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 */
31
32/*
33 * Machine independent bits of mutex implementation.
34 */
35
36#include <sys/cdefs.h>
37__FBSDID("$FreeBSD$");
38
39#include "opt_adaptive_mutexes.h"
40#include "opt_ddb.h"
41#include "opt_global.h"
42#include "opt_hwpmc_hooks.h"
43#include "opt_kdtrace.h"
44#include "opt_sched.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/bus.h>
49#include <sys/conf.h>
50#include <sys/kdb.h>
51#include <sys/kernel.h>
52#include <sys/ktr.h>
53#include <sys/lock.h>
54#include <sys/malloc.h>
55#include <sys/mutex.h>
56#include <sys/proc.h>
57#include <sys/resourcevar.h>
58#include <sys/sched.h>
59#include <sys/sbuf.h>
60#include <sys/sysctl.h>
61#include <sys/turnstile.h>
62#include <sys/vmmeter.h>
63#include <sys/lock_profile.h>
64
65#include <machine/atomic.h>
66#include <machine/bus.h>
67#include <machine/cpu.h>
68
69#include <ddb/ddb.h>
70
71#include <fs/devfs/devfs_int.h>
72
73#include <vm/vm.h>
74#include <vm/vm_extern.h>
75
76#if defined(SMP) && !defined(NO_ADAPTIVE_MUTEXES)
77#define	ADAPTIVE_MUTEXES
78#endif
79
80#ifdef HWPMC_HOOKS
81#include <sys/pmckern.h>
82PMC_SOFT_DEFINE( , , lock, failed);
83#endif
84
85/*
86 * Internal utility macros.
87 */
88#define mtx_unowned(m)	((m)->mtx_lock == MTX_UNOWNED)
89
90#define	mtx_destroyed(m) ((m)->mtx_lock == MTX_DESTROYED)
91
92#define	mtx_owner(m)	((struct thread *)((m)->mtx_lock & ~MTX_FLAGMASK))
93
94static void	assert_mtx(struct lock_object *lock, int what);
95#ifdef DDB
96static void	db_show_mtx(struct lock_object *lock);
97#endif
98static void	lock_mtx(struct lock_object *lock, int how);
99static void	lock_spin(struct lock_object *lock, int how);
100#ifdef KDTRACE_HOOKS
101static int	owner_mtx(struct lock_object *lock, struct thread **owner);
102#endif
103static int	unlock_mtx(struct lock_object *lock);
104static int	unlock_spin(struct lock_object *lock);
105
106/*
107 * Lock classes for sleep and spin mutexes.
108 */
109struct lock_class lock_class_mtx_sleep = {
110	.lc_name = "sleep mutex",
111	.lc_flags = LC_SLEEPLOCK | LC_RECURSABLE,
112	.lc_assert = assert_mtx,
113#ifdef DDB
114	.lc_ddb_show = db_show_mtx,
115#endif
116	.lc_lock = lock_mtx,
117	.lc_unlock = unlock_mtx,
118#ifdef KDTRACE_HOOKS
119	.lc_owner = owner_mtx,
120#endif
121};
122struct lock_class lock_class_mtx_spin = {
123	.lc_name = "spin mutex",
124	.lc_flags = LC_SPINLOCK | LC_RECURSABLE,
125	.lc_assert = assert_mtx,
126#ifdef DDB
127	.lc_ddb_show = db_show_mtx,
128#endif
129	.lc_lock = lock_spin,
130	.lc_unlock = unlock_spin,
131#ifdef KDTRACE_HOOKS
132	.lc_owner = owner_mtx,
133#endif
134};
135
136/*
137 * System-wide mutexes
138 */
139struct mtx blocked_lock;
140struct mtx Giant;
141
142void
143assert_mtx(struct lock_object *lock, int what)
144{
145
146	mtx_assert((struct mtx *)lock, what);
147}
148
149void
150lock_mtx(struct lock_object *lock, int how)
151{
152
153	mtx_lock((struct mtx *)lock);
154}
155
156void
157lock_spin(struct lock_object *lock, int how)
158{
159
160	panic("spin locks can only use msleep_spin");
161}
162
163int
164unlock_mtx(struct lock_object *lock)
165{
166	struct mtx *m;
167
168	m = (struct mtx *)lock;
169	mtx_assert(m, MA_OWNED | MA_NOTRECURSED);
170	mtx_unlock(m);
171	return (0);
172}
173
174int
175unlock_spin(struct lock_object *lock)
176{
177
178	panic("spin locks can only use msleep_spin");
179}
180
181#ifdef KDTRACE_HOOKS
182int
183owner_mtx(struct lock_object *lock, struct thread **owner)
184{
185	struct mtx *m = (struct mtx *)lock;
186
187	*owner = mtx_owner(m);
188	return (mtx_unowned(m) == 0);
189}
190#endif
191
192/*
193 * Function versions of the inlined __mtx_* macros.  These are used by
194 * modules and can also be called from assembly language if needed.
