1/*-
2 * Copyright (c) 1997 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.h,v 2.7.2.35 2000/04/27 03:10:26 cp Exp $
29 * $FreeBSD$
30 */
31
32#ifndef _SYS_MUTEX_H_
33#define _SYS_MUTEX_H_
34
35#include <sys/queue.h>
36#include <sys/_lock.h>
37#include <sys/_mutex.h>
38
39#ifdef _KERNEL
40#include <sys/pcpu.h>
41#include <sys/lock_profile.h>
42#include <sys/lockstat.h>
43#include <machine/atomic.h>
44#include <machine/cpufunc.h>
45
46/*
47 * Mutex types and options passed to mtx_init().  MTX_QUIET and MTX_DUPOK
48 * can also be passed in.
49 */
50#define	MTX_DEF		0x00000000	/* DEFAULT (sleep) lock */
51#define MTX_SPIN	0x00000001	/* Spin lock (disables interrupts) */
52#define MTX_RECURSE	0x00000004	/* Option: lock allowed to recurse */
53#define	MTX_NOWITNESS	0x00000008	/* Don't do any witness checking. */
54#define MTX_NOPROFILE   0x00000020	/* Don't profile this lock */
55
56/*
57 * Option flags passed to certain lock/unlock routines, through the use
58 * of corresponding mtx_{lock,unlock}_flags() interface macros.
59 */
60#define	MTX_QUIET	LOP_QUIET	/* Don't log a mutex event */
61#define	MTX_DUPOK	LOP_DUPOK	/* Don't log a duplicate acquire */
62
63/*
64 * State bits kept in mutex->mtx_lock, for the DEFAULT lock type. None of this,
65 * with the exception of MTX_UNOWNED, applies to spin locks.
66 */
67#define	MTX_RECURSED	0x00000001	/* lock recursed (for MTX_DEF only) */
68#define	MTX_CONTESTED	0x00000002	/* lock contested (for MTX_DEF only) */
69#define MTX_UNOWNED	0x00000004	/* Cookie for free mutex */
70#define	MTX_FLAGMASK	(MTX_RECURSED | MTX_CONTESTED | MTX_UNOWNED)
71
72/*
73 * Value stored in mutex->mtx_lock to denote a destroyed mutex.
74 */
75#define	MTX_DESTROYED	(MTX_CONTESTED | MTX_UNOWNED)
76
77/*
78 * Prototypes
79 *
80 * NOTE: Functions prepended with `_' (underscore) are exported to other parts
81 *	 of the kernel via macros, thus allowing us to use the cpp LOCK_FILE
82 *	 and LOCK_LINE or for hiding the lock cookie crunching to the
83 *	 consumers. These functions should not be called directly by any
84 *	 code using the API. Their macros cover their functionality.
85 *	 Functions with a `_' suffix are the entrypoint for the common
86 *	 KPI covering both compat shims and fast path case.  These can be
87 *	 used by consumers willing to pass options, file and line
88 *	 informations, in an option-independent way.
89 *
90 * [See below for descriptions]
91 *
92 */
93void	_mtx_init(volatile uintptr_t *c, const char *name, const char *type,
94	    int opts);
95void	_mtx_destroy(volatile uintptr_t *c);
96void	mtx_sysinit(void *arg);
97int	_mtx_trylock_flags_(volatile uintptr_t *c, int opts, const char *file,
98	    int line);
99void	mutex_init(void);
100void	__mtx_lock_sleep(volatile uintptr_t *c, uintptr_t tid, int opts,
101	    const char *file, int line);
102void	__mtx_unlock_sleep(volatile uintptr_t *c, int opts, const char *file,
103	    int line);
104#ifdef SMP
105void	_mtx_lock_spin_cookie(volatile uintptr_t *c, uintptr_t tid, int opts,
106	    const char *file, int line);
107#endif
108void	__mtx_lock_flags(volatile uintptr_t *c, int opts, const char *file,
109	    int line);
110void	__mtx_unlock_flags(volatile uintptr_t *c, int opts, const char *file,
111	    int line);
112void	__mtx_lock_spin_flags(volatile uintptr_t *c, int opts, const char *file,
113	     int line);
114void	__mtx_unlock_spin_flags(volatile uintptr_t *c, int opts,
115	    const char *file, int line);
116#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
117void	__mtx_assert(const volatile uintptr_t *c, int what, const char *file,
118	    int line);
119#endif
120void	thread_lock_flags_(struct thread *, int, const char *, int);
121
122#define	thread_lock(tdp)						\
123	thread_lock_flags_((tdp), 0, __FILE__, __LINE__)
124#define	thread_lock_flags(tdp, opt)					\
125	thread_lock_flags_((tdp), (opt), __FILE__, __LINE__)
126#define	thread_unlock(tdp)						\
127       mtx_unlock_spin((tdp)->td_lock)
128
129/*
130 * Top-level macros to provide lock cookie once the actual mtx is passed.
