xref: /netbsd-src/share/man/man9/mutex.9 (revision aef5eb5f59cdfe8314f1b5f78ac04eb144e44010)
1.\"	$NetBSD: mutex.9,v 1.32 2022/08/20 14:17:36 riastradh Exp $
2.\"
3.\" Copyright (c) 2007, 2009 The NetBSD Foundation, Inc.
4.\" All rights reserved.
5.\"
6.\" This code is derived from software contributed to The NetBSD Foundation
7.\" by Andrew Doran.
8.\"
9.\" Redistribution and use in source and binary forms, with or without
10.\" modification, are permitted provided that the following conditions
11.\" are met:
12.\" 1. Redistributions of source code must retain the above copyright
13.\"    notice, this list of conditions and the following disclaimer.
14.\" 2. Redistributions in binary form must reproduce the above copyright
15.\"    notice, this list of conditions and the following disclaimer in the
16.\"    documentation and/or other materials provided with the distribution.
17.\"
18.\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19.\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20.\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21.\" PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22.\" BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23.\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24.\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25.\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26.\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27.\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28.\" POSSIBILITY OF SUCH DAMAGE.
29.\"
30.Dd December 8, 2017
31.Dt MUTEX 9
32.Os
33.Sh NAME
34.Nm mutex ,
35.Nm mutex_init ,
36.Nm mutex_destroy ,
37.Nm mutex_enter ,
38.Nm mutex_exit ,
39.Nm mutex_ownable ,
40.Nm mutex_owned ,
41.Nm mutex_spin_enter ,
42.Nm mutex_spin_exit ,
43.Nm mutex_tryenter
44.Nd mutual exclusion primitives
45.Sh SYNOPSIS
46.In sys/mutex.h
47.Ft void
48.Fn mutex_init "kmutex_t *mtx" "kmutex_type_t type" "int ipl"
49.Ft void
50.Fn mutex_destroy "kmutex_t *mtx"
51.Ft void
52.Fn mutex_enter "kmutex_t *mtx"
53.Ft void
54.Fn mutex_exit "kmutex_t *mtx"
55.Ft int
56.Fn mutex_ownable "kmutex_t *mtx"
57.Ft int
58.Fn mutex_owned "kmutex_t *mtx"
59.Ft void
60.Fn mutex_spin_enter "kmutex_t *mtx"
61.Ft void
62.Fn mutex_spin_exit "kmutex_t *mtx"
63.Ft int
64.Fn mutex_tryenter "kmutex_t *mtx"
65.Pp
66.Cd "options DIAGNOSTIC"
67.Cd "options LOCKDEBUG"
68.Sh DESCRIPTION
69Mutexes are used in the kernel to implement mutual exclusion among LWPs
70(lightweight processes) and interrupt handlers.
71.Pp
72The
73.Vt kmutex_t
74type provides storage for the mutex object.
75This should be treated as an opaque object and not examined directly by
76consumers.
77.Pp
78Mutexes replace the
79.Xr spl 9
80system traditionally used to provide synchronization between interrupt
81handlers and LWPs.
82.Sh OPTIONS
83.Bl -tag -width abcd
84.It Cd "options DIAGNOSTIC"
85.Pp
86Kernels compiled with the
87.Dv DIAGNOSTIC
88option perform basic sanity checks on mutex operations.
89.It Cd "options LOCKDEBUG"
90.Pp
91Kernels compiled with the
92.Dv LOCKDEBUG
93option perform potentially CPU intensive sanity checks
94on mutex operations.
95.El
96.Sh FUNCTIONS
97.Bl -tag -width abcd
98.It Fn mutex_init "mtx" "type" "ipl"
99.Pp
100Dynamically initialize a mutex for use.
101.Pp
102No other operations can be performed on a mutex until it has been initialized.
103Once initialized, all types of mutex are manipulated using the same interface.
104Note that
105.Fn mutex_init
106may block in order to allocate memory.
107.Pp
108The
109.Fa type
110argument must be given as
111.Dv MUTEX_DEFAULT .
112Other constants are defined but are for low-level system use and are not
113an endorsed, stable part of the interface.
114.Pp
115The type of mutex returned depends on the
116.Fa ipl
117argument:
118.Bl -tag -width abcd
119.It IPL_NONE, or one of the IPL_SOFT* constants
120.Pp
121An adaptive mutex will be returned.
122Adaptive mutexes provide mutual exclusion between LWPs,
123and between LWPs and soft interrupt handlers.
124.Pp
125Adaptive mutexes cannot be acquired from a hardware interrupt handler.
126An LWP may either sleep or busy-wait when attempting to acquire
127an adaptive mutex that is already held.
128.It IPL_VM, IPL_SCHED, IPL_HIGH
129.Pp
130A spin mutex will be returned.
131Spin mutexes provide mutual exclusion between LWPs, and between LWPs
132and interrupt handlers.
133.Pp
134The
135.Fa ipl
136argument is used to pass a system interrupt priority level (IPL)
137that will block all interrupt handlers that may try to acquire the mutex.
138.Pp
139LWPs that own spin mutexes may not sleep, and therefore must not
140try to acquire adaptive mutexes or other sleep locks.
141.Pp
142A processor will always busy-wait when attempting to acquire
143a spin mutex that is already held.
144.Pp
145.Sy Note :
146Releasing a spin mutex may not lower the IPL to what it was when
147entered.
148If other spin mutexes are held, the IPL will not be lowered until the
149last one is released.
150.Pp
151This is usually not a problem because spin mutexes should held only for
152very short durations anyway, so blocking higher-priority interrupts a
153little longer doesn't hurt much.