195 */
196void
197_mtx_lock_flags(struct mtx *m, int opts, const char *file, int line)
198{
199
200	if (SCHEDULER_STOPPED())
201		return;
202	KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
203	    ("mtx_lock() by idle thread %p on sleep mutex %s @ %s:%d",
204	    curthread, m->lock_object.lo_name, file, line));
205	KASSERT(m->mtx_lock != MTX_DESTROYED,
206	    ("mtx_lock() of destroyed mutex @ %s:%d", file, line));
207	KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
208	    ("mtx_lock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
209	    file, line));
210	WITNESS_CHECKORDER(&m->lock_object, opts | LOP_NEWORDER | LOP_EXCLUSIVE,
211	    file, line, NULL);
212
213	__mtx_lock(m, curthread, opts, file, line);
214	LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
215	    line);
216	WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
217	curthread->td_locks++;
218}
219
220void
221_mtx_unlock_flags(struct mtx *m, int opts, const char *file, int line)
222{
223
224	if (SCHEDULER_STOPPED())
225		return;
226	KASSERT(m->mtx_lock != MTX_DESTROYED,
227	    ("mtx_unlock() of destroyed mutex @ %s:%d", file, line));
228	KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
229	    ("mtx_unlock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
230	    file, line));
231	curthread->td_locks--;
232	WITNESS_UNLOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
233	LOCK_LOG_LOCK("UNLOCK", &m->lock_object, opts, m->mtx_recurse, file,
234	    line);
235	mtx_assert(m, MA_OWNED);
236
237	if (m->mtx_recurse == 0)
238		LOCKSTAT_PROFILE_RELEASE_LOCK(LS_MTX_UNLOCK_RELEASE, m);
239	__mtx_unlock(m, curthread, opts, file, line);
240}
241
242void
243_mtx_lock_spin_flags(struct mtx *m, int opts, const char *file, int line)
244{
245
246	if (SCHEDULER_STOPPED())
247		return;
248	KASSERT(m->mtx_lock != MTX_DESTROYED,
249	    ("mtx_lock_spin() of destroyed mutex @ %s:%d", file, line));
250	KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
251	    ("mtx_lock_spin() of sleep mutex %s @ %s:%d",
252	    m->lock_object.lo_name, file, line));
253	if (mtx_owned(m))
254		KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
255	    ("mtx_lock_spin: recursed on non-recursive mutex %s @ %s:%d\n",
256		    m->lock_object.lo_name, file, line));
257	WITNESS_CHECKORDER(&m->lock_object, opts | LOP_NEWORDER | LOP_EXCLUSIVE,
258	    file, line, NULL);
259	__mtx_lock_spin(m, curthread, opts, file, line);
260	LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
261	    line);
262	WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
263}
264
265void
266_mtx_unlock_spin_flags(struct mtx *m, int opts, const char *file, int line)
267{
268
269	if (SCHEDULER_STOPPED())
270		return;
271	KASSERT(m->mtx_lock != MTX_DESTROYED,
272	    ("mtx_unlock_spin() of destroyed mutex @ %s:%d", file, line));
273	KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
274	    ("mtx_unlock_spin() of sleep mutex %s @ %s:%d",
275	    m->lock_object.lo_name, file, line));
276	WITNESS_UNLOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
277	LOCK_LOG_LOCK("UNLOCK", &m->lock_object, opts, m->mtx_recurse, file,
278	    line);
279	mtx_assert(m, MA_OWNED);
280
281	__mtx_unlock_spin(m);
282}
283
284/*
285 * The important part of mtx_trylock{,_flags}()
286 * Tries to acquire lock `m.'  If this function is called on a mutex that
287 * is already owned, it will recursively acquire the lock.