131 * They will also prevent passing a malformed object to the mtx KPI by
132 * failing compilation as the mtx_lock reserved member will not be found.
133 */
134#define	mtx_init(m, n, t, o)						\
135	_mtx_init(&(m)->mtx_lock, n, t, o)
136#define	mtx_destroy(m)							\
137	_mtx_destroy(&(m)->mtx_lock)
138#define	mtx_trylock_flags_(m, o, f, l)					\
139	_mtx_trylock_flags_(&(m)->mtx_lock, o, f, l)
140#define	_mtx_lock_sleep(m, t, o, f, l)					\
141	__mtx_lock_sleep(&(m)->mtx_lock, t, o, f, l)
142#define	_mtx_unlock_sleep(m, o, f, l)					\
143	__mtx_unlock_sleep(&(m)->mtx_lock, o, f, l)
144#ifdef SMP
145#define	_mtx_lock_spin(m, t, o, f, l)					\
146	_mtx_lock_spin_cookie(&(m)->mtx_lock, t, o, f, l)
147#endif
148#define	_mtx_lock_flags(m, o, f, l)					\
149	__mtx_lock_flags(&(m)->mtx_lock, o, f, l)
150#define	_mtx_unlock_flags(m, o, f, l)					\
151	__mtx_unlock_flags(&(m)->mtx_lock, o, f, l)
152#define	_mtx_lock_spin_flags(m, o, f, l)				\
153	__mtx_lock_spin_flags(&(m)->mtx_lock, o, f, l)
154#define	_mtx_unlock_spin_flags(m, o, f, l)				\
155	__mtx_unlock_spin_flags(&(m)->mtx_lock, o, f, l)
156#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
157#define	_mtx_assert(m, w, f, l)						\
158	__mtx_assert(&(m)->mtx_lock, w, f, l)
159#endif
160
161#define	mtx_recurse	lock_object.lo_data
162
163/* Very simple operations on mtx_lock. */
164
165/* Try to obtain mtx_lock once. */
166#define _mtx_obtain_lock(mp, tid)					\
167	atomic_cmpset_acq_ptr(&(mp)->mtx_lock, MTX_UNOWNED, (tid))
168
169/* Try to release mtx_lock if it is unrecursed and uncontested. */
170#define _mtx_release_lock(mp, tid)					\
171	atomic_cmpset_rel_ptr(&(mp)->mtx_lock, (tid), MTX_UNOWNED)
172
173/* Release mtx_lock quickly, assuming we own it. */
174#define _mtx_release_lock_quick(mp)					\
175	atomic_store_rel_ptr(&(mp)->mtx_lock, MTX_UNOWNED)
176
177/*
178 * Full lock operations that are suitable to be inlined in non-debug
179 * kernels.  If the lock cannot be acquired or released trivially then
180 * the work is deferred to another function.
181 */
182
183/* Lock a normal mutex. */
184#define __mtx_lock(mp, tid, opts, file, line) do {			\
185	uintptr_t _tid = (uintptr_t)(tid);				\
186									\
187	if (!_mtx_obtain_lock((mp), _tid))				\
188		_mtx_lock_sleep((mp), _tid, (opts), (file), (line));	\
189	else								\
190              	LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_LOCK_ACQUIRE, \
191		    mp, 0, 0, (file), (line));				\
192} while (0)
193
194/*
195 * Lock a spin mutex.  For spinlocks, we handle recursion inline (it
196 * turns out that function calls can be significantly expensive on
197 * some architectures).  Since spin locks are not _too_ common,
198 * inlining this code is not too big a deal.
