at %s:%d", 593 m->mtx_object.lo_name, file, line); 594 break; 595 default: 596 panic("unknown mtx_assert at %s:%d", file, line); 597 } 598} 599#endif 600 601/* 602 * The MUTEX_DEBUG-enabled mtx_validate() 603 * 604 * Most of these checks have been moved off into the LO_INITIALIZED flag 605 * maintained by the witness code. 606 */ 607#ifdef MUTEX_DEBUG 608 609void mtx_validate __P((struct mtx *)); 610 611void 612mtx_validate(struct mtx *m) 613{ 614 615/* 616 * XXX - When kernacc() is fixed on the alpha to handle K0_SEG memory properly 617 * we can re-enable the kernacc() checks. 618 */ 619#ifndef __alpha__ 620 /* 621 * Can't call kernacc() from early init386(), especially when 622 * initializing Giant mutex, because some stuff in kernacc() 623 * requires Giant itself. 624 */ 625 if (!cold) 626 if (!kernacc((caddr_t)m, sizeof(m), 627 VM_PROT_READ | VM_PROT_WRITE)) 628 panic("Can't read and write to mutex %p", m); 629#endif 630} 631#endif 632 633/* 634 * Mutex initialization routine; initialize lock `m' of type contained in 635 * `opts' with options contained in `opts' and description `description.' 636 */ 637void 638mtx_init(struct mtx *m, const char *description, int opts) 639{ 640 struct lock_object *lock; 641 642 MPASS((opts & ~(MTX_SPIN | MTX_QUIET | MTX_RECURSE | 643 MTX_SLEEPABLE | MTX_NOWITNESS)) == 0); 644 645#ifdef MUTEX_DEBUG 646 /* Diagnostic and error correction */ 647 mtx_validate(m); 648#endif 649 650 lock = &m->mtx_object; 651 KASSERT((lock->lo_flags & LO_INITIALIZED) == 0, 652 ("mutex %s %p already initialized", description, m)); 653 bzero(m, sizeof(*m)); 654 if (opts & MTX_SPIN) 655 lock->lo_class = &lock_class_mtx_spin; 656 else 657 lock->lo_class = &lock_class_mtx_sleep; 658 lock->lo_name = description; 659 if (opts & MTX_QUIET) 660 lock->lo_flags = LO_QUIET; 661 if (opts & MTX_RECURSE) 662 lock->lo_flags |= LO_RECURSABLE; 663 if (opts & MTX_SLEEPABLE) 664 lock->lo_flags |= LO_SLEEPABLE; 665 if ((opts & MTX_NOWITNESS) == 0) 666 lock->lo_flags |= LO_WITNESS; 667 668 m->mtx_lock = MTX_UNOWNED; 669 TAILQ_INIT(&m->mtx_blocked); 670 671 LOCK_LOG_INIT(lock, opts); 672 673 WITNESS_INIT(lock); 674} 675 676/* 677 * Remove lock `m' from all_mtx queue. We don't allow MTX_QUIET to be 678 * passed in as a flag here because if the corresponding mtx_init() was 679 * called with MTX_QUIET set, then it will already be set in the mutex's 680 * flags. 681 */ 682void 683mtx_destroy(struct mtx *m) 684{ 685 686 LOCK_LOG_DESTROY(&m->mtx_object, 0); 687 688 if (!mtx_owned(m)) 689 MPASS(mtx_unowned(m)); 690 else { 691 MPASS((m->mtx_lock & (MTX_RECURSED|MTX_CONTESTED)) == 0); 692 693 /* Tell witness this isn't locked to make it happy. */ 694 WITNESS_UNLOCK(&m->mtx_object, LOP_EXCLUSIVE | LOP_NOSWITCH, 695 __FILE__, __LINE__); 696 } 697 698 WITNESS_DESTROY(&m->mtx_object); 699} 700 701/* 702 * Encapsulated Giant mutex routines. These routines provide encapsulation 703 * control for the Giant mutex, allowing sysctls to be used to turn on and 704 * off Giant around certain subsystems. The default value for the sysctls 705 * are set to what developers believe is stable and working in regards to 706 * the Giant pushdown. Developers should not turn off Giant via these 707 * sysctls unless they know what they are doing. 708 * 709 * Callers of mtx_lock_giant() are expected to pass the return value to an 710 * accompanying mtx_unlock_giant() later on. If multiple subsystems are 711 * effected by a Giant wrap, all related sysctl variables must be zero for 712 * the subsystem call to operate without Giant (as determined by the caller). 713 */ 714 715SYSCTL_NODE(_kern, OID_AUTO, giant, CTLFLAG_RD, NULL, "Giant mutex manipulation"); 716 717static int kern_giant_all = 0; 718SYSCTL_INT(_kern_giant, OID_AUTO, all, CTLFLAG_RW, &kern_giant_all, 0, ""); 719 720int kern_giant_proc = 1; /* Giant around PROC locks */ 721int kern_giant_file = 1; /* Giant around struct file & filedesc */ 722SYSCTL_INT(_kern_giant, OID_AUTO, proc, CTLFLAG_RW, &kern_giant_proc, 0, ""); 723SYSCTL_INT(_kern_giant, OID_AUTO, file, CTLFLAG_RW, &kern_giant_file, 0, ""); 724 725int 726mtx_lock_giant(int sysctlvar) 727{ 728 if (sysctlvar || kern_giant_all) { 729 mtx_lock(&Giant); 730 return(1); 731 } 732 return(0); 733} 734 735void 736mtx_unlock_giant(int s) 737{ 738 if (s) 739 mtx_unlock(&Giant); 740} 741 742