288 */
289int
290_mtx_trylock(struct mtx *m, int opts, const char *file, int line)
291{
292#ifdef LOCK_PROFILING
293	uint64_t waittime = 0;
294	int contested = 0;
295#endif
296	int rval;
297
298	if (SCHEDULER_STOPPED())
299		return (1);
300
301	KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
302	    ("mtx_trylock() by idle thread %p on sleep mutex %s @ %s:%d",
303	    curthread, m->lock_object.lo_name, file, line));
304	KASSERT(m->mtx_lock != MTX_DESTROYED,
305	    ("mtx_trylock() of destroyed mutex @ %s:%d", file, line));
306	KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
307	    ("mtx_trylock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
308	    file, line));
309
310	if (mtx_owned(m) && (m->lock_object.lo_flags & LO_RECURSABLE) != 0) {
311		m->mtx_recurse++;
312		atomic_set_ptr(&m->mtx_lock, MTX_RECURSED);
313		rval = 1;
314	} else
315		rval = _mtx_obtain_lock(m, (uintptr_t)curthread);
316
317	LOCK_LOG_TRY("LOCK", &m->lock_object, opts, rval, file, line);
318	if (rval) {
319		WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE | LOP_TRYLOCK,
320		    file, line);
321		curthread->td_locks++;
322		if (m->mtx_recurse == 0)
323			LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_LOCK_ACQUIRE,
324			    m, contested, waittime, file, line);
325
326	}
327
328	return (rval);
329}
330
331/*
332 * _mtx_lock_sleep: the tougher part of acquiring an MTX_DEF lock.
333 *
334 * We call this if the lock is either contested (i.e. we need to go to
335 * sleep waiting for it), or if we need to recurse on it.
336 */
337void
338_mtx_lock_sleep(struct mtx *m, uintptr_t tid, int opts, const char *file,
339    int line)
340{
341	struct turnstile *ts;
342	uintptr_t v;
343#ifdef ADAPTIVE_MUTEXES
344	volatile struct thread *owner;
345#endif
346#ifdef KTR
347	int cont_logged = 0;
348#endif
349#ifdef LOCK_PROFILING
350	int contested = 0;
351	uint64_t waittime = 0;
352#endif
353#ifdef KDTRACE_HOOKS
354	uint64_t spin_cnt = 0;
355	uint64_t sleep_cnt = 0;
356	int64_t sleep_time = 0;
357#endif
358
359	if (SCHEDULER_STOPPED())
360		return;
361
362	if (mtx_owned(m)) {
363		KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
364	    ("_mtx_lock_sleep: recursed on non-recursive mutex %s @ %s:%d\n",
365		    m->lock_object.lo_name, file, line));
366		m->mtx_recurse++;
367		atomic_set_ptr(&m->mtx_lock, MTX_RECURSED);
368		if (LOCK_LOG_TEST(&m->lock_object, opts))
369			CTR1(KTR_LOCK, "_mtx_lock_sleep: %p recursing", m);
370		return;
371	}
372
373#ifdef HWPMC_HOOKS
374	PMC_SOFT_CALL( , , lock, failed);
375#endif
376	lock_profile_obtain_lock_failed(&m->lock_object,
377		    &contested, &waittime);
378	if (LOCK_LOG_TEST(&m->lock_object, opts))
379		CTR4(KTR_LOCK,
380		    "_mtx_lock_sleep: %s contested (lock=%p) at %s:%d",
381		    m->lock_object.lo_name, (void *)m->mtx_lock, file, line);
382
383	while (!_mtx_obtain_lock(m, tid)) {
384#ifdef KDTRACE_HOOKS
385		spin_cnt++;
386#endif
387#ifdef ADAPTIVE_MUTEXES
388		/*
389		 * If the owner is running on another CPU, spin until the
390		 * owner stops running or the state of the lock changes.