199 */
200#ifdef SMP
201#define __mtx_lock_spin(mp, tid, opts, file, line) do {			\
202	uintptr_t _tid = (uintptr_t)(tid);				\
203									\
204	spinlock_enter();						\
205	if (!_mtx_obtain_lock((mp), _tid)) {				\
206		if ((mp)->mtx_lock == _tid)				\
207			(mp)->mtx_recurse++;				\
208		else							\
209			_mtx_lock_spin((mp), _tid, (opts), (file), (line)); \
210	} else 								\
211              	LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_SPIN_LOCK_ACQUIRE, \
212		    mp, 0, 0, (file), (line));				\
213} while (0)
214#else /* SMP */
215#define __mtx_lock_spin(mp, tid, opts, file, line) do {			\
216	uintptr_t _tid = (uintptr_t)(tid);				\
217									\
218	spinlock_enter();						\
219	if ((mp)->mtx_lock == _tid)					\
220		(mp)->mtx_recurse++;					\
221	else {								\
222		KASSERT((mp)->mtx_lock == MTX_UNOWNED, ("corrupt spinlock")); \
223		(mp)->mtx_lock = _tid;					\
224	}								\
225} while (0)
226#endif /* SMP */
227
228/* Unlock a normal mutex. */
229#define __mtx_unlock(mp, tid, opts, file, line) do {			\
230	uintptr_t _tid = (uintptr_t)(tid);				\
231									\
232	if (!_mtx_release_lock((mp), _tid))				\
233		_mtx_unlock_sleep((mp), (opts), (file), (line));	\
234} while (0)
235
236/*
237 * Unlock a spin mutex.  For spinlocks, we can handle everything
238 * inline, as it's pretty simple and a function call would be too
239 * expensive (at least on some architectures).  Since spin locks are
240 * not _too_ common, inlining this code is not too big a deal.
241 *
242 * Since we always perform a spinlock_enter() when attempting to acquire a
243 * spin lock, we need to always perform a matching spinlock_exit() when
244 * releasing a spin lock.  This includes the recursion cases.
245 */
246#ifdef SMP
247#define __mtx_unlock_spin(mp) do {					\
248	if (mtx_recursed((mp)))						\
249		(mp)->mtx_recurse--;					\
250	else {								\
251		LOCKSTAT_PROFILE_RELEASE_LOCK(LS_MTX_SPIN_UNLOCK_RELEASE, \
252			mp);						\
253		_mtx_release_lock_quick((mp));				\
254	}                                                               \
255	spinlock_exit();				                \
256} while (0)
257#else /* SMP */
258#define __mtx_unlock_spin(mp) do {					\
259	if (mtx_recursed((mp)))						\
260		(mp)->mtx_recurse--;					\
261	else {								\
262		LOCKSTAT_PROFILE_RELEASE_LOCK(LS_MTX_SPIN_UNLOCK_RELEASE, \
263			mp);						\
264		(mp)->mtx_lock = MTX_UNOWNED;				\
265	}                                                               \
266	spinlock_exit();						\
267} while (0)
268#endif /* SMP */
269
270/*
271 * Exported lock manipulation interface.
272 *
273 * mtx_lock(m) locks MTX_DEF mutex `m'
274 *
275 * mtx_lock_spin(m) locks MTX_SPIN mutex `m'
276 *
277 * mtx_unlock(m) unlocks MTX_DEF mutex `m'
278 *
279 * mtx_unlock_spin(m) unlocks MTX_SPIN mutex `m'
280 *
281 * mtx_lock_spin_flags(m, opts) and mtx_lock_flags(m, opts) locks mutex `m'
282 *     and passes option flags `opts' to the "hard" function, if required.
283 *     With these routines, it is possible to pass flags such as MTX_QUIET
284 *     to the appropriate lock manipulation routines.
285 *
286 * mtx_trylock(m) attempts to acquire MTX_DEF mutex `m' but doesn't sleep if
287 *     it cannot. Rather, it returns 0 on failure and non-zero on success.
288 *     It does NOT handle recursion as we assume that if a caller is properly
289 *     using this part of the interface, he will know that the lock in question
290 *     is _not_ recursed.
291 *
292 * mtx_trylock_flags(m, opts) is used the same way as mtx_trylock() but accepts
293 *     relevant option flags `opts.'