154But it interferes with writing assertions that the IPL is
155.Em no higher than
156a specified level.
157.El
158.Pp
159See
160.Xr spl 9
161for further information on interrupt priority levels (IPLs).
162.It Fn mutex_destroy "mtx"
163.Pp
164Release resources used by a mutex.
165The mutex may not be used after it has been destroyed.
166.Fn mutex_destroy
167may block in order to free memory.
168.It Fn mutex_enter "mtx"
169.Pp
170Acquire a mutex.
171If the mutex is already held, the caller will block and not return until the
172mutex is acquired.
173.Pp
174All loads and stores after
175.Fn mutex_enter
176will not be reordered before it or served from a prior cache, and hence
177will
178.Em happen after
179any prior
180.Fn mutex_exit
181to release the mutex even on another CPU or in an interrupt.
182Thus, there is a global total ordering on all loads and stores under
183the same mutex.
184.Pp
185Mutexes and other types of locks must always be acquired in a
186consistent order with respect to each other.
187Otherwise, the potential for system deadlock exists.
188.Pp
189Adaptive mutexes and other types of lock that can sleep may
190not be acquired while a spin mutex is held by the caller.
191.Pp
192When acquiring a spin mutex, the IPL of the current CPU will be raised to
193the level set in
194.Fn mutex_init
195if it is not already equal or higher.
196.It Fn mutex_exit "mtx"
197.Pp
198Release a mutex.
199The mutex must have been previously acquired by the caller.
200Mutexes may be released out of order as needed.
201.Pp
202All loads and stores before
203.Fn mutex_exit
204will not be reordered after it or delayed in a write buffer, and hence
205will
206.Fn happen before
207any subsequent
208.Fn mutex_enter
209to acquire the mutex even on another CPU or in an interrupt.
210Thus, there is a global total ordering on all loads and stores under
211the same mutex.
212.It Fn mutex_ownable "mtx"
213.Pp
214When compiled with
215.Dv LOCKDEBUG
216(see
217.Xr options 4 ) ,
218ensure that the current process can successfully acquire
219.Ar mtx .
220If
221.Ar mtx
222is already owned by the current process, the system will panic
223with a "locking against myself" error.
224.Pp
225This function is needed because
226.Fn mutex_owned
227does not differentiate if a spin mutex is owned by the current process
228vs owned by another process.
229.Fn mutex_ownable
230is reasonably heavy-weight, and should only be used with
231.Xr KDASSERT 9 .
232.It Fn mutex_owned "mtx"
233.Pp
234For adaptive mutexes, return non-zero if the current LWP holds the mutex.
235For spin mutexes, return non-zero if the mutex is held, potentially by the
236current processor.
237Otherwise, return zero.
238.Pp
239.Fn mutex_owned
240is provided for making diagnostic checks to verify that a lock is held.
241For example:
242.Bd -literal
243	KASSERT(mutex_owned(&driver_lock));
244.Ed
245.Pp
246It should not be used to make locking decisions at run time.
247For spin mutexes, it must not be used to verify that a lock is not held.
248.It Fn mutex_spin_enter "mtx"
249.Pp
250Equivalent to
251.Fn mutex_enter ,
252but may only be used when it is known that
253.Ar mtx
254is a spin mutex.
255Implies the same memory ordering as
256.Fn mutex_enter .
257On some architectures, this can substantially reduce the cost of acquiring
258a spin mutex.
259.It Fn mutex_spin_exit "mtx"
260.Pp
261Equivalent to
262.Fn mutex_exit ,
263but may only be used when it is known that
264.Ar mtx
265is a spin mutex.
266Implies the same memory ordering as
267.Fn mutex_exit .
268On some architectures, this can substantially reduce the cost of releasing
269a spin mutex.
270.It Fn mutex_tryenter "mtx"
271.Pp
272Try to acquire a mutex, but do not block if the mutex is already held.
273Returns non-zero if the mutex was acquired, or zero if the mutex was
274already held.
275.Pp
276.Fn mutex_tryenter
277can be used as an optimization when acquiring locks in the wrong order.
278For example, in a setting where the convention is that
279.Dv first_lock
280must be acquired before
281.Dv second_lock ,
282the following can be used to optimistically lock in reverse order:
283.Bd -literal
284	/* We hold second_lock, but not first_lock. */
285	KASSERT(mutex_owned(&second_lock));
286
287	if (!mutex_tryenter(&first_lock)) {
288		/* Failed to get it - lock in the correct order. */
289		mutex_exit(&second_lock);
290		mutex_enter(&first_lock);
291		mutex_enter(&second_lock);
292
293		/*
294		 * We may need to recheck any conditions the code
295		 * path depends on, as we released second_lock
296		 * briefly.
297		 */
298	}
299.Ed
300.El
301.Sh CODE REFERENCES
302The core of the mutex implementation is in
303.Pa sys/kern/kern_mutex.c .
304.Pp
305The header file
306.Pa sys/sys/mutex.h
307describes the public interface, and interfaces that machine-dependent
308code must provide to support mutexes.
309.Sh SEE ALSO
310.Xr atomic_ops 3 ,
311.Xr membar_ops 3 ,
312.Xr lockstat 8 ,
313.Xr condvar 9 ,
314.Xr kpreempt 9 ,
315.Xr rwlock 9 ,
316.Xr spl 9
317.Pp
318.Rs
319.%A Jim Mauro
320.%A Richard McDougall
321.%T Solaris Internals: Core Kernel Architecture
322.%I Prentice Hall
323.%D 2001
324.%O ISBN 0-13-022496-0
325.Re
326.Sh HISTORY
327The mutex primitives first appeared in
328.Nx 5.0 .
329