391		 */
392		v = m->mtx_lock;
393		if (v != MTX_UNOWNED) {
394			owner = (struct thread *)(v & ~MTX_FLAGMASK);
395			if (TD_IS_RUNNING(owner)) {
396				if (LOCK_LOG_TEST(&m->lock_object, 0))
397					CTR3(KTR_LOCK,
398					    "%s: spinning on %p held by %p",
399					    __func__, m, owner);
400				while (mtx_owner(m) == owner &&
401				    TD_IS_RUNNING(owner)) {
402					cpu_spinwait();
403#ifdef KDTRACE_HOOKS
404					spin_cnt++;
405#endif
406				}
407				continue;
408			}
409		}
410#endif
411
412		ts = turnstile_trywait(&m->lock_object);
413		v = m->mtx_lock;
414
415		/*
416		 * Check if the lock has been released while spinning for
417		 * the turnstile chain lock.
418		 */
419		if (v == MTX_UNOWNED) {
420			turnstile_cancel(ts);
421			continue;
422		}
423
424#ifdef ADAPTIVE_MUTEXES
425		/*
426		 * The current lock owner might have started executing
427		 * on another CPU (or the lock could have changed
428		 * owners) while we were waiting on the turnstile
429		 * chain lock.  If so, drop the turnstile lock and try
430		 * again.
431		 */
432		owner = (struct thread *)(v & ~MTX_FLAGMASK);
433		if (TD_IS_RUNNING(owner)) {
434			turnstile_cancel(ts);
435			continue;
436		}
437#endif
438
439		/*
440		 * If the mutex isn't already contested and a failure occurs
441		 * setting the contested bit, the mutex was either released
442		 * or the state of the MTX_RECURSED bit changed.
443		 */
444		if ((v & MTX_CONTESTED) == 0 &&
445		    !atomic_cmpset_ptr(&m->mtx_lock, v, v | MTX_CONTESTED)) {
446			turnstile_cancel(ts);
447			continue;
448		}
449
450		/*
451		 * We definitely must sleep for this lock.
452		 */
453		mtx_assert(m, MA_NOTOWNED);
454
455#ifdef KTR
456		if (!cont_logged) {
457			CTR6(KTR_CONTENTION,
458			    "contention: %p at %s:%d wants %s, taken by %s:%d",
459			    (void *)tid, file, line, m->lock_object.lo_name,
460			    WITNESS_FILE(&m->lock_object),
461			    WITNESS_LINE(&m->lock_object));
462			cont_logged = 1;
463		}
464#endif
465
466		/*
467		 * Block on the turnstile.
468		 */
469#ifdef KDTRACE_HOOKS
470		sleep_time -= lockstat_nsecs();
471#endif
472		turnstile_wait(ts, mtx_owner(m), TS_EXCLUSIVE_QUEUE);
473#ifdef KDTRACE_HOOKS
474		sleep_time += lockstat_nsecs();
475		sleep_cnt++;
476#endif
477	}
478#ifdef KTR
479	if (cont_logged) {
480		CTR4(KTR_CONTENTION,
481		    "contention end: %s acquired by %p at %s:%d",
482		    m->lock_object.lo_name, (void *)tid, file, line);
483	}
484#endif
485	LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_LOCK_ACQUIRE, m, contested,
486	    waittime, file, line);
487#ifdef KDTRACE_HOOKS
488	if (sleep_time)
489		LOCKSTAT_RECORD1(LS_MTX_LOCK_BLOCK, m, sleep_time);
490
491	/*
492	 * Only record the loops spinning and not sleeping.
