Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
3. Berkeley Software Design Inc's name may not be used to endorse or
promote products derived from this software without specific prior
written permission.

THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``AS IS'' AND
ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
ARE DISCLAIMED. IN NO EVENT SHALL BERKELEY SOFTWARE DESIGN INC BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
SUCH DAMAGE.

from BSDI $Id: mutex.4,v 1.1.2.3 1998/04/27 22:53:13 ewv Exp $
$FreeBSD: head/share/man/man9/mutex.9 70001 2000-12-13 23:30:43Z jhb $

.Dd April 20, 1998 .Dt MUTEX 9 .Sh NAME .Nm mutex , .Nm mtx_init , .Nm mtx_enter , .Nm mtx_try_enter , .Nm mtx_exit , .Nm mtx_destroy , .Nm mtx_owned , .Nm mtx_assert , .Nm MUTEX_DECLARE .Nd kernel synchronization primitives .Sh SYNOPSIS .Fd #include <sys/mutex.h> .Ft int .Fn mtx_init "struct mtx *mutex" "char *name" "flags" .Ft void .Fn mtx_enter "struct mtx *mutex" "int flags" .Ft void .Fn mtx_try_enter "struct mtx *mutex" "int flags" .Ft void .Fn mtx_exit "struct mtx *mutex" "int flags" .Ft void .Fn mtx_destroy "struct mtx *mutex" .Ft int .Fn mtx_owned "struct mtx *mutex" .Ft void .Fn mtx_assert "struct mtx *mutex" "int what" .Fn MUTEX_DECLARE "modifiers" "name" .Sh DESCRIPTION Mutexes are the most basic and primary method of process synchronization. The major design considerations for mutexes are: l -enum t Acquiring and releasing uncontested mutexes should be as cheap as possible. t They must have the information and storage space to support priority propagation. t A process must be able to recursively acquire a mutex. .El

p There are currently two flavors of mutexes, those that context switch when they block and those that do not.

p By default mutexes will context switch when they are already held. As a machine dependent optimization they may spin for some amount of time before context switching. It is important to remember that since a process may be preempted at any time, the possible context switch introduced by acquiring a mutex is guaranteed to not break anything that isn't already broken.

p Mutexes which do not context switch are spin mutexes. These should only be used to protect data shared with device that require non-preemptive interrupts, and low level scheduling code. In most/all architectures both acquiring and releasing of a uncontested spin mutex is more expensive than the same operation on a non spin mutex. In order to protect an interrupt service routine from blocking against itself all interrupts are blocked on a processor while holding a spin lock. It is permissible to hold multiple spin mutexes. In this case it is a required that they be released in the opposite order to that which they were acquired.

p Once a spin mutex has been acquired it is not permissible to acquire a blocking mutex.

p The storage needed to implement a mutex is provided by a .Dv struct mtx . In general this should be treated as an opaque object and referenced only with the mutex primitives.

p The .Fn mtx_init function must be used to initialize a mutex before it can be passed to .Fn mtx_enter . The .Ar name argument is used by the witness code to classify a mutex when doing checks of lock ordering. The pointer passed in as the .Ar name is saved rather than the data it points to. The data pointed to must remain stable until the mutex is destroyed. Currently the .Ar flag argument is unused. In the future it will likely be at least used to identify spin mutexes so that debug code can verify consistent use of a mutex. It is not permissible to pass the same .Ar mutex to .Fn mtx_init multiple times without intervening calls to .Fn mtx_destroy .

p The .Fn mtx_enter function acquires a mutual exclusion lock on behalf of the currently running kernel thread. If another kernel thread is holding the mutex, the caller will be disconnected from the CPU until the mutex is available (i.e. it will sleep), spin wait for the mutex, or possibly a combination of both.

p It is possible for the same thread to recursively acquire a mutex with no ill effects; if recursion on a given mutex can be avoided, faster and smaller code will usually be generated.

p The .Fn mtx_try_enter function is used to acquire exclusive access to those objects protected by the mutex pointed to by .Ar mutex . The .Ar flag argument is used to specify various options, typically .Dv MTX_DEF is supplied. If the mutex can not be immediately acquired .Fn mtx_try_enter will return 0, otherwise the mutex will be acquired and a non-zero value will be returned.

p The .Fn mtx_exit function releases a mutual exclusion lock; if a higher priority thread is waiting for the mutex, the releasing thread may be disconnected to allow the higher priority thread to acquire the mutex and run.

p The .Fn mtx_destroy function is used to destroy .Ar mutex so the data associated with it may be freed or otherwise overwritten. Any mutex which is destroyed must previously have been initialized with .Fn mtx_init . It is permissible to have a single hold count on a mutex when it is destroyed. It is not permissible to hold the mutex recursively, or have another process blocked on the mutex when it is destroyed.

p The .Fn mtx_owned function returns non-zero if the current process holds .Ar mutex . If the current process does not hold .Ar mutex zero is returned.