294 *
295 * mtx_initialized(m) returns non-zero if the lock `m' has been initialized.
296 *
297 * mtx_owned(m) returns non-zero if the current thread owns the lock `m'
298 *
299 * mtx_recursed(m) returns non-zero if the lock `m' is presently recursed.
300 */
301#define mtx_lock(m)		mtx_lock_flags((m), 0)
302#define mtx_lock_spin(m)	mtx_lock_spin_flags((m), 0)
303#define mtx_trylock(m)		mtx_trylock_flags((m), 0)
304#define mtx_unlock(m)		mtx_unlock_flags((m), 0)
305#define mtx_unlock_spin(m)	mtx_unlock_spin_flags((m), 0)
306
307struct mtx_pool;
308
309struct mtx_pool *mtx_pool_create(const char *mtx_name, int pool_size, int opts);
310void mtx_pool_destroy(struct mtx_pool **poolp);
311struct mtx *mtx_pool_find(struct mtx_pool *pool, void *ptr);
312struct mtx *mtx_pool_alloc(struct mtx_pool *pool);
313#define mtx_pool_lock(pool, ptr)					\
314	mtx_lock(mtx_pool_find((pool), (ptr)))
315#define mtx_pool_lock_spin(pool, ptr)					\
316	mtx_lock_spin(mtx_pool_find((pool), (ptr)))
317#define mtx_pool_unlock(pool, ptr)					\
318	mtx_unlock(mtx_pool_find((pool), (ptr)))
319#define mtx_pool_unlock_spin(pool, ptr)					\
320	mtx_unlock_spin(mtx_pool_find((pool), (ptr)))
321
322/*
323 * mtxpool_lockbuilder is a pool of sleep locks that is not witness
324 * checked and should only be used for building higher level locks.
325 *
326 * mtxpool_sleep is a general purpose pool of sleep mutexes.
327 */
328extern struct mtx_pool *mtxpool_lockbuilder;
329extern struct mtx_pool *mtxpool_sleep;
330
331#ifndef LOCK_DEBUG
332#error LOCK_DEBUG not defined, include <sys/lock.h> before <sys/mutex.h>
333#endif
334#if LOCK_DEBUG > 0 || defined(MUTEX_NOINLINE)
335#define	mtx_lock_flags_(m, opts, file, line)				\
336	_mtx_lock_flags((m), (opts), (file), (line))
337#define	mtx_unlock_flags_(m, opts, file, line)				\
338	_mtx_unlock_flags((m), (opts), (file), (line))
339#define	mtx_lock_spin_flags_(m, opts, file, line)			\
340	_mtx_lock_spin_flags((m), (opts), (file), (line))
341#define	mtx_unlock_spin_flags_(m, opts, file, line)			\
342	_mtx_unlock_spin_flags((m), (opts), (file), (line))
343#else	/* LOCK_DEBUG == 0 && !MUTEX_NOINLINE */
344#define	mtx_lock_flags_(m, opts, file, line)				\
345	__mtx_lock((m), curthread, (opts), (file), (line))
346#define	mtx_unlock_flags_(m, opts, file, line)				\
347	__mtx_unlock((m), curthread, (opts), (file), (line))
348#define	mtx_lock_spin_flags_(m, opts, file, line)			\
349	__mtx_lock_spin((m), curthread, (opts), (file), (line))
350#define	mtx_unlock_spin_flags_(m, opts, file, line)			\
351	__mtx_unlock_spin((m))
352#endif	/* LOCK_DEBUG > 0 || MUTEX_NOINLINE */
353
354#ifdef INVARIANTS
355#define	mtx_assert_(m, what, file, line)				\
356	_mtx_assert((m), (what), (file), (line))
357
358#define GIANT_REQUIRED	mtx_assert_(&Giant, MA_OWNED, __FILE__, __LINE__)
359
360#else	/* INVARIANTS */
361#define mtx_assert_(m, what, file, line)	(void)0
362#define GIANT_REQUIRED
363#endif	/* INVARIANTS */
364
365#define	mtx_lock_flags(m, opts)						\
366	mtx_lock_flags_((m), (opts), LOCK_FILE, LOCK_LINE)
367#define	mtx_unlock_flags(m, opts)					\