493	 */
494	if (spin_cnt > sleep_cnt)
495		LOCKSTAT_RECORD1(LS_MTX_LOCK_SPIN, m, (spin_cnt - sleep_cnt));
496#endif
497}
498
499static void
500_mtx_lock_spin_failed(struct mtx *m)
501{
502	struct thread *td;
503
504	td = mtx_owner(m);
505
506	/* If the mutex is unlocked, try again. */
507	if (td == NULL)
508		return;
509
510	printf( "spin lock %p (%s) held by %p (tid %d) too long\n",
511	    m, m->lock_object.lo_name, td, td->td_tid);
512#ifdef WITNESS
513	witness_display_spinlock(&m->lock_object, td, printf);
514#endif
515	panic("spin lock held too long");
516}
517
518#ifdef SMP
519/*
520 * _mtx_lock_spin: the tougher part of acquiring an MTX_SPIN lock.
521 *
522 * This is only called if we need to actually spin for the lock. Recursion
523 * is handled inline.
524 */
525void
526_mtx_lock_spin(struct mtx *m, uintptr_t tid, int opts, const char *file,
527    int line)
528{
529	int i = 0;
530#ifdef LOCK_PROFILING
531	int contested = 0;
532	uint64_t waittime = 0;
533#endif
534
535	if (SCHEDULER_STOPPED())
536		return;
537
538	if (LOCK_LOG_TEST(&m->lock_object, opts))
539		CTR1(KTR_LOCK, "_mtx_lock_spin: %p spinning", m);
540
541#ifdef HWPMC_HOOKS
542	PMC_SOFT_CALL( , , lock, failed);
543#endif
544	lock_profile_obtain_lock_failed(&m->lock_object, &contested, &waittime);
545	while (!_mtx_obtain_lock(m, tid)) {
546
547		/* Give interrupts a chance while we spin. */
548		spinlock_exit();
549		while (m->mtx_lock != MTX_UNOWNED) {
550			if (i++ < 10000000) {
551				cpu_spinwait();
552				continue;
553			}
554			if (i < 60000000 || kdb_active || panicstr != NULL)
555				DELAY(1);
556			else
557				_mtx_lock_spin_failed(m);
558			cpu_spinwait();
559		}
560		spinlock_enter();
561	}
562
563	if (LOCK_LOG_TEST(&m->lock_object, opts))
564		CTR1(KTR_LOCK, "_mtx_lock_spin: %p spin done", m);
565
566	LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_SPIN_LOCK_ACQUIRE, m,
567	    contested, waittime, (file), (line));
568	LOCKSTAT_RECORD1(LS_MTX_SPIN_LOCK_SPIN, m, i);
569}
570#endif /* SMP */
571
572void
573_thread_lock_flags(struct thread *td, int opts, const char *file, int line)
574{
575	struct mtx *m;
576	uintptr_t tid;
577	int i;
578#ifdef LOCK_PROFILING
579	int contested = 0;
580	uint64_t waittime = 0;
581#endif
582#ifdef KDTRACE_HOOKS
583	uint64_t spin_cnt = 0;
584#endif
585
586	i = 0;
587	tid = (uintptr_t)curthread;
588
589	if (SCHEDULER_STOPPED())
590		return;
591
592	for (;;) {
593retry:
594		spinlock_enter();
595		m = td->td_lock;
596		KASSERT(m->mtx_lock != MTX_DESTROYED,
597		    ("thread_lock() of destroyed mutex @ %s:%d", file, line));
598		KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
599		    ("thread_lock() of sleep mutex %s @ %s:%d",
600		    m->lock_object.lo_name, file, line));
601		if (mtx_owned(m))
602			KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
603	    ("thread_lock: recursed on non-recursive mutex %s @ %s:%d\n",
604			    m->lock_object.lo_name, file, line));
605		WITNESS_CHECKORDER(&m->lock_object,
606		    opts | LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL);
607		while (!_mtx_obtain_lock(m, tid)) {
608#ifdef KDTRACE_HOOKS
609			spin_cnt++;
610#endif
611			if (m->mtx_lock == tid) {
612				m->mtx_recurse++;