p The .Fn mtx_assert function allows assertions to be made about .Ar mutex . If the assertions are not true and the kernel is compiled with .Dv INVARIANTS then the kernel will panic. Currently the following assertions are supported: l -enum t .Dv MA_OWNED Assert that the current thread holds the mutex pointed to by the first argument. t .Dv MA_NOTOWNED Assert that the current thread does not hold the mutex pointed to by the first argument. t .Dv MA_RECURSED Assert that the current thread has recursed on the mutex pointed to by the first argument. This assertion is only valid in conjuction with .Dv MA_OWNED . t .Dv MA_NOTRECURSED Assert that the current thread has not recursed on the mutex pointed to by the first argument. This assertion is only valid in conjuction with .Dv MA_OWNED . .El

p The .Fn MUTEX_DECLARE macro is used to declare a mutex that is initialized before .Xr malloc 9 is operating. Unfortunately, mutex initialization may require .Xr malloc 9 . However, some mutexes are intialized and used before .Xr malloc 9 can be used. Declaring these mutexes with the .Fn MUTEX_DECLARE macro and then using the .Dv MTX_COLD flag when calling .Fn mtx_init allows these early mutexes to be initialized and used before .Xr malloc 9 is available. The .Ar modifiers argument is a list of attributes to be applied to the mutex structure being declared such as .Dq static . The .Ar name argument is the name of the mutex structure to declare.

p The type of a mutex is not an attribute of the mutex, but instead a function of the .Fa flags argument passed to .Fn mtx_enter and .Fn mtx_exit ; this allows code to be generated for the specific mutex type at compile time and avoids wasting run time on the determination of lock features. This does place on the programmer, the burden of using matching forms of the .Fn mtx_enter and .Fn mtx_exit functions for a given mutex. It is an error to acquire a mutex in one mode (e.g. spin) and release it in another (e.g. default). It is also an error to get the lock in one mode and allow another thread to attempt to get the lock in another mode. A good general rule is to always use a given mutex in one mode only. .Ss The default Mutex Type Most kernel code should use the default lock type; the default lock type will allow the thread to be disconnected from the CPU if it cannot get the lock. The machine dependent implementation may treat the lock as a short term spin lock under some circumstances. However, it is always safe to use these forms of locks in an interrupt thread without fear of deadlock against an interrupted thread on the same CPU. .Ss The spin Mutex Type A spin mutex will not relinquish the CPU when it cannot immediately get the requested lock, but will loop, waiting for the mutex to be released by another CPU. This could result in deadlock if a thread interrupted the thread which held a mutex and then tried to acquire the mutex; for this reason spin locks will disable all interrupts (on the local CPU only) by default.

p Spin locks are fairly specialized locks that are intended to be held for very short periods of time; their primary purpose is to protect portions of the code that implement default (i.e. sleep) locks. .Ss Flags The flags passed to the .Fn mtx_enter and .Fn mtx_exit functions determine what type of mutex is being used and also provide various options used to generate more efficient code under certain circumstances.

p Both lock types (default and spin) can be acquired recursively by the same thread. This behavior can be changed with flags.

p The type of the mutex must always be specified: l -tag -width MTX_NORECURSE t Dv MTX_DEF Default lock type; will always allow the current thread to be suspended to avoid deadlock conditions against interrupt threads. The machine dependent implementation of this lock type may spin for a while before suspending the current thread. Most locks should be of this type. t Dv MTX_SPIN Spin lock; will never relinquish the CPU. By default all interrupts are disabled on the local CPU while any spin lock is held. .El

p Options that modify mutex behavior: l -tag -width MTX_NORECURSE t Dv MTX_NORECURSE If it is known, absolutely, that the mutex will not be recursively acquired at this invocation then this flag should be specified.

p If the lock is already held by the current thread, then a kernel with .Dv MUTEX_DEBUG defined will panic; without debugging enabled, the thread may deadlock against itself or leave the mutex in a corrupted state.

p This flag prevents generation of additional inline code to deal with recursive lock acquisitions and should be specified whenever possible in the interests of efficiency. Not specifying this flag will only cause the generated code to be a little larger than necessary; it will still operate correctly. t Dv MTX_RLIKELY This provides a hint that it is likely that this mutex will be held recursively at this invocation. The actual optimization used is machine dependent; generally, this will inline code to handle recursion where a function call would otherwise be needed.

p This is a hint only; leaving it out or specifying it inappropriately will not cause any great harm other than possibly generating less efficient code. t Dv MTX_TOPHALF This option applies to spin locks only. It indicates that the mutex is never acquired from an interrupt thread, so it is safe to leave interrupts enabled while holding the lock. Since an interrupt may occur while holding the lock, this may be detrimental to other processors spin waiting for the lock. Do not forget to include this option when the lock is released.

p This option should not be used in new code; it is documented here for completeness only. t Dv MTX_FIRST This option applies to spin locks only. It indicates this is the first spin lock acquired by the thread. No other spin locks may be held, and the requested lock also may not be currently held. Do not forget to include this option when the lock is released. t Dv MTX_NOSWITCH When releasing a mutex, this flag prevents a thread switch that might occur if another higher priority thread was waiting for the mutex. This may cause priority inversion and should be used carefully.

p This flag is used internally by the lock code. It should not be used in general kernel code and is documented here for completeness only. t Dv MTX_NOSPIN For default locks, this hint will prevent spinning before relinquishing the CPU. This should be specified when it is known that the lock will usually remain unavailable for some time when it is not immediately available (i.e.: coarse grained locks protecting large subsystems). t Dv MTX_COLD This option is only used in .Fn mtx_init and is used in conjunction with mutexes declared with .Fn MUTEX_DECLARE to initialize mutexes that are needed before .Xr malloc 9 is available for use. t Dv MTX_QUIET This option is used to quiet logging messages during mutex operations. This can be used to trim superfluous logging messages for debugging purposes. .El .Sh HISTORY These functions appeared in BSD/OS 4.1 and .Fx 5.0 .