xref: /netbsd-src/share/man/man9/mutex.9 (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1.\"	$NetBSD: mutex.9,v 1.31 2017/12/09 16:00:19 wiz 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.El
145.Pp
146See
147.Xr spl 9
148for further information on interrupt priority levels (IPLs).
149.It Fn mutex_destroy "mtx"
150.Pp
151Release resources used by a mutex.
152The mutex may not be used after it has been destroyed.
153.Fn mutex_destroy
154may block in order to free memory.
155.It Fn mutex_enter "mtx"
156.Pp
157Acquire a mutex.
158If the mutex is already held, the caller will block and not return until the
159mutex is acquired.
160.Pp
161All loads and stores after
162.Fn mutex_enter
163will not be reordered before it or served from a prior cache, and hence
164will
165.Em happen after
166any prior
167.Fn mutex_exit
168to release the mutex even on another CPU or in an interrupt.
169Thus, there is a global total ordering on all loads and stores under
170the same mutex.
171.Pp
172Mutexes and other types of locks must always be acquired in a
173consistent order with respect to each other.
174Otherwise, the potential for system deadlock exists.
175.Pp
176Adaptive mutexes and other types of lock that can sleep may
177not be acquired while a spin mutex is held by the caller.
178.Pp
179When acquiring a spin mutex, the IPL of the current CPU will be raised to
180the level set in
181.Fn mutex_init
182if it is not already equal or higher.
183.It Fn mutex_exit "mtx"
184.Pp
185Release a mutex.
186The mutex must have been previously acquired by the caller.
187Mutexes may be released out of order as needed.
188.Pp
189All loads and stores before
190.Fn mutex_exit
191will not be reordered after it or delayed in a write buffer, and hence
192will
193.Fn happen before
194any subsequent
195.Fn mutex_enter
196to acquire the mutex even on another CPU or in an interrupt.
197Thus, there is a global total ordering on all loads and stores under
198the same mutex.
199.It Fn mutex_ownable "mtx"
200.Pp
201When compiled with
202.Dv LOCKDEBUG
203(see
204.Xr options 4 ) ,
205ensure that the current process can successfully acquire
206.Ar mtx .
207If
208.Ar mtx
209is already owned by the current process, the system will panic
210with a "locking against myself" error.
211.Pp
212This function is needed because
213.Fn mutex_owned
214does not differentiate if a spin mutex is owned by the current process
215vs owned by another process.
216.Fn mutex_ownable
217is reasonably heavy-weight, and should only be used with
218.Xr KDASSERT 9 .
219.It Fn mutex_owned "mtx"
220.Pp
221For adaptive mutexes, return non-zero if the current LWP holds the mutex.
222For spin mutexes, return non-zero if the mutex is held, potentially by the
223current processor.
224Otherwise, return zero.
225.Pp
226.Fn mutex_owned
227is provided for making diagnostic checks to verify that a lock is held.
228For example:
229.Bd -literal
230	KASSERT(mutex_owned(&driver_lock));
231.Ed
232.Pp
233It should not be used to make locking decisions at run time.
234For spin mutexes, it must not be used to verify that a lock is not held.
235.It Fn mutex_spin_enter "mtx"
236.Pp
237Equivalent to
238.Fn mutex_enter ,
239but may only be used when it is known that
240.Ar mtx
241is a spin mutex.
242Implies the same memory ordering as
243.Fn mutex_enter .
244On some architectures, this can substantially reduce the cost of acquiring
245a spin mutex.
246.It Fn mutex_spin_exit "mtx"
247.Pp
248Equivalent to
249.Fn mutex_exit ,
250but may only be used when it is known that
251.Ar mtx
252is a spin mutex.
253Implies the same memory ordering as
254.Fn mutex_exit .
255On some architectures, this can substantially reduce the cost of releasing
256a spin mutex.
257.It Fn mutex_tryenter "mtx"
258.Pp
259Try to acquire a mutex, but do not block if the mutex is already held.
260Returns non-zero if the mutex was acquired, or zero if the mutex was
261already held.
262.Pp
263.Fn mutex_tryenter
264can be used as an optimization when acquiring locks in the wrong order.
265For example, in a setting where the convention is that
266.Dv first_lock
267must be acquired before
268.Dv second_lock ,
269the following can be used to optimistically lock in reverse order:
270.Bd -literal
271	/* We hold second_lock, but not first_lock. */
272	KASSERT(mutex_owned(&second_lock));
273
274	if (!mutex_tryenter(&first_lock)) {
275		/* Failed to get it - lock in the correct order. */
276		mutex_exit(&second_lock);
277		mutex_enter(&first_lock);
278		mutex_enter(&second_lock);
279
280		/*
281		 * We may need to recheck any conditions the code
282		 * path depends on, as we released second_lock
283		 * briefly.
284		 */
285	}
286.Ed
287.El
288.Sh CODE REFERENCES
289The core of the mutex implementation is in
290.Pa sys/kern/kern_mutex.c .
291.Pp
292The header file
293.Pa sys/sys/mutex.h
294describes the public interface, and interfaces that machine-dependent
295code must provide to support mutexes.
296.Sh SEE ALSO
297.Xr atomic_ops 3 ,
298.Xr membar_ops 3 ,
299.Xr lockstat 8 ,
300.Xr condvar 9 ,
301.Xr kpreempt 9 ,
302.Xr rwlock 9 ,
303.Xr spl 9
304.Pp
305.Rs
306.%A Jim Mauro
307.%A Richard McDougall
308.%T Solaris Internals: Core Kernel Architecture
309.%I Prentice Hall
310.%D 2001
311.%O ISBN 0-13-022496-0
312.Re
313.Sh HISTORY
314The mutex primitives first appeared in
315.Nx 5.0 .
316