613				break;
614			}
615#ifdef HWPMC_HOOKS
616			PMC_SOFT_CALL( , , lock, failed);
617#endif
618			lock_profile_obtain_lock_failed(&m->lock_object,
619			    &contested, &waittime);
620			/* Give interrupts a chance while we spin. */
621			spinlock_exit();
622			while (m->mtx_lock != MTX_UNOWNED) {
623				if (i++ < 10000000)
624					cpu_spinwait();
625				else if (i < 60000000 ||
626				    kdb_active || panicstr != NULL)
627					DELAY(1);
628				else
629					_mtx_lock_spin_failed(m);
630				cpu_spinwait();
631				if (m != td->td_lock)
632					goto retry;
633			}
634			spinlock_enter();
635		}
636		if (m == td->td_lock)
637			break;
638		__mtx_unlock_spin(m);	/* does spinlock_exit() */
639#ifdef KDTRACE_HOOKS
640		spin_cnt++;
641#endif
642	}
643	if (m->mtx_recurse == 0)
644		LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_SPIN_LOCK_ACQUIRE,
645		    m, contested, waittime, (file), (line));
646	LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
647	    line);
648	WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
649	LOCKSTAT_RECORD1(LS_THREAD_LOCK_SPIN, m, spin_cnt);
650}
651
652struct mtx *
653thread_lock_block(struct thread *td)
654{
655	struct mtx *lock;
656
657	THREAD_LOCK_ASSERT(td, MA_OWNED);
658	lock = td->td_lock;
659	td->td_lock = &blocked_lock;
660	mtx_unlock_spin(lock);
661
662	return (lock);
663}
664
665void
666thread_lock_unblock(struct thread *td, struct mtx *new)
667{
668	mtx_assert(new, MA_OWNED);
669	MPASS(td->td_lock == &blocked_lock);
670	atomic_store_rel_ptr((volatile void *)&td->td_lock, (uintptr_t)new);
671}
672
673void
674thread_lock_set(struct thread *td, struct mtx *new)
675{
676	struct mtx *lock;
677
678	mtx_assert(new, MA_OWNED);
679	THREAD_LOCK_ASSERT(td, MA_OWNED);
680	lock = td->td_lock;
681	td->td_lock = new;
682	mtx_unlock_spin(lock);
683}
684
685/*
686 * _mtx_unlock_sleep: the tougher part of releasing an MTX_DEF lock.
687 *
688 * We are only called here if the lock is recursed or contested (i.e. we
689 * need to wake up a blocked thread).
690 */
691void
692_mtx_unlock_sleep(struct mtx *m, int opts, const char *file, int line)
693{
694	struct turnstile *ts;
695
696	if (SCHEDULER_STOPPED())
697		return;
698
699	if (mtx_recursed(m)) {
700		if (--(m->mtx_recurse) == 0)
701			atomic_clear_ptr(&m->mtx_lock, MTX_RECURSED);
702		if (LOCK_LOG_TEST(&m->lock_object, opts))
703			CTR1(KTR_LOCK, "_mtx_unlock_sleep: %p unrecurse", m);
704		return;
705	}
706
707	/*
708	 * We have to lock the chain before the turnstile so this turnstile
709	 * can be removed from the hash list if it is empty.
710	 */
711	turnstile_chain_lock(&m->lock_object);
712	ts = turnstile_lookup(&m->lock_object);
713	if (LOCK_LOG_TEST(&m->lock_object, opts))
714		CTR1(KTR_LOCK, "_mtx_unlock_sleep: %p contested", m);
715	MPASS(ts != NULL);
716	turnstile_broadcast(ts, TS_EXCLUSIVE_QUEUE);
717	_mtx_release_lock_quick(m);
718
719	/*
720	 * This turnstile is now no longer associated with the mutex.  We can
721	 * unlock the chain lock so a new turnstile may take it's place.
722	 */
723	turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
724	turnstile_chain_unlock(&m->lock_object);
725}
726
727/*
728 * All the unlocking of MTX_SPIN locks is done inline.