368	mtx_unlock_flags_((m), (opts), LOCK_FILE, LOCK_LINE)
369#define	mtx_lock_spin_flags(m, opts)					\
370	mtx_lock_spin_flags_((m), (opts), LOCK_FILE, LOCK_LINE)
371#define	mtx_unlock_spin_flags(m, opts)					\
372	mtx_unlock_spin_flags_((m), (opts), LOCK_FILE, LOCK_LINE)
373#define mtx_trylock_flags(m, opts)					\
374	mtx_trylock_flags_((m), (opts), LOCK_FILE, LOCK_LINE)
375#define	mtx_assert(m, what)						\
376	mtx_assert_((m), (what), __FILE__, __LINE__)
377
378#define	mtx_sleep(chan, mtx, pri, wmesg, timo)				\
379	_sleep((chan), &(mtx)->lock_object, (pri), (wmesg),		\
380	    tick_sbt * (timo), 0, C_HARDCLOCK)
381
382#define	mtx_initialized(m)	lock_initalized(&(m)->lock_object)
383
384#define mtx_owned(m)	(((m)->mtx_lock & ~MTX_FLAGMASK) == (uintptr_t)curthread)
385
386#define mtx_recursed(m)	((m)->mtx_recurse != 0)
387
388#define mtx_name(m)	((m)->lock_object.lo_name)
389
390/*
391 * Global locks.
392 */
393extern struct mtx Giant;
394extern struct mtx blocked_lock;
395
396/*
397 * Giant lock manipulation and clean exit macros.
398 * Used to replace return with an exit Giant and return.
399 *
400 * Note that DROP_GIANT*() needs to be paired with PICKUP_GIANT()
401 * The #ifndef is to allow lint-like tools to redefine DROP_GIANT.
402 */
403#ifndef DROP_GIANT
404#define DROP_GIANT()							\
405do {									\
406	int _giantcnt = 0;						\
407	WITNESS_SAVE_DECL(Giant);					\
408									\
409	if (mtx_owned(&Giant)) {					\
410		WITNESS_SAVE(&Giant.lock_object, Giant);		\
411		for (_giantcnt = 0; mtx_owned(&Giant) &&		\
412		    !SCHEDULER_STOPPED(); _giantcnt++)			\
413			mtx_unlock(&Giant);				\
414	}
415
416#define PICKUP_GIANT()							\
417	PARTIAL_PICKUP_GIANT();						\
418} while (0)
419
420#define PARTIAL_PICKUP_GIANT()						\
421	mtx_assert(&Giant, MA_NOTOWNED);				\
422	if (_giantcnt > 0) {						\
423		while (_giantcnt--)					\
424			mtx_lock(&Giant);				\
425		WITNESS_RESTORE(&Giant.lock_object, Giant);		\
426	}
427#endif
428
429struct mtx_args {
430	void		*ma_mtx;
431	const char 	*ma_desc;
432	int		 ma_opts;
433};
434
435#define	MTX_SYSINIT(name, mtx, desc, opts)				\
436	static struct mtx_args name##_args = {				\
437		(mtx),							\
438		(desc),							\
439		(opts)							\
440	};								\
441	SYSINIT(name##_mtx_sysinit, SI_SUB_LOCK, SI_ORDER_MIDDLE,	\
442	    mtx_sysinit, &name##_args);					\
443	SYSUNINIT(name##_mtx_sysuninit, SI_SUB_LOCK, SI_ORDER_MIDDLE,	\
444	    _mtx_destroy, __DEVOLATILE(void *, &(mtx)->mtx_lock))
445
446/*
447 * The INVARIANTS-enabled mtx_assert() functionality.
448 *
449 * The constants need to be defined for INVARIANT_SUPPORT infrastructure
450 * support as _mtx_assert() itself uses them and the latter implies that
451 * _mtx_assert() must build.
452 */
453#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
454#define MA_OWNED	LA_XLOCKED
455#define MA_NOTOWNED	LA_UNLOCKED
456#define MA_RECURSED	LA_RECURSED
457#define MA_NOTRECURSED	LA_NOTRECURSED
458#endif
459
460/*
461 * Common lock type names.
462 */
463#define	MTX_NETWORK_LOCK	"network driver"
464
465#endif	/* _KERNEL */
466#endif	/* _SYS_MUTEX_H_ */
467