729 * See the __mtx_unlock_spin() macro for the details.
730 */
731
732/*
733 * The backing function for the INVARIANTS-enabled mtx_assert()
734 */
735#ifdef INVARIANT_SUPPORT
736void
737_mtx_assert(struct mtx *m, int what, const char *file, int line)
738{
739
740	if (panicstr != NULL || dumping)
741		return;
742	switch (what) {
743	case MA_OWNED:
744	case MA_OWNED | MA_RECURSED:
745	case MA_OWNED | MA_NOTRECURSED:
746		if (!mtx_owned(m))
747			panic("mutex %s not owned at %s:%d",
748			    m->lock_object.lo_name, file, line);
749		if (mtx_recursed(m)) {
750			if ((what & MA_NOTRECURSED) != 0)
751				panic("mutex %s recursed at %s:%d",
752				    m->lock_object.lo_name, file, line);
753		} else if ((what & MA_RECURSED) != 0) {
754			panic("mutex %s unrecursed at %s:%d",
755			    m->lock_object.lo_name, file, line);
756		}
757		break;
758	case MA_NOTOWNED:
759		if (mtx_owned(m))
760			panic("mutex %s owned at %s:%d",
761			    m->lock_object.lo_name, file, line);
762		break;
763	default:
764		panic("unknown mtx_assert at %s:%d", file, line);
765	}
766}
767#endif
768
769/*
770 * The MUTEX_DEBUG-enabled mtx_validate()
771 *
772 * Most of these checks have been moved off into the LO_INITIALIZED flag
773 * maintained by the witness code.
774 */
775#ifdef MUTEX_DEBUG
776
777void	mtx_validate(struct mtx *);
778
779void
780mtx_validate(struct mtx *m)
781{
782
783/*
784 * XXX: When kernacc() does not require Giant we can reenable this check
785 */
786#ifdef notyet
787	/*
788	 * Can't call kernacc() from early init386(), especially when
789	 * initializing Giant mutex, because some stuff in kernacc()
790	 * requires Giant itself.
791	 */
792	if (!cold)
793		if (!kernacc((caddr_t)m, sizeof(m),
794		    VM_PROT_READ | VM_PROT_WRITE))
795			panic("Can't read and write to mutex %p", m);
796#endif
797}
798#endif
799
800/*
801 * General init routine used by the MTX_SYSINIT() macro.
802 */
803void
804mtx_sysinit(void *arg)
805{
806	struct mtx_args *margs = arg;
807
808	mtx_init(margs->ma_mtx, margs->ma_desc, NULL, margs->ma_opts);
809}
810
811/*
812 * Mutex initialization routine; initialize lock `m' of type contained in
813 * `opts' with options contained in `opts' and name `name.'  The optional
814 * lock type `type' is used as a general lock category name for use with
815 * witness.
816 */
817void
818mtx_init(struct mtx *m, const char *name, const char *type, int opts)
819{
820	struct lock_class *class;
821	int flags;
822
823	MPASS((opts & ~(MTX_SPIN | MTX_QUIET | MTX_RECURSE |
824		MTX_NOWITNESS | MTX_DUPOK | MTX_NOPROFILE)) == 0);
825	ASSERT_ATOMIC_LOAD_PTR(m->mtx_lock,
826	    ("%s: mtx_lock not aligned for %s: %p", __func__, name,
827	    &m->mtx_lock));
828
829#ifdef MUTEX_DEBUG
830	/* Diagnostic and error correction */
831	mtx_validate(m);
832#endif
833
834	/* Determine lock class and lock flags. */
835	if (opts & MTX_SPIN)
836		class = &lock_class_mtx_spin;
837	else
838		class = &lock_class_mtx_sleep;
839	flags = 0;
840	if (opts & MTX_QUIET)
841		flags |= LO_QUIET;
842	if (opts & MTX_RECURSE)
843		flags |= LO_RECURSABLE;
844	if ((opts & MTX_NOWITNESS) == 0)
845		flags |= LO_WITNESS;
846	if (opts & MTX_DUPOK)
847		flags |= LO_DUPOK;
848	if (opts & MTX_NOPROFILE)
849		flags |= LO_NOPROFILE;
850
851	/* Initialize mutex. */
852	m->mtx_lock = MTX_UNOWNED;
853	m->mtx_recurse = 0;
854
855	lock_init(&m->lock_object, class, name, type, flags);
856}
857
858/*
859 * Remove lock `m' from all_mtx queue.  We don't allow MTX_QUIET to be
860 * passed in as a flag here because if the corresponding mtx_init() was
861 * called with MTX_QUIET set, then it will already be set in the mutex's
862 * flags.
863 */
864void
865mtx_destroy(struct mtx *m)
866{
867
868	if (!mtx_owned(m))
869		MPASS(mtx_unowned(m));
870	else {
871		MPASS((m->mtx_lock & (MTX_RECURSED|MTX_CONTESTED)) == 0);
872
873		/* Perform the non-mtx related part of mtx_unlock_spin(). */
874		if (LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin)
875			spinlock_exit();
876		else
877			curthread->td_locks--;
878
879		lock_profile_release_lock(&m->lock_object);
880		/* Tell witness this isn't locked to make it happy. */
881		WITNESS_UNLOCK(&m->lock_object, LOP_EXCLUSIVE, __FILE__,
882		    __LINE__);
883	}
884
885	m->mtx_lock = MTX_DESTROYED;
886	lock_destroy(&m->lock_object);
887}
888
889/*
890 * Intialize the mutex code and system mutexes.  This is called from the MD
891 * startup code prior to mi_startup().  The per-CPU data space needs to be
892 * setup before this is called.
893 */
894void
895mutex_init(void)
896{
897
898	/* Setup turnstiles so that sleep mutexes work. */
899	init_turnstiles();
900
901	/*
902	 * Initialize mutexes.
903	 */
904	mtx_init(&Giant, "Giant", NULL, MTX_DEF | MTX_RECURSE);
905	mtx_init(&blocked_lock, "blocked lock", NULL, MTX_SPIN);
906	blocked_lock.mtx_lock = 0xdeadc0de;	/* Always blocked. */
907	mtx_init(&proc0.p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK);
908	mtx_init(&proc0.p_slock, "process slock", NULL, MTX_SPIN | MTX_RECURSE);
909	mtx_init(&devmtx, "cdev", NULL, MTX_DEF);
910	mtx_lock(&Giant);
911}
912
913#ifdef DDB
914void
915db_show_mtx(struct lock_object *lock)
916{
917	struct thread *td;
918	struct mtx *m;
919
920	m = (struct mtx *)lock;
921
922	db_printf(" flags: {");
923	if (LOCK_CLASS(lock) == &lock_class_mtx_spin)
924		db_printf("SPIN");
925	else
926		db_printf("DEF");
927	if (m->lock_object.lo_flags & LO_RECURSABLE)
928		db_printf(", RECURSE");
929	if (m->lock_object.lo_flags & LO_DUPOK)
930		db_printf(", DUPOK");
931	db_printf("}\n");
932	db_printf(" state: {");
933	if (mtx_unowned(m))
934		db_printf("UNOWNED");
935	else if (mtx_destroyed(m))
936		db_printf("DESTROYED");
937	else {
938		db_printf("OWNED");
939		if (m->mtx_lock & MTX_CONTESTED)
940			db_printf(", CONTESTED");
941		if (m->mtx_lock & MTX_RECURSED)
942			db_printf(", RECURSED");
943	}
944	db_printf("}\n");
945	if (!mtx_unowned(m) && !mtx_destroyed(m)) {
946		td = mtx_owner(m);
947		db_printf(" owner: %p (tid %d, pid %d, \"%s\")\n", td,
948		    td->td_tid, td->td_proc->p_pid, td->td_name);
949		if (mtx_recursed(m))
950			db_printf(" recursed: %d\n", m->mtx_recurse);
951	}
952}
953#endif
954